Datasets:

nkandpa2

/

arxiv-papers

like 1

0704.3395

# General-Purpose Computing on a Semantic Network Substrate111Rodriguez, M.A., “General-Purpose Computing on a Semantic Network Substrate,” Emergent Web Intelligence: Advanced Semantic Technologies, Advanced Information and Knowledge Processing series, eds. R. Chbeir, A. Hassanien, A. Abraham, and Y. Badr, Springer-Verlag, pages 57-104, ISBN:978-1-84996-076-2, June 2010. Marko A. Rodriguez Digital Library Research and Prototyping Team Los Alamos National Laboratory Los Alamos, New Mexico 87545 (original: April 16, 2007 revision: October 7, 2007) ###### Abstract This article presents a model of general-purpose computing on a semantic network substrate. The concepts presented are applicable to any semantic network representation. However, due to the standards and technological infrastructure devoted to the Semantic Web effort, this article is presented from this point of view. In the proposed model of computing, the application programming interface, the run-time program, and the state of the computing virtual machine are all represented in the Resource Description Framework (RDF). The implementation of the concepts presented provides a practical computing paradigm that leverages the highly-distributed and standardized representational-layer of the Semantic Web. keywords: Resource Description Framework, Web Ontology Language, Virtual Machines, Object-Oriented Programming, Semantic Web Computing ## 1 Introduction This article discusses computing in semantic networks. A semantic network is a directed labeled graph ( ?). The thesis of this article is that the state of a computing machine, its low-level instructions, and the executing program can be represented as a semantic network. The computational model that is presented can be instantiated using any semantic network representation. However, given the existence of the Resource Description Framework (RDF) ( ?) and the popular Web Ontology Language (OWL) ( ?), this article presents the theory and the application in terms of these constructs. The computing model that is proposed is perhaps simple in theory, but in application, requires a relatively strong background in the computer sciences. This article discusses a wide breadth of concepts including those from computer architecture, the Semantic Web, and object-oriented programming. In order to accommodate all interested readers, each discipline’s concepts will be introduced at a tutorial level in this introduction. The remainder of the article presents a more in-depth practical application of the proposed model. The practical application includes a specification for an RDF virtual machine architecture (RVM) called Fhat (pronounced făt, like “fat”) and an RDF programming language designed specifically for that architecture called Neno (pronounced nēnō, like “knee-know”). The introduction to this article is split into three subsections. §1.1 provides a brief introduction to the field of computer architecture in order to elucidate those concepts which will be of primary interest later in the article. §1.2 discusses the Semantic Web and the RDF semantic network data model. Finally, §1.3 provides an overview of object-oriented programming and its relation to OWL. ### 1.1 General-Purpose Computing and the Virtual Machine A general-purpose computer is one that can support any known computation. The meaning of computability and what is required to compute was first developed by Alan Turning in the late 1930s ( ?). Since then, and with the large leaps in the engineering of computing machines, most computers of today are general- purpose computing machines. The two primary components of a computing machine are the central processing unit (CPU) and the main memory (RAM). The purpose of the CPU is to perform calculations (i.e. execute algorithms on data). Generally, the CPU reads instructions and data from RAM, performs a calculation, and re-inserts its results back into RAM. Most CPUs maintain a relatively small set of instructions that they can execute ( ?). Example instructions include add, sub, load, store, branch, goto, etc. However, these primitive instructions can be composed to perform any computing task desired by the programmer. Currently, the smallest unit of information in a digital computer is the bit. A bit can either be a $0$ or a $1$. Bits are combined to form bytes (8-bits) and words (machine architecture dependent). Most desktop machines of everyday use have a 32-bit word and are called 32-bit machines. 32-bits can be used to represent $2^{32}$ different “things”. For example, an unsigned 32-bit integer can represent the numbers 0 to 4,294,967,295. While instructions, like data, are represented as a series of 0s and 1s, it is possible to represent the instructions in a more human readable form. The abstraction above binary machine language is called assembly language ( ?). For example, the following three assembly instructions load 2, 3 load 1, 2 add 1, 2, 3 store 3, 2 instruct the CPU to 1) read the word in the memory cell at memory address $2$ in RAM and store it in CPU register $3$, 2) read the word at memory address $1$ and store it in register $2$, 3) add the contents of register $1$ and $2$ and store the result in register $3$, and finally 4) store the word in register $3$ into memory address $2$ of RAM. Modern day computer languages are written at a much higher level of abstraction than both machine and assembly language. For instance, the previous instructions could be represented by a single statement as z = y + x where register $1$ holds the value of variable y, register $2$ holds the value of variable x and memory address $3$ holds the value of variable z. To the modern-day programmer, the low-level CPU instructions are hidden. It is the role of the language compiler to translate the human readable/writeable source code into the machine code of the CPU. Simply stated, a compiler is a computer program that translates information written in one language to another language ( ?). In practice, the compiler translates the human code to a list of CPU instructions that are stored in RAM and executed by the CPU in sequential order as pointed to by the CPU’s program counter (PC). In some instances, the compiler will translate the human code to the native language of the CPU (the instruction set of the CPU). In other cases, the compiler will translate the human code to another language for a virtual machine to compute ( ?). Virtual machine language is called byte-code. A virtual machine is, for all practical purposes, a CPU represented in software, not hardware. However, depending on the complexity of the implementation, a virtual machine can either do a hard implementation (where an exact software replica of hardware is instantiated) or a soft implementation (where more of the hardware components are black-boxed in software). The virtual machine computes its byte-code instructions by using the underlying hardware CPU’s instruction set. Finally, to complete the computation stack, the CPU relies on the underlying physical laws of nature to compute its instructions. The laws of physics drive the CPU from state to state. The evolution of states and its effects on the world is computing. Perhaps the most popular virtual machine is the Java virtual machine (JVM) ( ?) of the Java programming language ( ?). The JVM is a piece of software that runs on a physical machine. The JVM has its own instruction set much like a hardware CPU has its own instruction set. The JVM resides in RAM and requires the physical CPU to compute its evolution. Thus, the JVM translates its instructions to the instruction set of the native CPU. The benefit of this model is that irrespective of the underlying hardware CPU, a JVM designed for that CPU architecture can read and process any Java software (i.e. any Java byte-code). The drawback, is that the computation is slower than when instructions are represented as instructions for the native CPU. ### 1.2 The Semantic Web and RDF The previous section described the most fundamental aspects of computing. This section presents one of the most abstract levels of computing: the Semantic Web. The primary goal of the Semantic Web effort is to provide a standardized framework for describing resources (both physical and conceptual) and their relationships to one another ( ?). This framework is called the Resource Description Framework (RDF) ( ?).222Note that RDF is a data model, not a syntax. RDF has many different syntaxes like RDF/XML ( ?), Notation 3 (N3) ( ?), the N-TRIPLE format ( ?), and TRiX ( ?). RDF maintains two central tenets. The first states that the lowest unit of representation is the Universal Resource Identifier (URI) ( ?) and the literal (i.e. strings, integers, floating point numbers, etc.).333There also exist blank or anonymous nodes. There will be no discussion of blank nodes in this article. A URI unambiguously identifies a resource. When two resources share the same URI, they are the same resource. However, note that when two resources do not share the same URI, this does not necessarily mean that they are not the same resource. In practice, URIs are namespaced to ensure that name conflicts do not occur across different organizations ( ?). For example, the URI http://www.newspaper.org/Article can have a different meaning, or connotation, than http://www.science.net/Article because they are from different namespaces.444For the sake of brevity, prefixes are usually used instead of the full namespace. For instance, http://www.w3.org/1999/02/22-rdf- syntax-ns# is prefixed as rdf:. The second tenet of RDF states that URIs and literal values are connected to one another in sets of triples, denoting edges of a directed labeled graph. A triple is the smallest relational fact that can be asserted about the world ( ?). For instance, the statement “I am”, can be denoted $(\texttt{I},\texttt{am},\texttt{I})$ in triple form. The first element of the triple is called the subject and can be any URI. The second element is called the predicate, and it can also be any URI. Finally, the third element is called the object, and it can be any URI or literal. If $U$ denotes the set of all URIs and $L$ denotes the set of all literals, then an RDF network denoted $G$ can be defined as $G\subseteq(U\times U\times(U\cup L)).$ RDF has attracted commercial and scholarly interest, not only because of the Semantic Web vision, but because RDF provides a unique way of modeling data. This enthusiasm has sparked the development and distribution of various triple-store applications dedicated to the storage and manipulation of RDF networks ( ?). Some triple-stores can support computations on RDF networks that are on the order of $10^{10}$ triples ( ?). A triple-store is analagous to a relational database. However, instead of representing data in relational tables, a triple-store represents its data as a semantic network. A triple- store provides an interface to an RDF network for the purpose of reading from and writing to the RDF network. The most implemented query language is the SPARQL Protocol and RDF Query Language (SPARQL) ( ?). SPARQL, loosely, is a hybrid of both SQL (a relational database language) and Prolog (a logic programming language) ( ?). As an example, the following SPARQL query returns all URIs that are both a type of CognitiveScientist and ComputerScientist. SELECT ?x WHERE { ?x <rdf:type> <ComputerScientist> . ?x <rdf:type> <CognitiveScientist> } The example SPARQL query will bind the variable ?x to all URIs that are the subject of the triples with a predicate of rdf:type and objects of ComputerScientist and CognitiveScientist. For the example RDF network diagrammed in Figure 1, ?x would bind to Marko. Thus, the query above would return Marko.555Many triple-store applications support reasoning about resources during a query (at run-time). For example, suppose that the triple (Marko, rdf:type, ComputerScientist) does not exist in the RDF network, but instead there exist the triples (Marko, rdf:type, ComputerEngineer) and (ComputerEngineer, owl:sameAs, ComputerScientist). With OWL reasoning, $?x$ would still bind to Marko because ComputerEngineer and ComputerScientist are the same according to OWL semantics. The RDF computing concepts presented in this article primarily focus on triple pattern matching and thus, beyond direct URI and literal name matching, no other semantics are used. Figure 1: An example RDF network. The previous query can be represented in its more set theoretic sense as $\displaystyle X=\;$ $\displaystyle\\{?x\;|\;(?x,\texttt{rdf:type},\texttt{ComputerScientist})\in G$ $\displaystyle\;\;\;\;\wedge\;(?x,\texttt{rdf:type},\texttt{CognitiveScientist})\in G\\},$ where $X$ is the set of URIs that bind to $?x$ and $G$ is the RDF network represented as an edge list. The above syntax’s semantics is “$X$ is the set of all elements $?x$ such that $?x$ is the head of the triple ending with rdf:type, ComputerScientist and the head of the triple ending with rdf:type, CognitiveScientist, where both triples are in the triple list $G$”. Only recently has there been a proposal to extend SPARQL to support writing and deleting triples to and from an RDF network. SPARQL/Update ( ?) can be used to add the fact that Marko is also an rdf:type of Human. INSERT { <Marko> <rdf:type> <Human> . } In a more set theoretic notation, this statement is equivalent to $G=G\cup(\texttt{Marko},\texttt{rdf:type},\texttt{Human}).$ The semantics of the previous statement is: “Set the triple list $G$ to the current triple list $G$ unioned with the triple (Marko, rdf:type, Human). Finally, it is possible to remove a triple using SPARQL/Update. For instance, DELETE { <I> <am> <I> . } In set theoretic notation, this is equivalent to $G=G\setminus(\texttt{I},\texttt{am},\texttt{I}),$ where the semantics are“set the triple list $G$ to the current triple list $G$ minus the triple (I, am, I)”. ### 1.3 Object-Oriented Programming and OWL OWL is an ontology modeling language represented completely in RDF. In OWL, it is possible to model abstract classes and their relationships to one another as well as to use these models and the semantics of OWL to reason about unspecified relationships. In OWL semantics, if Human is a class and there exists the triple (Marko, rdf:type, Human), then Marko is considered an instance of Human. The URI Human is part of the ontology-level of the RDF network and the URI Marko is part of the instance-level (also called the individual-level) of the RDF network. In OWL, it is possible to state that all Humans can have another Human as a friend. This is possible by declaring an owl:ObjectProperty named hasFriend that has an rdfs:domain of Human and an rdfs:range of Human. Furthermore, it is possible to restrict the cardinality of the hasFriend property and thus, state that a Human can have no more than one friend. This is diagrammed in Figure 2.666In this article, ontology diagrams will not explicitly represent the constructs rdfs:domain, rdfs:range, nor the owl:Restriction anonymous URIs. These URIs are assumed to be apparent from the diagram. For example, the restriction shown as [0..1] in Figure 2 is represented by an owl:Restriction for the hasFriend property where the maxCardinality is $1$ and Human is an rdfs:subClassOf of this owl:Restriction. Figure 2: An ontology and an instance is represented in an RDF network. A class specification in object-oriented programming is called an application programming interface (API) ( ?). OWL ontologies share some similarities to the object-oriented API. However, OWL ontologies also differ in many respects. OWL is a description logic language that is primarily focused on a means by which to reason on RDF data. An object-oriented API is primarily focused on concretely defining classes and their explicit relationships to one another and is thus, more in line with the frames modeling paradigm. Furthermore, OWL ontologies can contain instances (i.e. individuals), allow for multiple inheritance, do not support the unique name assumption, nor the closed world assumption ( ?, ?). Another aspect of OWL that differs from object-oriented APIs is that object- oriented APIs include the concept of a method. The method is an algorithmic “behavior” that forms the foundation of the evolutionary processes that drive the instances of these classes from state to state. One of the primary purposes of this article is to introduce an OWL ontology for modeling methods and their low-level machine instructions. While process information can be represented in Frame Logic (i.e. F-Logic) ( ?), this article is primarily interested in modeling methods in much the same way that they are represented in modern day object-oriented languages such as Java and C++ and in terms of their syntax, semantics, and low-level representation. In Java and C++, a method is defined for a class and is used to manipulate the properties (called fields) of an instance of that class. For example, class Human { Human hasFriend; void makeFriend(Human h) { this.hasFriend = h; } } declares that there exists an abstract class called Human. A Human has one field called hasFriend. The hasFriend field refers to an object of type Human. Furthermore, according to the class declaration, a Human has a method called makeFriend. The makeFriend method takes a single argument that is of type Human and sets its hasFriend field to the Human provided in the argument. The this keyword makes explicit that the hasFriend field is the field of the object for which the makeFriend method was invoked. In many object-oriented languages, an instance of Human is created with the new operator. For instance, Human Marko = new Human(); creates a Human named (referenced as) Marko. The new operator is analogous to the rdf:type property. Thus, after this code is executed, a similar situation exists as that which is represented in Figure 2. However, the ontological model diagrammed in the top half of Figure 2 does not have the makeFriend method URI. The relationship between object-oriented programming and OWL is presented in Table 1. | object-oriented | OWL | example ---|---|---|--- class specification | API | ontology | Human object property | field | rdf:Property | hasFriend object method | method | | makeFriend instantiate | new operator | rdf:type property | new/rdf:type Table 1: The relationship between object-oriented programming, OWL, and the section example. It is no large conceptual leap to attach a method URI to a class. Currently, there is no strong incentive to provide a framework for representing methods in OWL. RDF was originally developed as a data modeling framework, not a programming environment per se. However, in a similar vein, the Web Ontology Language for Services (OWL-S) has been proposed as a web services model to support the discovery, execution, and tracking of the execution of Semantic Web services ( ?, ?). An OWL-S service exposes a service profile that describes what the service does, a service grounding that describes how to invoke the service, and a service model that describes how the service works. While OWL-S does provide the notion of object-oriented method invocation on the Semantic Web, OWL-S is more at the agent-oriented level and its intended use is for more “client/server” type problems. Another interesting and related idea is to use RDF as a medium for communication between various computing devices and thus, utilize the Semantic Web as an infrastructure for distributed computing ( ?). Other object-oriented notions have been proposed within the context of RDF. For instance, SWCLOS ( ?) and ActiveRDF ( ?) utilize RDF as a medium for ensuring the long-term persistence of an object. Both frameworks allow their respective languages (CLOS and Ruby) to populate the fields of their objects for use in their language environments. Once their fields have been populated, the object’s methods can be invoked in their respective programming environments. ### 1.4 The Contributions of this Article This article unifies all of the concepts presented hitherto into a framework for computing on RDF networks. In this framework, the state of a computing virtual machine, the API, and the low-level instructions are all represented in RDF. Furthermore, unlike the current programming paradigm, there is no stack of representation. The lowest level of computing and the highest level of computing are represented in the same substrate: URIs, literals, and triples. This article proposes the concept of OWL APIs, RDF triple-code, and RDF virtual machines (RVM). Human readable/writeable source code is compiled to create an OWL ontology that abstractly represents how instructions should be united to form instruction sequences.777While OWL has many features that are useful for reasoning about RDF data, the primary purpose of OWL with respect to the concepts presented in this article is to utilize OWL for its ability to create highly restricted data models. These restricted models form the APIs and ensure that instance RDF triple-code can be unambiguously generated by an RVM. When objects and their methods are instantiated from an OWL API, RDF triple-code is created. RDF triple-code is analogous to virtual machine byte- code, but instead of being represented as bits, bytes, and words, it is represented as URIs and triples. In other words, a piece of executable software is represented as a traversable RDF network. The RVM is a virtual machine whose state is represented in RDF. The RVM’s stacks, program counter, frames, etc. are modeled as an RDF network. It is the role of the RVM to “walk” the traversable RDF triple-code and compute. In summary, software is written in human readable/writeable source code, compiled to an OWL API, instantiated to RDF triple-code, and processed by a computing machine whose state is represented in RDF. However, there is always a homunculus. There is always some external process that drives the evolution of the representational substrate. For the JVM, that homunculus is the hardware CPU. For the hardware CPU, the homunculus is the physical laws of nature. For the RVM, the homunculus is some host CPU whether that host CPU is another virtual machine like the JVM or a hardware CPU. Table 2 presents the different levels of abstraction in computing and how they are represented by the physical machine, virtual machine, and proposed RDF computing paradigms. level | machine paradigm | virtual machine paradigm | RDF paradigm ---|---|---|--- high-level code | source code | source code | source code machine code | native instructions | byte-code | triple-code instruction units | bits | bits | URIs and literals machine state | hardware | software | RDF machine execution | physics | hardware | software Table 2: The various levels of abstraction in current and proposed computing paradigms. ## 2 A High-Level Perspective Assume there exists an RDF triple-store. Internal to that triple-store is an RDF network. That RDF network is composed of triples. A triple is a set of three URIs and/or literals. Those URIs can be used as a pointer to anything. This article presents a model of computation that is represented by URIs and literals and their interrelation to one another (triples). Thus, computation is represented as an RDF network. Figure 3 presents a high-level perspective on what will be discussed throughout the remainder of this article. What is diagrammed in Figure 3 is a very compartmentalized model of the components of computing. This model is in line with the common paradigm of computer science and engineering. However, less traditional realizations of this paradigm can remove the discrete levels of representation to support multi-level interactions between the various computing components since all the components are represented in the same RDF substrate: as URIs, literals, and triples. Figure 3: A high-level perspective of the Semantic Web computing environment. Figure 3 shows 6 primary components. Two of these components are at the ontological level of the RDF network, two are at the instance level of the RDF network, and two are at the machine level external to the RDF network. While there are many benefits that emerge from this computing model that are currently seen and as of yet unseen, established interesting aspects are enumerated below. 1. 1. The total address space of the RVM is the space of all URIs and literals. In the RVM model of computing, the RVM state has no concept of the underlying hardware CPU’s address space because instructions and data are represented in RDF. This idea is discussed in §3.1. 2. 2. The Semantic Web is no longer an information gathering infrastructure, but a distributed information processing infrastructure (the process can move to the data, the data doesn’t have to move to the process). An RVM can be “GETed” from a web-server as an RDF/XML document or “SELECTed” from an RDF triple- store. RDF programs and RVM states are “first-class” web-entities. The ramifications of this is that an RVM can move between triple-store environments and can compute on local data sets without requiring moving the data to the processor. This idea is discussed in §4.1. 3. 3. This model maintains the “write once, run anywhere” paradigm of the JVM. The RVM model ensures that human readable/writeable source code is compiled down to an intermediate language that is independent of the underlying hardware CPU executing the RVM process. 4. 4. Languages built on a semantic network substrate can have unique constructs not found in other languages (e.g. inverse field referencing, multi-instance fields, field querying, etc.). While it is theoretically possible to add these constructs to other languages, they are not provided in the core of the languages as these languages do not have an underlying semantic network data model. These novel language constructs are discussed in §3.2. 5. 5. Currently, there already exists an infrastructure to support the paradigm (triple-stores, ontology modeling languages, query languages, etc.) and thus, requires very little investment by the community. The primary investment is the development of source-to-OWL API compilers, RVMs, and the standardization of RDF triple-code and RVM distribution/security protocols. 6. 6. An RVM can be engineered at any level of complexity. It is possible to move the complexity to the software implementing the RVM process to ease machine architecture development and speed up computing time. This idea is discussed in §4.3. 7. 7. In this model, language reflection exists at the API, software, and RVM level (everything is represented in RDF). This idea is discussed in §4.2. ### 2.1 The Ontological Level The ontological level of the RDF network diagrammed in Figure 3 is represented in OWL. This subsection will discuss the two primary ontological components: the API, and the RVM architecture. #### 2.1.1 The API OWL supports the specification of class interactions. However, class interactions are specified in terms of property relationships, not method invocations. OWL has no formal way of specifying class behaviors (i.e. methods). However, in OWL, it is possible to define method and instruction classes and formally specify restrictions that dictate how instructions should be interrelated within a method. The method and instruction ontology presented in this article makes RDF a programming framework and not just a data modeling framework. #### 2.1.2 The Machine Architecture The RDF machine architecture is modeled in OWL. The machine architecture ontology is an abstract description of an instance of a particular RVM. Depending on the level of abstraction required, different machine architectures can be implemented at varying levels of detail. ### 2.2 The Instance Level The instance level of an RDF network is constrained by the requirements specified in the ontological level of the RDF network. This subsection will present the two components of the instance layer of the diagram in Figure 3. #### 2.2.1 The Program An API abstractly defines a software application. When an API is instantiated, instance RDF triple-code is created. Triple-code represents the instructions used by an RVM to compute. #### 2.2.2 The Virtual Machine An instance of the machine architecture is an RDF virtual machine (RVM). The purpose of the RVM is to represent its state (stacks, program counter, etc.) in the same RDF network as the triple-code instructions. However, the RDF- based RVM is not a “true” computer. The RVM simply represents its state in RDF. The RVM requires a software implementation outside the triple-store to compute its instructions. This requires the machine level discussed next. ### 2.3 The Machine Level The machine level is where the actual computation is executed. An RDF network is a data structure. RDF is not a processor in the common sense—it has no way of evolving itself. In order to process RDF data, some external process must read and write to the RDF network. The reading and writing of the RDF network evolves the RVM and the objects on which it is computing. This section discusses the machine level that is diagrammed in Figure 3. #### 2.3.1 The Virtual Machine Process The virtual machine process is represented in software on a particular host machine. The RVM processor must be compatible with both the triple-store interface (e.g. SPARQL/Update) and the underlying host machine. The RVM’s host machine can be the physical machine (hardware CPU) or another virtual machine. For instance, if the RVM’s machine process is implemented in the Java language, then the machine process runs in the JVM. This is diagrammed in Figure 3 by the ... component in between the virtual machine process and the physical machine. #### 2.3.2 The Physical Machine The physical machine is the actual hardware CPU. The RVM implementation translates the RDF triple-code to the host machine’s instruction set. For example, if the RVM process is running on the Intel Core Duo, then it is the role of the RVM process to translate the RDF triple-code to that specified by the Intel Core Duo instruction set. Thus, portability of this architectural model relies on a per host implementation of the RVM. Finally, to complete the computational stack, the laws of physics compute the hardware CPU. Much like the RDF representation of the RVM is a “snap-shot” representation of a computation, the hardware CPU is a silicon/electron “snap-shot” representation of a computation. ## 3 The Neno Language This section presents the specification of a programming language designed to take advantage of a pure RDF computing environment. This language is called Neno. Neno is a high-level object-oriented language that is written in a grammar similar to other object-oriented languages such as Java and C++. However, Neno provides some functionality that is not possible with other languages (i.e. not explicit in the constructs of other object-oriented languages). This functionality is not due to the sophistication of the Neno language, but instead, is due to the fact that it is written for an RDF substrate and thus, can take advantage of the flexibility of RDF and its read/write interfaces. For this reason, Neno is in a class of languages that is coined semantic network programming languages. The Ripple programming language is another such semantic network programming language ( ?). Both Neno and Ripple are Turing complete and thus, can perform any classical (non- quantum) computation. Neno source code is written in human readable/writeable plain-text like the source code of many other high-level programming languages. Neno source code is compiled by a NenoFhat compiler. The NenoFhat compiler compiles Neno source code to a Fhat OWL API. The Fhat OWL API is analogous to the jar file of Java. A Fhat RVM instantiates (loads) aspects of the API into the instance layer of the RDF network. This instantiated aspect of the API is executable RDF triple- code. A Fhat RVM processes the triple-code and thus, computes. The analogies between the Neno and Java components are presented in Table 3. artifact | Neno | Java ---|---|--- source code | AClass.neno | AClass.java compiler | nenofhat | javac API | AClass.owl | AClass.class virtual machine | fhat | java program | RDF network | JVM memory Table 3: The mapping between Neno and Java components. The following examples will only namespace those entities that are not within the namespace http://neno.lanl.gov. Thus, the default namespace is http://neno.lanl.gov (prefixed as neno). The Neno programming language is engineered to be in compliance with OWL and the XML Schema Definition (XSD) namespaces. OWL provides the concept of classes, inheritance, datatype and class properties, and property restrictions. However, Neno restricts its compiled Fhat OWL APIs to single-parent classes (i.e. multiple-inheritance is not supported) and holds the closed world assumption (i.e. only properties that are stated in the ontology can be computed on in a Neno object). This is similar to what is assumed in Java. XSD provides the specification for the literal data types (e.g. string, integer, float, double, date, time, etc.). The XSD URI namespace prefix is xsd. The lexicon that will be used to express the following concepts is drawn from object-oriented programming, not OWL. OWL parlance will only be used when completely describing the “back-end” of a particular aspect of the language. Table 4 states the relationship between OWL terms and object-oriented programming terms. OWL | object-oriented languages ---|--- owl:Class | Class neno:Method | Method rdf:Property | Field subject of rdf:type | Object Table 4: The mapping between the terms in OWL and object-oriented programming. ### 3.1 The Universally Unique Identifier Address Space Throughout the remainder of this article, Universally Unique Identifiers (UUIDs) will be continually used ( ?). The set of all UUIDs is a subset of the set of all URIs. A UUID is a 128-bit (16-byte) string that can be created in disparate environments with a near zero probability of ever being reproduced. To understand the number of UUIDs that are possible at 128-bits, it would require 1 trillion unique UUIDs to be created every nanosecond for 10 billion years to exhaust the space of all possible UUIDs.888This fact was taken from Wikipedia at http://en.wikipedia.org/wiki/UUID. A UUID can be represented as a 36 character hexadecimal string. For example, 6c3f8afe- ec3d-11db-8314-0800200c9a66, is a UUID. The hexadecimal representation will be used in all the following examples. However, for the sake of brevity, since 36 characters is too lengthy for the examples and diagrams, only the first 8 characters will be used. Thus, 6c3f8afe-ec3d-11db-8314-0800200c9a66 will be represented as 6c3f8afe. Furthermore, UUIDs, when used as URIs are namespaced as urn:uuid:6c3f8afe-ec3d-11db-8314-0800200c9a66 and for diagrams and examples, is abbreviated as urn:uuid:6c3f8afe. When Neno source code is compiled to Fhat triple-code, a UUID is created for nearly everything; every instruction class and instruction instance is identified by a UUID. When a Fhat is instantiated, a UUID is created for all the rdfs:Resources that compose the machine (i.e. stacks, frames, etc.). In typical programming environments, the programming language and its computing machine are constrained by the size of RAM (and virtual memory with most modern day operating systems). For a 32-bit machine, the maximum size of RAM is approximately 4 gigabytes. This means that there are only $2^{32}$ possible addresses and thus, words in RAM. However, for Neno, no such constraints exist. The space of all UUIDs is the address space of a Fhat RVM (more generally, the space of all URIs and literals is the address space). Fhat does not use RAM for storing its data and instructions, Fhat uses an RDF network. Thus, Fhat does not have any hard constraint on how much memory it “allocates” for its processing. ### 3.2 Class Declarations in Neno Source Code Neno source code has a grammar that is very similar to other object-oriented languages. For instance, suppose the following simple class written in the Java programming language: package gov.lanl.neno.demo; import java.lang.*; import java.util.*; public class Human { private String hasName; private ArrayList<Human> hasFriend; public Human (String n) { this.hasName = n; } public void makeFriend(Human h) { if(h != this) this.hasFriend.add(h); } public void setName(String n) { this.hasName = n; } } The Human class has two fields named hasName and hasFriend. The field hasName takes a value of String (or java.lang.String to be more specific) and hasFriend takes a value of Human. The Human class has one constructor and one method. A constructor is used to create an object and is a type of method. In Java, a constructor tells the JVM to allocate memory for the object on the heap (i.e. an object “pool”) and set the object’s field values according to the statements in the body of the constructor. The constructor for Human takes a String called n and creates a new Human instance called an object. The Human constructor sets that object’s hasName field to n. The Human method is called makeFriend. This method takes a Human with variable name h as an argument. If the object referenced by h is not the Human for which this method was invoked, then the object for which this method was called has h added to its hasFriend field. Note, that unlike the example in Figure 2, it is possible for a Human object to have multiple friends because of the use of the ArrayList<Human>.999Java generics as represented by the $<\;>$ notation is supported by Java 1.5+. The Neno programming language is similar to Java. The following source code demonstrates how to declare nearly the same class in Neno.101010When there are no ambiguities in naming, the class declaration can be written without prefixes. prefix owl: <http://www.w3.org/2002/07/owl>; prefix xsd: <http://www.w3.org/2001/XMLSchema>; prefix demo: <http://neno.lanl.gov/demo>; owl:Thing demo:Human { xsd:string hasName[1]; demo:Human hasFriend[0..*]; !Human(xsd:string n) { this.hasName = n; } makeFriend(demo:Human h) { if(h != this) this.hasFriend =+ h; } setName(xsd:string n) { this.hasName = n; } } While the Human class declaration in Java and in Neno are nearly identical, there are a few constructs that make the two languages different. For one, instead of “importing” packages, in Neno, namespaces are declared and ontologies are imported.111111Note that languages such as Java and C++ do maintain the concept of package namespaces. To ease namespace declarations, prefixes are used (e.g. owl, xsd, and demo). All constructors are denoted by the class name prefixed by the ! symbol. Similarly, though not in the above example, all destructors are denoted by the class name prefixed by the `~` symbol. Notice that all datatype primitives (e.g. xsd:string) are from the XSD namespace. The Fhat RVM is engineered specifically for these datatypes. Perhaps the most unique aspect of the Neno language is the cardinality restriction specifier in the field declaration (e.g. [0..1]). Because Neno was designed for a semantic network substrate, there is nothing that prevents the same property (i.e. field) to point to multiple different URIs. In order to demand that there exist no more than one field, the [0..1] notation is used. Note that demo:Human is an rdfs:subClassOf owl:Thing as specified by the owl:Thing demo:Human class description. Class inheritance is specified by the prefix to the declaration of the class name. Note that in Neno, a class can only have a single parent even though OWL supports multiple- inhertiance.121212This constraint does not apply to owl:Restrictions as Neno classes utilize owl:Restrictions to make explicit property restrictions. Thus, excluding owl:Restriction subclassing, a Neno object class can only be the subclass of a single class. Furthermore, note that all class properties have a universal restriction on the class or datatype value. For example, in the above demo:Human class, the hasName property must have an xsd:string value and all hasFriend properties must have demo:Human values. In order to demonstrate the relationship between Neno source code and its compiled OWL API, the following simple class example is presented. The class prefix owl: <http://www.w3.org/2002/07/owl>; prefix xsd: <http://www.w3.org/2001/XMLSchema>; prefix demo: <http://neno.lanl.gov/demo>; owl:Thing demo:Example { xsd:integer t[0..1]; test(xsd:integer n) { for(xsd:integer i=0; i < n; i++) { this.t = this.t + 1; } } } has the following OWL RDF/XML representation: <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#" xmlns:owl="http://www.w3.org/2002/07/owl#"> <owl:Ontology rdf:about="http://neno.lanl.gov"/> <owl:Ontology rdf:about="http://neno.lanl.gov/demo"> <owl:imports rdf:resource="http://neno.lanl.gov"/> </owl:Ontology> ... <!-- A PUSHVALUE INSTRUCTION --> <owl:Class rdf:about="http://neno.lanl.gov/demo#2271ea72-877c-4090-9f89-..."> <rdfs:subClassOf rdf:resource="http://neno.lanl.gov#PushValue"/> <rdfs:subClassOf> <owl:Restriction> <owl:onProperty rdf:resource="http://neno.lanl.gov#hasValue"/> <owl:allValuesFrom> <owl:Class rdf:about="http://neno.lanl.gov/demo#9792cc3c-5600-4660-..."/> </owl:allValuesFrom> </owl:Restriction> </rdfs:subClassOf> <rdfs:subClassOf> <owl:Restriction> <owl:onProperty rdf:resource="http://neno.lanl.gov#nextInst"/> <owl:allValuesFrom> <owl:Class rdf:about="http://neno.lanl.gov/demo#a80ba54c-5344-4df1-..."/> </owl:allValuesFrom> </owl:Restriction> </rdfs:subClassOf> </owl:Class> <!-- THE PUSHED VALUE --> <owl:Class rdf:about="http://neno.lanl.gov/demo#9792cc3c-5600-4660-bc1f-..."> <rdfs:subClassOf rdf:resource="http://neno.lanl.gov#LocalDirect"/> <rdfs:subClassOf> <owl:Restriction> <owl:hasValue rdf:datatype="http://www.w3.org/2001/XMLSchema#integer" >1</owl:hasValue> <owl:onProperty rdf:resource="http://neno.lanl.gov#hasURI"/> </owl:Restriction> </rdfs:subClassOf> </owl:Class> <!-- THE NEXT INSTRUCTION AFTER THE PUSHVALUE INSTRUCTION: AN ADD INSTRUCTION --> <owl:Class rdf:about="http://neno.lanl.gov/demo#a80ba54c-5344-4df1-91a0-..."> <rdfs:subClassOf rdf:resource="http://neno.lanl.gov#Add"/> <rdfs:subClassOf> <owl:Restriction> <owl:onProperty rdf:resource="http://neno.lanl.gov#hasLeft"/> <owl:allValuesFrom rdf:resource="http://neno.lanl.gov/demo#4c715d16-b6e6-..."/> </owl:Restriction> </rdfs:subClassOf> <rdfs:subClassOf> <owl:Restriction> <owl:onProperty rdf:resource="http://neno.lanl.gov#hasRight"/> <owl:allValuesFrom rdf:resource="http://neno.lanl.gov/demo#fdde7f6f-b9c0-..."/> </owl:Restriction> </rdfs:subClassOf> <rdfs:subClassOf> <owl:Restriction> <owl:onProperty rdf:resource="http://neno.lanl.gov#nextInst"/> <owl:allValuesFrom rdf:resource="http://neno.lanl.gov/demo#e3b8a797-849b-..."/> </owl:Restriction> </rdfs:subClassOf> </owl:Class> ... </rdf:RDF> The most important idea to take away from the above Fhat OWL API subset is that the role of the compiler is to generate UUID-named instruction classes that are subclasses of particular Fhat instructions (e.g. PushValue). These generated instruction classes have owl:Restrictions on them that ensure that instances of these classes are connected to one another in an unambiguous way (e.g. owl:Restrictions on their respective nextInt property) and that their operand values are made explicit (e.g. owl:Restrictions on their respective operand properties). This unambiguous instantiation is the RDF triple-code that is created when a Fhat RVM instantiates the API. For example, in the above Fhat OWL API snippet, any demo:2271ea72 PushValue instruction instance must have one and only one hasValue property. The value of that property must be a demo:9792cc3c LocalDirect value with a hasURI property value of "1"∧∧<xsd:integer>. The instance of demo:2271ea72 must also have a nextInst property that is of rdf:type demo:a80ba54c, where demo:a80ba54c is an rdfs:subClassOf Add. An instance of this demo:a80ba54c Add instruction instructs the Fhat RVM to add its hasLeft operand and its hasRight operands together. This demo:a80ba54c Add also has a nextInst property value that must be an instance of demo:e3b8a797. Though not shown, the demo:e3b8a797 is an rdfs:subClassOf Set. In this way, through strict owl:Restrictions, the flow of triple-code can be generated in an unambiguous manner by the Fhat RVM. The remainder of this section will go over the more salient aspects of the Neno programming language. #### 3.2.1 Declaring Namespaces Namespaces promote the distributed nature of the Semantic Web by ensuring that there are no URI name conflicts in the ontologies and instances of different organizations ( ?). The Java language has a similar construct called packaging. The package specification in Java supports organizational namespacing. Neno supports the prefixing of namespaces. For example, demo:Human resolves to $\texttt{http://neno.lanl.gov/demo\\#Human}.$ #### 3.2.2 Datatypes Fhat is engineered to handle xsd:anySimpleType and provides specific support for any of its derived types ( ?). The XSD namespace maintains, amongst others: xsd:string, xsd:double, xsd:integer, xsd:date, etc. Example operations include, "neno"^^xsd:string + "fhat"^^xsd:string "2007-11-30"^^xsd:date < "2007-12-01"^^xsd:date "1"^^xsd:integer - "0"^^xsd:integer Neno has low-level support for high-level datatype manipulations such as string concatenation, data and time comparisons, date incrementing, etc. Exactly what operations are allowed with what datatypes will be discussed later when describing the Fhat instruction set. #### 3.2.3 The this Variable The this variable is used in many object-oriented languages to specify the field to be accessed or the method to be invoked. All methods inherently have this as a variable they can use. The same construct exists in Neno with no variation in meaning. #### 3.2.4 Field Cardinality While Neno is an object-oriented language, it is also a semantic network programming language. Neno is more in line with the concepts of RDF than it is with those of Java and C++. One of the major distinguishing features of an object in Neno is that objects can have multi-instance fields. This means that a single field (predicate) can have more than one value (object). For instance, in Java Human marko = new Human("Marko Rodriguez"); marko.setName("Marko Antonio Rodriguez"); will initially set the hasName field of the Human object referenced by the variable name marko to “Marko Rodriguez”. The invocation of the setName method of marko will replace “Marko Rodriguez” with “Marko Antonio Rodriguez”. Thus, the field hasName has a cardinality of 1. All fields in Java have a cardinality of 1 and are universally quantified for the specified class (though taxonomical subsumption is supported). In Neno, it is possible for a field to have a cardinality greater than one. In Neno, when a class’ fields are declared, the cardinality specifier is used to denote how many properties of this type are allowed for an instance of this class. Thus, in the Neno code at the start of this section, xsd:string hasName[1]; states that any Human object must have one and only one field (property) called hasName and that hasName field points to some xsd:string. Therefore, it is illegal for the Fhat RVM to add a new hasName property to the class marko. The original property must be removed before the new property can be added. The general grammar for field restrictions in Neno is [# (..(# | *))], where # refers to some integer value. Neno does not adopt any of the OWL semantics regarding cardinality and “semantically distinct” resources. The owl:sameAs relationship between resources is not considered when determining the cardinality of a property and thus, only the explicit number of properties (explicit triples) of a particular type (predicate) are acknowledged by the NenoFhat compiler and Fhat RVM. #### 3.2.5 Handling Fields Neno provides the following field and local variable operators: =+, =-, =/, and =. These operators are called “set plus”, “set minus”, “set clear”, and “set”, respectively. The definition of these operators is made apparent through examples that demonstrate their use. For instance, from the class declarations above, the Human class has the field hasFriend. For the Java example, the hasFriend field can have more than one Human value only indirectly through the use of the ArrayList<Human> class. In Neno, no ArrayList<Human> is needed because a field can have a cardinality greater than 1. The cardinality specifier [0..*] states that there are no restrictions on the number of friends a Human can have. In order to add more friends to a Human object, the =+ operator is used. If the Human instance has the URI urn:uuid:2db4a1d2 and the provided Human argument has the URI urn:uuid:47878dcc then the =+ operator instructs Fhat to execute INSERT { <urn:uuid:2db4a1d2> <demo:hasFriend> <urn:uuid:47878dcc> .} on the triple-store. On the other hand, if the = operator was used, then Fhat would issue the following commands to the triple-store: DELETE { <urn:uuid:2db4a1d2> <demo:hasFriend> ?x .} INSERT { <urn:uuid:2db4a1d2> <demo:hasFriend> <urn:uuid:47878dcc> .} For a multi-instance field, the $=$ is a very destructive operator. For a [0..1] or [1] field, = behaves as one would expect in any other object- oriented language. Furthermore, for a [0..1] or [1] field, =+ is not allowed as it will cause the insertion of more than one property of the same predicate. In order to control the removal of fields from a multi-instance field, the =- and =/ operators can be used. For example, suppose the following method declaration in Neno makeEnemy(Human h) { this.hasFriends =- h; } The makeEnemy method will remove the Human object identified by the variable name h from the hasFriend fields. If the h variable is a reference to the URI urn:uuid:4800e2c2, then at the Fhat level, Fhat will execute the following command on the triple-store: DELETE { <urn:uuid:2db4a1d2> <demo:hasFriend> <urn:uuid:4800e2c2> .} Finally, assume that there is a rogue Human that wishes to have no friends at all. In order for this one man army to sever his ties, the $=/$ operator is used. Assume the following overloaded method declaration for a Human. makeEnemy() { this.hasFriends =/; } The above statement statement would have Fhat execute the following delete command on the triple-store: DELETE { <urn:uuid:2db4a1d2> <demo:hasFriend> ?human } #### 3.2.6 Field Querying In many cases, a field (i.e. property) will have many instances. In computer programming terms, fields can be thought of as arrays. However, these “arrays” are not objects, but simply greater than one cardinality fields. In Java, arrays are objects and high-level array objects like the java.util.ArrayList provide functions to search an array. In Neno, there are no methods that support such behaviors since fields are not objects. Instead, Neno provides language constructs that support field querying. For example, suppose the following method boolean isFriend(Human unknown) { if(this.hasFriend =? unknown) { return true; } else { return false; } } In the above isFriend method, the provided Human argument referenced by the variable name unknown is checked against all the hasFriend fields. Again, the owl:sameAs property is not respected and thus, “sameness” is determined by exact URIs. The =? operator is a conditional operator and thus, always returns either "true"∧∧xsd:boolean or "false"∧∧xsd:boolean. At the Fhat level, if this references the UUID urn:uuid:2d386232 and unknown references urn:uuid:75e05c12, then the Fhat RVM executes the following query on the triple-store: ASK { <urn:uuid:2d386232> <demo:hasFriend> <urn:uuid:75e05c12> . } Similarly, imagine the following method, boolean isFriendByName(Human unknown) { if(this.hasFriend.hasName =? unknown.hasName) { return true; } else { return false; } } Assuming the same UUID references for this and unknown from previous examples, the =? operation would have the Fhat execute the following query on the RDF network ASK { <urn:uuid:2d386232> <demo:hasFriend> ?x . ?x <demo:hasName> ?y . <urn:uuid:75e05c12> <demo:hasName> ?y } Again, there is no reasoning involved in any of these triple-store operations; only “raw” triple and URI/literal matching is used. #### 3.2.7 Looping and Conditionals Looping and conditionals are nearly identical to the Java language. In Neno, there exists the for, while, and if/else constructs. For example, a for statement is for(xsd:integer i = "0"^^xsd:integer; i<"10"^^xsd:integer; i++) { /* for block */ } a while statement is while(xsd:integer i < "10"^^xsd:integer) { /* while block */ } and an if/else statement is if(xsd:integer i < "10"^^xsd:integer) { /* if block */ } { /* else block */} It is important to note that these statements need not have the literal type specifier (e.g. xsd:integer) on every hardcoded literal. The literal type can be inferred from its context and thus, is automatically added by the compiler. For example, since i is an xsd:integer, it is assumed that $10$ is also. #### 3.2.8 Field Looping In many cases it is desirable to loop through all the resources of a field for the purposes of searching or for manipulating each resource. For instance, suppose the following Human method: namelessFaces() { for(Human h : this.hasFriend) { h.hasName = "..."^^xsd:string; } for(xsd:integer i=0; i<this.hasFriend*; i++) { Human h = this.hasFriend[i]; h.hasName = "."^^xsd:string; } } The above namelessFaces method demonstrates two types of field looping mechanisms offered by Neno. The first is analogous to the Java 1.5 language specification. With the first for loop, the variable h is set to a single hasFriend of this. The second for loop uses the index i that goes from index 0 to the size of the “array” (this.hasFriend*). The * notation in this context returns the number of hasFriend properties of the this object. In other words * returns the cardinality of the this.hasFriend field. Finally, as field values are not stored in a vector, but instead as an unordered set, the field “arrays” in Neno are not guaranteed to be ordered. Thus, this.hasFriend[1] may not be the same value later in the code. Ordering is dependent upon the triple-store’s indexing algorithm and stability of a particular order is dependent upon how often re-indexing occurs in the triple- store. It is worth noting that higher-order classes can be created such as specialized rdf:Seq and rdf:List classes to provided ordered support for arrays. #### 3.2.9 Type Checking The typeof operator can be used to determine the class type of a URI. For instance, the following statement, xsd:boolean isType = urn:uuid:2db4a1d2 typeof Human would return true if urn:uuid:2db4a1d2 is rdf:type Human or rdf:type of some class that is an rdfs:subClassOf Human. Also, xsd:boolean isType = urn:uuid:2db4a1d2 typeof rdfs:Resource always returns true. Thus, RDFS subsumption semantics are respected and thus, Neno respects the subclassing semantics employed by modern objected-oriented languages. Similarly the typeOf? operator returns the type of the resource. For instance, xsd:anyURI type = urn:uuid:2db4a1d2 typeof? returns http://neno.lanl.gov/demo#Human. #### 3.2.10 Inverse Field Referencing In object-oriented languages the “dot” operator is used to access a method or field of an object. For instance, in this.hasName, on the left of the “dot” is the object and on the right of the “dot” is the field. Whether the right hand side of the operator is a field or method can be deduced by the compiler from its context. If this resolves to the URI urn:uuid:2db4a1d2, then the following Neno code Human h[0..*] = this.hasFriend; would instruct Fhat to execute the following query: SELECT ?h WHERE { <urn:uuid:2db4a1d2> <demo:hasFriend> ?h . } According to the previous query, everything that binds to ?h will be set to the variable h. The above query says “locate all Human hasFriends of this object.” However, Neno provides another concept not found in other object- oriented languages called the “dot dot” operator. The “dot dot” operator provides support for what is called inverse field referencing (and inverse method invocation discussed next). Assume the following line in some method of some class, Human h[0..*] = this..hasFriend; The above statement says, “locate all Humans that have this object as their hasFriend.” At the Fhat level, Fhat executes the following query on the triple-store: SELECT ?h WHERE { ?h <demo:hasFriend> <urn:uuid:2db4a1d2> .} Furthermore, if the statement is Human h[0..3] = this..hasFriend; Fhat would execute: SELECT ?h WHERE { ?h <demo:hasFriend> <urn:uuid:2db4a1d2> .} LIMIT 3 #### 3.2.11 Inverse Method Invocation Like inverse field referencing, inverse method invocation is supported by Neno. Inverse method invocation will invoke all the methods that meet a particular requirement. For instance, this..hasFriend.makeEnemy(this); will ensure that all objects that have this as their friend are no longer friends with this. #### 3.2.12 Variable Scoping Variable scoping in Neno is equivalent to Java. For example, in xsd:integer a = "11"^^xsd:integer; if(a < "10"^^xsd:integer) { xsd:integer b = "2"^^xsd:integer; } else { xsd:integer c = "3"^^xsd:integer; } the true and false block of the if statement can read the variable a, but the true block can not read the c in the false block and the false block can not read the b in the true block. Also, methods are out of scope from one another. The only way methods communicate are through parameter passing, return values, and object manipulations. #### 3.2.13 Constructors and Destructors Constructors and destructors are used in object-oriented languages to create and destroy object, respectively. The concept of a constructor in Neno is similar to that of Java and C++. The concept of a destructor does not exist in Java, but does in C++. It is very important in Neno to provide the programmer an explicit way of performing object destruction. Again, unlike Java, Neno is intended to be used on a persistent semantic network substrate. Thus, when a Fhat stops executing or an object is no longer accessible by a Fhat, that object should not be automatically removed. In short, Fhat does not provide automatic garbage collection ( ?). It is the role of the programmer to explicitly remove all unwanted objects from the RDF network. In order to create a new object, the constructor of a class is called using the new operator. For example, Human marko = new Human("Marko"^^xsd:string); will generate a sub-network in the RDF network equivalent to Figure 4. Figure 4: A Fhat instance maintains a variable reference to an object. The algorithm by which Fhat creates the RDF sub-network will be discussed in the next section. For now, understand that in the variable environment of a Fhat instance there exists a variable named marko that points to the newly created Human instance (e.g. $\langle\texttt{marko},\texttt{rdf:type},\texttt{Human}\rangle$). A destructor will instruct Fhat to destroy an object. A destructor is specified in the class declaration. For instance, suppose the following specification for demo:Human: Thing Human { string hasName[1]; Human hasFriend[0..*]; !Human(string n) { this.hasName = n; } ~Human() { this.hasName =/ this.hasFriend =/ this..hasFriend =/ } } In the above class declaration !Human(string n) is a constructor and `~`Human() is a destructor. A destructor is called using the delete operator. For instance, delete marko; calls marko’s `~`Human() destructor. A class can only have at most one destructor and the destructor takes no arguments. The `~`Human() destructor removes the reference to the object’s name, removes all the references to the object’s friends, and removes all hasFriend references to that object. Thus, if the Human object has the URI urn:uuid:55b2a3b0, Fhat would execute the following commands on the triple- store: DELETE { <urn:uuid:55b2a3b0> <demo:hasName> ?name .} DELETE { <urn:uuid:55b2a3b0> <demo:hasFriend> ?human .} DELETE { ?human <demo:hasFriend> <urn:uuid:55b2a3b0> .} Behind the scenes, Fhat would also remove all the method references of urn:uuid:55b2a3b0, internal variable references to urn:uuid:55b2a3b0, and the rdf:type relationships that relate the object to the ontological-layer. When an object is properly destroyed, only its instance is removed from the RDF network. The object’s class specification still exists in the ontological- layer. #### 3.2.14 General Query In many instances, Fhat will not have a reference to a particular object. Again, the environment anticipated is one in which objects persist in the RDF network. Thus, when code is executed, it is necessary to locate the URI of a particular object for processing. In order to make this easy for the programmer, a query operator is defined called the “network query” operator and is denoted by the symbol <?. For example, xsd:string x = "Marko Antonio Rodriguez"^^xsd:string; xsd:string query = "SELECT ?x WHERE { ?x <demo:hasName> <" + x + "> } LIMIT 1"^^xsd:string; Human h[0..1] <? query; will query the RDF network for at most one Human named “Marko Antonio Rodriguez”. Note that three statements above could have been written as one. However, to demonstrate string concatenation and variable use, three were used. ### 3.3 Starting a Program in Neno In Neno, there are no static methods. Thus, there does not exist something like the public static void main(String[] args) method in Java. Instead, Fhat is provided a class URI and a method for that class that takes no arguments. The class is automatically instantiated by Fhat and the specified no-argument method is invoked. For example, if Fhat is pointed to the following Test class and main method, then the main method creates a Human, changes its name, then exits. When main exits, Fhat halts. owl:Thing demo:Test { main() { demo:Human h = new Human("Marko Rodriguez"); h.setName("Marko Antonio Rodriguez"); } } ### 3.4 Typical Use Case This section describes how a developer would typically use the Neno/Fhat environment. The terminal commands below ensure that the NenoFhat compiler translates Neno source code to a Fhat OWL API, loads the Fhat OWL API into the triple-store, instantiates a Fhat RVM, and points the RVM to the demo:Test class with a main method. Note that the third command is broken into four lines for display purposes. Do not assume that there is a newline character at the end of the first three lines of the third statement. > nenofhat Human.neno -o ntriple -t http://www.triplestore.net/sparql > nenofhat Test.neno -o xml -t http://www.triplestore.net/sparql > fhat -vmc http://neno.lanl.gov/neno#Fhat -c http://neno.lanl.gov/neno/demo#Test -cm main -t http://www.triplestore.net/sparql The first terminal command compiles the Human.neno source code into a Fhat OWL API represented in N-TRIPLE format and then inserts the Human.ntriple triples into the triple-store pointed to by the “-t” URL. The second terminal command compiles the Test.neno source code and generates a Fhat OWL API in RDF/XML called Test.xml. That RDF/XML file is then loaded into the triple-store. The nenofhat compiler can produce any of the popular RDF syntaxes. While in most cases, one or another is chosen, two different syntaxes are shown to demonstrate what is possible with the compiler. Finally, a Fhat processor is initiated. The virtual machine process (fhat) is called with a pointer to an ontological model of the desired machine architecture. The machine architecture is instantiated. The instantiated Fhat then instantiates a Test object and calls its main method. The instantiated Test main method is executable RDF triple-code. In some instances, a Fhat RVM state may already exist in the triple-store. In such cases, the following command can be invoked to point the Fhat RVM process to the stored RVM state. In the example below, assume that urn:uuid:60ab17c2 is of rdf:type Fhat. > fhat -vmi urn:uuid:60ab17c2 -t http://www.triplestore.net/sparql When the Fhat RVM state is located, fhat processes the current instruction pointed to by its programLocation. The following list outlines the flags for the nenofhat compiler, * • -o : output type (ntriple $|$ n3 $|$ xml) * • -t : triple-store interface and the fhat RVM process, * • -vmi : virtual machine instance URI * • -vmc : virtual machine class URI * • -c : start class URI * • -cm : start class no-argument method * • -t : triple-store interface. ## 4 The Fhat Virtual Machine Architecture Fhat is an RVM that was specifically designed for RDF-based semantic network languages. Fhat is a semi-hard implementation of a computing machine. Table 5 presents an explanation of the various levels of virtual machine implementations. The concept of soft, semi-hard, and hard implementations are developed here and thus, are not part of the common lexicon. In the JVM, all of the “hardware” components are represented in software and the state of the machine is not saved outside the current run-time environment. For VHSIC Hardware Description Language (VHDL) machines, the hardware components are modeled at the level of logic gates (AND, OR, XOR, NOT, etc.) ( ?). In Fhat, the hardware components are modeled in RDF (the state), but component execution is modeled in software (the process). implementation type | requirements | example ---|---|--- soft | hardware methods | Java Virtual Machine, r-Fhat semi-hard | high-level components | Fhat hard | low-level components | VHDL designs Table 5: Different VM implementation types, their requirements, and an example. There are many reasons why a semi-hard implementation was desired for Fhat and these reasons will be articulated in the sections discussing the various components of the Fhat architecture. However, while this section presents the semi-hard implementation, a soft implementation of Fhat called reduced Fhat (r-Fhat) will be briefly discussed. In short, r-Fhat is faster than the Fhat virtual machine, but does not support run-time machine portability and machine-level reflection. In other words, r-Fhat does not support those functions that require an RDF representation of the machine state. Any high-level language can be written to take advantage of the Fhat architecture. While Neno and Fhat were developed in concert and thus, are strongly connected in their requirements of one another, any language that compiles to Fhat RDF triple-code can use a Fhat RVM. This section will discuss the Fhat RVM before discussing the Fhat instruction set. Figure 5 presents the Fhat machine architecture. This machine architecture is represented in OWL and is co-located with other resources in the ontology layer of the RDF network. Figure 5: The ontological model of the Fhat virtual machine. There are $8$ primary components to the Fhat RVM. These are enumerated below for ease of reference. Each component will be discussed in more detail in the following subsections. 1. 1. Fhat: the CPU that interprets instructions and uses its various components for processing those instructions. 2. 2. halt: suspends Fhat processing when false, and permits processing when true. 3. 3. methodReuse: determines whether or not method triple-code is reused amongst object instances. 4. 4. programLocation: a pointer to the current instruction being executed (i.e. a PC). 5. 5. BlockStack: an rdf:List that can be pushed and popped for entering and exiting blocks. 6. 6. OperandStack: an rdf:List that can be pushed and popped for arithmetic computations. 7. 7. Frame: a Method unique environment for storing local variables. 8. 8. ReturnStack: an rdf:List that provides a reference to the instruction that called a method and the frame of that method. While all of these components are represented in RDF, only Fhat has an external software component. The software implementation of Fhat is called the “virtual machine process” in Figure 3.131313When the term Fhat is used, it is referring to the entire virtual machine, when the teletyped term Fhat is used, it is referring to the virtual machine process identified by the URI Fhat. #### 4.0.1 Fhat Fhat is the primary component of the Fhat RVM. Fhat is the most complicated component in the entire Fhat architecture. The high-level Neno pseudo-code for the Fhat component is Thing Fhat { execute() { while(!this.halt && this.programLocation != null) { Instruction i = this.programLocation if(i typeof Block) { ... } else if(i typeof If) { ... } else if(i typeof Expression) { ... } else if(i typeof Set) { ... } ... /* update programLocation */ } } } The above pseudo-code should be implemented in the language of the virtual machine process and thus, for the executing hardware CPU. It is worth noting that a Fhat virtual machine process can be written in Neno as demonstrated in the Neno code above. For example, assume a Neno implemented Fhat instance called Fhat1. In such cases, another Fhat, called Fhat2, is processing Fhat1. Fhat2 can be run on yet another Fhat, called Fhat3, or grounded into some other language that is translating code to the native machine language. This is possible because Neno/Fhat is Turing complete and thus, can run a simulation of itself. When a simulation of itself is run, a complete RDF virtual machine is created. In this simulation environment, both the state and process of the Fhat RVM are represented in RDF. The current version of Fhat supports most common uses of the xsd:anySimpleType and a few of these uses are summarized below: * • xsd:boolean: Not, Equals * • xsd:integer, xsd:float, xsd:double: Arithmetic, Compare * • xsd:string: Add, Compare * • xsd:date, xsd:dateTime: Add, Subtract, Compare * • xsd:anyURI: Compare. #### 4.0.2 halt At any time, Fhat can be forced to halt by setting the halt property of Fhat to true∧∧xsd:boolean. Multi-threading can be simulated in this way. A Neno program can be engineered to run a master Fhat that has a reference to the halt property of all its slave Fhats. By setting the halt property, the Fhat master can control which Fhat slaves are able to process at any one time. In essence, the master Fhat serves as an operating system. #### 4.0.3 methodReuse When methodReuse is set to true∧∧xsd:boolean, Fhat will instantiate new objects with unique instructions for each method. When methodReuse is set to false∧∧xsd:boolean, Fhat will reuse method triple-code amongst the same methods for the different objects. This will be discussed in more detail in §5.2. #### 4.0.4 programLocation The programLocation is a pointer to the current instruction being executed by Fhat. Fhat executes one instruction at a time and thus, the programLocation must always point to a single instruction. The “while” loop of Fhat simply moves the programLocation from one instruction to the next. At each instruction, Fhat interprets what the instruction is (by its rdf:type “opcode”) and uses its various components appropriately. When there are no more instructions (i.e. when there no longer exists a programLocation property), Fhat halts. #### 4.0.5 BlockStack The BlockStack is important for variable setting. When a new variable is created in a block of code, it is necessary to associate that variable with that block. When the thread of execution exits the block, all variables created in that block are dereferenced (i.e. deallocated). #### 4.0.6 OperandStack The OperandStack is a LIFO (i.e. “last in, first out”) stack that supports any rdfs:Resource. The OperandStack is used for local computations such as x = 1 + (2 * 3). For example, when x = 1 + (2 * 3) is executed by Fhat, Fhat will 1. 1. push the value 1 on the OperandStack 2. 2. push the value 2 on the OperandStack 3. 3. push the value 3 on the OperandStack 4. 4. pop both 2 and 3 off the OperandStack, multiply the two operands, and push the value 6 on the OperandStack 5. 5. pop both 1 and 6 the OperandStack, add the two operands, and push the value 7 on the OperandStack 6. 6. set the current Frame FrameVariable x to the value $7$ popped off the OperandStack. The Neno statement x = 1 + (2 * 3) is actually multiple instructions when compiled to Fhat triple-code. The NenoFhat compiler would translate the statement to the triple-code represented in Figure 6. Figure 6: The triple-code representation of the statement x = 1 + (2 * 3). It is very important to represent such components as the OperandStack component in RDF and not simply in the memory of the host CPU. Suppose that a Fhat instance is to move to another physical machine or, by chance, lose its process “back-end”. If any of these two scenarios were the case, the state of the machine is always saved in RDF and thus, would simply “freeze” to await another virtual machine process to continue its execution. If the OperandStack was represented in software and thus, in RAM, then when the software halted, the OperandStack would be lost and the state of the machine would be inconsistent with its programLocation. With an RDF state, the RAM representation of the virtual machine process has a negligible effect on the consistency of the machine. #### 4.0.7 Frame Fhat is a frame-based processor. This means that each invoked method is provided a Frame, or local environment, for its variables (i.e. FrameVariables). Due to how variables are scoped in object-oriented languages and because Neno does not support global variables, each method can only communicate with one another through parameter (i.e. method arguments) passing, return value passing, or object manipulations. When method $A$ calls method $B$, the parameters passed by method $A$ are stored in method $B$’s Frame according to the variable names in the method description. For example, assume the following method, xsd:integer methodB(xsd:integer a) { return a + "1"^^xsd:integer; } If method $A$ calls method $B$, with the statement, xsd:integer x = marko.methodB("2"^^xsd:integer); the value $2$ is placed into the Frame of method $B$ with the associated variable a. Method $B$ adds $1$ to the value and pushes the value $3$ on the OperandStack. Method $A$ pops one value off the OperandStack and sets the local variable x to the value $3$. The OperandStack is used for the placement of method return values. #### 4.0.8 ReturnStack The ReturnStack is a LIFO stack that maintains pointers to the return location of a method and the method Frame. To support recursion, the ReturnStack maintains a pointer to the specific Frame that is being returned to. In order to explain how the ReturnStack is used, an example is provided. When method $A$ calls method $B$, the next instruction of method $A$ following the method invocation instruction is pushed onto the ReturnStack. When method $B$ has completed its execution (e.g. a return is called), Fhat pops the instruction off the ReturnStack and sets its programLocation to that instruction. In this way, control is returned to method $A$ to complete its execution. When return is called in method $B$, Fhat will delete (i.e. deallocate) all triples associated with the method $B$ Frame. If return has a value (e.g. return 2), that value is pushed onto the OperandStack for method $A$ to use in its computation. ### 4.1 Migrating Fhat Across Different Host CPUs An interesting aspect of Fhat is the ability to migrate a Fhat process across various host CPUs. A Fhat implementation has two primary components: an RDF state representation and a software process. Because both the RDF triple-code and the complete state of a Fhat instance is represented in the RDF network, it does not matter which Fhat process is executing a particular Fhat state. The Fhat RDF state representation ensures that there are no global variables in the software process. The only variables created in the software process are local to the instruction being executed. Because there are no global variables in the software process, any software process can execute the Fhat RDF state without requiring inter-software process communication. For example, one host CPU can be running the Fhat software process and halt. Another CPU can then start another Fhat software process that points to the URI of the originally halted Fhat RDF state and continue its execution. This concept is diagrammed in Figure 7 where $n=1$ refers to instruction 1. Figure 7: Migrating a Fhat execution across multiple host CPUs. In principle, each CPU can execute one instruction and then halt. In this way, it is possible to migrate the Fhat RVM across different host CPU’s. Thus, if a portion of the Semantic Web is needed for a particular computation, it may be best to have the physical computer supporting that RDF sub-network host the Fhat RVM. Once the Fhat RVM has completed computing that particular RDF sub- network, it can halt and another CPU can pick up the process on a yet another area of the Semantic Web that needs computing by the Fhat RVM. In this model of computing, data doesn’t move to the process, the process moves to the data. This idea is diagrammed in Figure 8, where both triple-store servers have Fhat process implementations. Figure 8: Migrating a Fhat state across different triple-stores. ### 4.2 Fhat Reflection A Fhat RVM and the triple-code that it is executing are in the same address space and thus, can reference one another. It is the UUID address space of Neno/Fhat that makes it a unique programming environment in that Neno is not only a completely reflective language, but also that it removes the representational stack found in most other programming environments. Language reflection means that the program can modify itself during its execution. Many scripting languages and even Java (through the java.lang.reflect package) support language reflection. However, not only does Neno/Fhat support language reflection, it also supports machine reflection. A Fhat can modify itself during its execution. There are no true boundaries between the various components of the computation. This idea is represented in Figure 9, where a Fhat RVM has its program counter (programLocation) pointing to a Push instruction. The Push instruction is instructing Fhat to push a reference to itself on its operand stack. With a reference to the Fhat instance in the Fhat operand stack, Fhat can manipulate its own components. Thus, the Fhat RVM is executing triple-code that is manipulating itself. Figure 9: A Fhat processor can process itself. ### 4.3 r-Fhat What has been presented thus far is a semi-hard implementation of Fhat. The semi-hard implementation explicitly encodes the state of a Fhat instance in RDF. While this has benefits such as fault tolerance due to virtual machine process failures, support for distributed computing in the form of processor migration, and support for machine-based evolutionary algorithms, it requires a large read/write overhead. Each instruction requires the virtual machine process to explicitly update the virtual machine state. A faster Fhat virtual machine can be engineered that does not explicitly encode the state of the machine in the RDF network. In such cases, the only read/write operations that occur are when an object is instantiated, destroyed, or a property manipulated. This faster Fhat is called reduced Fhat (r-Fhat). In r-Fhat, the operand stack, return stack, etc. are data structures in the implementing language. r-Fhat does not have an OWL machine architecture nor an RDF state. ## 5 The Fhat Instruction Set In order for Neno software to run on a Fhat machine instance, it must be compiled to a Fhat OWL API that is compliant with the Fhat instruction set (the Fhat OWL API owl:imports the Fhat instruction set ontology). A Fhat RVM uses the Fhat OWL API as a “blueprint” for constructing the instance-level representation of the RDF triple-code. It is the instance-level triple-code that the Fhat RVM “walks” when a program is executing. ### 5.1 The Method In Neno, the only process code that exists is that which is in a Method Block. Figure 10 defines the OWL ontology of a Method. Figure 10: The OWL Method ontology. A Method has an ArgumentDescriptor that is of rdfs:subClassOf rdf:Seq and a return descriptor that is of type rdfs:Resource. The sequence of the ArgumentDescriptor Argument denotes the placement of the Method parameter in the method declaration. For instance, xsd:integer exampleMethod(xsd:string n, Human h) { ... } would set the object of the hasReturnDescriptor property to the URI xsd:integer and the ArgumentDescriptor to the Arguments n (rdf:_1) and h (rdf:_2). The hasHumanCode property can be used, if desired, to point to the original human readable/writeable source code that describes that class and its methods. By using the hasHumanCode property, it is possible for “in-network” or run-time compiling of source code. In principle, a Neno compiler can be written in Neno and be executed by a Fhat RVM. The Neno compiler can compile the representation that results from resolving the URI that is the value of the xsd:anyURI. #### 5.1.1 A Block of Fhat Triple-Code A Method has a single Block. A Block is an rdfs:subClassOf Instruction and is composed of a sequence of Instructions. The Instruction sequence is denoted by the nextInst property. The Instruction rdf:type is the “opcode” of the Instruction. The set of all Instructions is the instruction set of the Fhat architecture. Figure 11 provides a collection of the super class Instructions that can exist in a Block of code and their relationship to one another. Figure 11: The OWL ontology for a Block of Instructions. Examples of these super classes are itemized below.141414Conditions are unique in that they have a trueInst and a falseInst property. If the Condition is true, the next Instruction is the one pointed to by the trueInst property, else the next Instruction is one pointed to by the falseInst property. * • Arithmetic: Add, Divide, Multiply, Not, Subtract. * • Condition: Equals, GreaterThan, GreaterThanEqual, LessThan, LessThanEqual. * • Setter: NetQuery, Set, SetClear, SetMinus, SetPlus, SetQuery. * • Invoke: Construct, Destruct. The Value class has a set of subclasses. These subclasses are itemized below. * • Direct: LocalDirect, PopDirect. * • Variable: LocalVariable, FieldVariable, ObjectVariable. When a Fhat instance enters a Method it creates a new Frame. When a Variable is declared, that Variable is specified in the Frame and according to the current Block of the Fhat instance as denoted by Fhat’s blockTop property. A Block is used for variable scoping. When Fhat leaves a Block, it destroys all the FrameVariables in the current Frame that have that Block as their fromBlock property (refer to Figure 5). However, entering a new Block is not exiting the old Block. Parent Block FrameVariables can be accessed by child Blocks. For instance, in the following Neno code fragment, xsd:integer x = "1"^^xsd:integer; if(x > 2) { xsd:integer y = x; } else{ xsd:integer y = x; } the two y Variables in the if and else Blocks are two different FrameVariables since they are from different Blocks. Furthermore, note that both the if and else Blocks can access the value of x since they are in the child Block of the Block declaring the variable x. When Fhat leaves a Method (i.e. returns), its Frame and its FrameVariables are destroyed through dereferencing. ### 5.2 A Method Instance There are two ways in which a Method instance is handled by Fhat: global and local instance models. In the global instance model, when a new object is instantiated, its methods are also instantiated. However, if the instantiated Method already exists in the RDF network, the newly created object points its hasMethod property to a previously created Method of the same hasMethodName and UUID. Thus, only one instance of a Method exists for all the objects of the same class type. While it is possible to have a unique Method instance for each object, by supporting method reuse amongst objects, Fhat limits the growth (in terms of the number of triples) in the RDF network. Furthermore, this increases the speed of the Fhat RVM since it does not need to create a new Method from the Fhat OWL API of that Method. The global instance model is diagrammed in Figure 12. To ensure global instances, the methodReuse property of the Fhat instance is set to "true"∧∧xsd:boolean (refer to Figure 5). Figure 12: Multiple object’s of the same type will share the same Method instance. In the local instance model, the methodReuse property of a Fhat instance is set to "false"∧∧xsd:boolean. In such cases, a new Method instance is created with each new instance of an owl:Thing. The benefit of this model is that method reflection can occur on a per-object basis. If an object is to manipulate its Method triple-code at run-time, it can do so without destroying the operation of its fellow owl:Things. The drawback of the local instance model is triple-store “bloat” and an increase in the time required to instantiate an object relative to the global instance model. The local instance model is diagrammed in Figure 13. Figure 13: Multiple object’s of the same type each have a unique Method instance. #### 5.2.1 An Example Method Instance Suppose the following code, owl:Thing demo:Human { xsd:int example(xsd:string a) { if(a == "marko"^^xsd:string) { return "1"^^xsd:int; } else { return "2"^^xsd:int; } } } When this code is compiled, it compiles to a Fhat OWL API. When an instance of demo:Human is created, the Fhat RVM will start its journey at the URI demo:Human and move through the ontology creating instance UUID URIs for all the components of the demo:Human class. This includes, amongst its hard-coded properties, its Methods, their Blocks, and their Instructions. When the demo:Human class is instantiated, an instance will appear in the RDF network as diagrammed in Figure 14. Figure 14: The RDF triple-code for the example(xsd:string) method. ## 6 Conclusion The primary drawback of the RVM is computing time because a virtual machine must not only read the RDF network to interpret the program instructions, it must also read/write to the RDF network to manipulate data (i.e. instance objects) in the RDF network. Moreover, the virtual machine must read and write to that sub-network of the RDF network that represents the virtual machine’s state (e.g. program counter, operand stack, etc.). In doing so, many read/write operations occur in order for the virtual machine to compute. However, much of this issue can be resolved through the use of r-Fhat. Imagine a world where virtual machines are as easy to distribute as an HTML document (e.g. an RDF/XML encoding of the virtual machine sub-network). Given that a virtualized machine encodes its state in the RDF network, think about how RVMs can “move” between physical machines in mid-execution. There is complete hardware independence as no physical machine maintains a state representation. Physical machines compute the RVM by reading its program location, its operand stack, its heap, etc. and update those data structures. In such situations, a personal computer can be encoded in the RDF network and be accessed anywhere. Thus, the underlying physical machine is only a hardware shell for the more “personal” machine encoded in the RDF network. These ideas are analogous to those presented in ( ?). In the RDF network, RVMs, APIs, and triple-code are “first-class” web entities. What happens when archiving services such as the Internet Archive, search engine caches, and digital libraries archive such RDF programs and “snap-shot” states of the executing RVMs ( ?, ?)? In theory, the state of computing world-wide, can be saved/archived and later retrieved to resume execution. The issues and novelties that archiving computations presents are many and are left to future work in this area. Much of the Semantic Web effort is involved in the distribution of knowledge between organizational boundaries ( ?). This is perhaps the primary purpose of the ontology. In this respect, organizations of a similar domain should utilize shared ontologies in order to make their information useable between their respective organizations. Procedural encodings support the distribution of not only the knowledge models, but also the algorithms that can be applied to compute on those models. In a non-disjoint manner, data and code are easily exchanged between organizational boundaries ( ?). Given that the RDF network is composed of triples and triples are composed of URIs and literals, the address space of any virtual machine in the RDF network is the set of all URIs and literals. Given that there are no bounds to the size of these resources, there are no realistic space limitations on the RVM. In other words, the amount of disk-space provided world-wide to support the Semantic Web is the actual memory constraints of this model. However, the success of this distributed computing paradigm relies on the consistent use of such standards as the Link Data specification ( ?). With further developments in Linked Data models and the RVM model of computing, the Semantic Web can be made to behave like a general-purpose computer. ## Acknowledgments This research was made possible by a generous grant from the Andrew W. Mellon Foundation. Herbert Van de Sompel, Ryan Chute, and Johan Bollen all provided much insight during the development of these ideas. ## References * Aasman Aasman, J. (2006). _Allegro graph_ (Tech. Rep. No. 1). Franz Incorporated. * Aho, Sethi, & Ullman Aho, A. W., Sethi, R., & Ullman, J. D. (1986). _Compilers: Principles, techniques, and tools._ Addison-Wesley. * Alesso & Smith Alesso, H. P., & Smith, C. F. (2005). _Developing Semantic Web services._ Wellesey, MA: A.K. Peters LTD. * Beckett Beckett, D. (2001). _N-Triples_ (Tech. Rep.). University of Bristol. * Berners-Lee Berners-Lee, T. (1998). _Notation 3_ (Tech. Rep.). World Wide Web Consortium. * Berners-Lee Berners-Lee, T. (2006). _Linked data_ (Tech. Rep.). World Wide Web Consortium. * Berners-Lee, Fielding, Software, Masinter, & Systems Berners-Lee, T., Fielding, R. T., Software, D., Masinter, L., & Systems, A. (2005, January). _Uniform Resource Identifier (URI): Generic Syntax._ * Biron & Malhotra Biron, P. V., & Malhotra, A. (2004). _XML schema part 2: Datatypes second edition_ (Tech. Rep.). World Wide Web Consortium. * Bray, Hollander, Layman, & Tobin Bray, T., Hollander, D., Layman, A., & Tobin, R. (2006). _Namespaces in XML 1.0_ (Tech. Rep.). World Wide Web Consortium. * Britton Britton, R. (2003). _MIPS assembly language programming._ Prentice Hall. * Carroll & Stickler Carroll, J. J., & Stickler, P. (2004). RDF triples in XML. In _Extreme markup languages._ Montréal, Québec: IDEAlliance. * Coelho & Stanculescu Coelho, D., & Stanculescu, A. (1988). A state-of-the-art VHDL simulator. In _Thirty-third IEEE computer society international conference_ (p. 320-323). San Francisco, CA. * Craig Craig, I. D. (2005). _Virtual machines._ Springer. * Davies, Fensel, & Harmelen Davies, J., Fensel, D., & Harmelen, F. van. (2003). _Towards the semantic web: Ontology-driven knowledge management._ Wiley. * Eckel Eckel, B. (2002). _Thinking in Java._ Prentice Hall. * Fensel Fensel, D. (2004). Triple-space computing: Semantic web services based on pesistent publication of information. In _Proceedings of the International Conference on Intelligence in Communication Systems_ (pp. 43–53). Bangkok, Thailand. * Fensel, Hendler, Lieberman, & Wahlster Fensel, D., Hendler, J. A., Lieberman, H., & Wahlster, W. (Eds.). (2003). _Spinning the Semantic Web: Bringing the World Wide Web to its full potential._ MIT Press. * Hennessy & Patterson Hennessy, J. L., & Patterson, D. A. (2002). _Computer architecture: A quantitative approach._ Morgan Kaufmann. * Kifer, Lausen, & Wu. Kifer, M., Lausen, G., & Wu., J. (1995). Logical foundations of object-oriented and logical foundations of object-oriented and. _Journal of the ACM_ , _42_(4), 741-843. * Koide & Kawamura Koide, S., & Kawamura, M. (2004). SWCLOS: A Semantic Web processor on Common Lisp object system. In _Proceedings of the International Semantic Web Conference._ Hiroshima, Japan. * Leach Leach, P. J. (2005). _A Universally Unique IDentifier (UUID) URN Namespace_ (Tech. Rep.). Network Working Group. * Lee Lee, R. (2004). _Scalability report on triple store applications_ (Tech. Rep.). Massachusetts Institute of Technology. * Lindholm & Yellin Lindholm, T., & Yellin, F. (1999). _The Java virtual machine specification._ Addison-Wesley. * Lorie Lorie, R. A. (2001). Long term preservation of digital information. In _Proceedings of the 1st ACM/IEEE-CS joint conference on digital libraries_ (pp. 346–352). New York, NY, USA: ACM Press. * Louden Louden, K. C. (2003). _Progamming languages: Principles and practice._ Brooks/Cole - Thomson Learning. * Manola & Miller Manola, F., & Miller, E. (2004, February). _RDF primer: W3C recommendation._ * Martin et al. Martin, D., Burstein, M., Hobbs, J., Lassila, O., McDermott, D., McIlraith, S., Narayanan, S., Paolucci, M., Parsia, B., Payne, T., Sirin, E., Srinivasan, N., & Sycara, K. (2004). _OWL-S: Semantic markup for web services_ (Tech. Rep.). World Wide Web Consortium. * McGuinness & Harmelen McGuinness, D. L., & Harmelen, F. van. (2004, February). _OWL web ontology language overview._ * Nelson, McCown, Smith, & Klein Nelson, M. L., McCown, F., Smith, J., & Klein, M. (2007). Using the web infrastructure to preserve web pages. _International Journal on Digital Libraries_. * Oren, Delbru, Gerke, Haller, & Decker Oren, E., Delbru, R., Gerke, S., Haller, A., & Decker, S. (2007). ActiveRDF: Object-oriented semantic web programming. In _Proceedings of the International World Wide Web Conference WWW07._ Banff, Canada. * Prud’hommeaux & Seaborne Prud’hommeaux, E., & Seaborne, A. (2004). _SPARQL query language for RDF_ (Tech. Rep.). World Wide Web Consortium. * Rodriguez, Watkins, Bollen, & Gershenson Rodriguez, M. A., Watkins, J. H., Bollen, J., & Gershenson, C. (2007). Using RDF to model the structure and process of systems. _InterJournal of Complex Systems_ , _2131_. * Satyanarayanan et al. Satyanarayanan, M., Gilbert, B., Toups, M., Tolia, N., Surie, A., O’Hallaron, D. R., Wolbach, A., Harkes, J., Perrig, A., Farber, D. J., Kozuch, M. A., Helfrich, C. J., Nath, P., & Lagar-Cavilla, H. A. (2007). Pervasive personal computing in an internet suspend/resume system. _IEEE Internet Computing_ , _11_(2), 16-25. * Seaborne & Manjunath Seaborne, A., & Manjunath, G. (2007). _SPARQL/Update: A language for updating RDF graphs_ (Tech. Rep.). Hewlett-Packard. * Sebesta Sebesta, R. W. (2005). _Concepts of programming languages._ Addison-Wesley. * Shinavier Shinavier, J. (2007). Functional programs as Linked Data. In _3rd Workshop on Scripting for the Semantic Web._ Innsbruck, Austria. * Sowa Sowa, J. F. (1991). _Principles of semantic networks: Explorations in the representation of knowledge._ San Mateo, CA: Morgan Kaufmann. * Sowa Sowa, J. F. (1999). _Knowledge representation: Logical, philosophical, and computational foundations._ Course Technology. * Turing Turing, A. M. (1937). On computable numbers, with an application to the entscheidungsproblem. _Proceedings of the London Mathematical Society_ , _42_(2), 230–265. * Wang et al. Wang, H. H., Noy, N., Rector, A., Musen, M., Redmond, T., Rubin, D., Tu, S., Tudorache, T., Drummond, N., Horridge, M., & Sedenberg, J. (2007). Frames and OWL side by side. In _10th International Protégé Conference._ Budapest, Hungary.

arxiv-papers

2024-09-04T02:48:52.820427

0705.1023

# Angles Between Infinite Dimensional Subspaces with Applications to the Rayleigh-Ritz and Alternating Projectors Methods111A preliminary version is available at http://arxiv.org/abs/0705.1023 Andrew Knyazev 222This material is based upon work supported by the NSF DMS award 0612751. andrew[dot]knyazev[at]ucdenver[dot]edu [ Department of Mathematical and Statistical Sciences University of Colorado Denver, P.O. Box 173364, Campus Box 170, Denver, CO 80217-3364 Abram Jujunashvili Abram[dot]Jujunashvili[at]na-net[dot]ornl[dot]gov Merico Argentati Merico[dot]Argentati[at]na-net[dot]ornl[dot]gov [ ###### Abstract We define angles from-to and between infinite dimensional subspaces of a Hilbert space, inspired by the work of E. J. Hannan, 1961/1962 for general canonical correlations of stochastic processes. The spectral theory of selfadjoint operators is used to investigate the properties of the angles, e.g., to establish connections between the angles corresponding to orthogonal complements. The classical gaps and angles of Dixmier and Friedrichs are characterized in terms of the angles. We introduce principal invariant subspaces and prove that they are connected by an isometry that appears in the polar decomposition of the product of corresponding orthogonal projectors. Point angles are defined by analogy with the point operator spectrum. We bound the Hausdorff distance between the sets of the squared cosines of the angles corresponding to the original subspaces and their perturbations. We show that the squared cosines of the angles from one subspace to another can be interpreted as Ritz values in the Rayleigh-Ritz method, where the former subspace serves as a trial subspace and the orthogonal projector of the latter subspace serves as an operator in the Rayleigh-Ritz method. The Hausdorff distance between the Ritz values, corresponding to different trial subspaces, is shown to be bounded by a constant times the gap between the trial subspaces. We prove a similar eigenvalue perturbation bound that involves the gap squared. Finally, we consider the classical alternating projectors method and propose its ultimate acceleration, using the conjugate gradient approach. The corresponding convergence rate estimate is obtained in terms of the angles. We illustrate a possible acceleration for the domain decomposition method with a small overlap for the 1D diffusion equation. ©2010 Knyazev, Jujunashvili, and Argentati. All rights reserved. ###### keywords: Hilbert space , gap , canonical correlations , angles , isometry , polar decomposition , Rayleigh-Ritz method , alternating projectors , conjugate gradient , domain decomposition ††journal: Journal of Functional Analysis url]http://math.ucdenver.edu/~aknyazev/ url]http://math.ucdenver.edu/~margenta/ ## 1 Introduction Principal angles, also referred to as canonical angles, or simply as angles, between subspaces represent one of the classical mathematical tools with many applications. The cosines of the angles are related to canonical correlations which are widely used in statistics. Angles between finite dimensional subspaces have become so popular that they can be found even in linear algebra textbooks. The angles between subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ are defined as $q=\min\\{\dim{\mathcal{F}},\dim{\mathcal{G}}\\}$ values on $[0,\pi/2]$ if $q<\infty$. In the case $q=\infty$, where both subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ are infinite dimensional, traditionally only single-valued angles are defined, which in the case $q<\infty$ would correspond to the smallest (Dixmier [11]), smallest non-zero (Friedrichs [13]), or largest (Krein et al. [29]), angles. We define angles from-to and between (infinite) dimensional subspaces of a Hilbert space using the spectra of the product of corresponding orthogonal projectors. The definition is consistent with the finite dimensional case $q<\infty$ and results in a _set_ , possibly infinite, of angles. Our definition is inspired by E.J. Hannan [16], where such an approach to canonical correlations of stochastic processes is suggested. Canonical correlations for stochastic processes and functional data often involve infinite dimensional subspaces. This paper is intended to revive the interest in angles between infinite dimensional subspaces. In functional analysis, the gap and the minimum gap are important concepts used, e.g., in operator perturbation theory ([19]). The gap between infinite dimensional subspaces bounds the perturbation of a closed linear operator by measuring the change in its graph. We show in Theorem 2.12 that the gap is closely connected to the sine of the largest angle. The minimum gap between infinite dimensional subspaces provides a necessary and sufficient condition to determine if the sum of two subspaces is closed. The minimum gap is applied, e.g., in [22] to prove wellposedness of degenerate saddle point problems. The minimum gap is precisely, see Theorem 2.15, the sine of the angle of Friedrichs, which, in its turn, as shown in Theorem 2.14, is the infimum of the set of nonzero angles. The Dixmier angle is simply the smallest of all angles in our definition. We consider a (real or complex) Hilbert space equipped with an inner product $(f,g)$ and a vector norm $\|f\|=(f,f)^{1/2}$. The angle between two unit vectors $f$ and $g$ is $\theta(f,g)=\arccos|(f,g)|\in[0,\pi/2].$ In §2 of the present paper, we replace $1D$ subspaces spanned by the vectors $f$ and $g$ with (infinite dimensional) subspaces, and introduce the concept of principal angles from one subspace to another and between subspaces using the spectral theory of selfadjoint operators. We investigate the basic properties of the angles, which are already known for finite dimensional subspaces, see [23], e.g., we establish connections between the angles corresponding to subspaces and their orthogonal complements. We express classical quantities: the gap and the minimum gap between subspaces, in terms of the angles. In § 2, we provide a foundation and give necessary tools for the rest of the paper, see also [5] and references there. In § 3, we introduce principal invariant subspaces and prove that they are connected by the isometry that appears in the polar decomposition of the product of corresponding orthogonal projectors. We define point angles by analogy with the point operator spectrum and consider peculiar properties of the invariant subspaces corresponding to a point angle. In § 4, the Hausdorff distance is used to measure the change in the principal angles, where one of the subspaces varies, extending some of our previous results of [23, 25] to infinite dimensional subspaces. We consider two applications of the angles: to bound the change in Ritz values, where the Rayleigh-Ritz method is applied to different infinite dimensional trial subspaces, in §5; and to analyze and accelerate the convergence of the classical alternating projectors method (e.g., [10, Chapter IX]) in the context of a specific example—a domain decomposition method (DDM) with an overlap, in §6. In computer simulations the subspaces involved are evidently finite dimensional; however, the assumption of the finite dimensionality is sometimes irrelevant in theoretical analysis of the methods. In §5, we consider the Rayleigh-Ritz method for a bounded selfadjoint operator $A$ on a trial subspace ${\mathcal{F}}$ of a Hilbert space, where the spectrum $\Sigma((P_{{\mathcal{F}}}A)|_{{\mathcal{F}}})$ of the restriction to the subspace ${\mathcal{F}}$ of the product of the orthoprojector $P_{\mathcal{F}}$ onto ${\mathcal{F}}$ and the operator $A$ is called the set of Ritz values, corresponding to $A$ and ${\mathcal{F}}$. In the main result of §5, we bound the change in the Ritz values, where one trial subspace ${\mathcal{F}}$ is replaced with another subspace ${\mathcal{G}}$, using the Hausdorff distance between the sets of Ritz values, by the spread of the spectrum times the gap between the subspaces. The proof of the general case is based on a specific case of one dimensional subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$, spanned by unit vectors $f$ and $g,$ correspondingly, where the estimate becomes particularly simple: $\left|(f,Af)-(g,Ag)\right|\leq\left(\lambda_{\max}-\lambda_{\min}\right)\sin(\theta(f,g));$ here $\lambda_{\max}-\lambda_{\min}$ is the spread of the spectrum of $A$, cf. [24]. If in addition $f$ or $g$ is an eigenvector of $A$, the same bound holds but with the sine squared—similarly, our Hausdorff distance bound involves the gap squared, assuming that one of the trial subspaces is $A$-invariant. The material of §5 generalizes some of the earlier results of [25, 26] and [27] for the finite dimensional case. The Rayleigh-Ritz method with infinite dimensional trial subspaces is used in the method of intermediate problems for determining two-sided bounds for eigenvalues, e.g., [36, 37]. The results of §5 may be useful in obtaining a priori estimates of the accuracy of the method of intermediate problems, but this is outside of the scope of the present paper. Our other application, in § 6, is the classical alternating projectors method: $e^{(i+1)}={P_{\mathcal{F}}P_{\mathcal{G}}}e^{(i)},$ $e^{(0)}\in{\mathcal{F}}$, where ${\mathcal{F}}$ and ${\mathcal{G}}$ are two given subspaces and $P_{\mathcal{F}}$ and $P_{\mathcal{G}}$ are the orthogonal projectors onto ${\mathcal{F}}$ and ${\mathcal{G}}$, respectively. If $\left\|\left.({P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{\mathcal{F}}\right\|<1$ then the sequence of vectors $e^{(i)}$ evidently converges to zero. Such a situation is typical if $e^{(i)}$ represents an error of an iterative method, e.g., a multiplicative DDM, so that the alternating projectors method describes the error propagation in the DDM, e.g., [38, 4]. If the intersection ${\mathcal{F}}\cap{\mathcal{G}}$ is nontrivial then the sequence of vectors $e^{(i)}$ converges under reasonable assumptions to the orthogonal projection of $e^{(0)}$ onto ${\mathcal{F}}\cap{\mathcal{G}}$ as in the von Neumann-Halperin method, see [34, 15], and [2]. Several attempts to estimate and accelerate the convergence of alternating projectors method are made, e.g., [9, 2], and [39]. Here, we use a different approach, known in the DDM context, e.g., [38, 4], but apparently novel in the context of the von Neumann-Halperin method, and suggest the ultimate, conjugate gradient based, acceleration of the von Neumann-Halperin alternating projectors method. Our idea of the acceleration is inspired by the following facts. On the one hand, every selfadjoint non-negative non-expansion $A,\,0\leq A\leq I$ in a Hilbert space ${\mathcal{H}}$ can be extended to an orthogonal projector $P_{\mathcal{G}}$ in the space ${\mathcal{H}}\times{\mathcal{H}}$, e.g., [14, 31], and, thus, is unitarily equivalent to a product of two orthogonal projectors ${P_{\mathcal{F}}P_{\mathcal{G}}}$ restricted to the subspace ${\mathcal{F}}={\mathcal{H}}\times\\{0\\}$. Any polynomial iterative method that involves as a main step a multiplication of a vector by $A$ can thus be called an “alternating projectors” method. On the other hand, the conjugate gradient method is the optimal polynomial method for computing the null space of $A,$ therefore the conjugate gradient approach provides the ultimate acceleration of the alternating projectors method. We give in §6 the corresponding convergence rate estimate in terms of the angles. We illustrate a possible acceleration for the DDM with a small overlap for the 1D diffusion equation. The convergence of the classical alternating projectors method degrades when the overlap gets smaller, but the conjugate gradient method we describe converges to the exact solution in two iterations. For a finite difference approximation of the 1D diffusion equation a similar result can be found in [12]. This paper is partially based on [18], where simple proofs that we skip here can be found. ## 2 Definition and Properties of the Angles Here we define angles from one subspace to another and angles between subspaces, and investigate the properties of the (sets of) angles, such as the relationship concerning angles between the subspaces and their orthogonal complements. We express the gap and the minimum gap between subspaces in terms of angles. We introduce principal invariant subspaces and prove that they are connected by an isometry that appears in the polar decomposition of the product of corresponding orthogonal projectors. We define point angles and their multiplicities by analogy with the point operator spectrum, and consider peculiar properties of the invariant subspaces corresponding to a point angle. ### 2.1 Preliminaries Let ${\mathcal{H}}$ be a (real or complex) Hilbert space and let ${\mathcal{F}}$ and ${\mathcal{G}}$ be proper nontrivial subspaces. A subspace is defined as a closed linear manifold. Let $P_{\mathcal{F}}$ and $P_{\mathcal{G}}$ be the orthogonal projectors onto ${\mathcal{F}}$ and ${\mathcal{G}}$, respectively. We denote by ${\mathcal{B}}({\mathcal{H}})$ the Banach space of bounded linear operators defined on ${\mathcal{H}}$ with the induced norm. We use the same notation $\|\cdot\|$ for the vector norm on ${\mathcal{H}}$, associated with the inner product $(\cdot,\cdot)$ on ${\mathcal{H}}$, as well as for the induced operator norm on ${\mathcal{B}}({\mathcal{H}})$. For $T\in{\mathcal{B}}({\mathcal{H}})$ we define $|T|=\sqrt{T^{*}T}$, using the positive square root. $T|_{U}$ denotes the restriction of the operator T to its invariant subspace $U$. By $\mathfrak{D}(T)$, $\mathfrak{R}(T)$, $\mathfrak{N}(T)$, $\Sigma(T)$, and $\Sigma_{p}(T)$ we denote the domain, range, null space, spectrum, and point spectrum, respectively, of the operator $T$. In this paper, we distinguish only between finite and infinite dimensions. If $q$ is a finite number then we set by definition $\min\\{q,\,\infty\\}=q$ and $\max\\{q,\,\infty\\}=\infty,$ and assume that $\infty\leq\infty$ holds. We use $\oplus$ to highlight that the sum of subspaces is orthogonal and for the corresponding sum of operators. We denote the $\ominus$ operation between subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ by ${\mathcal{F}}\ominus{\mathcal{G}}={\mathcal{F}}\cap{\mathcal{G}}^{\perp}.$ Introducing an orthogonal decomposition ${\mathcal{H}}=\mathfrak{M}_{00}\oplus\mathfrak{M}_{01}\oplus\mathfrak{M}_{10}\oplus\mathfrak{M}_{11}\oplus\mathfrak{M},$ where $\mathfrak{M}_{00}={\mathcal{F}}\cap{\mathcal{G}},\enspace\mathfrak{M}_{01}={\mathcal{F}}\cap{\mathcal{G}}^{\bot},\enspace\mathfrak{M}_{10}={\mathcal{F}}^{\bot}\cap{\mathcal{G}},\enspace\mathfrak{M}_{11}={\mathcal{F}}^{\bot}\cap{\mathcal{G}}^{\bot},$ (see, e.g., [14, 6]), we note that every subspace in the decomposition is $P_{\mathcal{F}}$ and $P_{\mathcal{G}}$ invariant. ###### Definition 2.1. (See [14]). Two subspaces ${\mathcal{F}}\subset{\mathcal{H}}$ and ${\mathcal{G}}\subset{\mathcal{H}}$ are said to be in _generic position_ within the space ${\mathcal{H}}$, if all four subspaces $\mathfrak{M}_{00},\,\mathfrak{M}_{01},\,\mathfrak{M}_{10}$, and $\mathfrak{M}_{11}$ are null-dimensional. Clearly, subspaces ${\mathcal{F}}\subset{\mathcal{H}}$ and ${\mathcal{G}}\subset{\mathcal{H}}$ are in generic position within the space ${\mathcal{H}}$ iff any of the pairs of subspaces: ${\mathcal{F}}\subset{\mathcal{H}}$ and ${\mathcal{G}}^{\bot}\subset{\mathcal{H}}$, or ${\mathcal{F}}^{\bot}\subset{\mathcal{H}}$ and ${\mathcal{G}}\subset{\mathcal{H}}$, or ${\mathcal{F}}^{\bot}\subset{\mathcal{H}}$ and ${\mathcal{G}}^{\bot}\subset{\mathcal{H}}$, is in generic position within the space ${\mathcal{H}}$. The fifth part, $\mathfrak{M}$, can be further orthogonally split in two different ways as follows: * 1. $\mathfrak{M}=\mathfrak{M}_{{\mathcal{F}}}\oplus\mathfrak{M}_{{\mathcal{F}}^{\bot}}$ with $\mathfrak{M}_{{\mathcal{F}}}={\mathcal{F}}\ominus(\mathfrak{M}_{00}\oplus\mathfrak{M}_{01}),\enspace\mathfrak{M}_{{\mathcal{F}}^{\bot}}={\mathcal{F}}^{\bot}\ominus(\mathfrak{M}_{10}\oplus\mathfrak{M}_{11}),$ or * 2. $\mathfrak{M}=\mathfrak{M}_{{\mathcal{G}}}\oplus\mathfrak{M}_{{\mathcal{G}}^{\bot}}$ with $\mathfrak{M}_{{\mathcal{G}}}={\mathcal{G}}\ominus(\mathfrak{M}_{00}\oplus\mathfrak{M}_{10}),\enspace\mathfrak{M}_{{\mathcal{G}}^{\bot}}={\mathcal{G}}^{\bot}\ominus(\mathfrak{M}_{01}\oplus\mathfrak{M}_{11}).$ We obtain orthoprojectors’ decompositions $P_{\mathcal{F}}=I_{\mathfrak{M}_{00}}\oplus I_{\mathfrak{M}_{01}}\oplus 0_{\mathfrak{M}_{10}}\oplus 0_{\mathfrak{M}_{11}}\oplus P_{\mathcal{F}}|_{\mathfrak{M}}\text{ and }P_{\mathcal{G}}=I_{\mathfrak{M}_{00}}\oplus 0_{\mathfrak{M}_{01}}\oplus I_{\mathfrak{M}_{10}}\oplus 0_{\mathfrak{M}_{11}}\oplus P_{\mathcal{G}}|_{\mathfrak{M}},$ and decompositions of their products: $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}=I_{\mathfrak{M}_{00}}\oplus 0_{\mathfrak{M}_{01}}\oplus({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathfrak{M}_{{\mathcal{F}}}},\text{ and }({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathcal{G}}=I_{\mathfrak{M}_{00}}\oplus 0_{\mathfrak{M}_{10}}\oplus({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathfrak{M}_{{\mathcal{G}}}}.$ These decompositions are very useful in the sequel. In the next theorem we apply them to prove the unitary equivalence of the operators ${P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}$ and ${P_{\mathcal{G}}P_{\mathcal{F}}P_{\mathcal{G}}}$. ###### Theorem 2.2. Let ${\mathcal{F}}$ and ${\mathcal{G}}$ be subspaces of ${\mathcal{H}}$. Then there exists a unitary operator $W\in{\mathcal{B}}({\mathcal{H}})$ such that ${P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}=W^{*}{P_{\mathcal{G}}P_{\mathcal{F}}P_{\mathcal{G}}}W.$ ###### Proof. Denote $T={P_{\mathcal{G}}P_{\mathcal{F}}}$. Then $T^{*}={P_{\mathcal{F}}P_{\mathcal{G}}}$ and $T^{*}T={P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}.$ Using, e.g., [31, §110, p. 286] or [19, §VI.2.7, p. 334], we introduce the polar decomposition, $T=U|T|$, where $|T|=\sqrt{T^{*}T}=\sqrt{{P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}}$ is selfadjoint and nonnegative and $U:\,{\mathfrak{R}(|T|)}\to{\mathfrak{R}(T)}$ is an isometry. We extend $U$ by continuity, keeping the same notation, to the isometry $U:\,\overline{\mathfrak{R}(|T|)}\to\overline{\mathfrak{R}(T)}$. It is easy to check directly that $\mathfrak{N}(|T|)=\mathfrak{N}(T)$, so $\overline{\mathfrak{R}(|T|)}=(\mathfrak{N}(T))^{\perp}$ since $|T|$ is selfadjoint. Taking also into account that $\overline{\mathfrak{R}(T)}=(\mathfrak{N}(T^{*}))^{\perp}$, we have $U:\,(\mathfrak{N}(T))^{\perp}\to(\mathfrak{N}(T^{*}))^{\perp}$. For a general operator $T\in{\mathcal{B}}({\mathcal{H}})$, the isometry $U$ is then typically extended to a partial isometry $U\in{\mathcal{B}}({\mathcal{H}})$ by setting $U=0$ on $\mathfrak{N}(T)$. For our special $T={P_{\mathcal{G}}P_{\mathcal{F}}}$, we can do better and extend $U$ to a unitary operator $W\in{\mathcal{B}}({\mathcal{H}})$. Indeed, we set $W=U$ on $(\mathfrak{N}(T))^{\perp}$ to make $W$ an extension of $U$. To make $W$ unitary, we set $W=V$ on $\mathfrak{N}(T)$, where $V:\,\mathfrak{N}(T)\to\mathfrak{N}(T^{*})$ must be an isometry. The specific form of $V$ is of no importance, since it evidently does not affect the validity of the formula ${P_{\mathcal{G}}P_{\mathcal{F}}}=W\sqrt{{P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}}$, which implies ${P_{\mathcal{F}}P_{\mathcal{G}}}=\sqrt{{P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}}W^{*}$. Multiplying these equalities we obtain the required ${P_{\mathcal{G}}P_{\mathcal{F}}P_{\mathcal{G}}}=W{P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}W^{*}$. For the existence of such $V$, it is sufficient (and, in fact, necessary) that $\mathfrak{N}(T^{*})=\mathfrak{N}({P_{\mathcal{F}}P_{\mathcal{G}}})$ and $\mathfrak{N}(T)=\mathfrak{N}({P_{\mathcal{G}}P_{\mathcal{F}}})$ be isomorphic. Using the five-parts decomposition, we get $\mathfrak{N}({P_{\mathcal{F}}P_{\mathcal{G}}})=\mathfrak{M}_{01}\oplus\mathfrak{M}_{10}\oplus\mathfrak{M}_{11}\oplus\mathfrak{N}(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathfrak{M}}),\,\mathfrak{N}({P_{\mathcal{G}}P_{\mathcal{F}}})=\mathfrak{M}_{01}\oplus\mathfrak{M}_{10}\oplus\mathfrak{M}_{11}\oplus\mathfrak{N}(({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathfrak{M}}).$ The first three terms in the decompositions of $\mathfrak{N}({P_{\mathcal{F}}P_{\mathcal{G}}})$ and $\mathfrak{N}({P_{\mathcal{G}}P_{\mathcal{F}}})$ are the same, so $\mathfrak{N}({P_{\mathcal{F}}P_{\mathcal{G}}})$ and $\mathfrak{N}({P_{\mathcal{G}}P_{\mathcal{F}}})$ are isomorphic iff the last terms $\mathfrak{N}(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathfrak{M}})=\mathfrak{M}_{{\mathcal{G}}^{\bot}}$ and $\mathfrak{N}(({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathfrak{M}})=\mathfrak{M}_{{\mathcal{F}}^{\bot}}$ are isomorphic. The subspaces $\mathfrak{M}_{{\mathcal{F}}}=P_{\mathcal{F}}\mathfrak{M}\subseteq\mathfrak{M}$ and $\mathfrak{M}_{{\mathcal{G}}}=P_{\mathcal{G}}\mathfrak{M}\subseteq\mathfrak{M}$ are in generic position within the space $\mathfrak{M}$, see [14], as well as their orthogonal in $\mathfrak{M}$ complements $\mathfrak{M}_{{\mathcal{F}}^{\bot}}$ and $\mathfrak{M}_{{\mathcal{G}}^{\bot}}$. According to [14, Proof of Theorem 1, p. 382], any two subspaces in generic position are isomorphic, thus $\mathfrak{N}({P_{\mathcal{F}}P_{\mathcal{G}}})$ and $\mathfrak{N}({P_{\mathcal{G}}P_{\mathcal{F}}})$ are isomorphic. ∎ ###### Corollary 2.3. The operators $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathfrak{M}_{{\mathcal{F}}}}$ and $({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathfrak{M}_{{\mathcal{G}}}}$ are unitarily equivalent. ###### Proof. We have that ${P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}=({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathfrak{M}_{{\mathcal{F}}}}\oplus I_{\mathfrak{M}_{00}}\oplus 0_{{\mathcal{H}}\ominus(\mathfrak{M}_{00}\oplus\mathfrak{M}_{{\mathcal{F}}})}$ and ${P_{\mathcal{G}}P_{\mathcal{F}}P_{\mathcal{G}}}=({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathfrak{M}_{{\mathcal{G}}}}\oplus I_{\mathfrak{M}_{00}}\oplus 0_{{\mathcal{H}}\ominus(\mathfrak{M}_{00}\oplus\mathfrak{M}_{{\mathcal{G}}})}$. The subspaces $\mathfrak{M}_{{\mathcal{F}}}$ and $\mathfrak{M}_{{\mathcal{G}}}$ are connected by $\mathfrak{M}_{{\mathcal{F}}}=W\mathfrak{M}_{{\mathcal{G}}}$, $\mathfrak{M}_{{\mathcal{G}}}=W^{*}\mathfrak{M}_{{\mathcal{F}}}$, and ${P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}=W^{*}{P_{\mathcal{G}}P_{\mathcal{F}}P_{\mathcal{G}}}W$. ∎ In the important particular case $\|P_{\mathcal{F}}-P_{\mathcal{G}}\|<1$, subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ are isometric and Riesz and Sz.-Nagy [31, §VII.105] explicitly describe a partial isometry $U=P_{\mathcal{G}}[I+P_{\mathcal{F}}(P_{\mathcal{G}}-P_{\mathcal{F}})P_{\mathcal{F}}]^{-1/2}P_{\mathcal{F}}$ that maps ${\mathcal{F}}$ one-to-one and onto ${\mathcal{G}}$. On ${\mathcal{F}}$, clearly $I+P_{\mathcal{F}}(P_{\mathcal{G}}-P_{\mathcal{F}})P_{\mathcal{F}}$ is just the same as ${P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}},$ so this $U$ represents the partial isometry in the polar decomposition in the proof of our Theorem 2.2, in this case. Let $V=(I-P_{\mathcal{G}})[I+(I-P_{\mathcal{F}})((I-P_{\mathcal{G}})-(I-P_{\mathcal{F}}))(I-P_{\mathcal{F}})]^{-1/2}(I-P_{\mathcal{F}})$ be another partial isometry that maps ${\mathcal{F}}^{\bot}$ one-to-one and onto ${\mathcal{G}}^{\bot}$, constructed in the same way as $U$. Setting $W=U+V$, we extend $U$ from the subspace ${\mathcal{F}}$ to a unitary operator $W$ on the whole space. The sum $W=U+V$ is the same as the unitary extension suggested in Kato [19, §I.4.6, §I.6.8] and Davis and Kahan [7]: $\displaystyle W$ $\displaystyle=$ $\displaystyle[{P_{\mathcal{G}}P_{\mathcal{F}}}+(I-P_{\mathcal{G}})(I-P_{\mathcal{F}})][I-(P_{\mathcal{F}}-P_{\mathcal{G}})^{2}]^{-1/2}$ $\displaystyle=$ $\displaystyle[(I-P_{\mathcal{G}})(I-P_{\mathcal{F}})+P_{\mathcal{F}}P_{\mathcal{G}}]^{-1/2}[P_{\mathcal{G}}P_{\mathcal{F}}+(I-P_{\mathcal{G}})(I-P_{\mathcal{F}})]$ (the second equality holds since the corresponding terms in square brackets are the same and $(P_{\mathcal{F}}-P_{\mathcal{G}})^{2}$ commutes both with $P_{\mathcal{F}}$ and $P_{\mathcal{G}}$), which is used there to prove the unitary equivalence $P_{\mathcal{F}}=W^{*}P_{\mathcal{G}}W$. It is easy to check directly that the operator $W$ is unitary and that on ${\mathcal{F}}$ it acts the same as the operator $U$, so it is indeed a unitary extension of $U$. If $\|P_{\mathcal{F}}-P_{\mathcal{G}}\|<1$, Theorem 2.2 holds with this choice of $W$. In the next subsection we define angles from-to and between subspaces using the spectrum of the product of two orthogonal projectors. Our goal is to develop a theory of angles from-to and between subspaces based on the well- known spectral theory of selfadjoint bounded operators. ### 2.2 Angles From–To and Angles Between Subspaces ###### Definition 2.4. $\hat{\Theta}({\mathcal{F}},{\mathcal{G}})=\\{\theta:\theta=\arccos(\sigma),\sigma\geq 0,{\sigma}^{2}\in\Sigma(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{{\mathcal{F}}})\\}\subseteq[0,{\pi}/{2}]$ is called the set of angles _from_ the subspace ${\mathcal{F}}$ to the subspace ${\mathcal{G}}$. Angles $\Theta({\mathcal{F}},{\mathcal{G}})=\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\cap\hat{\Theta}({\mathcal{G}},{\mathcal{F}})$ are called angles _between_ the subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$. Let the operator $T\in{\mathcal{B}}({\mathcal{H}})$ be a selfadjoint nonnegative contraction. Using an extension of $T$ to an orthogonal projector [31, §A.2, p. 461], there exist subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ in ${\mathcal{H}}^{2}$ such that $T$ is unitarily equivalent to $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{{\mathcal{F}}}$, where $P_{\mathcal{F}}$ and $P_{\mathcal{G}}$ are the corresponding orthogonal projectors in ${\mathcal{H}}^{2}$. This implies that the spectrum of the product of two orthogonal projectors is as general a set as the spectrum of an arbitrary selfadjoint nonnegative contraction, so the set of angles between subspaces can be a sufficiently general subset of $[0,{\pi}/{2}]$. ###### Definition 2.5. The angles $\hat{\Theta}_{p}({\mathcal{F}},{\mathcal{G}})=\left\\{\theta\in\hat{\Theta}({\mathcal{F}},{\mathcal{G}}):{\cos}^{2}(\theta)\in\Sigma_{p}\left(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}\right)\right\\}$ and $\Theta_{p}({\mathcal{F}},{\mathcal{G}})=\hat{\Theta}_{p}({\mathcal{F}},{\mathcal{G}})\cap\hat{\Theta}_{p}({\mathcal{G}},{\mathcal{F}})$ are called point angles. Angle $\theta\in\hat{\Theta}_{p}({\mathcal{F}},{\mathcal{G}})$ inherits its multiplicity from ${\cos}^{2}(\theta)\in\Sigma_{p}\left(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}\right)$. Multiplicity of angle $\theta\in\Theta_{p}({\mathcal{F}},{\mathcal{G}})$ is the minimum of multiplicities of $\theta\in\hat{\Theta}_{p}({\mathcal{F}},{\mathcal{G}})$ and $\theta\in\hat{\Theta}_{p}({\mathcal{G}},{\mathcal{F}})$. For two vectors $f$ and $g$ in the plane, and their orthogonal counterparts $f^{\perp}$ and $g^{\perp}$ we evidently have that $\theta(f,g)=\theta(f^{\perp},g^{\perp})$ and $\theta(f,g)+\theta(f,g^{\perp})=\pi/2.$ We now describe relationships for angles, corresponding to subspaces ${\mathcal{F}},{\mathcal{G}},{\mathcal{F}}^{\bot},$ and ${\mathcal{G}}^{\bot}.$ We first consider the angles from one subspace to another as they reveal the finer details and provide a foundation for statements on angles between subspaces. ###### Theorem 2.6. For any pair of subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ of ${\mathcal{H}}$: 1. 1. $\hat{\Theta}({\mathcal{F}},{\mathcal{G}}^{\bot})={\pi}/{2}-\hat{\Theta}({\mathcal{F}},{\mathcal{G}})$; 2. 2. $\hat{\Theta}({\mathcal{G}},{\mathcal{F}})\setminus\\{{\pi}/{2}\\}=\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\setminus\\{{\pi}/{2}\\}$; 3. 3. $\hat{\Theta}({\mathcal{F}}^{\bot},{\mathcal{G}})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})={\pi}/{2}-\\{\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})\\}$; 4. 4. $\hat{\Theta}({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})=\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})$; 5. 5. $\hat{\Theta}({\mathcal{G}},{\mathcal{F}}^{\bot})\setminus\\{0\\}={\pi}/{2}-\\{\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\setminus\\{{\pi}/{2}\\}\\}$; 6. 6. $\hat{\Theta}({\mathcal{G}}^{\bot},{\mathcal{F}})\setminus\\{{\pi}/{2}\\}={\pi}/{2}-\\{\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\setminus\\{0\\}\\}$; 7. 7. $\hat{\Theta}({\mathcal{G}}^{\bot},{\mathcal{F}}^{\bot})\setminus\\{0\\}=\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\setminus\\{0\\}$. Table 1: Multiplicities of $0$ and $\pi/2$ angles for different pairs of subspaces $\begin{array}[]{|c|c|c||c|c|c|}\hline\cr{\text{Pair}}&{\theta=0}&{\theta=\pi/2}&{\text{Pair}}&{\theta=0}&{\theta=\pi/2}\\\ \hline\cr{\hat{\Theta}({\mathcal{F}},{\mathcal{G}})}&{\dim\mathfrak{M}_{00}}&{\dim\mathfrak{M}_{01}}&{\hat{\Theta}({\mathcal{G}},{\mathcal{F}})}&{\dim\mathfrak{M}_{00}}&{\dim\mathfrak{M}_{10}}\\\ \hline\cr{\hat{\Theta}({\mathcal{F}},{\mathcal{G}}^{\bot})}&{\dim\mathfrak{M}_{01}}&{\dim\mathfrak{M}_{00}}&{\hat{\Theta}({\mathcal{G}},{\mathcal{F}}^{\bot})}&{\dim\mathfrak{M}_{10}}&{\dim\mathfrak{M}_{00}}\\\ \hline\cr{\hat{\Theta}({\mathcal{F}}^{\bot},{\mathcal{G}})}&{\dim\mathfrak{M}_{10}}&{\dim\mathfrak{M}_{11}}&{\hat{\Theta}({\mathcal{G}}^{\bot},{\mathcal{F}})}&{\dim\mathfrak{M}_{01}}&{\dim\mathfrak{M}_{11}}\\\ \hline\cr{\hat{\Theta}({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})}&{\dim\mathfrak{M}_{11}}&{\dim\mathfrak{M}_{10}}&{\hat{\Theta}({\mathcal{G}}^{\bot},{\mathcal{F}}^{\bot})}&{\dim\mathfrak{M}_{11}}&{\dim\mathfrak{M}_{01}}\\\ \hline\cr\end{array}$ The multiplicities of the point angles $\theta\in(0,{\pi}/{2})$ in $\hat{\Theta}({\mathcal{F}},{\mathcal{G}})$, $\hat{\Theta}({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})$, $\hat{\Theta}({\mathcal{G}},{\mathcal{F}})$ and $\hat{\Theta}({\mathcal{G}}^{\bot},{\mathcal{F}}^{\bot})$ are the same, and are equal to the multiplicities of the point angles ${\pi}/{2}-\theta\in(0,{\pi}/{2})$ in $\hat{\Theta}({\mathcal{F}},{\mathcal{G}}^{\bot})$, $\hat{\Theta}({\mathcal{F}}^{\bot},{\mathcal{G}})$, $\hat{\Theta}({\mathcal{G}},{\mathcal{F}}^{\bot})$ and $\hat{\Theta}({\mathcal{G}}^{\bot},{\mathcal{F}}).$ ###### Proof. (1) Using the equalities $(P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}})|_{\mathcal{F}}=P_{\mathcal{F}}|_{\mathcal{F}}-({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}=I|_{\mathcal{F}}-({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}$ and the spectral mapping theorem for $f(T)=I-T$ we have $\Sigma((P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}})|_{\mathcal{F}})=1-\Sigma(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}})$. Next, using the identity $\mathfrak{N}(T-\lambda I)=\mathfrak{N}((I-T)-(1-\lambda)I)$, we conclude that $\lambda$ is an eigenvalue of $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{{\mathcal{F}}}$ if and only if $1-\lambda$ is an eigenvalue of $(P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}})|_{\mathcal{F}}$, and that their multiplicities are the same. (2) The statement on nonzero angles follows from Corollary 2.3. The part concerning the zero angles follows from the fact that $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathfrak{M}_{00}}=({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathfrak{M}_{00}}=I|_{\mathfrak{M}_{00}}$. (3–7) All other statements can be obtained from the (1–2) by exchanging the subspaces. Table 1 entries are checked directly using the five-parts decomposition. ∎ Theorem 2.7 and Table 2 relate the sets of angles between pairs of subspaces: ###### Theorem 2.7. For any subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ of ${\mathcal{H}}$ the following equalities hold: 1. 1. $\Theta({\mathcal{F}},{\mathcal{G}})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})=\\{{\pi}/{2}-\Theta({\mathcal{F}},{\mathcal{G}}^{\bot})\\}\setminus(\\{0\\}\cup\\{{\pi}/{2}\\});$ 2. 2. $\Theta({\mathcal{F}},{\mathcal{G}})\setminus\\{0\\}=\Theta({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})\setminus\\{0\\};$ 3. 3. $\Theta({\mathcal{F}},{\mathcal{G}}^{\bot})\setminus\\{0\\}=\Theta({\mathcal{F}}^{\bot},{\mathcal{G}})\setminus\\{0\\}.$ Table 2: Multiplicities of $0$ and $\pi/2$ angles between subspaces $\begin{array}[]{|c|c|c|}\hline\cr{\text{Pair}}&{\theta=0}&{\theta=\pi/2}\\\ \hline\cr{\Theta({\mathcal{F}},{\mathcal{G}})}&{\dim\mathfrak{M}_{00}}&{\min\\{\dim\mathfrak{M}_{01},\,\dim\mathfrak{M}_{10}\\}}\\\ \hline\cr{\Theta({\mathcal{F}},{\mathcal{G}}^{\bot})}&{\dim\mathfrak{M}_{01}}&{\min\\{\dim\mathfrak{M}_{00},\,\dim\mathfrak{M}_{11}\\}}\\\ \hline\cr{\Theta({\mathcal{F}}^{\bot},{\mathcal{G}})}&{\dim\mathfrak{M}_{10}}&{\min\\{\dim\mathfrak{M}_{00},\,\dim\mathfrak{M}_{11}\\}}\\\ \hline\cr{\Theta({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})}&{\dim\mathfrak{M}_{11}}&{\min\\{\dim\mathfrak{M}_{01},\,\dim\mathfrak{M}_{10}\\}}\\\ \hline\cr\end{array}$ The multiplicities of the point angles $\theta$ in $\Theta({\mathcal{F}},{\mathcal{G}})$ and $\Theta({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})$ satisfying $0<\theta<{\pi}/{2}$ are the same, and equal to the multiplicities of point angles $0<{\pi}/{2}-\theta<{\pi}/{2}$ in $\Theta({\mathcal{F}},{\mathcal{G}}^{\bot})$ and $\Theta({\mathcal{F}}^{\bot},{\mathcal{G}}).$ ###### Proof. Statement (1) follows from Theorem 2.6 since $\displaystyle\Theta({\mathcal{F}},{\mathcal{G}})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})$ $\displaystyle=$ $\displaystyle\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})$ $\displaystyle=$ $\displaystyle\\{{\pi}/{2}-\hat{\Theta}({\mathcal{F}},{\mathcal{G}}^{\bot})\\}\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})$ $\displaystyle=$ $\displaystyle\\{{\pi}/{2}-\Theta({\mathcal{F}},{\mathcal{G}}^{\bot})\\}\setminus(\\{0\\}\cup\\{{\pi}/{2}\\}),$ Using Theorem 2.6(7) twice: first for ${\mathcal{F}}$ and ${\mathcal{G}}$, next for ${\mathcal{G}}$ and ${\mathcal{F}}$, and then intersecting them gives (2). Interchanging ${\mathcal{G}}$ and ${\mathcal{G}}^{\bot}$ in (2) leads to (3). The statements on multiplicities easily follow from Theorem 2.6 as the entries in Table 2 are just the minima between pairs of the corresponding entries in Table 1. ∎ ###### Remark 2.8. Theorem 2.6(1) allows us to introduce an equivalent sine-based definition: $\hat{\Theta}({\mathcal{F}},{\mathcal{G}})=\\{\theta:\enspace\theta=\arcsin(\mu),\enspace\mu\geq 0,\enspace{\mu}^{2}\in\Sigma((P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}})|_{{\mathcal{F}}})\\}\subseteq[0,{\pi}/{2}].$ ###### Remark 2.9. Theorem 2.6(2) implies $\Theta({\mathcal{F}},{\mathcal{G}})\setminus\\{{\pi}/{2}\\}=\hat{\Theta}({\mathcal{F}},{\mathcal{G}})\setminus\\{{\pi}/{2}\\}=\hat{\Theta}({\mathcal{G}},{\mathcal{F}})\setminus\\{{\pi}/{2}\\}.$ ###### Remark 2.10. We have $\overline{\Theta({\mathcal{F}},{\mathcal{G}})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})}=\Theta(P_{\mathfrak{M}}{\mathcal{F}},P_{\mathfrak{M}}{\mathcal{G}}),$ in other words, the projections $P_{\mathfrak{M}}{\mathcal{F}}=\mathfrak{M}_{{\mathcal{F}}}$ and $P_{\mathfrak{M}}{\mathcal{G}}=\mathfrak{M}_{{\mathcal{G}}}$ of the initial subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ onto their “fifth part” $\mathfrak{M}$ are in generic position within $\mathfrak{M}$, see [14], so the zero and right angles can not belong to the set of point angles $\Theta_{p}(P_{\mathfrak{M}}{\mathcal{F}},P_{\mathfrak{M}}{\mathcal{G}})$, but apart from $0$ and $\pi/2$ the angles $\Theta({\mathcal{F}},{\mathcal{G}})$ and $\Theta(P_{\mathfrak{M}}{\mathcal{F}},P_{\mathfrak{M}}{\mathcal{G}})$ are the same. ###### Remark 2.11. Tables 1 and 2 give the absolute values of the multiplicities of $0$ and $\pi/2$. If we need relative multiplicities, e.g., how many “extra” $0$ and $\pi/2$ values are in $\Theta({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})$ compared to $\Theta({\mathcal{F}},{\mathcal{G}})$, we can easily find the answers from Tables 1 and 2 by subtraction, assuming that we subtract finite numbers, and use identities such as $\dim\mathfrak{M}_{00}-\dim\mathfrak{M}_{11}=\dim{\mathcal{F}}-\dim{\mathcal{G}}^{\bot}$ and $\dim\mathfrak{M}_{01}-\dim\mathfrak{M}_{10}=\dim{\mathcal{F}}-\dim{\mathcal{G}}$. Indeed, for the particular question asked above, we observe that the multiplicity of $\pi/2$ is the same in $\Theta({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})$ and in $\Theta({\mathcal{F}},{\mathcal{G}})$, but the difference in the multiplicities of $0$ in $\Theta({\mathcal{F}}^{\bot},{\mathcal{G}}^{\bot})$ compared to in $\Theta({\mathcal{F}},{\mathcal{G}})$ is equal to $\dim\mathfrak{M}_{11}-\dim\mathfrak{M}_{00}=\dim{\mathcal{G}}^{\bot}-\dim{\mathcal{F}}$, provided that the terms that participate in the subtractions are finite. Some comparisons require both the dimension and the codimension of a subspace to be finite, thus, effectively requiring $\dim{\mathcal{H}}<\infty.$ ### 2.3 Known Quantities as Functions of Angles The gap bounds the perturbation of a closed linear operator by measuring the change in its graph, while the minimum gap between two subspaces determines if the sum of the subspaces is closed. We connect the gap and the minimum gap to the largest and to the nontrivial smallest principal angles. E.g., for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ in generic position, i.e., if $\mathfrak{M}={\mathcal{H}}$, we show that the gap and the minimum gap are the supremum and the infimum, correspondingly, of the sine of the set of angles between ${\mathcal{F}}$ and ${\mathcal{G}}$. The gap (aperture) between subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ defined as, e.g., [19], $\rm gap({\mathcal{F}},{\mathcal{G}})=\left\|P_{\mathcal{F}}-P_{\mathcal{G}}\right\|=\max\left\\{\left\|P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}}\right\|,\left\|P_{\mathcal{G}}P_{{\mathcal{F}}^{\bot}}\right\|\right\\}$ is used to measure the distance between subspaces. We now describe the gap in terms of the angles. ###### Theorem 2.12. $\min\left\\{\min\left\\{\cos^{2}(\hat{\Theta}({\mathcal{F}},{\mathcal{G}}))\right\\},\min\left\\{\cos^{2}(\hat{\Theta}({\mathcal{G}},{\mathcal{F}}))\right\\}\right\\}=1-\rm gap^{2}({\mathcal{F}},{\mathcal{G}}).$ ###### Proof. Let us consider both norms in the definition of the gap separately. Using Theorem 2.6, we have $\displaystyle\|P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}}\|^{2}$ $\displaystyle=$ $\displaystyle\sup_{\begin{subarray}{c}u\in{\mathcal{H}}\\\ \|u\|=1\end{subarray}}\|P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}}u\|^{2}=\sup_{\begin{subarray}{c}u\in{\mathcal{H}}\\\ \|u\|=1\end{subarray}}(P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}}u,P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}}u)$ $\displaystyle=$ $\displaystyle\sup_{\begin{subarray}{c}u\in{\mathcal{H}}\\\ \|u\|=1\end{subarray}}(P_{{\mathcal{G}}^{\bot}}P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}}u,u)=\|(P_{{\mathcal{G}}^{\bot}}P_{\mathcal{F}})|_{{\mathcal{G}}^{\bot}}\|=\max\\{\cos^{2}(\hat{\Theta}({\mathcal{G}}^{\bot},{\mathcal{F}}))\\}$ $\displaystyle=$ $\displaystyle\max\\{\sin^{2}(\hat{\Theta}({\mathcal{G}},{\mathcal{F}}))\\}=1-\min\\{\cos^{2}(\hat{\Theta}({\mathcal{G}},{\mathcal{F}}))\\}.$ Similarly, $\|P_{\mathcal{G}}P_{{\mathcal{F}}^{\bot}}\|^{2}=\max\\{\cos^{2}(\hat{\Theta}({\mathcal{F}}^{\bot},{\mathcal{G}}))\\}=1-\min\\{\cos^{2}(\hat{\Theta}({\mathcal{F}},{\mathcal{G}}))\\}.$ ∎ It follows directly from the above proof and the previous section that ###### Corollary 2.13. If $\rm gap({\mathcal{F}},{\mathcal{G}})<1$ or if the subspaces are in generic position then both terms under the minimum are the same and so $\rm gap({\mathcal{F}},{\mathcal{G}})=\max\\{\sin({\Theta}({\mathcal{F}},{\mathcal{G}}))\\}.$ Let $c({\mathcal{F}},{\mathcal{G}})=\sup\\{|(f,g)|:f\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}}),\|f\|\leq 1,g\in{\mathcal{G}}\ominus({\mathcal{F}}\cap{\mathcal{G}}),\|g\|\leq 1\\},$ as in [8], which is a definition of the cosine of the _angle of Friedrichs_. ###### Theorem 2.14. In terms of the angles, $c({\mathcal{F}},{\mathcal{G}})=\cos\left(\inf\left\\{\Theta({\mathcal{F}},{\mathcal{G}})\setminus\\{0\\}\right\\}\right).$ ###### Proof. Replacing the vectors $f=P_{\mathcal{F}}u$ and $g=P_{\mathcal{G}}v$ in the definition of $c({\mathcal{F}},{\mathcal{G}})$ with the vectors $u$ and $v$ and using the standard equality of induced norms of an operator and the corresponding bilinear form, we get $c({\mathcal{F}},{\mathcal{G}})=\sup_{\begin{subarray}{c}u\in{\mathcal{H}}\ominus\mathfrak{M}_{00}\\\ \|u\|=1\end{subarray}}\sup_{\begin{subarray}{c}v\in{\mathcal{H}}\ominus\mathfrak{M}_{00}\\\ \|v\|=1\end{subarray}}|(u,{P_{\mathcal{F}}P_{\mathcal{G}}}v)|=\|({P_{\mathcal{F}}P_{\mathcal{G}}})|_{{\mathcal{H}}\ominus\mathfrak{M}_{00}}\|.$ Using the five-parts decomposition, ${P_{\mathcal{F}}P_{\mathcal{G}}}=I_{\mathfrak{M}_{00}}\oplus 0_{\mathfrak{M}_{01}}\oplus 0_{\mathfrak{M}_{10}}\oplus 0_{\mathfrak{M}_{11}}\oplus({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathfrak{M}},$ thus “subtracting” the subspace $\mathfrak{M}_{00}$ from the domain of ${P_{\mathcal{F}}P_{\mathcal{G}}}$ excludes $1$ from the point spectrum of ${P_{\mathcal{F}}P_{\mathcal{G}}}$, and, thus, $0$ from the set of point angles from ${\mathcal{F}}$ to ${\mathcal{G}}$ and, by Theorem 2.6(2), from the set of point angles between ${\mathcal{F}}$ and ${\mathcal{G}}$. ∎ Let the _minimum gap_ , see [19, § IV.4], be defined as $\displaystyle\gamma({\mathcal{F}},{\mathcal{G}})=\inf_{\begin{subarray}{c}f\in{\mathcal{F}},\,f\notin{\mathcal{G}}\end{subarray}}\frac{\rm dist(f,\,{\mathcal{G}})}{\rm dist(f,\,{\mathcal{F}}\cap{\mathcal{G}})}.$ ###### Theorem 2.15. In terms of the angles, $\gamma({\mathcal{F}},{\mathcal{G}})=\sin\left(\inf\left\\{\Theta({\mathcal{F}},{\mathcal{G}})\setminus\\{0\\}\right\\}\right).$ ###### Proof. We have $f\in{\mathcal{F}}$ and $f\notin{\mathcal{G}}$, so we can represent $f$ in the form $f=f_{1}+f_{2}$, where $f_{1}\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}})$, $f_{1}\neq 0$ and $f_{2}\in{\mathcal{F}}\cap{\mathcal{G}}$. Then $\displaystyle\gamma({\mathcal{F}},{\mathcal{G}})$ $\displaystyle=$ $\displaystyle\inf_{\begin{subarray}{c}f\in{\mathcal{F}},\,f\notin{\mathcal{G}}\end{subarray}}\frac{\rm dist(f,\,{\mathcal{G}})}{\rm dist(f,\,{\mathcal{F}}\cap{\mathcal{G}})}$ $\displaystyle=$ $\displaystyle\inf_{\begin{subarray}{c}f_{1}\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}}),\,f_{2}\in{\mathcal{F}}\cap{\mathcal{G}}\end{subarray}}\frac{\|f_{1}+f_{2}-P_{\mathcal{G}}f_{1}-P_{\mathcal{G}}f_{2}\|}{\|f_{1}+f_{2}-P_{{\mathcal{F}}\cap{\mathcal{G}}}f_{1}-P_{{\mathcal{F}}\cap{\mathcal{G}}}f_{2}\|}$ $\displaystyle=$ $\displaystyle\inf_{\begin{subarray}{c}f_{1}\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}})\end{subarray}}\frac{\|f_{1}-P_{\mathcal{G}}f_{1}\|}{\|f_{1}-P_{{\mathcal{F}}\cap{\mathcal{G}}}f_{1}\|}$ $\displaystyle=$ $\displaystyle\inf_{\begin{subarray}{c}f\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}})\end{subarray}}\frac{\|f-P_{\mathcal{G}}f\|}{\|f-P_{{\mathcal{F}}\cap{\mathcal{G}}}f\|}.$ But $f\in({\mathcal{F}}\cap{\mathcal{G}})^{\perp}$ and $\|f-P_{{\mathcal{F}}\cap{\mathcal{G}}}f\|=\|f\|$. Since $\|\kappa f-P_{\mathcal{G}}(\kappa f)\|=|\kappa|\|f-P_{\mathcal{G}}f\|$, using the Pythagorean theorem we have $\displaystyle\gamma^{2}({\mathcal{F}},{\mathcal{G}})$ $\displaystyle=$ $\displaystyle\inf_{\begin{subarray}{c}f\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}}),\end{subarray}}\frac{\|f-P_{\mathcal{G}}f\|^{2}}{\|f\|^{2}}$ $\displaystyle=$ $\displaystyle\inf_{\begin{subarray}{c}f\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}}),\,\|f\|=1\end{subarray}}\|f-P_{\mathcal{G}}f\|^{2}$ $\displaystyle=$ $\displaystyle\inf_{\begin{subarray}{c}f\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}}),\,\|f\|=1\end{subarray}}{1-\|P_{\mathcal{G}}f\|^{2}}.$ Using the equality $\|P_{\mathcal{G}}f\|=\sup_{\begin{subarray}{c}g\in{\mathcal{G}},\,\|g\|=1\end{subarray}}|(f,g)|$ we get $\displaystyle\gamma^{2}({\mathcal{F}},{\mathcal{G}})$ $\displaystyle=$ $\displaystyle{1-\sup_{\begin{subarray}{c}f\in{\mathcal{F}}\ominus({\mathcal{F}}\cap{\mathcal{G}}),\,g\in{\mathcal{G}},\,\|f\|=\|g\|=1\end{subarray}}|(f,g)|^{2}}$ $\displaystyle=$ $\displaystyle{1-(c({\mathcal{F}},{\mathcal{G}}))^{2}}$ and finally we use Theorem 2.14. ∎ Let us note that removing $0$ from the set of angles in Theorems 2.14 and 2.15 changes the result after taking the $\inf$, only if $0$ is present as an isolated value in the set of angles, e.g., it has no effect for a pair of subspaces in generic position. ### 2.4 The Spectra of Sum and Difference of Orthogonal Projectors Sums and differences of a pair of orthogonal projectors often appear in applications. Here, we describe their spectra in terms of the angles between the ranges of the projectors, which provides a geometrically intuitive and uniform framework to analyze the sums and differences of orthogonal projectors. First, we connect the spectra of the product and of the difference of two orthogonal projectors. ###### Lemma 2.16. ([30, Theorem 1], [28, Lemma 2.4]). For proper subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ we have $\Sigma({P_{\mathcal{F}}P_{\mathcal{G}}})=\Sigma({P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}})\subseteq[0,1]$ and $\Sigma(P_{\mathcal{G}}-P_{\mathcal{F}})\setminus(\\{-1\\}\cup\\{0\\}\cup\\{1\\})=\\{\pm(1-{\sigma}^{2})^{1/2}:\enspace{\sigma}^{2}\in\Sigma({P_{\mathcal{F}}P_{\mathcal{G}}})\setminus(\\{0\\}\cup\\{1\\})\\}.$ Using Lemma 2.16, we now characterize the spectrum of the differences of two orthogonal projectors in terms of the angles between the corresponding subspaces. ###### Theorem 2.17. The multiplicity of the eigenvalue $1$ in $\Sigma(P_{\mathcal{G}}-P_{\mathcal{F}})$ is equal to $\dim\mathfrak{M}_{10}$, the multiplicity of the eigenvalue $-1$ is equal to $\dim\mathfrak{M}_{01}$, and the multiplicity of the eigenvalue $0$ is equal to $\dim\mathfrak{M}_{00}+\dim\mathfrak{M}_{11}$, where $\mathfrak{M}_{00}$, $\mathfrak{M}_{01}$, $\mathfrak{M}_{10}$ and $\mathfrak{M}_{11}$ are defined in § 2.1. For the rest of the spectrum, we have the following: $\Sigma(P_{\mathcal{F}}-P_{\mathcal{G}})\setminus(\\{-1\\}\cup\\{0\\}\cup\\{1\\})=\pm\sin(\Theta({\mathcal{F}},{\mathcal{G}}))\setminus(\\{-1\\}\cup\\{0\\}\cup\\{1\\}).$ ###### Proof. The last statement follows from Lemma 2.16 and Definition 2.4. To obtain the results concerning the multiplicity of eigenvalues $1$, $-1$ and $0$, it suffices to use the decomposition of these projectors into five parts, given in § 2.1. ∎ In some applications, e.g., in domain decomposition methods, see §6, the distribution of the spectrum of the sum of projectors is important. We directly reformulate [3, Corollary 4.9, p. 86], see also [33, p. 298], in terms of the angles between subspaces: ###### Theorem 2.18. For any nontrivial pair of orthogonal projectors $P_{\mathcal{F}}$ and $P_{\mathcal{G}}$ on ${\mathcal{H}}$ the spectrum of the sum $P_{\mathcal{F}}+P_{\mathcal{G}}$, with the possible exception of the point $0$, lies in the closed interval of the real line $[1-\|{P_{\mathcal{F}}P_{\mathcal{G}}}\|,1+\|{P_{\mathcal{F}}P_{\mathcal{G}}}\|]$, and the following identity holds: $\Sigma(P_{\mathcal{F}}+P_{\mathcal{G}})\setminus(\\{0\\}\cup\\{1\\})=\\{1\pm\cos(\Theta({\mathcal{F}},{\mathcal{G}}))\\}\setminus(\\{0\\}\cup\\{1\\}).$ ## 3 Principal Vectors, Subspaces and Invariant Subspaces In this section, we basically follow Jujunashvili [18, Section 2.8] to introduce principal invariant subspaces for a pair of subspaces by analogy with invariant subspaces of operators. Given the principal invariant subspaces (see Definition 3.1 below) of a pair of subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$, we construct the principal invariant subspaces for pairs ${\mathcal{F}}$ and ${\mathcal{G}}^{\bot}$, ${\mathcal{F}}^{\bot}$ and ${\mathcal{G}}$, ${\mathcal{F}}^{\bot}$ and ${\mathcal{G}}^{\bot}$. We describe relations between orthogonal projectors onto principal invariant subspaces. We show that, in particular cases, principal subspaces and principal vectors can be defined essentially as in the finite dimensional case, and we investigate their properties. Principal vectors, subspaces and principal invariant subspaces reveal the fine structure of the mutual position of a pair of subspaces in a Hilbert space. Except for Theorem 3.3, all other statements can be found in [18, sections 2.6-2.9], which we refer the reader to for detailed proofs and more facts. ### 3.1 Principal Invariant Subspaces Principal invariant subspaces for a pair of subspaces generalize the already known notion of principal vectors, e.g., [35]. We give a geometrically intuitive definition of principal invariant subspaces and connect them with invariant subspaces of the product of the orthogonal projectors. ###### Definition 3.1. A pair of subspaces ${\mathcal{U}}\subseteq{\mathcal{F}}$ and ${\mathcal{V}}\subseteq{\mathcal{G}}$ is called a pair of principal invariant subspaces for the subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$, if $P_{\mathcal{F}}{\mathcal{V}}\subseteq{\mathcal{U}}$ and $P_{\mathcal{G}}{\mathcal{U}}\subseteq{\mathcal{V}}.$ We call the pair ${\mathcal{U}}\subseteq{\mathcal{F}}$ and ${\mathcal{V}}\subseteq{\mathcal{G}}$ nondegenerate if $\overline{P_{\mathcal{F}}{\mathcal{V}}}={\mathcal{U}}\neq\\{0\\}$ and $\overline{P_{\mathcal{G}}{\mathcal{U}}}={\mathcal{V}}\neq\\{0\\}$ and strictly nondegenerate if $P_{\mathcal{F}}{\mathcal{V}}={\mathcal{U}}\neq\\{0\\}$ and $P_{\mathcal{G}}{\mathcal{U}}={\mathcal{V}}\neq\\{0\\}.$ This definition is different from that used in [18, Section 2.8, p. 57], where only what we call here strictly nondegenerate principal invariant subspaces are defined. The following simple theorem deals with enclosed principal invariant subspaces. ###### Theorem 3.2. Let ${\mathcal{U}}\subset{\mathcal{F}}$ and ${\mathcal{V}}\subset{\mathcal{G}}$ be a pair of principal invariant subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$, and $\underline{{\mathcal{U}}}\subset{\mathcal{U}}$, $\underline{{\mathcal{V}}}\subset{\mathcal{V}}$ be a pair of principal invariant subspaces for subspaces ${\mathcal{U}}$ and ${\mathcal{V}}$. Then $\underline{{\mathcal{U}}}$, $\underline{{\mathcal{V}}}$ form a pair of principal invariant subspaces for the subspaces ${\mathcal{F}}$, ${\mathcal{G}}$, and $\Theta(\underline{{\mathcal{U}}},\underline{{\mathcal{V}}})\subseteq\Theta({\mathcal{U}},{\mathcal{V}})\subseteq\Theta({\mathcal{F}},{\mathcal{G}}).$ Definition 3.1 resembles the notion of invariant subspaces. The next theorem completely clarifies this connection for general principal invariant subspaces. ###### Theorem 3.3. The subspaces ${\mathcal{U}}\subseteq{\mathcal{F}}$ and ${\mathcal{V}}\subseteq{\mathcal{G}}$ form a pair of principal invariant subspaces for the subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ if and only if ${\mathcal{U}}\subseteq{\mathcal{F}}$ is an invariant subspace of the operator $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}$ and ${\mathcal{V}}=\overline{P_{\mathcal{G}}{\mathcal{U}}}\oplus{\mathcal{V}}_{0}$, where ${\mathcal{V}}_{0}\subseteq\mathfrak{M}_{10}={\mathcal{G}}\cap{\mathcal{F}}^{\perp}$. ###### Proof. Conditions $P_{\mathcal{F}}{\mathcal{V}}\subseteq{\mathcal{U}}$ and $P_{\mathcal{G}}{\mathcal{U}}\subseteq{\mathcal{V}}$ imply $P_{\mathcal{F}}P_{\mathcal{G}}{\mathcal{U}}\subseteq P_{\mathcal{F}}{\mathcal{V}}\subseteq{\mathcal{U}}.$ Let us consider $v_{0}\in{\mathcal{V}}\ominus\overline{P_{\mathcal{G}}{\mathcal{U}}}={\mathcal{V}}\cap{\mathcal{U}}^{\perp}$ (the latter equality follows from $0=(v_{0},P_{\mathcal{G}}u)=(v_{0},u),\,\forall u\in{\mathcal{U}}$). We have $P_{\mathcal{F}}v_{0}\in{\mathcal{U}}^{\perp}$ since ${\mathcal{U}}\subseteq{\mathcal{F}}$, but our assumption $P_{\mathcal{F}}{\mathcal{V}}\subseteq{\mathcal{U}}$ assures that $P_{\mathcal{F}}v_{0}\in{\mathcal{U}}$, so $P_{\mathcal{F}}v_{0}=0$, which means that ${\mathcal{V}}_{0}\subseteq\mathfrak{M}_{10},$ as required. To prove the converse, let $P_{\mathcal{F}}P_{\mathcal{G}}{\mathcal{U}}\subseteq{\mathcal{U}}$ and ${\mathcal{V}}=\overline{P_{\mathcal{G}}{\mathcal{U}}}\oplus{\mathcal{V}}_{0}$. Then $P_{\mathcal{F}}{\mathcal{V}}=P_{\mathcal{F}}\overline{P_{\mathcal{G}}{\mathcal{U}}}\subseteq{\mathcal{U}}$ since ${\mathcal{U}}$ is closed. $P_{\mathcal{G}}{\mathcal{U}}\subseteq{\mathcal{V}}$ follows from the formula for ${\mathcal{V}}.$ ∎ If the subspace $\mathfrak{M}_{10}$ is trivial, the principal invariant subspace ${\mathcal{V}}$ that corresponds to ${\mathcal{U}}$ is clearly unique. The corresponding statement for ${\mathcal{U}}$, given ${\mathcal{V}}$, we get from Theorem 3.3 by swapping ${\mathcal{F}}$ and ${\mathcal{G}}.$ We now completely characterize (strictly) nondegenerate principal invariant subspaces using the corresponding angles. ###### Theorem 3.4. The pair ${\mathcal{U}}\subseteq{\mathcal{F}}$ and $\enspace{\mathcal{V}}\subseteq{\mathcal{G}}$ of principal invariant subspaces for the subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ is nondegenerate if and only if both operators $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}$ and $({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathcal{V}}$ are invertible, i.e., ${\pi}/{2}\notin\hat{\Theta}_{p}({\mathcal{U}},{\mathcal{V}})\cup\hat{\Theta}_{p}({\mathcal{V}},{\mathcal{U}}),$ and strictly nondegenerate if and only if each of the inverses is bounded, i.e., ${\pi}/{2}\notin\hat{\Theta}({\mathcal{U}},{\mathcal{V}})\cup\hat{\Theta}({\mathcal{V}},{\mathcal{U}}),$ or equivalently in terms of the gap, $\rm gap({\mathcal{U}},{\mathcal{V}})=\|P_{\mathcal{U}}-P_{\mathcal{V}}\|<1$. ###### Proof. We prove the claim for the operator $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}},$ and the claim for the other operator follows by symmetry. Definition 3.1 uses $\overline{P_{\mathcal{F}}{\mathcal{V}}}={\mathcal{U}}\neq\\{0\\}$ for nondegenerate principal invariant subspaces. At the same time, Theorem 3.3 holds, so ${\mathcal{V}}=\overline{P_{\mathcal{G}}{\mathcal{U}}}\oplus{\mathcal{V}}_{0}$, where ${\mathcal{V}}_{0}\subseteq\mathfrak{M}_{10}={\mathcal{G}}\cap{\mathcal{F}}^{\perp}$. So ${\mathcal{U}}=\overline{P_{\mathcal{F}}{\mathcal{V}}}=\overline{{P_{\mathcal{F}}P_{\mathcal{G}}}{\mathcal{U}}}.$ Also by Theorem 3.3, ${\mathcal{U}}\subseteq{\mathcal{F}}$ is an invariant subspace of the operator $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}$, so ${\mathcal{U}}=\overline{{P_{\mathcal{F}}P_{\mathcal{G}}}{\mathcal{U}}}=\overline{({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}{\mathcal{U}}}$. Since $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}$ is Hermitian, its null-space is trivial (as the orthogonal in ${\mathcal{U}}$ complement to its range which is dense in ${\mathcal{U}}$), i.e., the operator $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}$ is one-to-one and thus invertible. For strictly nondegenerate principal invariant subspaces, ${({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}{\mathcal{U}}}={\mathcal{U}}$, so the operator $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}$ by the open mapping theorem has a continuous and thus bounded inverse. Conversely, by Theorem 3.3 ${\mathcal{U}}\subseteq{\mathcal{F}}$ is an invariant subspace of the operator $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}$, so the restriction $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}$ is correctly defined. The operator $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}$ is invertible by assumption, thus its null-space is trivial, and so its range is dense: ${\mathcal{U}}=\overline{({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}}{\mathcal{U}}}=\overline{{P_{\mathcal{F}}P_{\mathcal{G}}}{\mathcal{U}}}$. By Theorem 3.3, ${\mathcal{V}}=\overline{P_{\mathcal{G}}{\mathcal{U}}}\oplus{\mathcal{V}}_{0}$, therefore $\overline{P_{\mathcal{F}}{\mathcal{V}}}=\overline{{P_{\mathcal{F}}P_{\mathcal{G}}}{\mathcal{U}}}={\mathcal{U}}.$ The other equality, $\overline{P_{\mathcal{G}}{\mathcal{U}}}={\mathcal{V}}\neq\\{0\\}$, of Definition 3.1 for nondegenerate principal invariant subspaces, is proved similarly using the assumption that $({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathcal{G}}$ is invertible. If, in addition, each of the inverses is bounded, the corresponding ranges are closed, ${\mathcal{U}}={P_{\mathcal{F}}P_{\mathcal{G}}}{\mathcal{U}}$ and ${\mathcal{V}}={P_{\mathcal{F}}P_{\mathcal{G}}}{\mathcal{V}}$ and we obtain $P_{\mathcal{F}}{\mathcal{V}}={\mathcal{U}}\neq\\{0\\}$ and $P_{\mathcal{G}}{\mathcal{U}}={\mathcal{V}}\neq\\{0\\}$ as is needed in Definition 3.1 for strictly nondegenerate principal invariant subspaces. The equivalent formulations of conditions of the theorem in terms of the angles and the gap follow directly from Definitions 2.4 and 2.5 and Theorem 2.12. ∎ Theorem 2.2 introduces the unitary operator $W$ that gives the unitary equivalence of ${P_{\mathcal{F}}P_{\mathcal{G}}P_{\mathcal{F}}}$ and ${P_{\mathcal{G}}P_{\mathcal{F}}P_{\mathcal{G}}}$ and, if $\rm gap({\mathcal{F}},{\mathcal{G}})<1$, the unitary equivalence by (2.1) of $P_{\mathcal{F}}$ and $P_{\mathcal{G}}$. Now we state that the same $W$ makes orthogonal projectors $P_{\mathcal{U}}$ and $P_{\mathcal{V}}$ unitarily equivalent for strictly nondegenerate principal invariant subspaces ${\mathcal{U}}\subset{\mathcal{F}}$ and ${\mathcal{V}}\subset{\mathcal{G}}$, and we obtain expressions for the orthogonal projectors. ###### Theorem 3.5. Let ${\mathcal{U}}\subseteq{\mathcal{F}}$ and ${\mathcal{V}}\subseteq{\mathcal{G}}$ be a pair of strictly nondegenerate principal invariant subspaces for the subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$, and $W$ be defined as in Theorem 2.2. Then ${\mathcal{V}}=W{\mathcal{U}}$ and ${\mathcal{U}}=W^{*}{\mathcal{V}},$ while the orthoprojectors satisfy $P_{\mathcal{V}}=WP_{\mathcal{U}}W^{*}=P_{\mathcal{G}}P_{\mathcal{U}}(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}})^{-1}P_{\mathcal{U}}P_{\mathcal{G}}$ and $P_{\mathcal{U}}=W^{*}P_{\mathcal{V}}W=P_{\mathcal{F}}P_{\mathcal{V}}(({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathcal{V}})^{-1}P_{\mathcal{V}}P_{\mathcal{F}}.$ The proof of Theorem 3.5 is straightforward and can be found in [18, §2.8]. Jujunashvili [18, §2.9] also develops the theory of principal invariant subspaces, using the spectral decompositions, e.g., below is [18, Theorem 2.108]: ###### Theorem 3.6. Let $\\{E_{1}\\}$ and $\\{E_{2}\\}$ be spectral measures of the operators $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}$ and $({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathcal{G}}$, respectively. Let $\Theta\subseteq{\Theta}({\mathcal{U}},{\mathcal{V}})\setminus\\{{\pi}/{2}\\}$ be a closed Borel set, and define $P_{{\mathcal{U}}(\Theta)}=\int_{\cos(\Theta)}dE_{1}(\lambda)$ and $P_{{\mathcal{V}}(\Theta)}=\int_{{\cos(\Theta)}}dE_{2}(\lambda)$. Then ${\mathcal{U}}(\Theta)\subset{\mathcal{F}}$ and ${\mathcal{V}}(\Theta)\subset{\mathcal{G}}$ is a pair of strictly nondegenerate principal invariant subspaces and $P_{{\mathcal{V}}({\Theta})}=P_{\mathcal{G}}\left\\{\int_{{\cos(\Theta)}}\frac{1}{\lambda}dE_{1}(\lambda)\right\\}P_{\mathcal{G}},$ and $\Theta=\hat{\Theta}({\mathcal{U}}(\Theta),{\mathcal{V}}(\Theta))=\hat{\Theta}({\mathcal{V}}(\Theta),{\mathcal{U}}(\Theta)).$ ###### Proof. We have $\int_{{\cos(\Theta)}}\frac{1}{\lambda}dE_{1}(\lambda)=\left((P_{\mathcal{F}}P_{\mathcal{U}})|_{{\mathcal{U}}}\right)^{-1}=P_{\mathcal{U}}\left((P_{\mathcal{F}}P_{\mathcal{G}})|_{{\mathcal{U}}}\right)^{-1}P_{\mathcal{U}}$ (where we denote ${\mathcal{U}}={\mathcal{U}}(\Theta)$), which we plug into the expression for the orthogonal projector $P_{\mathcal{V}}$ of Theorem 3.5. ∎ For a pair of principal invariant subspaces ${\mathcal{U}}\subset{\mathcal{F}}$ and ${\mathcal{V}}\subset{\mathcal{G}}$, using Theorems 3.3 and 3.4 we define the corresponding principal invariant subspaces in ${\mathcal{F}}^{\bot}$ and ${\mathcal{G}}^{\bot}$ as ${\mathcal{U}}_{\bot}=P_{{\mathcal{F}}^{\bot}}{\mathcal{V}}$ and ${\mathcal{V}}_{\bot}=P_{{\mathcal{G}}^{\bot}}{\mathcal{U}}$, and describe their properties in the next theorem. ###### Theorem 3.7. Let ${\mathcal{U}}$ and ${\mathcal{V}}$ be a pair of principal invariant subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ and $0,{\pi}/{2}\notin\hat{\Theta}({\mathcal{U}},{\mathcal{V}})\cup\hat{\Theta}({\mathcal{V}},{\mathcal{U}}).$ Then ${\mathcal{U}}_{\bot}=P_{{\mathcal{F}}^{\bot}}{\mathcal{V}}$ and ${\mathcal{V}}_{\bot}=P_{{\mathcal{G}}^{\bot}}{\mathcal{U}}$ are closed and * 1. ${\mathcal{U}}$ and ${\mathcal{V}}$ is a pair of strictly nondegenerate principal invariant subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$; * 2. ${\mathcal{U}}_{\bot}$ and ${\mathcal{V}}$ is a pair of strictly nondegenerate principal invariant subspaces for subspaces ${\mathcal{F}}^{\bot}$ and ${\mathcal{G}}$ and $P_{{\mathcal{U}}_{\bot}}$ and $P_{{\mathcal{V}}}$ are unitarily equivalent; * 3. ${\mathcal{U}}$ and ${\mathcal{V}}_{\bot}$ is a pair of strictly nondegenerate principal invariant subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}^{\bot}$ and $P_{{\mathcal{U}}}$ and $P_{{\mathcal{V}}_{\bot}}$ are unitarily equivalent; * 4. ${\mathcal{U}}_{\bot}$ and ${\mathcal{V}}_{\bot}$ is a pair of strictly nondegenerate principal invariant subspaces for subspaces ${\mathcal{F}}^{\bot}$ and ${\mathcal{G}}^{\bot}$ and $P_{{\mathcal{U}}_{\bot}}$ and $P_{{\mathcal{V}}_{\bot}}$ are unitarily equivalent. ###### Proof. The statements follow directly from Theorems 3.3 and 3.4 applied to the corresponding pairs of subspaces. The closedness of ${\mathcal{U}}_{\bot}$ and ${\mathcal{V}}_{\bot}$ can be alternatively derived from Theorem 2.14 and [8, Theorem 22]. ∎ ### 3.2 Principal Subspaces and Principal Vectors For a pair of principal invariant subspaces ${\mathcal{U}}\subset{\mathcal{F}}$ and ${\mathcal{V}}\subset{\mathcal{G}}$, if the spectrum $\Sigma(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{U}})$ consists of one number, which belongs to $(0,1]$ and which we denote by $\cos^{2}(\theta)$, we can use Theorem 3.5 to define a pair of principal subspaces corresponding to an angle $\theta$: ###### Definition 3.8. Let $\theta\in\Theta({\mathcal{F}},{\mathcal{G}})\setminus\\{{\pi}/{2}\\}$. Nontrivial subspaces ${\mathcal{U}}\subseteq{\mathcal{F}}$ and ${\mathcal{V}}\subseteq{\mathcal{G}}$ define a pair of principal subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ corresponding to the angle $\theta$ if $(P_{\mathcal{F}}P_{\mathcal{V}})|_{\mathcal{F}}={\cos}^{2}(\theta)P_{\mathcal{U}}$ and $(P_{\mathcal{G}}P_{\mathcal{U}})|_{\mathcal{G}}={\cos}^{2}(\theta)P_{\mathcal{V}}.$ Normalized vectors $u=u(\theta)\in{\mathcal{F}}$ and $v=v(\theta)\in{\mathcal{G}}$ form a pair of principal vectors for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ corresponding to the angle $\theta$ if $P_{\mathcal{F}}v=\cos(\theta)u$ and $P_{\mathcal{G}}u=\cos(\theta)v.$ We exclude $\theta=\pi/2$ in Definition 3.8 so that principal subspaces belong to the class of strictly nondegenerate principal invariant subspaces. We describe the main properties of principal subspaces and principal vectors that can be checked directly (for details, see [18]). The first property characterizes principal subspaces as eigenspaces of the products of the corresponding projectors. ###### Theorem 3.9. Subspaces ${\mathcal{U}}\subset{\mathcal{F}}$ and ${\mathcal{V}}\subset{\mathcal{G}}$ form a pair of principal subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ corresponding to the angle $\theta\in\Theta({\mathcal{F}},{\mathcal{G}})\setminus\\{{\pi}/{2}\\}$ if and only if $\theta\in\Theta_{p}({\mathcal{F}},{\mathcal{G}})\setminus\\{{\pi}/{2}\\}$ and ${\mathcal{U}}$ and ${\mathcal{V}}$ are the eigenspaces of the operators $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}$ and $({P_{\mathcal{G}}P_{\mathcal{F}}})|_{\mathcal{G}}$, respectively, corresponding to the eigenvalue ${\cos}^{2}(\theta)$. In such a case, $\Theta({\mathcal{U}},{\mathcal{V}})=\Theta_{p}({\mathcal{U}},{\mathcal{V}})=\\{\theta\\}$. All pairs of principal vectors $u$ and $v$ of subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ corresponding to the angle $\theta$ generate the largest principal subspaces ${\mathcal{U}}$ and ${\mathcal{V}}$ corresponding to the angle $\theta$. ###### Theorem 3.10. Let ${\mathcal{U}}(\theta),\,{\mathcal{U}}(\phi)\subset{\mathcal{F}}$, and ${\mathcal{V}}(\theta),\,{\mathcal{V}}(\phi)\subset{\mathcal{G}}$ be the principal subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$ corresponding to the angles $\theta,\phi\in\Theta_{p}({\mathcal{F}},{\mathcal{G}})\setminus\\{{\pi}/{2}\\}$. Then $P_{{\mathcal{U}}(\theta)}P_{{\mathcal{U}}(\phi)}=P_{{\mathcal{U}}(\theta)\cap\,{\mathcal{U}}(\phi)}$; $P_{{\mathcal{V}}(\theta)}P_{{\mathcal{V}}(\phi)}=P_{{\mathcal{V}}(\theta)\cap{\mathcal{V}}(\phi)}$ ; $P_{{\mathcal{U}}(\theta)}$ and $P_{{\mathcal{V}}(\phi)}$ are mutually orthogonal if $\theta\neq\phi$ (if $\theta=\phi$ we can choose ${\mathcal{V}}(\theta)$ such that $P_{{\mathcal{U}}(\theta)}P_{{\mathcal{V}}(\theta)}=P_{{\mathcal{U}}(\theta)}P_{\mathcal{G}}$); for given ${\mathcal{U}}(\theta)$ we can choose ${\mathcal{V}}(\theta)$ such that $P_{{\mathcal{V}}(\theta)}P_{{\mathcal{U}}(\theta)}=P_{{\mathcal{V}}(\theta)}P_{\mathcal{F}}$. ###### Corollary 3.11. [of Theorem 3.7] Let ${\mathcal{U}}$ and ${\mathcal{V}}$ be the principal subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$, corresponding to the angle $\theta\in\Theta_{p}({\mathcal{F}},{\mathcal{G}})\setminus(\\{0\\}\cup\\{{\pi}/{2}\\})$. Then ${\mathcal{U}}_{\bot}=P_{{\mathcal{F}}^{\bot}}{\mathcal{V}}$ and ${\mathcal{V}}_{\bot}=P_{{\mathcal{G}}^{\bot}}{\mathcal{U}}$ are closed and * 1. ${\mathcal{U}}_{\bot}$, ${\mathcal{V}}$ are the principal subspaces for subspaces ${\mathcal{F}}^{\bot}$ and ${\mathcal{G}}$, corresponding to the angle ${\pi}/{2}-\theta$; * 2. ${\mathcal{U}}$, ${\mathcal{V}}_{\bot}$ are the principal subspaces for subspaces ${\mathcal{F}}$ and ${\mathcal{G}}^{\bot}$, corresponding to the angle ${\pi}/{2}-\theta$; * 3. ${\mathcal{U}}_{\bot}$, ${\mathcal{V}}_{\bot}$ are the principal subspaces for subspaces ${\mathcal{F}}^{\bot}$ and ${\mathcal{G}}^{\bot}$, corresponding to the angle $\theta$. Let $u$ and $v$ form a pair of principal vectors for the subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$, corresponding to the angle $\theta$. Then $u_{\bot}=(v-\cos(\theta)u)/\sin(\theta)$ and $v_{\bot}=(u-\cos(\theta)v)/\sin(\theta)$ together with $u$ and $v$ describe the pairs of principal vectors. ## 4 Bounding the Changes in the Angles Here we prove bounds on the change in the (squared cosines of the) angles from one subspace to another where the subspaces change. These bounds allow one to estimate the sensitivity of the angles with respect to the changes in the subspaces. For the finite dimensional case, such bounds are known, e.g., [23, 24]. To measure the distance between two bounded real sets $S_{1}$ and $S_{2}$ we use the Hausdorff distance, e.g., [19], $\rm dist(S_{1},S_{2})=\max\\{\sup_{u\in S_{1}}\rm dist(u,S_{2}),\sup_{v\in S_{2}}\rm dist(v,S_{1})\\},$ where $\rm dist(u,S)=\inf_{v\in S}|u-v|$ is the distance from the point $u$ to the set $S$. The following theorem estimates the proximity of the set of squares of cosines of $\hat{\Theta}({\mathcal{F}},{\mathcal{G}})$ and the set of squares of cosines of $\hat{\Theta}({\mathcal{F}},\tilde{{\mathcal{G}}})$, where ${\mathcal{F}}$, ${\mathcal{G}}$ and $\tilde{{\mathcal{G}}}$ are nontrivial subspaces of ${\mathcal{H}}$. ###### Theorem 4.1. $\rm dist(\cos^{2}(\hat{\Theta}({\mathcal{F}},{\mathcal{G}})),\cos^{2}(\hat{\Theta}({\mathcal{F}},\tilde{{\mathcal{G}}})))\leq\rm gap({\mathcal{G}},\tilde{{\mathcal{G}}}).$ ###### Proof. $\cos^{2}(\hat{\Theta}({\mathcal{F}},{\mathcal{G}}))=\Sigma(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{{\mathcal{F}}})$ and $\cos^{2}(\hat{\Theta}({\mathcal{F}},\tilde{{\mathcal{G}}}))=\Sigma((P_{\mathcal{F}}P_{\tilde{\mathcal{G}}})|_{{\mathcal{F}}})$ by Definition 2.4. Both operators $({P_{\mathcal{F}}P_{\mathcal{G}}})|_{\mathcal{F}}$ and $(P_{\mathcal{F}}P_{\tilde{\mathcal{G}}})|_{{\mathcal{F}}}$ are selfadjoint. By [19, Theorem 4.10, p. 291], $\rm dist(\Sigma(({P_{\mathcal{F}}P_{\mathcal{G}}})|_{{\mathcal{F}}}),\Sigma((P_{\mathcal{F}}P_{\tilde{\mathcal{G}}})|_{{\mathcal{F}}}))\leq\|({P_{\mathcal{F}}P_{\mathcal{G}}})|_{{\mathcal{F}}}-(P_{\mathcal{F}}P_{\tilde{\mathcal{G}}})|_{{\mathcal{F}}}\|.$ Then, $\|({P_{\mathcal{F}}P_{\mathcal{G}}})|_{{\mathcal{F}}}-(P_{\mathcal{F}}P_{\tilde{\mathcal{G}}})|_{{\mathcal{F}}}\|\leq\|P_{\mathcal{F}}\|\|P_{\mathcal{G}}-P_{\tilde{\mathcal{G}}}\|\|P_{\mathcal{F}}\|\leq\rm gap({\mathcal{G}},\tilde{{\mathcal{G}}}).$ ∎ The same result holds also if the first subspace, ${\mathcal{F}}$, is changed in $\hat{\Theta}({\mathcal{F}},{\mathcal{G}}))$: ###### Theorem 4.2. $\rm dist(\cos^{2}(\hat{\Theta}({\mathcal{F}},{\mathcal{G}})),\cos^{2}(\hat{\Theta}(\tilde{{\mathcal{F}}},{\mathcal{G}})))\leq\rm gap({\mathcal{F}},\tilde{{\mathcal{F}}}).$ ###### Proof. The statement of the theorem immediately follows from Theorem 5.2, which is independently proved in the next section, where one takes $A=P_{\mathcal{G}}$. ∎ We conjecture that similar generalizations to the case of infinite dimensional subspaces can be made for bounds involving changes in the sines and cosines (without squares) of the angles extending known bounds [23, 24] for the finite dimensional case. ## 5 Changes in the Ritz Values and Rayleigh-Ritz error bounds Here we estimate how Ritz values of a selfadjoint operator change with the change of a vector, and then we extend this result to estimate the change of Ritz values with the change of a (infinite dimensional) trial subspace, using the gap between subspaces, $\rm gap({\mathcal{F}},{\mathcal{G}})=\|P_{{\mathcal{F}}}-P_{{\mathcal{G}}}\|.$ Such results are natural extensions of the results of the previous section that bound the change in the squared cosines or sines of the angles, since in the particular case where the selfadjoint operator is an orthogonal projector its Ritz values are exactly the squared cosines of the angles from the trial subspace of the Rayleigh-Ritz method to the range of the orthogonal projector. In addition, we prove a spectrum error bound that characterizes the change in the Ritz values for an invariant subspace, and naturally involves the gap squared; see [27, 1, 26] for similar finite dimensional results. Let $A\in{\mathcal{B}}({\mathcal{H}})$ be a selfadjoint operator. Denote by $\lambda(f)=(f,Af)/(f,f)$ the Rayleigh quotient of an operator $A$ at a vector $f\neq 0$. In the following lemma, we estimate changes in the Rayleigh quotient with the change in a vector. This estimate has been previously proven only for real finite dimensional spaces [24]. Here, we give a new proof that works both for real and complex spaces. ###### Lemma 5.1. Let $A\in{\mathcal{B}}({\mathcal{H}})$ be a selfadjoint operator on a Hilbert space ${\mathcal{H}}$ and $f,g\in{\mathcal{H}}$ with $f,g\neq 0$. Then $|\lambda(f)-\lambda(g)|\leq(\max\\{\Sigma(A)\\}-\min\\{\Sigma(A)\\})\sin(\theta(f,g)).$ (5.1) ###### Proof. We use the so-called “mini-dimensional” analysis, e.g., [20, 21]. Let $S=\rm span\\{f,g\\}\subset{\mathcal{H}}$ be a two dimensional subspace (if $f$ and $g$ are linearly dependent then the Rayleigh quotients are the same and the assertion is trivial). Denote $\tilde{A}=(P_{S}A)|_{S}$ and two eigenvalues of $\tilde{A}$ by $\lambda_{1}\leq\lambda_{2}$. By well known properties of the Rayleigh-Ritz method, we have $\lambda(f),\lambda(g)\in[\lambda_{1},\lambda_{2}]\subseteq[\max\\{\Sigma(A)\\},\min\\{\Sigma(A)\\}]$. In the nontrivial case $\lambda(f)\neq\lambda(g)$, we then have the strong inequality $\lambda_{1}<\lambda_{2}$. In this proof, we extend the notation of the Rayleigh quotient of an operator $A$ at a vector $f$ to $\lambda(f;\,A)=(f,Af)/(f,f)$ to explicitly include $A$. It is easy to see that $\lambda(f;\,A)=\lambda(f;\,\tilde{A})$ and that the same holds for vector $g$. Then, since $[\lambda_{1},\lambda_{2}]\subseteq[\max\\{\Sigma(A)\\},\min\\{\Sigma(A)\\}]$ the statement of the lemma would follow from the 2D estimate $|\lambda(f;\,\tilde{A})-\lambda(g;\,\tilde{A})|\leq(\lambda_{2}-\lambda_{1})\sin(\theta(f,g))$ that we now have to prove. The latter estimate is clearly invariant with respect to a shift and scaling of $\tilde{A}$. Let us use the transformation $\bar{A}=(\tilde{A}-\lambda_{1}I)/(\lambda_{2}-\lambda_{1})$ then the estimate we need to prove turns into $|\lambda(f;\,\bar{A})-\lambda(g;\,\bar{A})|\leq\sin(\theta(f,g)),$ but the operator $\bar{A}$ has two eigenvalues, zero and one, and thus is an orthoprojector on some one dimensional subspace $\rm span\\{h\\}\subset S$. Finally, $\lambda(f;\,\bar{A})=(f,P_{h}f)/(f,f)=\cos^{2}(\theta(h,f))$ and, similarly, $\lambda(g;\,\bar{A})=(g,P_{h}g)/(g,g)=\cos^{2}(\theta(h,g))$. But $|\cos^{2}(\theta(h,f))-\cos^{2}(\theta(h,g))|=|\,\|P_{h}P_{f}P_{h}||-\|P_{h}P_{g}P_{h}||\,|\leq\|P_{f}-P_{g}\|=\sin(\theta(f,g)).$ ∎ In the Rayleigh-Ritz method for a selfadjoint operator $A\in{\mathcal{B}}({\mathcal{H}})$ on a trial subspace ${\mathcal{F}}$ the spectrum $\Sigma((P_{{\mathcal{F}}}A)|_{{\mathcal{F}}})$ is called the set of Ritz values, corresponding to $A$ and ${\mathcal{F}}$. The next result of this section is an estimate of a change in the Ritz values, where one trial subspace, ${\mathcal{F}}$, is replaced with another, ${\mathcal{G}}$. For finite dimensional subspaces such a result is obtained in [24], where the maximal distance between pairs of individually ordered Ritz values is used to measure the change in the Ritz values. Here, the trial subspaces may be infinite dimensional, so the Ritz values may form rather general sets on the real interval $[\min\\{\Sigma(A)\\},\max\\{\Sigma(A)\\}]$ and we are limited to the use of the Hausdorff distance between the sets, which does not take into account the ordering and multiplicities. ###### Theorem 5.2. Let $A\in{\mathcal{B}}({\mathcal{H}})$ be a selfadjoint operator and ${\mathcal{F}}$ and ${\mathcal{G}}$ be nontrivial subspaces of ${\mathcal{H}}$. Then a bound for the Hausdorff distance between the Ritz values of $A$, with respect to the trial subspaces ${\mathcal{F}}$ and ${\mathcal{G}}$, is given by the following inequality $\rm dist(\Sigma((P_{{\mathcal{F}}}A)|_{{\mathcal{F}}}),\,\Sigma((P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}))\leq(\max\\{\Sigma(A)\\}-\min\\{\Sigma(A)\\})\,\rm gap({\mathcal{F}},{\mathcal{G}}).$ ###### Proof. If $\rm gap({\mathcal{F}},{\mathcal{G}})=1$ then the assertion holds since the both spectra are subsets of $[\min\\{\Sigma(A)\\},\max\\{\Sigma(A)\\}]$. Consequently we can assume without loss of generality that $\rm gap({\mathcal{F}},{\mathcal{G}})<1$. Then we have ${\mathcal{G}}=W{\mathcal{F}}$ with $W$ defined by (2.1). Operators $(P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}$ and $\left(W^{*}(P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}$ are unitarily equivalent, since $W$ is an isometry on ${\mathcal{F}}$, therefore, their spectra are the same. Operators $(P_{{\mathcal{F}}}A)|_{{\mathcal{F}}}$ and $\left(W^{*}(P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}$ are selfadjoint on the space ${\mathcal{F}}$ and using [19, Theorem 4.10, p. 291] we get $\displaystyle\rm dist(\Sigma((P_{{\mathcal{F}}}A)|_{{\mathcal{F}}}),\,\Sigma((P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}))$ $\displaystyle=$ $\displaystyle\rm dist(\Sigma((P_{{\mathcal{F}}}A)|_{{\mathcal{F}}}),\,\Sigma(\left(W^{*}(P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}))$ (5.2) $\displaystyle\leq$ $\displaystyle\|\left(P_{{\mathcal{F}}}A-W^{*}(P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}\|.$ Then $\displaystyle\|\left(P_{{\mathcal{F}}}A-W^{*}(P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}\|$ $\displaystyle=$ $\displaystyle\sup_{\begin{subarray}{c}\|f\|=1,\,f\in{\mathcal{F}}\end{subarray}}|((P_{{\mathcal{F}}}A-W^{*}(P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}W)f,f)|$ $\displaystyle=$ $\displaystyle\sup_{\begin{subarray}{c}\|f\|=1,\,f\in{\mathcal{F}}\end{subarray}}|(Af,f)-(AWf,Wf)|.$ We have $|(f,Af)-(Wf,AWf)|\leq(\max\\{\Sigma(A)\\}-\min\\{\Sigma(A)\\})\sqrt{1-|(f,Wf)|^{2}},$ $\forall f\in{\mathcal{F}}$, $\|f\|=1$ by Lemma 5.1. We need to estimate $|(f,Wf)|$ from below. From the polar decomposition $P_{{\mathcal{G}}}P_{{\mathcal{F}}}=W\sqrt{P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}}$, we derive the equalities $(f,Wf)=(P_{{\mathcal{G}}}P_{{\mathcal{F}}}f,Wf)=(W^{*}P_{{\mathcal{G}}}P_{{\mathcal{F}}}f,f)=(\sqrt{P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}}f,f),$ where we have $\sqrt{P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}}|_{{\mathcal{F}}}=\sqrt{(P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}})|_{{\mathcal{F}}}}=\sqrt{(P_{{\mathcal{F}}}P_{{\mathcal{G}}})|_{{\mathcal{F}}}},$ since ${\mathcal{F}}$ is an invariant subspace of the operator $P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}$. Thus, $(f,Wf)=(\sqrt{(P_{{\mathcal{F}}}P_{{\mathcal{G}}})|_{{\mathcal{F}}}}f,f)\geq\min\\{\cos(\hat{\Theta}({\mathcal{F}},{\mathcal{G}}))\\}$ by Definition 2.4. Finally, by assumption, $\rm gap({\mathcal{F}},{\mathcal{G}})<1$, thus Corollary 2.13 gives $\min\\{\cos^{2}(\hat{\Theta}({\mathcal{F}},{\mathcal{G}}))\\}=1-\rm gap^{2}({\mathcal{F}},{\mathcal{G}}).$ ∎ Finally, we assume that ${\mathcal{F}}$ is $A$-invariant, which implies that the set of the values $\Sigma((P_{{\mathcal{F}}}A)|_{{\mathcal{F}}})$ is a subset, namely $\Sigma(A|_{{\mathcal{F}}})$, of the spectrum of $A.$ The change in the Ritz values, bounded in Theorem 5.2, can now be interpreted as a spectrum error in the Rayleigh-Ritz method. The result of Theorem 5.2 here is improved since the new bound involves the gap squared as in [1, 26]. ###### Theorem 5.3. Under the assumptions of Theorem 5.2 let in addition ${\mathcal{F}}$ be an $A$-invariant subspace of ${\mathcal{H}}$ corresponding to the top (or bottom) part of the spectrum of $A$. Then $\rm dist(\Sigma(A|_{{\mathcal{F}}}),\,\Sigma((P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}))\leq(\max\\{\Sigma(A)\\}-\min\\{\Sigma(A)\\})\,\rm gap^{2}({\mathcal{F}},{\mathcal{G}}).$ ###### Proof. As the subspace ${\mathcal{F}}$ is $A$-invariant and $A$ is selfadjoint, the subspace ${\mathcal{F}}^{\bot}$ is also $A$-invariant, so $A=P_{{\mathcal{F}}}AP_{{\mathcal{F}}}+P_{{\mathcal{F}}^{\bot}}AP_{{\mathcal{F}}^{\bot}}$ and, with a slight abuse of the notation, $A=A|_{{\mathcal{F}}}+A|_{{\mathcal{F}}^{\bot}},$ corresponding to the decomposition ${\mathcal{H}}={\mathcal{F}}\oplus{\mathcal{F}}^{\bot}$, thus $\Sigma(A)=\Sigma(A|_{{\mathcal{F}}})\cup\Sigma(A|_{{\mathcal{F}}^{\bot}})$. We assume that ${\mathcal{F}}$ corresponds to the top part of the spectrum of $A$—the bottom part case can be treated by replacing $A$ with $-A$. Under this assumption, we have $\max\\{\Sigma(A|_{{\mathcal{F}}^{\bot}})\\}\leq\min\\{\Sigma(A|_{{\mathcal{F}}})\\}$. Let us also notice that the inequality we want to prove is unaltered by replacing $A$ with $A-\alpha I$ where $\alpha$ is an arbitrary real constant. Later in the proof we need $A|_{{\mathcal{F}}}$ to be nonnegative. We set $\alpha=\min\\{\Sigma(A|_{{\mathcal{F}}})\\}$ and substitute $A$ with $A-\alpha I$, so now $\max\\{\Sigma(A|_{{\mathcal{F}}^{\bot}})\\}\leq 0=\min\\{\Sigma(A|_{{\mathcal{F}}})\\}$, thus $\left\|A|_{{\mathcal{F}}}\right\|=\max\\{\Sigma(A|_{{\mathcal{F}}})\\}=\max\\{\Sigma(A)\\},\text{ and }\left\|A|_{{\mathcal{F}}^{\bot}}\right\|=-\min\\{\Sigma(A|_{{\mathcal{F}}^{\bot}})\\}=-\min\\{\Sigma(A)\\}.$ The constant in the bound we are proving then takes the following form: $\max\\{\Sigma(A)\\}-\min\\{\Sigma(A)\\}=\left\|A|_{{\mathcal{F}}}\right\|+\left\|A|_{{\mathcal{F}}^{\bot}}\right\|.$ (5.3) As in the proof of Theorem 5.2, if $\rm gap({\mathcal{F}},{\mathcal{G}})=1$ then the assertion holds since the both spectra are subsets of $[\min\\{\Sigma(A)\\},\max\\{\Sigma(A)\\}]$. Consequently we can assume without loss of generality that $\rm gap({\mathcal{F}},{\mathcal{G}})<1$. Then we have ${\mathcal{G}}=W{\mathcal{F}}$ with $W$ defined by (2.1). Operators $\left(W^{*}(P_{{\mathcal{G}}}P_{\mathcal{F}}AP_{\mathcal{F}})|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}$ and $(P_{{\mathcal{G}}}P_{\mathcal{F}}AP_{\mathcal{F}})|_{{\mathcal{G}}}$ are unitarily equivalent, since $W$ is an isometry on ${\mathcal{F}}$, thus their spectra are the same. Now, instead of (5.2), we use the triangle inequality for the Hausdorff distance: $\rm dist(\Sigma(A|_{{\mathcal{F}}}),\,\Sigma((P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}))\\\ \leq\rm dist(\Sigma((A|_{{\mathcal{F}}}),\,\Sigma(\left(W^{*}(P_{{\mathcal{G}}}P_{\mathcal{F}}AP_{\mathcal{F}})|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}))\\\ +\rm dist(\Sigma((P_{{\mathcal{G}}}P_{\mathcal{F}}AP_{\mathcal{F}})|_{{\mathcal{G}}})),\,\Sigma((P_{{\mathcal{G}}}A)|_{{\mathcal{G}}})).$ (5.4) The operator $\sqrt{P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}}|_{{\mathcal{F}}}=\sqrt{(P_{{\mathcal{F}}}P_{{\mathcal{G}}})|_{{\mathcal{F}}}}$ is selfadjoint and its smallest point of the spectrum is $\min\\{\cos(\hat{\Theta}({\mathcal{F}},{\mathcal{G}}))\\}$ by Definition 2.4, which is positive by Theorem 2.12 with $\rm gap({\mathcal{F}},{\mathcal{G}})<1.$ The operator $\sqrt{P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}}|_{{\mathcal{F}}}$ is invertible, so from the polar decomposition $P_{{\mathcal{G}}}P_{{\mathcal{F}}}=W\sqrt{P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}}$, which gives $P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}=P_{{\mathcal{F}}}P_{{\mathcal{G}}}W\sqrt{P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}}$, we obtain by applying the inverse on the right that $(P_{{\mathcal{F}}}P_{{\mathcal{G}}}W)|_{{\mathcal{F}}}=\sqrt{P_{{\mathcal{F}}}P_{{\mathcal{G}}}P_{{\mathcal{F}}}}|_{{\mathcal{F}}}=(W^{*}P_{{\mathcal{G}}}P_{\mathcal{F}})|_{{\mathcal{F}}}.$ Thus, $\displaystyle\left(W^{*}(P_{{\mathcal{G}}}P_{\mathcal{F}}AP_{\mathcal{F}})|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}$ $\displaystyle=$ $\displaystyle\left(\sqrt{P_{\mathcal{F}}P_{{\mathcal{G}}}P_{\mathcal{F}}}A\sqrt{P_{\mathcal{F}}P_{{\mathcal{G}}}P_{\mathcal{F}}}\right)|_{{\mathcal{F}}}$ $\displaystyle=$ $\displaystyle\sqrt{P_{\mathcal{F}}P_{{\mathcal{G}}}P_{\mathcal{F}}}|_{{\mathcal{F}}}\sqrt{A|_{{\mathcal{F}}}}\sqrt{A|_{{\mathcal{F}}}}\sqrt{P_{\mathcal{F}}P_{{\mathcal{G}}}P_{\mathcal{F}}}|_{{\mathcal{F}}}$ where the operator $A|_{{\mathcal{F}}}$ is already made nonnegative by applying the shift and the substitution. The spectrum of the product of two bounded operators, one of which is bijective, does not depend on the order of the multipliers, since both products are similar to each other. One of our operators, $\sqrt{P_{\mathcal{F}}P_{{\mathcal{G}}}P_{\mathcal{F}}}|_{{\mathcal{F}}}$, in the product is bijective, so $\Sigma(\left(W^{*}(P_{{\mathcal{G}}}P_{\mathcal{F}}AP_{\mathcal{F}})|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}})=\Sigma\left(\sqrt{A|_{\mathcal{F}}}(P_{\mathcal{F}}P_{{\mathcal{G}}})|_{\mathcal{F}}\sqrt{A|_{\mathcal{F}}}|_{{\mathcal{F}}}\right).$ Then the first term in the triangle inequality (5.4) for the Hausdorff distance is estimated using [19, Theorem 4.10, p. 291]: $\displaystyle\rm dist\big{(}\Sigma\left(A|_{{\mathcal{F}}}\right),\,$ $\displaystyle\Sigma\left(\left(W^{*}\left(P_{{\mathcal{G}}}P_{\mathcal{F}}AP_{\mathcal{F}}\right)|_{{\mathcal{G}}}W\right)|_{{\mathcal{F}}}\right)\big{)}$ $\displaystyle=$ $\displaystyle\rm dist\left(\Sigma\left(A|_{{\mathcal{F}}}\right),\,\Sigma\left(\sqrt{A|_{\mathcal{F}}}(P_{\mathcal{F}}P_{{\mathcal{G}}})|_{\mathcal{F}}\sqrt{A|_{\mathcal{F}}}\right)\right)$ $\displaystyle\leq$ $\displaystyle\left\|A|_{{\mathcal{F}}}-\sqrt{A|_{\mathcal{F}}}(P_{\mathcal{F}}P_{{\mathcal{G}}})|_{\mathcal{F}}\sqrt{A|_{\mathcal{F}}}\right\|$ $\displaystyle=$ $\displaystyle\left\|\sqrt{A|_{\mathcal{F}}}(P_{\mathcal{F}}-P_{\mathcal{F}}P_{{\mathcal{G}}})|_{\mathcal{F}}\sqrt{A|_{\mathcal{F}}}\right\|$ $\displaystyle\leq$ $\displaystyle\left\|A|_{\mathcal{F}}\right\|\left\|(P_{\mathcal{F}}P_{{\mathcal{G}}^{\bot}})|_{\mathcal{F}}\right\|=\left\|A|_{\mathcal{F}}\right\|\left\|P{{}_{\mathcal{F}}}P_{{\mathcal{G}}^{\bot}}\right\|^{2}.$ To estimate the second term in (5.4), we apply again [19, Theorem 4.10, p. 291]: $\rm dist\left(\Sigma\left.\left((P_{{\mathcal{G}}}P_{{\mathcal{F}}}AP_{{\mathcal{F}}}\right)\right|_{{\mathcal{G}}})),\,\Sigma((P_{{\mathcal{G}}}A)|_{{\mathcal{G}}})\right)\leq\|(P_{{\mathcal{G}}}P_{{\mathcal{F}}}AP_{{\mathcal{F}}})|_{{\mathcal{G}}}-(P_{{\mathcal{G}}}A)|_{{\mathcal{G}}}\|\\\ =\|(P_{{\mathcal{G}}}P_{{\mathcal{F}}^{\bot}}AP_{{\mathcal{F}}^{\bot}})|_{{\mathcal{G}}}\|=\|P_{{\mathcal{G}}}P_{{\mathcal{F}}^{\bot}}A|_{{\mathcal{F}}^{\bot}}P_{{\mathcal{F}}^{\bot}}P_{{\mathcal{G}}}\|\leq\|A|_{{\mathcal{F}}^{\bot}}\|\|P_{{\mathcal{G}}}P_{{\mathcal{F}}^{\bot}}\|^{2},$ where $A=P_{\mathcal{F}}AP_{\mathcal{F}}+P_{{\mathcal{F}}^{\bot}}AP_{{\mathcal{F}}^{\bot}}$. Plugging in bounds for both terms in (5.4) gives $\rm dist\left(\Sigma(A|_{{\mathcal{F}}}),\,\Sigma((P_{{\mathcal{G}}}A)|_{{\mathcal{G}}})\right)\leq\left\|A|_{\mathcal{F}}\right\|\left\|P_{{\mathcal{F}}}P_{{\mathcal{G}}^{\bot}}\right\|^{2}+\left\|A|_{{\mathcal{F}}^{\bot}}\right\|\left\|P_{{\mathcal{G}}}P_{{\mathcal{F}}^{\bot}}\right\|^{2}.$ Assumption $\rm gap({\mathcal{F}},{\mathcal{G}})<1$ implies that $\left\|P_{{\mathcal{F}}}P_{{\mathcal{G}}^{\bot}}\right\|=\|P_{{\mathcal{G}}}P_{{\mathcal{F}}^{\bot}}\|=\rm gap({\mathcal{F}},{\mathcal{G}}),$ e.g., see [19, §I.8, Theorem 6.34] and cf. Corollary 2.13. Thus we obtain $\rm dist\left(\Sigma((P_{{\mathcal{F}}}A)|_{{\mathcal{F}}}),\,\Sigma((P_{{\mathcal{G}}}A)|_{{\mathcal{G}}})\right)\leq\left(\left\|A|_{{\mathcal{F}}}\right\|+\|A|_{{\mathcal{F}}^{\bot}}\|\right)\rm gap^{2}({\mathcal{F}},{\mathcal{G}}).$ Taking into account (5.3) completes the proof. ∎ We conjecture that our assumption on the invariant subspace representing a specific part of the spectrum of $A$ is irrelevant, i.e., the statement of Theorem 5.3 holds without it as well, cf. Argentati et al. [1], Knyazev and Argentati [26]. ## 6 The ultimate acceleration of the alternating projectors method Every selfadjoint nonnegative non-expansion $A,\,0\leq A\leq I$ in a Hilbert space ${\mathcal{H}}$ can be extended to an orthogonal projector in the space ${\mathcal{H}}\times{\mathcal{H}}$, e.g., [14, 31], and, thus, can be implicitly written as (strictly speaking is unitarily equivalent to) a product of two orthogonal projectors ${P_{\mathcal{F}}P_{\mathcal{G}}}$ restricted to a subspace ${\mathcal{F}}\subset{\mathcal{H}}\times{\mathcal{H}}$. Any iterative method that involves as a main step a multiplication of a vector by $A$ can thus be called “an alternating projectors” method. In the classical alternating projectors method, it is assumed that the projectors are given explicitly and that the iterating procedure is trivially $e^{(i+1)}={P_{\mathcal{F}}P_{\mathcal{G}}}e^{(i)},\,e^{(0)}\in{\mathcal{F}}.$ (6.1) If $\left\|\left.({P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{\mathcal{F}}\right\|<1$ then the sequence of vectors $e^{(i)}$ evidently converges to zero. Such a situation is typical when $e^{(i)}$ represents an error of an iterative method, e.g., in a multiplicative DDM, and formula (6.1) describes the error propagation as in our DDM example below. If the subspace $\mathfrak{M}_{00}={\mathcal{F}}\cap{\mathcal{G}}$ is nontrivial and $\left\|\left.({P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{{\mathcal{F}}\ominus\mathfrak{M}_{00}}\right\|<1$ then the sequence of vectors $e^{(i)}$ converges to the orthogonal projection $e$ of $e^{(0)}$ onto $\mathfrak{M}_{00}$. The latter is called a von Neumann- Halperin ([34, 15]) method in [2] of alternating projectors for determining the best approximation to $e^{(0)}$ in $\mathfrak{M}_{00}$. We note that, despite the non-symmetric appearance of the error propagation operator ${P_{\mathcal{F}}P_{\mathcal{G}}}$ in (6.1), it can be equivalently replaced with the selfadjoint operator $\left.({P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{\mathcal{F}}$ since $e^{(0)}\in{\mathcal{F}}$ and thus all $e^{(i)}\in{\mathcal{F}}$. Several attempts to estimate and accelerate the convergence of iterations (6.1) are made, e.g., [9, 2, 39]. Here, we use a different approach, cf., e.g., [38, 4], to suggest the ultimate acceleration of the alternating projectors method. First, we notice that the limit vector $e\in\mathfrak{M}_{00}$ is a nontrivial solution of the following homogeneous equation $\left.(I-{P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{\mathcal{F}}e=0,\quad e\in{\mathcal{F}}.$ (6.2) Second, we observe that the linear operator is selfadjoint and nonnegative in the equation above, therefore, a conjugate gradient (CG) method can be used to calculate approximations to the solution $e$ in the null-space. The standard CG algorithm for linear systems $Ax=b$ can be formulated as follows, see, e.g., [17]: Initialization: set $\gamma=1$ and compute the initial residual $r=b-Ax$; Loop until convergence: $\gamma_{old}=\gamma,\,\gamma=(r,r)$; on the first iteration: $p=r$; otherwise: $\beta={\gamma}/{\gamma_{old}}$ (standard) or $\beta={(r-r_{old},r)}/{(r_{old},r_{old})}$ (the latter is recommended if an approximate application of $A$ is used) $p=r+\beta p,$ $r=Ap,$ $\alpha=\gamma/(r,p),$ $x=x+\alpha p,$ $r=r-\alpha r.$ End loop It can be applied directly to the homogeneous equation $Ae=0$ with $A=A^{\ast}\geq 0$ by setting $b=0$. We need $A=\left.(I-{P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{\mathcal{F}}$ for equation (6.2). Finally, we note that CG acceleration can evidently be applied to the symmetrized alternating projectors method with more than two projectors. The traditional theory of the CG method for non-homogeneous equations extends trivially to the computation of the null-space of a selfadjoint nonnegative operator $A$ and gives the following convergence rate estimate: $(e^{(k)},Ae^{(k)})\leq\min_{\deg p_{k}=k,\,p_{k}(0)=1}\sup_{\lambda\in\Sigma(A)\setminus\\{0\\}}|p_{k}(\lambda)|^{2}\quad(e^{(0)},Ae^{(0)}).$ (6.3) For equation (6.2), $A=\left.(I-{P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{\mathcal{F}}$ and thus $(e^{(k)},Ae^{(k)})=\|P_{{\mathcal{G}}^{\bot}}e^{(k)}\|^{2}$ and by Definition 2.4 we have $\Sigma(A)=1-\cos^{2}\hat{\Theta}({\mathcal{F}},{\mathcal{G}})$. Estimate (6.3) shows convergence if and only if zero is an isolated point of the spectrum of $A$, or, in terms of the angles, if and only if zero is an isolated point, or not present, in the set of angles $\hat{\Theta}({\mathcal{F}},{\mathcal{G}})$, which is the same as the condition for convergence of the original alternating projectors method (6.1), stated above. Method (6.1) can be equivalently reformulated as a simple Richardson iteration $e^{(k)}=(I-A)^{k}e^{(0)},\,e^{(0)}\in{\mathcal{F}},\text{ where }A=\left.(I-{P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{\mathcal{F}},$ and thus falls into the same class of polynomial methods as does the CG method. It is well known that the CG method provides the smallest value of the energy (semi-) norm of the error, in our case of $\|P_{{\mathcal{G}}^{\bot}}e^{(k)}\|$, where $e^{(k)}\in{\mathcal{F}}$, which gives us an opportunity to call it the “ultimate acceleration” of the alternating projectors method. A possible alternative to equation (6.2) is $(P_{{\mathcal{F}}^{\bot}}+P_{{\mathcal{G}}^{\bot}})e=0,$ (6.4) so we can take $A=P_{{\mathcal{F}}^{\bot}}+P_{{\mathcal{G}}^{\bot}}$ in the CG method for equation (6.4) and then $\Sigma(A)$ is given by Theorem 2.18. Equation (6.4) appears in the so-called additive DDM method, e.g., [32]. A discussion of (6.4) can be found in [18, §7.1, p. 127]. Estimate (6.3) guarantees the finite convergence of the CG method if the spectrum of $A$ consists of a finite number of points. At the same time, the convergence of the Richardson method can be slow in such a case, so that the CG acceleration is particularly noticeable. In the remainder of the section, we present a simple domain decomposition example for the one dimensional diffusion equation. Consider the following one dimensional diffusion equation $\int_{0}^{1}u^{\prime}v^{\prime}dx=\int_{0}^{1}fv^{\prime}dx,$ $\forall v\in H_{0}^{1}([0,1])$ with the solution $u\in H_{0}^{1}([0,1])$, where $H_{0}^{1}([0,1])$ is the usual Sobolev space of real-valued functions with the Lebesgue integrable squares of the first generalized derivatives and with zero values at the end points of the interval $[0,1]$. We use the bilinear form $\int_{0}^{1}u^{\prime}v^{\prime}dx$ as a scalar product on $H_{0}^{1}([0,1])$. We consider DDM with an overlap, i.e., we split $[0,1]=[0,\alpha]\cup[\beta,1]$, with $0<\beta<\alpha<1$ so that $[\beta,\alpha]$ is an overlap. We directly define orthogonal complements: ${\mathcal{F}}^{\bot}=\\{u\in H_{0}^{1}([0,1]):u(x)=0,x\in[\alpha,1]\\}\text{ and }{\mathcal{G}}^{\bot}=\\{v\in H_{0}^{1}([0,1]):v(x)=0,x\in[0,\beta]\\}$ of subspaces ${\mathcal{F}}\subset H_{0}^{1}([0,1])$ and ${\mathcal{G}}\subset H_{0}^{1}([0,1])$. Evidently, ${\mathcal{H}}={\mathcal{F}}^{\bot}+{\mathcal{G}}^{\bot}$, where the sum is not direct due to the overlap. It can be checked easily that the subspace ${\mathcal{F}}$ consists of functions, which are linear on the interval $[0,\alpha]$ and the subspace ${\mathcal{G}}$ consists of functions, which are linear on the interval $[\beta,1]$. Because of the overlap $[\beta,\alpha]$, the intersection $\mathfrak{M}_{00}={\mathcal{F}}\cap{\mathcal{G}}$ is trivial and the only solution of (6.2) and (6.4) is $e=0$. We now completely characterize all angles between ${\mathcal{F}}$ and ${\mathcal{G}}$. Let $f\in{\mathcal{F}}$ be linear on intervals $[0,\alpha]$ and $[\alpha,1]$. Similarly, let $g\in{\mathcal{G}}$ be linear on intervals $[0,\beta]$ and $[\beta,1]$. It is easy to see, cf. [18, §7.2], that all functions in the subspace ${\mathcal{F}}\ominus\rm span\\{f\\}$ vanish outside of the interval $[\alpha,1]$, while all functions in the subspace ${\mathcal{G}}\ominus\rm span\\{g\\}$ vanish outside of the interval $[0,\beta]$. Therefore, the subspaces ${\mathcal{F}}\ominus\rm span\\{f\\}$ and ${\mathcal{G}}\ominus\rm span\\{g\\}$ are orthogonal, since $\beta<\alpha$. We summarize these results in terms of the principal subspaces: $\hat{\Theta}({\mathcal{F}},{\mathcal{G}})=\Theta({\mathcal{F}},{\mathcal{G}})=\Theta_{p}({\mathcal{F}},{\mathcal{G}})=\theta(f,g)\cup{\pi}/{2},$ where $\cos^{2}\theta(f,g)=(\beta(1-\alpha))/(\alpha(1-\beta)),$ (the latter equality can be derived by elementary calculations, see [18, Theorem 7.2, p. 131]); $\rm span\\{f\\}$ and $\rm span\\{g\\}$ is one pair of principal subspaces and ${\mathcal{F}}\ominus\rm span\\{f\\}$ and ${\mathcal{G}}\ominus\rm span\\{g\\}$ is the other, corresponding to the angle $\pi/2.$ In multiplicative Schwarz DDM with an overlap for two subdomains, the error propagation of a simple iteration is given by (6.1) and the convergence rate is determined by the quantity $\left\|\left.({P_{\mathcal{F}}P_{\mathcal{G}}})\right|_{\mathcal{F}}\right\|=\cos^{2}\theta(f,g)=(\beta(1-\alpha))/(\alpha(1-\beta))<1,$ which approaches one when the overlap $\alpha-\beta$ becomes small. At the same time, however, the CG method described, e.g., in [4, 38], converges at most in two iterations, since the spectrum of $A$ in (6.2) consists of only two eigenvalues, $1-\cos^{2}\theta(f,g)=\sin^{2}\theta(f,g)$ and $1.$ In the additive DDM the error is determined by (6.4) and the spectrum of $A$, the sum of two orthoprojectors, by analogy with Theorem 2.18 consists of four eigenvalues, $1-\cos\theta(f,g)=2\sin^{2}(\theta(f,g)/2),\,1,\,1+\cos\theta(f,g)=2\cos^{2}(\theta(f,g)/2),\text{ and }2,$ therefore the CG method converges at most in four iterations. Similar results for a finite difference discretization of the 1D diffusion equation can be found in [12]. ## Acknowledgements We thank Ilya Lashuk for contributing to the proofs of Theorems 2.2 and 5.2. ## References * Argentati et al. [2006] M. E. Argentati, A. V. Knyazev, C. C. Paige, and I. Panayotov. Bounds on changes in Ritz values for a perturbed invariant subspace of a Hermitian matrix. _SIAM J. Matr. Anal. Appl._ , 30(2):548–559, 2008. MR2421459 (2009e:15021) * Bauschke et al. [2003] H. H. Bauschke, F. Deutsch, H. Hundal, and S.-H. Park. Accelerating the convergence of the method of alternating projections. _Trans. Amer. Math. Soc._ , 355(9):3433–3461 (electronic), 2003. MR1990157 (2004d:41062) * Bjørstad and Mandel [1991] P. E. Bjørstad and J. Mandel. On the spectra of sums of orthogonal projections with applications to parallel computing. _BIT_ , 31(1):76–88, 1991. MR1097483 (91m:65157) * Bjørstad and Widlund [1989] P. E. Bjørstad and O. B. Widlund. To overlap or not to overlap: a note on a domain decomposition method for elliptic problems. _SIAM J. Sci. Statist. Comput._ , 10(5):1053–1061, 1989. MR1009556 (90g:65139) * Bottcher and Spitkovsky [2010] A. Bottcher, I.M. Spitkovsky. A gentle guide to the basics of two projections theory. Linear Algebra and its Applications, 432(6) 1412-1459, 2010. DOI: 10.1016/j.laa.2009.11.002. * Davis [1958] C. Davis. Separation of two linear subspaces. _Acta Sci. Math. Szeged_ , 19:172–187, 1958. MR0098980 (20:5425) * Davis and Kahan [1970] C. Davis and W. M. Kahan. The rotation of eigenvectors by a perturbation. III. _SIAM J. Numer. Anal._ , 7:1–46, 1970. MR0264450 (41:9044) * Deutsch [1995] F. Deutsch. The angle between subspaces of a Hilbert space. In _Approximation theory, wavelets and applications_ , pp. 107–130. Kluwer Acad. Publ., Dordrecht, 1995. MR1340886 (96e:46027) * Deutsch [2001] F. Deutsch. Accelerating the convergence of the method of alternating projections via a line search: a brief survey. In _Inherently parallel algorithms in feasibility and optimization and their applications_ , v. 8 of _Stud. Comput. Math._ , pp. 203–217. North-Holland, Amsterdam, 2001. MR1853223 (2002g:90101) * Deutsch [2001] F. Deutsch. Best approximation in inner product spaces. CMS Books in Mathematics/Ouvrages de Mathématiques de la SMC, 7, Springer-Verlag, New York, 2001, 338 pp. ISBN = 0-387-95156-3. MR1823556 (2002c:41001). * Dixmier [1948] J. Dixmier Position relative de deux variétés linéaires fermées dans un espace de Hilbert. Revue Scientifique, 86: 387-399, 1948. MR0029095 (10,546e) * Efstathiou and Gander [2003] E. Efstathiou and M. J. Gander. Why restricted additive Schwarz converges faster than additive Schwarz. _BIT_ , 43:945–959, 2003. MR2058877 (2005c:65092) * Friedrichs [1937] K. Friedrichs On certain inequalities and characteristic value problems for analytic functions and for functions of two variables Trans. Amer. Math. Soc. 41 (1937), pp. 321–364. * Halmos [1969] P. R. Halmos. Two subspaces. _Trans. Amer. Math. Soc._ , 144:381–389, 1969. MR0251519 (40:4746) * Halperin [1962] I. Halperin. The product of projection operators. _Acta Sci. Math. (Szeged)_ , 23:96–99, 1962. MR0141978 (25:5373) * Hannan [1961/1962] E. J. Hannan. The general theory of canonical correlation and its relation to functional analysis. _J. Austral. Math. Soc._ , 2:229–242, 1961/1962. MR0166869 (29:4142) * Hestenes [1980] M. R. Hestenes. _Conjugate direction methods in optimization_. Springer-Verlag, New York, 1980. MR0561510 (81i:65052) * Jujunashvili [2005] A. Jujunashvili. _Angles Between Infinite Dimensional Subspaces_. PhD thesis, University of Colorado Denver, 2005. * Kato [1995] T. Kato. _Perturbation theory for linear operators_. Springer-Verlag, Berlin, 1995. MR1335452 (96a:47025) * Knyazev [1986] A. V. Knyazev. _Computation of eigenvalues and eigenvectors for mesh problems: algorithms and error estimates_. Dept. Numerical Math. USSR Academy of Sciences, Moscow, 1986. (In Russian). MR1111245 (92d:65005) * Knyazev [1987] A. V. Knyazev. Convergence rate estimates for iterative methods for mesh symmetric eigenvalue problem. _Soviet J. Numerical Analysis and Math. Modelling_ , 2(5):371–396, 1987. MR0915330 (88i:65057) * Knyazev [2007] A. V. Knyazev, Observations on degenerate saddle point problems. _Comput. Methods Appl. Mech.Engrg._. 196 (37-40), 3742–3749, 1997. MR2339999 (2008g:65164) * Knyazev and Argentati [2002] A. V. Knyazev and M. E. Argentati. Principal angles between subspaces in an A-based scalar product: Algorithms and perturbation estimates. _SIAM J. Sci. Computing_ , 23(6):2009–2041, 2002. MR1923723 (2003h:65060) * Knyazev and Argentati [2006a] A. V. Knyazev and M. E. Argentati. On proximity of Rayleigh quotients for different vectors and Ritz values generated by different trial subspaces. _Linear Algebra and its Applications_ , 415:82–95, 2006a. MR2214747 (2006m:65077) * Knyazev and Argentati [2006b] A. V. Knyazev and M. E. Argentati. Majorization for changes in angles between subspaces, Ritz values, and graph Laplacian spectra. _SIAM J. Matr. Anal. Appl._ , 29(1):15–32, 2006/07b. MR2288011 (2008f:15069) * Knyazev and Argentati [2007] A. V. Knyazev and M. E. Argentati. Rayleigh-Ritz majorization error bounds with applications to FEM. _SIAM J. Matr. Anal. Appl._ , 31(3):1521–1537, 2010 MR2587790 * Knyazev and Osborn [2006] A. V. Knyazev and J. Osborn. New A Priori FEM Error Estimates for Eigenvalues. _SIAM J. Numer. Anal._ , 43(6), 2647–2667, 2006. MR2206452 (2006k:65300) * Koliha and Rakočević [2004] J. J. Koliha and V. Rakočević. On the norm of idempotents in $C^{*}$-algebras. _Rocky Mountain J. Math._ , 34(2):685–697, 2004\. MR2072801 (2005c:46077) * Krein et al. [1948] M.G. Krein, M.A. Krasnoselski, and D.P. Milman. On the defect numbers of operators in Banach spaces and on some geometric questions, Trudy Inst. Mat. Akad. Nauk Ukrain. SSR 11 (1948), pp. 97–112 (in Russian). * Omladič [1987] M. Omladič. Spectra of the difference and product of projections. _Proc. Amer. Math. Soc._ , 99(2):317–318, 1987\. MR0870792 (88a:47001) * Riesz and Sz.-Nagy [1990] F. Riesz and B. Sz.-Nagy. _Functional analysis_. Dover Publications Inc., New York, 1990. MR1068530 (91g:00002) * Toselli and Widlund [2005] A. Toselli and O. Widlund. _Domain decomposition methods—algorithms and theory_. Springer-Verlag, Berlin, 2005. MR2104179 (2005g:65006) * Vidav [1977] I. Vidav. The norm of the sum of two projections. _Proc. Amer. Math. Soc._ , 65(2):297–298, 1977\. MR0442703 (56:1084) * von Neumann [1950] J. von Neumann. _Functional Operators. II. The Geometry of Orthogonal Spaces._ Princeton Univ. Press, Princeton, N. J., 1950. MR0034514 (11,599e) * Wedin [1983] P. A. Wedin. On angles between subspaces of a finite-dimensional inner product space. In _Matrix Pencils_ , pp. 263–285. Springer-Verlag, Berlin, 1983. * Weinberger [1974] H. F. Weinberger. _Variational methods for eigenvalue approximation_. SIAM, Philadelphia, Pa., 1974\. MR0400004 (53:3842) * Weinstein and Stenger [1972] A. Weinstein and W. Stenger. _Methods of intermediate problems for eigenvalues_. Academic Press, New York, 1972. MR0477971 (57:17469) * Widlund [1989] O. B. Widlund. Optimal iterative refinement methods. In _Domain decomposition methods_ , pp. 114–125. SIAM, Philadelphia, PA, 1989. MR0992008 (90m:65205) * Xu and Zikatanov [2002] J. Xu and L. Zikatanov. The method of alternating projections and the method of subspace corrections in Hilbert space. _J. Amer. Math. Soc._ , 15(3):573–597, 2002. MR1896233 (2003f:65095)

arxiv-papers

2024-09-04T02:48:52.857078

0705.1329

# Third Order Newton’s Method for Zernike Polynomial Zeros Richard J. Mathar http://www.strw.leidenuniv.nl/ mathar mathar@strw.leidenuniv.nl Leiden Observatory, Leiden University, P.O. Box 9513, 2300 RA Leiden, The Netherlands ###### Abstract. The Zernike radial polynomials are a system of orthogonal polynomials over the unit interval with weight $x$. They are used as basis functions in optics to expand fields over the cross section of circular pupils. To calculate the roots of Zernike polynomials, we optimize the generic iterative numerical Newton’s Method that iterates on zeros of functions with third order convergence. The technique is based on rewriting the polynomials as Gauss hypergeometric functions, reduction of second order derivatives to first order derivatives, and evaluation of some ratios of derivatives by terminating continued fractions. A PARI program and a short table of zeros complete up to polynomials of 20th order are included. ###### Key words and phrases: Zernike Polynomial, Jacobi Polynomial, circular pupil, root finding, Newton Method ###### 2000 Mathematics Subject Classification: Primary 26C10, 33C45; Secondary 78M25 Supported by the NWO VICI grant 639.043.201 to A. Quirrenbach, “Optical Interferometry: A new Method for Studies of Extrasolar Planets.” ## 1\. Classical Orthogonal Polynomials: Hofsommer’s Newton Method The generic third order Newton’s Method to compute roots $f(x)=0$ numerically improves solutions $x_{i}\rightarrow x_{i+1}=x_{i}+\Delta x$ iteratively, starting from initial guesses, via computation of corrections (1) $\Delta x=-\frac{f(x)}{f^{\prime}(x)}/\left(1-\frac{f(x)}{2f^{\prime}(x)}\,\frac{f^{\prime\prime}(x)}{f^{\prime}(x)}\right)$ where $f(x)$, $f^{\prime}(x)$ and $f^{\prime\prime}(x)$ are the function and its first and second derivatives at the current best approximation $x_{i}$ [8, 11, 14]. For some classes of orthogonal polynomials, $f^{\prime\prime}/f^{\prime}$ can be derived from $f/f^{\prime}$ [12, 24], which means the update can be done to third order at essentially no additional numerical expense. If we divide the differential equation of the classical orthogonal polynomials, for example as tabulated in [1, 22.6][17], (2) $h_{2}(x)f^{\prime\prime}+h_{1}(x)f^{\prime}+h_{0}(x)f=0,$ through $f^{\prime}$, (1) turns into (3) $\Delta x=-\frac{f(x)}{f^{\prime}(x)}/\left[1+\frac{1}{2h_{2}(x)}\frac{f(x)}{f^{\prime}(x)}\left(h_{0}(x)\frac{f(x)}{f^{\prime}(x)}+h_{1}(x)\right)\right].$ Structure relations [18] relate the ratio $f/f^{\prime}$ to ratios at shifted indices $n$ as tabulated for example in [1, 22.8], (4) $\displaystyle g_{2}(x)f_{n}^{\prime}(x)=g_{1}(x)f_{n}(x)+g_{0}(x)f_{n-1}(x);$ (5) $\displaystyle\Rightarrow$ $\displaystyle\frac{f_{n}(x)}{f_{n}^{\prime}(x)}=\frac{g_{2}(x)}{g_{1}(x)+g_{0}(x)\frac{f_{n-1}(x)}{f_{n}(x)}}.$ The benefit is that the three-term recurrence equations, in the notation of [1, 22.7] (6) $a_{1,n-1}f_{n}(x)=(a_{2,n-1}+a_{3,n-1}x)f_{n-1}(x)-a_{4,n-1}f_{n-2}(x),$ lead to terminating continued fraction representations for $f/f^{\prime}$ (7) $\frac{f_{n-1}(x)}{f_{n}(x)}=\frac{a_{1,n-1}}{a_{2,n-1}+a_{3,n-1}x-a_{4,n-1}\frac{f_{n-2}(x)}{f_{n-1}(x)}}.$ This is recursively inserted into the denominator of (5) to lower the index $n$ until $f_{0}/f_{1}$ is reached, which avoids problems with cancellation of digits. This work here implements this strategy for the family of Zernike polynomials, $f=R_{n}^{m}$, namely (i) fast calculation of $f^{\prime\prime}/f^{\prime}$ from $f/f^{\prime}$, (ii) calculation of $f/f^{\prime}$ from terminating continued fractions, both without evaluation of $f$ or its derivatives via direct methods like Horner schemes. ## 2\. Zernike Polynomials: Derivatives and Roots ### 2.1. Definition We define Zernike radial polynomials in Noll’s nomenclature [21, 23, 15, 26, 2, 25] for $n\geq 0$, $n-m=0\mod 2$, $m\leq n$ as (8) $\displaystyle R_{n}^{m}(x)$ $\displaystyle=$ $\displaystyle\sum_{s=0}^{(n-m)/2}\frac{(-1)^{s}(n-s)!}{s![(n+m)/2-s]![(n-m)/2-s]!}x^{n-2s}$ (9) $\displaystyle=$ $\displaystyle\sum_{s=0}^{(n-m)/2}(-1)^{s}{n-s\choose s}{n-2s\choose(n-m)/2-s}x^{n-2s}.$ Following the original notation, we will not put the upper index $m$ in $R_{n}^{m}$—which is not a power—into parentheses. The normalization integral is (10) $\int_{0}^{1}x\,R_{n}^{m}(x)R_{n^{\prime}}^{m}(x)dx=\frac{1}{2(n+1)}\delta_{n,n^{\prime}}.$ The complete set of examples for $n\leq 4$ is (11) $\displaystyle R_{0}^{0}(x)$ $\displaystyle=$ $\displaystyle 1;\qquad R_{1}^{1}(x)=x;$ (12) $\displaystyle R_{2}^{0}(x)$ $\displaystyle=$ $\displaystyle 2x^{2}-1;\qquad R_{2}^{2}(x)=x^{2};$ (13) $\displaystyle R_{3}^{1}(x)$ $\displaystyle=$ $\displaystyle 3x^{3}-2x;\qquad R_{3}^{3}(x)=x^{3};$ (14) $\displaystyle R_{4}^{0}(x)$ $\displaystyle=$ $\displaystyle 6x^{4}-6x^{2}+1;\qquad R_{4}^{2}(x)=4x^{4}-3x^{2};\qquad R_{4}^{4}(x)=x^{4}.$ The inversion of (9) decomposes powers $x^{i}$ into sums of $R_{n}^{m}(x)$, ($i\geq m$, $i-m$ even), (15) $\displaystyle x^{i}$ $\displaystyle\equiv$ $\displaystyle\sum_{n=m\mod 2}^{i}h_{i,n,m}R_{n}^{m}(x);\quad i-m=0,2,4,6,\ldots;$ (16) $\displaystyle h_{i,n,m}$ $\displaystyle=$ $\displaystyle 2(n+1)\sum_{s=0}^{(n-m)/2}\frac{(-1)^{s}}{n-2s+i+2}{n-s\choose s}{n-2s\choose(n-m)/2-s}.$ The basic examples are * $m=0$: (17) $\displaystyle x^{0}$ $\displaystyle=$ $\displaystyle R_{0}^{0}(x);$ (18) $\displaystyle x^{2}$ $\displaystyle=$ $\displaystyle\frac{1}{2}R_{0}^{0}(x)+\frac{1}{2}R_{2}^{0}(x);$ (19) $\displaystyle x^{4}$ $\displaystyle=$ $\displaystyle\frac{1}{3}R_{0}^{0}(x)+\frac{1}{2}R_{2}^{0}(x)+\frac{1}{6}R_{4}^{0}(x);$ (20) $\displaystyle x^{6}$ $\displaystyle=$ $\displaystyle\frac{1}{4}R_{0}^{0}(x)+\frac{9}{20}R_{2}^{0}(x)+\frac{1}{4}R_{4}^{0}(x)+\frac{1}{20}R_{6}^{0}(x);$ * $m=1$: (21) $\displaystyle x$ $\displaystyle=$ $\displaystyle R_{1}^{1}(x);$ (22) $\displaystyle x^{3}$ $\displaystyle=$ $\displaystyle\frac{2}{3}R_{1}^{1}(x)+\frac{1}{3}R_{3}^{1}(x);$ (23) $\displaystyle x^{5}$ $\displaystyle=$ $\displaystyle\frac{1}{2}R_{1}^{1}(x)+\frac{2}{5}R_{3}^{1}(x)+\frac{1}{10}R_{5}^{1}(x);$ (24) $\displaystyle x^{7}$ $\displaystyle=$ $\displaystyle\frac{2}{5}R_{1}^{1}(x)+\frac{2}{5}R_{3}^{1}(x)+\frac{6}{35}R_{5}^{1}(x)+\frac{1}{35}R_{7}^{1}(x);$ * $m=2$: (25) $\displaystyle x^{2}$ $\displaystyle=$ $\displaystyle R_{2}^{2}(x);$ (26) $\displaystyle x^{4}$ $\displaystyle=$ $\displaystyle\frac{3}{4}R_{2}^{2}(x)+\frac{1}{4}R_{4}^{2}(x);$ (27) $\displaystyle x^{6}$ $\displaystyle=$ $\displaystyle\frac{3}{5}R_{2}^{2}(x)+\frac{1}{3}R_{4}^{2}(x)+\frac{1}{15}R_{6}^{2}(x);$ (28) $\displaystyle x^{8}$ $\displaystyle=$ $\displaystyle\frac{1}{2}R_{2}^{2}(x)+\frac{5}{14}R_{4}^{2}(x)+\frac{1}{8}R_{6}^{2}(x)+\frac{1}{56}R_{8}^{2}(x).$ Much of this work is based on the representation as a terminating hypergeometric function, (29) $R_{n}^{m}(x)=\frac{n!x^{n}}{(\frac{n-m}{2})!(\frac{n+m}{2})!}F\left(-\frac{n-m}{2},-\frac{n+m}{2};-n;\frac{1}{x^{2}}\right),$ with the three negative integer parameters (30) $a\equiv-(n-m)/2;\quad b\equiv-(n+m)/2;\quad c\equiv-n,$ and the argument (31) $z\equiv\frac{1}{x^{2}}.$ In these variables, the three-term recurrence reads [16, 9, 23] (32) $\displaystyle 2n\left(a-1\right)\left(b-1\right)R_{n+2}^{m}(x)$ $\displaystyle=(n+1)\left[2n(n+2)x^{2}-m^{2}-n(n+2)\right]R_{n}^{m}(x)-2ab(n+2)R_{n-2}^{m}(x).$ $R_{n}^{m}(x)$ is also a product of $x^{m}$ times a polynomial of degree $n-m$, (33) $R_{n}^{m}(x)=(-1)^{a}{(n+m)/2\choose(n-m)/2}x^{m}\left[1-\frac{a(b-1)}{m+1}x^{2}+\frac{a(a+1)(b-1)(b-2)}{2(m+1)(m+2)}x^{4}-\cdots\right],$ which can be summarized in terms of Jacobi Polynomials (34) $R_{n}^{m}(x)=(-1)^{a}x^{m}P_{-a}^{(m,0)}(1-2x^{2})={n\choose-a}x^{m}G_{-a}(m+1,m+1,x^{2}).$ ### 2.2. Derivatives Derivatives of (29) are [1, 3.3.8] (35) $\displaystyle\frac{d}{dx}R_{n}^{m}(x)$ $\displaystyle=$ $\displaystyle{n\choose(n-m)/2}\Big{[}\frac{d}{dx}x^{n}F(a,b;c;z)+x^{n}\frac{d}{dx}F(a,b;c;z)\Big{]};$ $\displaystyle\frac{d^{2}}{dx^{2}}R_{n}^{m}(x)$ $\displaystyle=$ $\displaystyle{n\choose(n-m)/2}\Big{[}\frac{d^{2}}{dx^{2}}x^{n}F(a,b;c;z)+2\frac{d}{dx}x^{n}\frac{d}{dx}F(a,b;c;z)$ $\displaystyle\quad\quad+x^{n}\frac{d^{2}}{dx^{2}}F(a,b;c;z)\Big{]};$ $\displaystyle\frac{d^{3}}{dx^{3}}R_{n}^{m}(x)$ $\displaystyle=$ $\displaystyle{n\choose(n-m)/2}\Big{[}\frac{d^{3}}{dx^{3}}x^{n}F(a,b;c;z)+3\frac{d^{2}}{dx^{2}}x^{n}\frac{d}{dx}F(a,b;c;z)$ $\displaystyle\quad\quad+3\frac{d}{dx}x^{n}\frac{d^{2}}{dx^{2}}F(a,b;c;z)+x^{n}\frac{d^{3}}{dx^{3}}F(a,b;c;z)\Big{]}.$ Faà di Bruno’s Formula [10, 0.430.2][13] relegates the derivatives w.r.t. $x$ to derivatives w.r.t. $z$, $\displaystyle\frac{d}{dx}F(a,b;c;z)$ $\displaystyle=$ $\displaystyle-2x^{-3}F^{\prime}(a,b;c;z);$ $\displaystyle\frac{d^{2}}{dx^{2}}F(a,b;c;z)$ $\displaystyle=$ $\displaystyle 6x^{-4}F^{\prime}(a,b;c;z)+4x^{-6}F^{\prime\prime}(a,b;c;z);$ $\displaystyle\frac{d^{3}}{dx^{3}}F(a,b;c;z)$ $\displaystyle=$ $\displaystyle-24x^{-5}F^{\prime}(a,b;c;z)-36x^{-7}F^{\prime\prime}(a,b;c;z)-8x^{-9}F^{\prime\prime\prime}(a,b;c;z).$ After insertion of these three formulas into (35)–(2.2), the derivatives of $R_{n}^{m}\cong x^{n}F$ are (38) $\displaystyle{R_{n}^{m}}^{\prime}(x)$ $\displaystyle\cong$ $\displaystyle nx^{n-1}F-2x^{n-3}F^{\prime};$ (39) $\displaystyle{R_{n}^{m}}^{\prime\prime}(x)$ $\displaystyle\cong$ $\displaystyle n(n-1)x^{n-2}F-2(2n-3)x^{n-4}F^{\prime}+4x^{n-6}F^{\prime\prime};$ $\displaystyle{R_{n}^{m}}^{\prime\prime\prime}(x)$ $\displaystyle\cong$ $\displaystyle n(n-1)(n-2)x^{n-3}F-6(n-2)^{2}x^{n-5}F^{\prime}$ $\displaystyle\quad\quad+12(n-3)x^{n-7}F^{\prime\prime}-8x^{n-9}F^{\prime\prime\prime},$ where $\cong$ means the binomial factor and the argument list $(a,b;c;z)$ of the hypergeometric function have not been written down explicitly. Since $R_{n}^{m}(x)$ is a polynomial of order $n$, the $(n+1)$st derivatives equal zero. Backward elimination of $F$ and its derivatives with the aid of [1, 15.5.1] (41) $z(1-z)F^{\prime\prime}(a,b;c;z)+[c-(a+b+1)z]F^{\prime}(a,b;c;z)=abF(a,b;c;z)$ leads to the analog of (2), (42) $x^{2}(x^{2}-1)\frac{d^{2}}{dx^{2}}R_{n}^{m}(x)=\left[nx^{2}(n+2)-m^{2}\right]R_{n}^{m}(x)+x(1-3x^{2})\frac{d}{dx}R_{n}^{m}(x),$ which is one special case of differential equations that generate orthogonal functions [19], and could also be obtained by applying the derivatives of [1, 22.6.1] [5, 3] to (34). The derivative of this reaches out to the third derivatives, in which $R^{\prime\prime}$ is reduced to $R$ and $R^{\prime}$ with the help of the previous equation, (43) $\displaystyle x^{3}(x^{2}-1)^{2}\frac{d^{3}}{dx^{3}}R_{n}^{m}(x)=\left[x^{2}(n^{2}+2n+7m^{2})-5x^{4}n(2+n)-3m^{2}\right]R_{n}^{m}(x)$ $\displaystyle+x\left[6x^{2}(2x^{2}-1)-m^{2}(x^{2}-1)+2+x^{2}(x^{2}-1)n(n+2)\right]\frac{d}{dx}R_{n}^{m}(x).$ ### 2.3. Zeros #### 2.3.1. Ratios of Derivatives Installation of $f/f^{\prime}$ in (1) progresses by dividing $R_{n}^{m}\cong x^{n}F$ through (38), (44) $\frac{R_{n}^{m}(x)}{{R_{n}^{m}}^{\prime}(x)}=\frac{x}{n-2z\frac{F^{\prime}(a,b;c;z)}{F(a,b;c;z)}}.$ The analog of (5) is implemented by substituting [1, 15.2.1] (45) $F^{\prime}(a,b;c;z)=\frac{ab}{c}F(a+1,b+1;c+1;z)$ in the denominator. In lieu of (7) we find the continued fractions [7] (46) $\frac{F(a,b;c;z)}{F(a+1,b+1;c+1;z)}\equiv\frac{-bz}{c}+1-\frac{\frac{(a+1)(c-b)z}{c(c+1)}}{\frac{(a+1-b)z}{c+1}+1-\cdots}\,\frac{\frac{(a+2)(c+1-b)z}{(c+1)(c+2)}}{\frac{(a+2-b)z}{c+2}+1-\cdots}$ which terminate in our cases since $a$ is a negative integer and $c=a+b$. This already suffices to implement the standard Newton iteration, ie, to approximate (1) by $\Delta x=-f(x)/f^{\prime}(x)$. Division of (42) through ${R_{n}^{m}}^{\prime}(x)$ yields (47) $\frac{{R_{n}^{m}}^{\prime\prime}(x)}{{R_{n}^{m}}^{\prime}(x)}=\frac{1}{x^{2}-1}\left[\left(n(n+2)-\frac{m^{2}}{x^{2}}\right)\frac{R_{n}^{m}(x)}{{R_{n}^{m}}^{\prime}(x)}+\frac{1-3x^{2}}{x}\right].$ This is $f^{\prime\prime}(x)/f^{\prime}(x)$ of the generic formula, and can be quickly computed from $R_{n}^{m}(x)/{R_{n}^{m}}^{\prime}(x)=f(x)/f^{\prime}(x)$ of the lower order. #### 2.3.2. Initial Guesses For $n$ and $m$ fixed, the strategy adopted here is to compute the $(n-m)/2$ distinct roots in $(0,1)$ starting with the smallest, then bootstrapping the others in naturally increasing order. An approximation to the smallest root is found by equating the first three terms in the square bracket of (33) with zero—hoping that higher powers of $x$ become insignificant for small $x$—and solving the bi-quadratic equation for $x$. This guess may become unstable for $n$ approximately larger than 11 in the sense that the Newton iterations converge to another than this smallest root. Instead, the simple, heuristic initial guess (48) $x\approx\frac{1.46m+2.41}{n+0.46m+1.06}$ is used in general, but keeping the solution to the bi-quadratic equation when this is exact, ie, in the cases $n-m=2$ or $4$. A shooting method is useful to produce an initial estimate for one root supposed an adjacent one has already been found. The third order Taylor extrapolation from one root $x$ to the next one at $x+\Delta x$ is (49) $f(x+\Delta x)\approx f(x)+\Delta xf^{\prime}(x)+\frac{(\Delta x)^{2}}{2!}f^{\prime\prime}(x)+\frac{(\Delta x)^{3}}{3!}f^{\prime\prime\prime}(x)\approx 0.$ Division through $\Delta xf^{\prime}$ and exploiting $f(x)=0$ yields a quadratic equation for the approximate distance $\Delta x$ to the next one, (50) $1+\frac{\Delta x}{2}\frac{f^{\prime\prime}(x)}{f^{\prime}(x)}+\frac{(\Delta x)^{2}}{6}\frac{f^{\prime\prime\prime}(x)}{f^{\prime}(x)}\approx 0,$ from which the branch $\Delta x>0$ is systematically selected to start computation of the root adjacent to the previous one. The two ratios of derivatives are obtained by setting $R_{n}^{m}(x)=0$ in (42) and (43), then dividing both equations through ${R_{n}^{m}}^{\prime}(x)$. This aim to locate the next root with sufficient accuracy—and to prevent the Newton’s Method to be be drawn into the second next root which would call for more administrative care [4]—is the rationale to look into third derivatives; it might also guide the way to even higher order Newton’s methods employing $f^{\prime\prime\prime}$. ## 3\. Summary The Newton’s Method of third order convergence is implemented for Zernike Polynomials $R_{n}^{m}$ by computation of the ratios ${R_{n}^{m}}^{\prime\prime}/{R_{n}^{m}}^{\prime}$ and $R_{n}^{m}/{R_{n}^{m}}^{\prime}$ with relay to the generic formulas of associated, terminating hypergeometric functions. Adding knowledge on the derivative ${R_{n}^{m}}^{\prime\prime\prime}$, a shooting method is proposed which generates an initial guess for the adjacent root from each root found. ## Appendix A Table of Roots of Low-order Polynomials The roots $x_{i,n,m}$ of $R_{n}^{m}(x)$ are tabulated below for $2\leq n\leq 20$ in two major columns. Each column contains $n$, then $m$, then $(n-m)/2=|a|$ values of $x_{i,n,m}$. Only the roots $x>0$ are included, and only the standard parameter range for even, positive values of $n-m$ is considered. Because the $G_{-a}(m+1,m+1,y)$ mentioned in (34) build a system of orthogonal polynomials with weight $y^{m}$ over the unit interval $0\leq y\leq 1$, the squares $x_{i,n,m}^{2}=y_{i,n,m}$ are also the abscissae for Gaussian integration of moment $m$ [1, Tab. 25.8][6]. 2 0 0.7071067811865475727 14 4 0.8246570394661102421 4 0 0.4597008433809830485 14 4 0.9278396109654096779 4 0 0.8880738339771152567 14 4 0.9862121569592748882 6 0 0.3357106870197287818 16 4 0.4352810401596804435 6 0 0.7071067811865475727 16 4 0.6105335011400015999 6 0 0.9419651451198933767 16 4 0.7533739926853023627 8 0 0.2634992299855423159 16 4 0.8653533945767831748 8 0 0.5744645143153508382 16 4 0.9446613988337823065 8 0 0.8185294874300058643 16 4 0.9894368252983534173 8 0 0.9646596061808674349 18 4 0.3914606479189805532 10 0 0.2165873427295972042 18 4 0.5534677249323020076 10 0 0.4803804169063914387 18 4 0.6903188343084598610 10 0 0.7071067811865475727 18 4 0.8039833969179427386 10 0 0.8770602345636481223 18 4 0.8931498697389127495 10 0 0.9762632447087885579 18 4 0.9561484720295934103 12 0 0.1837532119404283737 18 4 0.9916369448881094950 12 0 0.4115766110542091183 20 4 0.3554897669109032265 12 0 0.6170011401597257361 20 4 0.5054988151106132310 12 0 0.7869622564275865484 20 4 0.6353277098425955671 12 0 0.9113751660173390334 20 4 0.7472059125768437671 12 0 0.9829724091252897145 20 4 0.8402437401628405356 14 0 0.1595181614381909196 20 4 0.9130561549207910632 14 0 0.3594918736220650279 20 4 0.9643628549735204780 14 0 0.5450480935764305812 20 4 0.9932085369576263423 14 0 0.7071067811865475727 7 5 0.9258200997725514192 14 0 0.8384047803350709316 9 5 0.7942238940183964369 14 0 0.9331482158798232174 9 5 0.9616464847987593600 14 0 0.9871949939963123866 11 5 0.6840930020506119646 16 0 0.1409080258581174028 11 5 0.8727107021799724862 16 0 0.3188522562146716699 11 5 0.9758129824224018867 16 0 0.4870665201405610101 13 5 0.5974058327888663866 16 0 0.6389700139694938219 13 5 0.7840085394020771536 16 0 0.7692316434259740543 13 5 0.9116353458812981314 16 0 0.8733648750425931917 13 5 0.9831524024611432155 16 0 0.9478044306220632098 15 5 0.5288602129223232140 16 0 0.9900226907746954019 15 5 0.7061939018640647214 18 0 0.1261740078074202742 15 5 0.8404366834692439392 18 0 0.2863292621034079777 15 5 0.9345601788393610443 18 0 0.4396752024502914580 15 5 0.9875190300435001678 18 0 0.5812686885581361818 17 5 0.4737815904024170188 18 0 0.7071067811865475727 17 5 0.6399479279255915198 18 0 0.8137116883158951319 17 5 0.7731681889583568168 18 0 0.8981568439589463493 17 5 0.8765920400095927878 18 0 0.9581312820607194025 17 5 0.9493949303895553582 18 0 0.9920081248426411147 17 5 0.9903520628752460198 20 0 0.1142223084227163565 19 5 0.4287525723551651180 20 0 0.2597466393536357887 19 5 0.5837872858381362162 20 0 0.4003688497504367394 19 5 0.7126994626270685140 20 0 0.5322614986408245041 19 5 0.8187572351505322255 20 0 0.6523517689936806363 19 5 0.9014368896323405878 20 0 0.7579163340968551044 19 5 0.9596126391703136971 20 0 0.8465800003925344486 19 5 0.9923041120472186893 20 0 0.9163540713880810040 8 6 0.9354143466934853324 20 0 0.9656768006659848247 10 6 0.8164965809277260345 20 0 0.9934552150241026114 10 6 0.9660917830792958849 3 1 0.8164965809277260345 12 6 0.7132667548676191460 5 1 0.5958615826865180098 12 6 0.8854995128634446377 5 1 0.9192110607898046348 12 6 0.9783359081211411290 7 1 0.4608042298407784010 14 6 0.6297279581530694781 7 1 0.7684615381131740808 14 6 0.8030111569365681046 7 1 0.9546790248493448594 14 6 0.9197999176484970008 9 1 0.3738447061866471688 14 6 0.9847470483288770504 9 1 0.6452980455813291938 16 6 0.5622633194744699470 9 1 0.8503863747508400017 16 6 0.7291002958196058925 9 1 0.9710282199223060351 16 6 0.8535798917458220503 11 1 0.3139029878781443572 16 6 0.9401303969875089983 11 1 0.5518475574344458012 16 6 0.9885994756607009437 11 1 0.7496833930084177977 18 6 0.5071545248007354179 11 1 0.8955370355972955831 18 6 0.6650990116151613840 11 1 0.9798929242261785744 18 6 0.7899637876841181239 13 1 0.2702856427564344077 18 6 0.8860816722252334854 13 1 0.4803812423169180335 18 6 0.9533796195297132847 13 1 0.6643255837527634045 18 6 0.9911212731459003722 13 1 0.8142575205172167818 20 6 0.4615059331053839586 13 1 0.9229958831606540626 20 6 0.6100885865453433698 13 1 0.9852327505925770890 20 6 0.7319787736696559133 15 1 0.2371973029714337655 20 6 0.8314254884667372503 15 1 0.4245476318823276363 20 6 0.9085207085711664021 15 1 0.5938221258198196351 20 6 0.9625678871955013483 15 1 0.7396983346814803850 20 6 0.9928726539379972849 15 1 0.8568606521572865731 9 7 0.9428090415820633563 15 1 0.9409149519691435426 11 7 0.8343946751715023424 15 1 0.9886964213353295339 11 7 0.9696128930707756943 17 1 0.2112674970031354627 13 7 0.7374505105030352281 17 1 0.3799555987772065824 13 7 0.8959410734924830866 17 1 0.5355602273735704522 13 7 0.9803818873685179458 17 1 0.6743984839815206911 15 7 0.6571568797605316092 17 1 0.7925073093774768207 15 7 0.8189060737060670503 17 1 0.8863924190810575920 15 7 0.9265772077286310582 17 1 0.9532451899171618948 15 7 0.9860657631628138020 17 1 0.9910701715078688023 17 7 0.5911284719636833440 19 1 0.1904148441776781775 17 7 0.7486326613042562084 19 1 0.3436262195904129513 17 7 0.8647030262108379439 19 1 0.4870081978929722277 17 7 0.9448234063345077871 19 1 0.6179666376955199603 17 7 0.9895076756006223606 19 1 0.7335050094709572033 19 7 0.5364117760083304542 19 1 0.8308624555679250401 19 7 0.6868800285074768697 19 1 0.9076801990993329516 19 7 0.8044060078556206639 19 1 0.9620876312949896425 19 7 0.8942045500841436789 19 1 0.9927676416720699892 19 7 0.9567806932331796022 4 2 0.8660254037844385966 19 7 0.9917768363310000002 6 2 0.6751652804971347566 10 8 0.9486832980505137680 6 2 0.9367417879781805290 12 8 0.8490975736565613552 8 2 0.5431369901889407936 12 8 0.9724710674756380513 8 2 0.8080818238035354373 14 8 0.7578370705278929531 8 2 0.9628114955311087853 14 8 0.9046299747290316162 10 2 0.4518280448392144044 14 8 0.9820745595413366003 10 2 0.6949479871872660253 16 8 0.6807458887789931135 10 2 0.8725819517089140609 16 8 0.8324053937687126981 10 2 0.9754483331027854476 16 8 0.9322943792644380334 12 2 0.3859349518416070879 16 8 0.9871744960883001019 12 2 0.6047036690554417060 18 8 0.6163473579853351314 12 2 0.7810976974325815059 18 8 0.7654979183523115127 12 2 0.9091312772247495122 18 8 0.8742424706423237435 12 2 0.9825584257853499093 18 8 0.9488319407257733706 14 2 0.3364437305441106418 18 8 0.9902817948430912010 14 2 0.5332155968164153936 20 8 0.5623002252838306125 14 2 0.7006879382687045688 20 8 0.7059431502741818631 14 2 0.8352024197624841051 20 8 0.8169643175952393532 14 2 0.9318991674757601817 20 8 0.9012382495289457118 14 2 0.9869627373443907725 20 8 0.9597179665033356288 16 2 0.2980215318345257325 20 8 0.9923422148430279810 16 2 0.4759020418201234115 11 9 0.9534625892455923513 16 2 0.6324385240645979955 13 9 0.8613939328976448762 16 2 0.7655049675886762550 13 9 0.9748375170636489617 16 2 0.8714619176716458249 15 9 0.7752632451107936973 16 2 0.9470538373566472767 15 9 0.9119748762120825081 16 2 0.9898824305070776930 15 9 0.9834982084062168228 18 2 0.2673780750923551164 17 9 0.7012604730894272942 18 2 0.4292182386813452322 17 9 0.8440171326522077910 18 2 0.5748458299193258680 17 9 0.9371828916889872740 18 2 0.7031379109502706726 17 9 0.9881197223055366852 18 2 0.8113864729725273062 19 9 0.6385852750050469151 18 2 0.8969530821125818454 19 9 0.7802151464898949840 18 2 0.9576530595755170516 19 9 0.8825163921882515083 18 2 0.9919184646169658670 19 9 0.9522959699168392911 20 2 0.2423925241972734457 19 9 0.9909494948025596717 20 2 0.3905933409401172729 12 10 0.9574271077563381027 20 2 0.5260526858211680423 14 10 0.8718317153731350855 20 2 0.6483100675208097741 14 10 0.9768291428674981125 20 2 0.7553451890441794658 16 10 0.7903347449834110527 20 2 0.8450496504116072893 16 10 0.9182661541320223941 20 2 0.9155547407096559231 16 10 0.9847122732288815516 20 2 0.9653576903957581390 18 10 0.7192722131418530784 20 2 0.9933952458125041574 18 10 0.8541140545067785750 5 3 0.8944271909999158554 18 10 0.9414112100240797920 7 3 0.7279134123608967943 18 10 0.9889351394546163077 7 3 0.9480050066727199187 20 10 0.6583509748272358131 9 3 0.6027143852742457009 20 10 0.7931750687612736384 9 3 0.8359493221264154839 20 10 0.8897625152667285597 9 3 0.9684648164078416555 20 10 0.9553196436506564693 11 3 0.5113489892733628084 20 10 0.9915313152497930993 11 3 0.7320153318669290199 13 11 0.9607689228305228424 11 3 0.8889787567592866147 15 11 0.8808037886787085657 11 3 0.9786966233548161087 15 11 0.9785284968679074380 13 3 0.4429582456583350258 17 11 0.8035014494300847243 13 3 0.6458329596901977165 17 11 0.9237159298241287564 13 3 0.8053384408042754128 17 11 0.9857598779860213822 13 3 0.9195679148240427647 19 11 0.7352174806998263978 13 3 0.9845992603558400003 19 11 0.8629761939750234534 15 3 0.3902219391376385849 19 11 0.9451049233245847336 15 3 0.5755870542650849409 19 11 0.9896457777687525104 15 3 0.7296653799715601130 14 12 0.9636241116594315148 15 3 0.8518232083769392560 16 12 0.8885993155618584494 15 3 0.9389444522639546209 16 12 0.9799955389607631906 15 3 0.9883295764481713208 18 12 0.8151048281585070443 17 3 0.3484639562887706932 18 12 0.9284828318578448592 17 3 0.5180380683998889735 18 12 0.9866730715175761057 17 3 0.6639777561205169043 20 12 0.7494357022184537920 17 3 0.7865138450349050681 20 12 0.8708181789128337513 17 3 0.8833196205032936010 20 12 0.9483595986475444883 17 3 0.9520307422269438380 20 12 0.9902706191583880990 17 3 0.9908430322275690871 15 13 0.9660917830792958849 19 3 0.3146454476233096487 17 13 0.8954359308455520639 19 3 0.4703652913957364068 17 13 0.9812748728586200286 19 3 0.6075556764710252633 19 13 0.8254089266585322715 19 3 0.7270113006012387524 19 13 0.9326878942873719769 19 3 0.8270388433632478442 19 13 0.9874761693913842731 19 3 0.9056952852248679742 16 14 0.9682458365518542553 19 3 0.9612979313921893310 18 14 0.9014806129319800077 19 3 0.9926194840347464243 18 14 0.9824003676126956686 6 4 0.9128709291752769017 20 14 0.8346212400166618250 8 4 0.7657261797294159233 20 14 0.9364250918976404492 8 4 0.9558574253919850383 20 14 0.9881879542766879299 10 4 0.6481612911435379321 17 15 0.9701425001453318764 10 4 0.8566734238949372804 19 15 0.9068636367673578169 10 4 0.9726240720110040927 19 15 0.9833982044520968024 12 4 0.5586908741397313971 18 16 0.9718253158075500497 12 4 0.7608650731649617693 20 16 0.9116881483483436632 12 4 0.9016049670538779370 20 16 0.9842889413994087011 12 4 0.9811849966449255334 19 17 0.9733285267845752653 14 4 0.4896856591758441124 20 18 0.9746794344808964450 14 4 0.6789209633908173114 ## Appendix B PARI implementation The full source code of the PARI interpreter program which computed the values shown in Appendix A is listed below. The language is similar to C/C++ and has inherent support for arbitrary precision computation [22]. HypergAugmRatio implements (46). HypergRatio implements (45). ZernikePrratio implements (44). Zernike2Prratio implements (47). Zernike3Prratio implements (43). ZernikeRoot implements (1). ZernikeRootEst implements (48), but (33) if $n=m+2$. ZernikeAllRoot implements a loop with guesses as in (50). main loops over $n$ and $m$ to tabulate the zeros up to a maximum $n$. /** Compute the quotient of a Gauss Hypergeometric Function over * the Function with the same argument but all the three parameters * increased by 1. * @param[in] a first parameter of the Gauss Hypergeometric Function F(a,b;c;z). * This must be a negative integer. * @param[in] b second parameter of F * @param[in] c third parameter of F * @param[in] z argument of F * @return the ratio F(a,b;c;z)/F(a+1,b+1;c+1;z) * @warning the function assumes that the parameter a is a negative integer */ HypergAugmRatio(a,b,c,z)={ local(ff=0.0) ; forstep(ap2= -1,a+1,-1, ff = z*(ap2-b)+(ap2+c-a)*(1.0-ff) ; ff = ap2*(ap2+c-a-b-1)*z/((ap2+c-a-1)*ff) ; ) ; return(-b*z/c+1-ff) ; } /** Compute the quotient of a Gauss Hypergeometric Function over * its first derivative. * @param[in] a first parameter of the Gauss Hypergeometric Function F(a,b;c;z). * This must be a negative integer. * @param[in] b second parameter of F * @param[in] c third parameter of F * @param[in] z argument of F * @return ratio of the value divided by the first derivative, F/F’. * @warning Checking a against being a negative integer is not done. */ HypergRatio(a,b,c,z)={ c*HypergAugmRatio(a,b,c,z)/(a*b) ; } /** Compute the ratio of the Zernike polynomial over its first derivative. * @param[in] n first parameter of the Zernike polynomial, a positive integer * @param[in] m second parameter of the polynomial, a positive integer * less than or equal to n, with n-m even. * @param[in] x the argument of the polynomial in the interval [0,1] * @param[in] ffprime the ratio F’(a,b;c;z)/F(a,b;c;z) of the Gauss * hypergeometric function, where a=-(n-m)/2, b=-(n+m)/2, c=a+b, * and z=1/x^2. * @return the ratio R/R’ * @warning no check is done that n-m is a positive even integer * or that n and m are individually positive or that x is between 0 * and one. */ ZernikePrratio(n,m,x,ffprime) = { local(z=1.0/x^2) ; /** To enhance stability in the case of ffprime close to * zero, we do not use the equivalent x/(n-2*z/ffprime) ; */ x*ffprime/(n*ffprime-2*z) ; } /** Compute the ratio of the second derivative of the Zernike polynomial * over the first derivative, with respect to the argument x. * @param[in] n first parameter of the Zernike polynomial, a positive integer * @param[in] m second parameter of the polynomial, a positive integer * less than or equal to n, with n-m even. * @param[in] x the argument of the polynomial in the interval [0,1] * @param[in] rrprime the ratio R(x)/R’(x) * @return the ratio of derivatives R’’/R’ * @warning no checking is done that n-m is a positive even integer * or that n and m are individually positive or that x is between 0 * and one. */ Zernike2Prratio(n,m,x,rrprime) = { local(xsq=x^2) ; ((n*(n+2)*xsq-m^2)*rrprime+x*(1-3*xsq))/(xsq*(xsq-1)) ; } /** Compute the ratio of the third derivative of the Zernike polynomial * over the first derivative. Derivatives are with respect to the argument x. * The ratio R’/R of the first derivative is to be provided. * @param[in] n first parameter of the Zernike polynomial, a positive integer * @param[in] m second parameter of the polynomial, a positive integer * less than or equal to n, with n-m even. * @param[in] x the argument of the polynomial in the interval [0,1] * @param[in] rrprime the ratio R’/R * @return the ratio R’’’/R’ * @warning no test is performed that n-m is a positive even integer * or that n and m are individually positive or that x is between 0 * and one. */ Zernike3Prratio(n,m,x,rrprime) = { local(x2=x^2,x2m=x^2-1, n2n=n*(n+2), m2=m^2) ; ( (x2*(n2n+7*m2)-5*x2^2*n2n-3*m2) *rrprime +x*(6*x2*(2*x2-1)-m2*x2m+2+x2*x2m*n2n) ) /(x*x2*x2m^2) ; } /** Compute a root of the Zernike polynomial within a specified error bar. * @param[in] n first parameter of the Zernike polynomial, a positive integer * @param[in] m second parameter of the polynomial, a positive integer * less than or equal to n, with n-m even. * @param[in] x an initial guess of the root * @param[in] eps the absolute accuracy of the result. The Newton iteration * will be terminated if two subsequent estimates agree within this limit. * @return a root x such that R(x)=0 * @warning no test is performed that n-m is a positive even integer * or that n and m are individually positive or that x is between 0 * and one. */ ZernikeRoot(n,m,x,eps) = { /** The variable root is used to keep a history of the most recent * approximation to the root. The variables a, b, and c are the * corresponding arguments to the associated Gauss Hypergeometric Function. */ local(root=x,z, a= -(n-m)/2, b= -(n+m)/2, c= -n, fprimef,ffprime,rrp,r2prp) ; if(n==m, 0.0, /** For an absolute accuracy of 1.e-30, up to 10 iterations * will be needed for n<=10. So we update the root with up to a * maximum of 20 Newton iterations. */ for(i=1,20, z=1/root^2 ; /* ffprime contains F’(a,b;c;z)/F(a,b;c;z) of the * associated Hypergeometric function. */ ffprime= HypergRatio( a, b, c, 1/root^2) ; /* rrp and r2prp are the ratios R/R’ and R’’/R’ of * the Zernike polynomial relative to its 1st and 2nd derivative. * This completes the relay F’/F -> R/R’ -> R’’/R’. */ rrp= ZernikePrratio(n,m,x,ffprime) ; r2prp= Zernike2Prratio(n,m,x,rrp) ; x=root ; /** Now perform the third order Newton update. The formula * is Delta(x) = -(f/f’)/[1-(f/f’)*(f’’/f’)/2]. Reduction to the * 2nd order would be implemented as the simpler * root = x-rrp ; */ root = x-rrp/(1.0-0.5*rrp*r2prp) ; /* terminate if the old and the new guess agree within eps. */ if( abs(x-root) < eps, break ; ) ; ) ; return(root) ; ) ; } /* ZernikeRoot */ /** Provide a guess of the smallest nonzero root of the Zernike Polynomial. * @param[in] n first parameter of the Zernike polynomial, a positive integer * @param[in] m second parameter of the polynomial, a positive integer * less than or equal to n, with n-m even. * @return an estimate of a root which is the smallest nontrivial (nonzero) * positive between 0 and 1 out of the total of (n-m)/2 . */ ZernikeRootEst(n,m) = { local(a= -(n-m)/2, b= -(n+m)/2, x2) ; if(n==m, 0, if(n == m+2, /** if n equals m+2, the resulting equation is * a linear equation in x^2. */ x2 = (m+1)/(a*(b-1)) ; return( sqrt(x2)) , /* if n does not equal m+2, a simple heuristic estimate * with a rational function of n and m is used. */ return((1.46*m+2.41)/(n+0.46*m+1.06)) ; ) ; ) ; } /* ZernikeRootEst */ /** * @param[in] n first parameter of the Zernike polynomial, a positive integer * @param[in] m second parameter of the polynomial, a positive integer * less than or equal to n, with n-m even. * @param[in] eps the desired absolute accuracy of each root. * @return the (n-m)/2 positive roots in the open interval from 0 to 1 * @note the m-fold degenerate root at zero is not returned, nor the * symmetric values on the negative real axis. * @warning no test is performed that n-m is a positive even integer * or that n and m are individually positive or that x is between 0 * and one. */ ZernikeAllRoot(n,m,eps) = { local(s=(n-m)/2,rs,x,r2pr,r3pr,disc) ; /* The result contains s=(n-m)/2 individual values, which are * collected in the vector rs. */ rs=vector(s) ; for(i=1,s, if(i==1, /** If this is the first root for a pair (n,m), * we attach to the leftmost (positive, but smallest in value) * of them, and call ZernikeRootEst() for an initial value. */ x=ZernikeRootEst(n,m), /** Otherwise, for the 2nd and higher roots, we take the * previous root, and solve the quadratic equation for the step * D to the next root which follows from a 3rd order Taylor * approximation at the known root. This estimate is then * used to call the Newton routine for the next root. */ x=rs[i-1] ; /** Compute R’’/2R’ at the old root. The general formula * x^2(x^2-1) [R’’/R’] = {n(n+2)x^2-m^2}*[R/R’]+x(1-3x^2) * is simplified because R and therefore R/R’ are known * to be zero for the recent root. */ r2pr= Zernike2Prratio(n,m,x,0)/2 ; /** Compute R’’’/6R’ at the old root. Application of the formula * x^3(x^2-1)^2 [R’’’/R’] * = {x^2(n^2+2n+7m^2...}*[R/R’]+x*[6x^2(2x^2-1)-m^2(x^2-1)+...] * uses that R and therefore R/R’ are known * to be zero for the recent root. */ r3pr= Zernike3Prratio(n,m,x,0)/6 ; /* The quadratic estimate from 1+(r2pr/2)*D+(r3pr/6)*D^2=0 * is in these variables 1+r2pr*D+r3pr*D^2=0, after division * 1/r3pr + r2pr/r3pr*D +D^2 = 0. disc is the discriminant * of the quadratic. */ disc = 1-4*r3pr/r2pr^2 ; /* This third order local approximation to the full polynomial * usually has a root to the left and another one to the right of * the current root. We select the sign of the square root of * the discriminant to lock into the larger of these two values. */ if(disc>0, if( r2pr/r3pr>0, x += r2pr/(2*r3pr)*(-1+sqrt(disc)) , x += r2pr/(2*r3pr)*(-1-sqrt(disc)) ; ) , /** If the discriminant was negative, just ignoring the * square root works also and keeps arithmetics real-valued. */ x -= r2pr/(2*r3pr) ; ) ; ) ; /** Given the starting value x of the next root, call the Newton * routine to converge to the root within eps, and store the result * as another value in the vector. */ rs[i] = ZernikeRoot(n,m,x,eps) ; ) ; return(rs) ; } /* ZernikeAllRoot */ /** The main routine tabulates the zeros of the Zernike Polynomials * of lowest order. * @param[in] nmax the maximum value of the parameter n to be used * @return 0 */ main(nmax)={ /** we loop with m and n over all cases that have non-zero roots. */ for(m=0,nmax, /* The parameter n runs from m up in steps of 2. */ forstep(n=m,nmax,2, /* Collect the roots of this (n,m) with an accuracy of 1.e-30 * in the vector rts. */ rts=ZernikeAllRoot(n,m,1.e-30) ; /* Print the (n-m)/2 results. */ for(i=1,(n-m)/2, print(n" "m" "rts[i]) ; ) ; ) ; ) ; return(0) ; } /* main */ { /** Collect all non-trivial roots up to some maximum explicit n. */ main(20) ; } ## References * [1] Milton Abramowitz and Irene A. Stegun (eds.), _Handbook of mathematical functions_ , 9th ed., Dover Publications, New York, 1972. MR 0167642 (29 #4914) * [2] G. Conforti, _Zernike aberration coefficients from Seidel and higher-order power-series coefficients_ , Opt. Lett. 8 (1983), no. 7, 407–408. * [3] E. H. Doha, _On the coefficients of differentiated expansions and derivatives of Jacobi polynomials_ , J. Phys. A: Math. Gen. 35 (2002), 3467–3478. MR 1907373 (2003e:33020) * [4] L. W. Ehrlich, _A modified Newton method for polynomials_ , Commun. ACM 10 (1967), no. 2, 107–108. * [5] David Elliott, _Uniform asymptotic expansions of the Jacobi polynomials and an associated function_ , Math. Comp. 25 (1971), no. 114, 309–315. * [6] Herbert Fishman, _Numerical integration constants_ , Math. Tables Aids Comp. 11 (1957), no. 57, 1–9. MR 0086391 (19,177g) * [7] Evelyn Frank, _A new class of continued fraction expansions for the ratios of hypergeometric functions_ , Trans. Am. Math. Soc. 81 (1956), no. 2, 453–476. * [8] Jürgen Gerlach, _Accelerated convergence in Newton’s method_ , SIAM Review 36 (1994), no. 2, 272–276. MR 1278637 (95e:65053) * [9] Amparo Gil, Javier Segura, and Nico M. Temme, _Numerically satisfactory solutions of hypergeometric recursions_ , Math. Comp. 76 (2007), no. 259, 1449–1468. * [10] I. Gradstein and I. Ryshik, _Summen-, Produkt- und Integraltafeln_ , 1st ed., Harri Deutsch, Thun, 1981. MR 0671418 (83i:00012) * [11] Eldon Hansen and Merrell Patrick, _A family of root finding methods_ , Numer. Math. 27 (1977), no. 3, 257–269. MR 0433858 (55 #6829) * [12] D. J. Hofsommer, _Note on the computation of the zeros of functions satisfying a second order differential equation_ , Math. Tabl. Aids Comput. 12 (1958), no. 61, 58–60, E: [20, (C2)]. MR 0099752 (20 #6190) * [13] Warren P. Johnson, _The curious history of Faà di Bruno’s formula_ , Am. Math. Monthly 109 (2002), no. 3, 217–234. MR 1903577 (2003d:01019) * [14] Bahman Kalantari, Iraj Kalantari, and Rahim Zaare-Nahandi, _A basic family of iteration functions for polynomial root finding and its characterizations_ , J. Comp. Appl. Math. 80 (1997), no. 2, 209–226. MR 1455244 (98d:65066) * [15] Eric C. Kintner, _On the mathematical properties of the Zernike polynomials_ , Optica Acta 23 (1976), no. 8, 679–680. * [16] Eric C. Kintner and Richard M. Sillitto, _A new analytic method for computing the optical transfer functions_ , Optica Acta 23 (1976), no. 8, 607–619. * [17] Stanisław Lewanowicz, _Recurrences for the coefficients of series expansions with respect to classical orthogonal polynomials_ , Applic. Mathemat. 29 (2002), no. 1, 97–116. MR 1907630 (2003d:33017) * [18] Francisco Marcellán and Ridha Sfaxi, _Second structure relation for semiclassical orthogonal polynomials_ , J. Comput. Appl. Math. 200 (2007), no. 2, 537–554. MR 2289233 * [19] Mohammad Masjed-Jamei, _A basic class of symmetric orthogonal polynomials using the extended Sturm-Liouville theorem for symmetric functions_ , J. Math. Anal. Appl. 325 (2007), no. 2, 753–775. MR 2270049 * [20] Richard J. Mathar, _Numerical representation of the incomplete gamma function of complex argument_ , arXiv:math.NA/0306184 (2003). * [21] Robert J. Noll, _Zernike polynomials and atmospheric turbulence_ , J. Opt. Soc. Am. 66 (1976), no. 3, 207–211. * [22] The PARI-Group, Bordeaux, _PARI/GP, version 2.3.2_ , 2007, available from http://pari.math.u-bordeaux.fr/. * [23] Aluizio Prata, Jr. and W. V. T. Rusch, _Algorithm for computation of Zernike polynomials expansion coefficients_ , Appl. Opt. 28 (1989), no. 4, 749–754. * [24] T. S. Shao, T. C. Chen, and R. M. Frank, _Tables of zeros and Gaussian weights of certain associated Laguerre polynomials and the related generalized Hermite polynomials_ , Math. Comp. 18 (1964), no. 88, 598–616. MR 0166397 (29 #3674) * [25] William J. Tango, _The circle polynomials of Zernike and their application in optics_ , Appl. Phys. A 13 (1977), no. 4, 327–332. * [26] Robert K. Tyson, _Conversion of Zernike aberration coefficients to Seidel and higher-order power series aberration coefficients_ , Opt. Lett. 7 (1972), no. 6, 262–264.

arxiv-papers

2024-09-04T02:48:52.869587

0705.4676

# Recursive n-gram hashing is pairwise independent, at best Daniel Lemire lemire@acm.org Owen Kaser o.kaser@computer.org LICEF, Université du Québec à Montréal (UQAM), 100 Sherbrooke West, Montreal, QC, H2X 3P2 Canada Dept. of CSAS, University of New Brunswick, 100 Tucker Park Road, Saint John, NB, Canada ###### Abstract Many applications use sequences of $n$ consecutive symbols ($n$-grams). Hashing these $n$-grams can be a performance bottleneck. For more speed, recursive hash families compute hash values by updating previous values. We prove that recursive hash families cannot be more than pairwise independent. While hashing by irreducible polynomials is pairwise independent, our implementations either run in time $O(n)$ or use an exponential amount of memory. As a more scalable alternative, we make hashing by cyclic polynomials pairwise independent by ignoring $n-1$ bits. Experimentally, we show that hashing by cyclic polynomials is twice as fast as hashing by irreducible polynomials. We also show that randomized Karp-Rabin hash families are not pairwise independent. ###### keywords: Rolling Hashing , Rabin-Karp Hashing , Hashing Strings ## 1 Introduction An $n$-gram is a consecutive sequence of $n$ symbols from an alphabet $\Sigma$. An $n$-gram hash function $h$ maps $n$-grams to numbers in $[0,2^{L})$. These functions have several applications from full-text matching [1, 2, 3], pattern matching [4], or language models [5, 6, 7, 8, 9, 10, 11] to plagiarism detection [12]. To prove that a hashing algorithm must work well, we typically need hash values to satisfy some statistical property. Indeed, a hash function that maps all $n$-grams to a single integer would not be useful. Yet, a single hash function is deterministic: it maps an $n$-gram to a single hash value. Thus, we may be able to choose the input data so that the hash values are biased. Therefore, we randomly pick a function from a family $\mathcal{H}$ of functions [13]. Such a family $\mathcal{H}$ is uniform (over $L$-bits) if all hash values are equiprobable. That is, considering $h$ selected uniformly at random from $\mathcal{H}$, we have $P(h(x)=y)=1/2^{L}$ for all $n$-grams $x$ and all hash values $y$. This condition is weak; the family of constant functions ($h(x)=c$) is uniform111 We omit families uniform over an arbitrary interval $[0,b)$—not of the form $[0,2^{L})$. Indeed, several applications [14, 15] require uniformity over $L$-bits. . Intuitively, we would want that if an adversary knows the hash value of one $n$-gram, it cannot deduce anything about the hash value of another $n$-gram. For example, with the family of constant functions, once we know one hash value, we know them all. The family $\mathcal{H}$ is pairwise independent if the hash value of $n$-gram $x_{1}$ is independent from the hash value of any other $n$-gram $x_{2}$. That is, we have $P(h(x_{1})=y\land h(x_{2})=z)=P(h(x_{1})=y)P(h(x_{2})=z)=1/4^{L}$ for all distinct $n$-grams $x_{1}$, $x_{2}$, and all hash values $y$, $z$ with $x_{1}\not=x_{2}$. Pairwise independence implies uniformity. We refer to a particular hash function $h\in\mathcal{H}$ as “uniform” or “a pairwise independent hash function” when the family in question can be inferred from the context. Moreover, the idea of pairwise independence can be generalized: a family of hash functions $\mathcal{H}$ is _$k$ -wise independent_ if given distinct $x_{1},\ldots,x_{k}$ and given $h$ selected uniformly at random from $\mathcal{H}$, then $P(h(x_{1})=y_{1}\land\cdots\land h(x_{k})=y_{k})=1/2^{kL}$. Note that $k$-wise independence implies $k-1$-wise independence and uniformity. (Fully) independent families are $k$-wise independent for arbitrarily large $k$. For applications, non-independent families may fare as well as fully independent families if the entropy of the data source is sufficiently high [16]. A hash function $h$ is _recursive_ [17]—or rolling[18]—if there is a function $F$ computing the hash value of the $n$-gram $x_{2}\ldots x_{n+1}$ from the hash value of the preceding $n$-gram ($x_{1}\ldots x_{n}$) and the values of $x_{1}$ and $x_{n+1}$. That is, we have $h(x_{2},\ldots,x_{n+1})=F(h(x_{1},\ldots,x_{n}),x_{1},x_{n+1}).$ Ideally, we could compute function $F$ in time $O(L)$ and not, for example, in time $O(Ln)$. The main contributions of this paper are: * 1. a proof that recursive hashing is no more than pairwise independent (§ 3); * 2. a proof that randomized Karp-Rabin can be uniform but never pairwise independent (§ 5); * 3. a proof that hashing by irreducible polynomials is pairwise independent (§ 7); * 4. a proof that hashing by cyclic polynomials is not even uniform (§ 9); * 5. a proof that hashing by cyclic polynomials is pairwise independent—after ignoring $n-1$ consecutive bits (§ 10). We conclude with an experimental section where we show that hashing by cyclic polynomials is faster than hashing by irreducible polynomials. Table 1 summarizes the algorithms presented. Table 1: A summary of the hashing function presented and their properties. For General and Cyclic, we require $L\geq n$. To make Cyclic pairwise independent, we need to discard some bits—the resulting scheme is not formally recursive. Randomized Karp-Rabin is uniform under some conditions. name | cost per $n$-gram | independence | memory use ---|---|---|--- non-recursive 3-wise (§ 4) | $O(Ln)$ | 3-wise | $O(nL|\Sigma|)$ Randomized Karp-Rabin (§ 5) | $O(L\log L2^{O(\log^{*}L)})$ | uniform | $O(L|\Sigma|)$ General (§ 7) | $O(Ln)$ | pairwise | $O(L|\Sigma|)$ RAM-Buffered General (§ 8) | $O(L)$ | pairwise | $O(L|\Sigma|+L2^{n})$ Cyclic (§ 9) | $O(L+n)$ | pairwise (§ 10) | $O((L+n)|\Sigma|)$ ## 2 Trailing-zero independence Some randomized algorithms [14, 15] merely require that the number of trailing zeroes be independent. For example, to estimate the number of distinct $n$-grams in a large document without enumerating them, we merely have to compute maximal numbers of leading zeroes $k$ among hash values [19]. Naïvely, we may estimate that if a hash value with $k$ leading zeroes is found, we have $\approx 2^{k}$ distinct $n$-grams. Such estimates might be useful because the number of distinct $n$-grams grows large with $n$: Shakespeare’s First Folio [20] has over 3 million distinct 15-grams. Formally, let $\textrm{zeros}(x)$ return the number of trailing zeros (0,1,…,$L$) of $x$, where $\textrm{zeros}(0)=L$. We say $h$ is _$k$ -wise trailing-zero independent_ if $P(\textrm{zeros}(h(x_{1}))\geq j_{1}\wedge\textrm{zeros}(h(x_{2}))\geq j_{2}\wedge\ldots\wedge\textrm{zeros}(h(x_{k}))\geq j_{k})=2^{-j_{1}-j_{2}-\cdots-j_{k}}$, for $j_{i}=0,1,\ldots,L$. If $h$ is $k$-wise independent, it is $k$-wise trailing-zero independent. The converse is not true. If $h$ is a $k$-wise independent function, consider $g\circ h$ where $g$ makes zero all bits before the rightmost 1 (e.g., $g(0101100)=0000100$). Hash $g\circ h$ is $k$-wise trailing-zero independent but not even uniform (consider that $P(g=0001)=8P(g=1000)$). ## 3 Recursive hash functions are no more than pairwise independent Not only are recursive hash functions limited to pairwise independence: they cannot be 3-wise _trailing-zero_ independent. ###### Proposition 1. There is no 3-wise trailing-zero independent hashing function that is recursive. ###### Proof Consider the ($n+2$)-gram $\texttt{a}^{n}\texttt{bb}$. Suppose $h$ is recursive and $3$-wise trailing-zero independent, then $\displaystyle P\left(\textrm{zeros}(h(\texttt{a},\ldots,\texttt{a}))\geq L\bigwedge\right.$ $\displaystyle\left.\textrm{zeros}(h(\texttt{a},\ldots,\texttt{a},\texttt{b}))\geq L\bigwedge\textrm{zeros}(h(\texttt{a},\ldots,\texttt{a},\texttt{b},\texttt{b}))\geq L\right)$ $\displaystyle=$ $\displaystyle P\left(h(\texttt{a},\ldots,\texttt{a})=0\bigwedge F(0,\texttt{a},\texttt{b})=0\bigwedge F(0,\texttt{a},\texttt{b})=0\right)$ $\displaystyle=$ $\displaystyle P\left(h(\texttt{a},\ldots,\texttt{a})=0\bigwedge F(0,\texttt{a},\texttt{b})=0\right)$ $\displaystyle=$ $\displaystyle P\left(\textrm{zeros}(h(\texttt{a},\ldots,\texttt{a}))\geq L\bigwedge\textrm{zeros}(h(\texttt{a},\ldots,\texttt{a},\texttt{b}))\geq L\right)$ $\displaystyle=$ $\displaystyle 2^{-2L}\mbox{\ by trailing-zero pairwise independence}$ $\displaystyle\not=$ $\displaystyle 2^{-3L}\mbox{\ as required by trailing-zero 3-wise independence.}$ Hence, we have a contradiction and no such $h$ exists. ## 4 A non-recursive 3-wise independent hash function A trivial way to generate an independent hash is to assign a random integer in $[0,2^{L})$ to each new value $x$. Unfortunately, this requires as much processing and storage as a complete indexing of all values. However, in a multidimensional setting this approach can be put to good use. Suppose that we have tuples in $K_{1}\times K_{2}\times\cdots\times K_{n}$ such that $|K_{i}|$ is small for all $i$. We can construct independent hash functions $h_{i}:K_{i}\rightarrow[0,2^{L})$ for all $i$ and combine them. The hash function $h(x_{1},x_{2},\ldots,x_{n})=h_{1}(x_{1})\oplus h_{2}(x_{2})\oplus\cdots\oplus h_{n}(x_{n})$ is then 3-wise independent ($\oplus$ is the “exclusive or” function, XOR). In time $O(\sum_{i=1}^{n}|K_{i}|)$, we can construct the hash function by generating $\sum_{i=1}^{n}|K_{i}|$ random numbers and storing them in a look-up table. With constant-time look-up, hashing an $n$-gram thus takes $O(Ln)$ time. Algorithm 1 is an application of this idea to $n$-grams. 0: $n$ $L$-bit hash functions $h_{1},h_{1},\ldots,h_{n}$ over $\Sigma$ from an independent hash family 1: $s\leftarrow$ empty FIFO structure 2: for each character $c$ do 3: append $c$ to $s$ 4: if length($s$)$=n$ then 5: yield $h_{1}(s_{1})\oplus h_{2}(s_{2})\oplus\ldots\oplus h_{n}(s_{n})$ {The yield statement returns the value, without terminating the algorithm.} 6: remove oldest character from $s$ 7: end if 8: end for Algorithm 1 The (non-recursive) 3-wise independent family. This new family is not 4-wise independent for $n>1$. Consider the $n$-grams ac,ad, bc, bd. The XOR of their four hash values is zero. However, the family is 3-wise independent. ###### Proposition 2. The family of hash functions $h(x)=h_{1}(x_{1})\oplus h_{2}(x_{2})\oplus\ldots\oplus h_{n}(x_{n})$, where the $L$-bit hash functions $h_{1},\ldots,h_{n}$ are taken from an independent hash family, is 3-wise independent. ###### Proof Consider any 3 distinct $n$-grams: $x^{(1)}=x_{1}^{(1)}\ldots x_{n}^{(1)}$, $x^{(2)}=x_{1}^{(2)}\ldots x_{n}^{(2)}$, and $x^{(3)}=x_{1}^{(3)}\ldots x_{n}^{(3)}$. Because the $n$-grams are distinct, at least one of two possibilities holds: Case A For some $i\in\\{1,\ldots,n\\}$, the three values $x_{i}^{(1)},x_{i}^{(2)},x_{i}^{(3)}$ are distinct. Write $\chi_{j}=h_{i}(x_{i}^{(j)})$ for $j=1,2,3$. For example, consider the three 1-grams: a,b,c. Case B (Up to a reordering of the three $n$-grams.) There are two values $i,j\in\\{1,\ldots,n\\}$ such that $x_{i}^{(1)}$ is distinct from the two identical values $x_{i}^{(2)},x_{i}^{(3)}$, and such that $x_{j}^{(2)}$ is distinct from the two identical values $x_{i}^{(1)},x_{i}^{(3)}$. Write $\chi_{1}=h_{i}(x_{i}^{(1)})$, $\chi_{2}=h_{j}(x_{j}^{(2)})$, and $\chi_{3}=h_{i}(x_{i}^{(3)})$. For example, consider the three 2-grams: ad,bc,bd. Recall that the XOR operation is invertible: $a\oplus b=c$ if and only if $a=b\oplus c$. We prove 3-wise independence for cases A and B. ### Case A Write $f^{(i)}=h(x^{(i)})\oplus\chi_{i}$ for $i=1,2,3$. We have that the values $\chi_{1},\chi_{2},\chi_{3}$ are mutually independent, and they are independent from the values $f^{(1)},f^{(2)},f^{(3)}$222The values $f^{(1)},f^{(2)},f^{(3)}$ are not necessarily mutually independent.: $\displaystyle P\left(\bigwedge_{i=1}^{3}\chi_{i}=y_{i}\land\bigwedge_{i=1}^{3}f^{(i)}=y^{\prime}_{i}\right)=\prod_{i=1}^{3}P(\chi_{i}=y_{i})P\left(\bigwedge_{i=1}^{3}f^{(i)}=y^{\prime}_{i}\right)$ for all values $y_{i},y^{\prime}_{i}$. Hence, we have $\displaystyle P\left(h(x^{(1)})=z^{(1)}\bigwedge h(x^{(2)})=z^{(2)}\bigwedge h(x^{(3)})=z^{(3)}\right)$ $\displaystyle=$ $\displaystyle P\left(\chi_{1}=z^{(1)}\oplus f^{(1)})\bigwedge\chi_{2}=z^{(2)}\oplus f^{(2)}\bigwedge\chi_{3}=z^{(3)}\oplus f^{(3)}\right)$ $\displaystyle=$ $\displaystyle\sum_{\eta,\eta^{\prime},\eta^{\prime\prime}}P\left(\chi_{1}=z^{(1)}\oplus\eta\bigwedge\chi_{2}=z^{(2)}\oplus\eta^{\prime}\bigwedge\chi_{3}=z^{(3)}\oplus\eta^{\prime\prime}\right)\times$ $\displaystyle P(f^{(1)}=\eta\land f^{(2)}=\eta^{\prime}\land f^{(3)}=\eta^{\prime\prime})$ $\displaystyle=$ $\displaystyle\sum_{\eta,\eta^{\prime},\eta^{\prime\prime}}\frac{1}{2^{3L}}P(f^{(1)}=\eta\land f^{(2)}=\eta^{\prime}\land f^{(3)}=\eta^{\prime\prime})$ $\displaystyle=$ $\displaystyle\frac{1}{2^{3L}}.$ Thus, in this case, the hash values are 3-wise independent. ### Case B Write $f^{(1)}=h(x^{(1)})\oplus\chi_{1}$, $f^{(2)}=h(x^{(2)})\oplus\chi_{2}\oplus\chi_{3}$, $f^{(3)}=h(x^{(3)})\oplus\chi_{3}$. Again, the values $\chi_{1},\chi_{2},\chi_{3}$ are mutually independent, and independent from the values $f^{(1)},f^{(2)},f^{(3)}$. We have $\displaystyle P\left(h(x^{(1)})=z^{(1)}\bigwedge h(x^{(2)})=z^{(2)}\bigwedge h(x^{(3)})=z^{(3)}\right)$ $\displaystyle=$ $\displaystyle P\left(\chi_{1}=z^{(1)}\oplus f^{(1)})\bigwedge\chi_{2}\oplus\chi_{3}=z^{(2)}\oplus f^{(2)}\bigwedge\chi_{3}=z^{(3)}\oplus f^{(3)}\right)$ $\displaystyle=$ $\displaystyle P\left(\chi_{1}=z^{(1)}\oplus f^{(1)})\bigwedge\chi_{2}=z^{(2)}\oplus f^{(2)}\oplus z^{(3)}\oplus f^{(3)}\bigwedge\chi_{3}=z^{(3)}\oplus f^{(3)}\right)$ $\displaystyle=$ $\displaystyle\sum_{\eta,\eta^{\prime},\eta^{\prime\prime}}P\left(\chi_{1}=z^{(1)}\oplus\eta\bigwedge\chi_{2}=z^{(2)}\oplus z^{(3)}\oplus\eta^{\prime}\oplus\eta^{\prime\prime}\bigwedge\chi_{3}=z^{(3)}\oplus\eta^{\prime\prime}\right)\times$ $\displaystyle P(f^{(1)}=\eta\land f^{(2)}=\eta^{\prime}\land f^{(3)}=\eta^{\prime\prime})$ $\displaystyle=$ $\displaystyle\sum_{\eta,\eta^{\prime},\eta^{\prime\prime}}\frac{1}{2^{3L}}P(f^{(1)}=\eta\land f^{(2)}=\eta^{\prime}\land f^{(3)}=\eta^{\prime\prime})$ $\displaystyle=$ $\displaystyle\frac{1}{2^{3L}}.$ This concludes the proof. ## 5 Randomized Karp-Rabin is not independent One of the most common recursive hash functions is commonly associated with the Karp-Rabin string-matching algorithm [21]. Given an integer $B$, the hash value over the sequence of integers $x_{1},x_{2},\ldots,x_{n}$ is $\sum_{i=1}^{n}x_{i}B^{n-i}$. A variation of the Karp-Rabin hash method is “Hashing by Power-of-2 Integer Division” [17], where $h(x_{1},\ldots,x_{n})=\sum_{i=1}^{n}x_{i}B^{n-i}\bmod{2^{L}}$. In particular, the hashcode method of the Java String class uses this approach, with $L=32$ and $B=31$ [22]. A widely used textbook [23, p. 157] recommends a similar Integer-Division hash function for strings with $B=37$. Since such Integer-Division hash functions are recursive, quickly computed, and widely used, it is interesting to seek a randomized version of them. Assume that $h_{1}$ is a random hash function over symbols uniform in $[0,2^{L})$, then define $h(x_{1},\ldots,x_{n})=B^{n-1}h_{1}(x_{1})+B^{n-2}h_{1}(x_{2})+\cdots+h_{1}(x_{n})\bmod{2^{L}}$ for some fixed integer $B$. We choose $B=37$ (calling the resulting randomized hash “ID37;” see Algorithm 2). Our algorithm computes each hash value in time O($M(L)$), where $M(L)$ is the cost of multiplying two $L$-bit integers. (We precompute the value $B^{n}\bmod{2^{L}}$.) In many practical cases, $L$ bits can fit into a single machine word and the cost of multiplication can be considered constant. In general, $M(L)$ is in $O(L\log L2^{O(\log^{*}L)})$ [24]. 0: an $L$-bit hash function $h_{1}$ over $\Sigma$ from an independent hash family 1: $B\leftarrow 37$ 2: $s\leftarrow$ empty FIFO structure 3: $x\leftarrow 0$ ($L$-bit integer) 4: $z\leftarrow 0$ ($L$-bit integer) 5: for each character $c$ do 6: append $c$ to $s$ 7: $x\leftarrow Bx-B^{n}z+h_{1}(c)\bmod{2^{L}}$ 8: if length($s$)$=n$ then 9: yield $x$ 10: remove oldest character $y$ from $s$ 11: $z\leftarrow h_{1}(y)$ 12: end if 13: end for Algorithm 2 The recursive ID37 family (Randomized Karp-Rabin). The randomized Integer-Division functions mapping $n$-grams to $[0,2^{L})$ are not pairwise independent. However, for some values of $B$ and $n$, they are uniform. ###### Proposition 3. Randomized Integer-Division hashing is not uniform for $n$-grams, if $n$ is even and $B$ is odd. Otherwise, it is uniform for $B$ even and any $n$, or $B$ odd and $n$ odd. However, there is no value of $B$ for which it is pairwise independent when $n\geq 2$. ###### Proof For $B$ odd, we see that $P(h(\texttt{a}^{2k})=0)>2^{-L}$ since $h(\texttt{a}^{2k})=h_{1}(\texttt{a})(B^{0}(1+B)+B^{2}(1+B)+\cdots+B^{2k-2}(1+B))\bmod 2^{L}$ and since $(1+B)$ is even, we have $P(h(\texttt{a}^{2k})=0)\geq P(h_{1}(x_{1})=2^{L-1}\lor h_{1}(x_{1})=0)=1/2^{L-1}$. Hence, for $B$ odd and $n$ even, we do not have uniformity. Suppose that $B$ and $n$ are both odd. Consider any string $x_{1},x_{2},\ldots,x_{n}$. We can find a character value $x_{j}$ which is repeated an odd number of times in the string. Let $I$ be the set of indexes $i$ such that $x_{i}=x_{j}$. We have that the equation $h(x_{1},x_{2},\ldots,x_{n})=y$ is equivalent to $(\sum_{i=1}^{n}B^{n-i}h_{1}(x_{i}))=y$. We can rewrite it as $(\sum_{i\in I}B^{n-i})h_{1}(x_{j})=y-(\sum_{i\not\in I}B^{n-i}h_{1}(x_{i}))$. There is a unique solution $h_{1}(x_{j})$ to this equation because $(\sum_{i\in I}B^{n-i})$ is odd: the sum of an odd number of odd integers is an odd integer. Hence, we have uniformity when $B$ and $n$ are odd. Consider $B$ even. Consider any string $x_{1},x_{2},\ldots,x_{n}$. We are interested in the last character $x_{n}$. It might be repeated several times in the string. Let $I$ be the set of indexes $i$ such that $x_{i}=x_{n}$. We have that $h(x_{1},x_{2},\ldots,x_{n})=y$ is equivalent to $(\sum_{i=1}^{n}B^{n-i}h_{1}(x_{i}))=y$ or $(\sum_{i\in I}B^{n-i})h_{1}(x_{n})=y-(\sum_{i\not\in I}B^{n-i}h_{1}(x_{i}))$. We want to show that there is a unique solution $h_{1}(x_{n})$ to this equation. This follows because we have that $(\sum_{i\in I}B^{n-i})$ is an odd number because $B$ is even and $n\in I$. Hence, we have uniformity when $B$ is even. To show it is not pairwise independent, first suppose that $B$ is odd. For any string $\beta$ of length $n-2$, consider $n$-grams $w_{1}=\beta\texttt{a}\texttt{a}$ and $w_{2}=\beta\texttt{b}\texttt{b}$ for distinct $\texttt{a},\texttt{b}\in\Sigma$. Then $P(h(w_{1})=h(w_{2}))=P(B^{2}h(\beta)+Bh_{1}(\texttt{a})+h_{1}(\texttt{a})=B^{2}h(\beta)+Bh_{1}(\texttt{b})+h_{1}(\texttt{b})\bmod 2^{L})=P((1+B)(h_{1}(\texttt{a})-h_{1}(\texttt{b}))\bmod 2^{L}=0)\geq P(h_{1}(\texttt{a})-h_{1}(\texttt{b})=0)+P(h_{1}(\texttt{a})-h_{1}(\texttt{b})=2^{L-1})$. Because $h_{1}$ is independent, $P(h_{1}(\texttt{a})-h_{1}(\texttt{b})=0)=\sum_{c\in[0,2^{L})}P(h_{1}(\texttt{a})=c)P(h_{1}(\texttt{b})=c)=\sum_{c\in[0,2^{L})}1/4^{L}=1/2^{L}$. Moreover, $P(h_{1}(\texttt{a})-h_{1}(\texttt{b})=2^{L-1})>0$. Thus, we have that $P(h(w_{1})=h(w_{2}))>1/2^{L}$ which contradicts pairwise independence. Second, if $B$ is even, a similar argument shows $P(h(w_{3})=h(w_{4}))>1/2^{L}$, where $w_{3}=\beta\texttt{a}\texttt{a}$ and $w_{4}=\beta\texttt{b}\texttt{a}$. $P(h(\texttt{a},\texttt{a})=h(\texttt{b},\texttt{a}))=P(Bh_{1}(\texttt{a})+h_{1}(\texttt{a})=Bh_{1}(\texttt{b})+h_{1}(\texttt{a})\bmod 2^{L})=P(B(h_{1}(\texttt{a})-h_{1}(\texttt{b}))\bmod 2^{L}=0)\geq P(h_{1}(\texttt{a})-h_{1}(\texttt{b})=0)+P(h_{1}(\texttt{a})-h_{1}(\texttt{b})=2^{L-1})>1/2^{L}$. Hence, as long as we consider strings of length $n>1$ and an alphabet $\Sigma$ containing at least two distinct characters, we can find two strings with a collision probability greater than $1/2^{L}$ whether $B$ is even or odd. A weaker condition than pairwise independence is 2-universality: a family is 2-universal if $P(h(x_{1})=h(x_{2}))\leq 1/2^{L}$ [16]. As a consequence of this proof, Randomized Integer-Division is not even 2-universal. These results also hold for any Integer-Division hash where the modulo is by an even number, not necessarily a power of 2. ## 6 Generating hash families from polynomials over Galois fields A practical form of hashing using the binary Galois field GF(2) is called “Recursive Hashing by Polynomials” and has been attributed to Kubina by Cohen [17]. GF(2) contains only two values (1 and 0) with the addition (and hence subtraction) defined by XOR, $a+b=a\oplus b$ and the multiplication by AND, $a\times b=a\wedge b$. $\textrm{GF}(2)[x]$ is the vector space of all polynomials with coefficients from GF(2). Any integer in binary form (e.g., $c=1101$) can thus be interpreted as an element of $\textrm{GF}(2)[x]$ (e.g., $c=x^{3}+x^{2}+1$). If $p(x)\in\textrm{GF}(2)[x]$, then $\textrm{GF}(2)[x]/p(x)$ can be thought of as $\textrm{GF}(2)[x]$ modulo $p(x)$. As an example, if $p(x)=x^{2}$, then $\textrm{GF}(2)[x]/p(x)$ is the set of all linear polynomials. For instance, $x^{3}+x^{2}+x+1=x+1\bmod{x^{2}}$ since, in $\textrm{GF}(2)[x]$, $(x+1)+x^{2}(x+1)=x^{3}+x^{2}+x+1$. As a summary, we compute operations over $\textrm{GF}(2)[x]/p(x)$—where $p(x)$ is of degree $L$—as follows: * 1. the polynomial $\sum_{i=0}^{L-1}q_{i}x^{i}$ is represented as the $L$-bit integer $\sum_{i=0}^{L-1}q_{i}2^{i}$; * 2. subtraction or addition of two polynomials is the XOR of their $L$-bit integers; * 3. multiplication of a polynomial $\sum_{i=0}^{L}q_{i}x^{i}$ by the monomial $x$ is represented either as $\sum_{i=0}^{L-1}q_{i}x^{i+1}$ if $q_{L-1}=0$ or as $p(x)+\sum_{i=0}^{L-1}q_{i}x^{i+1}$ otherwise. In other words, if the value of the last bit is 1, we merely apply a binary left shift, otherwise, we apply a binary left shift immediately followed by an XOR with the integer representing $p(x)$. In either case, we get an $L$-bit integer. Hence, merely with the XOR operation, the binary left shift, and a way to evaluate the value of the last bit, we can compute all necessary operations over $\textrm{GF}(2)[x]/p(x)$ using integers. Consider a hash function $h_{1}$ over characters taken from some independent family. Interpreting $h_{1}$ hash values as polynomials in $\textrm{GF}(2)[x]/p(x)$, and with the condition that $\textrm{degree}(p(x))\geq n$, we define a hash function as $h(a_{1},a_{2},\cdots,a_{n})=h_{1}(a_{1})x^{n-1}+h_{1}(a_{2})x^{n-2}+\cdots+h_{1}(a_{n})$. It _is_ recursive over the sequence $h_{1}(a_{i})$. The combined hash can be computed by reusing previous hash values: $h(a_{2},a_{3},\ldots,a_{n+1})=xh(a_{1},a_{2},\ldots,a_{n})-h_{1}(a_{1})x^{n}+h_{1}(a_{n+1}).$ Depending on the choice of the polynomial $p(x)$ we get different hashing schemes, including General and Cyclic, which are presented in the next two sections. ## 7 Recursive hashing by irreducible polynomials is pairwise independent 0: an $L$-bit hash function $h_{1}$ over $\Sigma$ from an independent hash family; an irreducible polynomial $p$ of degree $L$ in $\textrm{GF}(2)[x]$ 1: $s\leftarrow$ empty FIFO structure 2: $x\leftarrow 0$ ($L$-bit integer) 3: $z\leftarrow 0$ ($L$-bit integer) 4: for each character $c$ do 5: append $c$ to $s$ 6: $x\leftarrow\textrm{shift}(x)$ 7: $z\leftarrow\textrm{shift}^{n}(z)$ 8: $x\leftarrow x\oplus z\oplus h_{1}(c)$ 9: if length($s$)$=n$ then 10: yield $x$ 11: remove oldest character $y$ from $s$ 12: $z\leftarrow h_{1}(y)$ 13: end if 14: end for 1: function shift 2: input $L$-bit integer $x$ 3: shift $x$ left by 1 bit, storing result in an $L+1$-bit integer $x^{\prime}$ 4: if leftmost bit of $x^{\prime}$ is 1 then 5: $x^{\prime}\leftarrow x^{\prime}\oplus p$ 6: end if 7: {leftmost bit of $x^{\prime}$ is thus always 0} 8: return rightmost $L$ bits of $x^{\prime}$ Algorithm 3 The recursive General family. Table 2: Some irreducible polynomials over $\textrm{GF}(2)[x]$ degree | polynomial ---|--- 10 | $1+x^{3}+x^{10}$ 15 | $1+x+x^{15}$ 20 | $1+x^{3}+x^{20}$ 25 | $1+x^{3}+x^{25}$ 30 | $1+x+x^{4}+x^{6}+x^{30}$ We can choose $p(x)$ to be an irreducible polynomial of degree $L$ in $\textrm{GF}(2)[x]$: an irreducible polynomial cannot be factored into nontrivial polynomials (see Table 2). The resulting hash is called General (see Algorithm 3). The main benefit of setting $p(x)$ to be an irreducible polynomial is that $\textrm{GF}(2)[x]/p(x)$ is a field; in particular, it is impossible that $p_{1}(x)p_{2}(x)=0\bmod{p(x)}$ unless either $p_{1}(x)=0$ or $p_{2}(x)=0$. The field property allows us to prove that the hash function is pairwise independent. ###### Lemma 1. General is pairwise independent. ###### Proof If $p(x)$ is irreducible, then any non-zero $q(x)\in\textrm{GF}(2)[x]/p(x)$ has an inverse, noted $q^{-1}(x)$ since $\textrm{GF}(2)[x]/p(x)$ is a field. Interpret hash values as polynomials in $\textrm{GF}(2)[x]/p(x)$. Firstly, we prove that General is uniform. In fact, we show a stronger result: $P(q_{1}(x)h_{1}(a_{1})+q_{2}(x)h_{1}(a_{2})+\cdots+q_{n}(x)h_{1}(a_{n})=y)=1/2^{L}$ for any polynomials $q_{i}$ where at least one is different from zero. The result follows by induction on the number of non-zero polynomials: it is clearly true where there is a single non-zero polynomial $q_{i}(x)$, since $q_{i}(x)h_{1}(a_{i})=y\iff q_{i}^{-1}(x)q_{i}(x)h_{1}(a_{i})=q_{i}^{-1}(x)y$. Suppose it is true up to $k-1$ non-zero polynomials and consider a case where we have $k$ non-zero polynomials. Assume without loss of generality that $q_{1}(x)\neq 0$, we have $P(q_{1}(x)h_{1}(a_{1})+q_{2}(x)h_{1}(a_{2})+\cdots+q_{n}(x)h_{1}(a_{n})=y)=P(h_{1}(a_{1})=q_{1}^{-1}(x)(y-q_{2}(x)h_{1}(a_{2})-\cdots- q_{n}(x)h_{1}(a_{n})))=\sum_{y^{\prime}}P(h_{1}(a_{1})=q_{1}^{-1}(x)(y-y^{\prime}))P(q_{2}(x)h_{1}(a_{2})+\cdots+q_{n}(x)h_{1}(a_{n})=y^{\prime})=\sum_{y^{\prime}}\frac{1}{2^{L}}\frac{1}{2^{L}}=\frac{1}{2^{L}}$ by the induction argument. Hence the uniformity result is shown. Consider two distinct sequences $a_{1},a_{2},\ldots,a_{n}$ and $a^{\prime}_{1},a^{\prime}_{2},\ldots,a^{\prime}_{n}$. Write $H_{a}=h(a_{1},a_{2},\ldots,a_{n})$ and $H_{a^{\prime}}=h(a^{\prime}_{1},a^{\prime}_{2},\ldots,a^{\prime}_{n})$. We have that $P(H_{a}=y\land H_{a^{\prime}}=y^{\prime})=P(H_{a}=y|H_{a^{\prime}}=y^{\prime})P(H_{a^{\prime}}=y^{\prime})$. Hence, to prove pairwise independence, it suffices to show that $P(H_{a}=y|H_{a^{\prime}}=y^{\prime})=1/2^{L}$. Suppose that $a_{i}=a^{\prime}_{j}$ for some $i,j$; if not, the result follows since by the (full) independence of the hashing function $h_{1}$, the values $H_{a}$ and $H_{a^{\prime}}$ are independent. Write $q(x)=-(\sum_{k|a_{k}=a_{i}}x^{n-k})(\sum_{k|a^{\prime}_{k}=a^{\prime}_{j}}x^{n-k})^{-1}$, then $H_{a}+q(x)H_{a^{\prime}}$ is independent from $a_{i}=a^{\prime}_{j}$ (and $h_{1}(a_{i})=h_{1}(a^{\prime}_{j})$). In $H_{a}+q(x)H_{a^{\prime}}$, only hashed values $h_{1}(a_{k})$ for $a_{k}\neq a_{i}$ and $h_{1}(a^{\prime}_{k})$ for $a^{\prime}_{k}\neq a^{\prime}_{j}$ remain: label them $h_{1}(b_{1}),\ldots,h_{1}(b_{m})$. The result of the substitution can be written $H_{a}+q(x)H_{a^{\prime}}=\sum_{k}q_{k}(x)h_{1}(b_{k})$ where $q_{k}(x)$ are polynomials in $\textrm{GF}(2)[x]/p(x)$. All $q_{k}(x)$ are zero if and only if $H_{a}+q(x)H_{a^{\prime}}=0$ for all values of $h_{1}(a_{1}),\ldots,h_{1}(a_{n})$ and $h_{1}(a^{\prime}_{1}),\ldots,h_{1}(a^{\prime}_{n})$ (but notice that the value $h_{1}(a_{i})=h_{1}(a^{\prime}_{j})$ is irrelevant); in particular, it must be true when $h_{1}(a_{k})=1$ and $h_{1}(a^{\prime}_{k})=1$ for all $k$, hence $(x^{n}+\cdots+x+1)+q(x)(x^{n}\ldots+x+1)=0\Rightarrow q(x)=-1$. Thus, all $q_{k}(x)$ are zero if and only if $H_{a}=H_{a^{\prime}}$ for all values of $h_{1}(a_{1}),\ldots,h_{1}(a_{n})$ and $h_{1}(a^{\prime}_{1}),\ldots,h_{1}(a^{\prime}_{n})$ which only happens if the sequences $a$ and $a^{\prime}$ are identical. Hence, not all $q_{k}(x)$ are zero. Write $H_{y^{\prime},a^{\prime}}=(\sum_{k|a^{\prime}_{k}=a^{\prime}_{j}}x^{n-k})^{-1}(y^{\prime}-\sum_{k|a^{\prime}_{k}\neq a^{\prime}_{j}}x^{n-k}h_{1}(a^{\prime}_{k}))$. On the one hand, the condition $H_{a^{\prime}}=y^{\prime}$ can be rewritten as $h_{1}(a^{\prime}_{j})=H_{y^{\prime},a^{\prime}}$. On the other hand, $H_{a}+q(x)H_{a^{\prime}}=y+q(x)y^{\prime}$ is independent from $h_{1}(a^{\prime}_{j})=h_{1}(a_{i})$. Because $P(h_{1}(a^{\prime}_{j})=H_{y^{\prime},a^{\prime}})=1/2^{L}$ irrespective of $y^{\prime}$ and $h_{1}(a^{\prime}_{k})$ for $k\in\\{k|a^{\prime}_{k}\neq a^{\prime}_{j}\\}$, then $P(h_{1}(a^{\prime}_{j})=H_{y^{\prime},a^{\prime}}|H_{a}+q(x)H_{a^{\prime}}=y+q(x)y^{\prime})=P(h_{1}(a^{\prime}_{j})=H_{y^{\prime},a^{\prime}})$ which implies that $h_{1}(a^{\prime}_{j})=H_{y^{\prime},a^{\prime}}$ and $H_{a}+q(x)H_{a^{\prime}}=y+q(x)y^{\prime}$ are independent. Hence, we have $\displaystyle P(H_{a}=y|H_{a^{\prime}}=y^{\prime})$ $\displaystyle=$ $\displaystyle P(H_{a}+q(x)H_{a^{\prime}}=y+q(x)y^{\prime}|h_{1}(a^{\prime}_{j})=H_{y^{\prime},a^{\prime}})$ $\displaystyle=$ $\displaystyle P(H_{a}+q(x)H_{a^{\prime}}=y+q(x)y^{\prime})$ $\displaystyle=$ $\displaystyle P(\sum_{k}q_{k}(x)h_{1}(b_{k})=y+q(x)y^{\prime})$ and by the earlier uniformity result, this last probability is equal to $1/2^{L}$. This concludes the proof. ## 8 Trading memory for speed: RAM-Buffered General Unfortunately, General—as computed by Algorithm 3—requires $O(nL)$ time per $n$-gram. Indeed, shifting a value $n$ times in $\textrm{GF}(2)[x]/p(x)$ requires $O(nL)$ time. However, if we are willing to trade memory usage for speed, we can precompute these shifts. We call the resulting scheme RAM- Buffered General. ###### Lemma 2. Pick any $p(x)$ in $\textrm{GF}(2)[x]$. The degree of $p(x)$ is $L$. Represent elements of $\textrm{GF}(2)[x]/p(x)$ as polynomials of degree at most $L-1$. Given any $h$ in $\textrm{GF}(2)[x]/p(x)$. we can compute $x^{n}h$ in O($L$) time given an $O(L2^{n})$-bit memory buffer. ###### Proof Write $h$ as $\sum_{i=0}^{L-1}q_{i}x^{i}$. Divide $h$ into two parts, $h^{(1)}=\sum_{i=0}^{L-n-1}q_{i}x^{i}$ and $h^{(2)}=\sum_{i=L-n}^{L-1}q_{i}x^{i}$, so that $h=h^{(1)}+h^{(2)}$. Then $x^{n}h=x^{n}h^{(1)}+x^{n}h^{(2)}$. The first part, $x^{n}h^{(1)}$ is a polynomial of degree at most $L-1$ since the degree of $h^{(1)}$ is at most $L-1-n$. Hence, $x^{n}h^{(1)}$ as an $L$-bit value is just $q_{L-n-1}q_{L-n-2}\ldots q_{0}0\ldots 0$. which can be computed in time $O(L)$. So, only the computation of $x^{n}h^{(2)}$ is possibly more expensive than $O(L)$ time, but $h^{(2)}$ has only $n$ terms as a polynomial (since the first $L-n$ terms are always zero). Hence, if we precompute $x^{n}h^{(2)}$ for all $2^{n}$ possible values of $h^{(2)}$, and store them in an array with $O(L)$ time look-ups, we can compute $x^{n}h$ as an $L$-bit value in $O(L)$ time. When $n$ is large, this precomputation requires excessive space and precomputation time. Fortunately, we can trade back some speed for memory. Consider the proof of Lemma 2. Instead of precomputing the shifts of all $2^{n}$ possible values of $h^{(2)}$ using an array of $2^{n}$ entries, we can further divide $h^{(2)}$ into $K$ parts. For simplicity, assume that the integer $K$ divides $n$. The $K$ parts $h^{(2,1)},\ldots,h^{(2,K)}$ are made of the first $n/K$ bits, the next $n/K$ bits and so on. Because $x^{n}h^{(2)}=\sum_{i=1}^{K}x^{n}h^{(2,i)}$, we can shift $h^{(2)}$ by $n$ in $O(KL)$ operations using $K$ arrays of $2^{n/K}$ entries. To summarize, we have a time complexity of $O(KL)$ per $n$-gram using $O(L|\Sigma|+LK2^{n/K})$ bits. We implemented the case $K=2$. ## 9 Recursive hashing by cyclic polynomials is not even uniform Choosing $p(x)=x^{L}+1$ for $L\geq n$, for any polynomial $q(x)=\sum_{i=0}^{L-1}q_{i}x^{i}$, we have $x^{i}q(x)=x^{i}(q_{L-1}x^{L-1}+\cdots+q_{1}x+q_{0})=q_{L-i-1}x^{L-i-2}+\cdots+q_{L-i+1}x+q_{L-i}.$ Thus, we have that multiplication by $x^{i}$ is a bitwise rotation, a cyclic left shift—which can be computed in $O(L)$ time. The resulting hash (see Algorithm 4) is called Cyclic. It requires only $O(L)$ time per hash value. _Empirically_ , Cohen showed that Cyclic is uniform [17]. In contrast, we show that it is not formally uniform: 0: an $L$-bit hash function $h_{1}$ over $\Sigma$ from an independent hash family 1: $s\leftarrow$ empty FIFO structure 2: $x\leftarrow 0$ ($L$-bit integer) 3: $z\leftarrow 0$ ($L$-bit integer) 4: for each character $c$ do 5: append $c$ to $s$ 6: rotate $x$ left by 1 bit 7: rotate $z$ left by n bits 8: $x\leftarrow x\oplus z\oplus h_{1}(c)$ 9: if length($s$)$=n$ then 10: yield $x$ 11: remove oldest character $y$ from $s$ 12: $z\leftarrow h_{1}(y)$ 13: end if 14: end for Algorithm 4 The recursive Cyclic family. ###### Lemma 3. Cyclic is not uniform for $n$ even and never 2-universal, and thus never pairwise independent. ###### Proof If $n$ is even, use the fact that $x^{n-1}+\cdots+x+1$ is divisible by $x+1$ to write $x^{n-1}+\cdots+x+1=(x+1)r(x)$ for some polynomial $r(x)$. Clearly, $r(x)(x+1)(x^{L-1}+x^{L-2}+\cdots+x+1)=0\bmod{x^{L}+1}$ for any $r(x)$ and so $P(h(a_{1},a_{1},\ldots,a_{1})=0)=P((x^{n-1}+\cdots+x+1)h_{1}(a_{1})=0)=P((x+1)r(x)h_{1}(a_{1})=0)\geq P(h_{1}(a_{1})=0\lor h_{1}(a_{1})=x^{L-1}+x^{L-2}+\cdots+x+1)=1/2^{L-1}$. Therefore, Cyclic is not uniform for $n$ even. To show Cyclic is never pairwise independent, consider $n=3$ (for simplicity), then $P(h(a_{1},a_{1},a_{2})=h(a_{1},a_{2},a_{1}))=P((x+1)(h_{1}(a_{1})+h_{1}(a_{2}))=0)\geq P(h_{1}(a_{1})+h_{1}(a_{2})=0\lor h_{1}(a_{1})+h_{1}(a_{2})=x^{L-1}+x^{L-2}+\cdots+x+1)=1/2^{L-1}$, but 2-universal hash values are equal with probability $1/2^{L}$. The result is shown. Of the four recursive hashing functions investigated by Cohen [17], General and Cyclic were superior both in terms of speed and uniformity, though Cyclic had a small edge over General. For $n$ large, the benefits of these recursive hash functions compared to the 3-wise independent hash function presented earlier can be substantial: $n$ table look-ups is much more expensive than a single look-up followed by binary shifts. ## 10 Cyclic is pairwise independent if you remove $n-1$ consecutive bits Because Cohen found empirically that Cyclic had good uniformity [17], it is reasonable to expect Cyclic to be almost uniform and maybe even almost pairwise independent. To illustrate this intuition, consider Table 3 which shows that while $h(\texttt{a},\texttt{a})$ is not uniform ($h(\texttt{a},\texttt{a})=001$ is impossible), $h(\texttt{a},\texttt{a})$ minus any bit is indeed uniformly distributed. We will prove that this result holds in general. Table 3: Cyclic hash for various values of $h_{1}(\texttt{a})$ ($h(\texttt{a},\texttt{a})=xh_{1}(\texttt{a})+h_{1}(\texttt{a})\bmod{2^{L}+1}$) $h_{1}(\texttt{a})$ | $h(\texttt{a},\texttt{a})$ | $h(\texttt{a},\texttt{a})$ | $h(\texttt{a},\texttt{a})$ | $h(\texttt{a},\texttt{a})$ ---|---|---|---|--- | | (first two bits) | (last two bits) | (first and last bit) 000 | 000 | 00 | 00 | 00 100 | 110 | 11 | 10 | 10 010 | 011 | 01 | 11 | 01 110 | 101 | 10 | 01 | 11 001 | 101 | 10 | 01 | 11 101 | 011 | 01 | 11 | 01 011 | 110 | 11 | 10 | 10 111 | 000 | 00 | 00 | 00 The next lemma and the next theorem show that Cyclic is quasi-pairwise independent in the sense that $L-n+1$ consecutive bits (e.g., the first or last $L-n+1$ bits) are pairwise independent. In other words, Cyclic is pairwise independent if we are willing to sacrifice $n-1$ bits. (We say that $n$ bits are “consecutive modulo $L$” if the bits are located at indexes $i\bmod{L}$ for $n$ consecutive values of $i$ such as $i=k,k+1,\ldots,k+n-1$.) ###### Lemma 4. If $q(x)\in\textrm{GF}(2)[x]/(x^{L}+1)$ (with $q(x)\neq 0$) has degree $n<L$, then * 1. the equation $q(x)w=y\bmod{x^{L}+1}$ modulo the first $n$ bits333By “equality modulo $\langle$some specified set of bit positions$\rangle$”, we mean that the two quantities are bitwise identical, with exceptions permitted only at the specified positions. For our polynomials, “equality modulo the first $n$ bit positions” implies the difference of the two polynomials has degree at most $n-1$. has exactly $2^{n}$ solutions for all $y$; * 2. more generally, the equation $q(x)w=y\bmod{x^{L}+1}$ modulo any consecutive $n$ bits (modulo $L$) has exactly $2^{n}$ solutions for all $y$. ###### Proof Let $P$ be the set of polynomials of degree at most $L-n-1$. Take any $p(x)\in P$, then $q(x)p(x)$ has degree at most $L-n-1+n=L-1$ and thus if $q(x)\neq 0$ and $p(x)\neq 0$, then $q(x)p(x)\neq 0\bmod{x^{L}+1}$. Hence, for any distinct $p_{1},p_{2}\in P$ we have $q(x)p_{1}\neq q(x)p_{2}\bmod{x^{L}+1}$. To prove the first item, we begin by showing that there is always exactly one solution in $P$. Consider that there are $2^{L-n}$ polynomials $p(x)$ in $P$, and that all values $q(x)p(x)$ are distinct. Suppose there are $p_{1},p_{2}\in P$ such that $q(x)p_{1}=q(x)p_{2}\bmod{x^{L}+1}$ modulo the first $n$ bits, then $q(x)(p_{1}-p_{2})$ is a polynomial of degree at most $n-1$ while $p_{1}-p_{2}$ is a polynomial of degree at most $L-n-1$ and $q(x)$ is a polynomial of degree $n$, thus $p_{1}-p_{2}=0$. (If $p1-p2\neq 0$ then $\textrm{degree}(q(x)(p1-p2)\bmod{x^{L}+1})\geq\textrm{degree}(q(x))=n$, a contradiction.) Hence, all $p(x)$ in $P$ are mapped to distinct values modulo the first $n$ bits, and since there are $2^{L-n}$ such distinct values, the result is shown. Any polynomial of degree $L-1$ can be decomposed into the form $p(x)+x^{L-n}z(x)$ where $z(x)$ is a polynomial of degree at most $n-1$ and $p(x)\in P$. By the preceding result, for distinct $p_{1},p_{2}\in P$, $q(x)(x^{L-n}z(x)+p_{1})$ and $q(x)(x^{L-n}z(x)+p_{2})$ must be distinct modulo the first $n$ bits. In other words, the equation $q(x)(x^{L-n}z(x)+p)=y$ modulo the first $n$ bits has exactly one solution $p\in P$ for any $z(x)$ and since there are $2^{n}$ polynomials $z(x)$ of degree at most $n-1$, then $q(x)w=y$ (modulo the first $n$ bits) must have $2^{n}$ solutions. To prove the second item, choose $j$ and use the first item to find any $w$ solving $q(x)w=yx^{j}\bmod{x^{L}+1}$ modulo the first $n$ bits. $j$. Then $wx^{L-j}$ is a solution to $q(x)w=y\bmod{x^{L}+1}$ modulo the bits in positions $j,j+1,\ldots,j+n-1\bmod{L}$. We have the following corollary to Lemma 4. ###### Corollary 1. If $w$ is chosen uniformly at random in $\textrm{GF}(2)[x]/(x^{L}+1)$, then $P(q(x)w=y\bmod{n-1\textrm{\ bits}})=1/2^{L-n+1}$ where the $n-1$ bits are consecutive (modulo $L$). ###### Theorem 1. Consider the $L$-bit Cyclic $n$-gram hash family. Pick any $n-1$ consecutive bit locations, then remove these bits from all hash values. The resulting $L-n+1$-bit hash family is pairwise independent. ###### Proof We show $P(q_{1}(x)h_{1}(a_{1})+q_{2}(x)h_{1}(a_{2})+\cdots+q_{n}(x)h_{1}(a_{n})=y\bmod{n-1\textrm{\ bits}})=1/2^{L-n+1}$ for any polynomials $q_{i}$ where at least one is different from zero. It is true when there is a single non-zero polynomial $q_{i}(x)$ by Corollary 1. Suppose it is true up to $k-1$ non-zero polynomials and consider a case where we have $k$ non-zero polynomials. Assume without loss of generality that $q_{1}(x)\neq 0$, we have $P(q_{1}(x)h_{1}(a_{1})+q_{2}(x)h_{1}(a_{2})+\cdots+q_{n}(x)h_{1}(a_{n})=y\bmod{n-1\textrm{\ bits}})=P(q_{1}(x)h_{1}(a_{1})=y-q_{2}(x)h_{1}(a_{2})-\cdots- q_{n}(x)h_{1}(a_{n})\bmod{n-1\textrm{\ bits}})=\sum_{y^{\prime}}P(q_{1}(x)h_{1}(a_{1})=y-y^{\prime}\bmod{n-1\textrm{\ bits}})P(q_{2}(x)h_{1}(a_{2})+\cdots+q_{n}(x)h_{1}(a_{n})=y^{\prime}\bmod{n-1\textrm{\ bits}})=\sum_{y^{\prime}}\frac{1}{2^{L-n+1}}\frac{1}{2^{L-n+1}}=1/2^{L-n+1}$ by the induction argument, where the sum is over $2^{L-n+1}$ values of $y^{\prime}$. Hence the uniformity result is shown. Consider two distinct sequences $a_{1},a_{2},\ldots,a_{n}$ and $a^{\prime}_{1},a^{\prime}_{2},\ldots,a^{\prime}_{n}$. Write $H_{a}=h(a_{1},a_{2},\ldots,a_{n})$ and $H_{a^{\prime}}=h(a^{\prime}_{1},a^{\prime}_{2},\ldots,a^{\prime}_{n})$. To prove pairwise independence, it suffices to show that $P(H_{a}=y\bmod{n-1\textrm{\ bits}}|H_{a^{\prime}}=y^{\prime}\bmod{n-1\textrm{\ bits}})=1/2^{L-n+1}$. Suppose that $a_{i}=a^{\prime}_{j}$ for some $i,j$; if not, the result follows by the (full) independence of the hashing function $h_{1}$. Using Lemma 4, find $q(x)$ such that $q(x)\sum_{k|a^{\prime}_{k}=a^{\prime}_{j}}x^{n-k}=-\sum_{k|a_{k}=a_{i}}x^{n-k}\bmod{n-1\textrm{\ bits}}$, then $H_{a}+q(x)H_{a^{\prime}}\bmod{n-1\textrm{\ bits}}$ is independent from $a_{i}=a^{\prime}_{j}$ (and $h_{1}(a_{i})=h_{1}(a^{\prime}_{j})$). The hashed values $h_{1}(a_{k})$ for $a_{k}\neq a_{i}$ and $h_{1}(a^{\prime}_{k})$ for $a^{\prime}_{k}\neq a^{\prime}_{j}$ are now relabelled as $h_{1}(b_{1}),\ldots,h_{1}(b_{m})$. Write $H_{a}+q(x)H_{a^{\prime}}=\sum_{k}q_{k}(x)h_{1}(b_{k})\bmod{n-1\textrm{\ bits}}$ where $q_{k}(x)$ are polynomials in $\textrm{GF}(2)[x]/(x^{L}+1)$ (not all $q_{k}(x)$ are zero). As in the proof of Lemma 1, we have that $H_{a^{\prime}}=y^{\prime}\bmod{n-1\textrm{\ bits}}$ and $H_{a}+q(x)H_{a^{\prime}}=y+q(x)y^{\prime}\bmod{n-1\textrm{\ bits}}$ are independent444We use the shorthand notation $P(f(x,y)=c|x,y)=b$ to mean $P(f(x,y)=c|x=z_{1},y=z_{2})=b$ for all values of $z_{1},z_{2}$. : $P(H_{a^{\prime}}=y^{\prime}\bmod{n-1\textrm{\ bits}}|y^{\prime},b_{1},b_{2},\ldots,b_{m})=1/2^{L-n+1}$ by Corollary 1 since $H_{a^{\prime}}=y$ can be written as $r(x)h_{1}(a^{\prime}_{j})=y-\sum_{k}r_{k}(x)h_{1}(b_{k})$ for some polynomials $r(x),r_{1}(x),\ldots,r_{m}(x)$. Hence, we have $\displaystyle P(H_{a}=y\bmod{n-1\textrm{\ bits}}|H_{a^{\prime}}=y^{\prime}\bmod{n-1\textrm{\ bits}})$ $\displaystyle=$ $\displaystyle P(H_{a}+q(x)H_{a^{\prime}}=y+q(x)y^{\prime}\bmod{n-1\textrm{\ bits}}|H_{a^{\prime}}=y^{\prime}\bmod{n-1\textrm{\ bits}})$ $\displaystyle=$ $\displaystyle P(H_{a}+q(x)H_{a^{\prime}}=y+q(x)y^{\prime}\bmod{n-1\textrm{\ bits}})$ $\displaystyle=$ $\displaystyle P(\sum_{k}q_{k}(x)h_{1}(b_{k})=y+q(x)y^{\prime}\bmod{n-1\textrm{\ bits}})$ and by the earlier uniformity result, this last probability is equal to $1/2^{L-n+1}$. ## 11 Experimental comparison Irrespective of $p(x)$, computing hash values has complexity $\Omega(L)$. For General and Cyclic, we require $L\geq n$. Hence, the computation of their hash values is in $\Omega(n)$. For moderate values of $L$ and $n$, this analysis is pessimistic because CPUs can process 32- or 64-bit words in one operation. To assess their real-world performance, the various hashing algorithms555http://code.google.com/p/ngramhashing/. were written in C++. We compiled them with the GNU GCC 4.0.1 compiler on an Apple MacBook with two Intel Core 2 Duo processors (2.4 GHz) and 4 GiB of RAM. The -O3 compiler flag was used since it provided slightly better performance for all algorithms. All hash values are stored using 32-bit integers, irrespective of the number of bits used. All hashing functions generate 19-bit hash values, except for Cyclic which generates 19+$n$-bit hash values. We had Cyclic generate more bits to compensate for the fact that it is only pairwise independent after removal of $n-1$ consecutive bits. For General, we used the polynomial $p(x)=x^{19}+x^{5}+x^{2}+x+1$ [25]. For Randomized Karp-Rabin, we used the ID37 family. The character hash-values are stored in an array for fast look- up. We report wall-clock time in Fig. 1 for hashing the $n$-grams of the King James Bible [20] which contains 4.3 million ASCII characters. Cyclic is twice as fast as General. As expected, the running time of the non-recursive hash function (3-wise) grows linearly with $n$: for $n=5$, 3-wise is already seven times slower than Cyclic. Speed-wise, Randomized Karp-Rabin (ID37) is the clear winner, being nearly twice as fast as Cyclic. The performance of Cyclic and ID37 is oblivious to $n$ in this test. The RAM-Buffered General timings are—as expected—independent of $n$, but they are twice as large as the Cyclic timings. We do not show the modified version of RAM-Buffered General that uses two precomputed arrays instead of a single one. It was approximately 30% slower than ordinary RAM-Buffered General, even up to $n=25$. However, its RAM usage was 3 orders of magnitude smaller: from 135 MB down to 25 kB. Overall, we cannot recommend RAM-Buffered General or its modification considering that (1) its memory usage grows as $2^{n}$ and (2) it is slower than Cyclic. Figure 1: Wall-clock running time to hash all $n$-grams in the King James Bible ## 12 Conclusion Considering speed and pairwise independence, we recommend Cyclic—after discarding $n-1$ consecutive bits. If we require only uniformity, Randomized Integer-Division is twice as fast. ## Acknowledgments This work is supported by NSERC grants 155967, 261437 and by FQRNT grant 112381. The authors are grateful to the anonymous reviewers for their significant contributions. ## References * [1] J. D. Cohen, Hardware-assisted algorithm for full-text large-dictionary string matching using n-gram hashing, Information Processing and Management 34 (4) (1998) 443–464. * [2] J. D. Cohen, Massive query resolution for rapid selective dissemination of information, Journal of the American Society for Information Science 50 (3) (1999) 195–206. * [3] J. D. Cohen, An n-gram hash and skip algorithm for finding large numbers of keywords in continuous text streams, Softw. Pract. Exper. 28 (15) (1998) 1605–1635. * [4] T. Tan, S. Gould, D. Williams, E. Peltzer, R. Barrie, Fast pattern matching using large compressed databases, US Patent App. 11/326,131 (2006). * [5] A. Cardenal-Lopez, F. J. Diguez-Tirado, C. Garcia-Mateo, Fast LM look-ahead for large vocabulary continuous speech recognition using perfect hashing, in: ICASSP’02, 2002, pp. 705–708. * [6] X. Zhang, Y. Zhao, Minimum perfect hashing for fast N-gram language model lookup, in: Seventh International Conference on Spoken Language Processing, ISCA, 2002, pp. 401–404. * [7] H. Schwenk, Continuous space language models, Computer Speech & Language 21 (3) (2007) 492–518. * [8] X. Li, Y. Zhao, A fast and memory-efficient N-gram language model lookup method for large vocabulary continuous speech recognition, Computer Speech & Language 21 (1) (2007) 1–25. * [9] D. Talbot, M. Osborne, Smoothed Bloom filter language models: Tera-scale LMs on the cheap, in: EMNLP’07, 2007, pp. 468–476. * [10] D. Talbot, M. Osborne, Randomised language modelling for statistical machine translation, in: ACL’07, 2007, pp. 512–519. * [11] D. Talbot, T. Brants, Randomized language models via perfect hash functions, ACL’08 (2008) 505–513. * [12] R. L. Ribler, M. Abrams, Using visualization to detect plagiarism in computer science classes, in: INFOVIS’00, IEEE Computer Society, Washington, DC, USA, 2000, p. 173. * [13] L. Carter, M. N. Wegman, Universal classes of hash functions, Journal of Computer and System Sciences 18 (2) (1979) 143–154. * [14] P. Flajolet, G. N. Martin, Probabilistic counting algorithms for data base applications, Journal of Computer and System Sciences 31 (2) (1985) 182–209. * [15] P. B. Gibbons, S. Tirthapura, Estimating simple functions on the union of data streams, in: SPAA’01, 2001, pp. 281–291. * [16] M. Mitzenmacher, S. Vadhan, Why simple hash functions work: exploiting the entropy in a data stream, in: SODA ’08, 2008, pp. 746–755. * [17] J. D. Cohen, Recursive hashing functions for n-grams, ACM Trans. Inf. Syst. 15 (3) (1997) 291–320. * [18] S. Schleimer, D. S. Wilkerson, A. Aiken, Winnowing: local algorithms for document fingerprinting, in: SIGMOD’2003, 2003, pp. 76–85. * [19] M. Durand, P. Flajolet, Loglog counting of large cardinalities, in: ESA’03, Vol. 2832 of LNCS, 2003, pp. 605–617. * [20] Project Gutenberg Literary Archive Foundation, Project Gutenberg, http://www.gutenberg.org/ (checked 05-23-2011) (2009). * [21] R. M. Karp, M. O. Rabin, Efficient randomized pattern-matching algorithms, IBM Journal of Research and Development 31 (2) (1987) 249–260. * [22] Sun Microsystems, String (Java 2 Platform SE 5.0), online documentation: http://java.sun.com/j2se/1.5.0/docs/api/index.html (2004). * [23] M. Weiss, Data Structures and Algorithm Analysis in Java, Addison Wesley, 1999. * [24] M. Fürer, Faster integer multiplication, in: STOC ’07, 2007, pp. 57–66. * [25] F. Ruskey, The (combinatorial) object server, http://www.theory.cs.uvic.ca/~cos/cos.html, checked 2011-09-01 (2006).

arxiv-papers

2024-09-04T02:48:52.908865

0706.0058

# On Cuspidal Representations of General Linear Groups over Discrete Valuation Rings Anne-Marie Aubert C.N.R.S., Institut de Mathématiques de Jussieu, 175 rue du Chevaleret, 75013 Paris, France. , Uri Onn${}^{~{}\star}$ Department of Mathematics, Ben Gurion University of the Negev, Be’er Sheva 84105, Israel. , Amritanshu Prasad The Institute of Mathematical Sciences, CIT campus, Taramani, Chennai 600 113, India. and Alexander Stasinski† DPMMS, University of Cambridge, Wilberforce Road, Cambridge, CB3 0WB, U. K. ###### Abstract. We define a new notion of cuspidality for representations of $\mathrm{GL}_{n}$ over a finite quotient $\mathfrak{o}_{k}$ of the ring of integers $\mathfrak{o}$ of a non-Archimedean local field $F$ using geometric and infinitesimal induction functors, which involve automorphism groups $G_{\lambda}$ of torsion $\mathfrak{o}$-modules. When $n$ is a prime, we show that this notion of cuspidality is equivalent to strong cuspidality, which arises in the construction of supercuspidal representations of $\mathrm{GL}_{n}(F)$. We show that strongly cuspidal representations share many features of cuspidal representations of finite general linear groups. In the function field case, we show that the construction of the representations of $\mathrm{GL}_{n}(\mathfrak{o}_{k})$ for $k\geq 2$ for all $n$ is equivalent to the construction of the representations of all the groups $G_{\lambda}$. A functional equation for zeta functions for representations of $\mathrm{GL}_{n}(\mathfrak{o}_{k})$ is established for representations which are not contained in an infinitesimally induced representation. All the cuspidal representations for $\mathrm{GL}_{4}(\mathfrak{o}_{2})$ are constructed. Not all these representations are strongly cuspidal. ###### Key words and phrases: Cuspidal representations; general linear groups; local rings ###### 2000 Mathematics Subject Classification: 22E50, 11S37 $\star$ Supported by the Israel Science Foundation, ISF grant no. 555104, by the Edmund Landau Minerva Center for Research in Mathematical Analysis and Related Areas, sponsored by the Minerva Foundation (Germany). † Supported at various times by EPSRC Grants GR/T21714/01 and EP/C527402. ## 1\. Introduction The irreducible characters of $\mathrm{GL}_{n}(\mathbf{F}_{q})$ were computed by J. A. Green in 1955 [Gre55]. In Green’s work, parabolic induction was used to construct many irreducible characters of $\mathrm{GL}_{n}(\mathbf{F}_{q})$ from irreducible characters of smaller general linear groups over $\mathbf{F}_{q}$. The representations which could not be obtained in this way, known as _cuspidal representations_ , were shown to be in canonical bijective correspondence with Galois orbits of norm-primitive characters of $\mathbf{F}_{q^{n}}^{\times}$ (these are characters which do not factor through the norm map $\mathbf{F}_{q^{n}}^{\times}\to\mathbf{F}_{q^{d}}^{\times}$ for any proper factor $d$ of $n$). Let $F$ be a non-Archimedean local field with ring of integers $\mathfrak{o}$. Let $\mathfrak{p}$ be the maximal ideal in $\mathfrak{o}$, and $\mathfrak{o}_{k}=\mathfrak{o}/\mathfrak{p}^{k}$ for $k\geq 1$. Thus $\mathfrak{o}_{1}$ is a finite field, the residue field of $F$, which we take to be $\mathbf{F}_{q}$. In contrast with $\mathrm{GL}_{n}(\mathfrak{o}_{1})$, not much is known in general about the representation theory of $\mathrm{GL}_{n}(\mathfrak{o}_{k})$. Unlike general linear groups over fields, for which conjugacy classes are parameterized by Jordan canonical forms, the classification of conjugacy classes in $\mathrm{GL}_{n}(\mathfrak{o}_{k})$ for all $n$ and any $k\geq 2$ contains the matrix pair problem [Nag78, Section 4], which is a wild classification problem. The representations of $\mathrm{GL}_{n}(\mathfrak{o})$ received considerable attention after supercuspidal representations of $\mathrm{GL}_{n}(F)$ were constructed by induction from a compact-modulo-center subgroup [Shi68, How77, Kut78]. A class of representations (_représentations très cuspidales_) of the maximal compact-modulo-center subgroups which give rise to irreducible supercuspidal representations of $\mathrm{GL}_{n}(F)$ were identified by Carayol [Car84]. When the maximal compact subgroup modulo center in question is $F^{\times}\mathrm{GL}_{n}(\mathfrak{o})$, the restrictions of these representations to $\mathrm{GL}_{n}(\mathfrak{o})$ correspond to what we call _strongly cuspidal representations_ of $\mathrm{GL}_{n}(\mathfrak{o}_{k})$ for some $k$ (Definition 4.1). Carayol used these representations to construct all the supercuspidal representations of $\mathrm{GL}_{n}(F)$ when $n$ is prime. The classification of supercuspidal representations of $\mathrm{GL}_{n}(F)$ for $n$ arbitrary was completed by Bushnell and Kutzko [BK93]. Recently, Paskunas [Pas05] proved that given an irreducible supercuspidal representation $\pi$ of $\mathrm{GL}_{n}(F)$, there exists a unique (up to isomorphism) irreducible representation $\tau$ of $\mathrm{GL}_{n}(\mathfrak{o})$, such that $\tau$ is a type for the Bernstein component of $\pi$. Hence representations of $\mathrm{GL}_{n}(\mathfrak{o})$ occur naturally in the representation theory of $\mathrm{GL}_{n}(F)$. Nevertheless, with respect to $\mathrm{GL}_{n}(\mathfrak{o})$, since the general representation theory is unmanageably complicated, only those very special representations that are needed to understand the representations of the $p$-adic group itself have been considered. In this article, we take the point of view that the representation theory of $\mathrm{GL}_{n}(\mathfrak{o})$ is interesting in its own right, and while extremely complicated, does display a certain structure. To this end, a new definition of cuspidality is introduced for representations of $\mathrm{GL}_{n}(\mathfrak{o}_{k})$. This definition is closer in spirit to the characterization in [Gre55] of cuspidal representations as those which do not occur in representations obtained by parabolic induction. More specifically, let $\Lambda$ denote the set of all partitions of all positive integers. The isomorphism classes of finitely generated torsion $\mathfrak{o}$-modules are parameterized by $\Lambda$. For any $\mathfrak{o}$-module $\mathfrak{o}_{\lambda}=\oplus_{i=1}^{m}\mathfrak{o}_{\lambda_{i}}$ of type $\lambda=(\lambda_{1},\ldots,\lambda_{m})\in\Lambda$, let $G_{\lambda}=G_{\lambda,F}$ stand for its automorphism group. Thus, for example, $G_{k^{n}}=\mathrm{GL}_{n}(\mathfrak{o}_{k})$. Say that $\lambda\leq\mu$ if $\mathfrak{o}_{\lambda}$ can be embedded in $\mathfrak{o}_{\mu}$. We call an irreducible representation of $G_{k^{n}}$ cuspidal (see Definition 4.2) if it cannot be constructed from lower building blocks. By lower building blocks we mean the representations of $G_{\lambda}$, where $\lambda<k^{n}$. These automorphism groups play the role of Levi components of proper parabolic subgroups of $\mathrm{GL}_{n}(\mathfrak{o}_{1})$. Representations of $G_{k^{n}}$ are constructed from those of $G_{\lambda}$ using _infinitesimal_ and _geometric_ induction (Section 3). Our first result, which is proved in Section 4.2, compares cuspidality with strong cuspidality. ###### Theorem A. Every strongly cuspidal representation is cuspidal. When $n$ is prime every cuspidal representation is strongly cuspidal. When $n$ is not prime, it is not true that every cuspidal representation is strongly cuspidal. In Section 8, all the cuspidal representations of $\mathrm{GL}_{4}(\mathfrak{o}_{2})$ are constructed. Among these are representations which are not strongly cuspidal. The construction of strongly cuspidal representations is well-known [Shi68, Gér75, How77]. When $n$ is prime, then by Theorem A, all cuspidal representations are obtained in this manner. Moreover, for all $n$, the strongly cuspidal representations have properties analogous to cuspidal representations of $\mathrm{GL}_{n}(\mathfrak{o}_{1})$. Firstly, they can be parameterized in an analogous fashion. Suppose that $E$ is an unramified extension of $F$ of degree $n$, and $\mathfrak{O}$ is the integral closure of $\mathfrak{o}$ in $E$. Let $\mathfrak{P}$ denote the maximal ideal in $\mathfrak{O}$ and $\mathfrak{O}_{k}$ denote the finite quotient ring $\mathfrak{O}/\mathfrak{P}^{k}$. For $k>1$, a character $\mathfrak{O}_{k}^{\times}\to\mathbf{C}^{\times}$ is said to be _strongly primitive_ if its restriction to $\ker(\mathfrak{O}_{k}^{\times}\to\mathfrak{O}_{k-1}^{\times})\cong\mathfrak{O}_{1}$ does not factor through any proper subfield via the trace map. A character of $\mathfrak{O}_{1}^{\times}$ is said to be strongly primitive if it is norm- primitive. In Section 5.3 we prove ###### Theorem B. There is a canonical bijective correspondence between strongly cuspidal representations of $G_{k^{n}}$ and $\mathrm{Gal}(E/F)$-orbits of strongly primitive characters of $\mathfrak{O}_{k}^{\times}$. The group of units $\mathfrak{O}_{k}^{\times}$ is embedded in $G_{k^{n}}$ since $\mathfrak{O}_{k}\simeq\mathfrak{o}^{n}_{k}$ as $\mathfrak{o}_{k}$-modules. An element of $G_{k^{n}}$ is said to be regular elliptic if it is conjugate to an element of $\mathfrak{O}_{k}^{\times}$ whose image in $\mathfrak{O}_{1}^{\times}$ lies in no proper subfield. In section 5.2, we establish another property that strongly cuspidal representations share with cuspidal representations of $\mathrm{GL}_{n}(\mathfrak{o}_{1})$, which is that the correspondence of Theorem B is well-behaved with respect to character values on regular elliptic elements. ###### Theorem C. Let $\omega$ be a strongly primitive character of $\mathfrak{O}_{k}^{\times}$ and let $\Theta_{\omega}$ be the corresponding strongly cuspidal character of $G_{k^{n}}$. Then for all regular elliptic elements $u\in\mathfrak{O}_{k}^{\times}\subset G_{k^{n}}$ $\Theta_{\omega}(u)=(-1)^{(n-1)k}\sum_{\gamma\in\mathrm{Gal}(E/F)}\omega({{}^{\gamma}u}),$ Moreover, $\Theta_{\omega}$ vanishes on conjugacy classes which do not intersect $\mathfrak{O}_{k}^{\times}\cdot\mathrm{Ker}\\{G_{k^{n}}\to G_{\lceil k/2\rceil^{n}}\\}$. ###### Remark. Theorems B and C are due to Green when $k=1$. For $k>1$, the ideas used in the proofs can be found in the existing literature on supercuspidal representations of $\mathrm{GL}_{n}(F)$, the detailed account in Section 5 gives the complete picture, working entirely inside $\mathrm{GL}_{n}(\mathfrak{o})$. In particular, Theorem C is deduced from [Gér75, Theorem 1]. It is closely related to the result obtained by Henniart in [Hen93, Section 3.7]. We also observe that in [Lu04] Lusztig gave a geometric construction of representations (in the function field case) which is likely to include the description of strongly cuspidal representations of $G_{k^{n}}$ in terms of strongly primitive characters. There already is evidence that the representation theory of a group such as $G_{\lambda}$ can be studied by breaking up the problem into two parts. The first is to correctly define and understand the cuspidal representations. The second is to construct the remaining representations from cuspidal representations of $G_{\mu}$ with $\mu<\lambda$. This approach has been implemented successfully in [Onn07] for automorphism groups of modules of rank two. Theorems A, B and C provide further evidence of the validity of this approach when $\lambda=k^{n}$ and $n$ is a prime. The inevitability of the family of groups $G_{\lambda}$ in the representation theory of $G_{k^{n}}$ or even $G_{2^{n}}$ can be seen from another perspective. In Section 6, we prove ###### Theorem D. Let $F$ be a local function field. Constructing the irreducible representations of the family of groups $\\{G_{2^{n},F}=\mathrm{GL}_{n}(\mathfrak{o}_{2})~{}|~{}n\in\mathbf{N}\\}$ is equivalent to constructing the irreducible representations of the family $\\{G_{\lambda,E}~{}|~{}\lambda\in\Lambda,~{}E/F~{}\text{unramified extension}\\}$. Finally, we point out a suggestive connection to the Macdonald correspondence which might admit a higher level incarnation as well. Macdonald has established a correspondence between irreducible representations of $G_{1^{n}}$ and equivalence classes of $n$-dimensional tamely ramified representations of the Weil-Deligne group $W^{\prime}_{F}$ [Mac80]. One ingredient in this correspondence is a functional equation for the zeta function associated to $G_{1^{n}}$. It admits a straightforward generalization to $G_{k^{n}}$ for $k>1$. Let $\hat{f}$ denote a properly normalized additive Fourier transform of $f\in\mathbf{C}\left(M_{n}(\mathfrak{o}_{k})\right)$ with respect to $\psi\left(\mathrm{tr}(\cdot)\right)$, where $\psi:\mathfrak{o}_{k}\to\mathbf{C}$ is an additive character which does not factor through $\mathfrak{o}_{k-1}$. Let $\mathcal{Z}(f,\rho)=\sum_{g\in G_{k^{n}}}f(g)\rho(g)\in\text{End}_{\mathbf{C}}(V)$ where $f\in\mathbf{C}\left(M_{n}(\mathfrak{o}_{k})\right)$ and $(\rho,V)$ is an irreducible representation of $G_{k^{n}}$. Denote by $\check{\rho}$ the contragredient representation of $\rho$. In Section 7, we prove ###### Theorem E. If $\rho$ is not contained in an infinitesimally induced representation (in particular if $\rho$ is cuspidal), there exists a complex number $\varepsilon(\rho,\psi)$ and a such that ${{}^{t}\mathcal{Z}}(\hat{f},\check{\rho})=\varepsilon(\rho,\psi)\mathcal{Z}(f,\rho).$ ### 1.1. Acknowledgments The second author is grateful to Alex Lubotzky and Yakov Varshavsky for supporting this research. The third author acknowledges M. K. Vemuri for some very helpful discussions on Heisenberg groups. The second and third authors thank Shahar Mendelson, Amnon Neeman and the Australian National University in Canberra for giving them an opportunity to work together. The fourth author thanks S. Stevens for many helpful discussions which were instrumental for parts of the present work. The authors thank Robert Kottwitz and Dipendra Prasad, who read a draft of this article and provided some very valuable feedback. They are grateful to the referee for his careful reading of the article and valuable comments. ## 2\. Notations and preliminaries ### 2.1. Automorphism groups Let $\Lambda$ denote the set of all partitions of all positive integers. Any $\lambda\in\Lambda$ can be written in the form $(\lambda_{1}^{r_{1}},\ldots,\lambda_{l}^{r_{l}})$, where $\lambda_{1}>\cdots>\lambda_{l}$ and $r_{1},\ldots,r_{l}$ are positive integers. The sum $r_{1}+\cdots+r_{l}$ will be called the _length_ of the partition, and $\lambda_{1}$ will be called the _height_ of the partition. Every finitely generated torsion $\mathfrak{o}$-module is of the form $\mathfrak{o}_{\lambda}=\mathfrak{o}_{\lambda_{1}}^{r_{1}}\oplus\cdots\oplus\mathfrak{o}_{\lambda_{l}}^{r_{l}}$ for some $\lambda\in\Lambda$ of height no more than $k$. Consider the group $G_{\lambda}=\mathrm{Aut}_{\mathfrak{o}}(\mathfrak{o}_{\lambda})$. In particular, taking $\lambda=(k^{n})$, we have $G_{k^{n}}=\mathrm{GL}_{n}(\mathfrak{o}_{k})$. When it is necessary to specify the underlying non-Archimedean local field $F$, the notation $G_{\lambda,F}$ will be used for $G_{\lambda}$. Let $N_{r}$ denote the kernel of the natural map $G_{k^{n}}\to G_{r^{n}}$. Then, if $r\geq k/2$, the map $M_{n}(\mathfrak{o}_{k-r})\to N_{r}$ defined by $A\mapsto\mathrm{I}+\varpi^{r}A$, is an isomorphism of groups (it is a bijection of sets for all $r<k$). This results in a short exact sequence (2.1) $0\to M_{n}(\mathfrak{o}_{k-r})\to G_{k^{n}}\to G_{r^{n}}\to 1,$ for every $r\geq k/2$. In what follows, we identify $M_{n}(\mathfrak{o}_{k-r})$ with its image in $G_{k^{n}}$ for $r\geq k/2$. ### 2.2. Similarity classes associated to representations Assume that $r\geq k/2$. The action of $G_{k^{n}}$ on its normal subgroup $M_{n}(\mathfrak{o}_{k-r})$ factors through $G_{(k-r)^{n}}$. In fact, this is just the usual action by similarity transformations $g\cdot A=gAg^{-1},\quad g\in G_{(k-r)^{n}},\>A\in M_{n}(\mathfrak{o}_{k-r}).$ It results in an action of $G_{(k-r)^{n}}$ on the set of all characters of $M_{n}(\mathfrak{o}_{k-r})$. Now suppose that $\rho$ is an irreducible representation of $G_{k^{n}}$ on a vector space $V$. The restriction of $\rho$ to $M_{n}(\mathfrak{o}_{k-r})$ gives rise to a decomposition $V=\oplus V_{\chi}$, where $\chi$ ranges over the set of characters of $M_{n}(\mathfrak{o}_{k-r})$. Clifford theory then tells us that the set of characters $\chi$ for which $V_{\chi}$ is non-trivial consists of a single orbit for the action of $G_{(k-r)^{n}}$ on the characters of $M_{n}(\mathfrak{o}_{k-r})$. The group $M_{n}(\mathfrak{o}_{k-r})$ can be identified with its Pontryagin dual (as a $G_{(k-r)^{n}}$-space). For this, pick an additive character $\psi$ of $F\to\mathbf{C}^{\times}$ whose restriction to $\mathfrak{o}$ is trivial, but whose restriction to $\mathfrak{p}^{-1}$ is non-trivial. For each $A\in M_{n}(\mathfrak{o}_{k-r})$, define a character $\psi_{A}\colon M_{n}(\mathfrak{o}_{k-r})\to\mathbf{C}^{\times}$ by $\psi_{A}(B)=\psi(\varpi^{r-k}\mathrm{tr}(AB))$. The map $A\mapsto\psi_{A}$ identifies $M_{n}(\mathfrak{o}_{k-r})$ with its Pontryagin dual, and preserves the action of $G_{(k-r)^{n}}$. Thus we associate, for each $r\geq k/2$, to each irreducible representation $\rho$ of $G_{k^{n}}$, a similarity class $\Omega_{k-r}(\rho)\subset M_{n}(\mathfrak{o}_{k-r})$. ## 3\. Induction and restriction functors This section introduces the functors that will play the role of parabolic induction and restriction in the context of $\mathrm{GL}_{n}(\mathfrak{o}_{k})$. They were introduced in [Onn07, Section 2]. Geometric induction is an obvious analog of parabolic induction in the case of a field. Infinitesimal induction has no analog in that setting. ### 3.1. Geometric induction and restriction functors Given a direct sum decomposition $\mathfrak{o}_{k}^{n}=\mathfrak{o}_{k}^{n_{1}}\oplus\mathfrak{o}_{k}^{n_{2}}$, define $P_{n_{1},n_{2}}$ to be the subgroup of $G_{k^{n}}$ which preserves $\mathfrak{o}_{k}^{n_{1}}$. There is a natural surjection $\varphi\colon P_{n_{1},n_{2}}\to G_{k^{n_{1}}}\times G_{k^{n_{2}}}$. Denote the kernel by $U_{n_{1},n_{2}}$. Define the functor $i_{n_{1},n_{2}}\colon\mathrm{Rep}(G_{k^{n_{1}}})\times\mathrm{Rep}(G_{k^{n_{2}}})\to\mathrm{Rep}(G_{k^{n}})$ taking representations $\sigma_{1}$ and $\sigma_{2}$ of $G_{k^{n_{1}}}$ and $G_{k^{n_{2}}}$ respectively to the induction to $G_{k^{n}}$ of the pull-back under $\varphi$ of $\sigma_{1}\otimes\sigma_{2}$. The functor $r_{n_{1},n_{2}}\colon\mathrm{Rep}(G_{k^{n}})\to\mathrm{Rep}(G_{k^{n_{1}}})\times\mathrm{Rep}(G_{k^{n_{2}}})$ is defined by restricting a representation $\rho$ of $G_{k^{n}}$ to $P_{n_{1},n_{2}}$ and then taking the invariants under $U_{n_{1},n_{2}}$. By Frobenius reciprocity, these functors form an adjoint pair: $\mathrm{Hom}_{G_{k^{n}}}(\rho,i_{n_{1},n_{2}}(\sigma_{1},\sigma_{2}))=\mathrm{Hom}_{G_{k^{n_{1}}}\times G_{k^{n_{2}}}}(r_{n_{1},n_{2}}(\rho),\sigma_{1}\otimes\sigma_{2}).$ Following [Onn07], the functors $i_{n_{1},n_{2}}$ and $r_{n_{1},n_{2}}$ are called _geometric induction_ and _geometric restriction_ functors, respectively. Furthermore ###### Definition 3.1. An irreducible representation of $G_{k^{n}}$ will be said to lie in the geometrically induced series if it is isomorphic to a subrepresentation of $i_{n_{1},n_{n}}(\sigma_{1},\sigma_{2})$ for some decomposition $n=n_{1}+n_{2}$ with $n_{1}$ and $n_{2}$ strictly positive, and some representations $\sigma_{1}$ and $\sigma_{2}$ of $G_{k^{n_{1}}}$ and $G_{k^{n_{2}}}$ respectively. ### 3.2. Infinitesimal induction and restriction functors For two partitions $\lambda$ and $\mu$, say that $\lambda\leq\mu$ if there exists an embedding of $\mathfrak{o}_{\lambda}$ in $\mathfrak{o}_{\mu}$ as an $\mathfrak{o}$-module. This is equivalent to the existence of a surjective $\mathfrak{o}$-module morphism $\mathfrak{o}_{\mu}\to\mathfrak{o}_{\lambda}$. If $\lambda\leq k^{n}$, then the pair $(\lambda,k^{n})$ has the _unique embedding_ and _unique quotient_ properties, i.e., all embeddings of $\mathfrak{o}_{\lambda}$ in $\mathfrak{o}_{k^{n}}$ and all surjections of $\mathfrak{o}_{k^{n}}$ onto $\mathfrak{o}_{\lambda}$ lie in the same $G_{k^{n}}$-orbit. As a consequence the functors that are defined below will, up to isomorphism, not depend on the choices of embeddings and surjections involved (in the language of [BO07, Section 2], $k^{n}$ is a _symmetric type_). Given $\lambda\leq k^{n}$, take the obvious embedding of $\mathfrak{o}_{\lambda}$ in $\mathfrak{o}_{k}^{n}$ given on standard basis vectors by $\mathbf{f}_{i}\mapsto\pi^{k-\lambda_{h(i)}}\mathbf{e}_{i}$, where $h(i)$ is such that $r_{1}+\cdots+r_{h(i)-1}<i\leq r_{1}+\cdots+r_{h(i)}$. Define $P_{\lambda\hookrightarrow k^{n}}=\\{g\in G_{k^{n}}\;|\>g\cdot\mathfrak{o}_{\lambda}=\mathfrak{o}_{\lambda}\\},$ Restriction to $\mathfrak{o}_{\lambda}$ gives rise to a homomorphism $P_{\lambda\hookrightarrow k^{n}}\to G_{\lambda}$ which, due to the unique embedding property, is surjective. Let $U_{\lambda\hookrightarrow k^{n}}$ be the kernel. One may now define an induction functor $i_{\lambda\hookrightarrow k^{n}}\colon\mathrm{Rep}(G_{\lambda})\to\mathrm{Rep}(G_{k^{n}})$ as follows: given a representation of $G_{\lambda}$, pull it back to a representation of $P_{\lambda\hookrightarrow k^{n}}$ via the homomorphism $P_{\lambda\hookrightarrow k^{n}}\to G_{\lambda}$, and then induce to $G_{k^{n}}$. Its adjoint functor $r_{\lambda\hookrightarrow k^{n}}\colon\mathrm{Rep}(G_{k^{n}})\to\mathrm{Rep}(G_{\lambda})$ is obtained by taking a representation of $G_{k^{n}}$, restricting to $P_{\lambda\hookrightarrow k^{n}}$, and taking the vectors invariant under $U_{\lambda\hookrightarrow k^{n}}$. The adjointness is a version of Frobenius reciprocity: there is a natural isomorphism $\mathrm{Hom}_{G_{k^{n}}}\left(\rho,i_{\lambda\hookrightarrow k^{n}}(\sigma)\right)=\mathrm{Hom}_{G_{\lambda}}\left(r_{\lambda\hookrightarrow k^{n}}(\rho),\sigma\right)$ for representations $\rho$ and $\sigma$ of $G_{k^{n}}$ and $G_{\lambda}$ respectively. In terms of matrices, the groups $P_{\lambda\hookrightarrow k^{n}}$ and $U_{\lambda\hookrightarrow k^{n}}$ are $\displaystyle P_{\lambda\hookrightarrow k^{n}}=\\{(a_{ij})\in G_{k^{n}}\;|\>a_{ij}\in\pi^{\min\\{0,\lambda_{h(j)}-\lambda_{h(i)}\\}}\mathfrak{o}_{k}\\},$ $\displaystyle U_{\lambda\hookrightarrow k^{n}}=\\{(a_{ij})\in P_{\lambda\hookrightarrow k^{n}}\;|\>a_{ij}\in\delta_{ij}+\pi^{\lambda_{h(j)}}\mathfrak{o}_{k}\\}.$ Dually, fix the surjection of $\mathfrak{o}_{k}^{n}$ onto $\mathfrak{o}_{\lambda}$ given by $\mathbf{e}_{i}\mapsto\mathbf{f}_{i}$ and define $P_{k^{n}\twoheadrightarrow\lambda}=\\{g\in G_{k^{n}}\;|\>g\cdot\ker(\mathfrak{o}_{k}^{n}\to\mathfrak{o}_{\lambda})=\ker(\mathfrak{o}_{k}^{n}\to\mathfrak{o}_{\lambda})\\}.$ Taking the induced map on the quotient gives rise to a homomorphism $P_{k^{n}\twoheadrightarrow\lambda}\to G_{\lambda}$ which, by the unique quotient property, is surjective. Let $U_{k^{n}\twoheadrightarrow\lambda}$ denote the kernel. An adjoint pair of functors $i_{k^{n}\twoheadrightarrow\lambda}\colon\mathrm{Rep}(G_{\lambda})\to\mathrm{Rep}(G_{k^{n}})$ and $r_{k^{n}\twoheadrightarrow\lambda}\colon\mathrm{Rep}(G_{k^{n}})\to\mathrm{Rep}(G_{\lambda})$ are defined exactly as before. $P_{k^{n}\twoheadrightarrow\lambda}$ is conjugate to $P_{{\lambda^{\prime}\hookrightarrow k^{n}}}$ and $U_{k^{n}\twoheadrightarrow\lambda}$ is conjugate to $U_{{\lambda^{\prime}\hookrightarrow k^{n}}}$, where $\lambda^{\prime}$ is the partition that is complementary to $\lambda$ in $k^{n}$, i.e., the partition for which $\ker(\mathfrak{o}_{k}^{n}\to\mathfrak{o}_{\lambda})\cong\mathfrak{o}_{\lambda^{\prime}}$. Therefore, the collection of irreducible representations obtained as summands after applying either of the functors $i_{{\lambda\hookrightarrow k^{n}}}$ or $i_{{k^{n}\twoheadrightarrow\lambda}}$ is the same. Following [Onn07], the functors $i_{{\lambda\hookrightarrow k^{n}}}$ and $i_{{k^{n}\twoheadrightarrow\lambda}}$ are called _infinitesimal induction functors_. The functors $r_{\lambda\hookrightarrow k^{n}}$ and $r_{k^{n}\twoheadrightarrow\lambda}$ are called _infinitesimal restriction functors_. ###### Definition 3.2. An irreducible representation of $G_{k^{n}}$ will be said to lie in the infinitesimally induced series if it is isomorphic to a subrepresentation of $i_{\lambda\hookrightarrow k^{n}}\sigma$ for some partition $\lambda\leq k^{n}$ and some representation $\sigma$ of $G_{\lambda}$. ## 4\. Cuspidality and strong cuspidality ### 4.1. The definitions of cuspidality Recall from Section 2.2 that to every irreducible representation $\rho$ of $G_{k^{n}}$ is associated a similarity class $\Omega_{1}(\rho)\subset M_{n}(\mathfrak{o}_{1})$. The following definition was introduced in [Kut80] for $n=2$ and in [Car84] for general $n$. ###### Definition 4.1 (Strong cuspidality). An irreducible representation $\rho$ of $G_{k^{n}}$ is said to be _strongly cuspidal_ if either $k=1$ and $\rho$ is cuspidal, or $k>1$ and $\Omega_{1}(\rho)$ is an irreducible orbit in $M_{n}(\mathfrak{o}_{1})$. In the above definition, one says that an orbit is irreducible if the matrices in it are irreducible, i.e., they do not leave any non-trivial proper subspaces of $\mathfrak{o}_{1}^{n}$ invariant. This is equivalent to saying that the characteristic polynomial of any matrix in the orbit is irreducible. Another notion of cuspidality (which applies for any $G_{\lambda}$, however, we shall focus on $\lambda=k^{n}$) picks out those irreducible representations which can not be constructed from the representations of $G_{\lambda}$, $\lambda\leq k^{n}$ by using the functors defined in Section 3. ###### Definition 4.2 (Cuspidality). An irreducible representation $\rho$ of $G_{k^{n}}$ is said to be _cuspidal_ if no twist of it by a linear character lies in the geometrically or infinitesimally induced series. ### 4.2. Comparison between the definitions ###### Theorem 4.3. Every strongly cuspidal representation is cuspidal. When $n$ is a prime, every cuspidal representation is strongly cuspidal. ###### Proof. Let $\rho$ be an irreducible non-cuspidal representation of $G_{k^{n}}$. The linear characters of $G_{k^{n}}$ are of the form $\det\\!\circ\chi$ for some character $\chi\colon\mathfrak{o}_{k}^{\times}\to\mathbf{C}^{\times}$. Using the identification of $N_{k-1}\simeq M_{n}(\mathfrak{o}_{1})$ with its dual from Section 2.2, the restriction of $\det\\!\circ\chi$ to $N_{k-1}$ is easily seen to be a scalar matrix. Thus $\rho$ is strongly cuspidal if and only if $\rho(\chi)=\rho\otimes\det\\!\circ\chi$ is, since adding a scalar matrix does not effect the irreducibility of the orbit $\Omega_{1}(\rho)$. Since $\rho$ is non-cuspidal, there exists a character $\chi$ such that $\rho(\chi)^{U}$ is nonzero for some $U=U_{n_{1},n_{2}}$ or $U=U_{\lambda\hookrightarrow k^{n}}$. In either case this implies that the orbit $\Omega_{1}\left(\rho(\chi)\right)$ is reducible which in turn implies that $\rho(\chi)$ and hence $\rho$ are not strongly cuspidal. For the converse the following interesting result (for which the hypothesis that $n$ is prime is not necessary) plays an important role. A similar result was obtained by Kutzko in the context of supercuspidal representations of $\mathrm{GL}_{n}$ over a $p$-adic field [Kut85, Prop. 4.6]. Call a similarity class in $M_{n}(\mathfrak{o}_{1})$ _primary_ if its characteristic polynomial has a unique irreducible factor. ###### Proposition 4.4. Let $\rho$ be an irreducible representation of $G_{k^{n}}$. If $\Omega_{1}(\rho)$ is not primary then $\rho$ lies in the geometrically induced series. ###### Proof. If $\Omega_{1}(\rho)$ is not primary then it contains an element $\varphi=\left(\begin{smallmatrix}\hat{w}_{1}&0\\\ 0&\hat{w}_{2}\end{smallmatrix}\right)$ with $\hat{w}_{i}\in M_{n_{i}}(\mathfrak{o}_{1})$ and $n=n_{1}+n_{2}$, such that the characteristic polynomials of $\hat{w}_{1}$ and $\hat{w}_{2}$ have no common factor. It will be shown that $r_{n_{1},n_{2}}(\rho)\neq 0$. In what follows, matrices will be partitioned into blocks according to $n=n_{1}+n_{2}$. Let $P_{i}=P_{(k^{n_{1}},(k-i)^{n_{2}})\hookrightarrow k^{n}}$ for $i=0,\ldots,k$. Then $P_{i}$ consists of matrices in $G_{k^{n}}$ with blocks of the form $\left(\begin{smallmatrix}a&b\\\ \varpi^{i}c&d\end{smallmatrix}\right)$. Let $U_{i}$ be the normal subgroup of $P_{i}$ consisting of block matrices of the form $\left(\begin{smallmatrix}\mathrm{I}&\varpi^{k-i}u\\\ 0&\mathrm{I}\end{smallmatrix}\right)$. The $P_{i}$’s form a decreasing sequence of subgroups, while the $U_{i}$’s form increasing sequences. Given a representation $\rho_{i}$ of $P_{i}/U_{i}$ define $r_{i}(\rho_{i})$ to be the representation of $P_{i+1}/U_{i+1}$ obtained by taking the vectors in the restriction of $\rho_{i}$ to $P_{i+1}$ that are invariant under $U_{i+1}$. That is, $r_{i}\colon\text{Rep}(P_{i}/U_{i})\to\text{Rep}(P_{i+1}/U_{i+1}),\qquad{r}_{i}(\rho_{i})=\text{Inv}_{U_{i+1}/U_{i}}\circ\text{Res}^{P_{i}/U_{i}}_{P_{i+1}/U_{i}}(\rho_{i}).$ In particular, $P_{k}=P_{n_{1},n_{2}}$ and $U_{k}=U_{n_{1},n_{2}}$. Therefore, (see [Onn07, Lemma 7.1]) we have that $r_{n_{1},n_{2}}=r_{k-1}\circ\cdots\circ r_{0}$. We argue by induction that $r_{i}\circ\cdots\circ r_{0}(\rho)\neq 0$ for all $i=0,\ldots,k$. If $i=0$, then since $\varphi\in\Omega_{1}(\rho)$, we get that $\rho_{|U_{1}}$ contains the trivial character of $U_{1}$, hence, $r_{0}(\rho)\neq 0$. Denote $\rho_{i}=r_{i-1}\circ\cdots\circ r_{0}(\rho)$ and assume that $\rho_{i}\neq 0$. In order to show that $r_{i}(\rho_{i})\neq 0$, consider the normal subgroup $L_{i}$ of $P_{i}$ which consists of block matrices of the form $\mathrm{I}+\big{(}\begin{smallmatrix}\varpi^{k-1}w_{1}&\varpi^{k-i-1}u\\\ \varpi^{k-1}v&\varpi^{k-1}w_{2}\end{smallmatrix}\big{)}$. It is easily verified that $L_{i}/U_{i}\simeq M_{n}(\mathfrak{o}_{1})$, the isomorphism given by $\eta\colon\mathrm{I}+\left(\begin{matrix}\varpi^{k-1}w_{1}&\varpi^{k-i-1}u\\\ \varpi^{k-1}v&\varpi^{k-1}w_{2}\end{matrix}\right)\mod{U_{i}}~{}~{}\mapsto~{}~{}\left(\begin{matrix}w_{1}&u\\\ v&w_{2}\end{matrix}\right),$ where $w_{1},w_{2},u$ and $v$ are appropriate block matrices over $\mathfrak{o}_{1}$. It follows that we can identify the dual of ${L_{i}/U_{i}}$ with $M_{n}(\mathfrak{o}_{1})$: $\hat{x}\mapsto\psi_{\hat{x}}\circ\eta$, for $\hat{x}\in M_{n}(\mathfrak{o}_{1})$. The action of $P_{i}$ on the dual of $L_{i}/U_{i}$ is given by $\hat{x}\mapsto g\hat{x}$ where $\psi_{g\hat{x}}(\eta(l))=\psi_{\hat{x}}(\eta(g^{-1}lg))$. We shall not need the general action of elements of $P_{i}$, but rather of a small subgroup which is much easier to handle. If $g_{c}=\left(\begin{matrix}\mathrm{I}&\\\ \varpi^{i}c&\mathrm{I}\end{matrix}\right),\qquad\eta(l)=\left(\begin{matrix}w_{1}&u\\\ v&w_{2}\end{matrix}\right),\qquad\hat{x}=\left(\begin{matrix}\hat{w}_{1}&\hat{v}\\\ \hat{u}&\hat{w}_{2}\end{matrix}\right),$ then unraveling definitions gives (4.5) $\hat{x}\mapsto g_{c}\hat{x}=\left(\begin{matrix}\hat{w}_{1}&\hat{v}\\\ \hat{u}+c\hat{w}_{1}-\hat{w}_{2}c&\hat{w}_{2}\end{matrix}\right).$ As we have identifications $L_{0}/U_{1}=\cdots=L_{i}/U_{i+1}$ we infer that the restriction of $\rho_{i}$ to $L_{i}/U_{i+1}$ contains a character $\psi_{\hat{x}}=(\varphi_{|L_{0}/U_{1}},\hat{u})\colon L_{i}/U_{i+1}\times U_{i+1}/U_{i}=L_{i}/U_{i}\to\mathbf{C}^{\times},$ that is, $\psi_{\hat{x}}$ corresponds to $\hat{x}=\left(\begin{smallmatrix}\hat{w}_{1}&0\\\ \hat{u}&\hat{w}_{2}\end{smallmatrix}\right)$. We claim that there exist $g_{c}$ such that $g_{c}\hat{x}=\left(\begin{matrix}\hat{w}_{1}&0\\\ 0&\hat{w}_{2}\end{matrix}\right),$ therefore $\rho_{i|U_{i+1}/U_{i}}$ contains the trivial character of $U_{i+1}/U_{i}$ and hence $r_{i}(\rho_{i})\neq 0$. Indeed, using (4.5) it is enough to show that the map $c\mapsto c\hat{w}_{1}-\hat{w}_{2}c$ is surjective, hence $\hat{u}$ can be eliminated and the entry $(1,2)$ contains the trivial character. This map is surjective if and only if it is injective. So we show that its kernel is null. A matrix $c$ is in the kernel if and only if (4.6) $c\hat{w}_{1}=\hat{w}_{2}c.$ Let $p_{i}$ ($i=1,2$) be the characteristic polynomials of $\hat{w}_{i}$. Our assumption on the orbits is that $p_{1}$ and $p_{2}$ have disjoint set of roots. Using (4.6) we deduce that $cp_{1}(\hat{w}_{1})=p_{1}(\hat{w}_{2})c.$ By the Cayley-Hamilton theorem the left hand side of the above equation vanishes. Over an algebraic closure of $\mathfrak{o}_{1}$, $p_{1}(t)=\prod(t-\alpha_{j})$, where the $\alpha_{j}$ are the roots of $p_{1}$. The hypothesis on $\hat{w}_{1}$ and $\hat{w}_{2}$ implies that none of these is an eigenvalue of $\hat{w}_{2}$. Therefore, $\hat{w}_{2}-\alpha_{j}$ is invertible for each $j$. It follows that $p_{1}(\hat{w}_{2})=\prod(\hat{w}_{2}-\alpha_{j})$ is also invertible, hence $c=0$. This completes the proof of the proposition. ∎ Returning now to the proof of Theorem 4.3, assume that $\rho$ is not strongly cuspidal. There are two possibilities: 1. (a) Any element $\hat{\omega}\in\Omega_{1}(\rho)$ has eigenvalue in $\mathfrak{o}_{1}$. In such case, by twisting with a one-dimensional character $\chi$, we get a row of zeros in the Jordan canonical form of $\hat{\omega}$. Therefore, $\rho(\chi)$ is contained in a representation infinitesimally induced from $G_{(k^{n-1},k-1)}$. 2. (b) Elements in $\Omega_{1}(\rho)$ have no eigenvalue in $\mathfrak{o}_{1}$. Since $n$ is prime and since $\Omega_{1}(\rho)$ is reducible, the latter cannot be primary, and Proposition 4.4 implies that $\rho$ lies in the geometrically induced series. Thus, $\rho$ is non-cuspidal. ∎ ## 5\. Construction of strongly cuspidal representations The construction of strongly cuspidal representations of $\mathrm{GL}_{n}(\mathfrak{o}_{k})$ when $k>1$ can be found, for example, in [Shi68, Gér75, How77, Car84, BK93, Hil95a]. In this section, we recall this construction in a way that Theorems B and C are seen to follow from it. ### 5.1. Primitive characters Let $E$ denote an unramified extension of $F$ of degree $n$. Let $\mathfrak{O}$ be the integral closure of $\mathfrak{o}$ in $E$. The maximal ideal of $\mathfrak{O}$ is $\mathfrak{P}=\varpi\mathfrak{O}$. Let $\mathfrak{O}_{k}=\mathfrak{O}/\mathfrak{P}^{k}$. As an $\mathfrak{o}_{k}$-module, $\mathfrak{O}_{k}$ is isomorphic to a free $\mathfrak{o}_{k}$-module of rank $n$. Therefore, $G_{k^{n}}$ can be identified with $\mathrm{Aut}_{\mathfrak{o}_{k}}(\mathfrak{O}_{k})$. This identification is determined up to an inner automorphism of $G_{k^{n}}$. Thus, the strongly cuspidal representations constructed in this section are determined up to isomorphism. Left multiplication by elements of $\mathfrak{O}_{k}$ gives rise to $\mathfrak{o}_{k}$-module endomorphisms of $\mathfrak{O}_{k}$. Therefore, $\mathfrak{O}_{k}^{\times}$ can be thought of as a subgroup of $G_{k^{n}}$. Similarly, for each $r\geq k/2$, $\mathfrak{O}_{k-r}$ will be thought of as a subring of $M_{n}(\mathfrak{o}_{k-r})$. Strongly cuspidal representations of $G_{k^{n}}$ will be associated to certain characters of $\mathfrak{O}_{k}^{\times}$ which we will call _strongly primitive_. In order to define a strongly primitive character of $\mathfrak{O}_{k}^{\times}$ it is first necessary to define a primitive character of $\mathfrak{O}_{1}$. ###### Definition 5.1 (Primitive character of $\mathfrak{O}_{1}$). A _primitive character of $\mathfrak{O}_{1}$_ is a homomorphism $\phi\colon\mathfrak{O}_{1}\to\mathbf{C}^{\times}$ which does not factor through any proper subfield via the trace map. The map $\mathfrak{O}_{k}\to\mathfrak{O}_{k}^{\times}$ given by $a\mapsto 1+\varpi^{r}a$ induces an isomorphism $\mathfrak{O}_{k-r}\tilde{\to}\ker(\mathfrak{O}_{k}^{\times}\to\mathfrak{O}_{r}^{\times})$, for each $1\leq r<k$. ###### Definition 5.2 (Strongly primitive character of $\mathfrak{O}_{k}^{\times}$). When $k>1$, a _strongly primitive character of $\mathfrak{O}_{k}^{\times}$_ is a homomorphism $\omega\colon\mathfrak{O}_{k}^{\times}\to\mathbf{C}^{\times}$ whose restriction to $\ker(\mathfrak{O}_{k}^{\times}\to\mathfrak{O}_{k-1}^{\times})$ is a primitive character when thought of as a character of $\mathfrak{O}_{1}$ under the above identification. The above definition does not depend on the choice of uniformizing element $\varpi\in\mathfrak{p}$. Suppose that $r\geq k/2$. An identification $A\mapsto\psi_{A}$ of $M_{n}(\mathfrak{o}_{k-r})$ with its Pontryagin dual was constructed in Section 2.2. Given $a\in\mathfrak{O}_{k-r}$, view it as an element of $M_{n}(\mathfrak{o}_{k-r})$. Let $\phi_{a}$ denote the restriction of $\psi_{a}$ to $\mathfrak{O}_{k-r}$. Then $a\mapsto\phi_{a}$ is an isomorphism of $\mathfrak{O}_{k-r}$ with its Pontryagin dual. ### 5.2. Construction of strongly cuspidal representations from strongly primitive characters The reader may find it helpful to refer to (5.8) while navigating the construction. Let $l=\lceil k/2\rceil$ be the smallest integer not less than $k/2$ and $l^{\prime}=\lfloor k/2\rfloor$ be the largest integer not greater than $k/2$. Let $\omega$ be a strongly primitive character of $\mathfrak{O}_{k}^{\times}$. Let $a\in\mathfrak{O}_{k-l}$ be such that the restriction of $\omega$ to $N_{l}\cap\mathfrak{O}_{k}^{\times}$ (when identified with $\mathfrak{O}_{k-l}$) is of the form $\phi_{a}$. The strong primitivity of $\omega$ implies that the image of $a$ in $\mathfrak{O}_{1}$ does not lie in any proper subfield. The formula (5.3) $\tau_{\omega}(xu)=\psi_{a}(x)\omega(u)\text{ for all }x\in N_{l}\text{ and }u\in\mathfrak{O}_{k}^{\times},$ defines a homomorphism $\tau_{\omega}\colon N_{l}\mathfrak{O}_{k}^{\times}\to\mathbf{C}^{\times}$. Let $L$ denote the kernel of the natural map $\mathfrak{O}_{k}^{\times}\to\mathfrak{O}_{1}^{\times}$. Then $N_{l}L$ is a normal subgroup of $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$ (note that $N_{l}\mathfrak{O}_{k}^{\times}$ is not normal in $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$i, when $k$ is odd). Let $\sigma_{\omega}$ denote the restriction of $\tau_{\omega}$ to $N_{l}L$. We have (5.4) $\sigma_{\omega}(yxy^{-1})=\sigma_{\omega}(x)\text{ for all }y\in N_{l^{\prime}}\mathfrak{O}_{k}^{\times}\text{ and }x\in N_{l}L.$ Let $q$ denote the order and $p$ denote the characteristic of $\mathfrak{o}_{1}$. The quotient $V=N_{l^{\prime}}L/N_{l}L$ is naturally isomorphic to $M_{n}(\mathfrak{o}_{1})/\mathfrak{O}_{1}$ which111Here $M_{n}(\mathfrak{o}_{1})$ is identified with $\mathrm{End}_{\mathfrak{o}_{1}}(\mathfrak{O}_{1})$., being an abelian group where every non-trivial element has order $p$, can be viewed as a vector space over $\mathbf{F}_{p}$ of dimension $(n^{2}-n)\log_{p}q$. Then $\beta(xN_{l}L,yN_{l}L)=\sigma_{\omega}([x,y])\text{ for all }x,y\in N_{l^{\prime}}L,$ defines a non-degenerate alternating bilinear form $\beta\colon V\times V\to\mu_{p}$ , where $\mu_{p}$ denote the complex $p^{\text{th}}$ roots of unity, [Hil95b, Corollary 4.3]. The following lemma now follows from standard results on the representation theory of finite Heisenberg groups (see e.g., [Gér75, Proposition 3]). ###### Lemma 5.5. There exists a unique irreducible representation $\sigma^{\prime}_{\omega}$ of $N_{l^{\prime}}L$ whose restriction to $N_{l}L$ is $\sigma_{\omega}$ isotypic. This representation has dimension $q^{(l-l^{\prime})(n^{2}-n)/2}$. Its character is given by $\mathrm{tr}(\sigma^{\prime}_{\omega}(x))=\begin{cases}q^{(l-l^{\prime})(n^{2}-n)/2}\sigma_{\omega}(x)&\text{if }x\in N_{l}L,\\\ 0&\text{otherwise}.\end{cases}$ Recall from [Ser68, II.4, Proposition 8], that there is a unique multiplicative section $s\colon\mathfrak{O}_{1}^{\times}\to\mathfrak{O}_{k}^{\times}$. This allows us to realize $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$ as a semidirect product of $N_{l^{\prime}}L$ by $\mathfrak{O}_{1}^{\times}$. Recall also, that $x\in\mathfrak{O}_{k}^{\times}$ is called _regular elliptic_ if its image in $\mathfrak{O}_{1}^{\times}$ is not contained in any proper subfield. ###### Lemma 5.6. When $k$ is odd, there exists an irreducible representation $\tau^{\prime}_{\omega}$ of $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$, which is unique up to isomorphism, whose restriction to $N_{l}L$ is $\sigma_{\omega}$-isotypic, and such that for any $x\in N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$, $\mathrm{tr}(\tau^{\prime}_{\omega}(x))=\begin{cases}0\text{ when }x\text{ is not conjugate to an element of }N_{l}\mathfrak{O}_{k}^{\times}\\\ (-1)^{n-1}\omega(x)\text{ when }x\in\mathfrak{O}_{k}^{\times}\text{ is regular elliptic .}\end{cases}$ ###### Proof. The lemma is easily deduced from [Gér75, Theorem 1] as follows: the algebraic torus $T$ defined over $\mathbf{F}_{q}$ such that $T(\mathbf{F}_{q})=\mathfrak{O}_{1}^{\times}$ splits over the extension $\mathbf{F}_{q^{n}}$ of $\mathbf{F}_{q}$. The Galois group of this extension acts on the weights of $T(\mathbf{F}_{q^{n}})$ on $V\otimes\mathbf{F}_{q^{n}}$, which simply correspond to roots of $\mathrm{GL}_{n}$. The Frobenius automorphism which generates this group acts as a Coxeter element on this root system. One may see that, in the language of [Gér75, 1.4.9(b)], this action has a unique symmetric orbit and $(n-2)/2$ non- symmetric orbits if $n$ is even, and no symmetric orbits and $(n-1)/2$ non- symmetric orbits if $n$ is odd. The symmetric orbits contribute a factor of $(-1)$ to the character values. The hypothesis that $u$ is not an element of any proper subfield of $\mathfrak{O}_{1}$ implies that $u$ is regular semisimple, and that no weight vanishes on it. ∎ When $k$ is even, define the representation $\tau^{\prime}_{\omega}$ of $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$ to be just $\tau_{\omega}$ (see (5.3)). Then, for any $k>1$, if $u\in\mathfrak{O}_{k}^{\times}$ is an element whose image in $\mathfrak{O}_{1}^{\times}$ is a generates $\mathfrak{O}_{1}^{\times}$, we have (5.7) $\mathrm{tr}(\tau^{\prime}_{\omega}(u))=(-1)^{k(n-1)}\omega(u).$ Finally, define $\rho_{\omega}=\mathrm{Ind}_{N_{l^{\prime}}\mathfrak{O}_{k}^{\times}}^{G_{k^{n}}}\tau^{\prime}_{\omega}.$ This will be the strongly cuspidal representation associated to the strongly primitive character $\omega$ of $\mathfrak{O}_{k}^{\times}$. The representation $\rho_{\omega}$ is irreducible because $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$ is the centralizer of $\sigma_{\omega}$ in $G_{k^{n}}$. The steps in the construction of $\rho_{\omega}$ are described schematically below for the convenience of the reader. The diagram on the left describes the relation between the various groups involved. The position occupied by a group in the diagram on the left is occupied by the corresponding representation that appears in the construction in the diagram on the right. (5.8) $\textstyle{G_{k^{n}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$ $\textstyle{\rho_{\omega}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{N_{l^{\prime}}\mathfrak{O}_{k}^{\times}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$ $\textstyle{\tau^{\prime}_{\omega}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{N_{l^{\prime}}L\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{N_{l}\mathfrak{O}_{k}^{\times}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$ $\textstyle{\sigma^{\prime}_{\omega}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{\tau_{\omega}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{N_{l}L\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{\mathfrak{O}_{k}^{\times}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\qquad}$$\textstyle{\sigma_{\omega}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{\omega\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{N_{l}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{L\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{\qquad\psi_{a}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{\omega_{|L}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{N_{l}\cap L}$ $\textstyle{\phi_{a}}$ ###### Theorem 5.9. For each strongly primitive character $\omega$ of $\mathfrak{O}_{k}^{\times}$, $\rho_{\omega}$ is an irreducible representation such that 1. (1) $\mathrm{tr}(\rho_{\omega}(g))=0$ if $g$ is not conjugate to an element of $N_{l}\mathfrak{O}_{k}^{\times}$. 2. (2) if $u\in\mathfrak{O}_{k}^{\times}$ is such that its image in $\mathfrak{O}_{1}^{\times}$ is not contained in any proper subfield, then $\mathrm{tr}(\rho_{\omega}(u))=(-1)^{k(n-1)}\sum_{\gamma\in\mathrm{Gal}(E/F)}\omega({}^{\gamma}u)).$ for every $u\in\mathfrak{O}_{k}^{\times}$, whose image in $\mathfrak{O}_{1}^{\times}$ lies in no proper subfield. ###### Proof. The first assertion follows from Lemma 5.5. The second follows from the fact that the intersection of the conjugacy class of $u$ in $G_{k^{n}}$ with $\mathfrak{O}_{k}^{\times}$ consists only of the elements ${}^{\gamma}u$, for $\gamma\in\mathrm{Gal}(E/F)$. ∎ ### 5.3. The parameterization of strongly cuspidal representations of $G_{k^{n}}$ The following is a detailed version of Theorem B. ###### Theorem 5.10. 1. (1) For each strongly primitive character $\omega$ of $\mathfrak{O}_{k}^{\times}$, the representation $\rho_{\omega}$ of $G_{k^{n}}$ is irreducible and strongly cuspidal. 2. (2) Every strongly cuspidal representation of $G_{k^{n}}$ is isomorphic to $\rho_{\omega}$ for some strongly primitive character $\omega$ of $\mathfrak{O}_{k}^{\times}$. 3. (3) If $\omega^{\prime}$ is another strongly primitive character of $\mathfrak{O}_{k}^{\times}$, then $\rho_{\omega}$ is isomorphic to $\rho_{\omega^{\prime}}$ if and only if $\omega^{\prime}=\omega\circ\gamma$ for some $\gamma\in\mathrm{Gal}(E/F)$. ###### Proof of (1). The irreducibility of $\rho_{\omega}$ follows from standard results on induced representations. To see that $\rho_{\omega}$ is strongly cuspidal, observe that the restriction of $\rho_{\omega}$ to $N_{l}$ contains $\psi_{a}$. This means that its restriction to $N_{k-1}$ contains $\psi_{\overline{a}}$, where $\overline{a}$ is the image of $a$ in $\mathfrak{O}_{1}$. Since this image does not lie in any proper subfield, its minimal polynomial is irreducible of degree $n$. Therefore, as an element of $M_{n}(\mathfrak{o}_{1})$, its characteristic polynomial must be irreducible. ∎ ###### Proof of (2). Suppose that $\rho$ is an irreducible strongly cuspidal representation of $G_{k^{n}}$. Unwinding the definitions, one see that $\Omega_{1}(\rho)$ is just the image of $\Omega_{k-l}(\rho)$ under the natural map $M_{n}(\mathfrak{o}_{k-l})\to M_{n}(\mathfrak{o}_{1})$. Let $p(t)\in\mathfrak{o}_{k-l}[t]$ be the characteristic polynomial of the matrices in $\Omega_{k-l}(\rho)$. Denote its image in $\mathfrak{o}_{1}[t]$ by $\overline{p}(t)$. The hypothesis on $\rho$ implies that $\overline{p}(t)$ is irreducible. Let $\tilde{p}(t)$ be any polynomial in $\mathfrak{o}[t]$ whose image in $\mathfrak{o}_{k-l}[t]$ is $p(t)$. By Hensel’s lemma, there is a bijection between the roots of $\tilde{p}(t)$ in $E$ and the roots of $\overline{p}(t)$ in $\mathfrak{O}_{1}$. Consequently, $\mathrm{Hom}_{F}(F[t]/\tilde{p}(t),E)\cong\mathrm{Hom}_{\mathfrak{o}_{1}}(\mathfrak{o}_{1}[t]/\overline{p}(t),\mathfrak{O}_{1}).$ But we know that $\mathfrak{O}_{1}$ is isomorphic to $\mathfrak{o}_{1}[t]/\overline{p}(t)$. In fact there are exactly $n$ such isomorphisms. Each one of these gives an embedding of $F[t]/\tilde{p}(t)$ in $E$. Since both $F[t]/\tilde{p}(t)$ and $E$ have degree $n$, these embeddings must be isomorphisms. Any root $\tilde{a}$ of $\tilde{p}(t)$ in $E$ also lies in $\mathfrak{O}$. It is conjugate to the companion matrix of $\tilde{p}(t)$ in $\mathrm{GL}_{n}(\mathfrak{o})$. Therefore, its image $a\in\mathfrak{O}_{k-l}$ lies in $\Omega_{k-l}(\rho)$. It follows that $\rho_{|N_{l}}$ contains a $\psi_{a}$ isotypic vector. By applying the little groups method of Wigner and Mackey to the normal subgroup $N_{l}$ of $G_{k^{n}}$, we see that every representation of $\rho_{k}$ whose restriction to $N_{l}$ has a $\psi_{a}$ isotypic vector is induced from an irreducible representation of $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$ whose restriction to $N_{l}$ is $\psi_{a}$ isotypic. It is not difficult then to verify (by counting extensions at each stage) that the construction of $\tau^{\prime}_{\omega}$ in Section 5.2 gives all such representations. ∎ ###### Proof of (3). It follows from the proof of (2) that $\tau^{\prime}_{\omega_{1}}$ and $\tau^{\prime}_{\omega_{2}}$ are isomorphic if and only if $\omega_{1}=\omega_{2}$. The Galois group $\mathrm{Gal}(E/F)$ acts by inner automorphisms of $G_{k^{n}}$ (since we have identified it with $\mathrm{Aut}_{\mathfrak{o}_{k}}(\mathfrak{O}_{k})$) preserving $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$. Therefore, the restriction of $\rho_{\omega_{1}}$ to $N_{l^{\prime}}\mathfrak{O}_{k}^{\times}$ also contains $\tau_{\omega_{2}}$ whenever $\omega_{2}$ is in the $\mathrm{Gal}(E/F)$-orbit of $\omega_{1}$, hence $\rho_{\omega_{1}}$ is isomorphic to $\rho_{\omega_{2}}$. If $\omega_{1}$ and $\omega_{2}$ do not lie in the same $\mathrm{Gal}(E/F)$-orbit then Theorem 5.9 implies that that $\rho_{\omega_{1}}$ can not be isomorphic to $\rho_{\omega_{2}}$. ∎ ### 5.4. Connection with supercuspidal representations of $\mathrm{GL}_{n}(F)$ In [BK93, Theorem 8.4.1], Bushnell and Kutzko proved that all the irreducible supercuspidal representations of $\mathrm{GL}_{n}(F)$ can be obtained by compact induction from a compact subgroup modulo the center. One such subgroup is $F^{\times}\mathrm{GL}_{n}(\mathfrak{o})$. This group is a product of $\mathrm{GL}_{n}(\mathfrak{o})$ with the infinite cyclic group $Z_{1}$ generated by $\varpi\mathrm{I}$. Thus every irreducible representation of this group is a product of a character of $Z_{1}$ with an irreducible representation of $\mathrm{GL}_{n}(\mathfrak{o})$. An irreducible representation of $\mathrm{GL}_{n}(\mathfrak{o})$ is said to be of level $k-1$ if it factors through $\mathrm{GL}_{n}(\mathfrak{o}_{k})$, but not through $\mathrm{GL}_{n}(\mathfrak{o}_{k-1})$. When $n$ is prime, the representations of $\mathrm{GL}_{n}(\mathfrak{o})$ which give rise to supercuspidal representations are precisely those which are of level $k-1$, for some for $k>1$, and, when viewed as representations of $\mathrm{GL}_{n}(\mathfrak{o}_{k})$, are strongly cuspidal. For $k=1$, they are just the cuspidal representations of $\mathrm{GL}_{n}(\mathfrak{o}_{1})$. The corresponding representations of $Z\mathrm{GL}_{n}(\mathfrak{o})$ are called _très cuspidale de type $k$_ by Carayol in [Car84, Section 4.1]. The construction that Carayol gives for these representations is the same as the one given here, except that the construction here is made canonical by using Gérardin’s results. Let $\chi$ be any character of $Z_{1}$. Set $\pi_{\omega,\chi}:=\mathrm{c}\text{-}\mathrm{Ind}_{\mathrm{GL}_{n}(\mathfrak{o})F^{\times}}^{\mathrm{GL}_{n}(F)}(\rho_{\omega}\otimes\chi).$ These are the supercuspidal representations of $\mathrm{GL}_{n}(F)$ associated to $\rho_{\omega}$. Let $r\colon\mathrm{GL}_{n}(\mathfrak{o})\to\mathrm{GL}_{n}(\mathfrak{o}_{k})$ denote the homomorphism obtained by reduction modulo $\mathfrak{p}^{k}$. In the notation of [BK93], we have $r^{-1}(N_{l}L)=H^{1}(\beta,\mathfrak{A})$, $r^{-1}(N_{l^{\prime}}L)=J^{1}(\beta,\mathfrak{A})$ and $r^{-1}(N_{l^{\prime}}\mathfrak{O}_{k}^{\times})=J(\beta,\mathfrak{A})$, where $\mathfrak{A}=M_{n}(\mathfrak{o})$ and $\beta\in M_{n}(F)$ is minimal (see [BK93, (1.4.14)]). These groups are very special cases of the groups defined in [BK93, (3.1.14)]. The inflation $\eta$ of $\sigma_{\omega^{\prime}}$ to $J^{1}(\beta,\mathfrak{A})$ is a special case of the Heisenberg representation defined in [BK93, Prop. 5.1.1]. We will say that a supercuspidal representation $\pi$ of $\mathrm{GL}_{n}(F)$ belongs to the _unramified series_ if the field extension $F[\beta]$ of $F$ is unramified (by [BK93, (1.2.4), (6.2.3) (i)], this is equivalent to saying that the $\mathfrak{o}$-order $\mathfrak{A}$ occurring in the construction of $\pi$ is maximal). When $n$ is a prime number, Carayol has proved (see [Car84, Theorem 8.1 (i)]) that the representations $\pi_{\omega,\chi}$ give all the supercuspidal representations of $\mathrm{GL}_{n}(F)$ which belong to the unramified series. However, when $n$ is composite, the strongly cuspidal representations are not sufficient in order to build all the supercuspidal representations in the unramified series of $\mathrm{GL}_{n}(F)$ (see for instance Howe’s construction in [How77]). Since all the supercuspidal representations of $\mathrm{GL}_{n}(F)$ are known [BK93], it would be natural to try restricting them to $\mathrm{GL}_{n}(\mathfrak{o})$ and see if one get cuspidal representations among the components. On the other hand we observe that our notion of cuspidality is in a sense stronger than the usual notion of supercuspidality for representations of $\mathrm{GL}_{n}(F)$, since supercuspidality can only see geometric induction. ## 6\. Complexity of the classification problem In this section it will be shown that the representation theory of the family of groups $G_{k^{n}}$ actually involves the much larger family, $G_{\lambda,E}$ ($\lambda\in\Lambda$, $E/F$ unramified), which was defined in Section 2.1, even when $k=2$. ###### Theorem 6.1. Let $F=\mathbf{F}_{q}((\varpi))$ be a local function field. Then the problems of constructing all the irreducible representations of the following groups are equivalent: 1. (1) $G_{2^{n},F}$ for all $n\in\mathbf{N}$. 2. (2) $G_{k^{n},F}$ for all $k,n\in\mathbf{N}$. 3. (3) $G_{\lambda,E}$ for all partitions $\lambda$ and all unramified extensions $E$ of $F$. ###### Proof. Obviously (3) implies (2), which implies (1). That (1) implies (3) follows from the somewhat more precise formulation in Theorem 6.2. ∎ ###### Theorem 6.2. Let $F$ be a local function field. Then the problem of constructing all the irreducible representations of $G_{2^{n},F}$ is equivalent to the problem of constructing all the irreducible representations of all the groups $G_{\lambda,E}$, where $E$ ranges over all unramified extensions of $F$ of degree $d$ and $\lambda$ ranges over all partitions such that $d(\lambda_{1}r_{1}+\cdots+\lambda_{l}r_{l})\leq n$. ###### Proof. When $F$ is a local function field, $G_{2^{n}}$ is isomorphic to the semidirect product of $\mathrm{GL}_{n}(\mathfrak{o}_{1})$ by $M_{n}(\mathfrak{o}_{1})$. The _little groups method_ of Wigner and Mackey (see e.g., [Ser77, Prop. 25]) shows that constructing the irreducible representations of $G_{2^{n}}$ is equivalent to constructing the irreducible representations of the centralizers in $\mathrm{GL}_{n}(\mathfrak{o}_{1})$ of all the multiplicative characters of $M_{n}(\mathfrak{o}_{1})$. Pick any $\chi$ for which the space $V_{\chi}$ of $\chi$-isotypic vectors is non-zero. By the discussion in Section 2.2, these subgroups of $\mathrm{GL}_{n}(\mathfrak{o}_{1})$ are the same as the centralizer groups of matrices. We will see below that these centralizer groups are products of groups of the form $G_{\lambda,E}$ that appear in the statement of Theorem 6.2. Let $A\in M_{n}(\mathfrak{o}_{1})$. Then, $\mathfrak{o}_{1}^{n}$ can be thought of as a $\mathfrak{o}_{1}[\varpi]$-module where $\varpi$ acts through $A$. The centralizer of $A$ is the automorphism group of this $\mathfrak{o}_{1}[\varpi]$-module. For each irreducible monic polynomial $f(\varpi)\in\mathfrak{o}_{1}[\varpi]$ of degree $d$ which divides the characteristic polynomial of $A$, the $f$-primary part of this module is isomorphic to $(\mathfrak{o}_{1}[\varpi]/f(\varpi)^{\lambda_{1}})^{r_{1}}\oplus\cdots\oplus(\mathfrak{o}_{1}[\varpi]/f(\varpi)^{\lambda_{l}})^{r_{l}},$ for some partition $\lambda$. ###### Lemma 6.3. Let $\mathfrak{O}_{1}=\mathfrak{o}_{1}[\varpi]/f(\varpi)$. The rings $\mathfrak{o}_{1}[\varpi]/f(\varpi)^{k}$ and $\mathfrak{O}_{1}[u]/u^{k}$ are isomorphic for every $k>0$. ###### Proof. It will be shown by induction that there exists a sequence $\\{q_{k}(\varpi)\\}$, in $\mathfrak{o}_{1}[\varpi]$ such that 1. (1) $q_{1}(\varpi)=\varpi$, 2. (2) $q_{k+1}(\varpi)\equiv q_{k}(\varpi)\mod f(\varpi)^{k}$ for all $k>0$, and, 3. (3) $f(q_{k}(\varpi))\in f(\varpi)^{k}$ for all $k>0$. For $k=1$ the result is obvious. Suppose that $q_{k}(\varpi)$ has been constructed. Since $\mathfrak{o}_{1}$ is a perfect field and $f(\varpi)$ is irreducible, $f^{\prime}(\varpi)$ is not identically $0$. It follows that $f^{\prime}(\varpi)$ does not divide $f(\varpi)$. Since $q_{k}(\varpi)\equiv\varpi\mod f(\varpi)$, $f^{\prime}(q_{k}(\varpi))$ does not divide $f(\varpi)$. Therefore, the congruence $f(q_{k}(\varpi))+f(\varpi)^{k}h(\varpi)f^{\prime}(q_{k}(\varpi))\equiv 0\mod f(\varpi)^{k+1}$ can be solved for $h(\varpi)$. Let $h_{0}(\varpi)$ be a solution. Take $q_{k+1}(\varpi)=q_{k}(\varpi)+f(\varpi)^{k}h_{0}(\varpi)$. The sequence $\\{q_{k}(\varpi)\\}$ constructed in this manner has the required properties. Now note that $\mathfrak{O}_{1}[u]/u^{k}\cong\mathfrak{o}_{1}[\varpi,u]/(f(\varpi),u^{k})$. One may define a ring homomorphism $\mathfrak{o}_{1}[\varpi,u]/(f(\varpi),u^{k})\to\mathfrak{o}_{1}[\varpi]/f(\varpi)^{k}$ by $\varpi\mapsto q_{k}(\varpi)$ and $u\mapsto f(\varpi)$. Since $q_{k}(\varpi)\equiv\varpi\mod f(\varpi)$, $\varpi$ lies in the image of this map, so it is surjective. As vector spaces over $\mathfrak{o}_{1}$ both rings have dimension $kd$. Therefore, it is an isomorphism. ∎ It follows from Lemma 6.3 that the automorphism group of the $f$-primary part of $\mathfrak{o}_{1}^{n}$ is $G_{\lambda,E}$, where $E$ is an unramified extension of $F$ of degree $d$. The automorphism group of the $\mathfrak{o}_{1}[\varpi]$-module $\mathfrak{o}_{1}^{n}$ is the product of the automorphism groups of its $f$-primary parts. Therefore, the centralizer of $A$ in $G_{1^{n}}$ is a product of groups of the form $G_{\lambda,E}$. Considerations of dimension show that $d(\lambda_{1}r_{1}+\cdots+\lambda_{l}r_{l})\leq n$ for each $G_{\lambda,E}$ that occurs. Conversely given $\lambda$ and $d$ satisfying the above inequality, take an irreducible polynomial $f(\varpi)\in\mathfrak{o}_{1}[\varpi]$ of degree $d$. Define $J_{k}(f)=\begin{pmatrix}C_{f}&0&0&\cdots&0&0\\\ \mathrm{I}_{d}&C_{f}&0&\cdots&0&0\\\ 0&\mathrm{I}_{d}&C_{f}&\cdots&0&0\\\ \vdots&\vdots&\vdots&\ddots&\vdots&\vdots\\\ 0&0&0&\cdots&C_{f}&0\\\ 0&0&0&\cdots&\mathrm{I}_{d}&C_{f}\end{pmatrix}_{kd\times kd},$ where $C_{f}$ is any matrix with characteristic polynomial $f$. Let $A=J_{\lambda_{1}}(f)^{\oplus r_{1}}\oplus\cdots\oplus J_{\lambda_{l}}(f)^{\oplus r_{l}}\oplus J_{k}(\varpi-a),$ where $a\in\mathfrak{o}_{1}$ is chosen so that $\varpi-a\neq f(\varpi)$ and $k=n-d(\lambda_{1}r_{1}+\cdots+\lambda_{l}r_{l})$. The centralizer of $A$ contains $G_{\lambda,E}$ as a factor. ∎ ## 7\. The zeta function associated to $G_{k^{n}}$ In [Spr75], Springer attaches a zeta function to irreducible representations of $\mathrm{GL}_{n}(\mathfrak{o}_{1})$, and proves that for cuspidal representations it satisfies a functional equation. Later on, Macdonald [Mac80] shows that a functional equation holds for any irreducible representation, provided that it has no $1$-component, namely, it is not contained in $i_{n-1,1}(\rho,1)$ for any representation $\rho$ of $\mathrm{GL}_{n-1}(\mathfrak{o}_{1})$. Moreover, Macdonald establishes a bijection between irreducible representations of $\mathrm{GL}_{n}(\mathfrak{o}_{1})$, and equivalence classes of tamely ramified representations of the Weil-Deligne group $W^{\prime}_{F}$, which preserves certain $L$ and $\varepsilon$ factors. In this section we attach a zeta function to any irreducible representation of $G_{k^{n}}$ and show that it satisfies a functional equation, provided that $\rho$ does not lie in the infinitesimally induced series. We follow closely [Mac80] and make the necessary adaptations. The map $F\to\mathbf{C}^{\times}$ given by $x\mapsto\psi(\pi^{k}x)$, when restricted to $\mathfrak{o}$, factors through an additive character $\psi_{k}$ of $\mathfrak{o}_{k}$, which does not factor through $\mathfrak{o}_{k-1}$. Denote $G=G_{k^{n}}$ and $M=M_{k^{n}}=M_{n}(\mathfrak{o}_{k})$, and let $\mathbf{C}(M)$ denote complex valued functions on $M$. For $f\in\mathbf{C}(M)$ define its Fourier transform by $\hat{f}(x)=|M|^{-1/2}\sum_{y\in M}f(y)\psi_{k}\left(\mathrm{tr}(xy)\right),$ so that $\hat{\hat{f}}(x)=f(-x)$. Let $(\rho,V)$ be a finite dimensional representation of $G$. For each $f\in\mathbf{C}(M)$ define the zeta-function $\mathcal{Z}(f,\rho)=\sum_{g\in G}f(g)\rho(g)\in\text{End}_{\mathbf{C}}(V).$ Also, for $x\in M$ let $\mathcal{W}(\rho,\psi;x)=|M|^{-1/2}\sum_{g\in G}\psi_{k}(\mathrm{tr}(gx))\rho(g).$ The following lemma is straightforward. ###### Lemma 7.1. 1. (a) $\mathcal{Z}(f,\rho)=\sum_{x\in M}\hat{f}(-x)\mathcal{W}(\rho,\psi;x)$. 2. (b) $\mathcal{W}(\rho,\psi;xg)=\rho(g)^{-1}\mathcal{W}(\rho,\psi;x)$. 3. (c) $\mathcal{W}(\rho,\psi;gx)=\mathcal{W}(\rho,\psi;x)\rho(g)^{-1}$. In particular, setting $x=1$ in parts (b)-(c) of Lemma 7.1 shows that $\mathcal{W}(\rho,\psi;1)$ commutes with $\rho(g)$ for all $g\in G$. Therefore, if $\rho$ is irreducible, then $\mathcal{W}(\rho,\psi;1)$ is a scalar multiple of $\rho(1)$. Following [Mac80] we write $\varepsilon(\rho,\psi)\rho(1)=\mathcal{W}(\check{\rho},\psi;1)$, where $\check{\rho}$ is the contragredient of $\rho$, i.e. $\check{\rho}(g)={{}^{t}\rho(g^{-1})}$ and $\varepsilon(\rho,\psi)$ is a complex number. ###### Proposition 7.2. Let $\rho$ be an irreducible representation of $G$ which does not lie in the infinitesimally induced series. Then $\mathcal{W}(\rho,\psi;x)=0$ for all $x\in M\smallsetminus G$. ###### Proof. Let $H_{x}=\\{g\in G~{}|~{}gx=x\\}$. For $g\in H_{x}$ we have $\mathcal{W}(\rho,\psi;x)=\mathcal{W}(\rho,\psi;gx)=\mathcal{W}(\rho,\psi;x)\rho(g^{-1})=\mathcal{W}(\rho,\psi;x)\rho(e_{H_{x}}),$ where $\rho(e_{H_{x}})=|H_{x}|^{-1}\sum_{g\in H_{x}}\rho(g)$. Hence, it suffices to show that $\rho(e_{H_{x}})=0$ for $x\in M\smallsetminus G$. Since $\rho(e_{H_{x}})$ is the idempotent projecting $V$ onto $V^{H_{x}}$, it is enough to to show that the latter subspace is null. Let $\mu=(\mu_{1},\ldots,\mu_{n})$ be the divisor type of $x$. Namely, $0\leq\mu_{1}\leq\cdots\leq\mu_{n}\leq k$, such that acting with $G$ on the right and on the left gives: $gxh=d_{\mu}=\text{diag}(\varpi^{\mu_{1}},\ldots,\varpi^{\mu_{n}})$. Then $H_{x}=gH_{d_{\mu}}g^{-1}$. Now for any $\mu$ we have $H_{d_{\mu}}\supset H_{d_{\nu}}$, where $\nu=(0,0,\ldots,0,1)$. Therefore, it is enough to show that $V^{H_{d_{\nu}}}=(0)$. The subgroup $H_{d_{\nu}}$ is given explicitly by $H_{d_{\nu}}=\left[\begin{matrix}\mathrm{I}_{n-1}&\varpi^{k-1}\star\\\ 0&1+\varpi^{k-1}\star\end{matrix}\right]=U_{(k^{n-1},k-1)\hookrightarrow k^{n}}\text{ (see Section \ref{subsec:infinitesimal})}.$ It follows that $V^{H_{d_{\nu}}}=(0)$ if $\rho$ does not lie in the infinitesimally induced series. ∎ ###### Theorem 7.3. For all $f\in\mathbf{C}(M)$ and all irreducible representations $\rho$ of $G$ that do not lie in the infinitesimally induced series, we have ${{}^{t}\mathcal{Z}}(\hat{f},\check{\rho})=\varepsilon(\rho,\psi)\mathcal{Z}(f,\rho).$ ###### Proof. If $\rho$ does not lie in the infinitesimally induced series then nor does $\check{\rho}$, and hence $\mathcal{W}(\check{\rho},\psi;x)=0$ for all $x\in M\smallsetminus G$. We get $\begin{split}{{}^{t}\mathcal{Z}}(\hat{f},\check{\rho})&=\sum_{g\in G}{\hat{\hat{f}}(-g){{}^{t}\mathcal{W}}(\check{\rho},\psi;g)}\qquad\qquad\qquad\qquad\qquad\qquad\text{(by Lemma \ref{lem}(a))}\\\ &=\mathcal{W}(\check{\rho},\psi;1)\sum_{g\in G}f(g)\rho(g)=\varepsilon(\rho,\psi)\mathcal{Z}(f,\rho)\qquad\qquad\text{(by Lemma \ref{lem}(c))}.\end{split}$ ∎ The possibility of relating representations of $G_{k^{n}}$ with some equivalence classes of representations of the Weil-Deligne group $W^{\prime}_{F}$, and consequently extending Macdonald correspondence to higher level, seems very appealing. However, such correspondence, if exists, is expected to be much more involved in view of the complexity of the representation theory of $G_{k^{n}}$. ## 8\. Cuspidal representations which are not strongly cuspidal In this section we give a description of all the cuspidal representations of $G_{2^{4}}\cong\text{GL}_{4}(\mathfrak{o}_{2})$ in the sense of Definition 4.2. This shows in particular the existence of representations which are cuspidal, yet are not strongly cuspidal. Let $\lambda=(2^{4})$ and put $G=G_{\lambda}$. If $\pi$ is a cuspidal representation of $G$, then by Proposition 4.4 it is primary, that is, its orbit in $M_{4}(\mathbf{F}_{q})$ consists of matrices whose characteristic polynomial is of the form $f(X)^{a}$, where $f(X)$ is an irreducible polynomial. If $a=1$, then $\pi$ is strongly cuspidal (by definition), and such representations were described in Section 5. On the other hand, $f(X)$ cannot have degree 1, because then it would be infinitesimally induced from $G_{(2^{3},1)}$, up to $1$-dimensional twist (cf. the end of the proof of Theorem 4.3). We are thus reduced to considering representations whose characteristic polynomial is a reducible power of a non-linear irreducible polynomial. In the situation we are considering, there is only one such possibility, namely the case where $f(X)$ is quadratic, and $a=2$. Let $\eta$ denote an element which generates the extension $\mathbf{F}_{q^{2}}/\mathbf{F}_{q}$. We consider $M_{2}(\mathbf{F}_{q^{2}})$ as embedded in $M_{4}(\mathbf{F}_{q})$ via the embedding $\mathbf{F}_{q^{2}}\hookrightarrow M_{2}(\mathbf{F}_{q})$, by choosing the basis $\\{1,\eta\\}$ for $\mathbf{F}_{q^{2}}$ over $\mathbf{F}_{q}$. Rational canonical form implies that in $M_{4}(\mathbf{F}_{q})$ there are two conjugation orbits containing elements with two equal irreducible $2\times 2$ blocks on the diagonal, one regular, and one which is not regular (we shall call the latter _irregular_), represented by the following elements, respectively: $\beta_{1}=\begin{pmatrix}\eta&1\\\ 0&\eta\end{pmatrix},\qquad\beta_{2}=\begin{pmatrix}\eta&0\\\ 0&\eta\end{pmatrix}.$ Therefore, any irreducible cuspidal non-strongly cuspidal representation of $G$ has exactly one of the elements $\beta_{1}$ or $\beta_{2}$ in its orbit. Denote by $N_{1}\cong 1+\varpi M_{2}(\mathfrak{o}_{2})$ the kernel of the reduction map $G=G_{2^{4}}\rightarrow G_{1^{4}}$. As in Section 2.2, let $\psi$ be a fixed non-trivial additive character of $F$, trivial on $\mathfrak{o}$. Then for each $\beta\in M_{4}(\mathbf{F}_{q})$ we have a character $\psi_{\beta}:N_{1}\rightarrow\mathbf{C}^{\times}$ defined by $\psi_{\beta}(1+\varpi x)=\psi(\mathrm{\varpi^{-1}Tr}(\beta x)).$ The group $G$ acts on its normal subgroup $N_{1}$ via conjugation, and thus on the set of characters of $N_{1}$ via the “coadjoint action”. For any character $\psi_{\beta}$ of $N_{1}$, we write $G(\psi_{\beta}):=\operatorname{Stab}_{G}(\psi_{\beta}).$ By Proposition 2.3 in [Hil95b], the stabilizer $G(\psi_{\beta})$ is the preimage in $G$ of the centralizer $C_{G_{1^{4}}}(\beta)$, under the reduction mod $\mathfrak{p}$ map. By definition, an irreducible representation $\pi$ of $G$ is cuspidal iff none of its 1-dimensional twists $\pi\otimes\chi\circ\det$ has any non-zero vectors fixed under any group $U_{i,j}$ or $U_{\lambda\hookrightarrow 2^{4}}$, or equivalently (by Frobenius reciprocity), if $\pi\otimes\chi\circ\det$ does not contain the trivial representation $\mathbf{1}$ when restricted to $U_{i,j}$ or $U_{\lambda\hookrightarrow 2^{4}}$. The groups $U_{i,j}$ are analogs of unipotent radicals of (proper) maximal parabolic subgroups of $G$, and $U_{\lambda\hookrightarrow 2^{4}}$ are the infinitesimal analogs of unipotent radicals (cf. Section 3). Note that since $\mathrm{Ind}_{U_{i,j}}^{G}\mathbf{1}=\mathrm{Ind}_{U_{i,j}}^{G}(\mathbf{1}\otimes\chi\circ\det)=(\mathrm{Ind}_{U_{i,j}}^{G}\mathbf{1})\otimes\chi\circ\det$, for any character $\chi:\mathfrak{o}_{2}^{\times}\rightarrow\mathbf{C}^{\times}$, a representation is a subrepresentation of a geometrically induced representation if and only if all its one-dimensional twists are. In our situation, that is, for $n=4$ and $k=2$, there are three distinct geometric stabilizers, $P_{1,3}$, $P_{2,2}$, and $P_{3,1}$ with “unipotent radicals” $U_{1,3}$, $U_{2,2}$, and $U_{3,1}$, respectively. Thus a representation is a subrepresentation of a geometrically induced representation if and only if it is a component of $\mathrm{Ind}_{U_{i,j}}^{G}\mathbf{1}$, for some $(i,j)\in\\{(1,3),(2,2),(3,1)\\}$. Furthermore, there are three partitions, written in descending order, which embed in $2^{4}$ and give rise to non- trivial infinitesimal induction functors, namely $(2,1^{3}),\ (2^{2},1^{2}),\ (2^{3},1).$ Thus a representation is a subrepresentation of an infinitesimally induced representation if and only if it is a component of $\mathrm{Ind}_{U_{\lambda\hookrightarrow 2^{4}}}^{G}\mathbf{1}$, for some partition $\lambda$ as above. Because of the inclusions $U_{(2,1^{3})\hookrightarrow 2^{4}}\subset U_{(2^{2},1^{2})\hookrightarrow 2^{4}}\subset U_{(2^{3},1)\hookrightarrow 2^{4}},$ an irreducible representation of $G$ is a component of an infinitesimally induced representation if and only if it is a component of $\mathrm{Ind}_{U_{(2,1^{3})\hookrightarrow 2^{4}}}^{G}\mathbf{1}$. ###### Lemma 8.1. Suppose that $\pi$ is an irreducible representation of $G$ whose orbit contains either $\beta_{1}$ or $\beta_{2}$. Then $\pi$ is not an irreducible component of any representation geometrically induced from $P_{1,3}$ or $P_{3,1}$. Moreover, no $1$-dimensional twist of $\pi$ is an irreducible component of an infinitesimally induced representation. ###### Proof. If $\pi$ were a component of $\mathrm{Ind}_{U_{1,3}}^{G}\mathbf{1}$, then $\langle\pi|_{U_{1,3}},\mathbf{1}\rangle\neq 0$, so in particular $\langle\pi|_{N_{1}\cap U_{1,3}},\mathbf{1}\rangle\neq 0$, which implies that $\pi|_{N_{1}}$ contains a character $\psi_{b}$, where $b=(b_{ij})$ is a matrix such that $b_{i1}=0$ for $i=2,3,4$. This means that the characteristic polynomial of $b$ would have a linear factor, which contradicts the hypothesis. The case of $U_{3,1}$ is treated in exactly the same way, except that the matrix $b$ will have $b_{4j}=0$ for $j=1,2,3$. The case of infinitesimal induction is treated using the same kind of argument. Namely, if $\pi$ were a component of $\mathrm{Ind}_{U_{(2,1^{3})\hookrightarrow 2^{4}}}^{G}\mathbf{1}$, then $U_{(2,1^{3})\hookrightarrow 2^{4}}\subset N_{1}$ and $\langle\pi|_{U_{(2,1^{3})\hookrightarrow 2^{4}}},\mathbf{1}\rangle\neq 0$, which implies that $\pi|_{N_{1}}$ contains a character $\psi_{b}$, where $b=(b_{ij})$ is a matrix such that $b_{1j}=0$ for $j=1,\dots,4$. A $1$-dimensional twist of $\pi$ would then contain a character $\psi_{aI+b}$, where $a$ is a scalar and $I$ is the identity matrix. The matrix $aI+b$ has a linear factor in its characteristic polynomial, which contradicts the hypothesis. ∎ We now consider in order representations whose orbits contain $\beta_{1}$ or $\beta_{2}$, respectively. In the following we will write $\bar{P}_{2,2}$ and $\bar{U}_{2,2}$ for the images mod $\mathfrak{p}$ of the groups $P_{2,2}$ and $U_{2,2}$, respectively. ### 8.1. The regular cuspidal representations Assume that $\pi$ is an irreducible representation of $G$ whose orbit contains $\beta_{1}$. Since $\beta_{1}$ is a regular element, the representation $\pi$ can be constructed explicitly as an induced representation (cf. [Hil95b]). In particular, it is shown in [Hil95b] that there exists a $1$-dimensional representation $\rho$ of $G(\psi_{\beta_{1}})$ (uniquely determined by $\pi$) such that $\rho|_{N_{1}}=\psi_{\beta_{1}}$, and such that $\pi=\mathrm{Ind}_{G(\psi_{\beta_{1}})}^{G}\rho.$ ###### Proposition 8.1. The representation $\pi$ is cuspidal if and only if $\rho$ does not contain the trivial representation of $G(\psi_{\beta_{1}})\cap U_{2,2}$. ###### Proof. Lemma 8.1 shows that $\pi$ is cuspidal if and only if it is not a component of $\mathrm{Ind}_{U_{2,2}}^{G}\mathbf{1}$. By Mackey’s intertwining number theorem (cf. [CR62], 44.5), we have $\langle\pi,\mathrm{Ind}_{U_{2,2}}^{G}\mathbf{1}\rangle=\langle\mathrm{Ind}_{G(\psi_{\beta_{1}})}^{G}\rho,\mathrm{Ind}_{U_{2,2}}^{G}\mathbf{1}\rangle=\sum_{x\in G(\psi_{\beta_{1}})\backslash G/U_{2,2}}\langle\rho|_{G(\psi_{\beta_{1}})\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{1}\rangle,$ so this number is zero if and only if $\langle\rho|_{G(\psi_{\beta_{1}})\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{1}\rangle=0$ for each $x\in G$. Assume that $\pi$ is cuspidal. Then in particular, taking $x=1$, we have $\langle\rho|_{G(\psi_{\beta_{1}})\cap U_{2,2}},\mathbf{1}\rangle=0$. Conversely, assume that $\pi$ is not cuspidal. Then $\langle\rho|_{G(\psi_{\beta_{1}})\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{1}\rangle\neq 0$, for some $x\in G$, and in particular, $\langle\rho|_{N_{1}\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{\mathbf{1}}\rangle=\langle\psi_{\beta_{1}}|_{N_{1}\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{1}\rangle\neq 0$. Write $\bar{x}$ for $x$ modulo $\mathfrak{p}$. Now $\psi_{\beta_{1}}|_{N_{1}\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}}=\psi_{\beta_{1}}|_{\hskip 1.0pt{\vphantom{(N_{1}\cap U_{2,2})}}^{x}(N_{1}\cap U_{2,2})}$, and $\psi_{\beta_{1}}(\hskip 1.0pt{\vphantom{g}}^{x}g)=\psi_{\bar{x}^{-1}\beta_{1}\bar{x}}(g)$, for any $g\in N_{1}\cap U_{2,2}$. Let $\bar{x}^{-1}\beta_{1}\bar{x}$ be represented by the matrix $\begin{pmatrix}A_{11}&A_{12}\\\ A_{21}&A_{22}\end{pmatrix},$ where each $A_{ij}$ is a $2\times 2$-block. Then from the definition of $\psi_{\bar{x}^{-1}\beta_{1}\bar{x}}$ and the condition $\psi_{\bar{x}^{-1}\beta_{1}\bar{x}}(g)=1$, for all $g\in N_{1}\cap U_{2,2}$, it follows that $A_{21}=0$; thus $\bar{x}^{-1}\beta_{1}\bar{x}\in\bar{P}_{2,2}.$ Since $\bar{x}^{-1}\beta_{1}\bar{x}$ is a block upper-triangular matrix with the same characteristic polynomial as $\beta_{1}$, we must have $A_{11}=B_{1}\eta B_{1}^{-1}$, $A_{22}=B_{2}\eta B_{2}^{-1}$, for some $B_{1},B_{2}\in\mbox{GL}_{2}(\mathbf{F}_{q})$. Then there exists $p\in\bar{P}_{2,2}$ such that $(\bar{x}p)^{-1}\beta_{1}(\bar{x}p)=\begin{pmatrix}\eta&B\\\ 0&\eta\end{pmatrix},$ for some $B\in M_{2}(\mathbf{F}_{q})$ (in fact, we can take $p=\left(\begin{smallmatrix}B_{1}^{-1}&0\\\ 0&B_{2}^{-1}\end{smallmatrix}\right)$). The Levi decomposition $\bar{P}_{2,2}=\left(\begin{smallmatrix}*&0\\\ 0&*\end{smallmatrix}\right)\left(\begin{smallmatrix}1&*\\\ 0&1\end{smallmatrix}\right)$ (written in block matrix form) applied to $\beta_{1}$ and $(\bar{x}p)^{-1}\beta_{1}(\bar{x}p)$ implies that the semisimple parts $(\bar{x}p)^{-1}\left(\begin{smallmatrix}\eta&0\\\ 0&\eta\end{smallmatrix}\right)(\bar{x}p)$ and $\left(\begin{smallmatrix}\eta&0\\\ 0&\eta\end{smallmatrix}\right)$ are equal, that is, $\bar{x}p\in C_{G_{1^{4}}}(\left(\begin{smallmatrix}\eta&0\\\ 0&\eta\end{smallmatrix}\right))=G(\beta_{2})\cong\mbox{GL}_{2}(\mathbf{F}_{q^{2}}).$ Now, in $G(\beta_{2})$, the equation $(\bar{x}p)^{-1}\beta_{1}(\bar{x}p)=\left(\begin{smallmatrix}\eta&B\\\ 0&\eta\end{smallmatrix}\right)$ implies that $\bar{x}p\in\left(\begin{smallmatrix}*&*\\\ 0&*\end{smallmatrix}\right)\cap G(\beta_{2})\subset\bar{P}_{2,2}$, so $\bar{x}\in\bar{P}_{2,2}$, and hence $x\in N_{1}P_{2,2}$. The facts that $U_{2,2}$ is normal in $P_{2,2}$, and that $\langle\rho|_{G(\psi_{\beta_{1}})\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{1}\rangle$ only depends on the right coset of $x$ modulo $N_{1}$ then imply that $0\neq\langle\rho|_{G(\psi_{\beta_{1}})\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{1}\rangle=\langle\rho|_{G(\psi_{\beta_{1}})\cap U_{2,2}},\mathbf{1}\rangle.$ ∎ The preceding proposition shows that we can construct all the cuspidal representations of $G$ with orbit containing $\beta_{1}$ by constructing the corresponding $\rho$ on $G(\psi_{\beta_{1}})$. Since $\psi_{\beta_{1}}$ is trivial on $N_{1}\cap U_{2,2}$, we can extend $\psi_{\beta_{1}}$ to a representation of $(G(\psi_{\beta_{1}})\cap U_{2,2})N_{1}$, trivial on $G(\psi_{\beta_{1}})\cap U_{2,2}$. Then $\psi_{\beta_{1}}$ can be extended to a representation $\tilde{\psi}_{\beta_{1}}$ on the whole of $G(\psi_{\beta_{1}})$, such that $\tilde{\psi}_{\beta_{1}}$ is trivial on $G(\psi_{\beta_{1}})\cap U_{2,2}$ (this incidentally shows that there exist irreducible non-cuspidal representations of $G$ whose orbit contains $\beta_{1}$). Now let $\theta$ be a representation of $G(\psi_{\beta_{1}})$ obtained by pulling back a representation of $G(\psi_{\beta_{1}})/N_{1}$ that is non-trivial on $(G(\psi_{\beta_{1}})\cap U_{2,2})N_{1}/N_{1}$. Then $\rho:=\theta\otimes\tilde{\psi}_{\beta_{1}}$ is a representation of $G(\psi_{\beta_{1}})$ which is a lift of $\psi_{\beta_{1}}$, and which is non- trivial on $G(\psi_{\beta_{1}})\cap U_{2,2}$. By a standard fact in representation theory, all the lifts of $\psi_{\beta_{1}}$ to $G(\psi_{\beta_{1}})$ are of the form $\theta\otimes\tilde{\psi}_{\beta_{1}}$ for some $\theta$ trivial on $N_{1}$. Thus all the representations of $G(\psi_{\beta_{1}})$ which are lifts of $\psi_{\beta_{1}}$ and which are non- trivial on $G(\psi_{\beta_{1}})\cap U_{2,2}$, are of the form above, namely $\theta\otimes\tilde{\psi}_{\beta_{1}}$ where $\theta$ is trivial on $N_{1}$ but non-trivial on $G(\psi_{\beta_{1}})\cap U_{2,2}$. We note that in the regular case, distinct representations $\theta$ give rise to distinct lifts $\theta\otimes\tilde{\psi}_{\beta_{1}}$. This can be seen by a counting argument, in the following way. Because $\beta_{1}$ lies in a regular orbit, we can write $G(\psi_{\beta_{1}})=C_{G}(\hat{\beta}_{1})N_{1}$, for some element $\hat{\beta}_{1}\in M_{2}(\mathfrak{o}_{2})$ with image $\beta_{1}$ mod $\mathfrak{p}$. Then because $C_{G}(\hat{\beta}_{1})$ is abelian, there are exactly $(C_{G}(\hat{\beta}_{1}):C_{G}(\hat{\beta}_{1})\cap N_{1})=|G(\psi_{\beta_{1}})/N_{1}|$ characters $\chi$ of $C_{G}(\hat{\beta}_{1})$ which agree with $\psi_{\beta_{1}}$ on $C_{G}(\hat{\beta}_{1})\cap N_{1}$, and each of them gives rise to a representation $\chi\psi_{\beta_{1}}$ of $G(\psi_{\beta_{1}})$ defined by $\chi\psi_{\beta_{1}}(cn)=\chi(c)\psi_{\beta_{1}}(n)$, for $c\in C_{G}(\hat{\beta}_{1})$, $n\in N_{1}$. Clearly every lift of $\psi_{\beta_{1}}$ to $G(\psi_{\beta_{1}})$ must be equal to some such $\chi$ on $C_{G}(\hat{\beta}_{1})$, and distinct $\chi$ give rise to distinct representations $\chi\psi_{\beta_{1}}$. Since the number of lifts of $\psi_{\beta_{1}}$ to $G(\psi_{\beta_{1}})$ is thus equal to the number of representations of $G(\psi_{\beta_{1}})/N_{1}$, we see that distinct $\theta$ give rise to distinct representations $\theta\otimes\tilde{\psi}_{\beta_{1}}$. Now by a standard result in Clifford theory, distinct irreducible representations of $G(\psi_{\beta_{1}})$ containing $\psi_{\beta_{1}}$ (when restricted to $N_{1}$) induce to distinct irreducible representations of $G$. Thus, distinct representations $\theta$ give rise to distinct representations $\mathrm{Ind}_{G(\psi_{\beta_{1}})}^{G}\rho$, although the correspondence $\theta\mapsto\mathrm{Ind}_{G(\psi_{\beta_{1}})}^{G}\rho$ is by no means canonical, due to the choice of $\tilde{\psi}_{\beta_{1}}$. Similarly, if we are considering the lifts $\chi\psi_{\beta_{1}}$, then the construction depends on the choice of $\hat{\beta}_{1}$. The above parameterizations of representations of $G(\psi_{\beta_{1}})$ containing $\psi_{\beta_{1}}$, both involve non-canonical choices, although the set of representations obtained is certainly uniquely determined. Nevertheless, Proposition 8.1 shows that there is a canonical 1-1 correspondence (given simply by induction) between on the one hand irreducible representations of $G(\psi_{\beta_{1}})$ which contain $\psi_{\beta_{1}}$ and which are non-trivial on $G(\psi_{\beta_{1}})\cap U_{2,2}$, and on the other hand cuspidal representations of $G$ with $\beta_{1}$ in their respective orbits. We shall now extend this result to cuspidal representations which have $\beta_{2}$ in their respective orbits, and thus cover all cuspidal representations of $G$. ### 8.2. The irregular cuspidal representations Assume now that $\pi$ is an irreducible representation of $G$ whose orbit contains $\beta_{2}$. Although $\beta_{2}$ is not regular, it is strongly semisimple in the sense of [Hil95a], Definition 3.1, and thus $\pi$ can be constructed explicitly in a way similar to the regular case. More precisely, Proposition 3.3 in [Hil95a] implies that there exists an irreducible representation $\tilde{\psi}_{\beta_{2}}$ of $G(\psi_{\beta_{2}})$, such that $\tilde{\psi}_{\beta_{2}}|_{N_{1}}=\psi_{\beta_{2}}$, and any extension of $\psi_{\beta_{2}}$ to $G(\psi_{\beta_{2}})$ is of the form $\rho:=\theta\otimes\tilde{\psi}_{\beta_{2}}$, for some irreducible representation $\theta$ pulled back from a representation of $G(\psi_{\beta_{2}})/N_{1}$. Then $\pi=\mathrm{Ind}_{G(\psi_{\beta_{2}})}^{G}\rho$ is an irreducible representation, any representation of $G$ with $\beta_{2}$ in its orbit is of this form, and as in the regular case, $\rho$ is uniquely determined by $\pi$. We then have a result completely analogous to the previous proposition: ###### Proposition 8.2. The representation $\pi$ is cuspidal if and only if $\rho$ does not contain the trivial representation of $G(\psi_{\beta_{2}})\cap U_{2,2}$. ###### Proof. The proof of Proposition 8.1 with $\beta_{1}$ replaced by $\beta_{2}$, goes through up to the point where (under the assumption that $\pi$ is not cuspidal) we get $\bar{x}p\in C_{G_{1^{4}}}(\left(\begin{smallmatrix}\eta&0\\\ 0&\eta\end{smallmatrix}\right))=G(\psi_{\beta_{2}})/N_{1}$. It then follows that $x\in G(\psi_{\beta_{2}})P_{2,2}$, and since $U_{2,2}$ is normal in $P_{2,2}$, and $\langle\rho|_{G(\psi_{\beta_{2}})\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{1}\rangle$ only depends on the right coset of $x$ modulo $G(\psi_{\beta_{2}})$, we get $0\neq\langle\rho|_{G(\psi_{\beta_{2}})\cap\hskip 1.0pt{\vphantom{U_{2,2}}}^{x}U_{2,2}},\mathbf{1}\rangle=\langle\rho|_{G(\psi_{\beta_{2}})\cap U_{2,2}},\mathbf{1}\rangle.$ ∎ ## References * [BK93] Colin J. Bushnell and Philip C. Kutzko. The admissible dual of ${\rm GL}(N)$ via compact open subgroups, volume 129 of Annals of Mathematics Studies. Princeton University Press, Princeton, NJ, 1993. * [BO07] Uri Bader and Uri Onn. Geometric representations of ${\rm GL}(n,R)$, cellular Hecke algebras and the embedding problem. J. Pure Appl. Algebra, 208(3):905–922, 2007. * [Car84] H. Carayol. Représentations cuspidales du groupe linéaire. Ann. Sci. École Norm. Sup. (4), 17(2):191–225, 1984. * [CR62] C. W. Curtis and I. Reiner, _Representation Theory of Finite Groups and Associative Algebras_ , Pure and Applied Mathematics, Vol. XI, Interscience Publishers, New York-London, 1962. * [Gér75] Paul Gérardin. Construction de séries discrètes $p$-adiques. Springer-Verlag, Berlin, 1975. Sur les séries discrètes non ramifiées des groupes réductifs déployés $p$-adiques, Lecture Notes in Mathematics, Vol. 462. * [Gre55] J. A. Green. The characters of the finite general linear groups. Trans. Amer. Math. Soc., 80:402–447, 1955. * [Hen93] Guy Henniart. Correspondance de Jacquet-Langlands explicite. I. Le cas modéré de degré premier. In Séminaire de Théorie des Nombres, Paris, 1990–91, volume 108 of Progr. Math., pages 85–114. Birkhäuser Boston, Boston, MA, 1993. * [Hil95a] Gregory Hill. Semisimple and cuspidal characters of ${\rm GL}_{n}({\mathcal{O}})$. Comm. Algebra, 23(1):7–25, 1995. * [Hil95b] Gregory Hill. Regular elements and regular characters of ${\rm GL}_{n}({\mathcal{O}})$. J. Algebra, 174(2): 610–635, 1995. * [How77] Roger E. Howe. Tamely ramified supercuspidal representations of ${\rm Gl}_{n}$. Pacific J. Math., 73(2):437–460, 1977. * [Kut78] Philip Kutzko. On the supercuspidal representations of ${\rm Gl}_{2}$. Amer. J. Math., 100(1):43–60, 1978. * [Kut80] Philip Kutzko. The Langlands conjecture for ${\rm Gl}_{2}$ of a local field. Ann. of Math. (2), 112(2):381–412, 1980. * [Kut85] Philip Kutzko. On the restriction of supercuspidal representations to compact, open subgroups. _Duke Math. J._ 52(3):753–764, 1985. * [Lu04] George Lusztig. Representations of reductive groups over finite rings, Represent. Theory, 8:1088-4165, 2004. * [Mac80] I. G. Macdonald. Zeta functions attached to finite general linear groups. Math. Ann., 249(1):1–15, 1980. * [Nag78] S. V. Nagornyĭ. Complex representations of the general linear group of degree three modulo a power of a prime. (Russian) Rings and linear groups. Zap. Naučn. Sem. Leningrad. Otdel. Mat. Inst. Steklov. (LOMI), 75, 143–150, 1978. * [Onn07] Uri Onn. Representations of automorphism groups of rank two finite $\mathcal{O}$-modules. math.RT/0611383, 2007, to appear in Advances in Math. * [Pas05] Vytautas Paskunas. Unicity of types for supercuspidal representations of $\mathrm{GL}_{N}$. Proc. London Math. Soc., (3) 91:623–654, 2005. * [Ser68] Jean-Pierre Serre. Corps locaux. Hermann, Paris, 1968. Deuxième édition, Publications de l’Université de Nancago, No. VIII. * [Ser77] Jean-Pierre Serre. Linear Representations of finite groups. Springer-Verlag, 1977. * [Shi68] Takuro Shintani. On certain square-integrable irreducible unitary representations of some ${\mathfrak{p}}$-adic linear groups. J. Math. Soc. Japan, 20:522–565, 1968. * [Spr75] Tonny A. Springer. The zeta function of a cuspidal representation of a finite group ${\rm GL}_{n}(k)$. In Lie groups and their representations (Proc. Summer School, Bolyai János Math. Soc., Budapest, 1971), pages 645–648. Halsted, New York, 1975.

arxiv-papers

2024-09-04T02:48:52.920381

0707.0570

Gauss–Berezin integral operators, spinors over orthosymplectic supergroups, and Lagrangian super-Grassmannians Yuri A. Neretin111Supported by the grant FWF, project P19064, Russian Federal Agency for Nuclear Energy, the grant NWO.047.017.015, and the grant JSPS- RFBR-07.01.91209 We obtain explicit formulae for the spinor representation of the orthosymplectic supergroup $\mathrm{OSp}(2p|2q)$ and the corresponding Olshanski super semigroup. We also extend this representation to the Lagrangian super-Grassmannian.222This paper is a strongly revised variant of my preprint [23], however the result is the same, also the previous variant is shorter and explores another realization of spinors. 0.1. Berezin formulae. Apparently, first elements of a strange analogy between orthogonal and symplectic spinors333There is a lot of terms for symplectic spinors, namely, the Weil representation, the Shale–Weil representation, the Segal–Shale–Weil representation, the harmonic representation, the oscillator representation, the metaplectic representation. For references, see [21] were observed by K. O. Friedrichs in the early 1950s, see [11]. F. A. Berezin at the beginning of 1960s obtained explicit formulae [3] for both the representations. We shortly recall his results. First of all, let us realize the real symplectic group $\mathrm{Sp}(2n,{\mathbb{R}})$ as the group of complex $(n+n)\times(n+n)$ matrices $g=\begin{pmatrix}\Phi&\Psi\\\ \overline{\Psi}&\overline{\Phi}\end{pmatrix}$ satisfying the condition $g\begin{pmatrix}0&1\\\ -1&0\end{pmatrix}g^{t}=\begin{pmatrix}0&1\\\ -1&0\end{pmatrix}.$ In a similar way, we realize the real orthogonal group $\mathrm{O}(2n)$ as the group of matrices $g=\begin{pmatrix}\Phi&\Psi\\\ -\overline{\Psi}&\overline{\Phi}\end{pmatrix}$ satisfying $g\begin{pmatrix}0&1\\\ 1&0\end{pmatrix}g^{t}=\begin{pmatrix}0&1\\\ 1&0\end{pmatrix}.$ The ’spinor’ representation of $\mathrm{Sp}(2n,{\mathbb{R}})$ is realized by integral operators $W\begin{pmatrix}\Phi&\Psi\\\ \overline{\Psi}&\overline{\Phi}\end{pmatrix}f(z)=\\\ =\int_{{\mathbb{C}}^{n}}\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}z&\overline{u}\end{pmatrix}\begin{pmatrix}\overline{\Psi}\Phi^{-1}&\Phi^{t-1}\\\ \Phi^{-1}&-\Phi^{-1}\Psi\end{pmatrix}\begin{pmatrix}z^{t}\\\ \overline{u}^{t}\end{pmatrix}\Bigr{\\}}\,f(u)e^{-|u|^{2}}du\,d\overline{u}$ (0.1) in the space of holomorphic functions on ${\mathbb{C}}^{n}$. Here the symbol t denotes the transpose; $z=\begin{pmatrix}z_{1}&\dots&z_{n}\end{pmatrix}$, $u=\begin{pmatrix}u_{1}&\dots&u_{n}\end{pmatrix}$ are row vectors. On the other hand, Berezin obtained formulae for the spinor representation of the group $\mathrm{O}(2n,{\mathbb{R}})$. Exactly, he wrote certain ’integral operators’ of the form $\mathrm{spin}\begin{pmatrix}\Phi&\Psi\\\ \overline{\Psi}&\overline{\Phi}\end{pmatrix}f(\xi)=\\\ =\int\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}-\overline{\Psi}\Phi^{-1}&\Phi^{t-1}\\\ -\Phi^{-1}&-\Phi^{-1}\Psi\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}\,f(\eta)e^{-\eta\overline{\eta}^{t}}d\eta\,d\overline{\eta},$ (0.2) here $\xi=\begin{pmatrix}\xi_{1}&\dots&\xi_{n}\end{pmatrix}$, $\eta=\begin{pmatrix}\eta_{1}&\dots&\eta_{n}\end{pmatrix}$ are row-matrices; $\xi_{j}$, $\eta_{j}$, $\overline{\eta}_{j}$ are anti-commuting variables.444Actually, in both the cases Berezin considered $n=\infty$.,555However, such a formula can be written only for an open dense subset in $\mathrm{SO}(2n,{\mathbb{R}})\subset\mathrm{O}(2n,{\mathbb{R}})$; this difficulty produces some difficulties below. The integral in the right- hand side is the Berezin integral, see Section 1. In fact, Berezin in his book666On ’intellectual history’ of this book, its origins and influence, see [24]. declared that there is the analysis of Grassmann variables parallel to our human analysis. The book contains parallel exposition of the boson Fock space (usual analysis in infinite number of variables) and the fermion Fock space (Grassmann analysis in infinite number of variables), this parallel looks completely mysterious. However, the strangest elements of this analogy were the formulae (0.1) and (0.2). The mysterious analogy pushed him at the end of 60s – beginning of 70s to the invention of the ‘super analysis’, ‘supergroups’, ‘Lie superalgebras’777The supercommutator was known earlier in topology and the homological algebra; however the theory of Lie superalgebras investigates other superalgebras., which mix even (complex or real) variables and odd (Grassmann) variables. The crucial step that made a reality from shadow projects was the paper of F. A. Berezin and G. I. Kac [5], 1970, where formal supergroups were discovered. In 1971-73 Lie superalgebras appeared in the quantum field theory. 0.2. Purposes of the paper. Orthosymplectic spinors. We wish to unite the formulae (0.1)–(0.2) and to write explicitly the representation of the supergroup888Thus, this paper returns to the initial point of super-analysis. It seems strange that formula (0.3) was not written by Berezin himself. The author wrote it reading posthumous uncompleted book [4] of Berezin; it contains an interrupted attempt to do this (see also [6]). In the present paper I use some tools that were unknown to Berezin. A straightforward extension of [3] leads to cumbersome calculations. Certainly, these difficulties were surmountable. $\mathrm{OSp}(2p|2q)$, the operators of the representation have the form $T(g)f(z,\xi)=\iint\exp\left\\{\frac{1}{2}\begin{pmatrix}z&\xi&\overline{u}&\overline{\eta}\end{pmatrix}\cdot\mathfrak{R}(g)\cdot\begin{pmatrix}z^{t}\\\ \xi^{t}\\\ \overline{u}^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\right\\}\times\\\ \times f(u,\eta)\,e^{-u\overline{u}^{t}-\eta\overline{\eta}^{t}}\,du\,d\overline{u}\,d\eta\,d\overline{\eta},$ (0.3) where $g$ ranges in the supergroup $\mathrm{OSp}(2p|2q)$ and $\mathfrak{R}(g)$ is a certain matrix composed of elements a supercommutative algebra $\mathcal{A}$. 0.3. Purposes of the paper. Gauss–Berezin integral operators. In [18], [19], [20] it was shown that spinor representations are actually representations of categories. We obtain a (non perfect) super analog of these constructions. Let us explain this in more detail. First, let us consider the ’boson’ case. Consider a symmetric $(n+n)\times(n+n)$-matrix, $S=\begin{pmatrix}A&B\\\ B^{t}&C\end{pmatrix}.$ Consider a Gaussian integral operator $\mathfrak{B}\begin{bmatrix}A&B\\\ B^{t}&C\end{bmatrix}f(z)=\int_{{\mathbb{C}}^{n}}\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}z&\overline{u}\end{pmatrix}\begin{pmatrix}A&B\\\ B^{t}&C\end{pmatrix}\begin{pmatrix}z^{t}\\\ \overline{u}^{t}\end{pmatrix}\Bigr{\\}}\,f(u)e^{-|u|^{2}}du\,d\overline{u}.$ (0.4) These operators are more general than (0.1), in fact in Berezin’s formula (0.1) the symmetric matrix $\begin{pmatrix}A&B\\\ B^{t}&C\end{pmatrix}$ is unitary. It can be readily checked that bounded Gaussian operators form a semigroup, which includes the group $\mathrm{Sp}(2n,{\mathbb{R}})$; its algebraic structure is described below in Section 2. On the other hand, we can introduce Berezin operators that are fermion analogs of Gaussian operators. Namely, consider a skew-symmetric matrix $\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}$ and the integral operator $\mathfrak{B}\begin{bmatrix}A&B\\\ B^{t}&C\end{bmatrix}f(\xi)=\int_{{\mathbb{C}}^{n}}\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}\,f(u)e^{-\eta\overline{\eta}}d\eta\,d\overline{\eta}.$ (0.5) These operators are more general than (0.2), in Berezin’s formula a skew- symmetric matrix $\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}$ is contained in the pseudo-unitary group $\mathrm{U}(n,n)$. For Berezin operators in general position, a product of Berezin operators is a Gaussian operator. But sometimes this is not the case. However, it is possible to improve the definition (see our Section 1), and to obtain a semigroup of Berezin operators. In the paper we introduce ’Gauss–Berezin integral operators’, which unify boson ’Gaussian operators’ and fermion ’Berezin operators’. Our main result is a construction of a canonical bijection between set of Gauss–Berezin operators and a certain domain in Lagrangian super-Grassmannian; also we propose a geometric interpretation of products of Gauss–Berezin operators. Formula (0.3) for super-spinor representation of the supergroup $\mathrm{OSp}(2p|2q)$ is a byproduct of the geometric construction. 0.4. Purposes of the paper. Universality. The spinor representation and Weil representations, in particular, are important for two following reasons. The first reason is the Howe duality, see, for instance, [12], [13], [1]. Second, they are universal in the representation theory of infinite- dimensional groups, see [21]. Certainly, the orthosymplectic spinors furnishes us with the same tools for supergroups. But actual applications are beyond purposes of this work. 0.5. Structure of the paper. I tried to write a self-contained paper, no preliminary knowledge of the super-mathematics or representation theory is pre-assumed. But this implies a necessity of various preliminaries. We start with an exposition of orthogonal and symplectic spinors in Sections 1 and 2. These sections contain also a discussion of Gaussian integral operators and Berezin (fermion Gaussian) operators. In Sections 5 – 7 we discuss supergroups, super-Grassmannians, superlinear relations. I am trying to use minimally necessary tools, these sections are not an introduction to super-science. For generalities of super-mathematics, see [4], [17], [15], [28], [10], [27], [7]. Section 3 contains a discussion of super analogs of the Gaussian integral, this is a simple imitation of the well-known formula $\int_{{\mathbb{R}}^{n}}\exp\Bigl{\\{}-\frac{1}{2}xAx^{t}+bx^{t}\Bigr{\\}}\,dx=(2\pi)^{n/2}\det(A)^{-1/2}\exp\Bigl{\\{}-\frac{1}{2}bA^{-1}b^{t}\Bigr{\\}}.$ (0.6) Apparently, these calculations are written somewhere, but I do not know references. The Gauss–Berezin operators are introduced in Section 4. Our main construction is the canonical one-to-one correspondence between super-linear relations and Gauss–Berezin operators, which is obtained in Section 8. This immediately produces a representation of supergroups $\mathrm{OSp}(2p|2q)$. For $g$ being in an ’open dense’ subset in the supergroup, the operators of the representation has form (0.3). In the last section we discuss some details and unsolved problems. Acknowledgments. I am grateful for D. V. Alekseevski and A. S. Losev for explanations of superalgebra and superanalysis. The preliminary variant [23] of the paper was completely revised after a discussion with D. Westra. Also, he proposed numerous suggestions for an improvement of the text. ## 1 A survey of orthogonal spinors. Berezin operators and Lagrangian linear relations This Section is subdivided into 3 parts. In Part A we develop the standard formalism of Grassmann algebras $\Lambda_{n}$ and define Berezin operators $\Lambda_{n}\to\Lambda_{m}$, which are in a some sense morphisms of Grassmann algebras (but generally they are not morphisms in the category of algebras). In Part B we describe the ’geometrical’ category $\mathbf{GD}$, which is equivalent to the category of Berezin operators. Morphisms of the category $\mathbf{GD}$ are certain Lagrangian subspaces. In Part C we desribe explicitly the correspondence between Lagrangian subspaces and Berezin operators. In some cases we present proofs or explanations, for a coherent treatment see [21], Chapter 2. A. Grassmann algebras and Berezin operators. 1.1. Grassmann variables and Grassmann algebra. We denote by $\xi_{1}$, …, $\xi_{n}$ the standard Grassmann variables, $\xi_{i}\xi_{j}=-\xi_{j}\xi_{i},$ in particular, $\xi_{i}^{2}=0$. Denote by $\Lambda_{n}$ the algebra of polynomials in these variables, evidently, $\dim\Lambda_{n}=2^{n}$. The monomials $\xi_{j_{1}}\xi_{j_{2}}\dots\xi_{j_{\alpha}},\qquad\text{where $\alpha=0,1,\dots,n$ and $j_{1}<j_{2}<\dots<j_{\alpha}$}$ (1.1) form a basis of $\Lambda_{n}$. Below we call elements of Grassmann algebra by functions. 1.2. Derivatives. We define left differentiations in $\xi_{j}$ as usual. Exactly, if $f(\xi)$ does not depend on $\xi_{j}$, then $\frac{\partial}{\partial\xi_{j}}f(\xi)=0,\qquad\frac{\partial}{\partial\xi_{j}}\xi_{j}f(\xi)=f(\xi).$ Evidently, $\frac{\partial}{\partial\xi_{k}}\frac{\partial}{\partial\xi_{l}}=-\frac{\partial}{\partial\xi_{l}}\frac{\partial}{\partial\xi_{k}},\qquad\left(\frac{\partial}{\partial\xi_{k}}\right)^{2}=0.$ 1.3. Exponentials. Let $f(\xi)$ be an even function, i.e., $f(-\xi)=f(\xi)$. We define its exponential as always, $\exp\bigl{\\{}f(\xi)\bigr{\\}}:=\sum_{j=0}^{\infty}\frac{1}{j!}f(\xi)^{j}.$ Even functions $f$, $g$ commute, $fg=gf$, therefore $\exp\\{f(\xi)+g(\xi)\\}=\exp\\{f(\xi)\\}\cdot\exp\\{g(\xi)\\}.$ 1.4. Berezin integral. Let $\xi_{1}$, …, $\xi_{n}$ be Grassmann variables. The Berezin integral $\int f(\xi)\,d\xi=\int f(\xi_{1},\dots,\xi_{n})\,d\xi_{1}\dots d\xi_{n}$ is a linear functional on $\Lambda_{n}$ defined by $\int\xi_{n}\xi_{n-1}\dots\xi_{1}\,d\xi_{1}\dots d\xi_{n}=1$ The integral of all other monomials is zero. The following formula for integration by parts holds $\int f(\xi)\cdot\frac{\partial}{\partial\xi_{k}}\,g(\xi)\,d\xi=-\int\frac{\partial f(-\xi)}{\partial\xi_{k}}\cdot g(\xi)\,d\xi.$ 1.5. Integrals with respect ’to odd Gaussian measure’. Let $\xi_{1}$, …, $\xi_{q}$ be as above. Let $\overline{\xi}_{1}$, …, $\overline{\xi}_{q}$ be another collection of Grassmann variables, $\overline{\xi}_{k}\overline{\xi}_{l}=-\overline{\xi}_{l}\overline{\xi}_{k},\qquad\xi_{k}\overline{\xi}_{l}=-\overline{\xi}_{l}\xi_{k}.$ We need the following Gaussian expression $\exp\bigl{\\{}-\xi\overline{\xi}^{t}\bigr{\\}}:=\exp\\{-\xi_{1}\overline{\xi}_{1}-\xi_{2}\overline{\xi}_{2}-\dots\\}=\\\ =\exp\\{\overline{\xi}_{1}\xi_{1}\\}\exp\\{\overline{\xi}_{2}\xi_{2}\\}\dots=(1+\overline{\xi}_{1}\xi_{1})(1+\overline{\xi}_{2}\xi_{2})(1+\overline{\xi}_{3}\xi_{3})\dots{}$ Denote $d\overline{\xi}\,d\xi=d\overline{\xi}_{1}\,d\xi_{1}\,d\overline{\xi}_{2}\,d\xi_{2}\dots$ Therefore, $\int\Bigl{(}\prod_{k=1}^{m}(\overline{\xi}_{\alpha_{k}}\xi_{\alpha_{k}})\Bigr{)}\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi=1$ (1.2) and the integral is zero for all other monomials. For instance, $\displaystyle\int\xi_{1}\xi_{2}\overline{\xi}_{3}\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi=0,\qquad\int\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi=+1,$ $\displaystyle\int\xi_{1}\overline{\xi}_{1}\overline{\xi}_{33}\xi_{33}\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi=-\int(\overline{\xi}_{1}\xi_{1})\,(\overline{\xi}_{33}\xi_{33})\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi=-1.$ Evidently, $\displaystyle\int f(\overline{\xi})\,\cdot\,\frac{\partial}{\partial\xi_{k}}g(\xi)\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi=\int\overline{\xi}_{k}f(-\overline{\xi})\,\cdot\,g(\xi)\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi,$ $\displaystyle\int\frac{\partial}{\partial\overline{\xi}_{k}}f(\overline{\xi})\,\cdot\,g(\xi)\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi=\int f(-\overline{\xi})\cdot\xi_{k}g(\xi)\,e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi.$ 1.6. Integral operators. Now consider Grassmann algebras $\Lambda_{p}$ and $\Lambda_{q}$ consisting of polynomials in Grassmann variables $\xi_{1}$, …, $\xi_{p}$ and $\eta_{1}$, …, $\eta_{q}$ respectively. For a function $K(\xi,\overline{\eta})$ we define an integral operator $A_{K}:\Lambda_{q}\to\Lambda_{p}$ by $A_{K}f(\xi)=\int K(\xi,\overline{\eta})f(\eta)\,e^{-\eta\overline{\eta}^{t}}\,d\overline{\eta}\,d\eta.$ ###### Proposition 1.1 The map $K\mapsto A_{K}$ is a one-to-one correspondence of the set of all polynomials $K(\xi,\overline{\eta})$ and the set of all linear maps $\Lambda_{q}\to\Lambda_{p}$. Proposition 1.1 is completely trivial. Indeed, expand $K(\xi,\eta)=\sum a_{i_{1},\dots,i_{l}\,j_{1},\dots,j_{l}}\xi_{i_{1}}\dots\xi_{i_{k}}\overline{\eta}_{j_{1}}\dots\overline{\eta}{j_{l}}.$ Then $a_{\dots}$ are the matrix elements999up to signs of $A_{K}$ in the standard basis (1.1). ###### Proposition 1.2 If $A:\Lambda_{q}\to\Lambda_{p}$ is determined by the kernel $K(\xi,\eta)$ and $B:\Lambda_{p}\to\Lambda_{r}$ is determined by the kernel $L(\zeta,\overline{\xi})$, then the kernel of $BA$ is $M(\zeta,\overline{\eta})=\int L(\zeta,\overline{\xi})K(\xi,\overline{\eta})e^{-\xi\overline{\xi}^{t}}\,d\overline{\xi}\,d\xi.$ 1.7. Berezin operators in the narrow sense. A Berezin operator $\Lambda_{q}\to\Lambda_{p}$ in the narrow sense is an operator of the form $\mathfrak{B}\begin{bmatrix}A&B\\\ -B^{t}&C\end{bmatrix}f(\xi):=\int\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}f(\eta)\,e^{-\eta\overline{\eta}^{t}}\,d\overline{\eta}\,d\eta,$ (1.3) where $A=-A^{t}$, $C=-C^{t}$. Let us explain the notation. 1\. $\begin{pmatrix}\xi&\eta\end{pmatrix}$ denotes the row-matrix $\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}:=\begin{pmatrix}\xi_{1}&\dots&\xi_{p}&\overline{\eta}_{1}&\dots&\overline{\eta}_{q}\end{pmatrix}.$ Respectively, $\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}$ denotes the transposed column-matrix. 2\. The $(p+q)\times(p+q)$-matrix $\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}$ is skew-symmetric. The whole expression for the kernel has the form $\exp\Bigl{\\{}\frac{1}{2}\sum_{k\leqslant p,l\leqslant p}a_{kl}\xi_{k}\xi_{l}+\sum_{k\leqslant p,m\leqslant q}b_{km}\xi_{k}\overline{\eta}_{m}+\frac{1}{2}\sum_{m\leqslant q,j\leqslant q}c_{mj}\overline{\eta}_{m}\overline{\eta}_{j}\Bigr{\\}}.$ 1.8. Product formula. ###### Theorem 1.3 Let $\mathfrak{B}[S_{1}]=\mathfrak{B}\begin{bmatrix}P&Q\\\ -Q^{t}&R\end{bmatrix}:\Lambda_{p}\to\Lambda_{q},\qquad\mathfrak{B}[S_{2}]=\mathfrak{B}\begin{bmatrix}K&L\\\ -L^{t}&M\end{bmatrix}:\Lambda_{q}\to\Lambda_{r}$ be Berezin operators. Assume $\det(1-MP)\neq 0$. Then $\mathfrak{B}[S_{2}]\mathfrak{B}[S_{1}]=\mathrm{Pfaff}\begin{pmatrix}M&1\\\ -1&P\end{pmatrix}\mathfrak{B}[S_{2}\circ S_{1}],$ (1.4) where $\begin{pmatrix}K&L\\\ -L^{t}&M\end{pmatrix}\circ\begin{pmatrix}P&Q\\\ -Q^{t}&R\end{pmatrix}=\\\ =\begin{pmatrix}K+LP(1-MP)^{-1}L^{t}&L(1-PM)^{-1}Q\\\ -Q^{t}(1-MP)^{-1}L^{t}&R-Q^{t}(1-MQ)^{-1}MQ\end{pmatrix}.$ (1.5) The symbol $\mathrm{Pfaff}(\cdot)$ denotes the Pfaffian, see the next subsection. A calculation is not difficult, see Subsection 3. 1.9. Pfaffians. Let $R$ be a skew-symmetric $2n\times 2n$ matrix. Its Pfaffian $\mathrm{Pfaff}(R)$ is defined by the condition $\frac{1}{n!}\Bigl{(}\frac{1}{2}\sum_{kl}r_{kl}\xi_{k}\xi_{l}\Bigr{)}^{n}=\mathrm{Pfaff}(R)\,\xi_{1}\xi_{2}\dots\xi_{2n-1}\xi_{2n}.$ In other words, $\mathrm{Pfaff}(R)=\frac{1}{n!}\int\Bigl{(}\frac{1}{2}\sum_{kl}r_{kl}\xi_{k}\xi_{l}\Bigr{)}^{n}\,d\xi=\int\exp\Bigl{\\{}\frac{1}{2}\xi R\xi^{t}\Bigr{\\}}\,d\xi.$ Recall that $\mathrm{Pfaff}(R)^{2}=\det R.$ 1.10. The definition of Berezin operators. Theorem 1.3 suggests an extension of the definition of Berezin operators. Indeed, this theorem is perfect for operators in general position. However, if $\det(1-MP)=0$, then we get an indeterminacy of the type $0\cdot\infty$ in the product formula (1.4), (1.5). For this reason, consider the cone101010A cone is a subset invariant with respect to homotheties $z\mapsto sz$, where $s\in{\mathbb{C}}$. $\mathcal{C}$ of all the operators of the form $s\cdot\mathfrak{B}[S]$, where $s$ ranges in ${\mathbb{C}}$. Certainly, the cone $\mathcal{C}$ is not closed. Indeed, $\displaystyle\lim_{\varepsilon\to 0}\varepsilon\exp\Bigl{\\{}\frac{1}{\varepsilon}\xi_{1}\xi_{2}\Bigr{\\}}=\xi_{1}\xi_{2},$ $\displaystyle\lim_{\varepsilon\to 0}\varepsilon^{2}\exp\Bigl{\\{}\frac{1}{\varepsilon}(\xi_{1}\xi_{2}+\xi_{2}\xi_{4})\Bigr{\\}}=\xi_{1}\xi_{2}\xi_{3}\xi_{4}.$ This suggests the following definition. Berezin operators are operators $\Lambda_{q}\to\Lambda_{p}$, whose kernels have the form $s\cdot\prod_{j=1}^{m}(\xi u^{t}_{j}+\overline{\eta}v^{t}_{j})\cdot\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}},$ (1.6) where 1\. $s\in{\mathbb{C}}$; 2\. $u_{j}\in{\mathbb{C}}^{p}$, $v_{j}\in{\mathbb{C}}^{q}$ are row-matrices; 3\. $m$ ranges in the set $\\{0,1,\dots,p+q\\}$; 4\. $\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}$ is a skew-symmetric $(p+q)\times(p+q)$-matrix. 1.11. The space of Berezin operators. ###### Proposition 1.4 a) The cone of Berezin operators is closed in the space of all linear operators $\Lambda_{q}\to\Lambda_{p}$. b) Denote by $\mathop{\rm Ber}\nolimits^{[m]}$ the set of all Berezin operators with a given number $m$ of linear factors in (1.6). Then the closure of $\mathop{\rm Ber}\nolimits^{[m]}$ is $\mathop{\rm Ber}\nolimits^{[m]}\cup\mathop{\rm Ber}\nolimits^{[m+2]}\cup\mathop{\rm Ber}\nolimits^{[m+4]}\cup\dots$ Therefore the cone of all Berezin operators $\Lambda_{q}\to\Lambda_{p}$ consists of two components, namely $\bigcup\limits_{\text{$m$ is even}}\mathop{\rm Ber}\nolimits^{[m]}\qquad\text{and}\qquad\bigcup\limits_{\text{$m$ is odd}}\mathop{\rm Ber}\nolimits^{[m]}.$ They are closures of $\mathop{\rm Ber}\nolimits^{[0]}$ and $\mathop{\rm Ber}\nolimits^{[1]}$ respectively. Also, in these cases, the kernel $K(\xi,\overline{\eta})$ satisfies $K(-\xi,-\overline{\eta})=K(\xi,\overline{\eta}),\qquad\text{or}\qquad K(-\xi,-\overline{\eta})=-K(\xi,\overline{\eta})$ respectively. Remark. A kernel $K(\xi,\eta)$ of an operator $\Lambda_{q}\to\Lambda_{p}$ is canonically defined. However the expression (1.6) for a kernel of a Berezin operator is non-canonical if the number $m$ of linear factors is $\geqslant 1$. For instance, $(\xi_{1}+\xi_{33})(\xi_{1}+7\xi_{33})=6\xi_{1}\xi_{33},\qquad\xi_{1}\exp\\{\xi_{1}\xi_{2}+\xi_{3}\xi_{4}\\}=\xi_{1}\exp\\{\xi_{3}\xi_{4}\\}.$ 1.12. Examples of Berezin operators. a) The identity operator is a Berezin operator. Its kernel is $\exp\bigl{\\{}\sum\xi_{j}\overline{\eta}_{j}\\}$. b) More generally, an operator with kernel $\exp\bigl{\\{}\sum_{ij}b_{ij}\xi_{i}\overline{\eta}_{j}\bigr{\\}}$ is the natural substitution operator $\Lambda_{q}\to\Lambda_{p}$, $\eta_{i}=\sum_{j}b_{ji}\xi_{j}.$ c) The operator with kernel $\xi_{1}\exp\\{\sum\xi_{j}\overline{\eta}_{j}\\}$ is the operator $f\mapsto\xi_{1}f$. e) The operator with kernel $\overline{\eta}_{1}\exp\\{\sum\xi_{j}\overline{\eta}_{j}\\}$ is the operator $f\mapsto\frac{\partial}{\partial\xi_{1}}f$. f) The operator corresponding to the kernel $(\xi_{1}+\overline{\eta}_{1})(\xi_{2}+\overline{\eta}_{2})\dots$ is the Hodge $*$-operator. g) The operator with kernel $K(\xi,\eta)=1$ is the projection on the vector $f(\xi)=1$. h) The operator $\mathfrak{B}$ with the kernel $\exp\Bigl{\\{}\frac{1}{2}a_{ij}\xi_{i}\xi_{j}+\sum\xi_{j}\overline{\eta}_{j}\Bigr{\\}}$ is the multiplication operator $\mathfrak{B}f(\xi)=\exp\Bigl{\\{}\frac{1}{2}a_{ij}\xi_{i}\xi_{j}\Bigr{\\}}\,f(\xi).$ i) A product of Berezin operators is a Berezin operator (see below). j) Operators of the spinor representation of $\mathrm{O}(2n,{\mathbb{C}})$ are Berezin operators. 1.13. Another definition of Berezin operators. Denote by $\mathfrak{D}(\xi_{j})$ the following operators in $\Lambda_{p}$ $\mathfrak{D}[\xi_{j}]f(\xi):=\Bigl{(}\xi_{j}+\frac{\partial}{\partial\xi_{j}}\Bigr{)}\,f(\xi).$ (1.7) If $g$, $h$ do not depend on $\xi_{j}$, then $\mathfrak{D}[\xi_{j}]\bigl{(}g(\xi)+\xi_{j}h(\xi)\bigr{)}=\xi_{j}g(\xi)+h(\xi)$ In the same way, we define the operators $\mathfrak{D}[\eta_{j}]:\Lambda_{q}\to\Lambda_{q}$. Obviously $\mathfrak{D}[\xi_{i}]^{2}=1,\qquad\mathfrak{D}[\xi_{i}]\mathfrak{D}[\xi_{j}]=-\mathfrak{D}[\xi_{j}]\mathfrak{D}[\xi_{i}],\quad\text{for $i\neq j$}.$ (1.8) A Berezin operator $\Lambda_{q}\to\Lambda_{p}$ is an operator that can be represented in the form $\mathfrak{D}[\xi_{k_{1}}]\dots\mathfrak{D}[\xi_{k_{\alpha}}]\cdot\mathfrak{B}\cdot\mathfrak{D}[\eta_{m_{1}}]\dots\mathfrak{D}[\eta_{m_{\beta}}],$ (1.9) where — $\mathfrak{B}$ is a Berezin operator in the narrow sense, — $\alpha$ ranges in the set $\\{0,1,\dots,p\\}$ and $\beta$ ranges in $\\{0,1,\dots,q\\}$ By (1.8) we can assume $k_{1}<\dots<k_{\alpha}$, $m_{1}<\dots<m_{\beta}$. ###### Proposition 1.5 The two definitions of Berezin operators are equivalent. Remark. Usually a Berezin operator admits many representations in the the form (1.9). In fact the space of all Berezin operators is a smooth cone, and formula (1.9) determines $2^{p+q}$ coordinate systems on this cone. 1.14. Category of Berezin operators. Groups of automorphisms. ###### Theorem 1.6 Let $\mathfrak{B}_{1}:\Lambda_{q}\to\Lambda_{p},\qquad\mathfrak{B}_{2}:\Lambda_{p}\to\Lambda_{r}$ be Berezin operators. Then $\mathfrak{B}_{2}\mathfrak{B}_{1}:\Lambda_{q}\to\Lambda_{r}$ is a Berezin operator. By Theorem 1.6 we get a category, whose objects are Grassmann algebras $\Lambda_{0}$, $\Lambda_{1}$, $\Lambda_{2}$, …and whose morphisms are Berezin operators. Denote by $G_{n}$ the group of all invertible Berezin operators $\Lambda_{n}\to\Lambda_{n}$. By the definition, it contains the group ${\mathbb{C}}^{*}$ of all scalar operators. ###### Theorem 1.7 $G_{n}/{\mathbb{C}}^{*}\simeq\mathrm{O}(2n,{\mathbb{C}})$. Here $\mathrm{O}(2n,{\mathbb{C}})$ denotes the usual group of orthogonal transformations in ${\mathbb{C}}^{2n}$. Moreover, this isomorphism is nothing but the spinor representation of $\mathrm{O}(2n,{\mathbb{C}})$. Our next purpose is to describe explicitly the category of Berezin operators. In fact, we intend to clarify the strange matrix multiplication (1.5). B. Linear relations and category $\mathbf{GD}$. 1.15. Linear relations. Let $V$, $W$ be linear spaces over ${\mathbb{C}}$. A linear relation $P:V\rightrightarrows W$ is a linear subspace $P\subset V\oplus W$. Remark. Let $A:V\to W$ be a linear operator. Its graph $\mathop{\rm graph}\nolimits(A)\subset V\oplus W$ consists of all vectors $v\oplus Av$. By the definition, $\mathop{\rm graph}\nolimits(A)$ is a linear relation, $\dim\mathop{\rm graph}\nolimits(A)=\dim V.$ 1.16. Product of linear relations. Let $P:V\rightrightarrows W$, $Q:W\rightrightarrows Y$ be linear relations. Informally, a product $QP$ of linear relations is a product of many-valued maps. If $P$ takes a vector $v$ to a vector $w$ and $Q$ takes the vector $w$ to a vector $y$, then $QP$ takes $v$ to $y$. Now we present a formal definition. The product $QP$ is a linear relation $QP:V\rightrightarrows W$ consisting of all $v\oplus y\in V\oplus Y$ such that there exists $w\in W$ satisfying $v\oplus w\in P$, $w\oplus y\in Q$. In fact, the multiplication of linear relations extends the usual matrix multiplication. 1.17. Imitation of some standard definitions of matrix theory. $1^{\circ}$. The kernel $\ker P$ consists of all $v\in V$ such that $v\oplus 0\in P$. In other words, $\ker P=P\cap(V\oplus 0).$ $2^{\circ}$. The image $\mathop{\rm im}\nolimits P\subset W$ is the projection of $P$ onto $0\oplus W$. $3^{\circ}$. The domain $\mathop{\rm dom}\nolimits P\subset V$ of $P$ is the projection of $P$ onto $V\oplus 0$. $4^{\circ}$. The indefinity $\mathop{\rm indef}\nolimits P\subset W$ of $P$ is $P\cap(0\oplus W)$. The definitions of a kernel and an image extend the corresponding definitions for linear operators. The definition of a domain extends the usual definition of the domain of an unbounded operator in an infinite-dimensional space. For an actual operator, $\mathop{\rm indef}\nolimits=0$. 1.18. Lagrangian Grassmannian and orthogonal groups. First, let us recall some definitions. Let $V$ be a linear space equipped with a non-degenerate symmetric (or skew-symmetric) bilinear form $M$. A subspace $H$ is isotropic with respect to the bilinear form $M$, if $M(h,h^{\prime})=0$ for all $h$, $h^{\prime}\in H$. The dimension of an isotropic subspace satisfies $\dim H\leqslant\frac{1}{2}\dim V.$ A Lagrangian subspace111111Mostly, the term ’Lagrangian subspace’ is used for spaces equipped with a skew-symmetric bilinear forms; however, my usage is a common slang. Also, in the super case there is no distinction between symmetric and skew-symmetric forms. is an isotropic subspace whose dimension is precisely $\frac{1}{2}\dim V$. By $\mathop{\rm Lagr}\nolimits(V)$ we denote the Lagrangian Grassmannian, i.e., the space of all Lagrangian subspaces in $V$. Consider a space $\mathcal{V}_{2n}={\mathbb{C}}^{2n}$ equipped with the symmetric bilinear from $L$ determined by the matrix $\begin{pmatrix}0&1\\\ 1&0\end{pmatrix}$. Denote by $\mathrm{O}(2n,{\mathbb{C}})$ the group of all linear transformations $g$ of ${\mathbb{C}}^{2n}$ preserving $L$, i.e., $g$ must satisfy the condition $g\begin{pmatrix}0&1\\\ 1&0\end{pmatrix}g^{t}=\begin{pmatrix}0&1\\\ 1&0\end{pmatrix}.$ Equip the space $\mathcal{V}_{2n}\oplus\mathcal{V}_{2m}$ with the symmetric bilinear form $L^{\ominus}$ given by $\\!\\!L^{\ominus}(v\oplus w,v^{\prime}\oplus w^{\prime})=L(v,v^{\prime})-L(w,w^{\prime})\,\,,$ (1.10) where $v$, $v^{\prime}\in\mathcal{V}_{2n}$ and $w$, $w^{\prime}\in\mathcal{V}_{2m}$. ###### Observation 1.8 Let $g$ be an operator in ${\mathbb{C}}^{2n}$. Then $g\in\mathrm{O}(2n,{\mathbb{C}})$ iff its graph is an $L^{\ominus}$-Lagrangian subspace in $\mathcal{V}_{2n}\oplus\mathcal{V}_{2n}$. This is obvious. 1.19. Imitation of orthogonal groups. Category $\mathbf{GD}$. Now we define the category $\mathbf{GD}$. The objects are the spaces $\mathcal{V}_{2n}$, where $n=0,1,2,\dots$. There are two types of morphisms $\mathcal{V}_{2n}\to\mathcal{V}_{2m}$: a) $L^{\ominus}$-Lagrangian subspaces $P\subset\mathcal{V}_{2n}\oplus\mathcal{V}_{2m}$; we regard them as linear relations. b) a distinguished morphism121212It is not a linear relation. $\mathop{\mathsf{null}}\nolimits_{2n,2m}$. Now define a product of morphisms. — Product of $\mathop{\mathsf{null}}\nolimits$ and any morphism is $\mathop{\mathsf{null}}\nolimits$. — Let $P:\mathcal{V}_{2n}\rightrightarrows\mathcal{V}_{2m}$, $Q:\mathcal{V}_{2m}\rightrightarrows\mathcal{V}_{2k}$ be Lagrangian linear relations. Assume that $\ker Q\cap\mathop{\rm indef}\nolimits P=0\qquad\text{or, equivalently,\,\, $\mathop{\rm im}\nolimits P+\mathop{\rm dom}\nolimits Q=\mathcal{V}_{2m}$}$ (1.11) Then $QP$ is the product of linear relations. — If the condition (1.11) is not satisfied, then $QP=\mathop{\mathsf{null}}\nolimits$. ###### Theorem 1.9 The definition is self-consistent, i.e., a product of morphisms is a morphism and the multiplication is associative. The appearance of $\mathop{\mathsf{null}}\nolimits$ seems strange; however its necessity will be transparent immediately ($\mathop{\mathsf{null}}\nolimits$ correspond to zero operators in the next theorem). Also, $\mathop{\mathsf{null}}\nolimits$ will be the source of some our difficulties below. ###### Theorem 1.10 The category of Berezin operators defined up to scalar factors and the category $\mathbf{GD}$ are equivalent. In fact, there is a map that takes each Lagrangian linear relation $P:\mathcal{V}_{2n}\rightrightarrows\mathcal{V}_{2m}$ to a nonzero Berezin operator $\mathop{\rm spin}\nolimits(P):\Lambda_{n}\to\Lambda_{m}$ such that for each $R:\mathcal{V}_{2q}\rightrightarrows\mathcal{V}_{2p}$, $Q:\mathcal{V}_{2p}\rightrightarrows\mathcal{V}_{2r}$ $\mathop{\rm spin}\nolimits(Q)\mathop{\rm spin}\nolimits(R)=\lambda(Q,R)\mathop{\rm spin}\nolimits(QR)$ where $\lambda(Q,R)$ is a constant. Moreover $\lambda(Q,R)=0$ iff $QR=\mathop{\mathsf{null}}\nolimits$. We describe the correspondence between the category of Berezin operators and the category $\mathbf{GD}$ explicitly in Theorems 1.16, 1.17. First, we need some auxiliary facts concerning Lagrangian Grassmannians (for detailed introduction to Lagrangian Grassmannians see [2], [22]). C. Explicit correspondence. 1.20. Coordinates on the Lagrangian Grassmannian. We write $\mathcal{V}_{2n}={\mathbb{C}}^{2n}$ as $\mathcal{V}_{2n}=\mathcal{V}_{n}^{+}\oplus\mathcal{V}_{n}^{-}={\mathbb{C}}^{n}\oplus{\mathbb{C}}^{n}.$ ###### Lemma 1.11 Let $H\subset\mathcal{V}_{2n}$ be an $n$-dimensional subspace such that $H\cap\mathcal{V}_{n}^{+}=0$ (under this condition $H$ is a graph of an operator $T_{H}:\mathcal{V}_{n}^{-}\to\mathcal{V}_{n}^{+}$). The following conditions are equivalent — the matrix $T_{H}$ is skew-symmetric; — the subspace $H$ is Lagrangian Proof. We write the bilinear form $L$ as $L(v,w)=L(v^{+}\oplus v^{-},w^{+}\oplus w^{-})=v^{+}(w^{-})^{t}+w^{+}(v^{-})^{t}.$ By the definition, $v\in H$ iff $v^{-}=v^{+}T$. For $v$, $w\in P$ we evaluate $L(v^{+}\oplus v^{+}T,w^{+}\oplus w^{+}T)=v^{+}(w^{+}H)^{t}+v^{+}H(w^{+})^{t}=v^{+}(H+H^{t})w^{+}.$ Now the statement becomes obvious. $\square$ Lemma 1.11 defines a coordinate system on $\mathop{\rm Lagr}\nolimits(\mathcal{V}_{2n})$. Certainly, this coordinate system does not cover the whole space $\mathop{\rm Lagr}\nolimits(\mathcal{V}_{2n})$. 1.21. Atlas on the Lagrangian Grassmannian. Denote by $e_{j}^{+}$, $e_{l}^{-}$ the standard basis in ${\mathbb{C}}^{2n}$. Let $J$ be a subset in $\\{1,2,\dots,n\\}$. Denote by $\overline{J}$ its complement. We define the subspaces $\displaystyle\mathcal{V}_{n}^{+}[J]:=\bigl{(}\oplus_{j\in J}{\mathbb{C}}e_{j}^{+}\bigr{)}\oplus\bigl{(}\oplus_{j\notin J}{\mathbb{C}}e_{j}^{-}\bigr{)},$ $\displaystyle\mathcal{V}_{n}^{-}[J]:=\bigl{(}\oplus_{j\notin J}{\mathbb{C}}e_{j}^{+}\bigr{)}\oplus\bigl{(}\oplus_{j\in J}{\mathbb{C}}e_{j}^{-}\bigr{)},$ then $\mathcal{V}_{2n}=\mathcal{V}_{n}^{+}[J]\oplus\mathcal{V}_{n}^{-}[J]$. Denote by $\mathcal{M}[J]$ the set of all Lagrangian subspaces $H$ in $\mathcal{V}_{2n}$ such that $H\cap\mathcal{V}_{n}^{-}[J]=0$. Such subspaces are precisely graphs of symmetric operators $\mathcal{V}_{n}^{+}[J]\to\mathcal{V}_{n}^{-}[J]$. ###### Proposition 1.12 The $2^{n}$ maps $\mathcal{M}[J]$ cover the whole Lagrangian Grassmannian $\mathop{\rm Lagr}\nolimits(\mathcal{V}_{2n})$. 1.22. Atlas on the Lagrangian Grassmannian. Elementary reflections. We can describe the same maps in a slightly different way. For $i=1,2,\dots,n$ define the elementary reflection $\sigma_{i}:\mathcal{V}_{2n}\to\mathcal{V}_{2n}$ by $\sigma_{i}e_{i}^{+}=e_{i}^{-},\qquad\sigma_{i}e_{i}^{-}=e_{i}^{+},\qquad\sigma_{i}e_{j}^{\pm}=e_{j}^{\pm}\qquad\text{for $i\neq j$}.$ (1.12) ###### Observation 1.13 $\mathcal{M}[J]=\Bigl{(}\prod_{i\in J}\sigma_{i}\Bigr{)}\mathcal{M}.$ (1.13) 1.23. Components of Lagrangian Grassmannian. ###### Observation 1.14 The Lagrangian Grassmannian in the space $\mathcal{V}_{2n}$ consists of two connected components.131313Recall that $\mathcal{V}_{2n}$ is equipped with a symmetric bilinear form. The usual Lagrangian Grassmannian discussed in the next section is connected. The orthosymplectic Lagrangian Grassmannian (see Section 6) consists of two components. We propose two proofs to convince the reader. 1\. The group $\mathrm{O}(n,{\mathbb{C}})$ is dense in $\mathop{\rm Lagr}\nolimits(\mathcal{V}_{2n})$. This group has two components. 2\. It can be readily checked that $\mathop{\rm Lagr}\nolimits(\mathcal{V}_{2n})$ is a homogeneous space $\mathop{\rm Lagr}\nolimits(\mathcal{V}_{2n})\simeq\mathrm{O}(2n,{\mathbb{C}})/\mathrm{GL}(n,{\mathbb{C}}).$ The group $\mathrm{O}(2n,{\mathbb{C}})$ consists of two components and the group $\mathrm{GL}(n,{\mathbb{C}})$ is connected. $\square$ Remark. Two components of the Lagrangian Grassmannian correspond to two components of the space of Berezin operators. $\square$ 1.24. Coordinates on the set of morphisms of $\mathbf{GD}$. We can apply the reasoning of Section 1 to the $\mathop{\rm Lagr}\nolimits(\mathcal{V}_{2n}\oplus\mathcal{V}_{2m})$. Due to the minus in formula (1.10) we must take care of the signs in Lemma 1.11. ###### Lemma 1.15 Decompose $\mathcal{V}_{2n}\oplus\mathcal{V}_{2m}$ as141414We emphasis that the initial space $\mathcal{V}_{2n}$ and the target space $\mathcal{V}_{2m}$ are mixed in the next row. $\mathcal{V}_{2n}\oplus\mathcal{V}_{2m}=(\mathcal{V}_{n}^{-}\oplus\mathcal{V}_{m}^{+})\oplus(\mathcal{V}_{n}^{+}\oplus\mathcal{V}_{m}^{-}).$ Let $P$ be an $(m+n)$-dimensional subspace such that $P\cap(\mathcal{V}_{n}^{+}\oplus\mathcal{V}_{m}^{-})=0$, i.e., $P$ is a graph of an operator $\mathcal{V}_{n}^{-}\oplus\mathcal{V}_{m}^{+}\to\mathcal{V}_{n}^{+}\oplus\mathcal{V}_{m}^{-}.$ Then the following conditions are equivalent — $P\in\mathop{\rm Lagr}\nolimits(\mathcal{V}_{2n}\oplus\mathcal{V}_{2m})$; — $P$ is a graph of an operator having the form $\begin{pmatrix}A&B\\\ B^{t}&C\end{pmatrix},\qquad\text{where $A=-A^{t}$, $C=-C^{t}$}.$ (1.14) 1.25. Creation-annihilation operators. Let $\mathcal{V}_{2n}$ be as above. Decompose $\mathcal{V}_{2n}=\mathcal{V}_{n}^{+}\oplus\mathcal{V}_{n}^{-},\qquad\text{where $\mathcal{V}_{n}^{+}:={\mathbb{C}}^{n}\oplus 0$, \, $\mathcal{V}_{n}^{-}:=0\oplus{\mathbb{C}}^{n}$.}$ Let us write elements of $\mathcal{V}_{2m}$ as $v:=\begin{pmatrix}v_{1}^{+}&\dots&v_{n}^{+}&v_{1}^{-}&\dots&v_{n}^{-}\end{pmatrix}.$ For each $v\in\mathcal{V}_{2n}$ we define a creation-annihilation operator $\widehat{a}(v)$ in $\Lambda_{n}$ by $\widehat{a}(v)f(\xi):=\Bigl{(}\sum_{j}v_{j}^{+}\xi_{j}+\sum_{j}v_{j}^{-}\frac{\partial}{\partial\xi_{j}}\Bigr{)}f(\xi).$ Evidently, $\widehat{a}(v)\widehat{a}(w)+\widehat{a}(w)\widehat{a}(v)=L(v,w)\cdot 1.$ 1.26. Construction of correspondence. Let $\mathfrak{B}:\Lambda_{q}\to\Lambda_{p}$ be a nonzero Berezin operator. Consider the subspace $P=P[\mathfrak{B}]\subset\mathcal{V}_{2q}\oplus\mathcal{V}_{2p}$ consisting of $v\oplus w$ such that $\widehat{a}(w)\mathfrak{B}=\mathfrak{B}\widehat{a}(v).$ (1.15) ###### Theorem 1.16 a) $P[\mathfrak{B}]$ is a morphism of the category $\mathbf{GD}$, i.e., a Lagrangian subspace. b) The map $\mathfrak{B}\mapsto P$ is a bijection $\left\\{\begin{matrix}\text{Set of nonzero Berezin operators $\Lambda_{q}\to\Lambda_{p}$}\\\ \text{defined up to a scalar}\end{matrix}\right\\}\longleftrightarrow\\\ \longleftrightarrow\left\\{\begin{matrix}\text{The set of non-'$\mathop{\mathsf{null}}\nolimits$' morphisms $\mathcal{V}_{2q}\to\mathcal{V}_{2p}$}\\\ \text{of the category $\mathbf{GD}$}\end{matrix}\right\\}$ c) Let $\mathfrak{B}_{1}:\Lambda_{q}\to\Lambda_{p}$, $\mathfrak{B}_{2}:\Lambda_{p}\to\Lambda_{r}$ be Berezin operators. Then $\mathfrak{B}_{2}\mathfrak{B}_{1}=0\quad\text{iff}\quad P[\mathfrak{B}_{2}]P[\mathfrak{B}_{1}]=\mathop{\mathsf{null}}\nolimits.$ (1.16) d) Otherwise, $P[\mathfrak{B}_{2}\mathfrak{B}_{1}]=P[\mathfrak{B}_{2}]P[\mathfrak{B}_{1}].$ Sketch of proof. 1\. First, let us consider a Berezin operator in the narrow sense. We write the equation (1.15) $\Bigl{(}\sum w_{j}^{+}\xi_{j}+\sum w_{j}^{-}\frac{\partial}{\partial\xi_{j}}\Bigr{)}\int\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}f(\eta)e^{-\eta\overline{\eta}^{t}}\,d\overline{\eta}\,d\eta=\\\ =\int\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}\Bigl{(}\sum v_{k}^{+}\eta_{k}+\sum v_{k}^{-}\frac{\partial}{\partial\eta_{k}}\Bigr{)}f(\eta)e^{-\eta\overline{\eta}^{t}}\,d\overline{\eta}\,d\eta,$ or, equivalently, $\Bigl{(}\sum w_{j}^{+}\xi_{j}+\sum w_{j}^{-}\frac{\partial}{\partial\xi_{j}}-\sum v_{k}^{-}\overline{\eta}_{k}-\sum v_{k}^{+}\frac{\partial}{\partial\eta_{k}}\Bigr{)}\times\\\ \times\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}=0$ We differentiate the exponential and get $\Bigl{[}\sum_{j}w^{+}_{j}\xi_{j}+\sum_{j}w_{j}^{-}\Bigl{(}\sum_{i}a_{ji}\xi_{i}+\sum_{m}b_{jk}\overline{\eta}_{m}\Bigr{)}-\sum_{k}v_{k}^{-}\overline{\eta}_{k}-\sum_{k}v_{k}^{+}\Bigl{(}-\sum_{l}b_{lk}\xi_{l}+\sum_{m}c_{km}\overline{\eta}_{m}\Bigr{)}\Bigr{]}\times\\\ \times\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}=0.$ Therefore, $\begin{pmatrix}v^{+}&w^{-}\end{pmatrix}=\begin{pmatrix}v^{-}&w^{+}\end{pmatrix}\begin{pmatrix}A&-B\\\ -B^{t}&-C\end{pmatrix}.$ By Lemma 1.15, $P$ is a Lagrangian subspace in $\mathcal{V}_{2q}\oplus\mathcal{V}_{2p}$. 2\. Next, let $\mathfrak{C}$ be a Berezin operator having the form (1.9), $\mathfrak{C}=\mathfrak{D}[\xi_{k_{1}}]\dots\mathfrak{D}[\xi_{k_{\alpha}}]\cdot\mathfrak{B}\cdot\mathfrak{D}[\eta_{m_{1}}]\dots\mathfrak{D}[\eta_{m_{\beta}}]$ (1.17) We note that $\widetilde{a}(v)\mathfrak{D}[\xi_{i}]=\mathfrak{D}[\xi_{i}]\widetilde{a}(\sigma_{i}v),$ where $\sigma_{i}$ is the elementary reflection (1.12). Let the operator $\mathfrak{B}$ satisfy $\widehat{a}(w)\mathfrak{B}=\mathfrak{B}\widehat{a}(v),$ where $v\oplus w$ ranges in $P[\mathfrak{B}]$. Then the operator $\mathfrak{C}$ satisfies $\widehat{a}(\sigma_{k_{1}}\dots\sigma_{k_{\alpha}}w)\mathfrak{C}=\mathfrak{C}\widehat{a}(\sigma_{m_{1}}\dots\sigma_{m_{\beta}}v).$ Therefore the corresponding linear relation is $\sigma_{k_{1}}\dots\sigma_{k_{\alpha}}P\sigma_{m_{1}}\dots\sigma_{m_{\beta}}$ (1.18) and is Lagrangian. This proves a). 3\. By Proposition 1.12, the sets (1.18) sweep the whole Lagrangian Grassmannian. This proves b). 4\. Let $v\oplus w\in P[\mathfrak{B}_{1}]$, $w\oplus y\in P[\mathfrak{B}_{2}]$, i.e., $\widehat{a}(w)\mathfrak{B}_{1}=\mathfrak{B}_{1}\widehat{a}(v),\qquad\widehat{a}(y)\mathfrak{B}_{2}=\mathfrak{B}_{2}\widehat{a}(w).$ Then $\mathfrak{B}_{2}\mathfrak{B}_{1}\widehat{a}(v)=\mathfrak{B}_{2}\widehat{a}(w)\mathfrak{B}_{1}=\widehat{a}(y)\mathfrak{B}_{2}\mathfrak{B}_{1}$ and this proves c). We omit a proof of d), which is not completely obvious. $\square$ 1.27. Explicit correspondence. Another description. The previous proof implies also the following theorem. ###### Theorem 1.17 Let $P$ satisfy Lemma 1.15. Then the corresponding Berezin operator $\mathfrak{B}[P]$ has the kernel $\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A&-B\\\ B^{t}&-C\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}.$ ## 2 A survey of symplectic spinors. Category of Gaussian integral operators In Subsections 2–2, we define the boson Fock space. In Subsections 2–2 we introduce Gaussian integral operators. The algebraic structure of the category of Gaussian integral operators is described in Subsections 2–2. For a detailed exposition, see [21], Section 4. 2.1. Fock space. Denote by $d\lambda(z)$ the Lebesgue measure on ${\mathbb{C}}^{n}$ normalized as $d\lambda(z):=\pi^{-n}dx_{1}\dots dx_{n}\,dy_{1}\dots dy_{n},\qquad\text{where $z_{j}=x_{j}+iy_{j}$}$ The boson Fock space $\mathbf{F}_{n}$ is the space of entire functions on ${\mathbb{C}}^{n}$ satisfying the condition $\int_{{\mathbb{C}}^{n}}|f(z)|^{2}e^{-|z|^{2}}\,d\lambda(z)<\infty.$ We define the inner product in $\mathbf{F}_{n}$ by $\langle f,g\rangle:=\int_{{\mathbb{C}}^{n}}f(z)\overline{g(z)}e^{-|z|^{2}}\,d\lambda(z).$ ###### Theorem 2.1 The space $\mathbf{F}_{n}$ is complete, i.e., it is a Hilbert space. ###### Proposition 2.2 The monomials $z_{1}^{k_{1}}\dots z_{n}^{k_{n}}$ are pairwise orthogonal and $\|z_{1}^{k_{1}}\dots z_{n}^{k_{n}}\|^{2}=\prod k_{j}!$ 2.2. Operators. ###### Theorem 2.3 For each bounded operator $A:\mathbf{F}_{n}\to\mathbf{F}_{m}$ there is a function $K(z,\overline{u})$ on ${\mathbb{C}}^{m}\oplus{\mathbb{C}}^{n}$ holomorphic in $z\in{\mathbb{C}}^{m}$ and antiholomorphic in $u\in{\mathbb{C}}^{n}$ such that $Af(z)=\int_{{\mathbb{C}}^{n}}K(z,\overline{u})\,f(u)\,e^{-|u|^{2}}\,d\lambda(u)$ (the integral absolutely converges for all $f$). 2.3. Gaussian operators. Fix $m$, $n=0,1,2,\dots$ Let $S=\begin{pmatrix}K&L\\\ L^{t}&M\end{pmatrix}$ be a symmetric $(m+n)\times(m+n)$-matrix, i.e., $S=S^{t}$. A Gaussian operator $\mathfrak{B}[S]=\mathfrak{B}\begin{bmatrix}K&L\\\ L^{t}&M\end{bmatrix}:\mathbf{F}_{n}\to\mathbf{F}_{m}$ is defined by $\mathfrak{B}\begin{bmatrix}K&L\\\ L^{t}&M\end{bmatrix}f(z)=\int_{{\mathbb{C}}^{n}}\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}z&\overline{u}\end{pmatrix}\begin{pmatrix}K&L\\\ L^{t}&M\end{pmatrix}\begin{pmatrix}z^{t}\\\ \overline{u}^{t}\end{pmatrix}\Bigr{\\}}f(u)\,e^{-|u|^{2}}\,d\lambda(u),$ where $z:=\begin{pmatrix}z_{1}&\dots&z_{n}\end{pmatrix},\qquad\overline{u}:=\begin{pmatrix}\overline{u}_{1}&\dots&\overline{u}_{n}\end{pmatrix}$ are row-matrices ###### Theorem 2.4 (G. I. Olshanski) An operator $\mathfrak{B}[S]$ is bounded iff 1\. $\|S\|\leqslant 1$, 2\. $\|K\|<1$, $\|M\|<1$. 2.4. Product formula. ###### Theorem 2.5 Let $\mathfrak{B}[S_{1}]=\mathfrak{B}\begin{bmatrix}P&Q\\\ Q^{t}&R\end{bmatrix}:\mathbf{F}_{n}\to\mathbf{F}_{m},\qquad\qquad\mathfrak{B}[S_{2}]=\mathfrak{B}\begin{bmatrix}K&L\\\ L^{t}&M\end{bmatrix}:\mathbf{F}_{m}\to\mathbf{F}_{k}$ be bounded Gaussian operators. Then their product is $\det(1-MP)^{-1/2}\mathfrak{B}[S_{2}*S_{1}],$ where $S_{2}*S_{1}$ is given by $\begin{pmatrix}K&L\\\ L^{t}&M\end{pmatrix}*\begin{pmatrix}P&Q\\\ Q^{t}&R\end{pmatrix}=\\\ =\begin{pmatrix}K+LP(1-MP)^{-1}L^{t}&L(1-PM)^{-1}Q\\\ Q^{t}(1-MP)^{-1}L^{t}&R+Q^{t}(1-MQ)^{-1}MQ\end{pmatrix}.$ (2.1) ###### Theorem 2.6 Denote by $G_{n}$ the set of unitary $(n+n)\times(n+n)$ symmetric matrices $\begin{pmatrix}K&L\\\ L^{t}&M\end{pmatrix}$ satisfying $\|K\|<1$, $\|M\|<1$. Then $G_{n}$ is closed with respect to the $*$-multiplication. Moreover $G_{n}$ is isomorphic to the group $\mathrm{Sp}(2n,{\mathbb{R}})$. However, the formula (2.1) almost coincides with the formula (1.5). Again, it hides a product of linear relations. First, we define analogs of the spaces $\mathcal{V}_{2n}$ from the previous section. 2.5. Complexification of a linear space with bilinear form. Denote by $\mathcal{W}_{2n}$ the space ${\mathbb{C}}^{2n}={\mathbb{C}}^{n}\oplus{\mathbb{C}}^{n}$. Let us denote its elements by $v=\begin{pmatrix}v^{+}&v^{-}\end{pmatrix}$. We equip $\mathcal{W}_{2n}$ with two forms — the skew-symmetric bilinear form $\Lambda(v,w):=\begin{pmatrix}v^{+}&v^{-}\end{pmatrix}\begin{pmatrix}0&1\\\ -1&0\end{pmatrix}\begin{pmatrix}(w^{+})^{t}\\\ (w^{-})^{t}\end{pmatrix}=v^{+}(w^{-})^{t}-v^{-}(w^{+})^{t};$ — the indefinite Hermitian form $M(v,w):=\begin{pmatrix}v^{+}&v^{-}\end{pmatrix}\begin{pmatrix}1&0\\\ 0&-1\end{pmatrix}\begin{pmatrix}(\overline{w}^{+})^{t}\\\ (\overline{w}^{\,-})^{t}\end{pmatrix}=v^{+}(\overline{w}^{+})^{t}-v^{-}(\overline{w}^{\,-})^{t}.$ Remark. I wish to explain the origin of the definition. Consider a real space ${\mathbb{R}}^{2n}$ equipped with a nondegenerate skew-symmetric bilinear form $\\{\cdot,\cdot\\}$. Consider the space ${\mathbb{C}}^{2n}\supset{\mathbb{R}}^{2n}$. We can extend $\\{\cdot,\cdot\\}$ to ${\mathbb{C}}^{2n}$ in the following two ways. First, we can extend it as a bilinear form, $\widetilde{\Lambda}(x+iy,x^{\prime}+iy^{\prime}):=\\{x,x^{\prime}\\}-\\{y,y^{\prime}\\}+i\bigl{(}\\{x,y^{\prime}\\}+\\{x^{\prime},y\\}\bigr{)}.$ Next, we extend $\\{\cdot,\cdot\\}$ as a sesquilinear form $\widetilde{M}(x+iy,x^{\prime}+iy^{\prime}):=\\{x,x^{\prime}\\}+\\{y,y^{\prime}\\}+i\bigl{(}\\{y,x^{\prime}\\}-\\{x,y^{\prime}\\}\bigr{)}.$ Thus we get a space endowed with two forms. To achieve a complete correspondence with the space $\mathcal{W}_{2n}$, we put $\Lambda:=i\widetilde{\Lambda}$, $M:=i\widetilde{M}$. 2.6. The category $\mathbf{Sp}$. Objects of the category $\mathbf{Sp}$ are the spaces $\mathcal{W}_{2n}$, where $n=0$, $1$, $2$, …. We equip the direct sum $\mathcal{W}_{2n}\oplus\mathcal{W}_{2m}$ with two forms, $\displaystyle\Lambda^{\ominus}(v\oplus w,v^{\prime}\oplus w^{\prime}):=\Lambda(v,v^{\prime})-\Lambda(w,w^{\prime}),$ $\displaystyle M^{\ominus}(v\oplus w,v^{\prime}\oplus w^{\prime}):=M(v,v^{\prime})-M(w,w^{\prime}).$ A morphism $\mathcal{W}_{2n}\to\mathcal{W}_{2m}$ is a linear relation $P:\mathcal{W}_{2n}\rightrightarrows\mathcal{W}_{2m}$ satisfying the following conditions. 1\. $P$ is $\Lambda^{\ominus}$-Lagrangian. 2\. The form $M^{\ominus}$ is non-positive on $P$. 3\. The form $M=M_{\mathcal{W}_{2n}}$ is strictly negative on $\ker P$ and the form $M=M_{\mathcal{W}_{2m}}$ is strictly positive on $\mathop{\rm indef}\nolimits P$. A product of morphisms is the usual product of linear relations151515The condition 3 forbids an appearance of $\mathop{\mathsf{null}}\nolimits$.. ###### Observation 2.7 The group of automorphisms of $\mathcal{W}_{2n}$ is the real symplectic group $\mathrm{Sp}(2n,{\mathbb{R}})$. This follows from the remark given in the previous subsection. The group of operators preserving the both forms $\widetilde{\Lambda}$, $\widetilde{M}$ on ${\mathbb{C}}^{2n}$ preserves also the real subspace ${\mathbb{R}}^{2n}$. 2.7. Construction of Gaussian operators from linear relations. Recall that $\mathcal{W}_{2n}$ is ${\mathbb{C}}^{2n}={\mathbb{C}}^{n}\oplus{\mathbb{C}}^{n}$. Denote this decomposition by $\mathcal{W}_{2n}=\mathcal{W}_{n}^{+}\oplus\mathcal{W}_{n}^{-}.$ Represent a linear relation $P$ as graph of an operator $S=S(P):\mathcal{W}_{m}^{-}\oplus\mathcal{W}_{n}^{+}\to\mathcal{W}_{m}^{+}\oplus\mathcal{W}_{n}^{-}.$ This is possible, because $P$ is negative semi-definite and $\mathcal{W}_{m}^{+}\oplus\mathcal{W}_{n}^{-}$ is strictly positive; therefore $S\cap(\mathcal{W}_{m}^{+}\oplus\mathcal{W}_{n}^{-})=0$. ###### Proposition 2.8 A matrix $S$ has a form $S(P)$ iff it is symmetric and satisfies the Olshanski conditions from Theorem 2.4. ###### Theorem 2.9 For each morphisms $P:\mathcal{W}_{2n}\rightrightarrows\mathcal{W}_{2m},\,\qquad Q:\mathcal{W}_{2m}\rightrightarrows\mathcal{W}_{2k},$ the corresponding Gaussian operators $\mathfrak{B}\bigl{[}S(P)\bigr{]}:\mathbf{F}_{n}\to\mathbf{F}_{m},\qquad\mathfrak{B}\bigl{[}S(Q)\bigr{]}:\mathbf{F}_{m}\to\mathbf{F}_{k}$ satisfy $\mathfrak{B}\bigl{[}S(Q)\bigr{]}\mathfrak{B}\bigl{[}S(P)\bigr{]}=\lambda(Q,P)\mathfrak{B}\bigl{[}S(QP)\bigr{]},$ where $QP$ is a product of linear relations and $\lambda(Q,P)$ is a nonzero scalar. As formulated, the theorem can be proved by direct force. 2.8. Construction of linear relations from Gaussian operators. For $\begin{pmatrix}v_{1}^{+}&\dots&v_{n}^{+}&v_{1}^{-}&\dots&v_{n}^{-}\end{pmatrix}\in\mathcal{V}_{2n},$ we define the differential operator (a creation-annihilation operator) $\widehat{a}(v)f(z)=\Bigl{(}\sum_{j}v_{j}^{+}z_{j}+\sum_{j}v_{j}^{-}\frac{\partial}{\partial z_{j}}\Bigr{)}\,f(z).$ For a given Gaussian operator $\mathfrak{B}[S]:\mathbf{F}_{n}\to\mathbf{F}_{m}$ we consider the set $P$ of all $v\oplus w\in\mathcal{W}_{2n}\oplus\mathcal{W}_{2m}$ such that $\widehat{a}(w)\,\mathfrak{B}[S]=\mathfrak{B}[S]\,\widehat{a}(v).$ ###### Theorem 2.10 The linear relation $P$ is a morphism of the category $\mathbf{Sp}$. 2.9. Details. An analog of the Schwartz space. We define by $\mathcal{S}\mathbf{F}_{n}$ the subspace in $\mathbf{F}_{n}$ consisting of all $f(z)=\sum c_{j_{1},\dots,j_{n}}z^{j_{1}}\dots z^{j_{n}}$ such that for each $N$ $\sup_{j}|c_{j_{1},\dots,j_{n}}|\prod_{k}j_{k}!\,{j_{k}}^{N}<\infty.$ ###### Theorem 2.11 The subspace $\mathcal{S}\mathbf{F}_{n}$ is a common invariant domain for all Gaussian bounded operators and for all creation-annihilation operators. See [22], Chapter 5. 2.10. Details. The Olshanski semigroup $\Gamma\mathrm{Sp}(2n,{\mathbb{R}})$. The Olshanski semigroup $\Gamma\mathrm{Sp}(2n,{\mathbb{R}})$ is defined as the subsemigroup in $\mathrm{Sp}(2n,{\mathbb{C}})$ consisting of complex matrices $g$ satisfying the condition $g\begin{pmatrix}-1&0\\\ 0&1\end{pmatrix}g^{*}-\begin{pmatrix}-1&0\\\ 0&1\end{pmatrix}\leqslant 0,$ where $g^{*}=\overline{g}^{t}$ denotes the adjoint matrix. Equivalently, $g\in\Gamma\mathrm{Sp}(2n,{\mathbb{R}})$ iff $M(ug,ug)\leqslant M(u,u)\quad\text{for all $u\in{\mathbb{C}}^{2n}$}.$ The Olshanski semigroup is a subsemigroup in the semigroup of endomorphisms of the object $\mathcal{W}_{2n}$. ## 3 Gauss–Berezin integrals Here we discuss super analogs of Gaussian integrals. Final formulae actually are not used, but their structure is important for us. Apparently these integrals are evaluated somewhere, but I do not know a reference. Calculations in the fermion case are contained in [26]. 3.1. Phantom algebra. Phantom generators $\mathfrak{a}_{1}$, $\mathfrak{a}_{2}$, …are anti-commuting variables, $\mathfrak{a}_{k}\mathfrak{a}_{l}=-\mathfrak{a}_{l}\mathfrak{a}_{k}.$ We define a phantom algebra $\mathcal{A}$ as the algebra of polynomials in the variables $\mathfrak{a}_{j}$. For the sake of simplicity, we assume that the number of variables is infinite. We also call elements of $\mathcal{A}$ phantom constants. The phantom algebra has a natural ${\mathbb{Z}}$-gradation by degree of monomials, $\mathcal{A}=\oplus_{j=0}^{\infty}\mathcal{A}_{j}.$ Therefore, $\mathcal{A}$ admits a ${\mathbb{Z}}_{2}$-gradation, namely $\mathcal{A}_{\mathrm{even}}:=\oplus\mathcal{A}_{2j},\qquad\mathcal{A}_{\mathrm{odd}}:=\oplus\mathcal{A}_{2j+1}.$ We define the automorphism $\mu\mapsto\mu^{\sigma}$ of $\mathcal{A}$ by the rule $\mu^{\sigma}=\begin{cases}\mu\quad&\text{if $\mu$ is even,}\\\ -\mu\quad&\text{if $\mu$ is odd}\end{cases}$ (equivalently, $\mathfrak{a}_{j}^{\sigma}=-\mathfrak{a}_{j}$). The algebra $\mathcal{A}$ is supercommutative in the following sense: $\displaystyle\mu\in\mathcal{A},\,\nu\in\mathcal{A}_{\mathrm{even}}\quad\Longrightarrow\quad\mu\nu=\nu\mu,$ $\displaystyle\mu\in\mathcal{A}_{\mathrm{odd}},\,\nu\in\mathcal{A}_{\mathrm{odd}}\quad\Longrightarrow\quad\mu\nu=-\nu\mu.$ Also, $\mu\in\mathcal{A},\,\nu\in\mathcal{A}_{{\mathrm{odd}}}\Longrightarrow\quad\nu\mu=\mu^{\sigma}\nu,$ (3.1) Next, represent $\mu\in\mathcal{A}$ as $\mu=\sum_{j\geqslant 0}\mu_{j}$, where $\mu_{j}\in\mathcal{A}_{j}$. We define the map $\pi_{\downarrow}:\mathcal{A}\to{\mathbb{C}}$ by $\pi_{\downarrow}(\mu)=\pi_{\downarrow}\bigl{(}\sum\nolimits_{j\geqslant 0}\mu_{j}\bigr{)}:=\mu_{0}\in{\mathbb{C}}.$ Evidently, $\pi_{\downarrow}(\mu_{1}\mu_{2})=\pi_{\downarrow}(\mu_{1})\pi_{\downarrow}(\mu_{2}).$ Take $\varphi\in\mathcal{A}$ such that $\pi_{\downarrow}(\varphi)=0$. Then $\varphi^{N}=0$ for sufficiently large $N$. Therefore, $(1+\varphi)^{-1}:=\sum_{n\geqslant 0}(-\varphi)^{n},$ actually, the sum is finite. In particular, if $\pi_{\downarrow}(\mu)\neq 0$, then $\mu$ is invertible. 3.2. A comment for super-experts. The purpose of this paper is a specific construction and I use a minimal language necessary for this purpose. I think that a translation to the common super language is easy for an expert. Certainly we need not specify a supecommutative algebra $\mathcal{A}$. However, $\mathcal{A}$ is not completely arbitrary; in particular we need well-defined exponentials of even elements of $\mathcal{A}$. 3.3. Berezinian. Let $\begin{pmatrix}P&Q\\\ R&T\end{pmatrix}$ be a block $(p+q)\times(p+q)$-matrix, let $P$, $T$ be composed of even phantom constants, and $Q$, $R$ be composed of odd phantom constants. Then the Berezinian (or Berezin determinant) is161616The usual determinant of a block complex matrix is $\det P\det(T-QP^{-1}R)$. $\mathrm{ber}\begin{pmatrix}P&Q\\\ R&T\end{pmatrix}:=\det P^{-1}\det(T-QP^{-1}R).$ We note that $P$ and $T-QP^{-1}R$ are composed of elements of the commutative algebra $\mathcal{A}_{{\mathrm{even}}}$, therefore their determinants are well-defined. The Berezinian satisfies the multiplicative property of the usual determinant $\mathrm{ber}(A)\,\mathrm{ber}(B)=\mathrm{ber}(AB).$ 3.4. Functions. We consider 3 types of variables: — human (boson) variable, we denote them by $x_{i}$, $y_{j}$ (if they are real) and $z_{i}$, $u_{j}$ (if they are complex); — Grassmann (fermion) variables, we denote them by $\xi_{i}$, $\eta_{j}$ or $\overline{\eta}_{j}$; — phantom generators $\mathfrak{a}_{j}$ as above. Boson variables $x_{l}$ commute with the fermion variables $\xi_{j}$ and phantom constants $\mu\in\mathcal{A}$. We also assume that the fermion variables $\xi_{j}$ and the phantom generators $\mathfrak{a}_{l}$ anticommute, $\xi_{j}\mathfrak{a}_{l}=-\mathfrak{a}_{l}\xi_{j}.$ Fix a collection of boson variables $x_{1}$, …, $x_{p}$ and a collection of fermion variables $\xi_{1}$,…$\xi_{q}$. A function is an expression of the form $f(x,\xi):=\sum_{0<i_{1}<\dots<i_{k}\leqslant q}\mu_{i_{1}\dots i_{k}}h_{i_{1},\dots,i_{k}}(x_{1},\dots,x_{p})\,\xi_{i_{1}}\dots\xi_{i_{k}},$ (3.2) where $\mu$ are phantom constants and $h$ are smooth functions of a real variable $x\in{\mathbb{R}}^{p}$. We say, that a function $f$ is even (respectively odd) if it is an even expression in the total collection $\\{\xi_{i}\\}$, $\\{\mathfrak{a}_{k}\\}$. Remark. The reader can observe that the fermion variables and phantom variables have equal rights in our definition. However, below their roles are rather different; $\xi_{j}$ serve as variables and elements of $\mathcal{A}$ serve as constants (see (3.4)). More precisely, we regard ’functions’ as maps $\qquad\qquad\qquad\qquad\qquad\qquad\qquad C^{\infty}({\mathbb{R}}^{p})\otimes\Lambda_{q}\to\mathcal{A}.\qquad\qquad\qquad\qquad\qquad\qquad\qquad\square$ For a given $f$, we define the function $\pi_{\downarrow}(f)\in\,C^{\infty}({\mathbb{R}}^{p})\otimes\Lambda_{q}$ replacing phantom constants $\mu$ by $\pi_{\downarrow}(\mu)$ $\pi_{\downarrow}(f):=\sum_{0<i_{1}<\dots<i_{k}\leqslant q}\pi_{\downarrow}(\mu_{i_{1}\dots i_{k}})h_{i_{1},\dots,i_{k}}(x_{1},\dots,x_{p})\xi_{i_{1}}\dots\xi_{i_{k}}.$ 3.5. Integral. Now we define the symbols $\int f(x,\xi)\,dx\,\qquad\int f(x,\xi)\,d\xi,\qquad\int f(x,\xi)\,dx\,d\xi.$ The integration in $x$ is the usual termwise integration in (3.2), $\int_{{\mathbb{R}}^{p}}f(x,\xi)\,dx:=\sum_{0<i_{1}<\dots<i_{k}\leqslant q}\mu_{i_{1}\dots i_{k}}\,\xi_{i_{1}}\dots\xi_{i_{k}}\cdot\int_{{\mathbb{R}}^{p}}h_{i_{1},\dots,i_{k}}(x_{1},\dots,x_{p})\,dx.$ The integration in $\xi$ is the usual termwise Berezin integral. $\int f(x,\xi)\,d\xi:=\mu_{123\dots q}\cdot h_{123\dots q}(x_{1},\dots,x_{p}).$ 3.6. Exponential. Let $f(x,\xi)$ be an even expression in $\xi$, $\mathfrak{a}$, i.e., $f(x,\xi)=f(x,-\xi)^{\sigma}$. We define its exponential as usual, it satisfies the usual properties. Namely, $\exp\\{f(x,\xi)\\}:=\sum_{n=0}^{\infty}\frac{1}{n!}f(x,\xi)^{n}.$ (3.3) Since $f_{1}f_{2}=f_{2}f_{1}$, the identity $\exp\\{f_{1}+f_{2}\\}=\exp\\{f_{1}\\}\exp\\{f_{2}\\}$ holds. ###### Observation 3.1 The series (3.3) converges. This is obvious, however we present explanations. We represent $f$ as a polynomial in $\mathfrak{a}$, $\xi$ $f(x,\xi)=r_{\varnothing\varnothing}(x)+\sum_{\\{I,J\\}\neq\\{\varnothing,\varnothing\\}}r_{I,J}(x)\mathfrak{a}^{I}\xi^{J},$ here $I$ is a finite subset $i_{1}<i_{2}<\dots<i_{l}$ in ${\mathbb{N}}$ and $J$ is a subset $j_{1}<\dots<j_{s}$ in $\\{1,2,\dots,q\\}$. Since $f$ is even, only terms with even $|I|+|J|:=l+s$ are present. Then $\exp\bigl{\\{}f(x,\xi)\bigr{\\}}=\exp\bigl{\\{}r_{\varnothing\varnothing}(x)\bigr{\\}}\prod_{(I,J)\neq(\varnothing,\varnothing)}\exp\bigl{\\{}r_{I,J}(x)\mathfrak{a}^{I}\xi^{J}\bigr{\\}}=\\\ =\exp\bigl{\\{}r_{\varnothing\varnothing}(x)\bigr{\\}}\prod_{(I,J)\neq(\varnothing,\varnothing)}(1+r_{I,J}(x)\mathfrak{a}^{I}\xi^{J}).$ Opening brackets, we get a polynomial in $\mathfrak{a}_{j}$, $\xi_{k}$. 3.7. Gauss–Berezin integrals. A special case. Take $p$ real variables $x_{i}$ and $q$ Grassmann variables $\xi_{j}$. Consider the expression $I=\iint\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}x&\xi\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}x^{t}\\\ \xi^{t}\end{pmatrix}\Bigr{\\}}\,dx\,d\xi=\\\ =\iint\exp\Bigl{\\{}\frac{1}{2}\sum_{ij}a_{ij}x_{i}x_{j}+\sum_{ik}b_{ik}x_{i}\xi_{k}+\frac{1}{2}\sum_{kl}c_{kl}\xi_{k}\xi_{l}\Bigr{\\}}\,dx\,d\xi.$ (3.4) The notation t denotes the transpose as above, also $\begin{pmatrix}x&\xi\end{pmatrix}$ is the row-matrix composed of symbols $\begin{pmatrix}x&\xi\end{pmatrix}=\begin{pmatrix}x_{1}&\dots&x_{p}&\xi_{1}&\dots&\xi_{q}\end{pmatrix}.$ The matrices $A$ and $C$ are composed of even phantom constants, $A$ is symmetric, $C$ is skew-symmetric, and $B$ is a matrix composed of odd phantom constants.171717The argument of the exponential must be even in $\xi$, $\mathfrak{a}$. This constrain produces the conditions of parity for $A$, $B$, $C$. The symmetry conditions for $A$, $B$, $C$ are the natural conditions for coefficients of a quadratic form in $x$, $\xi$. ###### Observation 3.2 The integral converges iff the matrix $\mathop{\rm Re}\nolimits\pi_{\downarrow}(A)$ is negative definite. This is evident. Indeed, the integrand $\exp\\{\dots\\}$ is a finite sum of the form $\exp\Bigl{\\{}\frac{1}{2}\sum_{ij}\pi_{\downarrow}(a_{ij})x_{i}x_{j}\Bigr{\\}}\sum_{i_{1}<\dots<i_{k}}\,\,\sum_{j_{1}<\dots<j_{l}}P_{i_{1},\dots,i_{k};\,j_{1},\dots,j_{l}}(x)\xi_{i_{1}}\dots\xi_{i_{k}}\mathfrak{a}_{j_{1}}\dots\mathfrak{a}_{j_{l}},$ where $P(x)$ are polynomials. Under the condition of $\mathop{\rm Re}\nolimits\pi_{\downarrow}(A)<0$, a term-wise integration is possible. $\square$ 3.8. Evaluation of the Gauss–Berezin integral. ###### Theorem 3.3 Let $\mathop{\rm Re}\nolimits A<0$. Then $I=\begin{cases}(2\pi)^{p/2}\det(-A)^{-1/2}\mathrm{Pfaff}(C+B^{t}A^{-1}B)\quad&\text{if $q$ is even},\\\ 0,\quad&\text{otherwise}.\end{cases}$ Recall that $q$ is the number of Grassmann variables. Remark. The matrix $C+B^{t}A^{-1}B$ is skew-symmetric and composed of even phantom constants. Therefore, the Pfaffian is well defined. $\square$ Remark. Thus, for even $q$ we get a hybrid of a Pfaffian and a Berezinian (see also [14]), $I^{2}=-(2\pi)^{p}\,\mathrm{ber}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}.$ $\square$ Proof. First, we integrate in $x$, $\exp\Bigl{\\{}\frac{1}{2}\xi C\xi^{t}\Bigr{\\}}\int_{{\mathbb{R}}^{p}}\exp\Bigl{\\{}\frac{1}{2}xAx^{t}+xB\xi^{t}\Bigr{\\}}\,dx=\\\ =\exp\Bigl{\\{}\frac{1}{2}\xi C\xi^{t}\Bigr{\\}}\int_{{\mathbb{R}}^{p}}\exp\Bigl{\\{}\frac{1}{2}(x-\xi B^{t}A^{-1})A(x^{t}+A^{-1}B\xi^{t})\Bigr{\\}}\exp\Bigl{\\{}\frac{1}{2}\xi B^{t}A^{-1}B\xi^{t}\Bigr{\\}}\,dx.$ We substitute $y:=x-\xi B^{t}A^{-1},$ (3.5) get $\exp\Bigl{\\{}\frac{1}{2}\xi C\xi^{t}\Bigr{\\}}\cdot\exp\Bigl{\\{}\frac{1}{2}\xi B^{t}A^{-1}B\xi^{t}\Bigr{\\}}\int\exp\Bigl{\\{}\frac{1}{2}yAy^{t}\Bigr{\\}}\,dy,$ and come to the usual Gaussian integral (0.6). Integrating the result, we get $\det(-A)^{-1/2}(2\pi)^{p/2}\int\exp\Bigl{\\{}\frac{1}{2}\xi(C+B^{t}A^{-1}B)\xi^{t}\Bigr{\\}}\,d\xi,$ and come to the Pfaffian. We must justify the substitution (3.5). Let $\Phi$ be a function on ${\mathbb{R}}^{p}$, let $\nu$ be an even expression in $\mathfrak{a}$, $\xi$, assume that the constant term is 0. Then $\int_{{\mathbb{R}}^{p}}\Phi(x+\nu)\,dx=\int_{{\mathbb{R}}^{p}}\Phi(x)\,dx.$ Indeed, $\Phi(x+\nu):=\sum_{j=0}^{\infty}\frac{1}{j!}\nu^{j}\frac{d^{j}}{dx^{j}}\Phi(x)$ Actually, the summation is finite. A termwise integration in $x$ gives zero for all $j\neq 0$. $\square$ 3.9. A preliminary remark. Grassmann Gaussian integral. ###### Observation 3.4 Let $D$ be a complex skew-symmetric matrix, let $\zeta_{i}$ be Grassmann variables. Then the integral $\int\exp\Bigl{\\{}\frac{1}{2}\xi D\xi^{t}+\xi\zeta^{t}\Bigr{\\}}\,d\xi$ can be represented in the form $s\cdot\prod_{j=1}^{m}\bigl{(}\zeta h_{j}^{t}\bigr{)}\cdot\exp\Bigl{\\{}\frac{1}{2}\zeta Q\zeta^{t}\Bigr{\\}},$ where $s\in{\mathbb{C}}$, $Q$ is a skew-symmetric matrix, and $h_{j}$ are row- matrices; $m=0,1,2,\dots$. Indeed, one can find a linear substitution $\xi=\eta S$ such that181818In other words, one can reduce a skew-symmetric matrix over ${\mathbb{C}}$ to a canonical form. $\xi D\xi^{t}=\sum_{j=1}^{\gamma}\eta_{2j}\eta_{2j+1}.$ Then the integral can be reduced to $\det S\int\exp\bigl{\\{}\sum_{j=1}^{\gamma}\eta_{2j}\eta_{2j+1}+\sum_{j=2\gamma+2}^{q}\eta_{j}\nu_{j}\\}\,\,d\eta,$ where $\nu_{j}$ are linear expression in $\zeta_{k}$. We get $\det S\cdot\prod_{j=1}^{\gamma}\int\exp\bigl{\\{}\eta_{2j}\eta_{2j+1}+\eta_{2j}\nu_{2j}+\eta_{2j+}\nu_{2j+1}\bigl{\\}}\,d\eta_{2j}\,\,d\eta_{2j+1}\,\times\\\ \times\prod_{k\geqslant 2\gamma+2}\int\exp\bigl{\\{}\eta_{k}\nu_{k}\bigr{\\}}\,d\eta_{k}=\\\ =\pm\det S\cdot\exp\\{-\sum_{j=1}^{\gamma}\nu_{2j}\nu_{2j+1}\\}\prod_{k\geqslant 2\gamma+2}\nu_{k}.$ Recall that $\nu_{m}$ are certain linear expressions191919They are not canonically defined. in $\zeta_{i}$. $\square$. 3.10. More general Gauss–Berezin integrals. Consider an expression $J=\iint\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}x&\xi\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}x^{t}\\\ \xi^{t}\end{pmatrix}+xh^{t}+\xi g^{t}\Bigr{\\}}\,dx\,d\xi=\\\ =\iint\exp\Bigl{\\{}\frac{1}{2}\sum_{ij}a_{ij}x_{i}x_{j}+\sum_{ik}b_{ik}x_{i}\xi_{k}+\frac{1}{2}\sum_{kl}c_{kl}\xi_{k}\xi_{l}+\sum_{j}h_{j}x_{j}-\sum_{k}g_{k}\xi_{k}\Bigr{\\}}\,dx\,d\xi,$ (3.6) here $A$, $B$, $C$ are as above and $h^{t}$, $g^{t}$ are column-vectors, $h_{j}\in\mathcal{A}_{\mathrm{even}}$, $g_{k}\in\mathcal{A}_{\mathrm{odd}}$. We propose two ways of evaluation of this integral. 3.11. The first way of evaluation. Substituting $\begin{pmatrix}y&\eta\end{pmatrix}=\begin{pmatrix}x&\xi\end{pmatrix}+\begin{pmatrix}h&g\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}^{-1},$ we get $\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}h&g\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}^{-1}\begin{pmatrix}h^{t}\\\ g^{t}\end{pmatrix}\Bigr{\\}}\times\\\ \times\iint\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}y&\eta\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}y^{t}\\\ \eta^{t}\end{pmatrix}\Bigr{\\}}\,dy\,d\eta$ and come to Gauss–Berezin integral (3.4) evaluated above. This way is not perfect, because it uses an inversion of a matrix $\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}$. ###### Observation 3.5 A matrix $\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}$ is invertible iff $A$ and $C$ are invertible. The necessity is evident; to prove the sufficiency, we note that the matrix $T:=\begin{pmatrix}A^{-1}&0\\\ 0&C^{-1}\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}-\begin{pmatrix}1&0\\\ 0&1\end{pmatrix}$ is composed of nilpotent elements of $\mathcal{A}$. We write out $(1+T)^{-1}=1-T+T^{2}-\dots$. $\square$ The matrix $A$ is invertible, because $\mathop{\rm Re}\nolimits A<0$. But the matrix $C$ is skew-symmetric. — If $q$ is even, then a $q\times q$ skew-symmetric matrix $C$ in a general position is invertible. For noninvertible $C$ we have a chance to remove uncertainty. This way leads to an expression of the form (3.8). — If $q$ is odd, then $C$ is non-invertible; our approach is not suitable. 3.12. Second way of evaluation of Gauss–Berezin integrals. First, we integrate in $x$, $\exp\Bigl{\\{}\frac{1}{2}\xi C\xi^{t}+\xi g^{t}\Bigr{\\}}\int_{{\mathbb{R}}^{p}}\exp\Bigl{\\{}\frac{1}{2}xAx^{t}+xB\xi^{t}+xh^{t}\Bigr{\\}}\,dx=\\\ =\exp\Bigl{\\{}\frac{1}{2}\xi C\xi^{t}+\xi g^{t}\Bigr{\\}}\times\\\ \times\exp\Bigl{\\{}\frac{1}{2}(h-\xi B^{t}A^{-1})A(h^{t}+A^{-1}B\xi^{t})\Bigr{\\}}\int_{{\mathbb{R}}^{p}}\exp\Bigl{\\{}\frac{1}{2}(x+h-\xi B^{t})A(x^{t}+h^{t}+B\xi^{t})\Bigr{\\}}\,dx.$ Substituting $y=x+h-\xi B^{t}A^{-1}$ and integrating in $y$, we get $(2\pi)^{p/2}\det(-A)^{-1/2}\exp\Bigl{\\{}\frac{1}{2}(h-\xi B^{t}A^{-1})A(h^{t}+A^{-1}B\xi^{t})+\frac{1}{2}\xi C\xi^{t}+\xi g^{t}\Bigr{\\}}.$ Next, we must integrate in $\xi$, the integral has a form $\int\exp\Bigl{\\{}\frac{1}{2}\xi D\xi^{t}+\xi r^{t}\Bigr{\\}}\,d\xi,$ (3.7) where a matrix $D$ is composed of even phantom constants and a vector $r$ is odd. If $D$ is invertible, we shift the argument again $\eta^{t}:=\xi^{t}+D^{-1}r^{t}$ and get $\exp\Bigl{\\{}\frac{1}{2}rD^{-1}r^{t}\Bigr{\\}}\int\exp\Bigl{\\{}\frac{1}{2}\eta D\eta^{t}\Bigr{\\}}\,d\eta,$ the last integral is a Pfaffian. This way is equivalent to the approach discussed in the previous subsection. Now consider an arbitrary $D$. The calculation of Subsection 3 does not survive202020Let $D$ be a skew-symmetric matrix over $\mathcal{A}_{{\mathrm{even}}}$. If $\pi_{\downarrow}(D)$ is degenerate, then we can not reduce $D$ to a canonical form.. However, we can write (3.7) explicitly as follows. For any subset $I:\,i_{1}<\dots<i_{2k}$ in $\\{1,\dots,q\\}$ we consider the complementary subset $J:j_{1}<\dots<j_{q-2k}$. Define the constant $\sigma(I)=\pm 1$ as follows $\bigl{(}\xi_{i_{1}}\xi_{i_{2}}\dots\xi_{i_{2k}}\bigr{)}\bigl{(}\xi_{j_{1}}\xi_{j_{2}}\dots\xi_{j_{q-2k}}\bigr{)}=\sigma(I)\xi_{1}\xi_{2}\dots\xi_{q}.$ Evidently, $\int\exp\Bigl{\\{}\frac{1}{2}\xi D\xi^{t}+\xi r^{t}\Bigr{\\}}\,d\xi=\\\ =\sum_{I}\sigma(I)\mathrm{Pfaff}\begin{pmatrix}0&d_{i_{1}i_{2}}&\dots&d_{i_{1}i_{2k}}\\\ d_{i_{2}i_{1}}&0&\dots&d_{i_{2}i_{2k}}\\\ \vdots&\vdots&\ddots&\vdots\\\ d_{i_{2k}i_{1}}&d_{i_{2}i_{2k}}&\dots&0\end{pmatrix}r_{j_{1}}\dots r_{j_{q-2k}}.$ (3.8) Recall that $d_{pq}\in\mathcal{A}_{\mathrm{even}}$ and $r\in\mathcal{A}_{\mathrm{odd}}$. ## 4 Gauss–Berezin integral operators Here we define super hybrids of Gaussian operators and Berezin operators. 4.1. Fock–Berezin spaces. Fix $p$, $q=0$, $1$, $2$, …. Let $z_{1}$, …, $z_{p}$ be complex variables, $\xi_{1}$, …, $\xi_{q}$ be Grassmann variables. We consider expressions $f(z,\xi):=\sum_{0<i_{1}<\dots<i_{k}\leqslant q}\mu_{i_{1}\dots i_{k}}h_{i_{1},\dots,i_{k}}(z_{1},\dots,z_{p})\xi_{i_{1}}\dots\xi_{i_{k}},$ where $h$ are entire functions in $z$. We define the map $f\mapsto\pi_{\downarrow}(f)$ as above. Expand $f$ as above, $f(z,\xi)=:\sum_{I,J}r_{I,J}(z)\mathfrak{a}^{I}\xi^{J}.$ We define the Fock–Berezin $\mathbf{F}_{p,q}$ as the space of all functions $f(z,\xi)$ satisfying the condition: for each $I$, $J$, the function $r_{I,J}(z)$ is in the Schwartz–Fock space $\mathcal{S}\mathbf{F}_{p}$, see Subsection 2. Remark. This definition admits some variations, our choice is volitional. — We can assume that all the $r_{I,J}$ are in the Hilbert–Fock space $\mathbf{F}_{p}$. But some Gauss–Berezin operators defined below are unbounded in this space; therefore this point of view requires explanations of domain of definition of operators, meaning of products of operators, etc.. — We can assume that all $r_{I,J}$ are in the space dual to $\mathcal{S}\mathbf{F}_{p}$, see [22], Chapter 5. This is more-or-less equivalent to our approach, but requires more comments. $\square$ 4.2. Another form of Gauss–Berezin integral. Consider the integral $\int\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}z&\xi\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}z^{t}\\\ \xi^{t}\end{pmatrix}+z\alpha^{t}+\xi\beta^{t}\Bigr{\\}}\times\\\ \times\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\overline{z}&\overline{\xi}\end{pmatrix}\begin{pmatrix}K&L\\\ -L^{t}&M\end{pmatrix}\begin{pmatrix}\overline{z}\\\ \overline{\xi}\end{pmatrix}+\overline{z}\varkappa+\overline{\xi}\lambda^{t}\Bigr{\\}}\cdot e^{-z\overline{z}^{t}-\xi\overline{\xi}^{t}}\,dz\,\overline{d}z\,d\xi\,d\overline{\xi},$ (4.1) where two matrices are as above, row-vectors $\alpha$, $\varkappa$ are even, the vectors $\beta$, $\lambda$ are odd. Since ${\mathbb{C}}^{n}\simeq{\mathbb{R}}^{2n}$, this integral is a special case of the Gauss–Berezin integral. We get $\mathrm{const}\cdot\exp\left\\{\frac{1}{2}\begin{pmatrix}\alpha&\beta&\varkappa&\lambda\end{pmatrix}\begin{pmatrix}-A&-B&1&0\\\ B^{t}&-C&0&1\\\ 1&0&-K&-L\\\ 0&-1&L^{t}&-M\end{pmatrix}^{-1}\begin{pmatrix}\alpha^{t}\\\ \beta^{t}\\\ \varkappa^{t}\\\ \lambda^{t}\end{pmatrix}\right\\},$ (4.2) where the scalar factor is a hybrid of a Pfaffian and a Berezinian mentioned above. 4.3. Integral operators. We write operators $\mathbf{F}_{p,q}\to\mathbf{F}_{r,s}$ as $Af(z,\xi)=\int K(z,\xi;\overline{u},\overline{\eta})\,f(u,\eta)\,e^{-z\overline{z}^{t}-\eta\overline{\eta}^{t}}\,du\,d\overline{u}\,d\overline{\eta}\,d\eta.$ (4.3) 4.4. Linear and antilinear operators. We say that an operator $A:\mathbf{F}_{p,q}\to\mathbf{F}_{r,s}$ is linear if $A(f_{1}+f_{2})=A(f_{1}+f_{2}),\qquad A(\lambda f)=\lambda Af,\,\,\text{where $\lambda$ is a phantom constant},$ and antilinear if $A(f_{1}+f_{2})=A(f_{1}+f_{2}),\qquad A(\lambda f)=\lambda^{\sigma}Af,\,\,\text{where $\lambda$ is a phantom constant}.$ In fact, the operators $Af(z,\xi)=\xi_{j}f(z,\xi),\qquad Bf(z,\xi)=\frac{\partial}{\partial\xi_{j}}f(z,\xi),\qquad Cf(z,\xi)=\mathfrak{a}_{j}f(z,\xi)$ are antilinear. An integral operator (4.3) is linear if the kernel $K(z,\xi,\overline{u},\overline{\eta})$ is an even function in the totality of Grassmann variables $\xi$, $\overline{\eta}$, $\mathfrak{a}$. $K(x,\xi,\eta)=K(x,-\xi,-\overline{\eta})^{\sigma}.$ An operator is antilinear iff the function $K$ is odd. Below we meet only linear and antilinear operators. We define also the (antilinear) operator $\mathsf{S}$ of $\sigma$-conjugation, $\mathsf{S}\,f(z,\xi)=f(z,\xi)^{\sigma}$ or, more formally, $\mathsf{S}\sum\mu_{i_{1}\dots i_{k}}h_{i_{1},\dots,i_{k}}(z_{1},\dots,z_{p})\xi_{i_{1}}\dots\xi_{i_{k}}=\\\ =\sum\mu_{i_{1}\dots i_{k}}^{\sigma}h_{i_{1},\dots,i_{k}}(z_{1},\dots,z_{p})\xi_{i_{1}}\dots\xi_{i_{k}}.$ (4.4) Evidently, $\mathsf{S}^{2}f=f.$ 4.5. Gauss–Berezin vectors in the narrow sense. A Gauss–Berezin vector (in the narrow sense) in $\mathbf{F}_{p,q}$ is a vector of the form $\mathbf{b}\begin{bmatrix}A&B\\\ -B^{t}&C\end{bmatrix}=\lambda\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}z&\xi\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}z^{t}\\\ \xi^{t}\end{pmatrix}\Bigr{\\}},$ (4.5) where $A$, $B$, $C$ are as above, see Subsection 3. ###### Observation 4.1 $\mathbf{b}[\cdot]\in\mathbf{F}_{p,q}$ iff $\|\pi_{\downarrow}(A)\|<1$. 4.6. Gauss–Berezin operators in the narrow sense. A Gauss–Berezin integral operator in the narrow sense is an integral operator $\mathbf{F}_{p,q}\to\mathbf{F}_{r,s},$ whose kernel is a Gauss–Berezin vector. In other words, a Gauss–Berezin operator has the form $\mathfrak{B}f(z,\xi)=\\\ =\lambda\cdot\iint\exp\left\\{\frac{1}{2}\begin{pmatrix}z&\xi&\overline{u}&\overline{\eta}\end{pmatrix}\begin{pmatrix}A_{11}&A_{12}&A_{13}&A_{14}\\\ A_{21}&A_{22}&A_{23}&A_{24}\\\ A_{31}&A_{32}&A_{33}&A_{34}\\\ A_{41}&A_{42}&A_{43}&A_{44}\end{pmatrix}\begin{pmatrix}z^{t}\\\ \xi^{t}\\\ \overline{u}^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\right\\}\,f(u,\eta)\times\\\ \times e^{-\eta\overline{\eta}^{t}-u\overline{u}^{t}}\,du\,d\overline{u}\,d\eta d\overline{\eta},$ (4.6) where $\lambda$ is an even phantom constant, $A_{ij}$ is composed of even constants if $(i+j)$ is even, otherwise $A_{ij}$ is composed of odd constants. They also satisfy the natural conditions of the symmetry for a matrix of a quadratic form in the variables $z$, $\xi$, $\overline{u}$, $\overline{\eta}$. Remark. On the other hand, a Gauss–Berezin vector can be regarded as a Gauss–Berezin operator $\mathbf{F}_{0,0}\to\mathbf{F}_{p,q}$. $\square$ For Gauss–Berezin operators $\mathfrak{B}_{1}:\mathbf{F}_{p,q}\to\mathbf{F}_{p^{\prime},q^{\prime}},\qquad\mathfrak{B}_{2}:\mathbf{F}_{p^{\prime},q^{\prime}}\to\mathbf{F}_{p^{\prime\prime},q^{\prime\prime}}$ evaluation of their product is reduced to the Gauss–Berezin integral (4.1). For operators in general position, we can apply formula (4.2). Evidently, in this case the product is a Gauss–Berezin operator again. However, our final Theorem 8.4 avoids this calculation. Also, considerations of Section 1 suggest an extension of the definition of Gauss–Berezin operators. 4.7. General Gauss–Berezin operators. As above, we define first order differential operators $\mathfrak{D}[\xi_{j}]f:=\Bigl{(}\xi_{j}+\frac{\partial}{\partial\xi_{j}}\Bigr{)}f.$ If a function $f$ is independent on $\xi_{j}$, then $\mathfrak{D}[\xi_{j}]f=\xi_{j}f,\qquad\mathfrak{D}[\xi_{j}]\xi_{j}f=f.$ Evidently, $\mathfrak{D}[\xi_{j}]^{2}=1,\qquad\mathfrak{D}[\xi_{i}]\mathfrak{D}[\xi_{j}]=-\mathfrak{D}[\xi_{j}]\mathfrak{D}[\xi_{i}],\quad,i\neq j.$ These operators are antilinear. A Gauss–Berezin operator $\mathbf{F}_{p,q}\to\mathbf{F}_{r,s}$ is an operator of the form $\mathfrak{C}=\lambda\mathfrak{D}[\xi_{i_{1}}]\dots\mathfrak{D}[\xi_{i_{k}}]\,\mathfrak{B}\,\mathfrak{D}[\eta_{m_{1}}]\dots\mathfrak{D}[\eta_{m_{l}}]\cdot\mathsf{S}^{k+l},$ (4.7) where the operator $\mathsf{S}$ is given by (4.4) and — $\mathfrak{B}$ is a Gauss –Berezin operator in the narrow sense; — $i_{1}<i_{2}<\dots<i_{k}$, $m_{1}<m_{2}<\dots<m_{l}$, and $k$, $l\geqslant 0$; — $\lambda$ is an even invertible phantom constant. Note that a Gauss–Berezin operator is linear. Remark. We define the set of Gauss–Berezin operators as a union of $2^{p+q}$ sets. These sets are not disjoint. Actually, we get a (super)manifold consisting of two connected components (according the parity of $k+l$). Each set (4.7) is open and dense in the corresponding component. This will be obvious below. $\square$ 4.8. Remark. Operators $\pi_{\downarrow}(\mathfrak{B})$. Let $K(z,\xi,\overline{u},\overline{\eta})$ be the kernel of a Gauss–Berezin operator. Then the formula $\mathfrak{C}f(z,\xi)=\int\pi_{\downarrow}\Bigl{(}K(z,\xi;\overline{u},\overline{\eta})\Bigr{)}\,f(u,\eta)\,e^{-z\overline{z}^{t}-\eta\overline{\eta}^{t}}\,du\,d\overline{u}\,d\overline{\eta}\,d\eta$ determines an integral operator $\pi_{\downarrow}(\mathfrak{C}):\,\mathbf{F}_{p}\otimes\Lambda_{q}\to\mathbf{F}_{r}\otimes\Lambda_{s}.$ Evidently, this operator is a tensor product of a Gaussian operator $\pi_{\downarrow}^{+}(\mathfrak{C}):\mathbf{F}_{p}\to\mathbf{F}_{r}$ and a Berezin operator $\pi_{\downarrow}^{-}(\mathfrak{C}):\Lambda_{q}\to\Lambda_{s}.$ For instance, for an operator $\mathfrak{B}$ given by the standard formula (4.6), we get the Gaussian operator $\pi_{\downarrow}^{+}(\mathfrak{B})f(z)=\int_{{\mathbb{C}}^{n}}\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}z&\overline{u}\end{pmatrix}\begin{pmatrix}A_{11}&A_{13}\\\ A_{31}&A_{33}\end{pmatrix}\begin{pmatrix}z^{t}\\\ \overline{u}^{t}\end{pmatrix}\Bigr{\\}}\,f(z)\,e^{-z\overline{z}^{t}}\,dz\,d\overline{z}$ and the Berezin operator $\pi_{\downarrow}^{-}(\mathfrak{B})g(\xi)=\int\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}\xi&\overline{\eta}\end{pmatrix}\begin{pmatrix}A_{22}&A_{24}\\\ A_{42}&A_{44}\end{pmatrix}\begin{pmatrix}\xi^{t}\\\ \overline{\eta}^{t}\end{pmatrix}\Bigr{\\}}\,g(\xi)\,e^{-\xi\overline{\xi}^{t}}\,d\xi\,d\overline{\xi}.$ Next, $\pi_{\downarrow}^{\pm}(\mathfrak{C}_{1}\mathfrak{C}_{2})=\pi_{\downarrow}^{\pm}(\mathfrak{C}_{1})\pi_{\downarrow}^{\pm}(\mathfrak{C}_{2}).$ (4.8) 4.9. Products of Gauss–Berezin operators. ###### Theorem 4.2 For each Gauss–Berezin operators $\mathfrak{B}_{1}:\mathbf{F}_{p,q}\to\mathbf{F}_{p^{\prime},q^{\prime}},\qquad\mathfrak{B}_{2}:\mathbf{F}_{p^{\prime},q^{\prime}}\to\mathbf{F}_{p^{\prime\prime},q^{\prime\prime}},$ their product $\mathfrak{B}_{2}\mathfrak{B}_{1}$ is either a Gauss–Berezin operator or $\pi_{\downarrow}(\mathfrak{B}_{2}\mathfrak{B}_{1}f)=0$ (4.9) for all $f$. A proof is in Section 8. We also present an interpretation of a product in terms of linear relations. Remark. In the case (4.9) the kernel of the product has the form (3.8) but it is not a Gauss–Berezin operator in our sense. Possibly this requires to change our definitions. $\square$ 4.10. General Gauss–Berezin vectors. A Gauss–Berezin vector is a vector of the form $\mathfrak{D}[\xi_{i_{1}}]\dots\mathfrak{D}[\xi_{i_{k}}]\mathsf{S}^{k}\mathfrak{b},$ where $\mathfrak{b}$ is a Gauss–Berezin vector in the narrow sense. Remark. We write $\mathsf{S}^{k}$ for a correspondence with (4.7). However omitting this factor, we get the same definition. ## 5 Super-groups $\mathrm{OSp}(2p|2q)$ Here we define a super analog of the groups $\mathrm{O}(2n,{\mathbb{C}})$ and $\mathrm{Sp}(2n,{\mathbb{C}})$. For a general discussion of supergroups and super-Grassmannians, see books [4], [17], [15], [10]. 5.1. Modules $\mathcal{A}^{p|q}$. Let $\mathcal{A}^{p|q}:=\mathcal{A}^{p}\oplus\mathcal{A}^{q}$ be a direct sum of $(p+q)$ copies of $\mathcal{A}$. We regard elements of $\mathcal{A}^{p|q}$ as row-vectors $(v_{1},\dots,v_{p};w_{1},\dots,w_{q}).$ We define a structure of $\mathcal{A}$-bimodule on $\mathcal{A}^{p|q}$. The addition in $\mathcal{A}^{p|q}$ is natural. The left multiplication by $\lambda\in\mathcal{A}$ is also natural $\lambda\circ(v_{1},\dots,v_{p};w_{1},\dots,w_{q}):=(\lambda v_{1},\dots,\lambda v_{p};\lambda w_{1},\dots,\lambda w_{q}).$ The right multiplications by $\varkappa\in\mathcal{A}$ is $(v_{1},\dots,v_{p};w_{1},\dots,w_{q})*\varkappa:=(v_{1}\varkappa,\dots,v_{p}\varkappa;w_{1}\varkappa^{\sigma},\dots,w_{q}\varkappa^{\sigma}),$ where $\sigma$ is the involution of $\mathcal{A}$ defined above. We define the even part of $\mathcal{A}^{p|q}$ as $(\mathcal{A}_{{\mathrm{even}}})^{p}\oplus(\mathcal{A}_{{\mathrm{odd}}})^{q}$ and the odd part as $(\mathcal{A}_{{\mathrm{odd}}})^{p}\oplus(\mathcal{A}_{{\mathrm{even}}})^{q}$. 5.2. Matrices. Denote by $\mathrm{Mat}(p|q)$ the space of $(p+q)\times(p+q)$ matrices over $\mathcal{A}$, we represent such matrices in the block form $Q=\begin{pmatrix}A&B\\\ C&D\end{pmatrix}.$ We say that a matrix $Q$ is even if all matrix elements of $A$, $D$ are even and all matrix elements of $B$, $C$ are odd. A matrix is odd if elements of $A$, $D$ are odd and elements of $B$, $C$ are even. A matrix $Q$ acts on the space $\mathcal{A}^{p|q}$ as $v\to vQ.$ Such transformations are compatible with the left $\mathcal{A}$-module structure on $\mathcal{A}^{p|q}$, i.e., $(\lambda\circ v)Q=\lambda\circ(vQ)\qquad\lambda\in\mathcal{A},\quad v\in\mathcal{A}^{p|q}.$ However, even matrices also regard the right $\Lambda$-module structure, $(v*\lambda)Q=(vQ)*\lambda\qquad\lambda\in\mathcal{A}$ (we use the rule (3.1)). 5.3. Super-transposition. The supertranspose of $Q$ is defined by $Q^{st}=\begin{pmatrix}A&B\\\ C&D\end{pmatrix}^{st}:=\begin{cases}\begin{pmatrix}A^{t}&C^{t}\\\ -B^{t}&D^{t}\end{pmatrix}\quad\text{if $Q$ is even},\\\ \begin{pmatrix}A^{t}&-C^{t}\\\ B^{t}&D^{t}\end{pmatrix}\quad\text{if $Q$ is odd},\end{cases}$ and $(Q_{1}+Q_{2})^{st}:=Q_{1}^{st}+Q_{2}^{st}$ for even $Q_{1}$ and odd $Q_{2}$. The following identity holds $(QR)^{st}=\begin{cases}R^{st}Q^{st}\quad&\text{if $Q$ or $R$ are even},\\\ -R^{st}Q^{st}\quad&\text{if both $R$ or $Q$ are odd}.\end{cases}$ (5.1) Below we use only the first row. 5.4. The supergroups $\mathrm{GL}(p|q)$. The supergroup $\mathrm{GL}(p|q)$ is the group of even invertible matrices. The following lemma is trivial. ###### Lemma 5.1 An even matrix $Q\in\mathrm{Mat}(p|q)$ is invertible a) iff the matrices $A$, $D$ are invertible; b) iff the matrices $\pi_{\downarrow}(A)$, $\pi_{\downarrow}(D)$ are invertible. Here $\pi_{\downarrow}(A)$ denotes the matrix composed of elements $\pi_{\downarrow}(a_{kl})$. Also, the map $Q\mapsto\pi_{\downarrow}(Q)$ is a well-defined epimorphism $\pi_{\downarrow}:\mathrm{GL}(p|q)\to\mathrm{GL}(p,{\mathbb{C}})\times\mathrm{GL}(q,{\mathbb{C}})$ (because $\pi_{\downarrow}(B)=0$, $\pi_{\downarrow}(D)=0$). 5.5. The supergroup $\mathrm{OSp}(2p|2q)$. We define the standard orthosymplectic form on $\mathcal{A}^{2p|2q}$ by $\mathfrak{s}(u,v):=uJv^{st},$ where $J$ is a block $(p+p+q+q)\times(p+p+q+q)$ matrix $J:=\frac{1}{2}\begin{pmatrix}0&1&0&0\\\ -1&0&0&0\\\ 0&0&0&1\\\ 0&0&1&0\end{pmatrix}.$ (5.2) The supergroup $\mathrm{OSp}(2p|2q)$ is the subgroup in $\mathrm{GL}(2p|2q)$ consisting of matrices $g$ satisfying $\mathfrak{s}(u,v)=\mathfrak{s}(ug,vg).$ Equivalently, $gJg^{st}=J$ (for this conclusion, we use (5.1); since $g\in\mathrm{GL}(p|q)$ is even, a sign does not appear). For an element $g=\begin{pmatrix}A&B\\\ C&D\end{pmatrix}$, $\pi_{\downarrow}(A)\begin{pmatrix}0&1\\\ -1&0\end{pmatrix}\pi_{\downarrow}(A)^{t}=\begin{pmatrix}0&1\\\ -1&0\end{pmatrix},\quad\pi_{\downarrow}(D)\begin{pmatrix}0&1\\\ 1&0\end{pmatrix}\pi_{\downarrow}(D)^{t}=\begin{pmatrix}0&1\\\ 1&0\end{pmatrix},$ i.e., $\pi_{\downarrow}(A)\in\mathrm{Sp}(2n,{\mathbb{C}}),\quad\pi_{\downarrow}(D)\in\mathrm{O}(2n,{\mathbb{C}}).$ 5.6. The super Olshanski semigroup $\Gamma\mathrm{OSp}(2p|2q)$. We define the super semigroup (semi-supergroup) $\Gamma\mathrm{OSp}(2p|2q)$ as a subsemigroup in $\mathrm{OSp}(2p|2q)$ consisting of matrices $g=\begin{pmatrix}A&B\\\ C&D\end{pmatrix}$ such that $\pi_{\downarrow}(A)$ is contained in the Olshanski semigroup $\Gamma\mathrm{Sp}(2p,{\mathbb{R}})$, see Subsection 2. ## 6 Super-Grassmannians This section is a preparation to the definition of superlinear relations. 6.1. Super-Grassmannians. Let $u_{1}$,…, $u_{r}$ be even vectors and $v_{1}$,…, $v_{s}$ be odd vectors in $\mathcal{A}^{p|q}$. We suppose that — $\pi_{\downarrow}(u_{j})\in({\mathbb{C}}^{p}\oplus 0)$ are linear independent, — $\pi_{\downarrow}(v_{k})\in(0\oplus{\mathbb{C}}^{q})$ are linear independent. A subspace of dimension $\\{r|s\\}$ is a left $\mathcal{A}$-module generated by such vectors. Subspaces also are right $\mathcal{A}$-submodules. We define the (super-)Grassmannian $\mathrm{Gr}^{r|s}_{p|q}$ as the space of all subspaces in $\mathcal{A}^{p|q}$ of dimension $\\{r|s\\}$. By the definition, the map $\pi_{\downarrow}$ projects $\mathrm{Gr}^{r|s}_{p|q}$ to the product $\mathrm{Gr}^{r}_{p}\times\mathrm{Gr}^{s}_{q}$ of human complex Grassmannians. We denote by $\pi_{\downarrow}^{\pm}$ the natural projections $\pi_{\downarrow}^{+}:\mathrm{Gr}^{r|s}_{p|q}\to\mathrm{Gr}_{p}^{r},\quad\pi_{\downarrow}^{-}:\mathrm{Gr}^{r|s}_{p|q}\to\mathrm{Gr}^{s}_{q}.$ 6.2. Intersections of subspaces. ###### Lemma 6.1 Let $L$ be an $r|s$-dimensional subspace in $\mathcal{A}^{p|q}$, $M$ be a $\rho|\sigma$\- dimensional subspace. Let the following transversality conditions hold $\pi_{\downarrow}^{+}(L)+\pi_{\downarrow}^{+}(M)={\mathbb{C}}^{p},\quad\pi_{\downarrow}^{-}(L)+\pi_{\downarrow}^{-}(M)={\mathbb{C}}^{q}.$ (6.1) Then $L\cap M$ is a subspace and its dimension is $(r+\rho-p)|(s+\sigma-q)$. Remark. If the transversality conditions are not satisfied, then incidentally $L\cap M$ is not a subspace. For instance, consider $\mathcal{A}^{1|1}$ with a basis $e_{1}$, $e_{2}$ and subspaces $L:=\mathcal{A}(e_{1}+\xi_{1}e_{2}),\qquad M:=\mathcal{A}\cdot e_{1}.$ Then $L\cap M=\mathcal{A}\xi_{1}e_{1}$ is not a subspace. $\square$ Proof. Denote by $I\subset\mathcal{A}$ the ideal spanned by all $\mathfrak{a}_{j}$, i.e., $\mathcal{A}/I={\mathbb{C}}$. Evidently, $L+M=\mathcal{A}^{p|q}.$ (6.2) Let $v\in\pi_{\downarrow}^{+}(L)\cap\pi_{\downarrow}^{+}(M)$. Choose $x\in L$, $y\in M$ such that $\pi_{\downarrow}(x)=v$, $\pi_{\downarrow}(y)=v$. Then $x-y\in I\cdot\mathcal{A}^{p|q}$. However, $I\cdot L+I\cdot M=I\cdot\mathcal{A}^{p|q}$, therefore we can represent $x-y=a-b,\qquad\text{where $u\in I\cdot L$, $v\in I\cdot M$}.$ Then $(x-a)\in L,\,(y-b)\in M,\,\,\pi_{\downarrow}(x-a)=v=\pi_{\downarrow}(y-b).$ Thus, for any vector $v\in\pi_{\downarrow}^{+}(L)\cap\pi_{\downarrow}^{+}(M)$, there is a vector $v^{*}\in L\cap M$ such that $\pi_{\downarrow}(v^{*})=v$. The same is valid for vectors $w\in\pi_{\downarrow}^{-}(L)\cap\pi_{\downarrow}^{-}(M)$. Therefore, $L\cap M$ contains a super subspace of desired dimension generated by vectors $v^{*}$, $w^{*}$. It remains to show that there are no extra vectors. Now let us vary a phantom algebra $\mathcal{A}$. If $\mathcal{A}$ is an algebra in finite number of Grassmann constants $\mathfrak{a}_{1}$,…, $\mathfrak{a}_{n}$, then this completes a proof, since (6.2) gives the same dimension of the intersection over ${\mathbb{C}}$. Otherwise, we choose a basis in $L$ and a basis in $M$. Expressions for basis vectors contain only finite number of Grassmann constants $\mathfrak{a}_{1}$, …, $\mathfrak{a}_{k}$. After this, we apply the same reasoning to algebras $\mathcal{A}[l]$ generated by Grassmann constants $\mathfrak{a}_{1}$, …, $\mathfrak{a}_{l}$ for all $l\geqslant k$ and observe that $L\cap M$ does not contain extra vectors. $\square$ 6.3. Atlas on the super-Grassmannian. Define an atlas on the Grassmannian $\mathrm{Gr}^{r|s}_{p|q}$ as usual. Namely, consider the following complementary subspaces $V_{+}:=(\mathcal{A}^{r}\oplus 0)\oplus(\mathcal{A}^{s}\oplus 0)\quad\quad V_{-}:=(0\oplus\mathcal{A}^{p-r})\oplus(0\oplus\mathcal{A}^{q-s})$ in $\mathcal{A}^{p|q}$. Let $S:V_{+}\to V_{-}$ be an even operator. Then its graph is an element of the Grassmannian. Permutating coordinates is $\mathcal{A}^{p}$ and $\mathcal{A}^{q}$, we get an atlas that covers the whole Grassmannian $\mathrm{Gr}^{r|s}_{p|q}$. 6.4. Lagrangian super-Grassmannians. Now, equip the space $\mathcal{A}^{2p|2q}$ with the orthosymplectic form $\mathfrak{s}$ as above. We say that a subspace $L$ is isotropic if the form $\mathfrak{s}$ is zero on $L$. A Lagrangian subspace $L$ is an isotropic subspace of the maximal possible dimension, i.e., $\dim L=\\{p|q\\}$. ###### Observation 6.2 Let $L$ be a super-Lagrangian subspace. Then — $\pi_{\downarrow}^{+}(L)$ is Lagrangian subspace in ${\mathbb{C}}^{2p}$ with respect to the skew-symmetric bilinear form $\begin{pmatrix}0&1\\\ -1&0\end{pmatrix}$. — $\pi_{\downarrow}^{-}(L)$ is a Lagrangian subspace in ${\mathbb{C}}^{2q}$ with respect to the symmetric bilinear form $\begin{pmatrix}0&1\\\ 1&0\end{pmatrix}$. 6.5. Coordinates on Lagrangian super-Grassmannian. Consider the following complementary Lagrangian subspaces $V_{+}:=(\mathcal{A}^{p}\oplus 0)\oplus(\mathcal{A}^{q}\oplus 0),\quad\quad V_{-}:=(0\oplus\mathcal{A}^{p})\oplus(0\oplus\mathcal{A}^{q}).$ (6.3) ###### Proposition 6.3 Consider an even operator $S:V_{+}\to V_{-}$, $S=\begin{pmatrix}A&B\\\ C&D\end{pmatrix}.$ The graph of $S$ is a Lagrangian subspace iff $A=A^{t},\quad D=-D^{t},\quad C+B^{t}=0.$ (6.4) Remark. This statement is a super imitation of Lemma 1.11. Proof. We write out a vector $h\in\Lambda^{2p|2q}$ as $h=(u_{+},u_{-};v_{+},v_{-})\in\mathcal{A}^{p}\oplus\mathcal{A}^{p}\oplus\mathcal{A}^{q}\oplus\mathcal{A}^{q}.$ Then $\mathfrak{s}(h^{\prime},h)=u_{+}^{\prime}(u_{-})^{st}-u_{-}^{\prime}(u_{+})^{st}+v_{+}^{\prime}(v_{-})^{st}+v_{-}^{\prime}(v_{+})^{st}.$ Let $h$ be in the graph of $S$. Then $\begin{pmatrix}u_{-}&v_{-}\end{pmatrix}=\begin{pmatrix}u_{+}&v_{+}\end{pmatrix}\begin{pmatrix}A&B\\\ C&D\end{pmatrix}=\begin{pmatrix}u_{+}A+v_{+}C&u_{+}B+v_{+}D\end{pmatrix}$ and $\mathfrak{s}(h^{\prime},h)=u_{+}^{\prime}(u_{+}A+v_{+}C)^{st}-(u_{+}^{\prime}A+v_{+}^{\prime}C)u_{+}^{st}+\\\ +v_{+}^{\prime}(u_{+}B+v_{+}D)^{st}+(u^{\prime}_{+}B+v^{\prime}_{+}D)v_{+}^{st}.$ We emphasis that the matrices $A$, $B$, $C$, $D$ are even212121Recall that this means that $A$, $D$ are composed of even phantom constants and $C$, $B$ of odd phantom constants.; for this reason, we write $(u_{+}A)^{st}=A^{st}(u_{+})^{st}$ etc., see (5.1). We come to $u_{+}^{\prime}\bigl{[}A^{st}(u_{+})^{st}+C^{st}(v_{+})^{st}\bigr{]}-\big{[}(u_{+}^{\prime}A+v_{+}^{\prime}C\bigr{]}u_{+}^{st}+\\\ +v_{+}^{\prime}\bigl{[}B^{st}(u_{+})^{st}+D^{st}(v_{+})^{st}\bigr{]}+\bigl{[}u^{\prime}_{+}B+v^{\prime}_{+}D\bigr{]}v_{+}^{st}.$ Next, $A^{st}=A^{t},\quad B^{st}=-B^{t},\quad C^{st}=C^{t},\quad D^{st}=D^{t}.$ Therefore we convert our expression to the form $u_{+}^{\prime}(A-A^{t})(u_{+})^{st}+v_{+}^{\prime}(D+D^{t})(v_{+})^{st}+(u_{+}^{\prime})(B^{t}+C)v_{+}^{st}+(v^{+})(B+C^{t})u_{+}^{st}.$ This expression is zero iff the conditions (6.4) are satisfied. $\square$ 6.6. Atlas on the Lagrangian Grassmannian. Now we imitate the construction of Subsection 1. Denote by $e_{i}$, $e_{i}^{\prime}$, $f_{j}$, $f_{j}^{\prime}$, where $i\leqslant p$, $j\leqslant q$, the natural basis in $\mathcal{A}^{2p|2q}$. Consider subsets $I\subset\\{1,2,\dots,p\\}$, $J\subset\\{1,2,\dots,q\\}$. We define $\displaystyle V_{+}[I,J]=\bigl{(}\oplus_{i\in I}\mathcal{A}e_{i}\bigr{)}\oplus\bigl{(}\oplus_{k\notin I}\mathcal{A}e_{k}^{\prime}\bigr{)}\oplus\bigl{(}\oplus_{j\in J}\mathcal{A}f_{j}\bigr{)}\oplus\bigl{(}\oplus_{l\notin J}\mathcal{A}f_{l}^{\prime}\bigr{)},$ (6.5) $\displaystyle V_{-}[I,J]=\bigl{(}\oplus_{i\notin I}\mathcal{A}e_{i}\bigr{)}\oplus\bigl{(}\oplus_{k\in I}\mathcal{A}e_{k}^{\prime}\bigr{)}\oplus\bigl{(}\oplus_{j\notin J}\mathcal{A}f_{j}\bigr{)}\oplus\bigl{(}\oplus_{l\in J}\mathcal{A}f_{l}^{\prime}\bigr{)}.$ (6.6) We denote by $\mathcal{O}[I,J]$ the set of all the Lagrangian subspaces that are graphs of even operators $S:V_{+}[I,J]\to V_{-}[I,J].$ In fact, these operators satisfy the same conditions as in Proposition 6.3 (our initial map is $\mathcal{O}[\varnothing,\varnothing]$). Thus we get an atlas on the Lagrangian super-Grassmannian. 6.7. Elementary reflections. Now we repeat considerations of Subsection 1. We define elementary reflections $\sigma[e_{i}]$, $\sigma[f_{j}]$ in $\mathcal{A}^{2p|2q}$ by $\displaystyle\sigma[e_{i}]e_{i}^{+}=-e_{i}^{-},\quad\sigma[e_{i}]e_{i}^{-}=e_{i}^{+},\quad$ $\displaystyle\sigma[e_{i}]e_{k}^{\pm}=e_{k}^{\pm}\,\,\,\,\text{for $k\neq i$},\quad\sigma[e_{i}]f_{j}^{\pm}=f_{j}^{\pm},$ and $\displaystyle\sigma[f_{j}]f_{j}^{+}=f_{j}^{-},\quad\sigma[f_{j}]f_{j}^{-}=f_{j}^{+},\quad$ (6.7) $\displaystyle\sigma[f_{i}]f_{k}^{\pm}=e_{k}^{\pm}\,\,\,\,\text{for $k\neq j$},\quad\sigma[f_{j}]e_{i}^{\pm}=e_{i}^{\pm},$ (6.8) in the first row we have an extra change of a sign because we want to preserve the symplectic form. Then $\mathcal{O}[I,J]=\prod_{i\in I}\sigma[e_{i}]\cdot\prod_{j\in J}\sigma[f_{j}]\cdot\mathcal{O}[\varnothing,\varnothing].$ ## 7 Superlinear relations Gauss–Berezin integral operators are enumerated by contractive Lagrangian superlinear relations. These objects are defined in this section. 7.1. Superlinear relations. We define superlinear relations $P:\mathcal{A}^{p|q}\rightrightarrows\mathcal{A}^{r|s}$ as subspaces in $\mathcal{A}^{p|q}\oplus\mathcal{A}^{r|s}$. Products are defined as above, see Subsection 1. Next, for a superlinear relation we define complex linear relations $\pi_{\downarrow}^{+}(P):{\mathbb{C}}^{p}\rightrightarrows{\mathbb{C}}^{r},\quad\pi_{\downarrow}^{-}(P):{\mathbb{C}}^{q}\rightrightarrows{\mathbb{C}}^{s}$ in the natural way, we simply project the Grassmannian in $\mathcal{A}^{p|q}\oplus\mathcal{A}^{r|s}$ onto the product of the complex Grassmannians. 7.2. Transversality conditions. Let $V$, $W$, $Y$ be complex linear spaces. We say that linear relations $P:V\rightrightarrows W,\,\,Q:W\rightrightarrows Y$ are transversal if $\displaystyle\mathop{\rm im}\nolimits P+\mathop{\rm dom}\nolimits Q=W,$ (7.1) $\displaystyle\mathop{\rm indef}\nolimits P\cap\ker Q=0.$ (7.2) We met these conditions in Section 1, in what follows they are even more important. ###### Theorem 7.1 If $P:V\rightrightarrows W$, $Q:W\rightrightarrows Y$ are transversal, then $\dim QP=\dim Q+\dim P-\dim W.$ Proof. We rephrase the definition of the product $QP$ as follows. Consider the space $V\oplus W\oplus W\oplus Y$ and the following subspaces — $P\oplus Q$, — the subspace $H$ consisting of vectors $v\oplus w\oplus w\oplus y$, — the subspace $T\subset H$ consisting of vectors $0\oplus w\oplus w\oplus 0$. Let us project $(P\oplus Q)\cap H$ on $V\oplus W$ along $T$. The result is $QP\subset V\oplus W$. By the first transversality condition (7.1), $(P\oplus Q)+H=V\oplus W\oplus W\oplus Y,$ therefore we know the dimension of the intersection $S:=(P\oplus Q)\cap H$. By the second condition (7.2) the projection $H\to V\oplus W$ is injective on $S$. $\square$ 7.3. Transversality for superlinear relations. We say that superlinear relations $P:V\rightrightarrows W$ and $Q:W\rightrightarrows Y$ are transversal if $\pi_{\downarrow}^{+}(P)$ is transversal to $\pi_{\downarrow}^{+}(Q)$ and $\pi_{\downarrow}^{-}(P)$ is transversal to $\pi_{\downarrow}^{-}(Q)$. ###### Theorem 7.2 If $P:V\rightrightarrows W$, $Q:W\rightrightarrows Y$ are transversal superlinear relations, then their product is a superlinear relation and $\dim QP=\dim Q+\dim P-\dim W.$ Proof. We watch the proof of the previous theorem. $\square$ 7.4. Lagrangian superlinear relations. Consider the spaces $V=\mathcal{A}^{2p|2q}$, $W=\mathcal{A}^{2r|2s}$ endowed with the orthosymplectic forms $\mathfrak{s}_{V}$, $\mathfrak{s}_{W}$ respectively. Define the form $\mathfrak{s}^{\ominus}$ on $V\oplus W$ as $\mathfrak{s}^{\ominus}(v\oplus w,v^{\prime}\oplus w^{\prime}):=\mathfrak{s}_{V}(v,v^{\prime})-\mathfrak{s}_{W}(w,w^{\prime}).$ A Lagrangian superlinear relation $P:V\rightrightarrows W$ is a Lagrangian supersubspace in $V\oplus W$. ###### Observation 7.3 Let $g\in\mathrm{OSp}(2p|2q)$. Then the graph of $g$ is a Lagrangian superlinear relation $\mathcal{A}^{2p|2q}\rightrightarrows\mathcal{A}^{2p|2q}$. ###### Theorem 7.4 Let $P:V\rightrightarrows W$, $Q:W\rightrightarrows Y$ be transversal Lagrangian superlinear relations. Then $QP:V\rightrightarrows Y$ is a Lagrangian superlinear relation. Proof. Let $v\oplus w$, $v^{\prime}\oplus w^{\prime}\in P$ and $w\oplus y$, $w^{\prime}\oplus y^{\prime}\in Q$. By definition, $\mathfrak{s}_{V}(v,v^{\prime})=\mathfrak{s}_{W}(w,w^{\prime})=\mathfrak{s}_{Y}(y,y^{\prime}),$ therefore $QP$ is isotropic. By the virtue of Theorem 7.2, we know $\dim QP$. $\square$ 7.5. Components of Lagrangian super-Grassmannian. As we observed in Subsection 1, the orthogonal Lagrangian Grassmannian in the space ${\mathbb{C}}^{2n}$ consists of two components. The usual symplectic Lagrangian Grassmannian is connected. Therefore, the Lagrangian super-Grassmannian consists of two components. Below we must distinguish them. Decompose $V=V_{+}\oplus V_{-}$, $W=W_{+}\oplus W_{-}$ as above (6.3). We say that the component containing the linear relation $(V_{+}\oplus W_{-}):\,V\rightrightarrows W$ is even; the another component is odd. 7.6. Contractive Lagrangian linear relations. Now we again (see Section 2) consider the Hermitian form $M$ on ${\mathbb{C}}^{2p}$, it is defined by a matrix $\begin{pmatrix}1&0\\\ 0&-1\end{pmatrix}$. Then ${\mathbb{C}}^{2p}$ becomes an object of the category $\mathbf{Sp}$. We say that a Lagrangian super-linear relation $P:V\rightrightarrows W$ is contractive iff $\pi_{\downarrow}^{+}(P)$ is a morphism of the category $\mathbf{Sp}$. 7.7. Positive domain in the Lagrangian Grassmannian. We say that a Lagrangian subspace $P$ in ${\mathbb{C}}^{2p|2q}$ is positive if the form $M$ is positive on $\pi_{\downarrow}^{+}(P)$. ## 8 Correspondence between Lagrangian superlinear relations and Gauss–Berezin operators Here we prove our results, namely Theorems 8.3, 8.4. 8.1. Creation–annihilation operators. Let $V:=\mathcal{A}^{2p|2q}$ be a superlinear space endowed with the standard orthosymplectic bilinear form $\mathfrak{s}$. For a vector $v\oplus w:=v_{+}\oplus v_{-}\oplus w_{+}\oplus w_{-}\in\mathcal{A}^{2p|2q},$ we define the creation-annihilation operator in the Fock–Berezin space $\mathbf{F}_{p,q}$ by $\widehat{a}(v\oplus w)f(z,\xi)=\Bigl{(}\sum_{i}v_{+}^{(i)}\frac{\partial}{\partial z_{i}}+\sum_{i}v_{-}^{(i)}z_{i}+\sum_{j}w_{+}^{(j)}\frac{\partial}{\partial\xi_{i}}+\sum_{j}w_{-}^{(j)}\xi_{j}\Bigr{)}\,f(z,\xi).$ 8.2. Supercommutator. We say that a vector $v\oplus w$ is even if $v$ is even and $w$ is odd. It is odd if $v$ is odd and $w$ is even. This corresponds to the definition of even/odd for $(1|0)\times(2p|2q)$ matrices. Let $h=v\oplus w$, $h^{\prime}=v^{\prime}\oplus w^{\prime}$. We define the supercommutator $[\widehat{a}(h),\widehat{a}(h^{\prime})]_{\mathrm{super}}$ as $[\widehat{a}(h),\widehat{a}(h^{\prime})]_{\mathrm{super}}=\begin{cases}[\widehat{a}(h),\widehat{a}(h^{\prime})]=\widehat{a}(h)\widehat{a}(h^{\prime})-\widehat{a}(h^{\prime})\widehat{a}(h),\qquad\text{if $h$ or $h^{\prime}$ is even,}\\\ \\{\widehat{a}(h),\widehat{a}(h^{\prime})\\}=\widehat{a}(h)\widehat{a}(h^{\prime})+\widehat{a}(h^{\prime})\widehat{a}(h)\qquad\text{if $h$, $h^{\prime}$ are odd}.\end{cases}$ Then $[\widehat{a}(h),\widehat{a}(h^{\prime})]_{\mathrm{super}}=\mathfrak{s}(h,h^{\prime})\cdot 1,$ where $1$ denotes the unit operator. Also, note that an operator $\widehat{a}(h)$ is linear (see Subsection 4) if $h$ is even and antilinear if $h$ is odd. 8.3. Annihilators of Gaussian vectors. ###### Theorem 8.1 a) For a Gauss–Berezin vector $\mathbf{b}\in\mathbf{F}_{p,q}$ consider the set $L$ of all vectors $h\in\mathcal{A}^{2p|2q}$ such that $\widehat{a}(h)\mathbf{b}=0.$ Then $L$ is a positive Lagrangian subspace in $\mathcal{A}^{2p|2q}$. b) Moreover the map $\mathbf{b}\mapsto L$ is a bijection $\left\\{\begin{matrix}\text{The set of all Gauss--Berezin vectors}\\\ \text{defined up to an invertible scalar}\end{matrix}\right\\}\leftrightarrow\left\\{\begin{matrix}\text{The positive}\\\ \text{ Lagrangian Grassmannian}\end{matrix}\right\\}.$ Before a start of a formal proof we propose the following (nonsufficient) argument. Let $h$, $h^{\prime}\in L$. If one of them is even, then we write $\bigr{(}\widehat{a}(h)\widehat{a}(h^{\prime})-\widehat{a}(h^{\prime})\widehat{a}(h)\Bigl{)}\mathbf{b}.$ By the definition of $L$, this is 0. On the other hand, this is $\mathfrak{s}(h,h^{\prime})\mathbf{b}$. Therefore $\mathfrak{s}(h,h^{\prime})=0$. If both $h$, $h^{\prime}$ are odd, then we write $0=\bigr{(}\widehat{a}(h)\widehat{a}(h^{\prime})+\widehat{a}(h^{\prime})\widehat{a}(h)\Bigl{)}\mathbf{b}=\mathfrak{s}(h,h^{\prime})\mathfrak{b}$ and come to the same result. Proof. First, let $\mathfrak{b}(z,\xi)$ have the standard form (4.5). We write out $\widehat{a}(v\oplus w)\mathfrak{b}(z,\xi)=\Bigl{(}\sum_{i}v_{+}^{(i)}\frac{\partial}{\partial z_{i}}+\sum_{i}v_{-}^{(i)}z_{i}+\sum_{j}w_{+}^{(j)}\frac{\partial}{\partial\xi_{i}}+\sum_{j}w_{-}^{(j)}\xi_{j}\Bigr{)}\times\\\ \times\exp\Bigl{\\{}\frac{1}{2}\begin{pmatrix}z&\xi\end{pmatrix}\begin{pmatrix}A&B\\\ -B^{t}&C\end{pmatrix}\begin{pmatrix}z^{t}\\\ \xi^{t}\end{pmatrix}\Bigr{\\}}=\\\ =\Bigl{(}v_{+}(Ax^{t}+B\xi^{t})+v_{-}z^{t}+w_{+}(-B^{t}z^{t}+C\xi^{t})+w_{-}\xi^{t}\Bigr{)}\cdot\mathfrak{b}(z,\xi)=\\\ =\Bigl{(}(v_{+}A-w_{+}B^{t}+v_{-})z^{t}+(v_{+}B+w_{+}D+w_{-})\xi^{t}\Bigr{)}\cdot\mathfrak{b}(z,\xi).$ This is zero iff $\begin{cases}v_{-}=-(v_{+}A-w_{+}B^{t})\\\ w_{-}=-(v_{+}B+w_{+}D)\end{cases}.$ However, this system of equation determines a Lagrangian subspace. The positivity of a Lagrangian subspace is equivalent to $\|\pi_{\downarrow}(A)\|<1$ (see, for instance [22]). Next, consider an arbitrary Gauss–Berezin vector $\mathbf{b}(z,\xi)=\mathfrak{D}[\xi_{i_{1}}]\dots\mathfrak{D}[\xi_{i_{k}}]\,\mathsf{S}^{k}\,\,\mathfrak{b}[T],$ (8.1) where $\mathfrak{b}[T]$ is a standard Gauss–Berezin vector. We have $\widehat{a}\bigl{(}h)\,\mathfrak{D}[\xi_{1}]\,\mathsf{S}=\mathfrak{D}[\xi_{1}]\,\mathsf{S}\,\widehat{a}\bigl{(}\sigma[f_{1}]h\bigr{)},$ (8.2) where $\sigma[f_{1}]$ is the elementary reflection given by (6.7)–(6.8). If $h$ ranges in a Lagrangian subspace, then $\sigma[f_{1}]h$ also ranges in (another) Lagrangian subspace. Also, a map $\sigma[f_{1}]$ takes positive subspaces to positive subspaces. Therefore the statements a) for vectors $\mathfrak{D}[\xi_{i_{1}}]\mathfrak{D}[\xi_{i_{2}}]\dots\mathfrak{D}[\xi_{i_{k}}]\,\mathsf{S}^{k}\,\,\mathfrak{b}[T]\qquad\text{and}\qquad\mathfrak{D}[\xi_{i_{2}}]\dots\mathfrak{D}[\xi_{i_{k}}]\,\mathsf{S}^{k-1}\,\,\mathfrak{b}[T].$ are equivalent. In fact, for fixed $i_{1}$, …, $i_{k}$ all vectors of the form (8.1) correspond to a fixed map in the Lagrangian super-Grassmannian, namely to $\sigma[f_{i_{1}}]\cdots\sigma[f_{i_{k}}]\cdot\mathcal{O}[\varnothing,\varnothing]$ in notation of Subsection 6. But these maps cover the set of all positive Lagrangian subspaces. $\square$ ###### Theorem 8.2 For a positive Lagrangian subspace $L\subset\mathcal{A}^{2p|2q}$ consider the system of equations $\widehat{a}(v\oplus w)f(z,\xi)=0\quad\text{for all $v\oplus w\in L$},$ (8.3) in a function $f(z,\xi)$. All its solutions are of the form $\lambda\mathfrak{b}(z,\xi)$, where $\mathfrak{b}(z,\xi)$ is a Gauss–Berezin vector and $\lambda$ is a phantom constant. Proof. It suffices to prove the statement for $L$ being in the principal map. Put $\varphi(z,\xi):=f(z,\xi)/\mathfrak{b}(z,\xi),$ i.e., $f(z,\xi)=\mathfrak{b}(z,\xi)\cdot\varphi(z,\xi).$ By the Leibnitz rule, $0=\widehat{a}(v\oplus w)\bigl{(}\mathfrak{b}(z,\xi)\varphi(z,\xi)\bigr{)}=\\\ =\Bigl{(}\widehat{a}(v\oplus w)\mathfrak{b}(z,\xi)\Bigr{)}\cdot\varphi(x,\xi)+\mathfrak{b}(z,\xi)\cdot\Bigl{(}\sum_{j}v_{+}^{(j)}\frac{\partial}{\partial z_{i}}+\sum_{j}w_{+}^{(j)}\frac{\partial}{\partial\xi_{i}}\Bigr{)}.\varphi(z,\xi)$ The first summand is zero by the definition of $\mathfrak{b}(z,\xi)$. Since $v_{+}$, $w_{+}$ are arbitrary, we get $\frac{\partial}{\partial z_{i}}\varphi(z,\xi)=0,\quad\frac{\partial}{\partial\xi_{i}}\varphi(z,\xi)=0.$ Therefore, $\varphi(z,\xi)$ is a phantom constant. $\square$ 8.4. Gauss–Berezin operators and linear relations. Let $V=\mathcal{A}^{2p|2q}$, $\widetilde{V}=\mathcal{A}^{2r|2s}$ be spaces endowed with orthosymplectic forms. ###### Theorem 8.3 a) For each contractive Lagrangian linear relation $P:V\rightrightarrows\widetilde{V}$ there exists a linear operator $\mathfrak{B}(P):\mathbf{F}_{p,q}\to\mathbf{F}_{r,s}$ such that 1) The following condition is satisfied $\widehat{a}(\widetilde{h})\,\mathfrak{B}(P)=\mathfrak{B}(P)\,\widehat{a}(h)\quad\text{for all $h\oplus\widetilde{h}\in P$}$ 2) If $P$ is in the even component of Lagrangian Grassmannian, then $\mathfrak{B}(P)$ is an integral operator with an even222222with respect to totality of all Grassmann variables, kernel. If $P$ is in the odd component, then $\mathfrak{B}(P)\mathsf{S}$ is an integral operator with an odd kernel. Moreover, this operator is unique up to a scalar factor $\in\mathcal{A}_{even}$. b) The operators $\mathfrak{B}(P)$ are Gauss–Berezin operators and all Gauss–Berezin operators arise in this context. Proof. We must write out the differential equations for the kernel $K(z,\xi,\overline{y},\overline{\eta})$ of the operator $\mathfrak{B}(P)$. Denote $h=v_{+}\oplus v_{-}\oplus w_{+}\oplus w_{-},\quad\widetilde{h}=\widetilde{v}_{+}\oplus\widetilde{v}_{-}\oplus\widetilde{w}_{+}\oplus\widetilde{w}_{-}.$ Then $\widehat{a}(\widetilde{h})\int K(z,\xi,\overline{y},\overline{\eta})\,f(y,\eta)\,e^{-y\overline{y}^{t}-\eta\overline{\eta}^{t}}dy\,d\overline{y}\,d\overline{\eta}\,d\eta=\\\ =\int K(z,\xi,\overline{y},\overline{\eta})\,\widehat{a}(h)f(y,\eta)\,e^{-y\overline{y}^{t}-\eta\overline{\eta}^{t}}dy\,d\overline{y}\,d\overline{\eta}d\eta.$ Let $P$ be even. Integrating by parts in the right-hand side, we get: $\Bigl{(}\sum_{i}\widetilde{v}_{+}^{(i)}\frac{\partial}{\partial z_{i}}+\sum_{i}\widetilde{v}_{-}^{(i)}z_{i}+\sum_{j}\widetilde{w}_{+}^{(j)}\frac{\partial}{\partial\xi_{i}}+\sum_{j}\widetilde{w}_{-}^{(j)}\xi_{j}\Bigr{)}K(z,\xi,\overline{y},\overline{\eta})=\\\ =\Bigl{(}\sum_{i}v_{+}^{(i)}\overline{y}_{i}+\sum_{i}v_{-}^{(i)}\frac{\partial}{\partial\overline{y}_{i}}+\sum_{j}w_{+}^{(j)}\overline{\eta}_{j}+\sum_{j}w_{-}^{(j)}\frac{\partial}{\partial\overline{\eta}_{i}}\Bigr{)}K(z,\xi,\overline{y},\overline{\eta}).$ This system of equations has the form (8.3) and determines a Gaussian. Eveness condition was essentially used in this calculation. For instance for an odd kernel $K$ we must write $(v_{+}^{(i)})^{\sigma}$ instead of $v_{+}^{(i)}$ in the right hand side. Now let $P$ be odd. Let us try to find $\mathfrak{B}(P)$ as a product $\mathfrak{B}(P)=\mathfrak{C}\cdot\mathfrak{D}[\eta_{1}]\cdot\mathsf{S}.$ Let $L$ be the kernel of $\mathfrak{C}$. $\widehat{a}(\widetilde{h})\int L(z,\xi,\overline{y},\overline{\eta})\,\mathfrak{D}[\eta_{1}]\cdot\mathsf{S}\cdot f(y,\eta)\,e^{-y\overline{y}^{t}-\eta\overline{\eta}^{t}}dy\,d\overline{y}\,d\overline{\eta}\,d\eta=\\\ =\int L(z,\xi,\overline{y},\overline{\eta})\,\mathfrak{D}[\eta_{1}]\cdot\mathsf{S}\cdot\widehat{a}(h)f(y,\eta)\,e^{-y\overline{y}^{t}-\eta\overline{\eta}^{t}}dy\,d\overline{y}\,d\overline{\eta}d\eta.$ Next, we change the order $\mathfrak{D}[\eta_{1}]\mathsf{S}\widehat{a}(v)=\widehat{a}(\sigma(f_{1})v)\mathfrak{D}[\eta_{1}]\mathsf{S},$ where $\sigma$ is an elementary reflection of the type (6.7)–(6.8). We again get for $L$ a system of equations determining a Gaussian. $\square$ 8.5. Products of Gauss–Berezin operators. ###### Theorem 8.4 a) Let $P:\mathcal{A}^{p|q}\rightrightarrows\mathcal{A}^{p^{\prime}|q^{\prime}}$, $Q:\mathcal{A}^{p^{\prime}|q^{\prime}}\rightrightarrows\mathcal{A}^{p^{\prime\prime}|q^{\prime\prime}}$ be strictly contractive Lagrangian relations. Assume that $P$, $Q$ are transversal. Then $\mathfrak{B}(Q)\mathfrak{B}(P)=\lambda\cdot\mathfrak{B}(QP),$ (8.4) where $\lambda=\lambda(P,Q)$ is an even invertible phantom constant. b) If $P$, $Q$ are not transversal, then $\pi_{\downarrow}\bigl{(}\mathfrak{B}(Q)\mathfrak{B}(P)\bigr{)}=0.$ Proof. Let $v\oplus w\in P$, $w\oplus y\in Q$. Then $\mathfrak{B}(Q)\mathfrak{B}(P)\widehat{a}(v)=\mathfrak{B}(Q)\widehat{a}(w)\mathfrak{B}(P)=\widehat{a}(y)\mathfrak{B}(Q)\mathfrak{B}(P).$ On the other hand, $\widehat{a}(y)\mathfrak{B}(QP)=\mathfrak{B}(QP)\widehat{a}(v).$ By Theorem 8.3, these relations define a unique operator and we get (8.4). It remains to watch conditions of vanishing of $\pi_{\downarrow}\bigl{(}\mathfrak{B}(Q)\,\mathfrak{B}(P)\bigr{)}:\mathbf{F}_{p}\otimes\Lambda_{q}\to\mathbf{F}_{p^{\prime\prime}}\otimes\Lambda_{q^{\prime\prime}}.$ Here we refer to Subsection 4. Our operator is a tensor product of $\pi_{\downarrow}^{+}\bigl{(}\mathfrak{B}(Q)\bigr{)}\,\pi_{\downarrow}^{+}\bigl{(}\mathfrak{B}(P)\bigr{)}:\mathbf{F}_{p}\to\mathbf{F}_{p^{\prime\prime}}$ (8.5) and $\pi_{\downarrow}^{-}\bigl{(}\mathfrak{B}(Q)\bigr{)}\,\pi_{\downarrow}^{-}\bigl{(}\mathfrak{B}(P)\bigr{)}:\Lambda_{q}\to\Lambda_{q^{\prime\prime}}.$ (8.6) In the line (8.5) we have a product of Gaussian integral operators. By Theorem 2.5, it is nonzero. In the line (8.6) we observe a product of Berezin operators. It is nonzero iff $\pi_{\downarrow}^{-}(P)$ and $\pi_{\downarrow}^{-}(Q)$ are transversal, here we refer to Theorem 1.16.c. $\square$ ###### Corollary 8.5 We get a projective representation of the Olshanski super semigroup $\Gamma\mathrm{OSp}(2p|2q)$. Denote by $G(2p|2q)$ the group of invertible elements of Olshanski super semigroup. It is easier to say that $G(2p|2q)$ consists of all $g\in\mathrm{OSp}(2p|2q)$ such that $\pi_{\downarrow}(g)\in\mathrm{Sp}(2n,{\mathbb{R}})$232323This is not an ’official’ real form of $\mathrm{OSp}(2p|2q;{\mathbb{C}})$.. ###### Corollary 8.6 We get a representation of the supergroup $G(2p|2q)$. ## 9 Final remarks 9.1. Extension of Gaussian operators? Our main result is not complete, because Theorem 8.4 describes product of integral operators only if linear relations are transversal. However, a product of integral operators can be written explicitly in all cases (see Subsection 3). We get the following question — Is it possible to extend a definition of Gaussian operators and Lagrangian linear relations to get the formula (8.4) valid for all $P$, $Q$. We can also ask this question in another form. — Is it reasonable to consider the expressions (3.8) as Gaussians? 9.2. Howe duality. In a certain sense, the orthogonal and symplectic spinors are universal objects in the representation theory. In particular, one can produce numerous representations from spinors by the Howe duality. However, the Howe duality exists for supergroups, see [25], [8], [9], [16]. Therefore, there arise two questions. — Which infinite-dimensional representation of classical Lie superalgebras can be integrated to supergroups using our construction? — We extend the spinor representation to the Grassmannian. For which representations of supergroups such extensions are possible? ## References * [1] Adams J. D., Discrete spectrum of reductive dual pair $(O(p,q)$, $\text{\rm Sp}\,(2m))$, Inv. Math., 74, 449–475 (1983). * [2] Arnold, V. I. Mathematical methods of classical mechanics. Springer-Verlag, New York-Heidelberg, 1978 * [3] Berezin, F. A. The method of second quantization. Academic Press, New York-London, 1966 * [4] Berezin, F. A. Introduction to algebra and analysis with anticommuting variables. Moskov. Gos. Univ., Moscow, 1983; English transl.: Introduction to superanalysis Edited and with a foreword by A. A. Kirillov. With an appendix by V. I. Ogievetsky. Translated from the Russian by J. Niederle and R. Kotecký. Translation edited by Dimitri Leĭtes. D. Reidel Publishing Co., Dordrecht, 1987 * [5] Berezin, F. A.; Kac, G. I. Lie groups with commuting and anticommuting parameters. (Russian) Mat. Sb. (N.S.) 82 (124) 1970 343–359 English transl. in Math. USSR-Sb. 11, 311–325 (1971) * [6] Berezin, F. A.; Tolstoy, V. N. The group with Grassmann structure ${\rm UOSP}(1.2)$. Comm. Math. Phys. 78 (1980/81), no. 3, 409–428. * [7] Bernstein, I., Leites, D., Molotkov, Shander, V, Seminar on supersymmetries. MCCME publishers, to appear * [8] Cheng, Shun-Jen; Zhang, R. B. Howe duality and combinatorial character formula for orthosymplectic Lie superalgebras. Adv. Math. 182 (2004), no. 1, 124–172. * [9] Cheng, Shun-Jen; Wang, Weiqiang. Howe duality for Lie superalgebras. Compositio Math. 128 (2001), no. 1, 55–94. * [10] Caston, L., Fioresi R. Mathematical Foundations of Supersymmetry Available via http://xxx.arxiv.org/abs/0710.5742 * [11] Friedrichs, K. O. Mathematical aspects of the quantum theory of fields. Interscience Publishers, Inc., London, 1953. * [12] Howe, R. Transcending classical invariant theory. J. Amer. Math. Soc. 2 (1989), no. 3, 535–552. * [13] Kashiwara, M.; Vergne, M. On the Segal-Shale-Weil representations and harmonic polynomials. Invent. Math. 44 (1978), no. 1, 1–47 * [14] Lavaud, P. Superpfaffian. J. Lie Theory 16 (2006), no. 2, 271–296. * [15] Leites, D. A. Introduction to the theory of supermanifolds. Uspekhi Mat. Nauk 35 (1980), no. 1(211), 3–57, 255. * [16] Leites, D.; Shchepochkina, I. The Howe duality and Lie superalgebras. Noncommutative structures in mathematics and physics (Kiev, 2000), 93–111, NATO Sci. Ser. II Math. Phys. Chem., 22, Kluwer Acad. Publ., Dordrecht, 2001 * [17] Manin, Yu. I. Gauge field theory and complex geometry. Springer-Verlag, Berlin, 1988. * [18] Nazarov, M.; Neretin, Yu.; Olshanskii, G. Semi-groupes engendrés par la représentation de Weil du groupe symplectique de dimension infinie. C. R. Acad. Sci. Paris S?r. I Math. 309 (1989), no. 7, 443–446 * [19] Neretin, Yu. A. On a semigroup of operators in the boson Fock space. Funktsional. Anal. i Prilozhen. 24 (1990), no. 2, 63–73, 96; translation in Funct. Anal. Appl. 24 (1990), no. 2, 135–144 * [20] Neretin, Yu. A. Spinor representation of an infinite-dimensional orthogonal semigroup and the Virasoro algebra. Funktsional. Anal. i Prilozhen. 23 (1989), no. 3, 32–44, 96; translation in Funct. Anal. Appl. 23 (1989), no. 3, 196–207 (1990) * [21] Neretin, Yu. A. Categories of symmetries and infinite-dimensional groups. The Clarendon Press, Oxford University Press, New York, 1996. * [22] Neretin, Yu.A. Lectures on Gaussian integral operators and classical groups, available via www.mat.univie.ac.at/$\sim$neretin/lectures.htm * [23] Neretin, Yu.A. Gauss–Berezin integral operators and spinors over supergroups $\mathrm{OSp}(2p|2q)$., Preprint ESI–1930, 2007. * [24] Neretin, Yu. A. ”Method of second quantization” of Berezin. View 40 years after. In D. Leites, R. A. Minlos, I. Tyutin (eds.) ”Recollections about Felix Alexandrovich Berezin, the discoverer of supersymmetries”, Moscow, MCCME publishers, 2008; French translation of the book is in progress. * [25] Nishiyama K. Super dual pairs and highest weight modules of orthosymplectic algebras. Adv.Math., 104 (1994), 66-89 * [26] Sato, M., Miwa, T., Jimbo, M. Studies on holonomic quantum fields. I. Proc. Japan Acad. Ser. A Math. Sci. 53 (1977), no. 1, 6–10 * [27] Varadarajan, V. S. Supersymmetry for mathematicians: an introduction. American Mathematical Society, Providence, RI, 2004. * [28] DeWitt, B. Supermanifolds. Cambridge University Press, 1992. Math.Dept., University of Vienna, Nordbergstrasse, 15, Vienna, Austria & Institute for Theoretical and Experimental Physics, Bolshaya Cheremushkinskaya, 25, Moscow 117259, Russia & Moscow State University, MechMath, Vorob’ovy Gory, Moscow, Russia e-mail: neretin(at) mccme.ru URL:www.mat.univie.ac.at/$\sim$neretin :wwwth.itep.ru/$\sim$neretin

arxiv-papers

2024-09-04T02:48:52.985373

0707.2312

# Integer partitions and exclusion statistics: Limit shapes and the largest part of Young diagrams Alain Comtet1,2, Satya N. Majumdar1, Stéphane Ouvry1 and Sanjib Sabhapandit1 1Laboratoire de Physique Théorique et Modèles Statistiques, Université de Paris-Sud, CNRS UMR 8626, 91405 Orsay Cedex, France 2 Institut Henri Poincaré, 11 rue Pierre et Marie Curie, 75005 Paris, France ###### Abstract We compute the limit shapes of the Young diagrams of the minimal difference $p$ partitions and provide a simple physical interpretation for the limit shapes. We also calculate the asymptotic distribution of the largest part of the Young diagram and show that the scaled distribution has a Gumbel form for all $p$. This Gumbel statistics for the largest part remains unchanged even for general partitions of the form $E=\sum_{i}n_{i}i^{1/\nu}$ with $\nu>0$ where $n_{i}$ is the number of times the part $i$ appears. Journal-ref: J. Stat. Mech. (2007) P10001 ###### Contents 1. 1 Introduction 2. 2 Problems and outline 3. 3 Restricted grand partition function of MDP–$p$ problem 4. 4 Limit shapes of Young diagrams 5. 5 Largest part of Young diagrams 6. 6 Summary and remarks 7. Note added in proof 8. References ## 1 Introduction Exclusion statistics [1, 2, 3, 4, 5, 6, 7, 8]—a generalization of Bose and Fermi statistics—can be defined in the following thermodynamical sense. Let $Z(\beta,z)$ denote the grand partition function of a quantum gas of particles at inverse temperature $\beta$ and fugacity $z$. Such a gas is said to obey exclusion statistics with parameter $0\leq p\leq 1$, if $Z(\beta,z)$ can be expressed as an integral representation $\ln Z(\beta,z)=\int_{0}^{\infty}{\tilde{\rho}}(\epsilon)\ln y_{p}\left(z\rme^{-\beta\epsilon}\right)\,\rmd\epsilon,$ (1) where ${\tilde{\rho}}(\epsilon)$ denotes a single particle density of states and the function $y_{p}(x)$, which encodes fractional statistics, is given by the solution of the equation $y_{p}(x)-x\,y_{p}^{1-p}(x)=1.$ (2) In the cases $p=0$ and $p=1$, substituting $y_{p}(x)$ explicitly in (1) yield the standard grand partition functions of non-interacting bosons and fermions respectively. The fractional exclusion statistics with parameter $0<p<1$ (that corresponds to an interacting gas) smoothly interpolates between these two extreme cases. Two known microscopic quantum mechanical realizations of exclusion statistics are the Lowest Landau Level (LLL) anyon model [2, 3] and the Calogero model [6, 7], with ${\tilde{\rho}}(\epsilon)$ being, respectively, the LLL density of states and the free one-dimensional density of states. It is well known that a gas of non-interacting bosons ($p=0$) or fermions ($p=1$) occupying a single particle equidistant spectrum both have a combinatorial interpretation in terms of the integer partition problem [9]. A partition of a positive integer $E$ is a decomposition of $E$ as a sum of a nonincreasing sequence of positive integers $\\{h_{j}\\}$, i.e., $E=\sum_{j}h_{j}$ such that $h_{j}\geq h_{j+1}$, for $j=1,2\ldots$. For example, $4$ can be partitioned in $5$ ways: $4$, $3+1$, $2+2$, $2+1+1$, and $1+1+1+1$. Partitions can be graphically represented by Young diagrams (also called Ferrers diagrams), where $h_{j}$ corresponds to the height of the $j$-th column (see figure 1). Figure 1: (a) All the Young diagrams for the partitions of 4. (b) The Young diagram of the partition $91=18+16+13+13+9+6+5+5+3+3$, and the corresponding configuration of non-interacting bosons occupying energy levels $\epsilon_{i}=i$ for $i=1,2,\ldots,18$. In the Young diagram of a given partition of $E$, if $n_{i}$ denotes the number of columns having heights equal to $i$, then clearly $E=\sum_{i}n_{i}\epsilon_{i}$ —which can now be interpreted as the total energy of a non-interacting quantum gas of bosons where $\epsilon_{i}=i$ for $i=1,2,\ldots,\infty$ represent equidistant single particle energy levels and $n_{i}=0,1,2,\ldots,\infty$ represents the occupation number of the $i$-th level (see figure 1(b)). On the other hand, if one expresses a positive integer $E$ as a sum of strictly decreasing sequence of positive integers, i.e. $E=\sum_{j}h_{j}$ such that $h_{j}>h_{j+1}$ (e.g. allowed partitions of 4 are: $4$ and $3+1$), then the restricted partition problem corresponds to a non-interacting quantum gas of fermions, for which $n_{i}=0,1$. In the partitioning problems if one restricts the number of summands to be $N$, then clearly $N=\sum_{i}n_{i}$ represents the total number of particles. For example, if $E=4$ and $N=2$, the allowed partitions are $3+1$ and $2+2$ in the unrestricted problem, whereas the only allowed restricted partition is $3+1$. The number $\rho(E,N)$ of ways of partitioning $E$ into $N$ parts is simply the micro-canonical partition function of a gas of quantum particles with total energy $E$ and total number of particles $N$: $\rho(E,N)=\sum_{\\{n_{i}\\}}\delta\left(E-\sum_{i=1}^{\infty}n_{i}\epsilon_{i}\right)\,\delta\left(N-\sum_{i=1}^{\infty}n_{i}\right).$ (3) The grand partition functions, i.e., $Z(\beta,z)=\sum_{N}\sum_{E}z^{N}\rme^{-\beta E}\rho(E,N)$, for the unrestricted and restricted partitions are $Z(\beta,z)=\prod_{i=1}^{\infty}(1-z\rme^{-\beta i})^{-1}$ and $Z(\beta,z)=\prod_{i=1}^{\infty}(1+z\rme^{-\beta i})$ and hence $\ln Z(\beta,z)$ in the limit $\beta\rightarrow 0$ and $\tilde{\rho}(\epsilon)=1$ reduce to (1) with $p=0$ and $p=1$ respectively. Unlike Bose and Fermi statistics which describes non-interacting particles, for a quantum gas obeying exclusion statistics with parameter $0<p<1$, it is a priori not obvious how to provide a combinatorial description, since the underlying physical models with exclusion statistics describe interacting systems. However it has recently been shown [10] that a combinatorial description of exclusion statistics is possible in terms of a generalized partition problem known as the minimal difference $p$ partition (MDP–$p$), which we will define in the next section. Even though the parameter $p$ in MDP–$p$ is an integer, in [10] it has been shown that, when one analytically continues the results to non-integer values of $p$, for $0<p<1$, and in the limit $\beta\to 0$, the MDP–$p$ corresponds to a gas of quantum particles obeying exclusion statistics. This correspondence between exclusion statistics and MDP–$p$ motivates us to investigate some other aspects of the MDP–$p$ problem in this paper. ## 2 Problems and outline Figure 2: A typical Young diagram for MDP–$p$ problem. The thick solid border shows the height profile. $W_{h}$ is the width of the Young diagram at a height $h$, i.e., $W_{h}$ is the number of columns whose heights $\geq h$. In the MDP–$p$ problem, a positive integer $E$ is expressed as a sum of positive integers $E=\sum_{j}h_{j}$ such that $h_{j}-h_{j+1}\geq p$ (see figure 2). Therefore, $p=0$ corresponds to unrestricted partitions and $p=1$ to restricted partitions into distinct parts. The shortest part in the MDP–$p$ problem is usually taken to be $\geq 1$. However, for the calculation of certain specific quantities in this model, it is useful to consider a somewhat generalized version with the shortest part $\geq s$, where $s$ is considered to be a variable. The grand partition function of this problem was obtained recently in [10], which is given by (1) with constant density of states ${\tilde{\rho}}(\epsilon)=1$ and the lower limit of integration being $s$. One may also think of the MDP–$p$ in terms of a quantum system consisting of equidistant energy levels $\epsilon_{i}=i$ for $i=1,2,\ldots,\infty$. Now a given height $h_{j}=i$ corresponds the energy level $\epsilon_{i}=i$ and the number of columns with height $i$ is the occupation number $n_{i}$. Since the difference between two consecutive heights in the MDP–$p$ must be at least $p$, the gap between two adjacent occupied energy levels must be at least $p$. Clearly for $p=0$ this gap is zero, and hence each level can be occupied by any number of particles (bosons). For $p=1$, each level can be occupied by at most one particle (fermions). Again for $p>1$ a level can be occupied by at most one particle. However, in this case, when a energy level is occupied by a particle, the adjacent $p-1$ levels must remain unoccupied. One major issue in the partition problem is to study the limit shape, i.e., the average height profile of an ensemble of Young diagrams with a fixed but large $E$. The shape (height profile) can be defined by the width $W_{h}$ of the Young diagram at a height $h$ (see figure 2). In other words, $W_{h}$ is the number of columns of the Young diagram whose height is greater than or equal to $h$. In this corresponding quantum system, $W_{h}$ represents the total number of particles occupying energy levels above $h$. The height profile of the Young diagram of the unrestricted partition ($p=0$) was first studied by Temperley, who was interested in determining the equilibrium profile of a simple cubic crystal grown from the corner of three walls at right angles. The two dimensional version of the problem —where walls (two) are along the horizontal and the vertical axes and $E$ “bricks” (molecules) are packed into the first quadrant one by one such that each brick, when it is added, makes two contact along faces— corresponds to the $p=0$ partition problem. Temperley [11] computed the equilibrium profile of this two dimensional crystal. More recently the investigation of the limit shape of random partitions has been developed extensively by Vershik [12, 13, 14] and collaborators. The case of uniform random partitions was treated by Vershik who proved for the bosonic ($p=0$) as well as the fermionic ($p=1$) case that the rescaled $h/\sqrt{E}$ vs. $W_{h}/\sqrt{E}$ curves converge to limiting curves when $E\rightarrow\infty$, and obtained these limit shapes explicitly. These results were extended by Romik [15] to the MDP–$p$ for $p=2$. In this paper we compute the following two quantities: 1. (1) The limit shape of the Young diagrams of the MDP–$p$ for any $p$, from which the previously obtained results for $p=0,1,2$ follow as special cases. 2. (2) The distribution of the largest part of the Young diagrams of the MDP–$p$ problem for all $p$, whereas the earlier result existed only for the $p=0$ case [16]. The average height profile $\langle W_{h}\rangle$ of the Young diagrams of the partitions of a given integer $E$ is easier to compute in the grand canonical ensemble. Therefore one requires a restricted grand partition function $Z_{h}(\beta,z)$ which counts the columns whose heights $\geq h$, and the full grand partition function $Z(\beta,z)$ which counts all the columns. From the restricted grand partition function one finds $\langle W_{h}\rangle=z\frac{\partial}{\partial z}\ln Z_{h}(\beta,z)|_{z=1}$. For given large $E$, the parameter $\beta$ is fixed by the relation $E=-\frac{\partial}{\partial\beta}\ln Z(\beta,1)$. On the other hand, to compute the number of partitions $\rho_{p}(E,l)$ of an integer $E$ such that the largest part $\leq l$, it is useful to consider the partition function $Z_{l}(\beta)=\sum_{E}\rme^{-\beta E}\rho_{p}(E,l)$ first. Formally $\rho_{p}(E,l)$ can be obtained by inverting $Z_{l}(\beta)$ with respect to $\beta$, and for large $E$ the asymptotic behavior of $\rho_{p}(E,l)$ is obtained from the saddle point approximation, where the parameter $\beta$ is fixed in terms of given $E$ by the saddle point relation $E=-\frac{\partial}{\partial\beta}\ln Z_{l}(\beta)$. Thus, it is useful to consider a more general restricted grand partition function $Z(\beta,z,l,s)$ that counts the columns whose heights lie between $s$ and $l$. All the other partition functions we need for our calculations can be obtained from $Z(\beta,z,l,s)$ by taking various limits on $s$ and $l$. For example, by putting $s=1$ and taking the limit $l\rightarrow\infty$ one obtains $Z(\beta,z)$. Similarly $s=h$ and the limit $l\rightarrow\infty$ gives $Z_{h}(\beta,z)$ and putting $s=1$ and $z=1$ gives $Z_{l}(\beta)$. As we will see later in (14) and (23) that $\beta\sim E^{-1/2}$ for large $E$. Therefore, hereafter we will work in the limit $\beta\rightarrow 0$. The rest of the paper is organized as follows. We first obtain the generalized grand partition function $Z(\beta,z,l,s)$ of the MDP–$p$ problem in the next section. In section 4 we compute the limit shapes of the Young diagrams and also provide a simple physical interpretation of the result. In section 5 we calculate the distribution of the largest part of the MDP–$p$ . Finally, we conclude with a summary and some remarks in section 6. ## 3 Restricted grand partition function of MDP–$p$ problem Let $\rho_{p}(E,N,l,s)$ be the number of ways of partitioning an integer $E$ into $N$ parts in the MDP–$p$ problem such that the largest part is at most $l$ and the smallest part is at least $s$, i.e., $E=\sum_{j=1}^{N}h_{j}$ such that $h_{1}\leq l$, $h_{j+1}\leq h_{j}-p$ for all $j=1,2,\ldots,N-1$, and $h_{N}\geq s$. Then clearly, $[\rho_{p}(E,N,l,s)-\rho_{p}(E,N,l-1,s)]$ gives the number of MDP–$p$ of $E$, such that the largest part is exactly equal to $l$, and smallest part is at least $s$. Now, by eliminating the first part $h_{1}=l$ from the partition one immediately realizes that the above number is precisely $\rho_{p}(E-l,N-1,l-p,s)$, i.e., the number of MDP–$p$ of $E-l$ into $N-1$ parts such that the largest part is at most $l-p$ and the smallest part is at least $s$. Therefore, one has the recursion relation $\rho_{p}(E,N,l,s)=\rho_{p}(E,N,l-1,s)+\rho_{p}(E-l,N-1,l-p,s).$ (4) Following similar reasoning one can also derive another recursion relation in terms of the smallest part $s$, $\rho_{p}(E,N,l,s)=\rho_{p}(E,N,l,s+1)+\rho_{p}(E-s,N-1,l,s+p).$ (5) It follows from (4) and (5) that the grand partition function $Z(\beta,z,l,s)=\sum_{N}\sum_{E}z^{N}\rme^{-\beta E}\rho_{p}(E,N,l,s)$ satisfies the recursion relations: $\displaystyle Z(\beta,z,l,s)=Z(\beta,z,l-1,s)+z\rme^{-\beta l}Z(\beta,z,l-p,s),$ (6) $\displaystyle Z(\beta,z,l,s)=Z(\beta,z,l,s+1)+z\rme^{-\beta s}Z(\beta,z,l,s+p).$ (7) From these equations, it is evident that in the scaling limit $\beta\rightarrow 0$, and both $s$ and $l$ large, the correct scaling variables are $\beta s$ and $\beta l$, so that $\beta s$ and $\beta l$ remain finite. One knows from the statistical mechanics that the free energy $\beta^{-1}Z(\beta,z,l,s)$ becomes a function of the only the scaling variables in the limit $\beta\rightarrow 0$. Therefore in this limit it is natural to expect $Z(\beta,z,l,s)\approx\exp\left(\frac{1}{\beta}\Phi(\beta l,\beta s,z)\right).$ (8) Now to determine the scaling function $\phi(\beta l,\beta s,z)$, we substitute the ansatz (8) in (6) and (7), and then expand $\Phi(\beta l-\beta,\beta s,z)$ and $\Phi(\beta l-\beta p,\beta s,z)$ about $\beta l$, and $\Phi(\beta l,\beta s+\beta,z)$ and $\Phi(\beta l,\beta s+\beta p,z)$ about $\beta s$, respectively in Taylor series up to first order, which yields the equations: $\displaystyle\qquad\exp(-\Phi_{\beta l})$ $\displaystyle+z\rme^{-\beta l}\exp(-p\Phi_{\beta l})$ $\displaystyle=1,\quad\mbox{where}\quad\Phi_{\beta l}=\frac{\partial}{\partial u}\Phi(u,\beta s,z)\Big{|}_{u=\beta l},$ (9) $\displaystyle\qquad\exp(\Phi_{\beta s})$ $\displaystyle+z\rme^{-\beta s}\exp(p\Phi_{\beta s})$ $\displaystyle=1,\quad\mbox{where}\quad\Phi_{\beta s}=\frac{\partial}{\partial v}\Phi(\beta l,v,z)\Big{|}_{v=\beta s}.$ (10) It is evident from (9) and (10), that $\Phi_{\beta l}$ and $\Phi_{\beta s}$ are function of the arguments $z\rme^{-\beta l}$ and $z\rme^{-\beta s}$ respectively, and the solutions are $\Phi_{\beta l}=\ln\,y_{p}\left(z\rme^{-\beta l}\right)\quad\mbox{and}\quad\Phi_{\beta s}=-\ln\,y_{p}\left(z\rme^{-\beta s}\right)$ (11) where $y_{p}(x)$ satisfies the equation $y_{p}(x)-x\,y_{p}^{1-p}(x)=1$, which is the same equation (2) one encounters in exclusion statistics. Equation (11) implies, $\Phi(u,v,z)=\int_{v}^{u}\ln y_{p}\left(z\rme^{-\epsilon}\right)\,\rmd\epsilon.$ Therefore, (8) yields $\ln Z(\beta,z,l,s)=\frac{1}{\beta}\int_{\beta s}^{\beta l}\ln y_{p}\left(z\rme^{-\epsilon}\right)\,\rmd\epsilon,$ (12) i.e. (1) with constant density of states ${\tilde{\rho}}(\epsilon)=1$, and the lower and upper limits of integration being $s$ and $l$ respectively. This is the key equation, using which we compute the limit shapes and the largest parts of the Young diagrams in section 4 and section 5 respectively. The limit $\beta l\rightarrow\infty$ also provides a simpler derivation of an earlier result [10], which showed a link between the exclusion statistics and the MDP–$p$ problem. ## 4 Limit shapes of Young diagrams Let us consider all the MDP–$p$ of an integer $E$ with uniform measure. Then the number of columns having height between $s$ and $l$, averaged over all the Young diagrams of the MDP–$p$ of $E$, is obtained from (12) as $\qquad\left\langle N_{s}^{l}(z)\right\rangle=z\frac{\partial}{\partial z}\ln Z(\beta,z,l,s)=\frac{1}{\beta}\left[\ln y_{p}\left(z\rme^{-\beta s}\right)-\ln y_{p}\left(ze^{-\beta l}\right)\right].$ (13) Now to obtain the parameter $\beta$ in terms of the given large integer $E$ one again uses (12) with the limits $\beta l\rightarrow\infty$, $\beta s\rightarrow 0$, and $z=1$, i.e., $\qquad E=-\frac{\partial}{\partial\beta}\ln Z(\beta,1,\infty,0)=\frac{b^{2}(p)}{\beta^{2}},\quad\mbox{where}\quad b^{2}(p)=\int_{0}^{\infty}\ln y_{p}\left(\rme^{-\epsilon}\right)\,\rmd\epsilon$ (14) is a constant which depends on the parameter $p$. The average shape or the height profile of the Young diagrams $\langle W_{h}\rangle$ is simply given by (13) with $s=h$, $l\rightarrow\infty$ and $z=1$, i.e., $\beta\left\langle W_{h}\right\rangle=\ln y_{p}\left(\rme^{-\beta h}\right),\quad\mbox{where}\quad\beta=\frac{b(p)}{\sqrt{E}}.$ (15) For instance for $p=0,1$ and $2$, solving (2) yields $y_{0}(x)=1/(1-x)$, $y_{1}(x)=(1+x)$, and $y_{2}(x)=\left[1+\sqrt{1+4x}\right]/2$ respectively. From which using (14) one finds $b(0)=\pi/\sqrt{6}$, $b(1)=\pi/\sqrt{12}$ and $b(2)=\pi/\sqrt{15}$ in agreement with the earlier known results [12, 15]. Figure 3: Limit shapes for the minimal difference $p$ partitions with $p=0,1,2$, and $3$, where $b(0)=\pi/\sqrt{6}$, $b(1)=\pi/\sqrt{12}$, $b(2)=\pi/\sqrt{15}$, and $b(3)=0.752617\ldots$. The fluctuation about the average shape can be computed from (12) using $\langle W_{h}^{2}\rangle-\langle W_{h}\rangle^{2}=z\frac{\partial}{\partial z}z\frac{\partial}{\partial z}\ln Z(\beta,z,\infty,h)\Big{|}_{z=1},$ (16) which gives $\beta^{2}\left[\langle W_{h}^{2}\rangle-\langle W_{h}\rangle^{2}\right]=\beta\rme^{-\beta h}\left[\frac{y_{p}^{\prime}\left(\rme^{-\beta h}\right)}{y_{p}\left(\rme^{-\beta h}\right)}\right],$ (17) where $y_{p}^{\prime}(x)$ denotes the derivative of $y_{p}(x)$ with respect to its argument. This formula shows that the random variable $\beta W_{h}$ is strongly peaked around its mean value. Therefore, the curve $W_{h}/\sqrt{E}$ as a function of $h/\sqrt{E}$ converges to a limit curve when $E\rightarrow\infty$ (strictly speaking, to prove the existence of a limit curve, one needs to show that all the moments around the mean vanish when $E\rightarrow\infty$, which Vershik [13] showed for $p=0$ and $p=1$). Therefore hereafter we may replace $\langle\beta W_{h}\rangle$ by $\beta W_{h}$. Using (2) and (15), one can express $h$ in terms of $W_{h}$ as, $h=-\frac{1}{\beta}\ln\left(1-\rme^{-\beta W_{h}}\right)-pW_{h}.$ (18) Introducing the scaling variables $x=W_{h}/\sqrt{E}$ and $y=h/\sqrt{E}$, using (14) and taking $E\rightarrow\infty$, yields the equation of the limit shape $y=-\frac{1}{b(p)}\ln(1-\rme^{-xb(p)})-px.$ (19) Figure 3 shows the limit shapes for the MDP–$p$ with $p=0,1,2$, and $3$. Figure 4: (a) Transposed Young diagram for the unrestricted partition $91=18+16+13+13+9+6+5+5+3+3$. (b) Represents (a) in terms of non-interacting bosons (represented by ○) occupying energy levels $\epsilon_{i}=i$ for $i=1,2,\ldots,10$. (c) The configuration obtained from the bosonic configuration (b) by transferring particles from the higher levels to the lower levels such that in the final configuration all the levels below the highest occupied level $\epsilon_{5}=5$ receive 2 new particles (represented by ●) each, where ○represents the particles originally present in the initial bosonic configuration. (d) The Young diagram corresponding to the configuration (c). This is the transposed Young diagram of the partition $49=18+14+9+7+1$, in the minimal difference 2 partition problem. Equation (18) has a simple physical interpretation which we explain below. For $p=0$, any transposed Young diagram (see figure 4) provides a valid unrestricted partition. Therefore the transposed diagram also corresponds to a non interacting system of bosons occupying single particle equidistant energy levels. However this is no more true when $p>0$. In this case the transposed Young diagram(see figure 4) corresponds to a quantum system where there is a certain energy level (which differs from one realization to another) which is occupied by at least one particle, and above which all the levels are empty, and below which each of the levels must be occupied by at least $p$ particles. Therefore, $h$ in the limit shape expression (18) represents the number of particles above the energy level $W_{h}$. For bosons with total energy $E$, this number is precisely given by (18) with $p=0$ and $\beta$ has to be determined in terms of $E$. Now, a configuration for $p>0$, can be obtained from a bosonic configuration by transferring particles from the higher energy levels to the lower ones such that, in the final configuration, levels below the highest occupied level (which has at least one particle) receive exactly $p$ new particles each. Clearly, in the final configuration obtained by this procedure, each of the levels below the highest occupied level has at least $p$ particles. However, since transferring a particle from a higher energy level to a lower one decreases energy of the system, to obtain a configuration for $p>0$ with energy $E$ requires the initial bosonic configuration to be at a higher energy (i.e., lower inverse temperature $\beta$) than $E$. Now, while going from a initial bosonic configuration to a configuration for $p>0$, one transfers total of $pW_{h}$ particles from the levels above $W_{h}$ to below (i.e., $p$ particles to each level), the average number number of particles above level $W_{h}$ decreases from the corresponding bosonic system ($p=0$) precisely by $pW_{h}$, which is exactly the content of (18) . In fact, $\beta$ in (18) can directly be determined by using condition $h\geq 0$ and the normalization $\int_{0}^{W_{h}^{*}}h\left(W_{h}\right)\,\rmd W_{h}=E$, where $W_{h}^{*}$ is the solution of the equation $h\left(W_{h}^{*}\right)=0$. Writing $\exp\left(\beta W_{h}^{*}\right)=y^{*}$, it satisfies $y^{*}-y^{*1-p}=1$, and in terms of $y^{*}$ one finds $\beta=\frac{b(p)}{\sqrt{E}}\quad\mbox{with}\quad b^{2}(p)=\frac{\pi^{2}}{6}-\mathrm{Li_{2}}(1/y^{*})-\frac{p}{2}(\ln y^{*})^{2},$ (20) where $\mathrm{Li_{2}}(z)=\sum_{k=1}^{\infty}z^{k}k^{-2}$ is the dilogarithm function. The expression for $b(p)$ in (20) also follows directly from its integral representation given in (18). ## 5 Largest part of Young diagrams Equation (12) also allows one to compute the distribution of the largest part (i.e., the largest height in the Young diagram) in the MDP–$p$ problem. Let $\rho_{p}(E,l)$ be the number of partitions of the integer $E$ in MDP–$p$ problem, such that the largest part is at most $l$. Clearly, $\rho_{p}(E)=\rho_{p}(E,l\rightarrow\infty)$ gives the total number of partitions of $E$ and since the partitions are distributed with a uniform measure $C_{p}(l|E)=\rho_{p}(E,l)/\rho_{p}(E)$ gives the cumulative distribution of the largest height $l$. Note that the partition function $Z_{l}(\beta)=\sum_{E}\rme^{-\beta E}\rho_{p}(E,l)$ in the limit $\beta\rightarrow 0$ is obtained from (12) by simply taking the limit $\beta s\rightarrow 0$ and $z=1$. Therefore, formally inverting the Laplace transform (in the limit $\beta\rightarrow 0$, the sum over $E$ in the partition function of $Z_{l}(\beta)$ can be replaced by an integral), one can write $\rho_{p}(E,l)=\frac{1}{2\pi i}\int_{\gamma-i\infty}^{\gamma+i\infty}\exp\left[S_{E,l}(\beta)\right]\,\rmd\beta,$ (21) where $\gamma$ is a real constant chosen such that all singularities of integrand are to the left of the vertical contour in the complex plane, and the action $S_{E,l}(\beta)=\beta E+\frac{1}{\beta}\int_{0}^{\beta l}\ln y_{p}\left(\rme^{-\epsilon}\right)\,\rmd\epsilon.$ (22) For large $E$, the leading asymptotic behavior of $\rho_{p}(E,l)$ can be obtained from the saddle point approximation. Maximizing the action with respect to $\beta$, i.e., setting $\partial S/\partial\beta=0$ gives the saddle point equation $\beta^{2}E=\int_{0}^{\beta l}\ln y_{p}\left(\rme^{-\epsilon}\right)\,\rmd\epsilon-\beta l\ln y_{p}\left(\rme^{-\beta l}\right).$ (23) For large $E$, the saddle point $\beta^{*}$ is obtained implicitly solving the above equation and by substituting it back in the action $S_{E,l}(\beta^{*})$. Thus, to leading order, $\rho_{p}(E,l)\approx\exp\left[S_{E,l}(\beta^{*})\right],$ (24) where $S_{E,l}(\beta^{*})$ can be written as $S_{E,l}(\beta^{*})\approx\frac{1}{\beta^{*}}\left[2\int_{0}^{\beta^{*}l}\ln y_{p}\left(\rme^{-\epsilon}\right)\,\rmd\epsilon-\beta^{*}l\ln y_{p}\left(\rme^{-\beta^{*}l}\right)\right].$ (25) It is evident from the above equations that, in terms of $l$ and $E$, one has the scaling form $S_{E,l}(\beta^{*})=\sqrt{E}\,g_{p}(l/\sqrt{E})$, where the scaling function $g_{p}(x)$ can be determined as follows. We set $l/\sqrt{E}=x$ and $\beta^{*}l=H_{p}(x)$. In terms of these scaling variables, from the saddle point solution of (23) and the entropy (25) one has $\displaystyle\frac{H_{p}^{2}(x)}{x^{2}}$ $\displaystyle=\int_{0}^{H_{p}(x)}\ln y_{p}\left(\rme^{-\epsilon}\right)\,\rmd\epsilon-H_{p}(x)\ln y_{p}\left(\rme^{-H_{p}(x)}\right),$ (26) and $\displaystyle g_{p}(x)$ $\displaystyle=2\frac{H_{p}(x)}{x}+x\ln y_{p}\left(\rme^{-H_{p}(x)}\right),$ (27) respectively. Thus, given $x$, one has to find $H_{p}(x)$ by implicitly solving (26), then substitute it back in (27) to get $g_{p}(x)$, and finally $\rho_{p}(E,l)\approx\exp\left[\sqrt{E}\,g_{p}\left(\frac{l}{\sqrt{E}}\right)\right].$ (28) For large $x$, using (26) and (27), it can be shown that $g_{p}(x)\approx 2b(p)-\frac{1}{b(p)}\exp\left[-b(p)x\right]\qquad\mbox{as}\quad x\rightarrow\infty,$ (29) where $b(p)$ is given in (14) and (20). Thus, from (28), $\rho_{p}(E)=\rho_{p}(E,l\rightarrow\infty)\sim\exp[2b(p)\sqrt{E}]$ to leading order for large $E$, which is the generalization of the Hardy-Ramanujan formula [17] for $\rho_{0}(E)$, provided by Meinardus [18]. The normalized cumulative distribution of $l$, i.e., $C_{p}(l|E)=\rho_{p}(E,l)/\rho_{p}(E)$, for large $E$ and $l\gg\sqrt{E}$, is therefore $C_{p}(l|E)\approx\exp\left[-\frac{\sqrt{E}}{b(p)}\exp\left(-\frac{b(p)}{\sqrt{E}}l\right)\right]=F\left(\frac{b(p)}{\sqrt{E}}\Bigl{[}l-l^{*}(E)\Bigr{]}\right),$ (30) where the characteristic value of $l$ is $l^{*}(E)=[\sqrt{E}/b(p)]\ln(\sqrt{E}/b(p))$, and the scaling function has the Gumbel form, $F(z)=\exp[-\exp[-z]]$. The result for the $p=0$ case, i.e., for $C_{0}(l|E)$, was first derived Erdös and Lehner [16]. Equation (30) provides a generalization of their result, which is valid for all $p$. The probability distribution $P_{p}(l|E)=C_{p}(l|E)-C_{p}(l-1|E)\approx{\partial C_{p}(l|E)}/{\partial l}$, obtained from (30), $\quad P_{p}(l|E)\approx\frac{b(p)}{\sqrt{E}}\,F^{\prime}\left(\frac{b(p)}{\sqrt{E}}\Bigl{[}l-l^{*}(E)\Bigr{]}\right),\quad\mbox{where}\quad F^{\prime}(z)=\exp[-z-\exp[-z]],$ (31) is highly asymmetric around the peak at $l=l^{*}(E)$. This limiting distribution describes the probability of typical fluctuations of $\Or(\sqrt{E})$ of the random variable $l$ around the peak $l^{*}(E)$. ## 6 Summary and remarks In summary, we have obtained a generalized grand partition function for the minimal difference $p$ partition (MDP–$p$) of a positive integer $E$, where smallest part is at least $s$ and largest part is at most $l$, in the scaling limit $\beta\propto E^{-1/2}\rightarrow 0$, in terms of the scaling variables $\beta l$ and $\beta s$. The limit $\beta l\rightarrow\infty$ also provides a simpler derivation of an earlier result [10], which showed a link between the exclusion statistics and the MDP–$p$ problem, by showing that both problems are described by the same grand partition function in the limit $\beta\rightarrow 0$. Using the grand partition function we have computed the limiting shape of the Young diagram of the MDP–$p$ problem for all $p$, and also provided a simple physical interpretation of the result. Although the Young diagram is defined only for integer values of $p$, one can analytically continue the expression (15) for the width $W_{h}$ of the Young diagram to noninteger values of $p$. For $0<p<1$, $W_{h}$ corresponds to the number of particles each of which has energy at least $h$, in a system where the particles obeys exclusion statistics. We have also obtained the asymptotic distribution of the largest part of the Young diagram and showed that the scaled distribution has a Gumbel form for all $p$. When one analytically continues, for $0<p<1$, the largest part corresponds to the highest occupied energy level in exclusion statistics. Note that for $p=0$, the transposed Young diagram of a given partition gives another valid $p=0$ partition. This symmetry implies that the statistics of the largest part is the same as the statistics of the number of parts in the $p=0$ partition problem. The distribution of the number of parts for $p=0$ was computed by Erdös and Lehner [16] and in the appropriate scaling limit it has a Gumbel form. However, the symmetry between the number of parts and the largest part no longer holds when $p>0$, where the distribution of the number of parts become Gaussian (see [10] and references therein). Recently, the statistics of the number of parts for a general partitions of the form $E=\sum n_{i}i^{1/\nu}$ that corresponds to having a power-law density of states, $\tilde{\rho}(\epsilon)\sim\epsilon^{\nu-1}$, has been studied [19] in the bosonic sector ($p=0$). Clearly, $\nu=1$ corresponds to the usual unrestricted partition problem, where the number of parts obey Gumbel statistics. Interestingly, for $\nu\neq 1$, the authors in [19] also obtained the other two universal distribution laws of extreme value statistics, namely the Fréchet and Weibull distributions for $0<\nu<1$ and $\nu>1$ respectively. Therefore, the general partition problem can be defined in the parameter space of $(\nu,p)$ with $\nu>0$ and $p\geq 0$. In this parameter space the point $(\nu=1,p=0)$ is a very special one at which both the number of parts and the largest part obey the same statistics given by the Gumbel distribution. Along the line $\nu=1$, the limiting distribution of the number of parts becomes Gaussian as soon as $p>0$, whereas the limiting distribution of the largest part remains Gumbel for all $p$, as we have shown in this paper. On the other hand, along the $p=0$ line, for the number of parts one finds [19] all the three universal laws of the extreme value statistics, for the parameter $0<\nu<1$, $\nu=1$, and $\nu>1$ respectively. Therefore, it is interesting to ask whether there is any region in the $(\nu,p)$ parameter space, where the largest part obeys another statistics than the Gumbel one. The answer is negative. For a general density of states, (22) includes a factor of $\tilde{\rho}(\epsilon/\beta)$ in the integrand. Following the similar steps provided afterwards, it can be shown that even for the power-law density of states $\tilde{\rho}(\epsilon)\sim\epsilon^{\nu-1}$, the scaled distribution of the largest part remains Gumbel in the whole $(\nu,p)$ plane. Thus, the largest part obeys a more robust law, in contrast to the number of parts. ## Note added in proof We thank K. Hikami for pointing out [20] in which the author obtained the solution of a recursion relation similar to (6) with $s=0$ for arbitrary $\beta$. However, for the purpose of this paper we require the solution only in the limit $\beta\rightarrow 0$. In this limit it is simpler to obtain it using the method presented in this paper rather than obtaining by taking the limit $\beta\rightarrow 0$ in the solution of [20]. The average occupation number at a level $i$ for the exclusion statistics has been studied in [2, 3, 4, 21, 22], which also can be obtained from (13) simply through $\langle n_{i}\rangle=-\frac{\partial}{\partial s}\left\langle N_{s}^{l}(z)\right\rangle\Big{|}_{s=i}=\frac{\partial}{\partial l}\left\langle N_{s}^{l}(z)\right\rangle\Big{|}_{l=i},$ (32) which via elementary algebra yields $\langle n_{i}\rangle=\left(\frac{1}{y_{p}(z\rme^{-\beta i})-1}+p\right)^{-1}$ (33) AC, SNM and SS acknowledge the support of the Indo-French Centre for the Promotion of Advanced Research (IFCPAR/CEFIPRA) under Project 3404-2. ## References ## References * [1] Haldane F D M 1991 “Fractional statistics” in arbitrary dimensions: A generalization of the Pauli principle Phys. Rev. Lett. 67 937 * [2] Dasnières de Veigy A and Ouvry S 1994 Equation of state of an anyon gas in a strong magnetic field Phys. Rev. Lett. 72 600 * [3] Dasnières de Veigy A and Ouvry S 1995 One-dimensional statistical mechanics for identical particles: the Calogero and anyon cases Mod. Phys. Lett. B 9 271 * [4] Wu Y S 1994 Statistical distribution for generalized ideal gas of fractional-statistics particles Phys. Rev. Lett. 73 922 * [5] Murthy M V N and Shankar R 1994 Haldane exclusion statistics and second virial coefficient Phys. Rev. Lett. 72 3629 * [6] Murthy M V N and Shankar R 1994 Thermodynamics of a one-dimensional ideal gas with fractional exclusion statistics Phys. Rev. Lett. 73 3331 * [7] Isakov S B 1994 Fractional statistics in one dimension: modeling by means of $1/x^{2}$ interaction and statistical mechanics Int. J. Mod. Phys. A 9 2563 * [8] Bergère M C 2000 Fractional statistic J. Math. Phys. 41 7252 * [9] Andrews G E 1998 The Theory of Partitions (Cambridge University Press, Cambridge) * [10] Comtet A, Majumdar S N and Ouvry S 2007 Integer partitions and exclusion statistics J. Phys. A: Math. Theor. 40 11255 * [11] Temperley H N Y 1952 Statistical mechanics and the partition of numbers: the form of crystal surfaces Proc. Cambridge Philos. Soc. 48 683 * [12] Vershik A M 1996 Statistical mechanics of combinatorial partitions and their limit shapes Funct. Anal. Appl. 30 90 * [13] Freiman G, Vershik A M and Yakubovich Yu V 2000 A local limit theorem for random strict partitions Theory Probab. Appl 44 453 * [14] Vershik A M and Yakubovich Yu V 2001 The limit shape and fluctuations of random partitions of naturals with fixed number of summands Moscow Math. J. 1 457 * [15] Romik D 2003 Identities arising from limit shapes of constrained random partitions Preprint * [16] Erdös P and Lehner J 1951 The distribution of the number of summands in the partitions of a positive integer Duke Math. J. 8 335 * [17] Hardy G H and Ramanujan S 1918 Asymptotic formulaæ in combinatory analysis Proc. London. Math. Soc. 17 75 * [18] Meinardus G 1954 Über partitionen mit differenzenbedingungen Math. Zeitschr. 61 289 * [19] Comtet A, Leboeuf P and Majumdar S N 2007 Level density of a Bose gas and extreme value statistics Phys. Rev. Lett. 98 070404 * [20] Hikami K 1995 Character and TBA for an ideal $g$-on gas Phys. Lett. A 205 364 * [21] Hikami K 1998 Statistical mechanical interpretation of the inverse scattering method: level dynamics for exclusion statistics Phys. Rev. Lett. 80 4374 * [22] Hikami K 2000 Exclusion statistics and chiral partition function Physics and Combinatorics (Proc. Nagoya 2000 Int. Workshop) ed A N Kirillov and N Liskova (Singapore: World Scientific) pp 22–48

arxiv-papers

2024-09-04T02:48:53.012923

0707.3421

# The nature of electromagnetic energy Jerrold Franklin111Internet address: Jerry.F@TEMPLE.EDU Department of Physics Temple University, Philadelphia, PA 19122-6082 (Seotember 28, 2008) ###### Abstract The nature of the electromagnetic (EM) energy for general charge and current distributions is analyzed. We come to the conclusion that EM energy can reside only within charge-current distributions, and that EM energy density cannot exist in otherwise empty space. The form $\frac{1}{2}(\rho\phi+{\bf j\mbox{\boldmath$\cdot$}A})$ is a suitable EM energy density, but $({\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H})/8\pi$ is not. PACS numbers: 03.50.De, 41.20.-q ## 1 Introduction James Clerk Maxwell, in his 1864 paper “A Dynamical Theory of the Electromagnetic Field”[1], introduced the concept of electromagnetic (EM) field energy proportional to (${\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H}$) that existed throughout space. For Maxwell this was a natural property of the aether which he conceived of as a material substance permeating space. He thought of the EM energy in the aether as being similar to elastic energy in any other material object. Although most physicists have long discarded the concept of an aether, at least as a material substance, the tendency has been to follow Maxwell’s lead in the use of EM field energy. While most texts point out the alternative of EM energy residing in the charge and current distributions in the form of an energy density proportional to $(\rho\phi+{\bf j\mbox{\boldmath$\cdot$}\bf A})$, they usually settle on (${\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H}$) as the more useful form. Indeed, we have become comfortable with EM field energy permeating otherwise empty space. In this paper, we first analyze EM energy for specific examples of electrostatic (Sec. 2) and magnetostatic (Secs. 3 and 4) energy, and then general time dependent electromagnetic energy (Sec. 5). In each case we find that the EM energy must reside in matter, and reach the conclusion that $({\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H})/8\pi$ cannot be considered an energy density. EM energy resides in charge and current densities. Only $\frac{1}{2}(\rho\phi+{\bf j\mbox{\boldmath$\cdot$}A})$ is a suitable EM energy density, with no electromagnetic energy in otherwise free space. ## 2 Electrostatic Energy The electrostatic energy of a charge distribution $\rho$ in a static electric potential $\phi_{0}$ is given by $U_{\rho\phi_{0}}=\int\rho\phi_{0}d^{3}r,$ (1) with the understanding that the external potential $\phi_{0}$ is due to charges other than the distribution $\rho$. The subscript $\rho\phi_{0}$ is used here to emphasize that the energy given by this equation is due to a charge distribution in an external potential. This equation can be considered the definition of the electric potential as potential energy per unit charge. Note that the integral is taken over all space, but only regions with charge contribute to the integral. The electrostatic energy can also be given in terms of the electric fields by using Maxwell’s equation222We do not include polarizable matter in this and the following section. for $\mbox{\boldmath$\nabla\cdot$}E$ in Eq. (1): $\displaystyle U_{\rho\phi_{0}}$ $\displaystyle=$ $\displaystyle\int\rho\phi_{0}d^{3}r$ (2) $\displaystyle=$ $\displaystyle\frac{1}{4\pi}\int\phi_{0}(\mbox{\boldmath$\nabla\cdot$}{\bf E_{\rho}})d^{3}r$ $\displaystyle=$ $\displaystyle\frac{1}{4\pi}\int\left[\mbox{\boldmath$\nabla\cdot$}(\phi_{0}{\bf E_{\rho}})-{\bf E_{\rho}}\mbox{\boldmath$\cdot$}\mbox{\boldmath$\nabla$}\phi_{0}\right]d^{3}r$ $\displaystyle=$ $\displaystyle\frac{1}{4\pi}\oint{\bf dS\mbox{\boldmath$\cdot$}(\phi_{0}E_{\rho}})+\frac{1}{4\pi}\int{\bf E_{\rho}\mbox{\boldmath$\cdot$}E_{0}}d^{3}r.$ The surface integral is usually discarded by being taken over an infinite sphere for which it is presumed to vanish. This leads to the form $U_{\bf E\rho E_{0}}=\frac{1}{4\pi}\int{\bf E_{\rho}\mbox{\boldmath$\cdot$}E_{0}}d^{3}r,$ (3) for the electrostatic energy purely in terms of the electric field $\bf E_{\rho}$ of the charge distribution and the external electric field $\bf E_{0}$. (Note: We are using Gaussian units.) This integral is also over all space, however the regions of space that contribute to the integral are generally different for the integrals in Eqs. (1) and (3). The derivation of Eq. (3) from Eq. (1) is so simple that they are generally considered to be equivalent formulas for the same electrostatic energy. However we shall see that the two are not equivalent, and that Eq. (3) must be supplemented to give the correct energy for some cases. We first demonstrate this by the simple example of an electric dipole in a uniform electric field $\bf E_{0}$. The potential for this electric field is $\phi_{0}=-{\bf r\mbox{\boldmath$\cdot$}E_{0}}.$ (4) The energy $U_{\rho\phi_{0}}$ is given by $U_{\rho\phi_{0}}=\int\rho\phi_{0}d^{3}r=-\int\rho{\bf r\mbox{\boldmath$\cdot$}E_{0}}d^{3}r=-{\bf p\mbox{\boldmath$\cdot$}E_{0}}.$ (5) We see that in a uniform electric field, the energy of a charge distribution depends only on its electric dipole moment $\bf p$. For a point dipole at a position $\bf r$ in a spatially varying field, the energy is given by $U_{\rho\phi}=-{\bf p\mbox{\boldmath$\cdot$}E(r)},$ (6) For this case, the force on the point dipole will be the negative gradient of the energy: ${\bf F=\mbox{\boldmath$\nabla$}(p\mbox{\boldmath$\cdot$}E)}.$ (7) Equation (3) using the electric fields gives a different answer for the energy. For simplicity, we take the charge distribution to be a point dipole. Then the dipole electric field is given by ${\bf E_{\rho}}=\frac{\left[3{\bf(p\mbox{\boldmath$\cdot$}{\hat{r}}){\hat{r}}-p}\right]}{r^{3}}-4\pi{\bf(p\mbox{\boldmath$\cdot$}{\hat{r}}){\hat{r}}}\delta({\bf r}).$ (8) When this is put into Eq. (3), the first term cancels in the angular integration, and the second term leads to $U_{\bf E_{0}}=-\int{\bf(p\mbox{\boldmath$\cdot$}{\hat{r}})({\hat{r}}\mbox{\boldmath$\cdot$}E_{0})}\delta({\bf r})d^{3}r=-\frac{1}{3}{\bf p\mbox{\boldmath$\cdot$}E_{0}},$ (9) which is different than the energy $U_{\rho\phi_{0}}$. Of the two results, it is clear that $-{\bf p\mbox{\boldmath$\cdot$}E_{0}}$ is the correct energy, both from the definition of the potential and the fact that it leads to the correct form for the torque on the dipole. What went wrong with $U_{\bf E_{\rho}E_{0}}$? We can see this by looking at the surface integral which is usually discarded, but does not always vanish. For some cases, such as the present one, the surface integral In Eq. (2) must be included to get the correct energy. For the electric dipole in a uniform field, the surface integral, taken over a sphere of radius $r$, is $\displaystyle U_{\rm surf}$ $\displaystyle=$ $\displaystyle\frac{1}{4\pi}\oint{\bf dS\mbox{\boldmath$\cdot$}(\phi_{0}E_{\rho}})$ (10) $\displaystyle=$ $\displaystyle-\frac{1}{4\pi}\oint\frac{{\bf dS}\mbox{\boldmath$\cdot$}\left[3({\bf p\mbox{\boldmath$\cdot$}{\hat{r}}){\hat{r}}-p}\right]({\bf r\mbox{\boldmath$\cdot$}E_{0})}}{r^{3}}$ $\displaystyle=$ $\displaystyle-\frac{1}{4\pi}\oint d\Omega\left[3{\bf p\mbox{\boldmath$\cdot$}{\hat{r}}-p\mbox{\boldmath$\cdot$}{\hat{r}}}\right]({\bf{\hat{r}}\mbox{\boldmath$\cdot$}E_{0})}$ $\displaystyle=$ $\displaystyle-\frac{2}{3}{\bf p\mbox{\boldmath$\cdot$}E_{0}}.$ Including this surface contribution gives the correct electrostatic energy, in agreement with $U_{\rho\phi_{0}}$ We see that the surface integral in Eq. (2) must be included except in those cases where it does vanish. Since the surface integral involves the potential $\phi$, this means that the electrostatic energy cannot always be given solely in terms of the electric field. Note that the surface integral is independent of the radius of the sphere, which can be infinite. This means that even integrating ${\bf E}^{2}$ over “all space” does not exclude the need for the surface integral. This does not end the story, because there is an alternate derivation of Eq. (3) that seems unrelated to the potential. This derivation starts with the rate of energy put into matter by the electromagnetic field in a volume $V$, given by $\frac{dU_{\rm Matter}}{dt}=\int_{V}{\bf j\mbox{\boldmath$\cdot$}E}d^{3}r,$ (11) where $\bf j$ now represents all the currents and $\bf E$ is the electric field due to all sources. The standard textbook derivation,[2-4] using Maxwell’s equations and some vector identities, eventually leads to the result $\frac{dU_{\rm Matter}}{dt}+\frac{1}{8\pi}\int_{V}\partial_{t}({\bf E}^{2}+{\bf B}^{2})d^{3}r=-\frac{c}{4\pi}\oint_{S}{\bf dS\mbox{\boldmath$\cdot$}(E\mbox{\boldmath$\times$}B)}.$ (12) The surface integral on the right hand side is usually interpreted (because of its minus sign) as the rate in which electromagnetic energy enters the volume $V$ through the surface $S$ with the identification of the Poynting vector $\mbox{\boldmath$\cal P$}=\frac{c}{4\pi}{\bf(E\mbox{\boldmath$\times$}B)}$ (13) as the electromagnetic intensity (transmitted power per unit area). The first time derivative on the left hand side of Eq. (12) is the rate of change of the energy of the matter inside the volume. Assuming conservation of energy, the second term is then deduced to be the rate of change of the energy of the electromagnetic fields inside the volume. For the purely electric case, this has been taken to imply that the electrostatic energy can be given by $U_{\bf E}=\frac{1}{8\pi}\int{\bf E}^{2}d^{3}r.$ (14) The $\bf E$ in this integral is the total electric field, both external and that of the charge distribution $\rho$, so ${\bf E}^{2}={\bf E}^{2}_{\rho}+{\bf E}^{2}_{0}+2{\bf E_{\rho}\mbox{\boldmath$\cdot$}E_{0}}$. When the self energy contributions ${\bf E}^{2}_{\rho}$ and ${\bf E}^{2}_{0}$ are excluded from the integration, this gives the same result as Eq. (3). This will give the same wrong answer for the case of the electric dipole in a uniform electric field. What is missing now? To see this, let us first look at Eq. (11), the starting point for this second derivation,. We can consider the negative of the right hand side to be the rate of energy input to the electromagnetic field. If we start with a constant external field $\bf E_{0}$ and no initial charge distribution, the time integral of this rate will be the energy put into the field: $U(t)=-\int_{0}^{t}dt\int_{V}{\bf j\mbox{\boldmath$\cdot$}E_{0}}d^{3}r.$ (15) Now, consider a rod of length $\bf L$, originally uncharged. A current $\bf j$ in the rod in the direction of $\bf L$ will produce a dipole moment ${\bf p}(t)$ given by ${\bf p}(t)=q(t){\bf L}=\int_{0}^{t}dt\int_{V}{\bf j}d^{3}r.$ (16) Using this expression for the electric dipole moment of the rod, we get for its energy in an external field $\bf E_{0}$, $U=-\int_{0}^{t}dt\int_{V}{\bf j\mbox{\boldmath$\cdot$}E_{0}}d^{3}r=-{\bf p\mbox{\boldmath$\cdot$}E_{0}},$ (17) which is the correct answer. We see that the starting point for the derivation leading to Eq. (14) is correct. Where did the rest of the derivation go wrong? During the charging of the ends of the rod, there will be a current in the rod, which produces a magnetic field $\bf B$. Thus there will be an $\bf E\mbox{\boldmath$\times$}B$ vector, resulting in electromagnetic energy passing through the outer surface of the volume. (This is if we want to use the ${\bf E}^{2}$ integral for the energy.) We now show that this energy equals two thirds of the energy put into the field by the ${\bf j\mbox{\boldmath$\cdot$}E}$ integral. We consider the rod of length L to be along the $z$-axis and use spherical coordinates. The magnetic field due to a current $I$ in the rod is given by the law of Biot-Savart: ${\bf B}=\frac{I{\hat{\mbox{\boldmath$\phi$}}}}{cr\sin\theta}\left[\cos\theta-\frac{(r\cos\theta-L)}{\sqrt{r^{2}+L^{2}-2rL\cos\theta}}\right].$ (18) This is the magnetic field produced by the current in the rod as its ends are being charged. According to Eq. (12), the electromagnetic power passing through a sphere of radius R during the charging of the rod in the presence of the electric field $\bf E_{0}$ is given by $\displaystyle P$ $\displaystyle=$ $\displaystyle\frac{c}{4\pi}\oint{\bf dS\mbox{\boldmath$\cdot$}\left(E_{0}\mbox{\boldmath$\times$}B\right)}=\frac{c}{4\pi}\oint R^{2}d\Omega{\bf\left[{\hat{r}}\mbox{\boldmath$\cdot$}\left(E_{0}\mbox{\boldmath$\times$}{\hat{\mbox{\boldmath$\phi$}}}\right)\right]}B$ (19) $\displaystyle=$ $\displaystyle\frac{-cR^{2}}{4\pi}\oint d\Omega\frac{IE_{z}\sin\theta}{cR\sin\theta}\left[\cos\theta-\frac{(R\cos\theta-L)}{\sqrt{R^{2}+L^{2}-2RL\cos\theta}}\right]$ $\displaystyle=$ $\displaystyle\frac{-RIE_{z}}{2}\int_{-1}^{1}du\left[u-\frac{(Ru-L)}{\sqrt{R^{2}+L^{2}-2RLu}}\right].$ We have left out a term in $\cos\phi$ that integrates to zero. The remaining integral can be done by expanding in Legendre polynomials, giving $P=-\frac{2}{3}I{\bf E_{0}\mbox{\boldmath$\cdot$}L}.$ (20) This result is independent of the radius of the sphere. The energy that escapes beyond the sphere in producing the dipole moment (if we want to use the ${\bf E}^{2}$ form for energy) is then $U_{\rm escape}=\int Pdt=-\frac{2}{3}\int{I\bf E_{0}\mbox{\boldmath$\cdot$}L}dt=-\frac{2}{3}{\bf p\mbox{\boldmath$\cdot$}E_{0}}$ (21) Thus two thirds of the energy escapes and only one third of the energy is represented by the volume integral of ${\bf E}^{2}$. This is consistent with what we found by doing that integral. The energy passing through the surface is accounted for by including the surface integral term in Eq. (2). That is why the surface integral is necessary to give the correct electrostatic energy if the ${\bf E}^{2}$ form is used. We now consider the question of what combination of fields, potentials, or charges might be considered an electrostatic energy density. The electrostatic energy in terms of the total potential and charge distribution is usually calculated by starting from having no charge distribution, and then building up the final charge distribution in infinitesimal steps. Using Eq. (1) for the incremental energy at each step, leads to the result $U_{\rho\phi}=\frac{1}{2}\int\rho\phi d^{3}r.$ (22) The factor $\frac{1}{2}$ arises from the relation $\delta(\rho\phi)=\rho\delta\phi+\phi\delta\rho=2\phi\delta\rho,$ (23) which follows from $\rho\delta\phi=\phi\delta\rho$ in vacuum or a linear medium. The potential $\phi$ is now the potential due to all sources. Equation (22) gives the total electrostatic energy due to a charge distribution $\rho$ if the integral is taken over all space. However, contributions to the integral come only from those regions where there is charge. If we interpret the integral over a finite region as giving the electrostatic energy in that region, we can define an energy density as $u_{\rho\phi}=\frac{1}{2}\rho\phi.$ (24) With this definition, the electrostatic energy only exists in regions with a non-vanishing charge distribution. Starting from Eq. (22), we can repeat the steps that led to Eq. (2) with the result $U=\frac{1}{2}\int\rho\phi d^{3}r=\frac{1}{8\pi}\oint{\bf dS\mbox{\boldmath$\cdot$}(\phi E})+\frac{1}{8\pi}\int{\bf E}^{2}d^{3}r$ (25) for the electrostatic energy in terms of the total electric field. As before, the correct energy requires the inclusion of a surface integral involving the potential $\phi$ as well as $\bf E$. What if the integral of ${\bf E}^{2}$ is taken over a finite region? Can ${\bf E}^{2}/8\pi$ be considered an energy density? It is clear from Eq. (25) that this is not the case unless the surface integral vanishes. Correctly including the surface integral results in the electrostatic energy being the same as that given by the integral over $\frac{1}{2}\rho\phi$. In any event, the first equality in Eq. (25) shows that there is no electrostatic energy in any region without charge. The fairly common belief that there can be electrostatic field energy in otherwise empty space is wrong. Our conclusion is that ${\bf E}^{2}/8\pi$ cannot be considered an electrostatic energy density. ## 3 Magnetostatic Energy of a Current Distribution The energy of a current distribution in a external magnetic field $\bf B_{0}$ can be found by considering the hypothetical situation of turning on the magnetic field in the presence of the current, which is kept constant as the field is turned on. The constant current source that maintains the current provides the energy for this process. We start from Eq. (15) for the energy input to the EM field: $\displaystyle U_{\bf jA_{0}}(t)$ $\displaystyle=$ $\displaystyle-\int_{0}^{t}dt^{\prime}\int_{V}{\bf j\mbox{\boldmath$\cdot$}E(t^{\prime})}d^{3}r$ (26) $\displaystyle=$ $\displaystyle\frac{1}{c}\int_{0}^{t}dt^{\prime}\int_{V}{\bf j\mbox{\boldmath$\cdot$}(\partial_{t}^{\prime}A(t^{\prime}))}d^{3}r$ $\displaystyle=$ $\displaystyle\frac{1}{c}\int_{V}{\bf j\mbox{\boldmath$\cdot$}A_{0}}(t)d^{3}r.$ In deriving this, we used the fact that the current $\bf j$ does not change with time, and have assumed the absence of an electrostatic potential. Equation (26) gives the energy in a magnetostatic field in terms of the current density and the vector potential. The integral is over all space, but only regions with current contribute to the integral. We can introduce the magnetic field into Eq. (26) for the energy of a current distribution $\bf j$ in a static magnetic field $\bf B_{0}$ by using Maxwell’s $\mbox{\boldmath$\nabla\times$}{\bf B}$ equation: $\displaystyle U_{\bf jA_{0}}$ $\displaystyle=$ $\displaystyle\frac{1}{c}\int_{V}{\bf j\mbox{\boldmath$\cdot$}A_{0}}d^{3}r$ (27) $\displaystyle=$ $\displaystyle\frac{1}{4\pi}\int_{V}{\bf A_{0}\mbox{\boldmath$\cdot$}(\mbox{\boldmath$\nabla\times$}B_{j})}d^{3}r$ $\displaystyle=$ $\displaystyle\frac{1}{4\pi}\int_{V}\left[\mbox{\boldmath$\nabla\cdot$}({\bf B_{j}\mbox{\boldmath$\times$}A_{0})+B_{j}\mbox{\boldmath$\cdot$}(\mbox{\boldmath$\nabla\times$}A_{0})}\right]d^{3}r$ $\displaystyle=$ $\displaystyle\frac{1}{4\pi}\int_{S}{\bf dS\mbox{\boldmath$\cdot$}(B_{j}\mbox{\boldmath$\times$}A_{0})}+\frac{1}{4\pi}\int_{V}{\bf B_{j}\mbox{\boldmath$\cdot$}B_{0}}d^{3}r,$ where $\bf B_{j}$ is the magnetic field due to the current $\bf j$. As was the case for the electrostatic energy, there is a surface term that is usually discarded in describing the magnetostatic energy in terms of $\bf B$. Doing so leads to the expression $U_{\bf B_{j}B_{0}}=\frac{1}{4\pi}\int_{V}{\bf B_{j}\mbox{\boldmath$\cdot$}B_{0}}d^{3}r$ (28) for the magnetostatic energy purely in terms of the magnetic fields. Equations (26) and (27) give the magnetostatic energy of a current distribution in an external magnetic field. To derive the corresponding forms in terms of all currents and fields, we include an increase in the current in Eq. (26) utilizing the relation $\partial_{t}({\bf j\mbox{\boldmath$\cdot$}A})={\bf j\mbox{\boldmath$\cdot$}}(\partial_{t}{\bf A})+{\bf A\mbox{\boldmath$\cdot$}}(\partial_{t}{\bf j})=2{\bf j\mbox{\boldmath$\cdot$}}(\partial_{t}{\bf A}),$ (29) which holds in vacuum and within linear magnetic materials. This leads to the results $\displaystyle U_{\bf jA}(t)$ $\displaystyle=$ $\displaystyle\frac{1}{2c}\int_{V}{\bf j\mbox{\boldmath$\cdot$}A}d^{3}r$ (30) $\displaystyle=$ $\displaystyle\frac{1}{8\pi}\int_{S}{\bf dS\mbox{\boldmath$\cdot$}(B\mbox{\boldmath$\times$}A)}+\frac{1}{8\pi}\int_{V}{\bf B}^{2}d^{3}r,$ (31) where the currents and fields are due to all sources. Neglecting the surface integral leads to the expression $U_{\bf B}=\frac{1}{8\pi}\int_{V}{\bf B}^{2}d^{3}r$ (32) for the magnetostatic energy purely in terms of the magnetic field. To test the several forms for the magnetostatic energy, we consider a current distribution j in a uniform magnetic field $\bf B_{0}$. The vector potential for a uniform magnetic field is given by ${\bf A_{0}}=\frac{1}{2}{\bf B_{0}\mbox{\boldmath$\times$}r}.$ (33) Then the magnetostatic energy is $U_{\bf jA_{0}}=\frac{1}{2c}\int_{V}{\bf j\mbox{\boldmath$\cdot$}(B_{0}\mbox{\boldmath$\times$}r)}d^{3}r=\frac{1}{2c}{\bf B_{0}}\mbox{\boldmath$\cdot$}\int_{V}{\bf r\mbox{\boldmath$\times$}j}d^{3}r=\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\bf B_{0}},$ (34) where $\mu$ is the magnetic moment of the current distribution. We see that in a uniform magnetic field the magnetostatic energy depends only on the magnetic dipole moment. For a “point” magnetic dipole, having a current distribution of negligible extent, the energy in a spatially varying field would be given by $U_{\bf jA}={\bf\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B(r)},$ (35) where $\bf B(r)$ is the magnetic field evaluated at the position of the dipole. We note that the sign of the magnetic dipole energy is opposite that for the electric dipole. That is because the current producing the magnetic dipole is kept constant by a constant current source. The magnetic dipole would rotate so as to increase its energy, thus tending to align with the magnetic field. For this case, with the current kept constant by a constant current source, the force on the point dipole will be the positive gradient of the energy: ${\bf F=\mbox{\boldmath$\nabla$}(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B)}.$ (36) The magnetic dipole force thus has the same form and sign as the electric dipole force. What happens if we use Eq. (28) to evaluate the magnetostatic energy of a point magnetic dipole in a uniform magnetic field? The magnetic field due to a point magnetic dipole is given by ${\bf B}=\frac{\left[3{\bf(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}}){\hat{r}}-\mbox{\boldmath$\mu$}}\right]}{r^{3}}+4\pi\left[\mbox{\boldmath$\mu$}-{\bf(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}}){\hat{r}}}\right]\delta({\bf r}).$ (37) For a magnetic dipole in a uniform magnetic field, the volume integral in Eq. (28) gives $U_{\bf B_{0}}=\int\left[\mbox{\boldmath$\mu$}-{\bf(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}}){\hat{r}}}\right]\mbox{\boldmath$\cdot$}{\bf B_{0}}\delta({\bf r})d^{3}r=\frac{2}{3}{\bf\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B_{0}},$ (38) which is not the correct energy. The surface integral for this case gives $\displaystyle U_{\rm surf}$ $\displaystyle=$ $\displaystyle\frac{1}{4\pi}\int_{S}{\bf dS\mbox{\boldmath$\cdot$}(B_{j}\mbox{\boldmath$\times$}A_{0})}$ (39) $\displaystyle=$ $\displaystyle\frac{1}{8\pi}\int{\bf dS}\mbox{\boldmath$\cdot$}\frac{\left\\{\left[3({\bf\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}}){\hat{r}}-\mbox{\boldmath$\mu$}}\right]\mbox{\boldmath$\times$}({\bf B_{0}\mbox{\boldmath$\times$}r}\right\\}}{r^{3}}$ $\displaystyle=$ $\displaystyle-\frac{1}{8\pi}\int d\Omega{\bf{\hat{r}}}\mbox{\boldmath$\cdot$}\left[\mbox{\boldmath$\mu$}\mbox{\boldmath$\times$}({\bf B_{0}\mbox{\boldmath$\times$}{\hat{r}}})\right]$ $\displaystyle=$ $\displaystyle-\frac{1}{8\pi}\int d\Omega{\bf{\hat{r}}}\mbox{\boldmath$\cdot$}\left[{\bf B_{0}(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}})-{\hat{r}}(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B_{0}})\right]$ $\displaystyle=$ $\displaystyle\frac{1}{3}{\bf\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B_{0}}.$ As was the case for the electrostatic energy, adding the surface integral to the volume integral gives the correct value ${\bf\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B_{0}}$ for the magnetostatic energy. As was done for the electrostatic energy, the magnetostatic energy could also be inferred from Eq. (12) to be given purely in terms of $\bf B^{2}$ by $U_{\bf B}=\frac{1}{8\pi}\int_{V}{\bf B}^{2}d^{3}r,$ (40) in agreement with Eq. (32). This would also give the wrong answer $\frac{2}{3}{\bf\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B_{0}}$ if used to calculate the energy of a magnetic dipole in a uniform magnetic field. We now show (as in the electrostatic case) that the deficiency in magnetostatic energy if the integral of ${\bf B}^{2}$ is used occurs because energy would escape through the surface of the volume during the build up of the magnetic field.. We consider a uniform magnetic field $\bf B_{0}$ building up from zero to its final value in the presence of the magnetic dipole $\mu$, and calculate the energy that escapes through the bounding surface (assumed as a sphere of radius $R$) by integrating the Poynting vector over the surface. This gives $\displaystyle U_{\rm escape}$ $\displaystyle=$ $\displaystyle\frac{c}{4\pi}\int dt\oint{\bf dS\mbox{\boldmath$\cdot$}(E_{0}\mbox{\boldmath$\times$}B_{\mbox{\boldmath$\mu$}}})$ (41) $\displaystyle=$ $\displaystyle-\frac{1}{4\pi}\int dt\oint{\bf dS}\mbox{\boldmath$\cdot$}[(\partial_{t}{\bf A_{0})\mbox{\boldmath$\times$}B_{\mbox{\boldmath$\mu$}}}]$ $\displaystyle=$ $\displaystyle-\frac{1}{4\pi}\oint{\bf dS}\mbox{\boldmath$\cdot$}({\bf A_{0}\mbox{\boldmath$\times$}B_{\mbox{\boldmath$\mu$}}})$ The vector potential $\bf A_{0}$ for the uniform field $\bf B_{0}$ is given by Eq. (33), and the magnetic field of the magnetic dipole by Eq. (37). Using these in Eq. (41) leads to $\displaystyle U_{\rm escape}$ $\displaystyle=$ $\displaystyle-\frac{1}{8\pi}\oint{\bf dS}\mbox{\boldmath$\cdot$}[({\bf B_{0}\mbox{\boldmath$\times$}r)\mbox{\boldmath$\times$}B_{\mbox{\boldmath$\mu$}}}]$ (42) $\displaystyle=$ $\displaystyle-\frac{1}{8\pi}\oint R^{3}d\Omega{\bf{\hat{r}}\mbox{\boldmath$\cdot$}[{\hat{r}}(B_{0}\mbox{\boldmath$\cdot$}B_{\mbox{\boldmath$\mu$}})-B_{0}(B_{\mbox{\boldmath$\mu$}}\mbox{\boldmath$\cdot$}{\hat{r}}})]$ $\displaystyle=$ $\displaystyle-\frac{1}{8\pi}\oint d\Omega R^{3}{\bf[(B_{0}\mbox{\boldmath$\cdot$}B_{\mbox{\boldmath$\mu$}})-({\hat{r}}\mbox{\boldmath$\cdot$}B_{0})(B_{\mbox{\boldmath$\mu$}}\mbox{\boldmath$\cdot$}{\hat{r}}})]$ $\displaystyle=$ $\displaystyle-\frac{1}{8\pi}\oint d\Omega\\{{\bf({B_{0}}\mbox{\boldmath$\cdot$}[3{\hat{r}}(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}})-\mbox{\boldmath$\mu$}]-(B_{0}\mbox{\boldmath$\cdot$}{\hat{r}})[3(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}})-(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}})]}\\}$ $\displaystyle=$ $\displaystyle\frac{1}{8\pi}\oint d\Omega[{\bf({B_{0}}\mbox{\boldmath$\cdot$}\mbox{\boldmath$\mu$})-(B_{0}\mbox{\boldmath$\cdot$}{\hat{r}})(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}{\hat{r}})]}$ $\displaystyle=$ $\displaystyle\frac{1}{3}{\bf\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B_{0}}.$ This means that if the integral of ${\bf B}^{2}$ is used to determine the magnetostatic energy for this case, one third of the energy would escape through any bounding surface. This energy must be included, using the surface integral in Eq. (27), to get the correct total energy. What combination of currents and fields can we interpret as a magnetostatic energy density? As indicated by the steps in Eq. (26), the integral of $\bf j\mbox{\boldmath$\cdot$}A$ over a volume gives the magnetostatic energy within that volume. Thus $u_{\bf jA}=\frac{1}{2c}{\bf j\mbox{\boldmath$\cdot$}A}$ (43) can be considered the magnetostatic energy density. On the other hand, integrating ${\bf B}^{2}$ over a finite volume only gives the magnetostatic energy within the volume if the surface integral vanishes. Only in that case does the integral over ${\bf B}^{2}$ gives the same answer as that over $\bf j\mbox{\boldmath$\cdot$}A$. In any event, the first equality in Eq. (30) shows that there is no magnetostatic energy in any region without current. The fairly common belief that there can be magnetostatic field energy in regions without current is wrong. Our conclusion is that ${\bf B}^{2}/8\pi$ cannot be considered a magnetostatic energy density. ## 4 Magnetostatic Energy of a Permanent Magnetic Dipole The discussion above was for a current distribution held constant in time by a constant current source that provided energy. For strictly classical physics, if magnetism is assumed to come only from such currents, this would be a complete description. However elementary particles (leptons and quarks) have permanent point magnetic dipole moments that arise from relativistic quantum mechanics. These permanent magnetic moments do not arise from electric currents, and remain constant without the influence of an external source. Atomic and nuclear states with quantized spin and orbital angular momentum will also have permanent magnetic dipole moments. Then, a macroscopic object, such as a permanent magnet composed of polarized atoms locked in place, can have a permanent magnetic dipole moment. The force on a permanent magnetic dipole would be expected to be the same as for a dipole connected to a constant current source, and so be given by ${\bf F=\mbox{\boldmath$\nabla$}(\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B)}.$ (44) This also agrees with experiment for the force on a permanent magnetic dipole. In fact, Eq. (44) should be considered a separate force law for permanent magnetic dipoles on a par with the Lorentz force on charge or the $\bf j\mbox{\boldmath$\times$}B$ force on currents, since it cannot be derived from those force laws. As was the case for the electric dipole force, the magnetic force on a permanent magnetic dipole should equal the negative gradient of the magnetic energy of the dipole. Thus the energy of a permanent magnetic dipole in an external magnetic field $\bf B_{0}$ will be given by $U_{\mbox{\boldmath$\mu$}}=-{\bf\mbox{\boldmath$\mu$}\mbox{\boldmath$\cdot$}B_{0}}.$ (45) This has the opposite sign from the energy expression for a classical current loop but, as we have seen, that leads to the same force for each. Since the permanent magnetic dipole is not related to a classical current, we cannot derive any expression for the energy purely in terms of the fields, as in Eq. (27). The energy of a magnetic dipole in a magnetic field is given by Eq. (45). There is no consistent definition of energy within the magnetic dipole, even if it is finite in size. This is because of the inelasticity of the process of its magnetization. ## 5 Electromagnetic Energy We now derive general forms for the electromagnetic (EM) energy for charge and current densities, and electric and magnetic fields that may be time varying. In this section, we calculate the EM energy in the presence of polarizable material. We start with the power input to matter by an electromagnet field in a volume $V$, given by Eq. (11) $\displaystyle\frac{dU_{\rm matter}}{dt}$ $\displaystyle=$ $\displaystyle\int_{V}{\bf j\mbox{\boldmath$\cdot$}E}d^{3}r$ (46) $\displaystyle=$ $\displaystyle-\int_{V}{\bf j}\mbox{\boldmath$\cdot$}\left(\mbox{\boldmath$\nabla$}\phi+\frac{1}{c}\partial_{t}{\bf A}\right)d^{3}r$ $\displaystyle=$ $\displaystyle-\int_{V}\left[\mbox{\boldmath$\nabla\cdot$}({\bf j}\phi)-\phi(\mbox{\boldmath$\nabla\cdot$}{\bf j})+\frac{1}{c}{\bf j}\mbox{\boldmath$\cdot$}\partial_{t}{\bf A}\right]d^{3}r$ $\displaystyle=$ $\displaystyle-\oint_{S}{\bf dS\mbox{\boldmath$\cdot$}j\phi}-\int_{V}\left[\phi\left(\partial_{t}\rho\right)+\frac{1}{c}{\bf j}\mbox{\boldmath$\cdot$}\left(\partial_{t}{\bf A}\right)\right]d^{3}r$ $\displaystyle=$ $\displaystyle-\oint_{S}{\bf dS\mbox{\boldmath$\cdot$}j\phi}-\frac{1}{2}\int_{V}\left[\partial_{t}\left(\rho\phi\right)+\frac{1}{c}\partial_{t}\left({\bf j\mbox{\boldmath$\cdot$}\bf A}\right)\right]d^{3}r.$ The last step above assumes linear media so that $\phi\left(\partial_{t}\rho\right)=\rho\left(\partial_{t}\phi\right)=\frac{1}{2}\partial_{t}\left(\rho\phi\right)\quad{\rm and}\quad{\bf j}\mbox{\boldmath$\cdot$}\left(\partial_{t}{\bf A}\right)={\bf A}\mbox{\boldmath$\cdot$}\left(\partial_{t}{\bf j}\right)=\frac{1}{2}\partial_{t}\left({\bf j\mbox{\boldmath$\cdot$}\bf A}\right),$ (47) which hold if $\phi$ and $\bf A$ satsfy Poisson’s equation. These relations also hold for non-isotropic media as long as the tensors of permittivity and permeability are symmetric. The surface integral in Eq. (46) will vanish for contained current distributions, provided that the surface integral encloses the entire current distribution. It will also vanish for infinitely long wires in an infinite volume if the scalar potential $\phi$ vanishes at infinite radius fast enough. Then, conservation of the sum of the energy in matter and electromagnetic energy implies that the EM energy in the volume $V$ is given by $U_{\rm EM}=\frac{1}{2}\int_{V}\left(\rho\phi+\frac{1}{c}{\bf j\mbox{\boldmath$\cdot$}\bf A}\right)d^{3}r.$ (48) In a finite volume which does not encompass all of the current, the energy carried out of the volume by currents in the surface integral in Eq. (46) would have to be considered. We see that, in terms of time varying charge-current distributions and potentials, the expression for the EM energy is just the sum of the electrostatic and magnetostatic energies. Note that the presence of polarizable matter never entered the above derivation. Polarized matter does affect the EM energy through its affect on the scalar and vector potentials, but it does not affect the form of the energy density. The EM energy in a volume $V$ can be put in terms of the EM fields using Maxwell’s equations: $\displaystyle U_{\rm EM}$ $\displaystyle=$ $\displaystyle\frac{1}{2}\int_{V}\left(\rho\phi+\frac{1}{c}{\bf j\mbox{\boldmath$\cdot$}\bf A}\right)d^{3}r$ (49) $\displaystyle=$ $\displaystyle\frac{1}{8\pi}\int_{V}\left[\phi(\mbox{\boldmath$\nabla\cdot$}{\bf D})+{\bf A}\mbox{\boldmath$\cdot$}\left(\mbox{\boldmath$\nabla\times$}{\bf H}-\frac{1}{c}\partial_{t}{\bf D}\right)\right]d^{3}r$ $\displaystyle=$ $\displaystyle\frac{1}{8\pi}\int_{V}\left[{\bf\mbox{\boldmath$\nabla\cdot$}(\phi D)-D\mbox{\boldmath$\cdot$}\mbox{\boldmath$\nabla$}\phi+\mbox{\boldmath$\nabla\cdot$}(H\mbox{\boldmath$\times$}A)+H\mbox{\boldmath$\cdot$}(\mbox{\boldmath$\nabla\times$}A)}-\frac{1}{c}{\bf A}\mbox{\boldmath$\cdot$}(\partial_{t}{\bf D})\right]d^{3}r$ $\displaystyle=$ $\displaystyle\frac{1}{8\pi}\oint_{S}{\bf dS\mbox{\boldmath$\cdot$}(\phi D+H\mbox{\boldmath$\times$}A})+\frac{1}{8\pi}\int_{V}({\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H})d^{3}r$ $\displaystyle+\frac{1}{8\pi c}\int_{V}\left({\bf D}\mbox{\boldmath$\cdot$}\partial_{t}{\bf A}-{\bf A}\mbox{\boldmath$\cdot$}\partial_{t}{\bf D}\right)d^{3}r.$ The time derivative integral above is new. It vanishes for time constant fields, but the contribution of that integral and of the surface integral would have to be included in those cases for which they didn’t vanish. In any event, a correct evaluation of these last three integrals would have to reproduce the starting point, that is $U_{\rm EM}$ as given by Eq. (48). The commonly used relation $U_{\bf EB}=\frac{1}{8\pi}\int_{V}({\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H})d^{3}r$ (50) only equals the EM energy when the surface and time derivative integrals in Eq. (49) vanish. We look again at Eq. (12), rewritten for polarizable matter as $\frac{dU_{\rm Matter}}{dt}+\frac{1}{8\pi}\int_{V}\partial_{t}({\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H})d^{3}r=-\frac{c}{4\pi}\oint_{S}{\bf dS\mbox{\boldmath$\cdot$}(E\mbox{\boldmath$\times$}H)},$ (51) with the Poynting vector for EM energy flow given as $\mbox{\boldmath$\cal P$}=\frac{c}{4\pi}{\bf(E\mbox{\boldmath$\times$}H)}$ (52) Although Eq. (51) has been used to identify $({\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H})/8\pi$ as an EM energy density, we see that the Poynting vector is not related directly to the EM energy, but only to its time derivative. If Eq. (51) is integrated over time starting with no EM energy, then some of the energy put in by matter can escape through the surface due to the surface integral of the Poynting vector. This means that, just as in the static cases, if we try to use $({\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H})/8\pi$ as the EM energy density, its integral may not give the total energy put into the volume. This is why the additional integrals in Eq. (49) have to be included to get the total energy put into the volume. For integrals over finite volumes our conclusion is the same as in the static cases. Physical EM energy can only exist within a volume if the volume contains charge or current. There can be no EM field energy in a volume that does not contain charge or current. This means that the quantity $({\bf E}^{2}+{\bf B}^{2})/8\pi$ cannot be considered an EM energy density. If EM energy resides anywhere, it must be within charge or current distributions. The point of view is sometimes taken that $({\bf E}^{2}+{\bf B}^{2})/8\pi$ and $\frac{1}{2}(\rho\phi+{\bf j\mbox{\boldmath$\cdot$}A})$ are alternate expressions for the EM energy density, and either one can be chosen. This could work if both were considered as mathematical constructs with no direct physical significance. But, if the EM energy within a finite volume is an actual physical quantity, only one of the two choices can be applicable. A physical energy cannot have different locations, depending on our whim. Our preceding discussions for the static cases and the derivation of Eq. (46) show that only $\frac{1}{2}(\rho\phi+{\bf j\mbox{\boldmath$\cdot$}A})$ can be considered a physical EM energy density. What about the common use of the Poynting vector $\cal P$ of Eq. (52) to represent the intensity of an EM wave? To understand that, we have to enter a somewhat philosophical discussion about the difference between mathematically useful quantities and physical quantities. The use over many years of the Poynting vector to calculate angular distributions of the emission and scattering of EM radiation has given correct predictions for the expected detection of the radiation. However, the interpretation that the radiation passing through the bounding surface of a volume is given differentially by $\mbox{\boldmath$\cal P$}\mbox{\boldmath$\cdot$}{\bf dS}$ seems inconsistent with our discussion above. That is, we have shown that EM energy cannot leave a volume unless there is a charge or current distribution outside the volume. This means the energy cannot actually appear outside the volume until an antenna (or retina) detects the radiation by means of induced currents. The Poynting vector correctly gives the probability of radiated energy being observed in a particular direction, but the energy does not leave the volume until it registers in a detector. The situation is similar to that in quantum mechanics, 333The similarity between Classical Electromagnetism and Quantum Wave Mechanics is discussed in Sec. 16.4 of Ref. [4]. where the square of the wave function gives the probability of a particle interacting at a particular position, but the particle is not physically there until it is observed. In either case, the Poynting vector or the wave function (or probability current) is a useful mathematical construction, but it does not give the physical position of anything until the wave interacts with something. ## 6 Conclusion Our conclusion, from the specific and general cases studied above, is that electromagnetic energy can exist only within matter, and there is no EM energy density in empty space. The integral of $(\bf E\mbox{\boldmath$\cdot$}D+B\mbox{\boldmath$\cdot$}H)/8\pi$ only gives the EM energy within a volume if the surface and time derivative integrals in Eq. (49) vanish, in which case it gives the same result as the integral of $\frac{1}{2}(\rho\phi+{\bf j\mbox{\boldmath$\cdot$}A})$. This means the only valid choice for an EM energy density is $\frac{1}{2}(\rho\phi+{\bf j\mbox{\boldmath$\cdot$}A})$. As discussed above, the Poynting vector correctly gives the probability of radiated energy being observed in a particular direction, but the energy does not leave the volume until it registers in a detector. ## References * [1] J. C. Maxwell, Royal Society Transactions, Vol. CLV, XXV (1864). * [2] J. D. Jackson, Classical Electrodynamics 2nd Edn, (John Wiley & Sons, New York) (1975). * [3] D. J. Griffiths,1999 Introduction to Electrodynamics 3rd Edn. (Addison Wesley, San Francisco) (1999). * [4] J. Franklin, Classical Electromagnetism (Addison Wesley, San Francisco) (2005).

arxiv-papers

2024-09-04T02:48:53.029125

0707.4025

††thanks: Present address: School of Mathematics and Physics, Queen’s University, Belfast BT7 1NN, UK. E-mail address: adanehkar01@qub.ac.uk # Consistent Interactions of Yang-Mills Theory: A Review Ashkbiz Danehkar danehkar@central.ucv.ro Faculty of Physics, University of Craiova, 13 A. I. Cuza Str., Craiova 200585, Romania ###### Abstract We reconsider the interacting theory of the Yang-Mills model in the Lagrangian form. We obtain all consistent interactions through deformations of the master equation in the antifield formalism. The results determine deformed structures of the gauge transformation. ###### pacs: 11.10.Ef, 11.15.–q ## I Introduction Dirac’s workbib3a ; bib3b ; bib3c was a pioneer in constrained systems, which were used for several applications in quantum field theorybib7 ; bib8 ; bib26 . In this approach, we make the action in either Lagrangian or Hamiltonian forms bib9a ; bib9b , while they are equivalent to each otherbib10 . In this way, we can obtain the Hamiltonian quantization through canonical variables such as coordinate and momentum, which involve constrained dynamicsbib16 ; bib15 ; bib13 ; bib18 ; bib17 ; bib14 . The physical variables of a constrained system have gauge invariance and locally independent symmetry. Gauge transformations introduce some arbitrary time independent functions to the Hamilton’s equations of motion. We can see that the canonical variables are not all independent. Hence, we need to imply some conditions for canonical variables: the first-class and second-class constraints. We must also generalize the frame to include both bosonic and fermionic variables. BRST approach bib19a ; bib19b ; bib19c ; bib20 extended the local gauge symmetries in terms of BRST differential and co-/homological classes. It emerged as a replacement for the original gauge symmetry. We can construct the gauge symmetry from a nilpotent derivation. We see that the gauge action is invariant under a nilpotent symmetry, called the BRST symmetry. Replacing the gauge symmetry with the BRST symmetry introduces antifield, ghosts, and antighosts to each gauge variable bib21a ; bib21b . It presents the framework for solutions to the equations of motion bib24a ; bib24b . BRST cohomology also received some important extensions from the antifield formalism bib23a ; bib23b ; bib23c ; bib24a ; bib24b ; bib1 ; bib34a ; bib34b . Using the antifield formalism, we can construct consistent interactions among fields from coupling deformations of the master equation bib31 ; bib32 . In this paper, we reconsider the construction of consistent interactions of the Yang–Mills theory through all coupling deformations of the master equation. We can see that deformations stop at second-order. The resulting action identifies a deformed transformation as gauge symmetry and provides a commutator for it. The organization of this paper is as follows. Section II introduces the BRST differential and the antifield formalism. We consider the coupling deformations of the master equations in the antifield formalism. In § III, we describe the BRST transformation of the Yang–Mills theory. We obtain the deformation of the master equation and analyze its several orders. Eventually, we realize the entire gauge structure of the Yang-Mills theory. Section IV provides a conclusion. ## II BRST formalism The BRST differential $s$ is split into the Koszul-Tate resolution $\delta$ and the exterior derivative $\gamma$ along the gauge orbits bib24a ; bib24b : $s=\delta+\gamma.$ (1) The Koszul-Tate differential maintains the equation of motion (Euler-Lagrange equation). For any $X$ and $Y$ with Grassmann parity $\varepsilon_{X}$ and $\varepsilon_{Y}$, we have: $s(XY)=X(sY)+(-1)^{\varepsilon_{Y}}(sX)Y.$ (2) The BRST differential $s$ is a nilpotent derivation: $s^{2}=\delta^{2}=\gamma^{2}=0.$ (3) Moreover, the Koszul-Tate resolution $\delta$ commutes with $\gamma$: $\gamma\delta+\delta\gamma=0.$ (4) We denote the grading degree of $s$ by the ghost number (${\mathfrak{gh}}$), being equal to one: $\mathfrak{gh}{(s)}=\mathfrak{gh}{(\delta)}=\mathfrak{gh}{(\gamma)}=1,$ (5) with the following property $\mathfrak{gh}(XY)=\mathfrak{gh}(X)+\mathfrak{gh}(Y).$ (6) The ghost number consists of the pureghost number ($\mathfrak{pgh}$) and the antighost number ($\mathfrak{agh}$): $\mathfrak{gh}(X)=\mathfrak{pgh}(X)-\mathfrak{agh}(X).$ (7) We impose that $\begin{array}[]{cccc}{\mathfrak{pgh}(\delta)=0,}&{\mathfrak{agh}(\delta)=-1,}&{\mathfrak{pgh}(\gamma)=1,}&{\mathfrak{agh}(\gamma)=0.}\end{array}$ (8) This means that the differential $\delta$ reduces the antighost number by one, while it does not change the pureghost number. ### II.1 Antifield formalism Let us consider the Lagrangian action $S_{0}^{\mathrm{L}}[{\phi^{\alpha_{0}}]=}\int d^{4}x\mathcal{L}_{0}\left({\phi^{\alpha_{0}},\partial}_{\mu}{\phi}^{\alpha_{0}}\right).$ (9) It reads the equations of motion ${\delta}S_{0}^{\mathrm{L}}/{\delta\phi}^{\alpha_{0}}=0$. This has gauge symmetries as $\delta_{\varepsilon}\phi^{\alpha_{0}}=Z_{\alpha_{1}}^{\alpha_{0}}\varepsilon^{\alpha_{1}}$, where $Z_{\alpha_{1}}^{\alpha_{0}}$ is a structure of the gauge group. The field $\phi^{\alpha_{0}}$ with ghost number zero may imply ghost $C^{\alpha_{1}}$ with ghost number one, as well as the ghosts of ghost $C^{\alpha_{2}}$ with ghost number two, etc: $\begin{array}[]{ccc}C{{}^{A}=}\left\\{C^{\alpha_{1}}{,\ldots,}C^{\alpha_{k}}\right\\},&{\mathfrak{gh}(C^{\alpha_{k}})=k,}&{\varepsilon(C^{\alpha_{k}})=k\mathrm{~{}(mod~{}2)}.}\end{array}$ (10) We introduce antifield $\phi_{\alpha_{0}}^{\ast}$ and antighosts $C_{A}^{\ast}{=}\left\\{{C_{\alpha_{1}}^{\ast},\ldots,C_{\alpha_{k}}^{\ast}}\right\\}$ of opposite Grassmann parity: $\begin{array}[]{cc}{\mathfrak{gh}(\phi_{\alpha_{0}}^{\ast})=-\mathfrak{gh}(\phi^{\alpha_{0}})-1,}&{\varepsilon(\phi_{\alpha_{0}}^{\ast})=\varepsilon(\phi^{\alpha_{0}})+1\mathrm{~{}(mod~{}2),}}\end{array}$ (11) $\begin{array}[]{cc}{\mathfrak{gh}(C_{\alpha_{k}}^{\ast})=-(k+1),}&{\varepsilon(C_{\alpha_{k}}^{\ast})=k+1\mathrm{~{}(mod~{}2)}.}\end{array}$ (12) Therefore, we can define the gauge variables as $\begin{array}[]{cc}{\Phi^{A}=\left\\{{\phi^{\alpha_{0}},C^{A}}\right\\},}&{\Phi_{A}^{\ast}=\left\\{{\phi}_{\alpha_{0}}^{\ast}{,}C_{A}^{\ast}\right\\},}\end{array}$ (13) where ${\Phi}^{A}$ is a set of fields including the original field, the ghost, and the ghosts of ghosts, ${\Phi_{A}^{\ast}}$ provides the antifields definition. The action of the BRST differential $\bar{s}$ admits an antifield formalism $\bar{s}\cdot=(\cdot,S)$, where $S$ stands for its generator and $(,)$ is the antibracket defined in the space of fields ${\Phi^{A}}$ and antifields ${\Phi_{A}^{\ast}}$ by bib34a ; bib34b $(X,Y)\equiv\frac{{\partial_{r}X}}{{\partial\Phi^{A}}}\frac{{\partial_{l}Y}}{{\partial\Phi_{A}^{\ast}}}-\frac{{\partial_{r}X}}{{\partial\Phi_{A}^{\ast}}}\frac{{\partial_{l}Y}}{{\partial\Phi^{A}}}.$ (14) The nilpotent expression $\bar{s}^{2}=0$ becomes equivalent to the master equation $(S,S)=0$. ### II.2 Consistent interactions We may construct a consistent interaction from $S[{\Phi}^{A},{\Phi}_{A}^{\ast}]$ in a deformed solution in powers of the coupling constant $g$ bib31 : $\displaystyle S\rightarrow\bar{S}$ $\displaystyle=S_{0}+gS_{1}+g^{2}S_{2}+g^{3}S_{3}\cdots$ $\displaystyle=S+g\int d^{4}x\,a+g^{2}\int d^{4}x\,b+g^{3}\int d^{4}x\,c+\cdots$ (15) of the master equation for the interacting theory $\left(\bar{S},\bar{S}\right)=0.$ (16) On substituting Eq. (15) into the master equation (16), we obtain the deformations of the master equations: $\displaystyle(S_{0},S_{0})$ $\displaystyle=$ $\displaystyle 0,$ (17) $\displaystyle 2(S_{0},S_{1})$ $\displaystyle=$ $\displaystyle 0,$ (18) $\displaystyle 2(S_{0},S_{2})+(S_{1},S_{1})$ $\displaystyle=$ $\displaystyle 0,$ (19) $\displaystyle(S_{0},S_{3})+(S_{1},S_{2})$ $\displaystyle=$ $\displaystyle 0,$ $\displaystyle\vdots\ \ \ $ while $S_{0}$ reads as $S_{0}={S_{0}^{\mathrm{L}}}+\phi_{\alpha_{0}}^{\ast}Z_{\alpha_{1}}^{\alpha_{0}}C^{\alpha_{1}}+\ldots,$ (21) where ${S_{0}^{\mathrm{L}}}=S_{0}[{\Phi}^{A},{\Phi}_{A}^{\ast}=0]$ is a free action. We define the BRST differential $s$ of the field theory by $s\cdot=(\cdot,S_{0})$. Using the last definition, Eq. (17)–(II.2) are rewritten as: $\displaystyle s^{2}$ $\displaystyle=$ $\displaystyle 0,$ (22) $\displaystyle 2sS_{1}$ $\displaystyle=$ $\displaystyle 0,$ (23) $\displaystyle 2sS_{2}+(S_{1},S_{1})$ $\displaystyle=$ $\displaystyle 0,$ (24) $\displaystyle sS_{3}+(S_{1},S_{2})$ $\displaystyle=$ $\displaystyle 0,$ $\displaystyle\vdots\ \ \ $ We get all deformations of the master equation in the field theory. ## III Deformations of Master Equation We consider the Yang-Mills Lagrangian action involving a set of massless fields $A_{\mu}^{a}$ as: ${S_{0}^{\mathrm{L}}}[A_{\mu}^{a}]=-{{\textstyle\frac{1}{4}}}\int d^{4}xF_{\mu\nu}^{a}F_{a}^{\mu\nu},$ (26) where $F_{\mu\nu}^{a}$ is the abelian field strengths defined by $\begin{array}[]{cc}F_{\mu\nu}^{a}\equiv\partial_{\mu}A_{\nu}^{a}-\partial_{\nu}A_{\mu}^{a},&F_{a}^{\mu\nu}=\eta^{\mu\alpha}\eta^{\nu\beta}\delta_{ab}F_{\alpha\beta}^{b},\end{array}$ (27) and $\eta^{\mu\nu}$ is the $\mathrm{SO}(1,3)$ invariant flat metric in Minkowski space, and $\delta_{ab}$ is a given symmetric invertible matrix with the following properties $\begin{array}[]{cc}\delta_{ab}=\delta_{(ab)},&\delta^{ab}\delta_{bc}=\delta^{a}{}_{c}.\end{array}$ (28) The equation of motion of (26) reads $\frac{\delta{S_{0}^{\mathrm{L}}}}{\delta A_{\mu}^{a}}=\partial_{\nu}F_{a}^{\nu\mu}=0.$ (29) This reveals an irreducible transformation: $\delta_{\varepsilon}A_{\mu}^{a}=\partial_{\mu}\varepsilon^{a}.$ (30) The action (26) is close according to an abelian algebra, and invariant under the gauge transformations (30). The BRST transformation provides ghosts $C^{a}$, antifields $A_{a}^{\ast\mu}$ and antighosts $C_{a}^{\ast}$ bib21a ; bib21b : $\textstyle{\varepsilon=0}$$\textstyle{\mathop{A^{a}_{\mu}}\limits_{{\rm{gh=0}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\delta_{\varepsilon}}$$\textstyle{\mathop{\partial_{\mu}\varepsilon^{a}}\limits_{~{}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{\varepsilon=1}$$\textstyle{\mathop{A_{a}^{*\mu}}\limits_{{\rm{gh=-1}}}}$$\textstyle{\mathop{C^{a}}\limits_{{\rm{gh=1}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{\varepsilon=0}$$\textstyle{\mathop{C_{a}^{*}}\limits_{{\rm{gh=-2}}}}$ (31) The BRST differential $s$ consisting of $\delta$ and $\gamma$ acts on $A_{\mu}^{a}$, $A_{a}^{\ast\mu}$, $\rule{0.0pt}{12.91663pt}C^{a}$, and $C_{a}^{\ast}$: $\begin{array}[]{cccc}\delta A_{\mu}^{a}=0,&\delta A_{a}^{\ast\mu}=-\displaystyle\frac{\delta{S_{0}^{\mathrm{L}}}}{\delta A_{\mu}^{a}}=-\partial_{\nu}F_{a}^{\nu\mu},&\delta\rule{0.0pt}{12.91663pt}C^{a}=0,&\delta C_{a}^{\ast}=-\partial_{\mu}A_{a}^{\ast\mu},\end{array}$ (32) $\begin{array}[]{cccc}\gamma A_{\mu}^{a}=\partial_{\mu}C^{a},&\gamma A_{a}^{\ast\mu}=0,&\gamma C^{a}=0,&\gamma C_{a}^{\ast}=0.\end{array}$ (33) The classical master equation of the action (26) has the minimal solution by $S_{0}={S_{0}^{\mathrm{L}}}[A_{\mu}^{a}]+\int d^{4}xA_{a}^{\ast\mu}\partial_{\mu}C^{a}.$ (34) We will consider deformed solutions to the master equation (16) of the action (26). This comes to the minimal solution (34), when the coupling constant $g$ vanishes. ### III.1 First-order deformation We recognized that the first-order deformation satisfies (23). Here $S_{1}$ is bosonic function with ghost number zero. We assume $S_{1}=\int d^{4}x\,a,$ (35) where $a$ is a local function. The first-order deformation takes the local form $\begin{array}[]{ccc}sa=(\delta a+\gamma a)=\partial_{\mu}j^{\mu},&{\mathfrak{gh}}\left(a\right)=0,&\varepsilon\left(a\right)=0,\end{array}$ (36) where local current $j^{\mu}$ shows the nonintegrated density of the first- order deformation related to the local cohomology of $s$ at ghost number zero. To evaluate (36), we assume $\begin{array}[]{ccccc}a=\sum\limits_{i=1}^{I}a_{i},&{\mathfrak{agh}}\left(a_{i}\right)=i,&{\mathfrak{gh}}\left(a_{i}\right)=0,&\varepsilon\left(a_{i}\right)=0,&\forall i=1,\ldots,I,\end{array}$ (37) $\begin{array}[]{cccc}j^{\mu}=\sum\limits_{i=1}^{I}\overset{(i)}{{j}^{\mu}},&{\mathfrak{agh}}(\overset{(i)}{{j}^{\mu}})=i,&{\mathfrak{gh}}(\overset{(i)}{{j}^{\mu}})=0,&\varepsilon(\overset{(i)}{{j}^{\mu}})=0.\end{array}$ (38) where $\overset{(k)}{{j}^{\mu}}$ are some local currents. On substituting (37) and (38) into (36), we get $\begin{array}[]{ccc}\sum\limits_{i=1}^{I}\delta a_{i}+\sum\limits_{i=1}^{I}\gamma a_{i}=\sum\limits_{i=1}^{I}\overset{(i)}{\partial_{\mu}{j}^{\mu}},&{\mathfrak{agh}}(\delta a_{i})=i-1,&{\mathfrak{agh}}(\gamma a_{i})=i.\end{array}$ (39) It is decomposed into a number of antighost: $\begin{array}[]{c|cc}{\mathfrak{agh}}(Z)&Z&\\\ \hline\cr I&\gamma a_{I}=\partial_{\mu}\overset{(I)}{{j}^{\mu}},&\\\ I-1&\delta a_{I}+\gamma a_{I-1}=\partial_{\mu}\overset{(I-1)}{{j}^{\mu}},&\\\ k&\delta a_{k+1}+\gamma a_{k}=\partial_{\mu}\overset{(k)}{{j}^{\mu}},&\quad k=0,\ldots,I-2\end{array}$ (40) Although we strictly impose the first expression in (40) with positive antighost numbers vanishes: $\begin{array}[]{ccc}\gamma a_{I}=0;&I>0,&a_{I}\in H^{I}(\gamma).\end{array}$ (41) where $H^{I}(\gamma)$ is the local cohomology of $\gamma$ with pureghost number $I$. Term $a_{I}$ can exclusively reduce to $\gamma$-exact terms $a_{I}=\gamma b_{I}$ related to a trivial definition, that states $a_{I}=0$. This is plainly given by the second-order nilpotency of $\gamma$, which implies the unique solution for (41) up to $\gamma$-exact contributions, i. e. $a_{I}\rightarrow a_{I}+\gamma b_{I},$ (42) $\begin{array}[]{ccc}{\mathfrak{agh}}\left(b_{I}\right)=I,&{\mathfrak{pgh}}\left(b_{I}\right)=I-1,&\varepsilon\left(b_{I}\right)=1.\end{array}$ (43) So, the nontriviality of the first-order deformation $a_{I}$ purposes the cohomology of the longitudinal differential $\gamma$ at pureghost number equal to $I$, i. e. $a_{I}\in H^{I}(\gamma)$. To solve (40), we need to provide $H^{I}(\gamma)$ and $H_{I}\left(\delta|d\right)$: $\begin{array}[]{cc}\delta a_{I}=\partial_{\mu}m_{I}^{\mu},&a_{I}\in H_{I}(\delta|d).\end{array}$ (44) where $H_{I}\left(\delta|d\right)$ is the local homology of the Koszul-Tate differential $\delta$ with antighost number $I$. For an irreducible situation, where gauge generators are field independent, we assume $H_{I}(\delta|d)=0,\quad I>2.$ (45) We then obtain $\displaystyle\gamma a_{2}$ $\displaystyle=$ $\displaystyle 0,$ (46) $\displaystyle\delta a_{2}+\gamma a_{1}$ $\displaystyle=$ $\displaystyle\partial_{\mu}\overset{(1)}{{j}^{\mu}},$ (47) $\displaystyle\delta a_{1}+\gamma a_{0}$ $\displaystyle=$ $\displaystyle\partial_{\mu}\overset{(0)}{{j}^{\mu}}.$ (48) This affords the first-order deformation as follows $a=a_{0}+a_{1}+a_{2}.$ (49) Let us consider (32) and (33). The local cohomology of $\gamma$ at pureghost number one has a ghost $C^{a}$, while pureghost number two shows two ghosts $C^{a}C^{b}$, i. e. $\\{C^{a}\\}\in H^{1}(\gamma)$ and $\\{C^{a}C^{b}\\}\in H^{2}(\gamma)$. From here, we solve (46): $a_{2}={{\textstyle\frac{1}{2}}}C_{a}^{\ast}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}\rule{0.0pt}{12.91663pt}C^{c},$ (50) where $f_{bc}^{a}\rule{0.0pt}{12.91663pt}$ are the structure constants and antisymmetric on indices $bc$: $\begin{array}[]{ccc}f_{bc}^{a}=f_{[bc]}^{a},&f_{mbc}=\delta_{am}f_{bc}^{a},&f_{mbc}=-f_{bmc}.\end{array}$ (51) We solve (47) by taking $\delta$ from (50): $\delta a_{2}=-{{\textstyle\frac{1}{2}}}\partial_{\mu}(A_{a}^{\ast\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}\rule{0.0pt}{12.91663pt}C^{c})+\gamma(A_{a}^{\ast\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}A_{\mu}^{c})$ This provides $\begin{array}[]{cc}a_{1}=-A_{a}^{\ast\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}A_{\mu}^{c},&\overset{(1)}{{j}^{\mu}}=-\frac{1}{2}A_{a}^{\ast\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}\rule{0.0pt}{12.91663pt}C^{c}.\end{array}$ (52) We also solve (48) by taking $\delta$ from $a_{1}$: $\delta a_{1}=\partial_{\nu}(-F_{a}^{\nu\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}A_{\mu}^{c})+\gamma({{\textstyle\frac{1}{2}}}F_{a}^{\nu\mu}f_{bc}^{a}A_{\mu}^{b}A_{\mu}^{c})+{{\textstyle\frac{1}{2}}}F_{a}^{\nu\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}F_{\nu\mu}^{c}.$ (53) The last term in (53) vanishes due to antisymmetric property (51c). We obtain $\delta a_{1}-\gamma({{\textstyle\frac{1}{2}}}F_{a}^{\nu\mu}f_{bc}^{a}A_{\mu}^{b}A_{\mu}^{c})=\partial_{\nu}(-F_{a}^{\nu\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}A_{\mu}^{c}).$ (54) This shows $\begin{array}[]{cc}a_{0}=-{{\textstyle\frac{1}{2}}}F_{a}^{\nu\mu}f_{bc}^{a}A_{\mu}^{b}A_{\mu}^{c},&\overset{(0)}{{j}^{\mu}}=-F_{a}^{\nu\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}A_{\mu}^{c}.\end{array}$ (55) Therefore, we get the first-order deformation up to antighost number two: $S_{1}=-{{\textstyle\frac{1}{2}}}\int d^{4}x\,\left(F_{a}^{\nu\mu}f_{bc}^{a}A_{\mu}^{b}A_{\mu}^{c}+2A_{a}^{\ast\mu}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}A_{\mu}^{c}-C_{a}^{\ast}f_{bc}^{a}\rule{0.0pt}{12.91663pt}C^{b}\rule{0.0pt}{12.91663pt}C^{c}\right).$ (56) Here the gauge generators are field independent, and are reduced to a sum of terms with antighost numbers from zero to two. ### III.2 Higher-order deformations We now solve the second-order deformation of the master equation, (24). We shall assume $S_{2}=\int d^{4}x\,b,$ (57) which takes the local form $\Delta+2sb=\partial_{\mu}m^{\mu}.$ (58) We shall use (56) to compute $(S_{1},S_{1})\equiv\int d^{4}x\,\Delta.$ This provides the following results: $\Delta=\Delta_{0}+\Delta_{1}+\Delta_{2},$ (59) namely, $\Delta_{0}\equiv- f_{e[m}^{a}f_{np]}^{e}F_{a}^{\alpha\beta}A_{\alpha}^{m}A_{\beta}^{n}C^{p}+2f_{bcm}f_{np}^{m}\eta^{\alpha\rho}\eta^{\beta\mu}(\partial_{\rho}C^{b})A_{\mu}^{c}A_{\alpha}^{n}A_{\beta}^{p}.$ (60) $\Delta_{1}\equiv- f_{e[m}^{a}f_{np]}^{e}A_{a}^{\ast\mu}C^{m}C^{n}A_{\mu}^{p},$ (61) $\Delta_{2}\equiv-{{\textstyle\frac{1}{3}}}f_{e[m}^{a}f_{np]}^{e}C_{a}^{\ast}C^{m}C^{n}C^{p}.$ (62) If we define $b\equiv b_{0}+b_{1}+b_{2},$ (63) we get a set of equations $\displaystyle\Delta_{2}+2\gamma b_{2}$ $\displaystyle=$ $\displaystyle\partial_{\mu}\overset{(2)}{{m}^{\mu}},$ (64) $\displaystyle\Delta_{1}+\delta b_{2}+2\gamma b_{1}$ $\displaystyle=$ $\displaystyle\partial_{\mu}\overset{(1)}{{m}^{\mu}},$ (65) $\displaystyle\Delta_{0}+\delta b_{1}+2\gamma b_{0}$ $\displaystyle=$ $\displaystyle\partial_{\mu}\overset{(0)}{{m}^{\mu}}.$ (66) Eqs. (62) and (64) lead to $\begin{array}[]{cc}\Delta_{2}=0,&b_{2}=0.\end{array}$ (67) This implies the Jacobi identity: $f_{e[m}^{a}f_{np]}^{e}=0.$ (68) We also derive $\begin{array}[]{cc}\Delta_{1}=0,&b_{1}=0.\end{array}$ (69) Equation (66) gives $2f_{bc}^{a}k_{ma}f_{np}^{m}\eta^{\alpha\rho}\eta^{\beta\mu}(\partial_{\rho}C^{b})A_{\mu}^{c}A_{\alpha}^{n}A_{\beta}^{p}+2\gamma b_{0}=\partial_{\mu}\overset{(0)}{{m}^{\mu}}.$ (70) We can solve it by substituting $\gamma$ of vector fields $A_{\mu}^{a}$ into $\partial_{\mu}C^{a}$: $2f_{bc}^{a}k_{ma}f_{np}^{m}\eta^{\alpha\rho}\eta^{\beta\mu}(\partial_{\rho}C^{b})A_{\mu}^{c}A_{\alpha}^{n}A_{\beta}^{p}=\gamma\left(-{{\textstyle\frac{1}{2}}}f_{bcm}f_{np}^{m}\eta^{\alpha\rho}\eta^{\beta\mu}A_{\rho}^{b}A_{\mu}^{c}A_{\alpha}^{n}A_{\beta}^{p}\right).$ It provides $b_{0}=-{{\textstyle\frac{1}{4}}}f_{bcm}f_{np}^{m}\eta^{\alpha\rho}\eta^{\beta\mu}A_{\rho}^{b}A_{\mu}^{c}A_{\alpha}^{n}A_{\beta}^{p}.$ We accordingly obtain the second-order deformation $S_{2}=-{{\textstyle\frac{1}{4}}}\int d^{4}x\,f_{bcm}f_{np}^{m}\eta^{\alpha\rho}\eta^{\beta\mu}A_{\rho}^{b}A_{\mu}^{c}A_{\alpha}^{n}A_{\beta}^{p}.$ (71) The Jacobi identity (68) shows $(S_{1},S_{2})=0\rightarrow S_{3}=0.$ We then find out that all orders higher than second shall vanish: $S_{k}=0,\ \forall k\geqslant 3.$ We solve the Yang-Mills action by the first- and second-order deformations: ${S}={S}_{0}+g{S}_{1}+g^{2}{S}_{2}.$ (72) This includes the gauge structures decomposed into terms with antighost number from zero to two. We can see the part with antighost number zero in the Lagrangian forms. The antighost number one corresponds to the gauge generators. Higher antighost numbers shows the reducibility functions appearing in the ghosts of ghosts. All functions with order higher than second will vanish. ### III.3 Lagrangian and gauge structure Setting ${\Phi}_{A}^{\ast}=0$ in (72), we read the entire Lagrangian action ${S_{0}^{\mathrm{L}}}$: $\displaystyle{S_{0}^{\mathrm{L}}}[A_{\mu}^{a}]$ $\displaystyle=$ $\displaystyle-{{\textstyle\frac{1}{4}}}\int d^{4}xF_{\mu\nu}^{a}F_{a}^{\mu\nu}-{{\textstyle\frac{1}{2}}}g\int d^{4}x\,F_{a}^{\nu\mu}f_{bc}^{a}A_{\mu}^{b}A_{\mu}^{c}$ (73) $\displaystyle-{{\textstyle\frac{1}{4}}}g^{2}\int d^{4}x\,f_{bcm}f_{np}^{m}\eta^{\alpha\rho}\eta^{\beta\mu}A_{\rho}^{b}A_{\mu}^{c}A_{\alpha}^{n}A_{\beta}^{p}.$ It determines the following action ${S_{0}^{\mathrm{L}}}[A_{\mu}^{a}]=-{{\textstyle\frac{1}{4}}}\int d^{4}x\mathcal{F}_{\mu\nu}^{a}\mathcal{F}_{a}^{\mu\nu},$ (74) where $\mathcal{F}_{\mu\nu}^{a}$ is the field strengths defined by $\mathcal{F}_{\mu\nu}^{a}=F_{\mu\nu}^{a}+gf_{bc}^{a}A_{\mu}^{b}A_{\nu}^{c},$ (75) and $f_{bc}^{a}$ are the structure constants of the Lie algebra. The gauge symmetries read $\hat{\delta}_{\varepsilon}A_{\mu}^{a}=\partial_{\mu}\varepsilon^{a}-gf_{bc}^{a}\varepsilon^{b}A_{\mu}^{c}\equiv D_{\mu}\varepsilon^{a},$ (76) which holds the following commutator: $[\hat{\delta}_{\varepsilon_{1}},\hat{\delta}_{\varepsilon_{2}}]A_{\mu}^{a}=\hat{\delta}_{\varepsilon}A_{\mu}^{a}.$ (77) The gauge transformations remain abelian after consistent deformation. The antighost number one identifies the gauge transformations (76) by substituting $C^{a}$ with gauge parameter $\varepsilon^{a}$. The antighost number two reads the commutator (77). The resulting model is a non-abelian Yang-Mills model constructed by abelian vector fields $A_{\mu}^{a}$. ## IV Conclusion In this paper, we studied a consistently deforming Lagrangian action of the Yang-Mills model in the framework of the antifield formalism. We used the BRST differential to rewrite the deformations of the master equation. The analysis showed that all orders higher than two are trivial. The deformations stopped at second-order provide the consistent interactions being abelian to order $g$. Upon dismissing antifields, the entire gauge structures of the interacting theory is being realized. ## Acknowledgment I have been partially supported by a grant from the Marie Curie European Community Programme during my stay at the University of Craiova. ## Appendix A Antibracket Structure For a function $X(\psi)$ in a generic space, commutative or anticommutative, we state: ${\frac{{\partial_{l}X}}{{\partial\psi}}=\frac{{\vec{\partial}}}{{\partial\psi}}X,}~{}~{}~{}{\frac{{\partial_{r}X}}{{\partial\psi}}=X\frac{{\mathord{\buildrel{\lower 3.0pt\hbox{$\scriptscriptstyle\leftarrow$}}\over{\partial}}}}{{\partial\psi}}.}$ (78) The left derivative ${\partial_{l}}$ is an ordinary derivative (left to right). The right derivative ${\partial_{r}}$ is the derivative action from right to left. For any $X(\psi)$ in a generic space, we get $\frac{{\partial_{l}X}}{{\partial\psi}}=(-1)^{\varepsilon_{\psi}(\varepsilon_{X}+1)}\frac{{\partial_{r}X}}{{\partial\psi}}.$ (79) Considering Eqs. (14) and (79), it follows that $(X,Y)=-(-1)^{(\varepsilon_{X}+1)(\varepsilon_{Y}+1)}(Y,X).$ Assuming $X=Y$, one can find $\frac{{\partial_{r}X}}{{\partial\Phi^{A}}}\frac{{\partial_{l}X}}{{\partial\Phi_{A}^{\ast}}}=(-1)^{(\varepsilon_{X}+1)(\varepsilon_{X}+1)}\frac{{\partial_{r}X}}{{\partial\Phi_{A}^{\ast}}}\frac{{\partial_{l}X}}{{\partial\Phi^{A}}},$ (80) For bosonic (commutative) and fermionic (anticommutative) variables, we have $(X,X)=\left\\{{\begin{array}[]{cc}{\displaystyle 2\frac{{\partial_{r}X}}{{\partial\Phi^{A}}}\frac{{\partial_{l}X}}{{\partial\Phi_{A}^{\ast}}}}&{X\mathrm{~{}is~{}commutative,}}\\\ 0&{X\mathrm{~{}is~{}anticommutative}\mathrm{.}}\end{array}}\right.$ (81) For any $X$, we have $\begin{array}[]{cc}{((X,X),X)=0,}&{\forall X.}\end{array}$ (82) Furthermore, the antibracket has the following properties: $(X,YZ)=(X,Y)Z+(-1)^{\varepsilon_{Y}\varepsilon_{Z}}(X,Z)Y,$ (83) $(XY,Z)=X(Y,Z)+(-1)^{\varepsilon_{X}\varepsilon_{Y}}Y(X,Z),$ (84) $\displaystyle((X,Y),Z)+(-1)^{(\varepsilon_{X}+1)(\varepsilon_{Y}+\varepsilon_{Z})}((Y,Z),X)$ $\displaystyle+(-1)^{(\varepsilon_{Z}+1)(\varepsilon_{X}+\varepsilon_{Y})}((Z,X),Y)$ $\displaystyle=$ $\displaystyle 0.$ (85) ## References * (1) P. A. M. Dirac, Can. J. Math. 2, 129 (1950). * (2) P. A. M. Dirac, Proc. Roy. Soc. Lond. A 246, 326 (1958). * (3) P. A. M. Dirac, Lectures on Quantum Mechanics (New York: Yeshiva University, 1964). * (4) J. L. Anderson and P. G. Bergmann, Phys. Rev. 83, 1018 (1951). * (5) P. G. Bergmann and I. Goldberg, Phys. Rev. 98, 531 (1955). * (6) S. Weinberg, The Quantum Theory of Fields (Cambridge: Cambridge University Press, 1995). * (7) M. J. Gotay and J. M. Nester, Ann. Inst. H. Poincaré A 30, 129 (1979). * (8) M. J. Gotay and J. M. NesterAnn. Inst. H. Poincaré A 32, 1 (1980). * (9) C. Batlle, J. Gomis, J. M. Pons, and N. Roman-Roy J. Math. Phys. 27, 2953 (1986). * (10) D. M. Gitman and I. V. Tyutin, Quantization of Fields with Constraints (Berlin: Springer Verlag, 1990). * (11) J. Govaerts, Hamiltonian Quantisation and Constrained Dynamics (Leuven: Leuven University Press, 1991). * (12) A. Hanson and T. Regge and C. Teitelboim, Constrained Hamiltonian Systems (Rome: Accad. Naz. dei Lincei, 1976). * (13) L. D. Landau and E. M. Lifshitz, Mechanics, (trans.) J. B. Sykes and J. S. Bell (Oxford: Pergamon Press, 1976). * (14) E. C. G. Sudarshan and N. Mukunda, Classical Dynamics (New York: Wiley, 1974). * (15) K. Sundermeyer, Constrained Dynamics, Lecture Notes in Physics Vol. 169 (Berlin: Springer, 1982). * (16) C. Becchi , A. Rouet and R. Stora, Phys. Lett. B 52, 344 (1974). * (17) C. Becchi , A. Rouet and R. Stora, Commun. Math. Phys. 42, 127 (1975). * (18) C. Becchi , A. Rouet and R. Stora, Ann. Phys. (NY) 98 287, (1976). * (19) I. V. Tyutin, Gauge Invariance in Field Theory and Statistical Physics in Operator Formalism (LEBEDEV-75-39, 1975). * (20) F. Brandt, N. Dragon and M. Kreuzer, Phys. Lett. B 231, 263 (1989). * (21) F. Brandt, N. Dragon and M. Kreuzer, All Solutions of the Consistency Equations (DESY 89-076, ITP-UH 2/89, 1989). * (22) J. M. L. Fisch and M. Henneaux, Commun. Math. Phys. 128, 627 (1990). * (23) M. Henneaux, Nucl. Phys. B (Proc. Suppl.) 18A, 47 (1990). * (24) I. A. Batalin and G. A. Vilkovisky, Phys. Lett. B 102, 27 (1981). * (25) I. A. Batalin and G. A. Vilkovisky, Phys. Rev. D 28, 2567 (1983). * (26) I. A. Batalin and G. A. Vilkovisky, Phys. Rev. D 30, 508 (1984). * (27) M. Henneaux and C. Teitelboim, Quantization of Gauge Systems (Princeton: Princeton University Press, 1991). * (28) J. Gomis and J. París, Nucl. Phys. B 395, 288 (1993). [arXiv:hep-th/9204065] * (29) J. Gomis and J. París and S. Samuel, Phys. Rept. 259, 1 (1995). [arXiv:hep-th/9412228] * (30) G. Barnich and M. Henneaux, Phys. Lett. B 311, 123 (1993) [arXiv:hep-th/9304057] * (31) M. Henneaux, Contemp. Math. 219, 93 (1998). [arXiv:hep-th/9712226]

arxiv-papers

2024-09-04T02:48:53.039956

0707.4476

11institutetext: Dipartimento di Matematica e Informatica, Università degli Studi di Salerno, Via Ponte don Melillo, I-84084 Fisciano (SA), Italy INFN Sezione di Napoli, Gruppo collegato di Salerno, Baronissi (SA), Italy CNR-INFM Coherentia, Napoli, Italy ISI Foundation for Scientific Interchange, Viale Settimio Severo 65, I-10133 Torino, Italy Entanglement and quantum nonlocality Entanglement production, characterization, and manipulation Neutrino mass and mixing # Entanglement in neutrino oscillations Massimo Blasone 1122 Fabio Dell’Anno 112233 Silvio De Siena 112233 Fabrizio Illuminati 1122334411223344 ###### Abstract Flavor oscillations in elementary particle physics are related to multi-mode entanglement of single-particle states. We show that mode entanglement can be expressed in terms of flavor transition probabilities, and therefore that single-particle entangled states acquire a precise operational characterization in the context of particle mixing. We treat in detail the physically relevant cases of two- and three-flavor neutrino oscillations, including the effective measure of $CP$ violation. We discuss experimental schemes for the transfer of the quantum information encoded in single-neutrino states to spatially delocalized two-flavor charged lepton states, thus showing, at least in principle, that single-particle entangled states of neutrino mixing are legitimate physical resources for quantum information tasks. ###### pacs: 03.65.Ud ###### pacs: 03.67.Mn ###### pacs: 14.60.Pq Various branches of condensed matter, atomic physics, and quantum optics have evolved in recent years towards the investigation and development of schemes for quantum information and computation science [1]. To this aim, entanglement is a key ingredient and a crucial physical resource. Different forms of entanglement have been proven to be equivalent to observable quantifiers of performance success in quantum information protocols either with discrete [2] or continuous variables [3]. In the present work, we investigate the operational meaning of entanglement in the context of elementary particles physics. We will show that single-particle (mode) entanglement associated to particle mixing can be expressed in terms of transition probabilities in flavor oscillations, and can be exploited for quantum informational tasks. The concept of single-particle entanglement has been introduced and elucidated in a recent series of important theoretical papers [4, 5, 6]. Its use has been discussed in various contexts of quantum information, including teleportation, quantum cryptography, and violation of Bell inequalities [7, 8], and later experimentally demonstrated with single-photon systems [9, 10, 11]. Existing schemes to probe non-locality in single-photon states have been subsequently generalized to include massive particles [12]. In the present work we extend the discussion to the arena of elementary particles and provide a general operational characterization of single-particle entanglement in this context by showing its connection with the transition probabilities in any quantum system oscillating between different modes. We then show how this form of entanglement is in principle a real physical resource for the realization of quantum information protocols by discussing explicit experimental schemes for transferring it to spatially separated modes of stable leptonic particles. These results allow to place mode entanglement in neutrino oscillations on equal footing with mode entanglement in single-particle atomic and optical systems. Furthermore, we show how the single-particle entanglement quantifies $CP$ violation in neutrino mixing. Flavor mixing of neutrinos for three generations is described by the $3\times 3$ Pontecorvo-Maki-Nakagawa-Sakata (PMNS) unitary mixing matrix $\mathbf{U}(\tilde{\theta},\delta)$ [13], $\mathbf{U}(\tilde{\theta},\delta)\,=\,\left(\begin{array}[]{ccc}c_{12}c_{13}&s_{12}c_{13}&s_{13}e^{-i\delta}\\\ -s_{12}c_{23}-c_{12}s_{23}s_{13}e^{i\delta}&c_{12}c_{23}-s_{12}s_{23}s_{13}e^{i\delta}&s_{23}c_{13}\\\ s_{12}s_{23}-c_{12}c_{23}s_{13}e^{i\delta}&-c_{12}s_{23}-s_{12}c_{23}s_{13}e^{i\delta}&c_{23}c_{13}\end{array}\right)\,\,,$ (1) where $(\tilde{\theta},\delta)\equiv(\theta_{12},\theta_{13},\theta_{23};\delta)$ and $c_{ij}\equiv\cos\theta_{ij}$, $s_{ij}\equiv\sin\theta_{ij}$. The parameters $\theta_{ij}$ are the mixing angles, and $\delta$ is the phase responsible for $CP$ violation. Here, without loss of generality we consider only Dirac neutrinos. In the instance of Majorana neutrinos, two additional CP-violating phases are present, which, however, as it is well known, do not affect the physics of neutrino oscillations. The three-flavor neutrino states are defined as $|\underline{\nu}^{(f)}\rangle\,=\,\mathbf{U}(\tilde{\theta},\delta)\,|\underline{\nu}^{(m)}\rangle$ (2) where $|\underline{\nu}^{(f)}\rangle\,=\,\left(|\nu_{e}\rangle,|\nu_{\mu}\rangle,|\nu_{\tau}\rangle\right)^{T}$ are the states with definite flavor and $|\underline{\nu}^{(m)}\rangle\,=\,\left(|\nu_{1}\rangle,|\nu_{2}\rangle,|\nu_{3}\rangle\right)^{T}$ those with definite mass. Let us recall that both $|\nu_{\alpha}\rangle$ $(\alpha=e,\mu,\tau)$ and $|\nu_{j}\rangle$ $(j=1,2,3)$ are orthonormal, i.e. $\langle\nu_{\alpha}|\nu_{\beta}\rangle=\delta_{\alpha,\beta}$ and $\langle\nu_{j}|\nu_{k}\rangle=\delta_{j,k}$. Neutrino oscillations are due to neutrino mixing and neutrino mass differences. The neutrino states $|\nu_{j}\rangle$ have definite masses $m_{j}$ and definite energies $E_{j}$. Their propagation can be described by plane wave solutions of the form $|\nu_{j}(t)\rangle=e^{-iE_{j}t}|\nu_{j}\rangle$. The time evolution of the flavor neutrino states is given by: $\displaystyle|\underline{\nu}^{(f)}(t)\rangle$ $\displaystyle=$ $\displaystyle\mathbf{\widetilde{U}}(t)|\underline{\nu}^{(f)}\rangle\,,$ $\displaystyle\mathbf{\widetilde{U}}(t)$ $\displaystyle\equiv$ $\displaystyle\mathbf{U}(\tilde{\theta},\delta)\,\mathbf{U}_{0}(t)\,\mathbf{U}(\tilde{\theta},\delta)^{-1}\,,$ (3) where $|\underline{\nu}^{(f)}\rangle$ are the flavor states at $t=0$, $\mathbf{U}_{0}(t)=diag(e^{-iE_{1}t},e^{-iE_{2}t},e^{-iE_{3}t})$, and $\mathbf{\widetilde{U}}(t=0)=1{\rm I}$. At time $t$ the transition probability for $\nu_{\alpha}\rightarrow\nu_{\beta}$ is $P_{\nu_{\alpha}\rightarrow\nu_{\beta}}(t)\,=\,|\langle\nu_{\beta}|\nu_{\alpha}(t)\rangle|^{2}\,=\,|\mathbf{\widetilde{U}}_{\alpha\beta}(t)|^{2}\,,$ (4) where $\alpha,\beta=e,\mu,\tau\,.$ The transition probability $P_{\nu_{\alpha}\rightarrow\nu_{\beta}}(t)$ is a function of the energy differences $\Delta E_{jk}=E_{j}-E_{k}$ $(j,k=1,2,3)$ and of the mixing angles. Since the current neutrino experiments deal with ultra-relativistic neutrinos, the standard adopted approximation is $\Delta E_{jk}\simeq\frac{\Delta m_{jk}^{2}}{2E}$, where $\Delta m_{jk}^{2}=m_{j}^{2}-m_{k}^{2}$ and $E=|\overrightarrow{p}|$ is the energy of a massless neutrino (all massive neutrinos are assumed to have the same momentum $\overrightarrow{p}$). Flavor neutrinos are identified via charged current weak interaction processes, together with their associated charged leptons. In the Standard Model (SM), where neutrinos are taken to be massless, flavor is strictly conserved in such processes. On the other hand, neutrino mixing, consisting in a mismatch between flavor and mass, is at the basis of neutrino oscillations and CP violation. The introduction of neutrino masses as a correction to the SM is a necessary condition to explain such effects. When neutrino mixing is considered, loop corrections produce violations of lepton flavor in the charged current vertices: however, these effects are extremely small and essentially vanish in the relativistic limit[14]. Consequently, neutrino states entering weak interaction processes, like the ones where flavor neutrinos are created or detected, must be eigenstates of flavor neutrino charges. The corresponding operators can be rigorously defined together with their eigenstates in the context of Quantum Field Theory (QFT) [15]. In the relativistic limit, the exact QFT flavor states reduce to the usual Pontecorvo flavor states, which define the flavor modes as legitimate and physically well-defined individual entities. Mode entanglement can thus be defined and studied in analogy with the static case [16]. Equipped with the above setting, one can establish the following correspondence with three-qubit states: $|\nu_{e}\rangle\equiv|1\rangle_{\nu_{e}}|0\rangle_{\nu_{\mu}}|0\rangle_{\nu_{\tau}}$, $|\nu_{\mu}\rangle\equiv|0\rangle_{\nu_{e}}|1\rangle_{\nu_{\mu}}|0\rangle_{\nu_{\tau}}$, $|\nu_{\tau}\rangle\equiv|0\rangle_{\nu_{e}}|0\rangle_{\nu_{\mu}}|1\rangle_{\nu_{\tau}}$. States $|0\rangle_{\nu_{\alpha}}$ and $|1\rangle_{\nu_{\alpha}}$ correspond, respectively, to the absence and the presence of a neutrino in mode $\alpha$. Entanglement is thus established among flavor modes, in a single-particle setting. Eq. (3) can then be recast as $\displaystyle|\nu_{\alpha}(t)\rangle$ $\displaystyle=$ $\displaystyle\mathbf{\widetilde{U}}_{\alpha e}(t)|1\rangle_{\nu_{e}}|0\rangle_{\nu_{\mu}}|0\rangle_{\nu_{\tau}}+\mathbf{\widetilde{U}}_{\alpha\mu}(t)|0\rangle_{\nu_{e}}|1\rangle_{\nu_{\mu}}|0\rangle_{\nu_{\tau}}$ (5) $\displaystyle+\,\mathbf{\widetilde{U}}_{\alpha\tau}(t)|0\rangle_{\nu_{e}}|0\rangle_{\nu_{\mu}}|1\rangle_{\nu_{\tau}}\,,$ with the normalization condition $\sum_{\beta}|\mathbf{\widetilde{U}}_{\alpha\beta}(t)|^{2}=1$ $(\alpha,\beta=e,\mu,\tau)$. The time-evolved states $|\underline{\nu}^{(f)}(t)\rangle$ are entangled superpositions of the three flavor eigenstates with time-dependent coefficients. It is important to remark that, although dealing with similar physical systems, both the framework and the aim of the present paper differ substantially from those of Ref. [16]. In the latter, by exploiting the wave packet approach, the multipartite entanglement, associated with the multiqubit space of mass modes, has been analyzed in connection with the “decoherence” effects induced by free evolution. In the present work, by exploiting the plane-wave approximation, the entanglement is quantified with respect to the multiqubit space associated with flavor modes, and is related to the quantum information encoded in the neutrino flavor states, which is in principle experimentally accessible, as we will show by devising an explicit scheme for the information transfer. States of the form Eq. (5) belong to the class of $W$ states. These, together with the $GHZ$ states, define the two possible sets of states with tripartite entanglement that are inequivalent under local operations and classical communication [17]. In some instances, only two neutrinos are significantly involved in the mixing. For example, only the transition $\nu_{\mu}\leftrightarrow\nu_{\tau}$ is relevant for atmospheric neutrinos, while only the transitions of the type $\nu_{e}\leftrightarrow\nu_{\alpha}$ are relevant for solar neutrinos. For two-flavor mixing the mixing matrix $\mathbf{U}(\tilde{\theta},\delta)$ reduces to the $2\times 2$ rotation matrix $\mathbf{U}(\theta)$, $\mathbf{U}(\theta)=\left(\begin{array}[]{cc}\cos\theta&\sin\theta\\\ -\sin\theta&\cos\theta\end{array}\right)\,,$ (6) the evolution operator reads $\mathbf{U}_{0}(t)=diag(e^{-iE_{1}t},e^{-iE_{2}t})$, and the time-evolved flavor states yield the Bell-like superposition ($\alpha=e,\mu$): $|\nu_{\alpha}(t)\rangle\,=\,\mathbf{\widetilde{U}}_{\alpha e}(t)|1\rangle_{\nu_{e}}|0\rangle_{\nu_{\mu}}+\mathbf{\widetilde{U}}_{\alpha\mu}(t)|0\rangle_{\nu_{e}}|1\rangle_{\nu_{\mu}}.$ (7) Bipartite entanglement of pure states is unambiguously quantified by the von Neumann entropy or by any other monotonic function of the former [18]. Among entanglement monotones, the linear entropy has a special physical significance because it is directly linked to the purity of the reduced states, and enters in the fundamental monogamy inequalities for distributed entanglement in the multipartite setting [18]. As one moves from the two- to the three-flavor scenario, multipartite entanglement measures are readily available in terms of functions of bipartite measures [19, 20, 21]. Representative of this type of measures is the global entanglement. It is defined as the sum of all the two- qubit entanglements between a single subsystem and each of the remaining ones [19], and can be expressed as the average subsystem linear entropy [20]. Global entanglement can then be generalized by constructing the set of mean linear entropies associated to all possible bi-partitions of the entire system [21]. An alternative characterization of multipartite entanglement is given in Refs.[22, 23]. Let $\rho=|\psi\rangle\langle\psi|$ be the density operator corresponding to a pure state $|\psi\rangle$, describing the system $S$ partitioned into $N$ parties. Consider the bipartition of the $N$-partite system $S=\\{S_{1},S_{2},\ldots,S_{N}\\}$ in two subsystems $S_{A_{n}}=\\{S_{i_{1}},S_{i_{2}},\ldots,S_{i_{n}}\\}$, with $1\leq i_{1}<i_{2}<\ldots<i_{n}\leq N$ $(1\leq n<N)$, and $S_{B_{N-n}}=\\{S_{j_{1}},S_{j_{2}},\ldots,S_{j_{N-n}}\\}$, with $1\leq j_{1}<j_{2}<\ldots<j_{N-n}\leq N$, and $i_{q}\neq j_{p}$. Let $\rho_{A_{n}}\equiv\rho_{i_{1},i_{2},\ldots,i_{n}}\,=\,Tr_{B_{N-n}}[\rho]\,=\,Tr_{j_{1},j_{2},\ldots,j_{N-n}}[\rho]\,$ (8) denote the reduced density matrix of subsystem $S_{A_{n}}$ after tracing over subsystem $S_{B_{N-n}}$. The linear entropy associated to such a bipartition is defined as $S_{L}^{(A_{n};B_{N-n})}(\rho)\,=\,\frac{d}{d-1}(1-Tr_{A_{n}}[\rho_{A_{n}}^{2}])\,,$ (9) where the $d$ is the Hilbert-space dimension given by $d=\min\\{\dim S_{A_{n}}\,,\dim S_{B_{N-n}}\\}=\min\\{2^{n},2^{N-n}\\}$. Finally, we introduce the average linear entropy $\langle S_{L}^{(n:N-n)}(\rho)\rangle\,=\,\left(\begin{array}[]{c}N\\\ n\\\ \end{array}\right)^{-1}\;\sum_{A_{n}}S_{L}^{(A_{n};B_{N-n})}(\rho)\,,$ (10) where the sum is intended over all the possible bi-partitions of the system in two subsystems, respectively with $n$ and $N-n$ elements $(1\leq n<N)$ [21]. We can now compute the linear entropies (9) and (10) for the two-qubit Bell state $|\nu_{\alpha}(t)\rangle$, i.e. Eq. (7), with density matrix $\rho^{(\alpha)}=|\nu_{\alpha}(t)\rangle\langle\nu_{\alpha}(t)|$. The linear entropy associated to the reduced state after tracing over one mode (flavor) can be computed straightforwardly: $\displaystyle S_{L\alpha}^{(\mu;e)}\,=\,S_{L\alpha}^{(e;\mu)}$ $\displaystyle=$ $\displaystyle 4|\mathbf{\widetilde{U}}_{\alpha e}(t)|^{2}\,|\mathbf{\widetilde{U}}_{\alpha\mu}(t)|^{2}$ (11) $\displaystyle=$ $\displaystyle 4|\mathbf{\widetilde{U}}_{\alpha e}(t)|^{2}\,(1-|\mathbf{\widetilde{U}}_{\alpha e}(t)|^{2})$ $\displaystyle=$ $\displaystyle 4|\mathbf{\widetilde{U}}_{\alpha\mu}(t)|^{2}\,(1-|\mathbf{\widetilde{U}}_{\alpha\mu}(t)|^{2})\,.$ In Eq. (11) and in the following, we use the notation $S_{L\alpha}^{(e;\mu)}\equiv S_{L}^{(e;\mu)}(\rho^{(\alpha)})$, where the subscript $\alpha$ refers to the time-evolved state (channel), and the superscripts $(e;\mu)$ refer to the considered modes (flavors). Clearly, for the two-flavor state (7), and in general for any two-qubit system, symmetry imposes $S_{L\alpha}^{(e;\mu)}=S_{L\alpha}^{(\mu;e)}=\langle S_{L\alpha}^{(1:1)}\rangle$. Expression (11) establishes that the linear entropy of the reduced state is equal to the product of the two-flavor transition probabilities. Moreover, for any reduced state $\rho$ of a two- level system one has that $S_{L}=2[1-Tr(\rho^{2})]=4Det\rho=4\lambda_{1}(1-\lambda_{1})$, where $\lambda_{1}$ is one of the two non-negative eigenvalues of $\rho$, and the relation $\lambda_{1}+\lambda_{2}=1$ has been exploited. Comparing with Eq. (11), one sees that the transition probabilities coincide with the eigenvalues of the reduced state density matrix. In Fig. 1 we show the behavior of $S_{Le}^{(e;\mu)}$ as a function of the scaled, dimensionless time $T=\frac{2Et}{\Delta m_{12}^{2}}$. In the same figure, we also report the behavior of the transition probabilities $P_{\nu_{e}\rightarrow\nu_{e}}$ and $P_{\nu_{e}\rightarrow\nu_{\mu}}$. Figure 1: (Color online) Linear entropy $S_{Le}^{(e;\mu)}$ (full) as a function of the scaled time $T=\frac{2Et}{\Delta m_{12}^{2}}$. The mixing angle $\theta$ is fixed at the experimental value $\sin^{2}\theta=0.314$. The transition probabilities $P_{\nu_{e}\rightarrow\nu_{e}}$ (dashed) and $P_{\nu_{e}\rightarrow\nu_{\mu}}$ (dot-dashed) are reported as well for comparison. The plots have a clear physical interpretation. At time $T=0$, the entanglement is zero, the global state of the system is factorized, and the two flavors are not mixed. For $T>0$, flavors start to oscillate and the entanglement is maximal at largest mixing: $P_{\nu_{e}\rightarrow\nu_{e}}=P_{\nu_{e}\rightarrow\nu_{\mu}}=0.5$, and minimum at $T=\pi$. We can now investigate three-flavor oscillations, and the associated three- qubit $W$-like states (5). Similarly to Eq. (11), tracing, e. g., over mode $\tau$, one has $\displaystyle S_{L\alpha}^{(e,\mu;\tau)}$ $\displaystyle=$ $\displaystyle 4|\mathbf{\widetilde{U}}_{\alpha\tau}(t)|^{2}\,(|\mathbf{\widetilde{U}}_{\alpha e}(t)|^{2}+|\mathbf{\widetilde{U}}_{\alpha\mu}(t)|^{2})$ (12) $\displaystyle=$ $\displaystyle 4|\mathbf{\widetilde{U}}_{\alpha\tau}(t)|^{2}\,(1-|\mathbf{\widetilde{U}}_{\alpha\tau}(t)|^{2})\,.$ The linear entropies for the two remaining bi-partitions are easily obtained by permuting the indexes $e,\mu,\tau$. The average linear entropy for the state (5) is then $\displaystyle\langle S_{L\alpha}^{(2:1)}\rangle$ $\displaystyle=$ $\displaystyle\frac{8}{3}(|\mathbf{\widetilde{U}}_{\alpha e}(t)|^{2}|\mathbf{\widetilde{U}}_{\alpha\mu}(t)|^{2}+|\mathbf{\widetilde{U}}_{\alpha e}(t)|^{2}|\mathbf{\widetilde{U}}_{\alpha\tau}(t)|^{2}$ (13) $\displaystyle+|\mathbf{\widetilde{U}}_{\alpha\mu}(t)|^{2}|\mathbf{\widetilde{U}}_{\alpha\tau}(t)|^{2}).$ In Fig. 2, we show $S_{Le}^{(\alpha,\beta;\gamma)}$ and $\langle S_{Le}^{(2;1)}\rangle$ as functions of the scaled time $T=\frac{2Et}{\Delta m_{12}^{2}}$. The mixing angles $\theta_{ij}$ and the squared mass differences are fixed at the most recent experimental values reported in Ref. [24]. Figure 2: (Color online) Linear entropies $S_{Le}^{(\alpha,\beta;\gamma)}$ and $\langle S_{Le}^{(2;1)}\rangle$ as functions of the scaled time $T$. Curves correspond to the partial linear entropies $S_{Le}^{(e,\mu;\tau)}$ (long- dashed), $S_{Le}^{(e,\tau;\mu)}$ (dashed), $S_{Le}^{(\mu,\tau;e)}$ (dot- dashed), and to the average linear entropy $\langle S_{Le}^{(2;1)}\rangle$ (full). Parameters $\theta_{ij}$ and $\Delta m_{ij}^{2}$ are fixed at the central experimental values [24]. In order to track the behavior of the entanglement, we plot in Fig. 3 the transition probabilities $P_{\nu_{e}\rightarrow\nu_{\alpha}}$ $(\alpha=e,\mu,\tau)$. Figure 3: (Color online) Transition probabilities $P_{\nu_{e}\rightarrow\nu_{\alpha}}$ as functions of the scaled time $T$. Parameters $\theta_{ij}$ and $\Delta m_{ij}^{2}$ are fixed at the central experimental values as in Fig. 2. Curves correspond to $P_{\nu_{e}\rightarrow\nu_{e}}$ (long-dashed), $P_{\nu_{e}\rightarrow\nu_{\mu}}$ (dashed), and $P_{\nu_{e}\rightarrow\nu_{\tau}}$ (dot-dashed). Comparing Fig. 2 and Fig. 3, we observe that, as one may expect, the more mixed are the flavors, the higher is the global multipartite entanglement of the system. Moreover, the partial linear entropies $S_{Le}^{(e,\mu;\tau)}$ and $S_{Le}^{(e,\tau;\mu)}$ measuring the reduced bipartite entanglement, exhibit a similar behavior due to the strong correlation between the components $\nu_{\mu}$ and $\nu_{\tau}$. As $T>0$ the probabilities $P_{\nu_{e}\rightarrow\nu_{\mu}}$ and $P_{\nu_{e}\rightarrow\nu_{\tau}}$ increase and oscillate while remaining close. Similar considerations hold for states $|\nu_{\mu}(t)\rangle$ and $|\nu_{\tau}(t)\rangle$. Entanglement and flavor transition probabilities for these states exhibit very fast oscillating behaviors, related to the experimentally measured values of the mixing parameters. Because of $CPT$ invariance, the $CP$ asymmetry $\Delta_{CP}^{\alpha,\beta}$ is equal to the asymmetry under time reversal, defined as $\displaystyle\Delta_{T}^{\alpha,\beta}$ $\displaystyle=$ $\displaystyle P_{\nu_{\alpha}\rightarrow\nu_{\beta}}(t)-P_{\nu_{\beta}\rightarrow\nu_{\alpha}}(t)$ (14) $\displaystyle=$ $\displaystyle P_{\nu_{\alpha}\rightarrow\nu_{\beta}}(t)-P_{\nu_{\alpha}\rightarrow\nu_{\beta}}(-t)\,.$ In the three-flavor instance, such a quantity is different from zero for a nonvanishing phase $\delta$. It is worth noticing that $\sum_{\beta}\Delta_{CP}^{\alpha\beta}=0$ with $\alpha,\beta=e,\mu,\tau$. Introducing the “imbalances”, i.e. the difference between the linear entropies and their time-reversed expressions: $\Delta S_{L\lambda}^{(\alpha,\beta;\gamma)}=S_{L\lambda}^{(\alpha,\beta;\gamma)}(t)-S_{L\lambda}^{(\alpha,\beta;\gamma)}(-t)\,,$ (15) we can compute, e. g., $\Delta S_{Le}^{(e,\mu;\tau)}$, and obtain: $\Delta S_{Le}^{(e,\mu;\tau)}=4\Delta_{CP}^{e,\mu}(|\mathbf{\widetilde{U}}_{e\tau}(t)|^{2}+|\mathbf{\widetilde{U}}_{\tau e}(t)|^{2}-1)\,,$ (16) where the last factor is $CP$-even. In Fig. 4 we show the behavior of the imbalances $\Delta S_{Le}^{(\alpha,\beta;\gamma)}$ as functions of time, and see how they effectively measure $CP$ violation. Figure 4: (Color online) The imbalances $\Delta S_{Le}^{(\alpha,\beta;\gamma)}$ as functions of the scaled time $T$. Curves correspond to $\Delta S_{Le}^{(e,\mu;\tau)}$ (long-dashed) and $\Delta S_{Le}^{(e,\tau;\mu)}$ (dot-dashed). The quantity $\Delta S_{Le}^{(\mu,\tau;e)}$ is vanishing. Parameters $\theta_{ij}$ and $\Delta m_{ij}^{2}$ are fixed at the central experimental values as in Fig. 2. The $CP$-violating phase is set at the value $\delta=\pi/2$. In order to demonstrate that the form of single-particle entanglement encoded in the time-evolved flavor states $|\underline{\nu}^{(f)}(t)\rangle$ is a real physical resource that can be legitimately used, at least in principle, for protocols of quantum information, we discuss an experimental scheme for the transfer of the flavor entanglement of a neutrino beam into that of a single- particle system with spatially separated modes. For simplicity, we will restrict the analysis to two flavors $\alpha=e,\mu$. Consider the elementary charged-current interaction between a neutrino $\nu_{\alpha}$ with flavor $\alpha$ and a nucleon $N$ [13]. The quasi-elastic scattering interaction yields the production of a lepton $\alpha^{-}$ and of an outgoing baryon $X$, according to the reaction: $\nu_{\alpha}+N\longrightarrow\alpha^{-}+X\,.$ (17) In the simplest instance, the nucleon $N$ is a neutron and the baryon $X$ is a proton $p$; the corresponding scheme is illustrated in Fig. 5. Figure 5: (Color online) Scheme for the generation of single-particle entangled lepton states. A neutrino beam is focused on a target where events of the following charged-current interaction may occur: $\nu_{\alpha}+n\longrightarrow\alpha^{-}+p$ with $\alpha=e\,,\mu$. The beam is assumed to have energy higher than the threshold value necessary for the creation of a muon. A spatially nonuniform magnetic field $\mathbf{B}(\mathbf{r})$ is then applied to limit the momentum of the outgoing lepton within a certain solid angle $\Omega_{i}$, and to ensure the spatial separation between the electron and muon spatial paths. The reaction produces a superposition of electronic and muonic spatially separated states. Given the initial Bell-like superposition (7), the unitary process associated with the weak interaction (17) produces the superposition $|\alpha(t)\rangle\,=\,\Lambda_{e}|1\rangle_{e}|0\rangle_{\mu}+\Lambda_{\mu}|0\rangle_{e}|1\rangle_{\mu}\,,$ (18) where $|\Lambda_{e}|^{2}+|\Lambda_{\mu}|^{2}=1$, and $|k\rangle_{\alpha}$, with $k=0,1$, represents the lepton qubit. The coefficients $\Lambda_{\alpha}$ are proportional to $\mathbf{\widetilde{U}}_{\alpha\beta}(t)$ and to the cross sections associated with the creation of an electron or a muon. Comparing our single-lepton system with the single-photon system, the quantum uncertainty on “which path” of the photon at the output of an unbalanced beam splitter is replaced by the uncertainty on “which flavor” of the produced lepton. The coefficients $\Lambda_{\alpha}$ play the role of the transmissivity and of the reflectivity of the beam splitter. Moreover, by exploiting the mass difference between the two leptons, the desired spatial separation between the flavors can be achieved by applying a nonuniform magnetic field. It is also important to remark that the approach proposed in the present work can be applied even in extended neutrino models including one or more sterile neutrinos. In such cases, from a mathematical point of view the main difference is that one deals with more than three modes (flavors), while, from an operational point of view, the presence of sterile neutrinos (undetectable to date) would introduce a mechanism of loss of quantum information by making the (observed) mixing matrix non unitary. Therefore we can conclude that, at least in principle, the quantum information encoded in the neutrino flavor states can be transferred to a delocalized two-flavor lepton state, and the single-particle mode entanglement acquires an operational characterization that can be exploited for quantum information tasks using systems of elementary particle physics. ###### Acknowledgements. We acknowledge financial support from MIUR, CNR-INFM Research and Development Center “Coherentia”, INFN, and from ISI Foundation for Scientific Interchange. ## References * [1] Nielsen M. A. Chuang I. L. Quantum Computation and Quantum Information CUP, Cambridge, UK 2001\. * [2] Horodecki M., Horodecki P. Horodecki R. Phys. Rev. A 6019991888. * [3] Adesso G. Illuminati F. Phys. Rev. Lett. 952005150503. * [4] Zanardi P. Phys. Rev. A 652002042101; Shi Y. Phys. Rev. A 672003024301; Zanardi P., Lidar D. A. Lloyd S. Phys. Rev. Lett. 922004060402; * [5] van Enk S. J. Phys. Rev. A 722005064306; ibid.742006026302; * [6] Terra Cunha M. O.,Dunningham J. A. Vedral V. Proc. Royal Soc. A 46320072277. * [7] Björk G., Jonsson P. Sánchez-Soto L. L. Phys. Rev. A 642001042106; * [8] Lee J. W. et al. Phys. Rev. A 682003012324; Nha H. Kim J. Phys. Rev. A 752007012326; * [9] Lombardi E. et al. Phys. Rev. Lett.882002070402; * [10] Lvovsky A. I. et al. Phys. Rev. Lett.872001050402; Babichev S. A. , Appel J. Lvovsky A. I. Phys. Rev. Lett.922004193601; * [11] Hessmo B. et al. Phys. Rev. Lett.922004180401; * [12] Dunningham J. Vedral V. Phys. Rev. Lett.992007180404; * [13] Cheng T. Li L. Gauge Theory of Elementary Particle Physics Clarendon Press 1989\. * [14] Casas J. A. Ibarra A. Nucl. Phys. B6182001171; * [15] Blasone M. Vitiello G. Ann. Phys. (N.Y.) 2441995283; Blasone M., Jizba P. Vitiello G. Phys. Lett. B5172001471. * [16] Blasone M., Dell’Anno F., De Siena S., Di Mauro M. Illuminati F. Phys. Rev. D 772008096002. * [17] Dür W., Vidal G. Cirac J. I. Phys. Rev. A622000062314. * [18] Amico L., Fazio R., Osterloh A. Vedral V. Rev. Mod. Phys.802008517; Horodecki R., Horodecki P., Horodecki M. Horodecki K. quant-ph/0702225; * [19] Meyer D. A. Wallach N. R. J. Math. Phys. 4320024273. * [20] Brennen G. K. Quantum Inf. Comp.32003619. * [21] de Oliveira T. R., Rigolin G. de Oliveira M. C. Phys. Rev. A732006010305(R). * [22] Wei T. C. Goldbart P. M. Phys. Rev. A682003042307. * [23] Blasone M., Dell’Anno F., De Siena S. Illuminati F. Phys. Rev. A772008062304. * [24] Fogli G. L. et al. Phys. Rev. D 752007053001.

arxiv-papers

2024-09-04T02:48:53.048929

0708.0580

11institutetext: Department of Combinatorics & Optimization, University of Waterloo, Canada # Efficient Divide-and-Conquer Implementations Of Symmetric FSAs David A. G. Pritchard daveagp@gmail.com ###### Abstract A deterministic finite-state automaton (FSA) is an abstract sequential machine that reads the symbols comprising an input word one at a time. An FSA is _symmetric_ if its output is independent of the order in which the input symbols are read, i.e., if the output is invariant under permutations of the input. We show how to convert a symmetric FSA $\mathcal{A}$ into an automaton- like divide-and-conquer process whose intermediate results are no larger than the size of $\mathcal{A}$’s memory. In comparison, a similar result for general FSA’s has been long known via functional composition, but entails an exponential increase in memory size. The new result has applications to parallel processing and symmetric FSA networks. ###### keywords: divide and conquer, FSA, network, parallel processing, PRAM, sequential automaton, symmetry ## 1 Introduction One of the simplest models of computation is the _deterministic finite state automaton_ (FSA). Although FSAs are often considered to act as solitary computing devices (e.g., in the classical string matching algorithm of Knuth, Morris, and Pratt [6]) they can also be connected together to form a computing network (e.g., in cellular automata and the models of [1, 8]). A _symmetric_ automaton is one that will produce the same output even if its inputs are permuted. Symmetric FSAs are natural building blocks for fault- tolerant computation networks. In previous work with Vempala [10] we showed there are symmetric FSAs implementing fault-tolerant randomized algorithms for the following tasks: implicit 2-approximate census (via an algorithm due to Milgram [8]), network search (via breadth-first search and greedy traversal), basic connectivity problems (finding all bridges), and leader election. Precisely, the _finite-state symmetric graph automaton_ (FSSGA) model introduced in [10] is that a copy of the same symmetric FSA is placed at every node of a graph (network); when a node “activates” to advance its state, it obtains one input symbol from each neighbour without regard to order. In sum, FSSGA are like symmetric cellular automata but generalized in the sense that the underlying graph does not have to be regular. Three models of symmetric automata are given in [10]: in the _sequential_ model each node is a (sequential) FSA, in the _parallel_ model each node uses divide-and-conquer on its inputs (in a way that will be defined precisely later), and in the _mod- thresh_ model each node applies a finite-size formula (analogous to a regular expression) to update its state. One of the main results of [10] is that these three models are equivalent; e.g., for any symmetric FSA there exists a divide-and-conquer process to compute the same function. Unfortunately, for the particular construction given in [10], an exponential increase in the size of the state space is required. More generally, an efficient way to simulate _any_ FSA with divide-and-conquer has been known for decades. The basic technique is sometimes called _functional composition_ as applied to _parallel prefix_. Ladner and Fischer used the technique in 1977 [7] on the PRAM model of parallel computing; see also [9] for an implementation in mesh networks. The basic idea is that for any single character $\sigma$, the transition of the FSA on that character can be viewed as function $f_{\sigma}$ from the FSA’s state space back to itself, and the computation of the FSA on a string $w=w_{1}w_{2}\dotsb w_{k}$ is essentially determined by the composition of functions $f_{w}:=f_{w_{k}}\circ\dotsb\circ f_{w_{2}}\circ f_{w_{1}}.$ In turn, this composition problem lends itself easily to divide-and-conquer: break the string into two parts $w=uv$, compute the compositions $f_{u}$ and $f_{v}$ for the two parts, and return $f_{v}\circ f_{u}$. Like the transformation of [10] for symmetric automata, the size of intermediate results increases exponentially, since for a state space $Q$ there are $|Q|^{|Q|}$ functions from $Q$ to $Q$. The main contribution of this paper is that for a _symmetric_ FSA, no increase in the state space size is necessary. We present the result (Theorem 7) after introducing our notation. The resulting small-state-space divide-and-conquer process is applicable to the PRAM setting, so e.g. for symmetric FSAs we are able to decrease the working memory used by the divide-and-conquer approaches of [7, 9]. For high-degree FSSGAs and the special case of symmetric cellular automata, divide-and-conquer is a natural way for each node to read its neighbours’ states, as we will illustrate in Section 2; our main result permits such divide-and-conquer processes to be more memory-efficient. ## 2 Preliminaries We denote an FSA by the tuple $(\Sigma,Q,q_{0},\\{f_{\sigma}\\}_{\sigma\in\Sigma},O,\beta)$. Here $\Sigma$ is a finite set called the _input alphabet_ , $Q$ is a finite set called the _state space_ , $q_{0}$ is an element of $Q$ called the _initial state_ , each $f_{\sigma}$ is a function from $Q$ to $Q$ called the _transition function of $\sigma$_, $O$ is a finite _output set_ , and $\beta$ is an _output function_ from $Q$ to $O$. ###### Definition 1 (FSA). An _FSA_ is any tuple $\mathcal{A}=(\Sigma,Q,q_{0},\\{f_{\sigma}\\}_{\sigma\in\Sigma},O,\beta)$ of the form described above. Let $\Sigma^{*}$ denote the set of all strings over $\Sigma$, and let $f\circ g$ denote the functional composition of $f$ and $g$, defined by $(f\circ g)(x)=f(g(x)).$ It is convenient to extend the definition of $f$ to strings via functional composition. Namely, for a string $w=w_{1}w_{2}\dotsb w_{k},$ define $f_{w}:=f_{w_{k}}\circ f_{w_{k-1}}\circ\dotsb\circ f_{w_{2}}\circ f_{w_{1}},$ and by convention, where $\lambda$ denotes the empty string, let $f_{\lambda}$ be the identity function on $Q$. In particular, we obtain the identity $f_{uv}(q)=f_{v}(f_{u}(q))$ for any strings $u,v\in\Sigma^{*}$ and any $q\in Q$. Let $\Sigma^{+}$ denote the set of nonempty strings over $\Sigma$; the empty string is excluded to agree with the divide-and-conquer model later on. Our definition of $f_{w}$ affords a concise description of computation for an FSA. ###### Definition 2 (FSA computation). An FSA $\mathcal{A}=(\Sigma,Q,q_{0},\\{f_{\sigma}\\}_{\sigma\in\Sigma},O,\beta)$ _computes_ the function $\nu_{\mathcal{A}}:\Sigma^{+}\to O$ defined by $\nu_{\mathcal{A}}(w):=\beta(f_{w}(q_{0})).$ Note that the traditional model where the FSA accepts or rejects strings depending on the final state can be modeled by setting $O=\\{accept,reject\\}$ and defining $\beta(q)=accept$ iff $q$ is an accepting state. We use the multi-output version because it is more natural in some settings, e.g., the FSSGA model. We represent a divide-and-conquer automaton by a tuple $(\Sigma,Q,\alpha,c,O,\beta)$. As before $\Sigma$ is the input alphabet, $Q$ is the state space, $O$ is the output set and $\beta$ is the output function. Here $\alpha$ is an _input function_ from $\Sigma$ to $Q$ and $c$ is a _combining function_ from $Q\times Q$ to $Q$. Informally, the divide-and- conquer automaton runs according to the following rules: 1. 1. apply $\alpha$ to all input characters 2. 2. combine states arbitrarily using $c$ until only one state $q^{*}$ is left 3. 3. output $\beta(q^{*})$. Our definition will require that the end result of the computation is the same no matter how the arbitrary choices of combination are made. To give our formal definition, we use a set-valued function $\chi$ that maps each nonempty string to a subset of $Q$ so that $q^{*}\in\chi(w)$ iff, dividing inputs arbitrarily, the input $w$ could produce $q^{*}$ as the final state. We denote the length of $w$ by $|w|$. ###### Definition 3 (DCA). Let $\mathcal{A}^{\prime}$ denote the tuple $(\Sigma,Q,\alpha,c,O,\beta)$ as described above. Define $\chi_{\mathcal{A}^{\prime}}(w)$ for $w\in\Sigma^{+}$ recursively as follows: if $|w|=1$, say $w$ consists of the character $\sigma$, then $\chi_{\mathcal{A}^{\prime}}(w):=\\{\alpha(\sigma)\\};$ otherwise (for $|w|\geq 2$) $\chi_{\mathcal{A}^{\prime}}(w):=\bigcup_{\begin{subarray}{c}(u,v):uv=w\end{subarray}}\\{c(q^{*}_{u},q^{*}_{v})\mid q^{*}_{u}\in\chi_{\mathcal{A}^{\prime}}(u),q^{*}_{v}\in\chi_{\mathcal{A}^{\prime}}(v)\\}$ (1) where $(u,v)$ ranges over all partitions of $w$ into two nonempty substrings. We say that $\mathcal{A}^{\prime}$ is a _divide-and-conquer automaton (DCA)_ if for all $w\in\Sigma^{+}$, $|\\{\beta(q^{*})\mid q^{*}\in\chi_{\mathcal{A}^{\prime}}(w)\\}|=1.$ (2) The previous definition amounts to saying that the output of a divide-and- conquer automaton should be well-defined regardless of how the dividing is performed. For a singleton set $S$ let $\mathtt{the.member}(S)$ be a function that returns the element of $S$, i.e., it “unwraps” the set. ###### Definition 4 (DCA computation). A DCA $\mathcal{A}^{\prime}=(\Sigma,Q,\alpha,c,O,\beta)$ _computes_ the function $\nu_{\mathcal{A}^{\prime}}:\Sigma^{+}\to O$ defined by $\nu_{\mathcal{A}^{\prime}}(w)=\mathtt{the.member}(\\{\beta(q)\mid q\in\chi_{\mathcal{A}^{\prime}}(w)\\}).$ (3) Figure 1 illustrates how a node in an FSA-based computing network could make use of the divide-and-conquer methodology. Specifically, when reading the states of all neighbours the node can process and combine inputs from its neighbours in parallel rather than one-by-one. As a function of the neighbourhood size $|\Gamma|$ (i.e. the degree) the circuit depicted has depth $O(\log|\Gamma|)$ and hence this approach would lead to efficient physical implementation for large neighbourhoods. $v$$n$$n$$n$$n$$n$$n$$n$$n$$\alpha$$\alpha$$c$$\alpha$$\alpha$$c$$c$$c$$\beta$$\alpha$$\alpha$$\alpha$$\alpha$$c$$c$$c$ Figure 1: An FSA in a network updates its state via divide-and-conquer. The node $v$ is activating and its neighbours are labeled $n$. The lines carry values from tail to head, and the boxes apply functions, like in a circuit diagram. Each neighbour supplies an input symbol and the divide-and-conquer process produces an output symbol which is used by $v$ to update its state. We denote by $Q^{Q}$ the set of all functions from $Q$ to $Q$. We mentioned the following well-known (e.g., [7]) result earlier: ###### Theorem 5. Given any FSA $\mathcal{A}$, there is a DCA $\mathcal{A}^{\prime}$ such that $\nu_{\mathcal{A}}=\nu_{\mathcal{A}^{\prime}},$ i.e., $\mathcal{A}$ and $\mathcal{A}^{\prime}$ compute the same function. ###### Proof. Define $\mathcal{A}^{\prime}=(\Sigma,Q^{Q},\sigma\mapsto f_{\sigma},(f_{1},f_{2})\mapsto f_{2}\circ f_{1},O,\beta)$. ∎ Conversely, as was observed in [10], any divide-and-conquer automaton can be easily rewritten in sequential form since a sequential FSA can be thought of as conquering one input at a time. The particular result we want to prove pertains only to symmetric automata, which we now define formally. ###### Definition 6. Suppose that $\mathcal{A}$ is an FSA or a DCA. We say that $\mathcal{A}$ is _symmetric_ if for every $w\in\Sigma^{+}$ and every permutation $w^{\prime}$ of $w,$ $\nu_{\mathcal{A}}(w)=\nu_{\mathcal{A}}(w^{\prime}).$ The main result of the present paper is the following, which is a more efficient version of Theorem 5 for symmetric FSA’s. ###### Theorem 7. Given any symmetric FSA $\mathcal{A}=(\Sigma,Q,q_{0},f,O,\beta)$, there is a DCA $\mathcal{A}^{\prime}=(\Sigma,Q^{\prime},\alpha,c,O,\beta^{\prime})$ such that $\nu_{\mathcal{A}}=\nu_{\mathcal{A}^{\prime}}$ _and $|Q^{\prime}|\leq|Q|$_. In the next section, we prove a supporting lemma for later use. In Section 4 we complete the proof of Theorem 7. In Section 5 we mention some ideas for future investigation. ## 3 Looking Inside A Symmetric FSA The key to Theorem 7 is to focus on automata with specific irredundany properties. Symmetry of an automaton is a _black-box property_ — the definition only cares about the correspondence of final outputs when the inputs are permutations of one another, regardless of the internal structure of the automaton. We now describe how this black-box property (symmetry), when combined with irredundancy requirements, implies a structural property — namely, that the transition functions must commute. ###### Definition 8 ([4]). Let $\mathcal{A}=(\Sigma,Q,q_{0},f,O,\beta)$ be an FSA and let $q\in Q$. The state $q$ is said to be _accessible_ if for some string $w\in\Sigma^{*},$ $f_{w}(q_{0})=q$. We say $\mathcal{A}$ is _accessible_ if every state in $Q$ is accessible. ###### Definition 9 ([4]). Let $\mathcal{A}=(\Sigma,Q,q_{0},f,O,\beta)$ be an FSA and let $q,q^{\prime}\in Q$. The states $q$ and $q^{\prime}$ are said to be _distinguishable_ if for some string $w\in\Sigma^{*},$ $\beta(f_{w}(q))\neq\beta(f_{w}(q^{\prime}))$. We say $\mathcal{A}$ is _distinguishable_ if every pair of states in $Q$ is distinguishable. As we will later make precise, every FSA can be rewritten in an accessible, distinguishable way. This gives some general applicability to the following lemma. ###### Lemma 10 (Commutativity Lemma). Let $\mathcal{A}=(\Sigma,Q,q_{0},f,O,\beta)$ be a symmetric FSA that is accessible and distinguishable. Then the functions $\\{f_{\sigma}\\}_{\sigma\in\Sigma}$ commute. We defer the proof of the lemma to the end of this section. In order to explain how it is useful, we recall the following additional definitions. ###### Definition 11 ([4]). Two automata $\mathcal{A},\mathcal{A}^{\prime}$ are _equivalent_ if they compute the same function, i.e. if $\nu_{\mathcal{A}}=\nu_{\mathcal{A}^{\prime}}$. An FSA $\mathcal{A}$ is _minimal_ if for every FSA $\mathcal{A}^{\prime}$ equivalent to $\mathcal{A}$, $\mathcal{A}^{\prime}$ has at least as many states as $\mathcal{A}$. It is not hard to see that any minimal FSA must be accessible (or else we could remove some states) and distinguishable (or else we could collapse some states)111Interestingly, the converse is also true: any accessible, distinguishable FSA is minimal. See [4] for a derivation of this result as a corollary of the Myhill-Nerode theorem; adapting the proof from accept/reject automata to our more general model is straightforward.. It is also not hard to see that for every FSA $\mathcal{A}$ there exists a minimal equivalent FSA $\mathcal{A}^{\prime}$; such minimization can be performed algorithmically in $poly(|Q|,|\Sigma|)$ time, e.g. using an approach of Hopcroft [5]. In sum, for any FSA we can efficiently obtain an equivalent FSA meeting the conditions of Lemma 10, which we now prove. ###### Proof of Lemma 10. Suppose for the sake of contradiction that not all of the functions $f$ commute. Then $f_{\sigma_{1}}(f_{\sigma_{2}}(q))\neq f_{\sigma_{2}}(f_{\sigma_{1}}(q))$ for some $\sigma_{1},\sigma_{2}\in\Sigma,q\in Q.$ We want to show that this discrepancy can be “continued” to a violation of symmetry. Let $q_{1}$ denote $f_{\sigma_{2}}(f_{\sigma_{1}}(q))$ and $q_{2}$ denote $f_{\sigma_{1}}(f_{\sigma_{2}}(q)).$ First, since $q$ is accessible, there exists some string $w_{\ell}$ such that $f_{w_{\ell}}(q_{0})=q.$ Second, since $q_{1}$ and $q_{2}$ are distinguishable, there exists some string $w_{r}$ such that $\beta(f_{w_{r}}(q_{1}))\neq\beta(f_{w_{r}}(q_{2}))$. Now putting things together we have $\beta(f_{w_{\ell}\sigma_{1}\sigma_{2}w_{r}}(q_{0}))=\beta(f_{\sigma_{1}\sigma_{2}w_{r}}(q))=\beta(f_{w_{r}}(q_{1})).$ Similarly $\beta(f_{w_{\ell}\sigma_{2}\sigma_{1}w_{r}}(q_{0}))=\beta(f_{w_{r}}(q_{2}))\neq\beta(f_{w_{r}}(q_{1})).$ Hence $\mathcal{A}$ outputs different values under the inputs $w_{\ell}\sigma_{1}\sigma_{2}w_{r}$ and $w_{\ell}\sigma_{2}\sigma_{1}w_{r}$; since these inputs are permutations of one another, this means $\mathcal{A}$ is not symmetric. ∎ ## 4 Proof of Theorem 7 We are given that $\mathcal{A}=(\Sigma,Q,q_{0},f,O,\beta)$ is a symmetric FSA and without loss of generality it is minimal. For each $q\in Q,$ define $r[q]\in\Sigma^{*}$ to be a fixed _representative string_ that generates state $q$ from $q_{0}$, i.e., such that $f_{r[q]}(q_{0})=q$ holds. Each $r[q]$ is guaranteed to exist since $q$ is accessible. These $r[q]$ remain fixed for the remainder of the proof. We need the following claim, which roughly says that every string $w$ is interchangeable with the representative string $r[f_{w}(q_{0})]$. We know they are interchangeable when they are read first, but using the commutativity of the $f$’s, we can show they are interchangeable when read later. ###### Claim 12. For each $w\in\Sigma^{*}$ we have $f_{w}=f_{r[f_{w}(q_{0})]}.$ ###### Proof. For any $q\in Q,$ alternately applying the definition of $r[\cdot]$ and the commutativity of the $f$’s, we have $\displaystyle f_{w}(q)$ $\displaystyle=f_{w}(f_{r[q]}(q_{0}))=f_{r[q]}(f_{w}(q_{0}))$ $\displaystyle=f_{r[q]}(f_{r[f_{w}(q_{0})]}(q_{0}))=f_{r[f_{w}(q_{0})]}(f_{r[q]}(q_{0}))=f_{r[f_{w}(q_{0})]}(q).$ ∎ ### 4.1 The Construction Here we define the divide-and-conquer automaton $\mathcal{A}^{\prime}=(\Sigma,Q^{\prime},\alpha,c,O,\beta^{\prime})$. Namely, let $Q^{\prime}=Q$, $\beta^{\prime}=\beta$, define $\alpha(\sigma):=f_{\sigma}(q_{0})$ and define $c(q,q^{\prime}):=f_{r[q^{\prime}]}(q).$ It remains to prove that the construction is correct, i.e., that $\nu_{\mathcal{A}}=\nu_{\mathcal{A}^{\prime}}.$ Our recursive proof uses the idea outlined previously, that each string $w$ is essentially interchangeable with $r[f_{w}(q_{0})].$ ###### Claim 13. For any nonempty string $w\in\Sigma^{+},$ the set $\chi_{\mathcal{A}^{\prime}}(w)$ is a singleton and $\mathtt{the.member}(\chi_{\mathcal{A}^{\prime}}(w))=f_{w}(q_{0})$. ###### Proof. We proceed by induction on $|w|$. Base case: If $w$ has length 1, say it consists of the character $\sigma$, then $f_{w}(q_{0})=f_{\sigma}(q_{0})$, and by the definition of $\chi$, we have $\chi_{\mathcal{A}^{\prime}}(w)=\\{\alpha(\sigma)\\}=\\{f_{\sigma}(q_{0})\\}.$ Thus the claim is satisfied. Inductive step: Now $w$ has length 2 or more. The induction statement to be proved is $\chi_{\mathcal{A}^{\prime}}(w)=\\{f_{w}(q_{0})\\}.$ Recalling Equation (1), which defines $\chi$ in this case, this is equivalent to saying that for all partitions $w=uv$ of $w$ into two nonempty substrings, $\displaystyle c(\mathtt{the.member}(\chi_{\mathcal{A}^{\prime}}(u)),\mathtt{the.member}(\chi_{\mathcal{A}^{\prime}}(v)))=f_{w}(q_{0}).$ (4) By the induction hypothesis, the left-hand side of (4) is equal to $c(f_{u}(q_{0}),f_{v}(q_{0})).$ (5) Applying the definition of $c$, we find that the value (5) is in turn equal to $f_{r[f_{v}(q_{0})]}(f_{u}(q_{0})).$ Finally, applying Claim 12 we see that the value (5) is equal to $f_{v}(f_{u}(q_{0}))=f_{w}(q_{0})$, as desired. ∎ ###### Proof of Theorem 7. As outlined previously, minimizing $\mathcal{A}$ makes it accessible and distinguishable, without changing $\nu_{\mathcal{A}}$. Now consider the DCA $\mathcal{A}^{\prime}$ as defined previously. On any input $w\in\Sigma^{+}$, using Claim 13, $\nu_{\mathcal{A}^{\prime}}(w)=\beta(\mathtt{the.member}(\chi_{\mathcal{A}^{\prime}}(w)))=\beta(f_{w}(q_{0}))=\nu_{\mathcal{A}}(w).$ Hence $\mathcal{A}$ and $\mathcal{A}^{\prime}$ compute the same function (i.e., they are equivalent). Since the state space of $\mathcal{A}^{\prime}$ is $Q$, and since $Q$ could only have gotten smaller when $\mathcal{A}$ was minimized, the state space of the DCA $\mathcal{A}^{\prime}$ is indeed no larger than the state space of the original FSA. ∎ One might question whether any result similar to Theorem 7 is possible if we discard the symmetry requirement. The following result gives a negative answer to this question and shows that the exponential state space increase of Theorem 5 is best possible. ###### Proposition 14. For any integer $n\geq 1$, there is an $n$-state FSA $\mathcal{A}$ on a three- symbol alphabet $\Sigma$ so that any DCA equivalent to $\mathcal{A}$ has at least $n^{n}$ states. ###### Proof. Let $Q$ be a set of $n$ states and $\Sigma$ a set of size 3. Dénes [2] showed that $Q^{Q}$, viewed as a semigroup under the operation of composition, has a generating set of size 3. We choose $\\{f_{\sigma}\\}_{\sigma\in\Sigma}$ to be this generating set; this implies that for every function $g:Q\to Q$, there is a string $w[g]\in\Sigma^{*}$ so that $f_{w[g]}=g$. We define $O=Q$, $\beta$ to be the identity function, and we choose $q_{0}\in Q$ arbitrarily; this completes the definition of the FSA $\mathcal{A}$. Suppose for the sake of contradiction that there exists a DCA $\mathcal{A}^{\prime}$ that computes $\nu_{\mathcal{A}}$, and that this DCA’s state space $Q^{\prime}$ has $|Q^{\prime}|<n^{n}$. By the pigeonhole principle there are two distinct functions $g_{1},g_{2}\in Q^{Q}$ so that $\chi_{\mathcal{A}^{\prime}}(w[g_{1}])\cap\chi_{\mathcal{A}^{\prime}}(w[g_{2}])\neq\emptyset$, since each $\chi_{\mathcal{A}^{\prime}}(\cdot)$ is a nonempty subset of $Q^{\prime}$. Let $\hat{q}\in Q$ denote a state for which $g_{1}(\hat{q})\neq g_{2}(\hat{q})$ and let $q^{\prime}\in Q^{\prime}$ denote any element of $\bigcap_{i=1,2}\chi_{\mathcal{A}^{\prime}}(w[g_{i}])$. Now let $h:Q\to Q$ be any function for which $h(q_{0})=\hat{q}$. We claim that the two input strings $w[h]w[g_{i}]$ for $i=1,2$ produce different outputs under $\mathcal{A}$ and the same output under $\mathcal{A}^{\prime}$, providing the contradiction. To see that the outputs under $\mathcal{A}$ are different, observe that $\nu_{\mathcal{A}}(w[h]w[g_{i}])=\beta(f_{w[h]w[g_{i}]}(q_{0}))=\beta(g_{i}(h(q_{0})))=g_{i}(\hat{q})$ and since $g_{1}(\hat{q})\neq g_{2}(\hat{q})$, we are done. To see that the outputs under $\mathcal{A}^{\prime}$ are the same, let $\overline{q}^{\prime}$ denote any element of $\chi_{\mathcal{A}^{\prime}}(w[h])$ and notice that $c(\overline{q}^{\prime},q^{\prime})\in\chi_{\mathcal{A}^{\prime}}(w[h]w[g_{i}])$ for $i=1,2$; then recalling Equations (2) and (3), we see that $\nu_{\mathcal{A}^{\prime}}(w[h]w[g_{1}])=\nu_{\mathcal{A}^{\prime}}(w[h]w[g_{2}])$ as claimed. ∎ ## 5 Extensions We mention in this sections some extensions of FSAs and ask if analogues of Theorem 7 hold for them. Some of these issues were raised previously in [10]. First, the main result of this paper is not suitable in the following natural situation. Suppose the input alphabet and state space are both the set of all $k$-bit binary strings, i.e. $\Sigma=Q=\\{0,1\\}^{k}$, and that the transition function $f_{\sigma}(q)$ is some polynomial-time Turing-computable function of $\sigma$ and $q$ (and similarly for $\beta$). For such an FSA, $\nu_{\mathcal{A}}(w)$ can be computed in $|w|\cdot poly(k)$ time. If $\mathcal{A}$ is symmetric we can simulate it by a DCA using Theorem 7 but this approach takes exponential time in $k$, since minimizing $\mathcal{A}$ requires looking at all of its $2^{k}$ states. Functional composition (Theorem 5) has the same issue. Thus, the open problem is to determine if a $poly(k)$-time technique exists to convert a symmetric FSA of this type into a DCA. Second, a variant of the above model might allow the string lengths to grow as some function $k(m)$ of the total number of inputs $m$. Since the original submission of this paper and independently of our work, Feldman et al. [3] showed that for this sort of model, an analogue of Theorem 7 holds where the divide-and-conquer version uses strings of length at most $k^{2}(m)$. Their construction, like ours, takes exponential time in $k(m)$. Finally, the functional composition view of FSAs (e.g., in the proof of Theorem 5) also works for nondeterministic automata and probabilistic automata. A result obtained by Feldman et al. [3] shows that an analogue of Theorem 7 for probabilistic automata is false, while the nondeterministic version appears to be an open problem. ## References * [1] Dana Angluin, James Aspnes, Zoë Diamadi, Michael J. Fischer, and René Peralta. (2006). Computation in networks of passively mobile finite-state sensors. Distributed Computing, 18(4):235–253. Preliminary version (2004) appeared in _Proc. 23rd PODC_ , pages 290–299. * [2] J. Dénes. (1968). On transformations, transformation-semigroups and graphs. In P. Erdős and G. Katona, editors, Theory of Graphs (Proc. Colloq., Tihany, 1966), pages 65–75. Academic Press. * [3] J. Feldman, S. Muthukrishnan, A. Sidiropoulos, C. Stein, and Z. Svitkina. (2008). On distributing symmetric streaming computations. In Proc. 19th SODA, pages 710–719. * [4] J.E. Hopcroft and J.D. Ullman. (1979). Introduction to Automata Theory, Languages, and Computation. Addison-Wesley, Reading, Massachusetts. * [5] John E. Hopcroft. (1971). An $n$ log $n$ algorithm for minimizing the states in a finite automaton. In Z. Kohavi, editor, The Theory of Machines and Computations, pages 189–196. Academic Press. * [6] Donald E. Knuth, James H. Morris Jr., and Vaughan R. Pratt. (1977). Fast pattern matching in strings. SIAM J. Comput., 6(2):323–350. * [7] Richard E. Ladner and Michael J. Fischer. (1980). Parallel prefix computation. J. ACM, 27(4):831–838. Preliminary version (1977) appeared in _Proc. 6th International Conf. Parallel Processing_ , pages 218–223. * [8] David L. Milgram. (1975). Web automata. Information and Control, 29(2):162–184. * [9] Z. George Mou and Sevan G. Ficici. (1995). A scalable divide-and-conquer parallel algorithm for finite state automata and its applications. In Proc. 6th Conf. Parallel Processing for Scientific Computing, pages 193–194. * [10] David Pritchard and Santosh Vempala. (2006). Symmetric network computation. In Proc. 18th SPAA, pages 261–270.

arxiv-papers

2024-09-04T02:48:53.065090

0708.2658

# On positive face structures and positive-to-one computads Marek Zawadowski Instytut Matematyki, Uniwersytet Warszawski ul. S.Banacha 2, 00-913 Warszawa, Poland zawado@mimuw.edu.pl ###### Abstract We introduce the notion of a positive face structure. The positive face structures to positive-to-one computads are like simple graphs, c.f. [MZ], to free ${\omega}$-categories over ${\omega}$-graphs. In particular, they allow to give an explicit combinatorial description of positive-to-one computads. Using this description we show, among other things, that positive-to-one computads form a presheaf category with the exponent category being the category of principal positive face structures. We also present the Harnik argument in this context showing that the ${\omega}$-categories are monadic over positive-to-one computads with the ’free functor’ being the inclusion ${\bf Comp}^{+/1}\rightarrow{{\omega}Cat}$. ###### Contents 1. 1 Introduction 2. 2 Positive hypergraphs 3. 3 Positive face structures 4. 4 Atlas for $\gamma$ and $\delta$ 5. 5 Combinatorial properties of positive face structures 6. 6 The ${\omega}$-categories generated by the positive face structures 7. 7 Normal positive face structures 8. 8 Decomposition of positive face structures 9. 9 $S^{*}$ is a positive-to-one computad 10. 10 The inner-outer factorization in ${{\cal C}types}^{+/1}_{\omega}$ 11. 11 The terminal positive-to-one computad 12. 12 A description of the positive-to-one computads 13. 13 Positive-to-one computads form a presheaf category 14. 14 The principal pushouts 15. 15 Yet another full nerve of the ${\omega}$-categories 16. 16 A monadic adjunction 17. 17 Appendix ## 1 Introduction In this paper we present a combinatorial description of the category of the positive-to-one computads ${\bf Comp}^{+/1}$. We show, that this category is a presheaf category and we describe its exponent category in a very simple combinatorial way as the category of positive principal face structures ${\bf pFs}^{+/1}$, see section 3. However the proof of that requires some extended studies of the category of all positive face structures. Intuitively, the (isomorphism classes of) positive face structures correspond to the types of arbitrary cells in positive-to-one computads. The notion of a positive face structure in the main notion introduced in this paper. We describe in a combinatorial way, the embedding functor ${\bf e}:{\bf Comp}^{+/1}\rightarrow{{\omega}Cat}$ of the category of positive-to-one computads into the category into the ${\omega}$-categories as the left Kan extension along a suitable functor ${\bf j}$, and its right adjoint as the restriction along ${\bf j}$. We end by adopting an argument due to V.Harnik to show that the right adjoint to ${\bf e}$ is monadic. This approach does not cover the problem of the cells with empty domains which is important for both Makkai’s multitopic categories and Baez-Dolan opetopic categories. However it keeps something from the simplicity of the Joyal’s $\theta$-categories, i.e. the category of positive face structures or rather the category of positive computypes ${{\cal C}types}^{+/1}_{\omega}$, the full image of the former in ${{\omega}Cat}$ is not much more complicated than the category of simple ${\omega}$-categories, the dual of the category of disks, c.f. [J], [MZ], [Be]. In this sense this paper may be considered as a step towards a comparison of these two approaches. ### Positive face structures Positive face structures represent all possible shapes of cells in positive- to-one computads. A positive face structure $S$ of dimension $2$ can be pictured as a figure $s_{6}\vphantom{s_{4}}$$\vphantom{s_{6}}s_{4}$$x_{6}$$s_{5}$$x_{8}$$x_{7}$$\Downarrow a_{2}$$\Downarrow a_{1}$$s_{7}\vphantom{\phantom{s_{7}}}$$x_{4}$$x_{9}$$x_{5}$$s_{2}\vphantom{s_{0}}$$\vphantom{s_{2}}s_{0}$$x_{0}$$s_{1}$$x_{2}$$x_{1}$$\Downarrow a_{0}$$S:$$s_{3}\vphantom{\phantom{s_{2}}}$$x_{3}$ and a positive face structure $T$ of dimension $3$ can be pictured as a figure $t_{3}\vphantom{t_{1}}$$\vphantom{t_{3}}t_{1}$$y_{3}$$t_{2}$$y_{5}$$y_{4}$$\Downarrow\\!b_{3}$$t_{4}\vphantom{}$$y_{6}$$\vphantom{}t_{0}$$y_{1}$$y_{0}$$y_{2}$$\Downarrow\\!b_{2}$$\Downarrow\\!b_{1}$$\Longrightarrow$$\beta$$t_{3}\vphantom{t_{1}}$$\vphantom{t_{3}}t_{1}$$t_{2}$$y_{5}$$y_{4}$$\Downarrow\\!b_{0}$$t_{4}\vphantom{}$$y_{6}$$\vphantom{}t_{0}$$y_{1}$$y_{0}$ They have faces of various dimensions that fit together so that it make sense to compose them in a unique way. By $S_{n}$ we denote faces of dimension $n$ in $S$. Each faces $a$ has a face $\gamma(a)$ as its codomain and a non-empty set of faces $\delta(a)$ as its domain. In $S$ above we have for $a_{1}$ $\gamma(a_{1})=x_{4}\;\;\;{\rm and}\;\;\;\delta(a_{1})=\\{x_{5},x_{6},x_{9}\\}$ and in $T$ we have for $\beta$ $\gamma(\beta)=b_{0}\;\;\;{\rm and}\;\;\;\delta(\beta)=\\{b_{1},b_{2},b_{3}\\}$ This is all the data we need. Moreover, these (necessarily finite) data satisfy four conditions (see Section 3 for details). Below we explain them in an intuitive way. Globularity. This is the main condition. It relate the sets that are obtained by duble application of $\gamma$ and $\delta$. They are $\gamma\gamma(a)=\gamma\delta(a)-\delta\delta(a),\hskip 42.67912pt\delta\gamma(a)=\delta\delta(a)-\gamma\delta(a).$ Let us look how it works for two faces $a_{1}$ and $\beta$. In case of the face $a_{1}$ we have $\gamma\delta(a_{1})=\\{s_{3},s_{4},s_{6}\\},\hskip 42.67912pt\delta\delta(a_{1})=\\{s_{4},s_{6},s_{7}\\}$ $\gamma\gamma(a_{1})=s_{3},\hskip 42.67912pt\delta\gamma(a_{1})=\\{s_{7}\\}$ So we have indeed $\delta\delta(a_{1})-\gamma\delta(a_{1})=\\{s_{4},s_{6},s_{7}\\}-\\{s_{3},s_{4},s_{6}\\}=\\{s_{7}\\}=\delta\gamma(a_{1})$ $\gamma\delta(a_{1})-\delta\delta(a_{1})=\\{s_{3},s_{4},s_{6}\\}-\\{s_{4},s_{6},s_{7}\\}=\\{s_{3}\\}=\\{\gamma\gamma(a_{1})\\}$ Similarly for the face $\beta$ we have $\gamma\gamma(\beta)=y_{0},\hskip 42.67912pt\delta\gamma(\delta)=\\{y_{1},y_{4},y_{5},y_{6}\\}$ $\gamma\delta(\beta)=\\{y_{0},y_{2},y_{3}\\},\hskip 42.67912pt\delta\delta(\beta)=\\{y_{1},y_{2},y_{3},y_{4},y_{5},y_{6}\\}$ and hence $\gamma\delta(\beta)-\delta\delta(\beta)=\\{y_{0},y_{2},y_{3}\\}-\\{y_{1},y_{2},y_{3},y_{4},y_{5},y_{6}\\}=\\{y_{0}\\}=\\{\gamma\gamma(\beta)\\}$ $\delta\delta(\beta)-\gamma\delta(\beta)=\\{y_{1},y_{2},y_{3},y_{4},y_{5},y_{6}\\}-\\{y_{0},y_{2},y_{3}\\}=\\{y_{1},y_{4},y_{5},y_{6}\\}=\delta\gamma(\beta)$ As we see in both cases $a_{1}$ and $\beta$ the first actual formula is a bit more baroque (due to the curely brackets around $\gamma\gamma(a_{1})$, $\gamma\gamma(\beta)$) that in the globularity condition stated above. However in the following we omit curely bracket on purpose to have a simpler notation hoping that it will contribute to simplicity without messing things up. Using $\delta$’s and $\gamma$’s we can define two binary relations $<^{+}$ and $<^{-}$ on faces of the same dimension which are transitive closures of the relations $\lhd^{+}$ and $\lhd^{-}$, respectively. $a\lhd^{+}b$ holds iff there is a face $\alpha$ such that $a\in\delta(\alpha)$ and $\gamma(\alpha)=b$, and $a\lhd^{-}b$ holds iff $\gamma(a)\in\delta(b)$. We call $<^{+}$ the upper order and $<^{-}$ the lower order. The following three conditions refer to these relations. Strictness. In each dimension, the relation $<^{+}$ is a strict order. The relation $<^{+}$ on $0$-dimensional faces is required to be a linear order. Disjointness. This condition says that no two faces can be comparable with respect to both orders $<^{+}$ and $<^{-}$. Pencil linearity. This final condition says that the sets of cells with common codomain ($\gamma$-pencil) and the sets of cells that have the same distinguished cell in the domain ($\delta$-pencil) are linearly ordered by $<^{+}$. The morphism of positive face structures are functions that preserves dimensions and operations $\gamma$ and $\delta$. The size of a positive face structure $S$ is defined as an infinite sequence of natural numbers $size(S)=\\{size(S)_{k}\\}_{k\in{\omega}}=\\{S_{k}-\delta(S_{k+1})\\}_{k\in{\omega}}$ (almost all equal $0$). We order the sequences lexicographically with higher dimensions being more important. The induction on the size of face structures is a convenient way of reasoning about positive face structures. Dimension of a face structure $S$ is the index of the largest non-zero number in the sequence $size(S)$. If for $k\leq dim(S)$ ($k<dim(S)$), $size(S)_{k}=1$ then $S$ is principal (normal). The normal positive face structures plays role of the pasting diagram in [HMP] and the principal positive face structures plays role of the (positive) multitopes. Note that, contrary to [HMP], we do not consider either the empty-domain multitopes or the pasting diagrams. The precise connection between these two approaches will be described elsewhere. On positive face structures we define operations of the domain, codomain, and special pushouts which plays the role of composition. With these operations (isomorphisms classes of) the positive face structures form the terminal positive-to-one computad, and at the same time the monoidal globular category in the sense of Batanin. ### Categories and functors We shall define the following categories ${\bf Comp}^{+/1}\vphantom{{{\omega}Cat}}$$\vphantom{{\bf Comp}^{+/1}}{{\omega}Cat}$${\bf e}$${\bf Fs}^{+/1}\vphantom{{{\cal C}types}^{+/1}_{\omega}}$$\vphantom{{\bf Fs}^{+/1}}{{\cal C}types}^{+/1}_{\omega}$${\bf j}$$(-)^{*}$${\bf pFs}^{+/1}\vphantom{{\cal S}}$$\vphantom{{\bf pFs}^{+/1}}{\cal S}$${\bf i}$${\bf k}$ where ${\bf pFs}^{+/1}$ is the category of principal positive face structures, ${\bf Fs}^{+/1}$ is the category of positive face structures, ${\cal S}$ is the category of simple categories c.f. [MZ], $(-)^{*}$ is the embedding functor of positive face structures into positive-to-one computads, ${\bf e}$ is the inclusion functor, ${{\cal C}types}^{+/1}_{\omega}$ is the full image of the composition functor $(-)^{*};{\bf e}$, with the non-full embedding ${\bf j}$. Having these functor we can form the following diagram $Set^{({\bf pFs}^{+/1})^{op}}\vphantom{sPb({\cal S}^{op},Set)}$$\vphantom{Set^{({\bf pFs}^{+/1})^{op}}}sPb({\cal S}^{op},Set)$$sPb(({\bf Fs}^{+/1})^{op},Set)\vphantom{sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)}$$\vphantom{sPb(({\bf Fs}^{+/1})^{op},Set)}sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$$Lan_{\bf j}$${\bf i}^{*}$$Ran_{\bf k}$${\bf k}^{*}$$Ran_{\bf i}$${\bf Comp}^{+/1}\vphantom{{{\omega}Cat}}$$\vphantom{{\bf Comp}^{+/1}}{{\omega}Cat}$${\bf e}$$\widehat{(-)}$$\widetilde{(-)}$$\widehat{(-)}$$\widetilde{(-)}$${\bf j}^{*}$ in which all the vertical arrows comes in pairs and they are adjoint equivalences of categories. The unexplained categories in the diagram above are: $sPb(({\bf Fs}^{+/1})^{op},Set)$ \- the category of the special pullback preserving morphisms from $({\bf Fs}^{+/1})^{op}$ to $Set$ and natural transformations, $sPb({\cal S}^{op},Set)$ \- the category of the special pullback preserving morphisms from ${\cal S}^{op}$ to $Set$ and natural transformations, $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ \- the category of the special pullbacks preserving morphisms from $({{\cal C}types}^{+/1}_{\omega})^{op}$ to $Set$ and natural transformations. The functors $\widehat{(-)}:{\bf Comp}^{+/1}\rightarrow sPb(({\bf Fs}^{+/1})^{op},Set)$ and $\widehat{(-)}:{{\omega}Cat}\rightarrow sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ are defined similarly, due to the embeddings $(-)^{*}$ in the previous diagram. For a computad $Q$ and an ${\omega}$-category $C$, $\widehat{Q}$ and $\widehat{C}$ are presheaves so that for a positive face structure $S$ we have $\widehat{Q}={\bf Comp}^{+/1}(S^{*},Q)\hskip 42.67912pt\widehat{C}={{\omega}Cat}(S^{*},C)$ The adjoint functors $\widetilde{(-)}$ that produce ${\omega}$-categories are slightly more complicated. They are defined in Sections 13 and 15. The other functors are standard. The functors ${\bf i}^{*}$, ${\bf j}^{*}$, ${\bf k}^{*}$ are inverse image functors. $Ran_{\bf i}$ and $Ran_{\bf k}$ are the right Kan extensions along ${\bf i}$, ${\bf k}$, respectively and $Lan_{\bf j}$ is the left Kan extension along ${\bf j}$. Since we have ${\bf e};\widehat{(-)}=\widehat{(-)};Lan_{\bf j}$, and $\widehat{(-)}$’s are equivalences of categories, the functor $Lan_{\bf j}$ is like ${\bf e}$ but moved into a more manageable context. In fact we have a very neat description of this functor. ### The content Since the paper is quite long I describe below the content of each section to help the reading. Sections 2 and 3 introduce the notion of a positive hyper- graph and positive face structure. Section 4 is concerned with establishing what kind of inclusions hold between iterated applications of $\gamma$’s and $\delta$’s. Section 5 contains many technical statements concerning positive face structure. All of them are there because they are needed afterwards. Section 6 describes the embedding $(-)^{*}:{\bf Fs}^{+/1}\rightarrow{{\omega}Cat}$ i.e. it’s main goal is to define an ${\omega}$-category $S^{*}$ for any positive face structure $S$. Section 7 describes rather technical but useful properties of normal positive face structures. In section 8 we study a way we can decompose positive face structures if they are at all decomposable. Any positive face structure is either principal or decomposable. This provides a way of proving the properties of positive face structures by induction on the size. Using this in section 9 we show that the ${\omega}$-category $S^{*}$ and in fact the whole functor $(-)^{*}$ end up in ${\bf Comp}^{+/1}$. The next two short sections 10, 11 describe just what is in their titles: factorization in ${{\cal C}types}^{+/1}_{\omega}$, and the terminal positive-to-one computad in terms of positive face structures. Section 12 gives an explicite description of all the cells in a given positive-to-one computad with the help of positive face structures. In other words, it describe in concrete terms the functor $\overline{(-)}:{\bf Comma}^{+/1}_{n}\longrightarrow{{\bf Comp}^{+/1}_{n}}$. Section 13 establishes the equivalence of categories between ${\bf Comp}^{+/1}$ and the category of presheaves over ${\bf pFs}^{+/1}$. In Section 14 the principal pullbacks are introduced and the V.Harnik’s argument in the present context is presented. Section 15 describes a full nerve functor $\widehat{(-)}:{{\omega}Cat}\longrightarrow Set^{({{\cal C}types}^{+/1}_{\omega})^{op}}$ and identifies its essential image as the special pullbacks preserving functors. Section 16 describes the inclusion functor as the left Kan extension $Lan_{\bf j}:sPb(({\bf Fs}^{+/1})^{op},Set)\longrightarrow sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ with the formulas involving just coproduct (and no other colimits). This gives as a corollary the fact that ${\bf e}:{\bf Comp}^{+/1}\rightarrow{{\omega}Cat}$ preserves connected limits. Then it is shown that the right adjoint to $Lan_{\bf j}$ ${\bf j}^{*}:sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)\longrightarrow sPb(({\bf Fs}^{+/1})^{op},Set)$ (and hence the right adjoint to ${\bf e}:{\bf Comp}^{+/1}\rightarrow{{\omega}Cat}$) is monadic. In Appendix we recall the definition of the category of positive-to-one computads. ### Acknowledgments I am very grateful to Victor Harnik and Mihaly Makkai for the conversations concerning matters of this paper. The diagrams for this paper were prepared with a help of catmac1 of Michael Barr. ### Notation In the paper we will use both directions of compositions of morphisms in categories. But each time we will write which way we compose the morphism. So, for the composition of a pair of morphism $x\stackrel{{\scriptstyle f}}{{\longrightarrow}}y\stackrel{{\scriptstyle g}}{{\longrightarrow}}z$ we can write either $f;g$ or $g\circ f$. ${\omega}$ is the set of natural numbers. ## 2 Positive hypergraphs A positive hypergraph $S$ is a family $\\{S_{k}\\}_{k\in{\omega}}$ of finite sets of faces, a family of functions $\\{\gamma_{k}:S_{k+1}\rightarrow S_{k}\\}_{k\in{\omega}}$, and a family of total relations $\\{\delta_{k}:S_{k+1}\rightarrow S_{k}\\}_{0\leq k<n}$. Moreover $\delta_{0}:S_{1}\rightarrow S_{0}$ is a function and only finitely many among sets $\\{S_{k}\\}_{k\in{\omega}}$ are non-empty. As it is always clear from the context we shall never use the indices of the functions $\gamma$ and $\delta$. A morphism of positive hypergraphs $f:S\longrightarrow T$ is a family of functions $f_{k}:S_{k}\longrightarrow T_{k}$, for $k\in{\omega}$, such that the diagrams $S_{k}\vphantom{T_{k}}$$\vphantom{S_{k}}T_{k}$$f_{k}$$S_{k+1}\vphantom{T_{k+1}}$$\vphantom{S_{k+1}}T_{k+1}$$f_{k+1}$$\gamma$$\gamma$$S_{k}\vphantom{T_{k}}$$\vphantom{S_{k}}T_{k}$$f_{k}$$S_{k+1}\vphantom{T_{k+1}}$$\vphantom{S_{k+1}}T_{k+1}$$f_{k+1}$$\delta$$\delta$ commute, for $k\in{\omega}$. The commutation of the left hand square is the commutation of the diagram of sets an functions but in case of the right hand square we mean more than commutation of a diagram of relations, i.e. we demand that for any $a\in S_{\geq 1}$, $f_{a}:\delta(a)\longrightarrow\delta(f(a))$ be a bijection, where $f_{a}$ is the restriction of $f$ to $\delta(a)$. The category of positive hypergraphs is denoted by ${\bf Hg}^{+1}$. When convenient and does not lead to confusions, for $a\in S_{k}$ we sometime treat $\gamma(a)$ as an element of $S_{k-1}$ and sometimes as a subset $\\{\gamma(a)\\}$ of $S_{k-1}$. Before we go on, we need some notation. Let $S$ be a positive hypergraph. 1. 1. The dimension of $S$ is $max\\{k\in{\omega}:S_{k}\neq\emptyset\\}$, and it is denoted by $dim(S)$. 2. 2. The sets of faces of different dimensions are assumed to be disjoint (i.e. $S_{k}\cap S_{l}=\emptyset$, for $k\neq l$). $S$ is also used to mean the set of all faces of $S$, i.e. $\bigcup_{k=0}^{n}S_{k}$; the notation $A\subseteq S$ mean that $A$ is a set of some faces of $S$; $A_{k}=A\cap S_{k}$, for $k\in{\omega}$. 3. 3. If $a\in S_{k}$ then the face $a$ has dimension $k$ and we write $dim(a)=k$. 4. 4. For $a\in S_{\geq 1}$ the set $\theta(a)=\delta(a)\cup\gamma(a)$ is the set of codimension $1$ faces in $a$. 5. 5. $S_{\geq k}=\bigcup_{i\geq k}S_{i}$, $\;\;\;\;\;S_{\leq k}=\bigcup_{i\leq k}S_{i}$. The set $S_{\leq k}=\bigcup_{i\leq k}S_{i}$ is closed under $\delta$ and $\gamma$ so it is a sub-hypergraph of $S$, called $k$-truncation of $S$. 6. 6. The image of $A\subseteq S$ under $\delta$ and $\gamma$ will be denoted by $\delta(A)=\bigcup_{a\in A}\delta(a),\hskip 14.22636pt\gamma(A)=\\{\gamma(a)\,:\,a\in A\\},$ respectively. In particular $\delta\delta(a)=\bigcup_{x\in\delta(a)}\delta(x)$, $\gamma\delta(a)=\\{\gamma(x):x\in\delta(a)\\}$. 7. 7. $\iota(a)=\delta\delta(a)\cap\gamma\delta(a)$ is the set of internal faces of the face $a\in S_{\geq 2}$. 8. 8. On each set $S_{k}$ we introduce two binary relations $<^{S_{k},-}$ and $<^{S_{k},+}$, called lower and upper order, respectively. We usually omit $k$ and even $S$ in the superscript. 1. (a) $<^{S_{0},-}$ is the empty relation. For $k>0$, $<^{S_{k},-}$ is the transitive closure of the relation $\lhd^{S_{k},-}$ on $S_{k}$, such that $a\lhd^{S_{k},-}b$ iff $\gamma(a)\in\delta(b)$. We write $a\perp^{-}b$ iff either $a<^{-}b$ or $b<^{-}a$, and we write $a\leq^{-}b$ iff either $a=b$ or $a<^{-}b$. 2. (b) $<^{S_{k},+}$ is the transitive closure of the relation $\lhd^{S_{k},+}$ on $S_{k}$, such that $a\lhd^{S_{k},+}b$ iff there is $\alpha\in S_{k+1}$, such that $a\in\delta(\alpha)$ and $\gamma(\alpha)=b$. We write $a\perp^{+}b$ iff either $a<^{+}b$ or $b<^{+}a$, and we write $a\leq^{+}b$ iff either $a=b$ or $a<^{+}b$. 3. (c) $a\not\perp b$ if both conditions $a\not\perp^{+}b$ and $a\not\perp^{-}b$ hold. 9. 9. Let $a,b\in S_{k}$. A lower path $a_{0},\ldots,a_{m}$ from $a$ to $b$ in $S$ is a sequence of faces $a_{0},\ldots,a_{m}\in S_{k}$ such that $a=a_{0}$, $b=a_{m}$ and for $\gamma(a_{i-1})\in\delta(a_{i})$, $i=1,\ldots,m$. 10. 10. Let $x,y\in S_{k}$. An upper path $x,a_{0},\ldots,a_{m},y$ from $x$ to $y$ in $S$ is a sequence of faces $a_{0},\ldots,a_{m}\in S_{k+1}$ such that $x\in\delta(a_{0})$, $y=\gamma(a_{m})$ and $\gamma(a_{i-1})\in\delta(a_{i})$, for $i=1,\ldots,m$. 11. 11. The iterations of $\gamma$ and $\delta$ will be denoted in two different ways. By $\gamma^{k}$ and $\delta^{k}$ we mean $k$ applications of $\gamma$ and $\delta$, respectively. By $\gamma^{(k)}$ and $\delta^{(k)}$ we mean the application as many times $\gamma$ and $\delta$, respectively, to get faces of dimension $k$. For example in $a\in S_{5}$ then $\delta^{3}(a)=\delta\delta\delta(a)\subseteq S_{2}$ and $\delta^{(3)}(a)=\delta\delta(a)\subseteq S_{3}$. 12. 12. For $l\leq k$, $a,b\in S_{k}$ we define $a<_{l}b$ iff $\gamma^{(l)}(a)<^{-}\gamma^{(l)}(b)$. 13. 13. A face $a$ is unary iff $\delta(a)$ is a singleton. We have ###### Lemma 2.1 If $S$ is an hypergraph and $k\in{\omega}$, then $<^{S_{k+1},-}$ is a strict partial order iff $<^{S_{k},+}$ is a strict partial order. ## 3 Positive face structures To simplify the notation, we treat both $\delta$ and $\gamma$ as functions acting on faces as well as on sets of faces, which means that sometimes we confuse elements with singletons. Clearly, both $\delta$ and $\gamma$ when considered as functions on sets are monotone. A positive hypergraph $S$ is a positive face structure if it is non-empty, i.e. $S_{0}\neq\emptyset$ and 1. 1. Globularity: for $a\in S_{\geq 2}$: $\gamma\gamma(a)=\gamma\delta(a)-\delta\delta(a),\hskip 42.67912pt\delta\gamma(a)=\delta\delta(a)-\gamma\delta(a);$ 2. 2. Strictness: for $k\in{\omega}$, the relation $<^{S_{k},+}$ is a strict order; $<^{S_{0},+}$ is linear; 3. 3. Disjointness: for $k>0$, $\perp^{S_{k},-}\cap\perp^{S_{k},+}=\emptyset$ 4. 4. Pencil linearity: for any $k>0$ and $x\in S_{k-1}$, the sets $\\{a\in S_{k}\;|\;x=\gamma(a)\\}\;\;\;\;{\rm and}\;\;\;\;\\{a\in S_{k}\;|\;x\in\delta(a)\\}$ are linearly ordered by $<^{S_{k},+}$. $S$ is a weak positive face structure if $S$ is a globular, strict, disjoint, positive hypergraph (i.e. pencil linearity is not required to hold). The category of (weak) positive face structures is the full subcategory of ${\bf Hg}^{+1}$ whose objects are the (weak) positive face structures and is denoted by denoted by ${\bf Fs}^{+/1}$ (${\bf wFs}^{+/1}$). Remarks. 1. 1. The reason why we call the first condition ’globularity’ is that it will imply the usual globularity condition in the ${\omega}$-categories generated by positive face structures. 2. 2. For each $k\in{\omega}$, the $k$-truncation of a weak positive face structure $S$ is again a weak positive face structure $S_{\leq k}$. In particular, any $k$-truncation of a positive face structure $S$ is a weak positive face structure $S_{\leq k}$, but it does not necessarily satisfy local linearity condition. 3. 3. Note that if we were to assume that each positive face structure has a single cell of dimension $-1$ then linearity of $<^{S_{0},+}$ would become a special case of pencil linearity. 4. 4. The fact that, for $x\in S_{k-1}$, the set $\\{a\in S_{k}\;|\;x=\gamma(a)\\}$ is linearly ordered is sometimes referred to as $\gamma$-linearity of $<^{S_{k},+}$, and the fact that the set $\\{a\in S_{k}\;|\;\;x\in\delta(a)\\}$ is linearly ordered is sometimes referred to as $\delta$-linearity of $<^{S_{k},+}$. 5. 5. If $S$ has dimension $n$, as hypergraph, then we say that $S$ is an $n$-face structure. 6. 6. The size of positive face structure $S$ is the sequence natural numbers $size(S)=\\{|S_{n}-\delta(S_{n+1})|\\}_{n\in{\omega}}$, with almost all being equal $0$. We have an order $<$ on such sequences, so that $\\{x_{n}\\}_{n\in{\omega}}<\\{y_{n}\\}_{n\in{\omega}}$ iff there is $k\in{\omega}$ such that $x_{k}<y_{k}$ and for all $l>k$, $x_{l}=y_{l}$. This order is well founded and many facts about positive face structures will be proven by induction on the size. 7. 7. Let $S$ be a positive face structure. $S$ is $k$-principal iff $size(S)_{l}=1$, for $l\leq k$. $S$ is principal iff $S$ is $dim(S)$-principal. $S$ is normal iff $S$ is $(dim(S)-1)$-principal. $S$ is $k$-normal iff $S$ is normal and $dim(S)=k$. By ${\bf pFs}^{+/1}$ (${\bf nFs}^{+/1}$) we denote full subcategories of ${\bf Fs}^{+/1}$ whose objects are principal (normal) positive face structures. ## 4 Atlas for $\gamma$ and $\delta$ We have an easy ###### Lemma 4.1 Let $S$ be a positive face structure, $a\in S_{n}$, $n>1$. Then 1. 1. the sets $\delta\gamma(a)$, $\iota(a)$, and $\gamma\gamma(a)$ are disjoint; $\delta\gamma$$\iota$$\gamma\gamma$ 2. 2. $\delta\delta(a)=\delta\gamma(a)\cup\iota(a)$; 3. 3. $\gamma\delta(a)=\gamma\gamma(a)\cup\iota(a)$. Proof. These are immediate consequences of globularity. $~{}~{}\Box$ Moreover ###### Lemma 4.2 Let $S$ be a positive face structure, $a\in S_{n}$, $n>2$. Then we have 1. 1. $\delta\gamma\gamma(a)\subseteq\delta\gamma\delta(a)\subseteq\delta\gamma\gamma(a)\cup\iota\gamma(a)=\delta\delta\gamma(a)=\delta\delta\delta(a)$; 2. 2. $\gamma\gamma\gamma(a)\subseteq\gamma\gamma\delta(a)\subseteq\gamma\gamma\gamma(a)\cup\iota\gamma(a)=\gamma\delta\gamma(a)=\gamma\delta\delta(a)$. Proof. From globularity we have $\gamma\gamma(\alpha)\subseteq\gamma\delta(\alpha)$. Thus by monotonicity of $\delta$ and $\gamma$ we get $\gamma\gamma\gamma(\alpha)\subseteq\gamma\gamma\delta(\alpha)\hskip 14.22636pt{\rm and}\hskip 14.22636pt\delta\gamma\gamma(\alpha)\subseteq\delta\gamma\delta(\alpha)\hskip 14.22636pt{\rm and}\hskip 14.22636pt\gamma\gamma\delta(\alpha)\subseteq\gamma\delta\delta(\alpha).$ Similarly, as we have from globularity: $\delta\gamma(\alpha)\subseteq\delta\delta(\alpha)$ it follows by monotonicity of $\delta$ and $\gamma$: $\gamma\delta\gamma(\alpha)\subseteq\gamma\delta\delta(\alpha)\hskip 14.22636pt{\rm and}\hskip 14.22636pt\delta\delta\gamma(\alpha)\subseteq\delta\delta\delta(\alpha)\hskip 14.22636pt{\rm and}\hskip 14.22636pt\delta\gamma\delta(\alpha)\subseteq\delta\delta\delta(\alpha).$ The equalities $\delta\gamma\gamma(a)\cup\iota\gamma(a)=\delta\delta\gamma(a)\hskip 14.22636pt{\rm and}\hskip 14.22636pt\gamma\gamma\gamma(a)\cup\iota\gamma(a)=\gamma\delta\gamma(a)$ follow easily from Lemma 4.1. Thus it remains to show that: 1. 1. $\delta\delta\gamma(a)\supseteq\delta\delta\delta(a)$, 2. 2. $\gamma\delta\gamma(a)\supseteq\gamma\delta\delta(a)$. Both inclusions can be proven similarly. We shall show the first only. Suppose contrary, that there is $u\in\delta\delta\delta(a)-\delta\delta\gamma(a)$. Let $x\in\delta(a)$ be $<^{-}$-minimal element in $\delta(a)$ such that there is $s\in\delta(x)$ with $u\in\delta(s)$. If $s\in\delta\gamma(a)$ then $u\in\delta\delta\gamma(a)$ contrary to the supposition. Thus $s\not\in\delta\gamma(a)$. Since $\delta\gamma(a)=\delta\delta(a)-\gamma\delta(a)$ it follows that $s\in\gamma\delta(a)$. Hence there is $x^{\prime}\in\delta(a)$ with $\gamma(x^{\prime})=s$. In particular $x^{\prime}<^{-}x$. Moreover $u\in\delta(s)=\delta\gamma(x^{\prime})\subseteq\delta\delta(x^{\prime}).$ Then there is $s^{\prime}\in\delta(x^{\prime})$ so that $u\in\delta(s^{\prime})$. This contradicts the $<^{-}$-minimality of $x$. $~{}~{}\Box$ From Lemma 4.2 we get ###### Corollary 4.3 Let $S$ be a positive face structure, $a\in S_{n}$, $n>2$, $k<n$. Then, with $\xi^{l}$ and $\xi^{\prime l}$ being two fixed strings of $\gamma$’s and $\delta$’s of length $l$, we have 1. 1. $\gamma^{k}(a)\subseteq\gamma\xi^{k-1}(a)$; 2. 2. $\delta\xi^{k-1}(a)\subseteq\delta^{k}(a)$; 3. 3. $\delta^{k}(a)\cap\gamma^{k}(a)=\emptyset$; 4. 4. $\xi^{k}(a)\subseteq\gamma^{k}(a)\cup\delta^{k}(a)$; 5. 5. $\delta^{2}\xi^{k-2}(a)=\delta^{2}\xi^{\prime k-2}(a)$, (e.g. $\delta^{k}(a)=\delta^{2}\gamma^{k-2}(a)$); 6. 6. $\gamma\delta\xi^{k-2}(a)=\gamma\delta\xi^{\prime k-2}(a)$, (e.g. $\gamma\delta^{k-1}(a)=\gamma\delta\gamma^{k-2}(a)$); 7. 7. $\xi^{k-2}\delta\gamma(a)=\xi^{k-2}\delta^{2}(a)$, for $k>2$; 8. 8. $\delta^{k}(a)=\delta\gamma^{k-1}(a)\cup\iota\gamma^{k-2}(a)$, for $k>1$. ## 5 Combinatorial properties of positive face structures ### Local properties ###### Proposition 5.1 Let $S$ be a positive face structure, $k>0$ and $\alpha\in S_{k}$, $a_{1},a_{2}\in\delta(\alpha)$, $a_{1}\neq a_{2}$. Then we have 1. 1. $a_{1}\not\perp^{+}a_{2}$; 2. 2. $\delta(a_{1})\cap\delta(a_{2})=\emptyset$ and $\gamma(a_{1})\neq\gamma(a_{2})$. Proof. Ad 1. Suppose contrary that there are $a_{1},a_{2}\in\delta(\alpha)$ such that $a_{1}<^{+}a_{2}$. So we have an upper path $a_{1},\beta_{1},\ldots,\beta_{r},a_{2}$ and hence a lower path $\beta_{1},\ldots,\beta_{r},\alpha.$ In particular $\beta_{1}<^{-}\alpha$. As $a_{1}\in\delta(\beta_{1})\cap\delta(\alpha)$ by $\delta$-linearity we have $\beta_{1}\perp^{+}\alpha$. But then $(\alpha,\beta_{1})\in\perp^{+}\cap\perp^{-}\neq\emptyset$ i.e. $S$ does not satisfy the disjointness. This shows 1. Ad 2. This is an immediate consequence of 1. If $a_{1},a_{2}\in\delta(\alpha)$ and either $\gamma(a_{1})=\gamma(a_{2})$ or $\delta(a_{1})\cap\delta(a_{2})\neq\emptyset$ then by pencil linearity we get that $a_{1}\perp^{+}a_{2}$, contradicting 1. $~{}~{}\Box$ After proving the above proposition we can introduce more notation. Let $S$ be a positive face structure, $n\in{\omega}$. 1. 1. For a face $\alpha\in S_{n+2}$, we shall denote by $\rho(\alpha)\in\delta(\alpha)$ be the only face in $\delta(\alpha)$, such that $\gamma(\rho(\alpha))=\gamma\gamma(\alpha)$. 2. 2. $X\subseteq S_{n+1}$, $a,b\in S_{n}$ and $a,\alpha_{1},\ldots,\alpha_{k},b$ be an upper path in $S$. We say that it is a path in $X$ (or $X$-path) if $\\{\alpha_{k},\ldots,\alpha_{k}\\}\subseteq X$. ###### Lemma 5.2 Let $S$ be a positive face structure, $n\in{\omega}$, $\alpha\in S_{n+2}$, $a,b\in S_{n+1}$, $y\in\delta\delta(\alpha)$. Then 1. 1. there is a unique upper $\delta(\alpha)$-path from $y$ to $\gamma\gamma(\alpha)$; 2. 2. there is a unique $x\in\delta\gamma(\alpha)$ and an upper $\delta(\alpha)$-path from $x$ to $y$ such that $\gamma(x)=\gamma(y)$; 3. 3. if $t\in\delta(y)$ there is an $x\in\delta\gamma(\alpha)$ and an upper $\delta(\alpha)$-path from $x$ to $y$ and $t\in\delta(x)$; 4. 4. If $a<^{+}b$ then $\gamma(a)\leq^{+}\gamma(b)$. Proof. Ad 1. Uniqueness follows from Lemma 5.1. To show the existence, let us suppose contrary that there is no $\delta(\alpha)$-path from $y$ to $\gamma\gamma(\alpha)$. We shall construct an infinite upper $\delta(\alpha)$-path from $y$ $y,a_{1},a_{2},\ldots$ As $y\in\delta\delta(\alpha)$ there is $a_{1}\in\delta(\alpha)$ such that $y\in\delta(a_{1})$. So now suppose that we have already constructed $a_{1},\ldots,a_{k}$. By assumption $\gamma(a_{k})\neq\gamma\gamma(\alpha)$, so $\gamma(a_{k})\in\delta\delta(\alpha)$. Hence there is $a_{k+1}\in\delta(\alpha)$ such that $\gamma(a_{k})\in\delta(a_{k+1})$. This ends the construction of the path. As in positive face structure there are no infinite paths, this is a contradiction and if fact there is a $\delta(\alpha)$-path from $y$ to $\gamma\gamma(\alpha)$. Ad 2. Suppose not, that there is no $x\in\delta\gamma(\alpha)$ as claimed. We shall construct an infinite descending lower $\delta(\alpha)$-path $\ldots\lhd^{-}a_{1}\lhd^{-}a_{0}$ such that $\gamma(a_{0})=y$, $\gamma\gamma(a_{n})=\gamma(y)=t$, for $n\in{\omega}$. By assumption $y\not\in\delta\gamma(\alpha)=\delta\delta(\alpha)-\gamma\delta(\alpha)$. So $y\in\gamma\delta(\alpha)$. Hence there is $a_{0}\in\delta(\alpha)$ such, that $\gamma(a_{0})=y$. Now, suppose that the lower $\delta(\alpha)$-path $a_{k}\lhd^{-}a_{k-1}\lhd^{-}\ldots\lhd^{-}a_{0}$ has been already constructed. By globularity, we can pick $z\in\delta(a_{k})$, such that $\gamma(z)=t$. By assumption $z\not\in\delta\gamma(\alpha)=\delta\delta(\alpha)-\gamma\delta(\alpha)$. So $z\in\gamma\delta(\alpha)$. Hence there is $a_{k+1}\in\delta(\alpha)$ such, that $\gamma(a_{k+1})=z\in\delta(a_{k})$. Clearly, $\gamma\gamma(a_{k+1})=t$. This ends the construction of the path. But by strictness such a path has to be finite, so there is $x$ as needed. Ad 3. This case is similar. We put it for completeness. Suppose not, that there is no $x\in\delta\gamma(\alpha)$ as above. We shall construct an infinite descending lower $\delta(\alpha)$-path $\ldots\lhd^{-}a_{1}\lhd^{-}a_{0}$ such that $\gamma(a_{0})=y$, $t\in\delta\gamma(a_{n})$, for $n\in{\omega}$. By assumption $y\not\in\delta\gamma(\alpha)=\delta\delta(\alpha)-\gamma\delta(\alpha)$. So $y\in\gamma\delta(\alpha)$. Hence there is $a_{0}\in\delta(\alpha)$ such, that $\gamma(a_{0})=y$. Now, suppose that the lower $\delta(\alpha)$-path $a_{k}\lhd^{-}a_{k-1}\lhd^{-}\ldots\lhd^{-}a_{0}$ has been already constructed. By globularity, we can pick $z\in\delta(a_{k})$, such that $t\in\delta(z)$. By assumption $z\not\in\delta\gamma(\alpha)=\delta\delta(\alpha)-\gamma\delta(\alpha)$. So $z\in\gamma\delta(\alpha)$. Hence there is $a_{k+1}\in\delta(\alpha)$ such, that $\gamma(a_{k+1})=z\in\delta(a_{k})$. Clearly, $t\in\delta\gamma(a_{k+1})$. This ends the construction of the path. But by strictness such a path has to be finite, so there is $x$ as needed. Ad 4. The essential case is when $a\lhd^{+}b$. This follows from 1. Then use the induction. $~{}~{}\Box$ ###### Lemma 5.3 Let $S$ be a positive face structure, $n>1$, $\alpha\in S_{n+1}$, and $a,b\in S_{n}$ such that $a<^{+}b$. Then 1. 1. $\iota\delta(\alpha)=\iota\gamma(\alpha)$ ; 2. 2. $\iota(a)\subseteq\iota(b)$; 3. 3. $\iota(a)\cup\gamma\gamma(a)\subseteq\iota(b)\cup\gamma\gamma(b)$; 4. 4. $\delta\delta(a)\subseteq\delta\delta(b)$; 5. 5. $\theta\theta(a)\subseteq\theta\theta(b)$. Proof. Ad 1. $\iota\delta(\alpha)\subseteq\iota\gamma(\alpha)$: Fix $a\in\delta(\alpha)$ and $t\in\iota(a)$. Thus there are $x,y\in\delta(a)$ such, that $\gamma(x)=t\in\delta(y)$. By Lemma 5.2 2,3 there are $x^{\prime},y^{\prime}\in\delta\gamma(\alpha)$ such, that $x^{\prime}\leq^{+}x$, $y^{\prime}\leq^{+}y$ and $\gamma(x^{\prime})=t\in\delta(y^{\prime})$. Thus $t\in\iota\gamma(\alpha)$ and the first inclusion is proved. $\iota\delta(\alpha)\supseteq\iota\gamma(\alpha)$: Fix $t\in\iota\gamma(\alpha)$. In particular, there are $x,y\in\delta\gamma(\alpha)$, so that $\gamma(x)=t\in\delta(y)$. Suppose that $t\not\in\iota\delta(\alpha)$. We shall build an infinite $\delta(\alpha)$-path $a_{1}\lhd^{-}a_{2}\ldots$ such that $\gamma\gamma(a_{i})=t$ for $i\in{\omega}$. Since $\delta\gamma(\alpha)\subseteq\delta\delta(\alpha)$, there is $a_{1}\in\delta(\alpha)$ such, that $x\in\delta(a_{1})$. Since $t\not\in\iota\delta(\alpha)$, it follows that $\gamma\gamma(a_{1})=t$. Suppose now that we have already constructed the path $a_{1}\lhd^{-}a_{2}\ldots a_{k}$ as above. We have $\gamma\gamma(a_{k})=t\lhd^{+}\gamma(y)\leq^{+}\gamma\gamma\gamma(\alpha)$. So $\gamma(a_{k})\neq\gamma\gamma(\alpha)$ and $\gamma(a_{k})\in\delta\delta(\alpha)$. Then there is $a_{k+1}\in\delta(\alpha)$ such, that $\gamma(a_{k})\in\delta(a_{k+1})$. Again, as $t\not\in\iota\delta(\alpha)$, it follows that $\gamma\gamma(a_{k+1})=t$. This ends the construction of the path. Since, by strictness, such a path cannot exists we get the other inclusion. Ad 2. Since the inclusion is transitive, it is enough to consider the case $a\lhd^{+}b$, i.e. there is an $\alpha\in S_{n+1}$ such, that $a\in\delta(\alpha)$ and $b=\gamma(\alpha)$. Then by 1. we have $\iota(a)\subseteq\iota\delta(\alpha)=\iota\gamma(\alpha)=\iota(b)$ Ad 3. As above it is enough to consider the case $a\lhd^{+}b$, i.e. that there is $\alpha\in S_{n+1}$ such that $a\in\delta(\alpha)$ and $\gamma(\alpha)=b$. By 2. we need to show that $\gamma\gamma(a)\subseteq\iota(b)\cup\gamma\gamma(b)$. We have $\gamma\gamma(a)\in\gamma\gamma\delta(\alpha)\subseteq\iota\gamma(\alpha)\cup\gamma\gamma\gamma(\alpha)=\iota(b)\cup\gamma\gamma(b).$ Ad 4. Again it is enough to consider the case $a\lhd^{+}b$, i.e. that there is $\alpha\in S_{n+1}$ such that $a\in\delta(\alpha)$ and $\gamma(\alpha)=b$. We have $\delta\delta(a)\subseteq\delta\delta\delta(\alpha)=\delta\delta\gamma(\alpha)=\delta\delta(b).$ Ad 5. This follows from 3. and 4. $~{}~{}\Box$ ### Global properties ###### Lemma 5.4 Let $S$ be a positive face structure, $n\in{\omega}$, $a,b\in S_{n}$, $a<^{+}b$. Then, there is an upper $S_{n+1}-\gamma(S_{n+2})$-path from $a$ to $b$. Proof. Let $a,\alpha_{1},\ldots,\alpha_{k},b$ be an upper path is $S$. By Lemma 5.2 we can replace each face $\alpha_{i}$ in this path which is not in $S-\gamma(S)$ by a sequence of faces which are $<^{+}$-smaller. Just take $\Gamma\in S_{n+2}$, such that $\gamma(\Gamma)=\alpha_{i}$ and take instead of $\alpha_{i}$ a path in $\delta(\Gamma)$ from $\gamma(\alpha_{i-1})$ (if $i=0$ then from $a$) to $\gamma(\alpha_{i})$. Repeated application of this procedure will eventually yield the required path. $~{}~{}\Box$ ###### Lemma 5.5 Let $S$ be a positive face structure, $n>0$, $a\in S_{n}$, $\alpha\in S_{n+1}$, and either $\gamma(a)\in\iota(\alpha)$ or $\delta(a)\cap\iota(\alpha)\neq\emptyset$. Then $a<^{+}\gamma(\alpha)$. Moreover, if $\alpha\in S-\gamma(S)$ then there is a unique $a^{\prime}\in\delta(\alpha)$ such that $a\leq^{+}a^{\prime}$. Proof. If $a\in\delta(\alpha)$ there is nothing to prove. So we assume that $a\not\in\delta(\alpha)$. We begin with the second part of Lemma, i.e. we assume that $\alpha\in S_{n+1}-\gamma(S_{n+2})$. Let $\gamma(a)\in\iota(\alpha)$. Thus there are $b,c\in\delta(\alpha)$ such that $\gamma(a)=\gamma(b)\in\delta(c)$. In particular $a<^{-}c$. By $\gamma$-linearity either $b<^{+}a$ or $a<^{+}b$. Suppose that $b<^{+}a$. Then we have an $(S-\gamma(S))$-upper path $b,\beta_{0},\ldots,\beta_{r},a$. As $b\in\alpha\cap\beta_{0}$ and $\alpha,\beta_{0}\in S-\gamma(S)$, we have $\alpha=\beta_{0}$. But then $c\in\delta(\alpha)=\delta(\beta_{0})$ and hence $c<^{+}\gamma(\beta_{0})\leq^{+}a$. So we get $a<^{-}c$ and $c<^{+}a$ contradicting the disjointness of $\perp^{+}$ and $\perp^{-}$. Thus we can put $a^{\prime}=b$ and we have $a<^{+}a^{\prime}$. The uniqueness of $a^{\prime}$ follows from the fact that $\gamma(a)=\gamma(a^{\prime})$. The case $\delta(a)\cap\iota(\alpha)\neq\emptyset$ is similar and we put it for completeness. Thus there are $b,c\in\delta(\alpha)$ such that $\gamma(b)\in\delta(a)\cap\delta(c)$. In particular $b<^{-}a$. By $\delta$-linearity either $c<^{+}a$ or $a<^{+}c$. Suppose that $c<^{+}a$. Then we have an $(S-\gamma(S))$-upper path $c,\beta_{0},\ldots,\beta_{r},a$. As $c\in\alpha\cap\beta_{0}$ and $\alpha,\beta_{0}\in S-\gamma(S)$, we have $\alpha=\beta_{0}$. But then $b\in\delta(\alpha)=\delta(\beta_{0})$ and hence $b<^{+}\gamma(\beta_{0})\leq^{+}a$. So we get $b<^{-}a$ and $b<^{+}a$ contradicting the disjointness of $\perp^{+}$ and $\perp^{-}$. Thus we can put $a^{\prime}=c$ and we have $a<^{+}a^{\prime}$. The uniqueness of $a^{\prime}$ follows from the fact that $\gamma(b)\in\delta(a^{\prime})$ and $a^{\prime}\in\delta(\alpha)$ and Lemma 5.1. The first part of the Lemma follows from the above, Lemma 5.2.4 and the following Claim. Claim. If $\alpha\in S_{n+1}$ and $x\in\iota(\alpha)$ then there is an $\alpha^{\prime}\in S_{n+1}$ such that $\alpha^{\prime}\leq^{+}\alpha$, $x\in\iota(\alpha^{\prime})$ and $\alpha^{\prime}\not\in\gamma(S_{n+2})$. Proof of the Claim. Suppose that Claim is not true. To get a contradiction, we shall build an infinite descending $\gamma(S_{n+2})$-path $\ldots\lhd^{+}\alpha_{1}\lhd^{+}\alpha_{0}=\alpha$ such that $x\in\iota(\alpha_{i})$, for $i\in{\omega}$. We put $\alpha_{0}=\alpha$. Suppose that we have already constructed $\alpha_{0},\ldots,\alpha_{k}\in\gamma(S_{n+2})$. Hence there is $\beta\in S_{n+2}$ such, that $\gamma(\beta)=\alpha_{k}$. Since $\iota\delta(\beta)=\iota\gamma(\beta)=\iota(\alpha_{k})$, there is $\alpha_{k+1}\in\delta(\beta)$ such, that $x\in\iota(\alpha_{k+1})$. This ends the construction of the infinite path and the proof of the Claim and the Lemma. $~{}~{}\Box$ ###### Corollary 5.6 Let $S$ be a positive face structure. If $a\in S-\delta(S)$ then $\gamma(a)\in S-\iota(S)$ and $\delta(a)\subseteq S-\iota(S)$. Proof. Let $a\in S_{n}$ and $\alpha\in S_{n+2}$. If either $\gamma(a)\in\iota(\alpha)$ or $\delta(a)\cap\iota(\alpha)\neq\emptyset$ then by Lemma 5.5 $a<^{+}\gamma(\alpha)$. Thus $a\in\delta(S)$. $~{}~{}\Box$ A lower path $b_{0},\ldots,b_{m}$ is a maximal path if $\delta(b_{0})\subseteq\delta(S)-\gamma(S)$ and $\gamma(b_{m})\in\gamma(S)-\delta(S)$, i.e. if it can’t be extended either way. ###### Lemma 5.7 (Path Lemma) Let $k\geq 0$, $B=(a_{0},\ldots,a_{k})$ be a maximal $S_{n}$-path in a positive face structure $S$, $b\in S_{n}$, $0\leq s\leq k$, $a_{s}<^{+}b$. Then there are $0\leq l\leq s\leq p\leq k$ such, that 1. 1. $a_{i}<^{+}b$ for $i=l,\ldots,p$; 2. 2. $\gamma(a_{p})=\gamma(b)$; 3. 3. either $l=0$ and $\delta(a_{0})\subseteq\delta(b)$ or $l>0$ and $\gamma(a_{l-1})\in\delta(b)$; 4. 4. $\gamma(a_{i})\in\iota(S)$, for $l\leq i<p$. Proof. Let $0\leq l\leq p\leq k$ be such that $a_{i}<^{+}b$ for $l\leq i\leq p$ and either $l=0$ or $a_{l-1}\not<^{+}b$ and either $p=k$ or $a_{k+1}\not<^{+}b$. We shall show that $l$ and $p$ have the properties stated in the Lemma. From the very definition the property $1$ holds. We shall show 2. Take an upper $(S-\gamma(S))$-path from $a_{p}$ to $b$: $a_{p},\beta_{0},\ldots,\beta_{r},b$. If $\gamma(a_{p})=\gamma\gamma(\beta_{i})$, for $i=0,\ldots,r$ then $\gamma(a_{p})=\gamma\gamma(\beta_{r})=\gamma(b)$ and we are done. So suppose contrary and let $i_{0}=min\\{i:\gamma(a_{p})\neq\gamma\gamma(\beta_{i})\\}$ Then there are $a,c\in\delta(\beta_{i_{0}})$ such that $\gamma(a_{p})=\gamma(a)\in\delta(c)$ (NB. $a=a_{p}$ if $i_{0}=0$ and $a=\gamma(\beta_{i_{0}-1})$ otherwise). In particular $\gamma(a_{p})\in\iota(\beta_{i_{0}})$. As $\gamma(a_{p})\in\delta(S)$, we have $p<k$. Thus $\gamma(a_{p})\in\delta(a_{p+1})\cap\iota(\beta_{i_{0}})$, and by Lemma 5.5 $a_{p+1}<^{+}c<^{+}b$. But this contradicts the choice of $p$. So the property 2. holds. Now we shall show 3. Take an upper $(S-\gamma(S))$-path from $a_{l}$ to $b$: $a_{l},\beta_{0},\ldots,\beta_{r},b$. We have two cases: $l=0$ and $l>0$. If $l=0$ then there is no face $a\in S$ such that $\gamma(a)\in\delta(a_{l})$. As $\delta(a_{l})\subseteq\delta\delta(\beta_{0})$ we must have $\delta(a_{l})\in\delta\gamma(\beta_{i})$, for $i=0,\ldots,r$. Hence $\delta(a_{l})\subseteq\delta\gamma(\beta_{r})=\delta(b)$ and 3. holds in this case. Now suppose that $l>0$. If $\gamma(a_{l-1})\in\delta\gamma(\beta_{i})$, for $i=0,\ldots,r$, then $\gamma(a_{l-1})\in\delta\gamma(\beta_{r})=\delta(b)$ and 3. holds again. So suppose contrary , and let $i_{1}=min\\{i:\gamma(a_{l-1})\not\in\delta\gamma(\beta_{i})\\}$ Then there are $a,c\in\delta(\beta_{i_{1}})$ such that $\gamma(a_{l-1})=\gamma(a)\in\delta(c)$ (NB: $c=a_{l}$ if $i_{1}=0$ and $c=\gamma(\beta_{i_{1}-1})$ otherwise). In particular $\gamma(a_{l-1})\in\iota(\beta_{i_{1}})$, and by Lemma 5.5 we have $a_{l-1}<^{+}a<^{+}b$ contrary to the choice of $l$. Thus 3. holds in this case as well. Finally, we shall show 4. Let $l\leq j\leq p$ and $a_{j},\beta_{0},\ldots,\beta_{r},b$ be an upper $(S-\delta(S))$-path from $a_{j}$ to $b$. As $a_{j}<^{-}a_{p}$ and $a_{p}<^{+}b$ we have $\gamma(a_{j})\neq\gamma(b)$. So we can put $i_{2}=min\\{i:\gamma(a_{j})\neq\gamma\gamma(\beta_{i})\\}$ But then $\gamma(a_{j})\in\gamma\delta(\beta_{i_{2}})-\gamma\gamma(\beta_{i_{2}})=\iota(\beta_{i_{2}})$. Therefore $\gamma(a_{j})\in\iota(S)$ and 4. holds. $~{}~{}\Box$ ###### Lemma 5.8 Let $S$ be a positive face structure, $n\in{\omega}$, $x,y\in S_{n}$, $x<^{+}y$. Then if $x,y\not\in\iota(S_{n+2})$, then there is an upper path from $x$ to $y$ contained in $S_{n+1}-\delta(S_{n+2})$. Proof. Assume $x,y\in(S-\iota(S))$ and $x<^{+}y$. Let $x,b_{0},\ldots,b_{k},y$ be an upper path from $x$ to $y$ with the longest possible initial segment $b_{0},\ldots,b_{l}$ in $S-\delta(S)$. We need to show that $k=l$. So suppose contrary, that $l<k$. By Corollary 5.6 $\gamma(b_{l})not\in\iota(S)$. Let $\alpha$ be a face such that $b_{l+1}\in\iota(\alpha)$. In particular $b_{l+1}<^{+}\gamma(\alpha)$. As $\gamma(b_{l})\not\in\iota(S)$ we have $\gamma(b_{l})\in\delta\gamma(\alpha)\;(=\delta\delta(\alpha)-\iota(\alpha))$. So $\gamma(b_{l})\in\delta(b_{l+1}\cap\delta\gamma(\alpha)$. Let $a$ be the $<^{+}$-largest element of the set $\\{b\in S:\gamma(b_{l})\in\delta(b)\\}$. Then $b_{l+1}\leq a$. By a similar argument as above for $b_{l+1}$ we get that $a\not\in\delta(S)$. By Lemma 5.7 there is $p$ such that $l+1\leq p\leq k$ such that $\gamma(b_{p})=\gamma(a)$ (Note that the facts that $y\not\in\iota(S)$ and Lemma 5.7.4 are needed here.). Thus we have an upper path from $x$ to $y$, $x,b_{0},\ldots,b_{l},a,b_{p+1},\ldots,b_{k},y$ with a longer initial segment in $S-\delta(S)$. But this means in fact that in fact $l=k$, as required. $~{}~{}\Box$ ### Order ###### Lemma 5.9 Let $S$ be a positive face structure, $n\in{\omega}$, $a,b\in S_{n}$. Then we have 1. 1. If $a<^{+}b$ then for any $x\in\delta(a)$ there is $y\in\delta(b)$ such that $y\leq^{+}x$; 2. 2. If $a<^{+}b$ and $\gamma(a)=\gamma(b)$ then for any $y\in\delta(b)$ there is $x\in\delta(a)$ such that $y\leq^{+}x$; 3. 3. If $\gamma(a)=\gamma(b)$ then either $a=b$ or $a\perp^{+}b$; 4. 4. If $\gamma(a)<^{+}\gamma(b)$ then either $a<^{+}b$ or $a<^{-}b$; 5. 5. If $a<^{+}b$ then $\gamma(a)\leq^{+}\gamma(b)$; 6. 6. If $a<^{-}b$ then $\gamma(a)<^{+}\gamma(b)$; 7. 7. If $\gamma(a)\perp^{-}\gamma(b)$ then $a\not\perp^{-}b$ and $a\not\perp^{+}b$. Proof. Ad 1. Let $a<^{+}b$ and $x\in\delta(a)$. We have two cases: either $x\in\gamma(S)$ or $x\not\in\gamma(S)$. In the first case there is $a^{\prime}\in$, such that $\gamma(a^{\prime})=x$. Let $a_{0},\ldots,a_{k}$ be a maximal path containing $a^{\prime},a$, say $a_{s-1}=a^{\prime}$ and $a_{s}=a$, where $0<s\leq k$. As $a_{s}<^{+}b$, by Lemma 5.7 there is $l\leq s$ and $y\in\delta(a_{l})\cap\delta(b)$. Clearly, $y\leq^{+}x$. In the second case consider an upper path from $a$ to $b$: $a,\beta_{0},\ldots,\beta_{r},b$. We have $x\in\delta(a)\subseteq\delta\delta(\beta_{0})$. As $x\not\in\gamma(S)$ so $x\not\in\gamma\delta(\beta_{0})$, and hence $x\in\delta\delta(\beta_{0})-\gamma\delta(\beta_{0})=\delta\gamma(\beta_{0})$. Thus we can define $r^{\prime}=max\\{i:x\in\delta\gamma(\beta_{i})\\}$ If we had $r^{\prime}<r$ then again we would have $x\in\delta\delta(\beta_{r^{\prime}+1})-\gamma\delta(\beta_{r^{\prime}+1})=\delta\gamma(\beta_{r^{\prime}+1})$, contrary to the choice of $r^{\prime}$. So $r^{\prime}=r$ and $x\in\delta\gamma(\beta_{r})=\delta(b)$. Thus we can put $y=x$. Ad 2. Fix $a<^{+}b$ such that $\gamma(a)=\gamma(b)$ and $y\in\delta(b)$. We need to find $x\in\delta(a)$ with $y\leq^{=}$. Take a maximal $(S-\gamma(S))$-path $a_{0},\ldots,a_{k}$ passing through $y$, i.e. there is $0\leq j\leq k$ such that $y\in\delta(a_{j})$ and if $y\in\gamma(S)$ then moreover $j>0$ and $\gamma(a_{j-1})=y$. Since $a_{j}\not\in\gamma(S)$ by $\delta$-linearity $a_{j}<^{+}b$. Thus by Lemma 5.7 there is $j\leq p\leq k$ such that $\gamma(a_{p})=\gamma(b)=\gamma(a)$. Since $a_{p}\not\in\gamma(S)$ by $\gamma$-linearity we have $a_{p}\leq^{+}a$. If $a_{p}=a$ then we can take as the face $x$ either $y$ if $p=0$ or $\gamma(a_{p-1})$ if $p>0$. So assume now that $a_{p}<^{+}a$. Again by Lemma 5.7 there is $0\leq l\leq p$ such that either $l=0$ and $\delta(a_{0})\subseteq\delta(a)$ or $l>0$ and $\gamma(a_{l-1})\in\delta(a)$. As $a_{l}$ is the first face in the path $a_{0},\ldots,a_{k}$ such that $a_{l}<^{+}a$ and $a_{j}$ is the first face in the path $a_{0},\ldots,a_{k}$ such that $a_{l}<^{+}b$ and moreover $a<^{+}b$ it follows that $j\leq l$. Thus in this case we can take as the face $x$ either $y$ if $l=0$ or $\gamma(a_{l-1})$ if $l>0$. Ad 3. This is an immediate consequence of $\gamma$-linearity. Ad 4. Suppose $\gamma(a)<^{+}\gamma(b)$. So there is an upper path $\gamma(a),c_{1},\ldots,c_{k},\gamma(b)$ with $k>0$. We put $c_{0}=a$. We have $\gamma(c_{k})=\gamma(b)$ so by $\gamma$-linearity $c_{k}\perp^{+}b$. So we have two cases: either $b<^{+}c_{k}$ or $c_{k}<^{+}b$. If $b<^{+}c_{k}$ then by Lemma 5.7 for any maximal path that contains $b$ and the face $c_{k}$ we get that $c_{k-1}<^{-}b$. Thus we have $a<^{-}b$. If $c_{k}<^{+}b$ then by Lemma 5.7 for any maximal path that extends $c_{0},c_{1},\ldots,c_{k}$ and face $b$ we get that either there is $0\leq i<k$ such that $\gamma(c_{i})\in\delta(b)$ and then $a<^{-}b$ or else $a=c_{0}<^{+}b$. Ad 5. This is repeated from Lemma 5.2. Ad 6. Suppose $a<^{-}b$. Then there is a lower path $a=a_{0},a_{1},\ldots,a_{k}=b$ with $k>0$. Then we have an upper path $\gamma(a)=\gamma(a_{0}),a_{1},\ldots,a_{k},\gamma(a_{k})=\gamma(b).$ Hence $\gamma(a)<^{+}\gamma(b)$. Ad 7. Easily follows from 5 and 6. $~{}~{}\Box$ ###### Proposition 5.10 Let $S$ be a positive face structure, $a,b\in S_{n}$, $a\neq b$. Let $\\{a_{i}\\}_{0\leq i\leq n}$, $\\{b_{i}\\}_{0\leq i\leq n}$ be two sequences of codomains of $a$ and $b$, respectively, so that $a_{i}=\gamma^{(i)}(a)\hskip 56.9055ptb_{i}=\gamma^{(i)}(b)$ (i.e. $dim(a_{i})=i$), for $i=0,\ldots,n$. Then there are two numbers $0\leq l\leq k\leq n$ such that either 1. 1. $a_{i}=b_{i}$ for $i<l$, 2. 2. $a_{i}<^{+}b_{i}$ for $l\leq i\leq k$, 3. 3. $a_{i}<^{-}b_{i}$ for $k+1=i\leq n$, 4. 4. $a_{i}\not\perp b_{i}$ for $k+2\leq i\leq n$, or the roles of $a$ and $b$ are interchanged. Proof. The above conditions we can present more visually as: $a_{0}=b_{0},\ldots,a_{l-1}=b_{l-1},\hskip 14.22636pta_{l}<^{+}b_{l},\ldots a_{k}<^{+}b_{k},$ $a_{k+1}<^{-}b_{k+1},\hskip 14.22636pta_{k+2}\not\perp b_{k+2},\ldots,a_{n}\not\perp b_{n}.$ These conditions we will verify from the bottom up. Note that by strictness $<^{S_{0},+}$ is a linear order. So either $a_{0}=b_{0}$ or $a_{0}\perp^{+}b_{0}$. In the later case $l=0$. As $a\neq b$ then there is $i\leq n$ such that $a_{i}\neq b_{i}$. Let $l$ be minimal such, i.e. $l=min\\{i:a_{i}\neq b_{i}\\}$. By Lemma 5.9 3., $a_{l}\perp^{+}b_{l}$. So assume that $a_{l}<^{+}b_{l}$. We put $k=max\\{i\leq n:a_{i}<^{+}b_{i}\\}$. If $k=n$ we are done. If $k<n$ then by Lemma 5.9 4., we have $a_{k+1}<^{-}b_{k+1}$. Then if $k+1<n$, by Lemma 5.9 5\. 6. 7., $a_{i}\not\perp b_{i}$ for $k+2\leq i\leq n$. This ends the proof. $~{}~{}\Box$ From the above Proposition we get immediately ###### Corollary 5.11 Let $S$ be a positive face structure, $a,b\in S_{n}$, $a\neq b$. Then either $a\perp^{+}b$ or there is a unique $0\leq l\leq k$ such that $a\perp^{-}_{l}b$, but not both. The above Corollary allows us to define an order $<^{S}$ (also denoted $<$) on all cells of $S$ as follows. For $a,b\in S_{n}$, $a<^{S}b\;\;\;{\rm iff}\;\;\;a<^{+}b\;\;{\rm or}\;\;\exists_{l}\;\;a<^{-}_{l}b.$ ###### Corollary 5.12 For any positive face structure $S$, and $n\in{\omega}$, the relation $<^{S}$ restricted to $S_{n}$ is a linear order. Proof. We need to verify that $<^{S}$ is transitive. Let $a,b,c\in S_{n}$. There are some cases to consider. If $a<^{+}b<^{+}c$ then clearly $a<^{+}c$. If $a<^{+}b<^{-}_{l}c$ then we have $\gamma^{(l)}(a)<^{+}\gamma^{(l)}(b)<^{-}\gamma^{(l)}(c)$. By Lemma 5.2 4., and transitivity of $<^{-}$ we have $\gamma^{(l)}(a)<^{-}\gamma^{(l)}(c)$. Hence $a<^{-}_{l}c$. If $a<^{-}_{l}b<^{+}c$ then, by use Lemma 5.7, either the maximal lower path from that contains both $\gamma^{(l)}(a)$ and $\gamma^{(l)}(b)$ passes between $\gamma^{(l)}(a)$ and $\gamma^{(l)}(b)$ through a face in $\delta\gamma^{(l+1)}(c)$ and $a<^{-}_{l}c$ or it does not passes through a face $\delta\gamma^{(l+1)}(c)$ and then $\gamma^{(l)}(a)<^{+}\gamma^{(l)}(c)$. Thus by Proposition 5.10 either there is $l\leq l^{\prime}\leq n$ such that $\gamma^{(l^{\prime})}(a)<^{+}\gamma^{(l^{\prime})}(c)$ and hence $a<^{-}_{l^{\prime}}c$ or $a<^{+}c$. Finally suppose that $a<^{-}_{k}b<^{-}_{l}c$. If $k=l$ then clearly $a<_{l}c$. If $k>l$ then $\gamma^{(l)}(a)<^{+}\gamma^{(l)}(b)<^{-}\gamma^{(l)}(c)$ and, by the previous argument, $\gamma^{(l)}(a)<^{-}\gamma^{(l)}(c)$, i.e. $a<^{-}_{l}c$. Finally, assume that $k<l$, i.e. $\gamma^{(k)}(a)<^{-}\gamma^{(k)}(b)<^{+}\gamma^{(k)}(c)$. Then, by Path Lemma, either $\gamma^{(k)}(a)<^{-}\gamma^{(k)}(c)$ or $\gamma^{(k)}(a)<^{+}\gamma^{(k)}(c)$. In the latter case, by Proposition 5.10, either $a<^{+}b$ or there is $k^{\prime}$, such that $k<k^{\prime}\leq n$ and $\gamma^{(k^{\prime})}(a)<^{+}\gamma^{(k^{\prime})}(c)$. In any case we have $a<c$, as required. $~{}~{}\Box$ From the proof of the above corollary we get ###### Lemma 5.13 Let $S$ be a positive face structure, $a\in S_{n}$, $a$. Then the set $\\{b\in S_{n}\,:\,a\leq^{+}b\\}$ is linearly ordered by $\leq^{+}$. Proof. Suppose $a\leq^{+}b,b^{\prime}$. If we where to have $b<^{-}_{l}b^{\prime}$ for some $l\leq n$ then, by Corollary 5.12 we would have $a<^{-}_{l}b^{\prime}$ which is a contradiction. $~{}~{}\Box$ ###### Corollary 5.14 Any morphism of positive face structures is one-to-one. Moreover any automorphism of positive face structures is an identity. Proof. By Corollary 5.12, the (strict, linear in each dimension) order $<^{S}$ is defined internally using relations $<^{-}$ and $<^{+}$ that are preserved by any morphism. Hence $<^{S}$ must be preserved by any morphism, as well. From this observation the Corollary follows. $~{}~{}\Box$ ###### Lemma 5.15 Let $S$ be a positive face structure, $a,b\in S_{n}$, $a$. Then 1. 1. if $\iota(a)\cap\iota(b)\neq\emptyset$ then $a\perp^{+}b$; 2. 2. if $\emptyset\neq\iota(a)\subseteq\iota(b)\neq\iota(a)$ then $a<^{+}b$; 3. 3. if $a\perp^{-}b$ then $\iota(a)\cap\iota(b)=\emptyset$. Proof. 2\. is an easy consequence of 1. and Lemma 5.3. 3. is an easy consequence of 1. and Disjointness. We shall show 1. Assume that $u\in\iota(a)\cap\iota(b)$. Thus there are $x,y\in\delta(a)$ and $x^{\prime},y^{\prime}\in\delta(b)$ such that $\gamma(x)=\gamma(x^{\prime})=u\in\delta(y)\cap\delta(y^{\prime})$. If $x=x^{\prime}$ then by Local linearity $a\perp^{+}b$, as required. So assume that $x\neq x^{\prime}$. Again by Local linearity $x\perp^{+}x^{\prime}$, say $x^{\prime}\perp^{+}x$. Thus there is a $T-\gamma(T)$-path $x^{\prime},a_{1},\ldots,a_{k},x$. As, for $i=1,\ldots,k$, $\gamma\gamma(a_{i})=u$ and $\gamma\gamma(b)\not\in\iota(b)\ni u$, we have that $\gamma(a_{i})\neq\gamma(b)$ and $a_{i}\neq b$. Once again by Local linearity $a_{0}\perp^{+}b$ and by Path Lemma $a_{i}<b$, for $i=1,\ldots,k$ with $\gamma(a_{k})\neq\gamma(b)$. As $\gamma(a_{k})=x\in\delta(a)$, again by Path Lemma $a<^{+}b$, as well. $~{}~{}\Box$ ###### Proposition 5.16 Let $S$ be a positive face structure, $a,b\in S_{k}$, $\alpha\in S_{k+1}$, so that $\alpha$ is a $<^{+}$-minimal element in $S_{k+1}$, and $a\in\delta(\alpha)$, $b=\gamma(\alpha)$. Then $b$ is the $<^{+}$-successor of $a$. Proof. Assume that $\alpha$ is a $<^{+}$-minimal element in $S_{k+1}$. Suppose that there is $c\in S_{k}$ such that $a<^{+}c<^{+}b$. Thus we have an upper path $a,\beta_{1},\ldots,\beta_{i},c,\beta_{i+1},\ldots,\beta_{l},b.$ Hence $\beta_{1}<^{-}\beta_{l}$. Moreover $a\in\delta(\beta_{1})\cap\delta(\alpha)$ and $\gamma(\beta_{l})=b=\gamma(\alpha)$. Thus both $\beta_{1}$ and $\beta_{l}$ are $<^{+}$-comparable with $\alpha$. Since $\alpha$ is $<^{+}$-minimal we have $\alpha<^{+}\beta_{1},\beta_{l}$. By Lemma 5.13, $\beta_{1}\perp^{+}\beta_{l}$. But then we have $(\beta_{1},\beta_{l})\in\perp^{+}\cap\perp^{-}\neq\emptyset$, contradicting strictness. $~{}~{}\Box$ ###### Proposition 5.17 Let $T$ be a positive face structure and $X\subseteq T$ a subhypergraph of $T$. Then $X$ is a positive face structure iff the relation $<^{X,+}$ is the restriction of $<^{T,+}$ to $X$. Proof. Assume that $X$ is a subhypergraph of a positive face structure $T$. Then $X$ satisfies axioms of globularity, disjointness, and strictness of the relations $<^{X_{k},+}$ for $k>0$. Clearly, if $<^{X_{k},+}\;=\;<^{T_{k},+}\cap(X_{k})^{2}$ then the relation $<^{X_{0},+}$ is linear, the relations $<^{X_{k},+}$, for $k>0$, satisfy pencil linearity, i.e. $X$ is a positive face structure. Now we assume that the subhypergraph $X$ of positive face structure $T$ is a positive face structure. We shall show that for $k\in{\omega}$, $a,b\in X_{k}$ we have $a<^{X_{k},+}b$ iff $a<^{T_{k},+}b$. Since $X$ is a subhypergraph $a<^{X_{k},+}b$ implies $a<^{T_{k},+}b$. Thus it is enough to show that if $a<^{T_{k},+}b$ then $a\perp^{X_{k},+}b$. We shall prove this by induction on $k$. For $k=0$ it is obvious, since $<^{X_{0},+}$ is linear. So assume that for faces $x,y\in X_{l}$, with $l<k$ we already know that $x<^{X_{l},+}y$ iff $x<^{T_{l},+}y$. Fix $a,b\in T_{k}$ such that $a<^{T_{k},+}b$. Then by Lemma 5.9.2 $\gamma(a)\leq^{T_{k-1},+}\gamma(b)$ and hence by inductive hypothesis $\gamma(a)\leq^{X_{k-1},+}\gamma(b)$. Thus we have an upper $X$-path $a=a_{r},\gamma(a),a_{r-1}\ldots,a_{1},\gamma(b)$, with $r\geq 1$. As $a<^{T_{k},+}b$, by Path Lemma $a_{i}<^{T_{k},+}b$ for $i=1,\ldots,r$. Again by induction on $r$ we shall show that $a_{i}<^{X_{k},+}b$, for $i=1,\ldots,r$. As $\gamma(a_{1})=\gamma(b)$ by pencil linearity we have $a_{1}\perp^{X_{k},+}b$. So $a_{1}<^{X_{k},+}b$. Suppose that $a_{i}<^{X_{k},+}b$, for $i\leq l<r$. Let $a_{l},\alpha_{s},\ldots,\alpha_{1},b$ be an upper $T$-path. As $a_{l+1}<^{T_{k},+}b$, we cannot have $\gamma(a_{l+1})\in\delta(b)$. Therefore, for some $j\geq 1$, $\gamma(a_{l+1})\in\iota(\alpha_{j})$. So by Lemma 5.5, $a_{l+1}<^{X_{k},+}\gamma(\alpha_{j})\leq^{X_{k},+}b$, as required. $~{}~{}\Box$ ###### Lemma 5.18 Let $T$ be a positive face structure, $a,b,\alpha\in T$. If $a\in\delta(\alpha)$ and $a<^{+}b<^{+}\gamma(\alpha)$ then $b\in\iota(T)$. Proof. Assume that $a,b,\alpha\in T$ are as in the assumption of the Lemma. Thus we have an upper path $a,\alpha_{0},\ldots,\alpha_{r},b$. As $a\in\delta(\alpha)\cap\delta(\alpha_{0})$, by pencil linearity we have $\alpha\perp^{+}\alpha_{0}$. If $\alpha<^{+}\alpha_{0}<^{-}\alpha_{r}$ then $\gamma(\alpha)\leq^{+}\gamma(\alpha_{r})=b$ contradicting our assumption. Thus $\alpha_{0}<^{+}\alpha$. Then by Path Lemma, since $b=\gamma(\alpha_{r})<^{+}\gamma(\alpha)$, we have $\alpha_{r}<^{+}\alpha$ and $b\in\iota(T)$, as required. $~{}~{}\Box$ ### Some equations ###### Proposition 5.19 Let $S$ be a positive face structure $0<k\in{\omega}$. Then 1. 1. $\iota(S_{k+1})=\iota(S_{k+1}-\delta(S_{k+2}))$; 2. 2. $\delta(S_{k})=\delta(S_{k}-\gamma({S_{k+1}}))$; 3. 3. $\gamma(S_{k})=\gamma(S_{k}-\gamma({S_{k+1}}))$; 4. 4. $\delta(S_{k})=\delta(S_{k}-\iota({S_{k+2}}))$; 5. 5. $\delta(S_{k})=\delta(S_{k}-\delta({S_{k+1}}))\cup\iota(S_{k+1})$. Proof. In all the above equations the inclusion $\supseteq$ is obvious. So in each case we need to check the inclusion $\subseteq$ only. Ad 1. Let $s\in\iota(S_{k+1})$, i.e. there is $a\in S_{k+1}$ such that $s\in\iota(a)$. By strictness, there is $b\in S_{k+1}$ such that $a<^{+}b$ and $b\not\in\delta(S_{k+1})$. By Lemma 5.3, we have $s\in\iota(a)\subseteq\iota(b)\subseteq\iota(S_{k+1}-\delta(S_{k+2}))$ as required. Ad 2. Let $x\in\delta(S_{k})$. Let $a\in S_{k}$ be the $<^{+}$-minimal element in $S_{k}$ such that $x\in\delta(a)$. We shall show that $a\in S_{k}-\gamma(S_{k+1})$. Suppose contrary that there is an $\alpha\in S_{k+1}$ such that $a=\gamma(\alpha)$. The by globularity $x\in\delta(a)=\delta\gamma(\alpha)=\delta\delta(\alpha)-\gamma\delta(\alpha).$ So there is $b\in\delta(\alpha)$ such that $x\in\delta(b)$. As $b<^{+}a$ this contradicts the minimality of $a$. Ad 3. This is similar to the previous one but simpler. Ad 4. Since $\iota({S_{k+2}})\subseteq\gamma({S_{k+1}})$ 4\. follows from 2. Ad 5. Let $x\in\delta(S_{k})$. Let $a\in S_{k}$ be the $<^{+}$-maximal element in $S_{k}$ such that $x\in\delta(a)$. If $a\not\in\delta(S_{k+1})$ then $x\in\delta(S_{k}-\delta({S_{k+1}}))$ as required. So assume that $a\in\delta(S_{k+1})$, i.e. there is $\alpha\in S_{k+1}$ such that $a\in\delta(\alpha)$. Thus $x\in\delta\delta(\alpha)$. As $a<^{+}\gamma(\alpha)$, by choice of $a$ we have $x\not\in\delta\gamma(\alpha)\;(=\delta\delta(\alpha)-\gamma\delta(\alpha))$. So $x\in\gamma\delta(\alpha)$ and hence $x\in\iota(\alpha)\subseteq\iota(S_{k+1})$, as required. $~{}~{}\Box$ ## 6 The ${\omega}$-categories generated by the positive face structures Let $T^{*}$ ($T^{*}_{n}$) be the set of all face substructures of the face structure $T$ (of dimension at most $n$). We introduce operations ${\bf d}^{(n)},{\bf c}^{(n)}:T^{*}\longrightarrow T^{*}_{n}$ of the $n$-th domain and the $n$-th codomain. For $S$ in $(T^{*})_{\geq n}$ the faces $n$-th domain ${\bf d}^{(n)}S$ are: 1. 1. $({\bf d}^{(n)}S)_{k}=\emptyset,$ for $k>n$, 2. 2. $({\bf d}^{(n)}S)_{n}=S_{n}-\gamma(S_{n+1})$, 3. 3. $({\bf d}^{(n)}S)_{k}=S_{k}$, for $k<n$. and faces $n$-th codomain ${\bf c}^{(n)}S$ are: 1. 1. $({\bf c}^{(n)}S)_{k}=\emptyset,$ for $k>n$, 2. 2. $({\bf c}^{(n)}S)_{n}=S_{n}-\delta(S_{n+1})$, 3. 3. $({\bf c}^{(n)}S)_{n-1}=S_{n-1}-\iota(S_{n+1})$, 4. 4. $({\bf c}^{(n)}S)_{k}=S_{k}$, for $k<n-1$. If $k\in{\omega}$, $dim(S)=k+1$, we write ${\bf d}S$ for ${\bf d}^{(k)}(S)$, and ${\bf c}S$ for ${\bf c}^{(k)}(S)$. We have ###### Lemma 6.1 Let $S$ be a positive face structure. Then 1. 1. if $dim(S)\geq k$ then ${\bf d}^{(k)}(S)$, ${\bf c}^{(k)}(S)$ are positive face structures of dimension $k$; 2. 2. if $dim(S)>k$ then ${\bf d}{\bf d}^{(k+1)}(S)={\bf d}^{(k)}(S)$ , ${\bf d}{\bf d}^{(k+1)}(S)={\bf d}^{(k)}(S)$; 3. 3. if $dim(S)>1$ then ${\bf d}{\bf d}S={\bf d}{\bf c}S$, ${\bf c}{\bf d}S={\bf c}{\bf c}S$. 4. 4. For any $k\in{\omega}$ and $\alpha\in S_{k}$, the least sub-hypergraph of $S$ containing the face $\alpha$ is again a positive face structure of dimension $k$; it is denoted by $[\alpha]$. Moreover, if $k>0$, then ${\bf c}[\alpha]=[\gamma(\alpha)],\hskip 28.45274pt{\bf d}[\alpha]=[\delta(\alpha)]$ where $[\delta(\alpha)]$ is the least sub-hypergraph of $S$ containing the set of face $\delta(\alpha)$. Proof. Ad 1. Clearly, ${\bf d}^{(k)}S$ is a sub-hypergraph $S$ and ${\bf c}^{(k)}S$ is a sub-hypergraph $S$ by Corollary 5.6. Any sub-hypergraph $T$ of a positive face structure $S$ satisfies the conditions of globularity, strictness (possibly without $<^{T_{0},+}$ being linear), and disjointness. By Lemma 5.4, for $a,b\in{\bf d}^{(k)}S_{l}$ we have $a<^{S_{l},+}b$ iff $a<^{{\bf d}^{(k)}S_{l},+}b$. Moreover, by Lemma 5.8, for $a,b\in{{\bf c}^{(k)}S}_{l}$ we have $a<^{S_{l},+}b$ iff $a<^{{{\bf c}^{(k)}S}_{l},+}b$. Hence by Lemma 5.17 both ${\bf d}^{(k)}S$ and ${\bf c}^{(k)}S$ are positive face structures. Ad 2. Fix a positive face structure $S$ and $k\in{\omega}$ such that $dim(S)>k$. Then the faces of ${\bf c}^{(k+1)}(S)$, ${\bf c}{\bf c}^{(k+1)}(S)$, and ${\bf c}^{(k)}(S)$ are as in the table $\begin{array}[]{|c|c|c|c|}dim&{\bf c}^{(k+1)}(S)&{\bf c}{\bf c}^{(k+1)}(S)&{\bf c}^{(k)}(S)\\\ \hline\cr k+1&S_{k+1}-\delta(S_{k+2})&\emptyset&\emptyset\\\ \hline\cr k&S_{k}-\iota(S_{k+2})&(S_{k}-\iota(S_{k+2}))-\delta(S_{k+1}-\delta(S_{k+2}))&S_{k}-\delta(S_{k+1})\\\ \hline\cr k-1&S_{k-1}&S_{k-1}-\iota(S_{k+1}-\delta(S_{k+2}))&S_{k-1}-\iota(S_{k+1})\\\ \hline\cr l&S_{l}&S_{l}&S_{l}\end{array}$ where $l<k-1$. Moreover the faces of ${\bf d}^{(k+1)}(S)$, ${\bf d}{\bf d}^{(k+1)}(S)$, and ${\bf d}^{(k)}(S)$ are as in the table $\begin{array}[]{|c|c|c|c|}dim&{\bf d}^{(k+1)}(S)&{\bf d}{\bf d}^{(k+1)}(S)&{\bf d}^{(k)}(S)\\\ \hline\cr k+1&S_{k+1}-\gamma(S_{k+2})&\emptyset&\emptyset\\\ \hline\cr k&S_{k}&S_{k}-\gamma(S_{k+1}-\gamma(S_{k+2}))&S_{k}-\gamma(S_{k+1})\\\ \hline\cr l&S_{l}&S_{l}&S_{l}\end{array}$ where $l<k$. Thus the equalities in question all follow from Lemma 5.19. Ad 3. Let $dim(S)=n>1$. Note that both $({\bf d}{\bf d}S)_{n-2}$ and $({\bf d}{\bf c}S)_{n-2}$ are the sets of all $<^{+}$-minimal elements in $S_{n-2}$, i.e. they are equal and the equation ${\bf d}{\bf d}S={\bf d}{\bf c}S$ holds. To see that ${\bf c}{\bf d}S={\bf c}{\bf c}S$ holds, note first that both $({\bf c}{\bf d}S)_{n-2}$ and $({\bf c}{\bf c}S)_{n-2}$ are the sets of all $<^{+}$-maximal elements in $S_{n-2}$. Moreover $({\bf c}{\bf d}S)_{n-3}=S_{n-3}-\iota(S_{n-1}-\gamma(S_{n})),$ $({\bf c}{\bf c}S)_{n-3}=S_{n-3}-\iota(S_{n-1}-\delta(S_{n})).$ Now the equality ${\bf c}{\bf d}S={\bf c}{\bf c}S$ follows from the following equalities $\iota(S_{n-1}-\gamma(S_{n}))=\iota(S_{n-1})=\iota(S_{n-1}-\delta(S_{n})).$ Both equalities follow from Lemma 5.3. We shall show the first equality only. Suppose contrary, that there is $x\in\iota(S_{n-1})$ such that $x\not\in\iota(S_{n-1}-\gamma(S_{n}))$. Let $a\in S_{n-1}$ be a $<^{+}$-minimal face such that $x\in\iota(a)$. Since $x\not\in\iota(S_{n-1}-\gamma(S_{n}))$, there is $\alpha\in S_{n}$ such that $a=\gamma(\alpha)$. By Lemma 5.3 we have $\iota\delta(\alpha)=\iota\gamma(\alpha)=\iota(a).$ Therefore, there is $a^{\prime}\in\delta(\alpha)$ such that $x\in\iota(a^{\prime})$. Clearly $a^{\prime}\lhd^{+}a$, and hence $a$ is not $<^{+}$-minimal contrary to the supposition. This ends the proof of the first equality above. Ad 4. Fix $\alpha\in S_{k}$. We need to show that $[\alpha]$ is a positive face structure. The globularity, strictness (except for linearity of $<^{[\alpha]_{0},+}$), and disjointness are clear. The linearity of $<^{[\alpha]_{0},+}$. If $k\leq 2$ it is obvious. Put $a=\gamma^{(k+2)}(\alpha)$. Using Corollary 4.3, we have $[\alpha]_{0}=\delta^{(k)}(\alpha)\cup\gamma^{(k)}(\alpha)=$ $=\delta\delta(\gamma^{(n-2)}(\alpha))\cup\gamma\gamma(\gamma^{(n-2)}(\alpha))=\delta\delta(a)\cup\gamma\gamma(a)$ Thus it is enough to assume that $k=2$. But in this case, as we mentioned, the linearity of $<^{[\alpha]_{0},+}$ is obvious. The $\gamma$-linearity of $[\alpha]$. The proof proceeds by induction on $k=dim(\alpha)$. For $k\leq 2$, the $\gamma$-linearity is obvious. So assume that $k>2$ and that for $l<k$ and $a\in S_{l}$. First we shall show that ${\bf c}([\alpha])=[\gamma(\alpha)]$. We have ${\bf c}([\alpha])_{k-1}=(\gamma(\alpha)\cup\delta(\alpha))-\delta(\alpha)=\gamma(\alpha)=[\gamma(\alpha)]_{k-1}$ ${\bf c}([\alpha])_{k-2}=(\gamma\gamma(\alpha)\cup\delta\delta(\alpha))-\iota(\alpha)=\delta\gamma(\alpha)\cup\gamma\gamma(\alpha)=[\gamma(\alpha)]_{k-2}$ and for $l<k-2$ ${\bf c}([\alpha])_{l}=\gamma^{(l)}(\alpha)\cup\delta\gamma^{(l)}(\alpha))=\gamma^{(l)}(\alpha)\cup\delta^{(l)}\gamma(\alpha)=[\gamma(\alpha)]_{l}$ Note that the definition of ${\bf c}(H)$ make sense for any positive hypergraph $H$ and in the above argument we haven’t use the fact (which we don’t know yet) that $[\alpha]$ is a positive face structure. Thus, for $l<k-2$, $[\alpha]_{l}=[\gamma(\alpha)]_{l}$. By induction, $[\gamma(\alpha)]$ is a positive face structure, and hence $[\alpha]_{l}$ is $\gamma$-linear $l<k-2$. Clearly $[\alpha]_{l}$ is$\gamma$-linear, for $l=k-1,k$. Thus it remains to show the $\gamma$-linearity of $k-2$-cells in $[\alpha]$. Fix $t\in[\alpha]_{k-3}$, and let $\Gamma_{t}=\\{x\in[\alpha]_{k-2}:\gamma(x)=t\\}.$ We need to show that $\Gamma_{t}$ is linearly ordered by $<^{+}$. We can assume that $t\in\gamma([\alpha]_{n-2})=\gamma\delta\delta(\alpha)=\gamma\delta\gamma(\alpha)$ (otherwise $\Gamma_{t}=\emptyset$ is clearly linearly ordered by $<^{+}$). By Proposition 5.1 there is a unique $x_{t}\in\delta\gamma(\alpha)$ such that $\gamma(x_{t})=t$. From Lemma 5.2.2 we get easily the following Claim. Claim 1. For every $x\in\Gamma_{t}$ there is a unique upper $\delta(\alpha)$-path from $x_{t}$ to $x$. Now fix $x,x^{\prime}\in\Gamma_{t}$. By the Claim 1, we have the unique upper $\delta(\alpha)$-path $x_{t},a_{0},\ldots,a_{l},x,\;\;\;\;x_{t},a^{\prime}_{0},\ldots,a^{\prime}_{l},x^{\prime}.$ Suppose $l\leq l^{\prime}$. By Proposition 5.1, for $i\leq l$, $a_{i}=a^{\prime}_{i}$. Hence either $l=l^{\prime}$ and $x=x^{\prime}$ or $l<l^{\prime}$ and $x,a_{l+1},\ldots,a_{l^{\prime}},x^{\prime}$ is a $\delta(\alpha)$-upper path. Hence either $x=x^{\prime}$ or $x\perp^{+}x^{\prime}$ and $[\alpha]_{k-2}$ satisfy the $\gamma$-linearity, as required. The proof is of the $\delta$-linearity of $[\alpha]$ is very similar to the one above. For the same reasons the only non-trivial thing to check is the condition for $(k-2)$-faces. We pick $t\in\delta\delta(\alpha)$ and consider the set $\Delta_{t}=\\{x\in[\alpha]_{k-2}:t\in\delta(x)\\}.$ Then we have a unique $y_{t}\in\delta\gamma(\alpha)$ such that $t\in\delta(y_{t})$. From Lemma 5.2.3 we get the following Claim. Claim 2. For every $y\in\Delta_{t}$ there is a unique upper $\delta(\alpha)$-path from $y_{t}$ to $y$. The $\delta$-linearity of the $(k-2)$-faces in $[\alpha]$ can be proven from Claim 2 similarly as the $\gamma$-linearity from Claim 1. It remains to verify the equalities ${\bf c}[\alpha]=[\gamma(\alpha)],\hskip 28.45274pt{\bf d}[\alpha]=[\delta(\alpha)].$ The first one we already checked on the way. To see that the second equality also hold we calculate ${\bf d}[\alpha]_{k-1}=(\gamma(\alpha)\cup\delta(\alpha))-\gamma(\alpha)=\delta(\alpha)=[\delta(\alpha)]_{k-1}$ ${\bf d}[\alpha]_{k-2}=(\gamma\gamma(\alpha)\cup\delta\delta(\alpha))=\gamma\delta(\alpha)\cup\delta\delta(\alpha)=[\delta(\alpha)]_{k-2}$ and for $l<k-2$ ${\bf d}[\alpha]_{l}=\gamma^{(l)}(\alpha)\cup\delta^{(l)}(\alpha))=\gamma^{(l)}\delta(\alpha)\cup\delta^{(l)}(\alpha)=[\delta(\alpha)]_{l}$ So the second equality holds as well.$~{}~{}\Box$ ###### Lemma 6.2 Let $S$ and $T$ be positive face structures such that ${\bf c}^{(k)}S\subseteq{\bf d}^{(k)}T$. Then the pushout $S+_{k}T$ in ${\bf Fs}^{+/1}$ of $S$ and $T$ over ${{\bf c}^{(k)}S}$ exists. Moreover, if ${\bf c}^{(k)}S=S\cap T$ then the diagram of inclusions in ${\bf Fs}^{+/1}$ ${\bf c}^{(k)}S\vphantom{T}$$\vphantom{{\bf c}^{(k)}S}T$$S\vphantom{S\cup T}$$\vphantom{S}S\cup T$ is the pushout. Proof. Assume that ${\bf c}^{(k)}S=S\cap T\subseteq{\bf d}^{(k)}T$. Let $S\cup T$ be the obvious sum of $S$ and $T$ as positive hypergraphs. The fact that $S\cup T$ is a pushout in ${\bf Hg}^{+1}$ is obvious. Thus the only thing we need to verify that $S\cup T$ is a positive face structure. First we write in details the condition ${\bf c}^{(k)}S=S\cap T\subseteq{\bf d}^{(k)}T$: 1. 1. $S_{l}\cap T_{l}=\emptyset$, for $l>k$, 2. 2. $S_{k}-\delta(S_{k+1})\subseteq T_{k}-\gamma(T_{k+1})$, 3. 3. $S_{k-1}-\iota(S_{k+1})\subseteq T_{k-1}$, 4. 4. $S_{l}\subseteq T_{l}$, for $l<k-1$. Now we describe the orders $<^{+}$ in $S\cup T$: $<^{(S\cup T)_{l},+}=\left\\{\begin{array}[]{ll}<^{S_{l},+}+<^{T_{l},+}&\mbox{for $l>k$,}\\\ <^{S_{l},+}+_{(S_{k}-\delta(S_{k+1}))}<^{T_{l},+}&\mbox{for $l=k$,}\\\ <^{S_{l},+}+_{(S_{k-1}-\iota(S_{k+1}))}<^{T_{l},+}&\mbox{for $l=k-1$,}\\\ <^{T_{l},+}&\mbox{for $l<=k-1$.}\end{array}\right.$ We shall comment on these formulas. For $l>k$ the formulas say that the order $<^{+}$ in $(S\cup T)_{l}$ is the disjoint sum of the orders in $S_{l}$ and $T_{l}$. This is obvious. For $l<k-1$ the order $<^{+}$ in $(S\cap T)_{l}$ is just the order $<^{T_{l},+}$. The only case that requires an explanation is $l=k-2$. So suppose that $a,b\in T_{k-2}$ and $a<^{(S\cup T)_{k-2},+}b$. So we have an upper path $a,\alpha_{1},\ldots,\alpha_{m},b$ such that $\alpha_{i}\in(S\cup T)_{k-1}=\iota(S_{k+1})\cup T_{k-1}$. By Lemma 5.4, we can assume that if $\alpha_{i}\in S_{k-1}$ then $\alpha_{i}\not\in\gamma(S_{k})$. But then $\alpha_{i}\not\in\iota(S_{k+1})$. So in fact $\alpha_{i}\in T_{k-1}$, as required. The most involved are the formulas for $<^{(S\cap T)_{l},+}$ for $l=k$ and $l=k-1$. In both cases the comparison in $S\cup T$ involves orders both from $S$ and $T$. In the former case we have that, for $a,b\in(S\cup T)_{k}$, we have $a<^{(S\cup T)_{k},+}b\mbox{ iff }\hskip 199.16928pt$ $\left\\{\begin{array}[]{ll}\mbox{either }a,b\in T_{k}&\mbox{ and }a<^{T_{k},+}b,\\\ \mbox{or }a,b\in S_{k}&\mbox{ and }a<^{S_{k},+}b,\\\ \mbox{or }a\in\delta(S_{k+1}),\;b\in T_{k}&\mbox{ and }\exists_{a^{\prime}\in S_{k}-\delta(S_{k+1})}a<^{S_{k},+}a^{\prime}\mbox{ and }a^{\prime}\leq^{T_{k},+}b.\end{array}\right.$ The orders $<^{S_{k},+}$ and $\leq^{T_{k},+}$ are glued together along the set $S_{k}-\delta(S_{k+1})$ which is the set of $<^{S_{k},+}$-maximal elements in $S_{k}$ and at the same time it is contained in the set of $<^{T_{k},+}$-minimal elements $T_{k}-\gamma(T_{k+1})$. This is obvious when we realize that $\delta(S_{k+1})\cap\gamma(T_{k+1})=\emptyset$. In the later case we have for $x,y\in(S\cup T)_{k-1}$ we have $x<^{(S\cup T)_{k-1},+}y\mbox{ iff }\hskip 199.16928pt$ $\left\\{\begin{array}[]{ll}\mbox{either }x,y\in S_{k-1}&\mbox{ and }x<^{S_{k-1},+}y,\\\ \mbox{or }x,y\in T_{k-1}&\mbox{ and }x<^{T_{k-1},+}y,\\\ \mbox{or }x\in\iota(S_{k+1}),\;y\in T_{k}&\mbox{ and }\exists_{x^{\prime}\in S_{k-1}-\iota(S_{k+1})}x<^{S_{k},+}x^{\prime}\mbox{ and }x^{\prime}\leq^{T_{k},+}y,\\\ \mbox{or }x\in T_{k},\;y\in\iota(S_{k+1})&\mbox{ and }\exists_{x^{\prime}\in S_{k-1}-\iota(S_{k+1})}x<^{T_{k},+}x^{\prime}\mbox{ and }x^{\prime}\leq^{S_{k},+}y.\end{array}\right.$ The order $<^{S_{k-1},+}$ is ’plugged into’ the order $\leq^{T_{k-1},+}$ along the set $S_{k}-\iota(S_{k+1})$. To show that these formulas hold true we argue by cases. Assume that $x,y\in(S\cup T)_{k-1}$ and that $x<^{(S\cup T)_{k-1},+}y$ i.e. there is an upper path $x,a_{1},\ldots,a_{m},y$ with $a_{i}\in(S\cup T)_{k},$ for $i=1,\ldots,m$. First suppose that $x,y\in S_{k-1}$ and that the set $\\{a_{i}\\}_{i}\not\subseteq S_{k}$. Let $a_{i_{0}},a_{i_{0}+1},\ldots,a_{i_{1}}$ be a maximal subsequence of consecutive elements of the path $a_{1},\ldots,a_{m}$ such that $\\{a_{i}\\}_{i_{0}\leq i\leq i_{1}}\subseteq T_{k}$. Thus it is an upper path in $T_{k}$ from $\bar{x}$ to $\bar{y}=\gamma(a_{i_{1}})$, where $\bar{x}=\left\\{\begin{array}[]{ll}x&\mbox{if }i_{0}=1,\\\ \gamma(a_{i_{0}-1})&\mbox{otherwise.}\end{array}\right.$ Note that from maximality of the path $a_{i_{0}},\ldots,a_{i_{1}}$ follows that both $\bar{x},\bar{y}\in S_{k-1}-\iota(S_{k+1})$. As we have $\bar{x}<^{T_{k-1},+}\bar{y}$ from Corollary 5.11 we have $\bar{x}\not\perp^{T_{l},-}\bar{y}$, for all $l<k-1$. Clearly $\perp^{S_{l},-}\subseteq\perp^{T_{l},-}$. Thus $\bar{x}\not\perp^{S_{l},-}\bar{y}$, for all $l<k-1$, as well. But then again by Corollary 5.11 we have that $\bar{x}\perp^{S_{k-1},+}\bar{y}$. If we were to have $\bar{y}<^{S_{k-1},+}\bar{x}$ then, as $\bar{x},\bar{y}\in S_{k-1}-\iota(S_{k+1})$, we would have $\bar{y}<^{T_{k-1},+}\bar{x}$. But this would contradict the strictness of $<^{T_{k-1},+}$. So we must have $\bar{x}<^{S_{k-1},+}\bar{y}$. In this way we can replace the upper path $a_{1},\ldots,a_{m}$ in $(S\cup T)_{k}$ from $x$ to $y$ by an upper path from $x$ to $y$ in $S_{k}$. Next, suppose that $x,y\in T_{k-1}$ and that the set $\\{a_{i}\\}_{i}\not\subseteq T_{k}$. Let $a_{i_{0}},a_{i_{0}+1},\ldots,a_{i_{1}}$ be a maximal subsequence of consecutive elements of the path $a_{1},\ldots,a_{m}$ such that $\\{a_{i}\\}_{i_{0}\leq i\leq i_{1}}\subseteq S_{k}$. Thus it is an upper path in $S_{k}$ from $\bar{x}$ to $\bar{y}=\gamma(a_{i_{1}})$, where $\bar{x}=\left\\{\begin{array}[]{ll}x&\mbox{if }i_{0}=1,\\\ \gamma(a_{i_{0}-1})&\mbox{otherwise.}\end{array}\right.$ Note that from maximality of the sequence $a_{i_{0}},\ldots,a_{i_{1}}$ follows that both $\bar{x},\bar{y}\in S_{k-1}-\iota(S_{k+1})\subseteq T_{k-1}$. Thus by Lemma 5.8 there is an upper path from $\bar{x}$ to $\bar{y}$ in $S_{k-1}-\delta(S_{k})\subseteq T_{k-1}$. In this way we can replace the upper path $a_{1},\ldots,a_{m}$ in $(S\cup T)_{k}$ from $x$ to $y$ by an upper path from $x$ to $y$ in $T_{k}$. Thus we have justified the first two cases of the above formula. The following two cases are easy consequences these two. This end the description of the orders in $S\cup T$. From these descriptions follows immediately that $<^{(S\cup T),+}$ is strict for all $l$. It remains to show the pencil linearity. Both $\gamma$\- and $\delta$-linearity of $l$-cells, for $l<k-1$ or $l>k$, are obvious. To see the $\gamma$-linearity of $k$-cells assume $a\in S_{k}$ and $b\in T_{k}$, such that $\gamma(a)=\gamma(b)$. Let $\bar{a}\in S_{k}$ be the $<^{S_{k},+}$-maximal $k$-cells, such that $\gamma(a)=\gamma(\bar{a})$. Then $\bar{a}\in{\bf c}^{(k)}(S)_{k}\subseteq{\bf d}^{(k)}(T)_{k}$. So $\bar{a}\in T_{k}$ is $<^{T_{k},+}$-minimal $k$-cells, such that $\gamma(\bar{a})=\gamma(b)$. Thus $a\leq^{S_{k},+}\bar{a}\leq^{T_{k},+}b.$ Thus the $\gamma$-linearity of $k$-cells holds. The proof of $\delta$-linearity of $k$-cells is similar. Finally, we need to establish the $\gamma$\- and $\delta$\- linearity of $(k-1)$-cells in $S\cup T$. In order to prove the $\gamma$-linearity, let $x\in\iota(S_{k+1})$ and $y\in T_{k-1}$ such that $\gamma(x)=\gamma(y)$. We need to show that $x\perp^{(S\cup T)_{k-1},+}y$. Let $\alpha_{0}\in S_{k+1}$ such that $x\in\iota(\alpha_{0})$, $a\in\delta(\alpha_{0})$ such that $x=\gamma(a)$ and let $\alpha_{0},\ldots,\alpha_{l}$ be a lower path in $S_{k+1}$ such that $\gamma(\alpha_{l})\in T_{k}$. Since $x\in\iota(\alpha_{0})$, then $x\in\gamma\delta(\alpha_{0})$ and, by Lemma 4.2 $\gamma(x)\in\gamma\gamma\delta(\alpha_{0})\subseteq\iota\gamma(\alpha_{0}).$ As $\gamma(\alpha_{0})\leq^{+}\gamma(\alpha_{l})$, by Lemma 5.3, we have $\gamma(x)\in\iota\gamma(\alpha_{l})\cup\gamma\gamma\gamma(\alpha_{l})$. Thus we have two cases: 1. 1. $\gamma(x)\in\iota\gamma(\alpha_{l})$, 2. 2. $\gamma(x)=\gamma\gamma\gamma(\alpha_{l})$. Case 1: $\gamma(x)\in\iota\gamma(\alpha_{l})$. By Lemma 5.2.2, there is a unique $z\in\delta\gamma(\alpha_{l})$ such that $\gamma(z)=\gamma(x)$ and $z<^{+}x$. As $\gamma(\alpha_{l})\in T_{k}$, so $z\in T_{k-1}$. If $y<^{T_{k-1},+}z$ then indeed $y<^{(S\cup T)_{k-1},+}z$, as required. By $\gamma$-linearity in $T_{k-1}$, it is enough to show that it is impossible to have $z<^{T_{k-1},+}y$. Suppose contrary, that there is an upper path $z,b_{0},\ldots,b_{r},y$ in $T$. Since $\gamma(\alpha_{l})$ is $<^{+}$-minimal in $T$ (as $\alpha_{l}\in S$) and $z\in\delta\gamma(\alpha_{l})\cap\delta(b_{0})$, by $\delta$-linearity in $T_{k}$ we have $\gamma(\alpha_{l})<^{+}b_{0}$. By Lemma 5.3, we have $\gamma(x)\in\iota\gamma(\alpha_{l})\subseteq\iota(b_{0})\subseteq\iota(b_{r})$ But $\gamma(b_{r})=y$ so $\gamma\gamma(b_{r})=\gamma(y)=\gamma(x)$. In particular $\gamma(x)\not\in\iota(b_{r})$ and we get a contradiction. Case 2: $\gamma(x)=\gamma\gamma\gamma(\alpha_{l})$. By Lemma 5.2.2 there is $z\in\delta\gamma(\alpha_{l})$ such that $\gamma(x)=\gamma(z)(=\gamma\gamma\gamma(\alpha_{l}))$, so that we have $z<^{S_{k-1},+}x<^{S_{k-1},+}\gamma\gamma(\alpha_{l}).$ As $\gamma(\alpha_{l})\in T_{k}$ and it is $<^{+}$-minimal in $T_{k}$, by Proposition 5.16, there is no face $y\in T_{k-1}$ so that $z<^{T_{k-1},+}y<^{T_{k-1},+}\gamma\gamma(\alpha_{l}).$ So if $y\in T_{k-1}$ and $\gamma(y)=\gamma(x)$ then either $y\leq^{T_{k-1},+}z<^{S_{k-1},+}x\;\;\;\mbox{ or }\;\;\;x<^{S_{k-1},+}\gamma\gamma(\alpha_{l})\leq^{S_{k-1},+}y.$ In either case $x\perp^{{(S\cup T)_{k-1},+}}y$, as required. This ends the proof of $\gamma$-linearity of $(k-1)$-faces in $(S\cup T)$. Finally, we prove the $\delta$-linearity of $(k-1)$-faces in $S\cup T$. Let $x\in\iota(S_{k+1})$ and $y\in T_{k-1}$, $t\in T_{k-2}$ such that $t\in\delta(x)\cap\delta(y)$. We need to show that $x\perp^{(S\cup T)_{k-1},+}y$. Let $\alpha\in S_{k+1}$ such that $x\in\iota(\alpha_{0})$, $a\in\delta(\alpha_{0})$ such that $x=\gamma(a)$, and let $\alpha_{0},\ldots,\alpha_{l}$ be a lower path in $S_{k+1}$ such that $\gamma(\alpha_{l})\in T_{k}$. As $x\in\iota(\alpha_{0})$, using Lemma 4.2 we have $t\in\delta(x)\subseteq\delta\gamma\delta(\alpha_{0})\subseteq\delta\gamma\gamma(\alpha_{0})\cup\iota\gamma(\alpha_{0}).$ As $\gamma(\alpha_{0})<^{+}\gamma(\alpha_{l})$, by Lemma 5.3, we have two cases: 1. 1. $t\in\iota\gamma(\alpha_{l})$, 2. 2. $t\in\delta\gamma\gamma(\alpha_{l})$. Case 1: $t\in\iota\gamma(\alpha_{l})$. By Lemma 5.2.3, there is a unique $z\in\delta\gamma(\alpha_{l})$ such that $t\in\delta(z)$ and $z<^{+}x$. As $\gamma(\alpha_{l})\in T_{k}$, so $z\in T_{k-1}$. If $y<^{T_{k-1},+}z$ then indeed $y<^{(S\cup T)_{k-1},+}z$, as required. By $\delta$-linearity in $T_{k}$, it is enough to show that it is impossible to have $z<^{T_{k-1},+}y$. Suppose contrary, that there is an upper path in $T$ $z,b_{0},\ldots,b_{r},y.$ Since $\gamma(\alpha_{l})$ is $<^{+}$-minimal in $T_{k}$ and $z\in\delta\gamma(\alpha_{l})\cap\delta(b_{0})$, by $\delta$-linearity of $k$-faces in $T$ we have $\gamma(\alpha_{l})<^{+}b_{0}$. By Lemma 5.3, we have $t\in\iota\gamma(\alpha_{l})\subseteq\iota(b_{0})\subseteq\ldots\subseteq\iota(b_{r}).$ But $\gamma(b_{r})=y$, so $t\in\delta(y)\subseteq\delta\gamma(b_{r})$. In particular $t\not\in\iota(b_{r})$ and we get a contradiction. Case 2: $t\in\delta\gamma\gamma(\alpha_{l})$. By Lemma 5.2.3 there is $z\in\delta\gamma\gamma(\alpha_{l})$ such that $t\in\delta(z)$ and we have $z<^{S_{k-1},+}x<^{S_{k-1},+}\gamma\gamma(\alpha_{l}).$ As $\gamma(\alpha_{l})\in T_{k}$, and it is $<^{+}$-minimal face in $T_{k}$, by Lemma 5.16, there is no face $y\in T_{k-1}$ such that $z<^{T_{k-1},+}y<^{T_{k-1},+}\gamma\gamma(\alpha_{l}).$ So if $y\in T_{k-1}$ and $t\in\delta(y)$ then either $y\leq^{T_{k-1},+}z<^{S_{k-1},+}x\;\;\;\mbox{ or }\;\;\;x<^{S_{k-1},+}\gamma\gamma(\alpha_{l})\leq^{S_{k-1},+}y.$ In either case $x\perp^{{(S\cup T)_{k-1},+}}y$, as required. This ends the proof of $\delta$-linearity of $(k-1)$-faces in $(S\cup T)$ and the whole proof that $S\cup T$ is a positive face structure. $~{}~{}\Box$ Let $S$ and $T$ be positive face structures such that ${\bf c}^{(k)}S={\bf d}^{(k)}T$. Then the pushout ${\bf c}^{(k)}S\vphantom{T}$$\vphantom{{\bf c}^{(k)}S}T$$S\vphantom{S+_{k}T}$$\vphantom{S}S+_{k}T$ is called special pushouts Now we shall describe an ${\omega}$-category $S^{*}$ generated by the face structure $S$. The set $S^{*}_{n}$ of $n$-cell of $S^{*}$ is the set of all positive face sub-structures of $S$ of dimension at most $n$, for $n\in{\omega}$. The domain and codomain operations in $S^{*}$ are restricted operations ${\bf d}^{(k)},{\bf c}^{(k)}:S^{*}_{n}\longrightarrow S^{*}_{k}$ of the $k$-th domain and the $k$-th codomain, for $k\leq n$. The identity operation ${\bf i}^{(n)}:S^{*}_{k}\longrightarrow S^{*}_{n}$ is an inclusion and the composition map ${\bf m}_{n,k,n}:S^{*}_{n}\times_{S^{*}_{k}}S^{*}_{n}\longrightarrow S^{*}_{n}$ is the sum, i.e. if $X$, $Y$ are sub-face structures of $S$ of dimension at most $n$ such that ${\bf c}^{(k)}X={\bf d}^{(k)}Y$ then ${\bf m}_{n,k,n}(X,Y)=X+_{k}Y=X\cup Y.$ ###### Corollary 6.3 Let $S$ be a weak positive face structure. Then $S^{*}$ is an ${\omega}$-category. In fact, we have a functor $(-)^{*}:{\bf Fs}^{+/1}\longrightarrow{{\omega}Cat}$ Proof. The fact that the operation on $S^{*}$ defined above satisfy the laws of ${\omega}$-category is obvious. The image $f(X)$ of a sub-face structure $X$ of a positive face structure $S$ under a morphism $f:S\rightarrow T$ is a sub-face structure of $T$. The association $X\mapsto f(X)$ is easily seen to be an ${\omega}$-functor. $~{}~{}\Box$ Let $S$ be a positive face structure. We have a functor $\Sigma^{S}:{\bf pFs}^{+/1}\downarrow S\longrightarrow{\bf Fs}^{+/1}$ such that $\Sigma^{S}(f:B\rightarrow S)=B$ and a cocone $\sigma^{S}:\Sigma^{S}\longrightarrow S$ such that $\sigma^{S}_{(f:B\rightarrow S)}=f:\Sigma^{S}(f:B\rightarrow S)=B\longrightarrow S$ We have ###### Lemma 6.4 The cocone $\sigma^{S}:\Sigma^{S}\stackrel{{\scriptstyle}}{{\longrightarrow}}S$ is a colimiting cocone in ${\bf Fs}^{+/1}$. Such colimiting cones are called special colimits. Any special limit in ${\bf Fs}^{+/1}$ can be obtained via some special pushouts and vice versa any special pushout can be obtained from special limits. In particular, a functor from ${\bf Fs}^{+/1}$ preserves special limits if and only if it preserves special pushouts.$~{}~{}\Box$ ## 7 Normal positive face structures Let $S$ be a $k$-normal positive face structure $S$. By ${\bf p}_{l}^{S}$ we denote the unique element of the set $S_{l}-\delta(S_{l+1})$, for $l<k$. Moreover, as we shall show below ${\bf p}_{k-1}\in\gamma(S_{k})$ and hence the set $\\{x\in S_{k}:\gamma(x)={\bf p}_{k-1}\\}$ is not empty. We denote by ${\bf p}_{k}$ the $<^{+}$-maximal element of this set. We shall omit the superscript $S$ if it does not lead to a confusion. ###### Lemma 7.1 Let $S$ be a $(k-1)$-principal positive face structure of dimension at least $k$, $k>0$. Then 1. 1. $S_{l}=\delta^{(l)}(S_{k})\cup\gamma^{(l)}(S_{k})=\delta^{(l)}(S_{k})\cup\\{{\bf p}_{l}\\}$, for $l<k$. 2. 2. $\delta(S_{l+1})=\delta^{(l)}(S_{k})$, for $l<k$. 3. 3. ${\bf p}_{k}$ is $<^{-}$-largest element in $S_{k}-\delta(S_{k+1})$. 4. 4. $\gamma({\bf p}_{l})={\bf p}_{l-1}$, for $0<l\leq k$. 5. 5. $\delta({\bf p}_{l})=\delta(S_{l})-\gamma(S_{l})$, for $0<l<k$. 6. 6. $S_{l}=\delta^{(l)}({\bf p}_{k-1})\cup\gamma^{(l)}({\bf p}_{k-1})$, for $l<k-2$. Proof. Ad 1. If $H$ is a hypergraph of dimension greater than $l$ and $\gamma(H_{l+1})\subseteq\delta(H_{l+1})$ then there is an infinite lower path in $H_{l+1}$, i.e. $<^{H_{l},+}$ is not strict. Thus, if $S$ is a positive face structure of dimension greater than $l$, we have $\delta(S_{l+1})\subseteq\hskip-11.38109pt_{\not}\hskip 8.53581ptS_{l}$. A positive face structure is normal iff this difference $S_{l}-\delta(S_{l+1})$ is minimal possible (i.e. one-element set), for $l<k$. Thus, by the above, we must have $S_{l}=\delta(S_{l+1})\cup\gamma(S_{l+1})$ (1) The first equation of the statement 1. we shall show by the downward induction on $l$. Suppose that we have $S_{l+1}=\delta^{(l)}(S_{k})\cup\gamma^{(l)}(S_{k})$ (for $l=k-2$ it is true by the above). Then $S_{l}=\delta^{(l)}(S_{k})\cup\gamma^{(l)}(S_{k})=$ $=\delta(\delta^{(l+1)}(S_{k})\cup\gamma^{(l+1)}(S_{k}))\cup\gamma(\delta^{(l+1)}(S_{k})\cup\gamma^{(l+1)}(S_{k}))=$ $=\delta\delta^{(l+1)}(S_{k})\cup\delta\gamma^{(l+1)}(S_{k})\cup\gamma\delta^{(l+1)}(S_{k})\cup\gamma\gamma^{(l+1)}(S_{k})=$ $=\delta^{(l)}(S_{k})\cup\delta\gamma^{(l)}(S_{k})\cup\gamma\delta^{(l)}(S_{k})\cup\gamma^{(l)}(S_{k})=$ $=\delta^{(l)}(S_{k})\cup\gamma^{(l)}(S_{k})$ where the last equation follows from Corollary 4.3. The second equation of 1. is obvious for $l=k-1$. So assume that $l<k-1$. We have $\\{{\bf p}_{l}\\}=S_{l}-\delta(S_{l+1})=$ $=S_{l}-\delta(\delta^{(l+1)}(S_{k})\cup\gamma^{(l+1)}(S_{k}))=$ $=S_{l}-(\delta^{(l)}(S_{k})\cup\delta\gamma^{(l+1)}(S_{k}))=$ $S_{l}-\delta^{(l)}(S_{k}).$ Thus $S_{l}=\delta^{(l)}(S_{k})\cup\\{{\bf p}_{l}\\}$ as required. Ad 2. Let $l<k$ Then using 1. we have $\delta^{(l)}(S_{k})\subseteq\delta(S_{l+1})\subseteq\hskip-11.38109pt_{\not}\hskip 8.53581pt\delta^{(l)}(S_{k})\cup\\{{\bf p}_{l}\\}$ Hence $\delta^{(l)}(S_{k})=\delta(S_{l+1}).$ Ad 3. First we shall show that ${\bf p}_{k}\in S_{k}-\delta(S_{k+1})$. Suppose contrary, that there is $\alpha\in S_{k+1}$ such that ${\bf p}_{k}\in\delta(\alpha)$. Then $\gamma({\bf p}_{k})\in\gamma\delta(\alpha)=\gamma\gamma(\alpha)\cup\iota(\alpha)$. If $\gamma({\bf p}_{k})=\gamma\gamma(\alpha)$ then ${\bf p}_{k}<^{+}\gamma(\alpha)$ i.e. $\gamma(\alpha)$ is $<^{+}$-smaller element than ${\bf p}_{k}$ such that $\gamma(\gamma(\alpha))={\bf p}_{k-1}$. This contradicts the choice of ${\bf p}_{k}$. If $\gamma({\bf p}_{k})=\iota(\alpha)$ then there is $a\in\delta(\alpha)$ such that $\gamma({\bf p}_{k})\in\delta(a)$. But this means that ${\bf p}_{k-1}=\gamma({\bf p}_{k})\in\delta(S_{k})$ contradicting the choice of ${\bf p}_{k-1}\in S_{k-1}-\delta(S_{k})$. This shows that ${\bf p}_{k}\in S_{k}-\delta(S_{k+1})$. We need to prove that any maximal lower $(S_{k}-\delta(S_{k+1}))$-path ends at ${\bf p}_{k}$. By strictness, it is enough to show that if $x\in S_{k}-\delta(S_{k+1})$ and $x\neq{\bf p}_{k}$ then there is $x^{\prime}\in S_{k}-\delta(S_{k+1})$ such that $\gamma(x)\in\delta(x^{\prime})$. So fix $x\in S_{k}-\delta(S_{k+1})$. If we were to have $\gamma(x)\in\iota(\beta)$ for some $\beta\in S_{k+1}$, then by Lemma 5.5 we would have $x<^{+}\gamma(\beta)$. In particular, $x\in\delta(S_{k+1})$, contrary to the assumption. Therefore $\gamma(x)\in S_{k-1}-\iota(S_{k+1})$. As $x,{\bf p}_{k}\in S_{k}-\delta(S_{k+1})$, by $\gamma$-linearity we have $\gamma(x)\neq\gamma({\bf p}_{k})={\bf p}_{k-1}$. Hence by 1. the set $\Delta_{\gamma(x)}=\\{y\in S_{k}:\gamma(x)\in\delta(y)\\}$ is not empty. Let $x^{\prime}$ be the $<^{+}$-maximal element of this set. It remains to show that $x^{\prime}\not\in\delta(S_{k+1})$. Suppose contrary, that there is $\alpha\in S_{k+1}$ such that $x^{\prime}\in\delta(\alpha)$. As $\gamma(x)\not\in\iota(S_{k+1})$ and $\gamma(x)\in\delta(x^{\prime})$ so $\gamma(x)\not\in\iota(\alpha)$ and $\gamma(x)\neq\gamma\gamma(\alpha)$. Thus $\gamma(x)\in\delta\gamma(\alpha)$. But this means that $x^{\prime}<^{+}\gamma(\alpha)$ and $\gamma(\alpha)\in\Delta_{\gamma(x)}$. This contradicts the choice of $x^{\prime}$. This ends the proof of 3. Ad 4. $\gamma({\bf p}_{k})={\bf p}_{k-1}$ by definition. Fix $0<l<k$. As $S_{l}=\delta(S_{l+1})\cup\\{{\bf p}_{l}\\}$, ${\bf p}_{l}$ is $<^{+}$-greatest element in $S_{l}$. Assume that $\gamma({\bf p}_{l})\neq{\bf p}_{l-1}$. Thus $\gamma({\bf p}_{l})<^{+}{\bf p}_{l-1}$. Let $x\in S_{l}$. Then $x\leq{\bf p}_{l}$ and, by Lemma 5.9, $\gamma(x)\leq^{+}\gamma({\bf p}_{l})<^{+}{\bf p}_{l-1}$. Thus ${\bf p}_{l-1}\not\in\gamma(S_{l})$. So $\gamma(S_{l})\subseteq\delta(S_{l})$. But this is impossible in a positive face structure as we noticed in the proof of 1. This ends 4. Ad 5. Fix $l<k$. First we shall show that $\delta({\bf p}_{l})\cap\gamma(S_{l})=\emptyset$ (2) Let $z\in\gamma(S_{l})$, i.e. there is $a\in S_{l}$ such that $\gamma(a)=z$. By 1. $a\leq^{+}{\bf p}_{l}$. By Lemma 5.7, there are $x\in\delta({\bf p}_{l})$ and $y\in\delta(a)$ such that $x\leq^{+}y$. Hence $x<^{+}\gamma(a)=z$. By Proposition 5.1, since $x\in\delta({\bf p}_{l})$ it follows that $z\not\in\delta({\bf p}_{l})$. This shows (2). By Lemma 5.19, we have $\delta(S_{l})=\delta(S_{l}-\delta({S_{l+1}}))\cup\iota(S_{l+1})$ (3) Since $\delta({\bf p}_{l})=\delta(S_{l}-\delta({S_{l+1}}))$ and $\iota(S_{l+1})\subseteq\gamma(S_{l})$ we have by (2) $\delta(S_{l}-\delta({S_{l+1}}))\cap\iota(S_{l+1})=\emptyset$ (4) Next we shall show that $\iota(S_{l+1})=\gamma(S_{l})\cap\delta(S_{l})$ (5) The inclusion $\subseteq$ is obvious. Let $x\in\gamma(S_{l})\cap\delta(S_{l})$. Hence there are $a,b\in S_{l}$ such that $\gamma(a)=x\in\delta(b)$. We can assume that $a$ is $<^{+}$-maximal with this property. As $a<^{-}b$, neither $a$ nor $b$ is equal to the $<^{+}$-greatest element ${\bf p}_{l}\in S_{l}$. Therefore there is $\alpha\in S_{l+1}$ such that $a\in\delta(\alpha)$. If we were to have $x=\gamma(a)=\gamma\gamma(\alpha)$ then $\gamma(\alpha)$ would be a $<^{+}$-greater element than $a$ with $\gamma(\gamma(\alpha))=x$. So $\gamma(a)\neq\gamma\gamma(\alpha)$. Clearly, $x\in\gamma\delta(\alpha)$. By globularity, $x\in\delta\delta(\alpha)$, as well. Thus $x\in\iota(\alpha)$, and (5) is shown. Using (2), (3), (4), and (5) we have $\delta({\bf p}_{l})=\delta(S_{l}-\delta(S_{l+1}))=$ $=\delta(S_{l})-\iota(S_{l+1})=$ $=\delta(S_{l})-(\gamma(S_{l})\cap\delta(S_{l}))=$ $=\delta(S_{l})-\gamma(S_{l})$ as required. Ad 6. By 1. and 2. it is enough to show $\delta^{(l)}(S_{k-1})=\delta^{(l)}({\bf p}_{k-1}),$ for $l<k-2$. The inclusion $\supseteq$ is obvious. Pick $x\in S_{k-1}$ We have an upper path $x,a_{1},\ldots,a_{r},{\bf p}_{k-1}$. By Corollary 4.3, as $\gamma(a_{i})\in\delta(a_{i+1})$, we have $\delta^{(l)}(a_{i})=\delta^{(l)}\gamma(a_{i})\subseteq\delta^{(l)}(\delta(a_{i+1}))=\delta^{(l)}(a_{i+1})$ for $i=0,\ldots,r-1$. Then, by transitivity of $\subseteq$ and again Corollary 4.3 we get $\delta^{(l)}(x)\subseteq\delta^{(l)}(a_{1})\subseteq\delta^{(l)}(a_{r})\subseteq\delta^{(l)}(\gamma(a_{r}))=\delta^{(l)}({\bf p}_{k-1}).$ This ends the proof of the inclusion $\subseteq$ and 6. $~{}\Box$ ###### Lemma 7.2 Let $S$ be a positive face structure of dimension at least $k$. Then 1. 1. $S$ is $(k-1)$-principal iff $\;{\bf d}^{(k)}(S)$ is normal iff $\;{\bf c}^{(k-1)}(S)$ is principal, 2. 2. if $S$ is normal, so is ${\bf d}(S)$, 3. 3. if $S$ is principal, so is ${\bf c}(S)$. Proof. The whole Lemma is an easy consequence of Lemma 5.19. We shall show 1. leaving 2. and 3. for the reader. First note that all three conditions in 1. imply that, $|S_{l}-\delta(S_{l+1}|=1$ for $l<k-2$. In addition to this they say: 1. 1. $S$ is $(k-1)$-principal iff $|S_{l}-\delta(S_{l+1})|=1$ for $l=k-2,k-1$. 2. 2. ${\bf d}^{(k)}(S)$ is normal iff 1. (a) $|S_{k-1}-\delta(S_{k}-\gamma(S_{k+1}))|=1$, and 2. (b) $|S_{k-2}-\delta(S_{k-1})|=1$. 3. 3. ${\bf c}^{(k-1)}(S)$ is principal iff 1. (a) $|(S_{k-1}-\iota(S_{k+1}))-\delta(S_{k}-\delta({S_{k+1}}))|=1$, and 2. (b) $|S_{k-2}-\delta(S_{k-1}-\iota({S_{k+1}}))|=1$. So the equivalence of these conditions follows directly from Lemma 5.19. $~{}~{}\Box$ Let $N$ be a $n$-normal positive face structure. We define a $(n+1)$-hypergraph $N^{\bullet}$, that contains two additional faces: ${\bf p}_{n+1}^{N^{\bullet}}$ of dimension $n+1$, and ${\bf p}_{n}^{N^{\bullet}}$ of dimension $n$. We shall drop superscripts if it does not lead to confusions. We also put $\delta({\bf p}_{n+1})=N_{n},\;\;\;\;\gamma({\bf p}_{n+1})={\bf p}_{n},$ $\delta({\bf p}_{n})=\delta(N_{n})-\gamma(N_{n}),\;\;\;\;\gamma({\bf p}_{n})={\bf p}_{n-1}(=\gamma(N_{n})-\delta(N_{n})).$ As $N$ is normal the $\gamma(N_{n})-\delta(N_{n})$ has one element so $\gamma({\bf p}_{n})$ is well defined. This determines $N^{\bullet}$ uniquely. $N^{\bullet}$ is called a simple extension of $N$. Example. For a normal positive face structure $N$ like this $x_{2}\vphantom{x_{1}}$$\vphantom{x_{2}}x_{1}$$f_{1}$$f_{2}$$f_{0}$$x_{3}\vphantom{x_{0}}$$\vphantom{x_{3}}x_{0}$ the hyper-graph $N^{\bullet}$ looks like this $x_{2}\vphantom{x_{1}}$$\vphantom{x_{2}}x_{1}$$f_{1}$$f_{2}$$f_{0}$$x_{3}\vphantom{x_{0}}$$\vphantom{x_{3}}x_{0}$${\bf p}^{N^{\bullet}}_{1}$$\Downarrow{\bf p}^{N^{\bullet}}_{2}$ We have ###### Proposition 7.3 Let $N$ be a normal positive face structure of dimension $n$. Then 1. 1. $N^{\bullet}$ is a principal positive face structure of dimension $n+1$. 2. 2. We have ${\bf d}(N^{\bullet})\cong N$, ${\bf c}(N^{\bullet})\cong({\bf d}N)^{\bullet}$. 3. 3. If $N$ is a principal, then $N\cong({\bf d}N)^{\bullet}$. 4. 4. If $T$ is a positive sub-face structure of $N^{\bullet}$ then either $T=N^{\bullet}$ or $T={\bf c}(N^{\bullet})$ or $T\subseteq N$. Proof. Ad 1. We shall check globularity of the new added cells. The other conditions are simple. For ${\bf p}_{n+1}$, we have: $\gamma\gamma({\bf p}_{n+1})=\gamma({\bf p}_{n})=$ $=\gamma(N_{n})-\delta(N_{n})=\gamma\delta({\bf p}_{n+1})-\delta\delta({\bf p}_{n+1})$ and $\delta\gamma({\bf p}_{n+1})=\delta({\bf p}_{n})=$ $=\delta(N_{n})-\gamma(N_{n})=\delta\delta({\bf p}_{n+1})-\gamma\delta({\bf p}_{n+1}).$ So globularity holds for ${\bf p}_{n+1}$. For ${\bf p}_{n}$, using Lemmas 7.1, 5.19 and normality of $N$, we have: $\gamma\gamma({\bf p}_{n})=\gamma({\bf p}_{n-1})={\bf p}_{n-2}=$ $=\gamma(N_{n-1})-\delta(N_{n-1})=$ $=\gamma(N_{n-1}-\gamma(N_{n}))-\delta(N_{n-1}-\gamma(N_{n}))=$ $=\gamma(\delta(N_{n})-\gamma(N_{n}))-\delta(\delta(N_{n})-\gamma(N_{n}))=$ $=\gamma\delta({\bf p}_{n})-\delta\delta({\bf p}_{n})$ and similarly $\delta\gamma({\bf p}_{n})=\delta({\bf p}_{n-1})=$ $=\delta(N_{n})-\gamma(N_{n})=$ $=\delta(\delta(N_{n})-\gamma(N_{n}))-\gamma(\delta(N_{n})-\gamma(N_{n}))=$ $=\delta\delta({\bf p}_{n})-\gamma\delta({\bf p}_{n})$ So globularity for ${\bf p}_{n}$ holds, as well. Ad 2. The first isomorphism is obvious. The faces of $(N^{\bullet})$, ${\bf c}(N^{\bullet})$, ${\bf d}N$, and $({\bf d}N)^{\bullet}$ are as in the tables $\begin{array}[]{|c|c|c|}dim&(N^{\bullet})&{\bf c}(N^{\bullet})\\\ \hline\cr n+1&\\{{\bf p}_{n+1}^{N^{\bullet}}\\}&\emptyset\\\ \hline\cr n&N_{n}\cup\\{{\bf p}_{n}^{N^{\bullet}}\\}&\\{{\bf p}_{n}^{N^{\bullet}}\\}\\\ \hline\cr n-1&N_{n-1}&N_{n-1}-(\gamma(N_{n})\cap\delta(N_{n}))\\\ \hline\cr n-2&N_{n-2}&N_{n-2}\end{array}$ and $\begin{array}[]{|c|c|c|}dim&{\bf d}N&({\bf d}N)^{\bullet}\\\ \hline\cr n+1&\emptyset&\emptyset\\\ \hline\cr n&\emptyset&\\{{\bf p}_{n}^{({\bf d}N)^{\bullet}}\\}\\\ \hline\cr n-1&N_{n-1}-\gamma(N_{n})&(N_{n-1}-\gamma(N_{n}))\cup\\{{\bf p}_{n-1}^{({\bf d}N)^{\bullet}}\\}\\\ \hline\cr n-2&N_{n-2}&N_{n-2}\end{array}$ We define the isomorphism $f:{\bf c}(N^{\bullet})\longrightarrow({\bf d}N)^{\bullet}$ as follows $f_{n}({\bf p}_{n+1}^{N^{\bullet}})={\bf p}_{n+1}^{({\bf d}N)^{\bullet}},$ $f_{n-1}(x)=\left\\{\begin{array}[]{ll}{\bf p}_{n-1}^{({\bf d}N)^{\bullet}}&\mbox{if $x=\gamma({\bf p}_{n}^{N^{\bullet}})$,}\\\ x&\mbox{otherwise.}\end{array}\right.$ and $f_{l}=1_{N_{l}}$ for $l<n-1$. Clearly, all $f_{i}$’s are bijective. The preservation of the domains and codomains is left for the reader. 3\. is left as an exercise. Ad 4. If ${\bf p}_{n+1}\in T_{n+1}$ then $T=N^{\bullet}$. If ${\bf p}_{n}\not\in T_{n}$ then $T\subseteq N$. Suppose that ${\bf p}_{n+1}\not\in T_{n+1}$ but ${\bf p}_{n}\in T_{n}$. Since $N^{\bullet}=[{\bf p}_{n+1}]$, by Lemma 6.1 it is enough to show that $T=[{\bf p}_{n}]$. Clearly $[{\bf p}_{n}]\subseteq T$. As $[{\bf p}_{n}]_{l}=N_{l}$, for $l<n-1$ we have $[{\bf p}_{n}]_{l}=T_{l}$, for $l<n-1$, as well. Fix $x\in N_{n}$. As $x\in\delta({\bf p}_{n+1})$ and $\gamma({\bf p}_{n+1})={\bf p}_{n}$, we have $x<^{N^{\bullet},+}{\bf p}_{n}$. So by Corollary 5.11 $x\not\perp^{N^{\bullet},-}_{l}{\bf p}_{n}$, for any $l\leq n$. Thus we cannot have $x\perp^{T,-}_{l}{\bf p}_{n}$, for any $l\leq n$, as well. As $T$ is a positive face structure, again by Corollary 5.11, $x\not\in T$. Since $x$ was an arbitrary element of $N_{n}$, we have $T_{n}=\\{{\bf p}_{n}\\}=[{\bf p}_{n}]_{n}$. It remains to show that $T_{n-1}=[{\bf p}_{n}]_{n-1}$. Suppose that $x\in N_{n-1}-(\delta({\bf p}_{n})\cup\gamma({\bf p}_{n}))$. Then $x<^{N,+}\gamma({\bf p}_{n})$ and hence $x\not\perp^{N^{\bullet},-}_{l}\gamma({\bf p}_{n})$, for $l\leq n$. So $x$ and $\gamma({\bf p}_{n})$ cannot be $<^{T,-}_{l}$ comparable, for $l\leq n$. Since, as we have shown, $N_{n}\cap T_{n}=\emptyset$, it follows that $x$ and $\gamma({\bf p}_{n})$ cannot be $<^{T,+}$ comparable. So by Lemma 5.11, $x\not\in T_{n-1}$, i.e. $T_{n-1}=\delta({\bf p}_{n})\cup\gamma({\bf p}_{n})=[{\bf p}_{n}]_{n-1}$. $~{}~{}\Box$ ## 8 Decomposition of positive face structures Let $T$ be a positive face structure, $X\subseteq T$ a subhypergraph of $T$, $k\in{\omega}$, $a\in(T_{k}-\iota(T_{k+2}))$. We define two subhypergraphs of $T$, $X^{\downarrow a}$ and $X^{\uparrow a}$, as follows: $X^{\downarrow a}_{l}=\left\\{\begin{array}[]{ll}\\{\alpha\in X_{l}:\gamma^{(k)}(\alpha)\leq^{+}a\\}&\mbox{for $l>k$,}\\\ \\{b\in X_{k}:b\leq^{+}a\;{\rm or}\;\;b\not\in\gamma(X_{k+1})\\}&\mbox{for $l=k$}\\\ X_{l}&\mbox{for $l<k$.}\end{array}\right.$ $X^{\uparrow a}_{l}=\left\\{\begin{array}[]{ll}\\{\alpha\in X_{l}:\gamma^{(k)}(\alpha)\not\leq^{+}a\\}&\mbox{for $l>k$,}\\\ \\{b\in X_{k}:b\not<^{+}a\mbox{ or }b\not\in\delta(X_{k+1})\\}&\mbox{for $l=k$}\\\ X_{k-1}-\iota(X^{\downarrow a}_{k+1})&\mbox{for $l=k-1$}\\\ X_{l}&\mbox{for $l<k-1$.}\end{array}\right.$ Intuitively, if $X$ is a positive face substructure, $X^{\downarrow a}$ is the least positive face substructure of $X$ that contains faces ’smaller or equal’ $a$ and can be $k$-pre-composed with the ’rest’ to get $X$. $X^{\uparrow a}$ is this ’rest’ or in other words it is the largest positive face substructure of $X$ that can be $k$-post-composed with $X^{\downarrow a}$ to get $X$ (or largest positive face substructure of $X$ that do not contains faces ’smaller’ than $a$). Examples. If $X$ is a hypergraph $a\in T$ then $X^{\downarrow a}$ is a hypergraph as well. However, this is not the case with $X^{\uparrow a}$, if $a\in\iota(T)$, as we can see below: $\cdot\vphantom{\cdot}$$\vphantom{\cdot}\cdot$$a$$\Downarrow$$X$:$\cdot\vphantom{\cdot}$$\vphantom{\cdot}\cdot$$a$$\Downarrow$$X^{\uparrow a}$: Here $X=T$. The faces in the domain of the 2-dimensional face are not in $X^{\uparrow a}$, i.e. $X^{\uparrow a}$ is not closed under $\delta$. To see some real decompositions let fix a positive face structure $T$ as follows: $T$$\bullet\vphantom{}$$\bullet$$\vphantom{}y$$\bullet$$\Downarrow$$\Downarrow$$b$$a$$\bullet\vphantom{\bullet}$$\vphantom{\bullet}\bullet$$\bullet$$\Downarrow$$x\vphantom{}$$\vphantom{}\bullet$$\Downarrow$$\Downarrow$ Clearly $x,y,a,b\in T-\iota(T)$. Then $T^{\downarrow a}$$\bullet\vphantom{}$$\bullet$$\vphantom{}y$$\bullet$$b$$a$$\bullet\vphantom{\bullet}$$\vphantom{\bullet}\bullet$$\bullet$$\Downarrow$$x\vphantom{}$$\vphantom{}\bullet$$\Downarrow$ and $T^{\uparrow a}$$\bullet\vphantom{}$$\bullet$$\vphantom{}y$$\bullet$$\Downarrow$$\Downarrow$$a$$\vphantom{}\bullet$$x\vphantom{}$$\vphantom{}\bullet$$\Downarrow$ Moreover with $X_{1}$$\bullet\vphantom{}$$\bullet$$\vphantom{}y$$\bullet$$\Downarrow$$\Downarrow$$a$$\vphantom{}\bullet$$x\vphantom{}$$\vphantom{}\bullet$$\Downarrow$ we have $X_{1}^{\uparrow b}=X_{1}$ and $X_{1}^{\downarrow b}$$\bullet\vphantom{}$$\bullet$$\vphantom{}y$$\bullet$$a$$\vphantom{}\bullet$$x\vphantom{}$$\vphantom{}\bullet$ i.e. $X_{1}^{\downarrow b}={\bf d}^{(1)}(X_{1})$. For $X_{2}$$\bullet\vphantom{}$$\bullet$$\vphantom{}y$$\bullet$$\Downarrow$$\Downarrow$ we have $X_{2}^{\downarrow x}=X_{2}$ and $X_{2}^{\uparrow x}=\\{y\\}$. We have ###### Lemma 8.1 Let $T$ be a positive face structure, $X\subseteq T$ a subhypergraph of $T$, $a\in(T-\iota(T))$, $a\in X_{k}$. Then 1. 1. $X^{\downarrow a}$ and $X^{\uparrow a}$ are positive face structures; 2. 2. ${\bf c}^{(k)}(X^{\downarrow a})={\bf d}^{(k)}(X^{\uparrow a})=X^{\downarrow a}\cap X^{\uparrow a}$; 3. 3. ${\bf d}^{(k)}(X^{\downarrow a})={\bf d}^{(k)}(X)$, ${\bf c}^{(k)}(X^{\uparrow a})={\bf c}^{(k)}(X)$; 4. 4. $X=X^{\downarrow a}+_{k}X^{\uparrow a}=X^{\downarrow a}\cup X^{\uparrow a}$. Proof. Ad 1. The verification that both $X^{\downarrow a}$ and $X^{\uparrow a}$ are closed under $\gamma$ and $\delta$ is routine. For any $k$, if $x,y\in X^{\downarrow a}_{k}$ then $x<^{+,X}y$ iff $x<^{+,X^{\downarrow a}}y$. Similarly, for any $k$, if $x,y\in X^{\uparrow a}_{k}$ then $x<^{+,X}y$ iff $x<^{+,X^{\uparrow a}}y$. Thus by Lemma 5.17 both $X^{\downarrow a}$ and $X^{\uparrow a}$ positive face structures. Ad 2. Let as spell in details both sides of the equation. ${\bf c}^{(k)}(X^{\downarrow a})$ is: 1. 1. ${\bf c}^{(k)}(X^{\downarrow a})_{l}=\emptyset$, for $l>k$; 2. 2. ${\bf c}^{(k)}(X^{\downarrow a})_{k}=\\\ (\\{b\in X_{k}:b\leq^{+}a\\}\cup(X_{k}-\gamma(X_{k+1})))-\delta(\\{\alpha\in X_{k+1}:\gamma(\alpha)\leq^{+}a\\})$; 3. 3. ${\bf c}^{(k)}(X^{\downarrow a})_{k-1}=X_{k-1}-\iota(X^{\downarrow a}_{k+1})$; 4. 4. ${\bf c}^{(k)}(X^{\downarrow a})_{l}=X_{l}$, for $l<k-1$. and ${\bf d}^{(k)}(X^{\uparrow a})$ is: 1. 1. ${\bf d}^{(k)}(X^{\uparrow a})_{l}=\emptyset$, for $l>k$; 2. 2. ${\bf d}^{(k)}(X^{\uparrow a})_{k}=\\\ \\{b\in X_{k}:b\not<^{+}a\mbox{ or }b\not\in\delta(X_{k+1})\\}-\gamma(X_{k+1}-\\{\alpha\in X_{k+1}:\gamma(\alpha)\leq^{+}a\\})$; 3. 3. ${\bf d}^{(k)}(X^{\uparrow a})_{k-1}=X_{k-1}-\iota(X^{\downarrow a}_{k+1})$; 4. 4. ${\bf d}^{(k)}(X^{\uparrow a})_{l}=X_{l}$, for $l<k-1$. Thus to show that ${\bf c}^{(k)}(X^{\downarrow a})={\bf d}^{(k)}(X^{\uparrow a})$ we need to verify that ${\bf c}^{(k)}(X^{\downarrow a})_{k}={\bf d}^{(k)}(X^{\uparrow a})_{k}$. As both sets are contained in $X_{k}$, we can compare their complements. We have $X_{k}-{\bf c}^{(k)}(X^{\downarrow a})_{k}=\\{b\in\delta(X_{k+1}):b<^{+}a\\}\cup\gamma(X_{k+1}-\\{\alpha\in X_{k+1}:\gamma(\alpha)\not\leq^{+}a\\})$ and $X_{k}-{\bf d}^{(k)}(X^{\uparrow a})_{k}=\\{b\in\gamma(X_{k+1}):b\not\leq^{+}a\\}\cup\delta(\\{\alpha\in X_{k+1}\gamma(\alpha)\leq^{+}a\\}).$ But it easy to see that $\\{b\in\delta(X_{k+1}):b<^{+}a\\}=\delta(\\{\alpha\in X_{k+1}\gamma(\alpha)\leq^{+}a\\})$ and $\gamma(X_{k+1}-\\{\alpha\in X_{k+1}:\gamma(\alpha)\not\leq^{+}a\\})=\\{b\in\gamma(X_{k+1}):b\not\leq^{+}a\\}.$ The second equality uses the fact that $a\not\in\iota(T)$. Thus ${\bf c}^{(k)}(X^{\downarrow a})_{k}={\bf d}^{(k)}(X^{\uparrow a})_{k}$, as required. Ad 3. To see that ${\bf c}^{(k)}(X^{\uparrow a})={\bf c}^{(k)}(X)$ it is enough to note that $\iota(X_{k+1})=\iota(X^{\downarrow a}_{k+1})\cup\iota(X^{\uparrow a}_{k+1})$. The equation ${\bf d}^{(k)}(X^{\downarrow a})={\bf d}^{(k)}(X)$ is even simpler. Ad 4. Obvious. $~{}~{}\Box$ ###### Corollary 8.2 Let $T$ be a positive face structure, $k\in{\omega}$, $a\in(T_{k}-\iota(T_{k+2}))$. Then the square ${\bf c}^{(k)}(T^{\downarrow a})\vphantom{T^{\uparrow a}}$$\vphantom{{\bf c}^{(k)}(T^{\downarrow a})}T^{\uparrow a}$$T^{\downarrow a}\vphantom{T}$$\vphantom{T^{\downarrow a}}T$ is a special pushout in ${\bf Fs}^{+/1}$. Proof. Follows immediately from Lemmas 6.2 and 8.1. $~{}~{}\Box$ We need some notions and notations. Let $X$, $T$ be a positive face structures $X\subseteq T$, $a\in(T-\iota(T))$. The decomposition $X=X^{\downarrow a}\cup X^{\uparrow a}$ is said to be proper iff $size(X^{\downarrow a}),size(X^{\uparrow a})<size(X)$. If the decomposition $X=X^{\downarrow a}\cup X^{\uparrow a}$ is proper then $a$ is said to be a saddle face of $X$. $Sd(X)$ is the set of saddle faces of $X$; $Sd(X)_{k}=Sd(X)\cap X_{k}$. ###### Lemma 8.3 Let $X$, $S$, $T$ be positive face structures, $X\subseteq T$, $l\in{\omega}$. Then 1. 1. if $a\in(T_{l}-\iota(T))$ then $a\in Sd(X)$ iff there are $\alpha,\beta\in X_{l+1}$ such that $\gamma(\alpha)\leq^{+}a$ and $\gamma(\beta)\not\leq^{+}a$; 2. 2. if ${\bf c}^{(k)}(S)={\bf d}^{(k)}(T)$ then $size(S+_{k}T)_{l}=\left\\{\begin{array}[]{ll}size(S)_{l}+size(T)_{l}&\mbox{ if $l>k$,}\\\ size(T)_{l}&\mbox{ if $l\leq k$;}\end{array}\right.$ 3. 3. $size(S)_{k}\geq 1$ iff $k\leq dim(S)$; 4. 4. if $a\in Sd(S)_{k}$ then $size(S)_{k+1}\geq 2$; 5. 5. $S$ is principal iff $Sd(S)$ is empty. Proof. We shall show 5. The rest is easy. If there is $a\in Sd(S)_{k}$ then by 2., 3. and Lemma 8.1 we have that $size(S)_{k+1}=size(S^{\downarrow a})_{k+1}+size(S^{\uparrow a})_{k+1}\geq 1+1>1$. So in that case $S$ is not principal. For the converse, assume that $S$ is not principal. Fix $k\in{\omega}$ such, that $size(S)_{k+1}>1$. Thus there are $a,b\in S_{k+1}$, that $a\neq b$. Suppose $\gamma(a)\in\iota(\alpha)$ for some $\alpha\in S_{k+2}$. Then by Lemma 5.5, $a<^{+}\gamma(\alpha)$ contrary to the assumption on $a$. Hence $a\in S-\iota(S)$ and for similar reasons $b\in S-\iota(S)$. We have $a\not\perp^{+}b$ and, by pencil linearity, $\gamma(a)\neq\gamma(b)$. Then either $\gamma(a)\not<^{+}\gamma(b)$ and then $\gamma(b)\in Sd(S)_{k}$ or $\gamma(b)\not<^{+}\gamma(a)$ and then $\gamma(a)\in Sd(S)_{k}$. In either case $Sd(S)$ is not empty, as required. $~{}\Box$ ###### Lemma 8.4 Let $T$, $X$ be positive face structures, $X\subseteq T$, and $a,x\in X-\iota(X)$, $k=dim(x)<dim(a)=m$. 1. 1. We have the following equations of positive face structures: $X^{\downarrow x\downarrow a}=X^{\downarrow a\downarrow x},\;\;\;X^{\downarrow x\uparrow a}=X^{\uparrow a\downarrow x},\;\;\;X^{\uparrow x\downarrow a}=X^{\downarrow a\uparrow x},\;\;\;X^{\uparrow x\uparrow a}=X^{\uparrow a\uparrow x},$ i.e. ’the decompositions of different dimension commute’. 2. 2. If $x\in Sd(X)$ then $x\in Sd(X^{\downarrow a})\cap Sd(X^{\uparrow a})$. 3. 3. Moreover, we have the following equations concerning domains and codomains ${\bf c}^{(k)}(X^{\downarrow x\downarrow a})={\bf c}^{(k)}(X^{\downarrow x\uparrow a})={\bf d}^{(k)}(X^{\uparrow x\downarrow a})={\bf d}^{(k)}(X^{\uparrow x\uparrow a})$ ${\bf c}^{(m)}(X^{\downarrow x\downarrow a})={\bf d}^{(m)}(X^{\downarrow x\uparrow a}),\;\;\;\;{\bf c}^{(m)}(X^{\uparrow x\downarrow a})={\bf d}^{(m)}(X^{\uparrow x\uparrow a}).$ 4. 4. Finally, we have the following equations concerning compositions $X^{\downarrow x\downarrow a}+_{m}X^{\downarrow x\uparrow a}=X^{\downarrow x},\;\;\;\;X^{\uparrow x\downarrow a}+_{m}X^{\uparrow x\uparrow a}=X^{\uparrow x}$ $X^{\downarrow x\downarrow a}+_{k}X^{\uparrow x\downarrow a}=X^{\downarrow a},\;\;\;\;X^{\downarrow x\uparrow a}+_{k}X^{\uparrow x\uparrow a}=X^{\uparrow a}.$ Proof. Simple check. $~{}\Box$ ###### Lemma 8.5 Let $T$, $X$ be positive face structures, $X\subseteq T$, and $a,b\in X-\iota(X)$, $dim(a)=dim(b)=m$. 1. 1. We have the following equations of positive face structures: $X^{\downarrow a\downarrow b}=X^{\downarrow b\downarrow a},\;\;\;X^{\uparrow a\uparrow b}=X^{\uparrow b\uparrow a},$ i.e. ’the decompositions in the same dimension and the same directions commute’. 2. 2. Assume $a<^{+}b$. Then we have the following farther equations of positive face structures: $X^{\downarrow a}=X^{\downarrow a\downarrow b},\;\;\;X^{\uparrow b}=X^{\uparrow a\uparrow b},\;\;\;X^{\downarrow b\uparrow a}=X^{\uparrow a\downarrow b}.$ Moreover, if $a,b\in Sd(X)$ then $a\in Sd(X^{\downarrow b})$ and $b\in Sd(X^{\uparrow a})$. 3. 3. Assume $a<^{-}_{l}b$, for some $l<m$. Then $X^{\uparrow b\downarrow a}$, $X^{\uparrow a\downarrow b}$, are positive face structures, and $X^{\downarrow a}+_{m}X^{\uparrow a\downarrow b}=X^{\downarrow b}+_{m}X^{\uparrow b\downarrow a}$ Moreover, if $a,b\in Sd(X)$ then either there is $k$ such that $l-1\leq k<m$ and $\gamma^{(k)}(a)\in Sd(X)$ or $a\in Sd(X^{\uparrow b})$ and $b\in Sd(X^{\uparrow a})$. Proof. Simple check. $~{}\Box$ ###### Lemma 8.6 Let $T$, $X$ be positive face structures, $X\subseteq T$, $dim(X)=n$, $l<n-1$, $a\in Sd(X)_{l}$. Then 1. 1. $a\in Sd({\bf c}X)\cap Sd({\bf d}X)$; 2. 2. ${\bf d}(X^{\downarrow a})=({\bf d}X)^{\downarrow a}$; 3. 3. ${\bf d}(X^{\uparrow a})=({\bf d}X)^{\uparrow a}$; 4. 4. ${\bf c}(X^{\downarrow a})=({\bf c}X)^{\downarrow a}$; 5. 5. ${\bf c}(X^{\uparrow a})=({\bf c}X)^{\uparrow a}$. Proof. The proof is again by a long and simple check. We shall check part of 5. We should consider separately cases: $l=n-2$, $l=n-3$, and $l<n-3$, but we shall check the case $l=n-3$ only. The other cases can be also shown by similar, but easier, check. $({\bf c}X)^{\uparrow a}$ is: 1. 1. $({\bf c}X)^{\uparrow a}_{l}=\emptyset$, for $l\geq n$; 2. 2. $({\bf c}X)^{\uparrow a}_{n-1}=\\{x\in X_{n-1}:\gamma^{(n-3)}(x)\not\leq^{+}a,\;\;x\not\in\delta(X_{n})\\}$; 3. 3. $({\bf c}X)^{\uparrow a}_{n-2}=\\{x\in X_{n-2}:\gamma(x)\not\leq^{+}a,\;\;x\not\in\iota(X_{n})\\}$; 4. 4. $({\bf c}X)^{\uparrow a}_{n-3}=\\{x\in X_{n-3}:x\not<^{+}a\mbox{ or }x\not\in\delta(X_{n-2}-\iota(X_{n}))\\}$; 5. 5. $({\bf c}X)^{\uparrow a}_{n-4}=X_{n-4}-\iota(\\{x\in X_{n-2}:x\not\in\iota(X_{n}),\;\gamma(x)\leq^{+}a\\})$; 6. 6. $X^{\downarrow a}_{l}=X_{l}$, for $l<n-4$. and ${\bf c}(X^{\uparrow a})$ is: 1. 1. ${\bf c}(X^{\uparrow a})_{l}=\emptyset$, for $l\geq n$; 2. 2. ${\bf c}(X^{\uparrow a})_{n-1}=\\{x\\!\in\\!X_{n-1}:\gamma^{(n-3)}(x)\not\leq^{+}a\\}-\delta(\\{z\\!\in\\!X_{n}:\gamma^{(n-3)}(z)\not\leq^{+}a\\})$; 3. 3. ${\bf c}(X^{\uparrow a})_{n-2}=\\{x\\!\in\\!X_{n-2}:\gamma(x)\not\leq^{+}a\\}-\iota(\\{z\\!\in\\!X_{n}:\gamma^{(n-3)}(z)\not\leq^{+}a\\})$; 4. 4. ${\bf c}(X^{\uparrow a})_{n-3}=\\{x\in X_{n-3}:x\not<^{+}a\mbox{ or }x\not\in\delta(X_{n-2})\\}$; 5. 5. ${\bf c}(X^{\uparrow a})_{n-4}=X_{n-4}-\iota(X^{\downarrow a}_{n-2})$; 6. 6. ${\bf c}(X^{\uparrow a})_{l}=X_{l}$, for $l<n-4$. We need to verify the equality $({\bf c}X)^{\uparrow a}_{l}={\bf c}(X^{\uparrow a})_{l}$ for $l=n-1,\ldots,n-4$. In dimension $n-1$, it is enough to show that if $x\in X_{n-1}$ and $z\in X_{n}$ so that $\gamma^{(n-3)}(x)\not\leq^{+}a$ and $x\in\delta(z)$ then $\gamma^{n-3}(z)\not\leq^{+}a$. So assume that $x\in X_{n-1}$, $\gamma^{(n-3)}(x)\not\leq^{+}a$, $z\in X_{n}$ such that $x\in\delta(z)$. Hence $x\lhd^{+}\gamma(z)$. By Lemma 5.9.5 $\gamma^{(n-3)}(x)\leq^{+}\gamma^{(n-3)}(z)$. Therefore $\gamma^{(n-3)}(z)\not\leq^{+}a$ (otherwise we would have $\gamma^{(n-3)}(x)\not\leq^{+}a$), as required. In dimension $n-2$, it is enough to show that if $x\in X_{n-2}$ and $z\in X_{n}$ so that $x\not\leq^{+}a$ and $x\in\iota(z)$ then $\gamma^{n-3}(z)\not\leq^{+}a$. So assume that $x\in X_{n-2}$, $z\in X_{n}$ so that $x\not\leq^{+}a$ and $x\in\iota(z)$. Hence $x\leq^{+}\gamma\gamma(z)$. By Lemma 5.9.5 $\gamma(x)\leq^{+}\gamma^{(n-3)}(z)$. Therefore $\gamma^{(n-3)}(z)\not\leq^{+}a$, as required. The equality in dimension $n-3$ follows immediately from Lemma 5.19.4. To show that in dimension $n-4$, the above equation also holds we shall show that $\iota(X^{\downarrow a}_{n-2})\subseteq\iota(\\{x\in X_{n-2}:x\not\in\iota(X_{n}),\;\gamma(x)\leq^{+}a\\})$ Note that, by Lemma 5.3.1, if $t\in X_{n-4}$ and $x\in X_{n-2}$, $y\in X_{n-1}$, $t\in\iota(x)$ and $\gamma(x)\leq^{+}a$ and $x=\gamma(y)$ then there is $x^{\prime}\in\delta(y)$ (i.e. $x^{\prime}\lhd^{+}x$ and hence $\gamma(x^{\prime})\leq^{+}a$) such that $t\in\iota(x^{\prime})$. Thus, as $<^{+}$ is well founded, from the above observation follows that, for any $t\in X_{n-4}$ and $x\in X_{n-2}$ that $t\in\iota(x)$ and $\gamma(x)\leq^{+}a$, there is $x^{\prime\prime}\leq^{+}x$ such that $t\in\iota(x^{\prime\prime})$ and $x^{\prime\prime}\not\in\gamma(X)$. Then we clearly have that $x^{\prime\prime}\not\in\iota(X)$ and $\gamma(x^{\prime\prime})\leq^{+}a$, as required. $~{}\Box$ ###### Lemma 8.7 Let $T,T_{1},T_{2}$ be positive face structures, $dim(T_{1}),dim(T_{2})>k$, such that ${\bf c}^{(k)}(T_{1})={\bf d}^{(k)}(T_{2})$ and $T=T_{1}+_{k}T_{2}$. Then ${\bf c}^{(k)}(T_{1})_{k}\cap\gamma(T_{1})\neq\emptyset$. For any $a\in{\bf c}^{(k)}(T_{1})_{k}\cap\gamma(T_{1})$ we have $a\in Sd(T)_{k}$ and either $T_{1}=T^{\downarrow a}$ and $T_{2}=T^{\uparrow a}$ or $a\in Sd(T_{1})_{k}$, $T^{\downarrow a}=T_{1}^{\downarrow a}$ and $T^{\uparrow a}=T_{1}^{\uparrow a}+_{k}T_{2}$. Proof. By assumption $(T_{1})_{k+1}\neq\emptyset$ and $(T_{2})_{k+1}\neq\emptyset$. So ${\bf c}^{(k)}(T_{1})\cap\gamma(T_{1})\neq\emptyset$. Fix $a\in{\bf c}^{(k)}(T_{1})\cap\gamma(T_{1})\neq\emptyset$. Then $T^{\downarrow a}_{k+1}\neq\emptyset$. As $T^{\downarrow a}_{k+1}\cap(T_{2})_{k+1}=\emptyset$ we must have $a\in Sd(T)_{k}$. Assume that $T_{1}\neq T^{\downarrow a}$. Then $T^{\downarrow a}\subseteq\hskip-11.38109pt_{\not}\hskip 8.53581ptT_{1}$. Hence $(T_{1})-(T^{\downarrow a})\neq\emptyset$. But this means that $a\in Sd(T_{1})_{k}$. The verification that the equalities $T^{\downarrow a}=T_{1}^{\downarrow a}$ and $T^{\uparrow a}=T_{1}^{\uparrow a}+_{k}T_{2}$ hold in this case is left as an exercise. $~{}\Box$ ## 9 $S^{*}$ is a positive-to-one computad ###### Proposition 9.1 Let $S$ be a weak positive face structure. Then $S^{*}$ is a positive-to-one computad. Proof. The proof is by induction on dimension $n$ of the weak positive face structure $S$. For $n=0,1$ the Proposition is obvious. So assume that for any weak positive face structure $T$ of dimension $n$, $T^{*}$ is a positive-to-one computad of dimension $n$, generated by faces in $T$. Suppose that $S$ is a weak positive face structure of dimension $n+1$. We shall show that $S^{*}$ is a computad generated by faces in $S$. Since $S_{\leq n}$ is a weak positive face structure, by inductive assumption $S_{\leq n}^{*}$ is the computad generated by faces in $S_{\leq n}$. So we need to verify that for any ${\omega}$-functor $f:S_{\leq n}^{*}\longrightarrow C$ to any ${\omega}$-category $C$, and any function $|f|:S_{n+1}\longrightarrow C_{n+1}$ such, that for $a\in S_{n+1}$ $d_{C}(|f|(a))=f({\bf d}([a])),\;\;\;\;\;c_{C}(|f|(a))=f({\bf c}([a])),$ there is a unique ${\omega}$-functor $F:S^{*}\longrightarrow C$, such that $F_{n+1}([a])=|f|(a),\;\;\;\;\;F_{\leq n}=f$ as in the diagram $S_{\leq n}\vphantom{S_{\leq n}^{*}}$$\vphantom{S_{\leq n}}S_{\leq n}^{*}$$[-]$$S_{n+1}\vphantom{S^{*}}$$\vphantom{S_{n+1}}S^{*}$$\delta$$\gamma$${\bf d}$${\bf c}$$F$$f$$C$$[-]$$|f|$ We define $F_{n+1}$ as follows. For $X\in S^{*}_{n+1}$ $F_{n+1}(X)=\left\\{\begin{array}[]{ll}id_{f(X)}&\mbox{ if $dim(X)\leq n$,}\\\ |f|(a)&\mbox{ if $dim(X)=n+1$, $X$ is principal and $X=[a]$,}\\\ F_{n+1}(X^{\downarrow a});_{l}F_{n+1}(X^{\uparrow a})&\mbox{ if $dim(X)=n+1$, $a\in Sd(X)_{l}$.}\end{array}\right.$ $;_{l}$ refers to the composition in $C$. Clearly $F_{k}=f_{k}$ for $k\leq n$. The above morphism, if well defined, clearly preserves identities. We need to verify, for $X\in S^{*}_{n+1}$ and $dim(X)=n+1$, three conditions: * I $F$ is well defined, i.e. $F_{n+1}(X)=F_{n+1}(X^{\downarrow a});_{l}F_{n+1}(X^{\uparrow a})$ does not depend on the choice of the saddle face $a\in Sd(X)$; * II $F$ preserves the domains and codomains i.e., $F_{n}(d(X))=d(F_{n+1}(X))$ and $F_{n}(c(X))=c(F_{n+1}(X))$; * III $F$ preserves compositions i.e., $F_{n+1}(X)=F_{n+1}(X_{1});_{k}F_{n+1}(X_{2})$ whenever $X=X_{1}+_{k}X_{2}$ and $dim(X_{1}),dim(X_{2})>k$. We have an embedding $[-]:S_{\leq n}\longrightarrow S_{\leq n}^{*}$. So let assume that for positive face substructures of $S$ of size less than $size(X)$ the above assumption holds. If $size(X)_{n+1}=0$ or $X$ is principal all three conditions are obvious. So assume that $X$ is not principal and $dim(X)=n+1$. To save on notation we write $F$ for $F_{n+1}$. Ad I. First we will consider two saddle faces $a,x\in Sd(X)$ of different dimension $k=dim(x)<dim(a)=m$. Using Lemma 8.4 we have $\displaystyle F(X^{\downarrow a});_{m}F(X^{\uparrow a})=$ $\displaystyle ind.\;hyp.\;I$ $\displaystyle=(F(X^{\downarrow a\downarrow x});_{k}F(X^{\downarrow a\uparrow x}));_{m}(F(X^{\uparrow a\downarrow x});_{k}F(X^{\uparrow a\uparrow x}))=$ $\displaystyle MEL$ $\displaystyle=(F(X^{\downarrow a\downarrow x});_{m}F(X^{\uparrow a\downarrow x}));_{k}(F(X^{\downarrow a\uparrow x});_{m}F(X^{\uparrow a\uparrow x}))=$ $\displaystyle=(F(X^{\downarrow x\downarrow a});_{m}F(X^{\downarrow x\uparrow a}));_{k}(F(X^{\uparrow x\downarrow a});_{m}F(X^{\uparrow x\uparrow a}))=$ $\displaystyle ind.\;hyp.\;III$ $\displaystyle=F(X^{\downarrow x});_{m}F(X^{\uparrow x})$ Now we will consider two saddle faces $a,b\in Sd(X)$ of the same dimension $dim(a)=dim(b)=m$. We shall use Lemma 8.5. Assume that $a<^{-}_{l}b$, for some $l<m$. If $\gamma^{(k)}(a)\in Sd(X)$, for some $k<m$, then this case reduces to the previous one for two pairs $a,\gamma^{(k)}(a)\in Sd(X)$ and $b,\gamma^{(k)}(a)\in Sd(X)$. Otherwise $a\in Sd(X^{\uparrow b})$ and $a\in Sd(X^{\uparrow b})$ and we have $\displaystyle F(X^{\downarrow a});_{k}F(X^{\uparrow a})=$ $\displaystyle ind.\;hyp\;I$ $\displaystyle=F(X^{\downarrow a});_{k}(F(X^{\uparrow a\downarrow b});_{k}F(X^{\uparrow a\uparrow b}))=$ $\displaystyle=(F(X^{\downarrow a});_{k}F(X^{\uparrow a\downarrow b}));_{k}F(X^{\uparrow b\uparrow a})=$ $\displaystyle ind\;hyp\;III$ $\displaystyle=F(X^{\downarrow a};_{k}X^{\uparrow a\downarrow b});_{k}F(X^{\uparrow b\uparrow a})=$ $\displaystyle=F(X^{\downarrow b};_{k}X^{\uparrow b\downarrow a});_{k}F(X^{\uparrow b\uparrow a})=$ $\displaystyle ind\;hyp\;III$ $\displaystyle=(F(X^{\downarrow b});_{k}F(X^{\uparrow b\downarrow a}));_{k}F(X^{\uparrow b\uparrow a})=$ $\displaystyle=F(X^{\downarrow b});_{k}(F(X^{\uparrow b\downarrow a});_{k}F(X^{\uparrow b\uparrow a}))=$ $\displaystyle ind\;hyp\;I$ $\displaystyle=F(X^{\downarrow b});_{k}F(X^{\uparrow b})$ Finally, we consider the case $a<^{+}b$. We have $\displaystyle F(X^{\downarrow a});_{k}F(X^{\uparrow a})=$ $\displaystyle ind.\;hyp\;I$ $\displaystyle=F(X^{\downarrow a});_{k}(F(X^{\uparrow a\downarrow b});_{k}F(X^{\uparrow a\uparrow b}))=$ $\displaystyle=F(X^{\downarrow b\downarrow a});_{k}(F(X^{\downarrow b\uparrow a}));_{k}F(X^{\uparrow b})=$ $\displaystyle=(F(X^{\downarrow b\downarrow a});_{k}F(X^{\downarrow b\uparrow a}));_{k}F(X^{\uparrow b})=$ $\displaystyle ind\;hyp\;I$ $\displaystyle=F(X^{\downarrow b});_{k}F(X^{\uparrow b})$ This shows that $F(X)$ is well defined. Ad II. We shall show that the domains are preserved. The proof that, the codomains are preserved, is similar. The fact that if $Sd(X)=\emptyset$ then $F$ preserves domains and codomains follows immediately from the assumption on $f$ and $|f|$. So assume that $Sd(X)\neq\emptyset$ and let $a\in Sd(X)$, $dim(a)=k$. We use Lemma 8.6. We have to consider two cases $k<n$, and $k=n$. If $k<n$ then $\displaystyle F_{n}(d(X))=F_{n}(d(X^{\downarrow a}+_{k}X^{\uparrow a}))=$ $\displaystyle=F_{n}(d(X^{\downarrow a});_{k}d((X^{\uparrow a}))=$ $\displaystyle=F_{n}(d(X)^{\downarrow a};_{k}d(X)^{\uparrow a})=$ $\displaystyle ind\;hyp\;III$ $\displaystyle=F_{n}(d(X)^{\downarrow a});_{k}F_{n}(d(X)^{\uparrow a})=$ $\displaystyle=F_{n}(d(X^{\downarrow a}));_{k}F_{n}(d(X^{\uparrow a}))=$ $\displaystyle ind\;hyp\;II$ $\displaystyle=d(F_{n+1}(X^{\downarrow a}));_{k}d(F_{n+1}(X^{\uparrow a}))=$ $\displaystyle=d(F_{n+1}(X^{\downarrow a});_{k}F_{n+1}(X^{\uparrow a}))=$ $\displaystyle ind\;hyp\;I$ $\displaystyle=d(F_{n+1}(X))$ If $k=n$ then $\displaystyle F_{n}(d(X))=F_{n}(d(X^{\downarrow a}+_{n}X^{\uparrow a}))=$ $\displaystyle=F_{n}(d(X^{\downarrow a}))=$ $\displaystyle ind\;hyp\;II$ $\displaystyle=d(F_{n+1}(X^{\downarrow a}))=$ $\displaystyle=d(F_{n+1}(X^{\downarrow a});_{n}F_{n+1}(X^{\uparrow a}))=$ $\displaystyle ind\;hyp\;I$ $\displaystyle=d(F_{n+1}(X))$ Ad III. Suppose that $X=X_{1};_{k}X_{2}$ and $dim(X)\leq n+1$. We shall show that $F$ preserves this composition. If $dim(X_{1})=k$ then $X=X_{2}$, $X_{1}={\bf d}^{(k)}(X_{2})$. We have $F_{n+1}(X)=F_{n+1}(X_{2})=1^{(n+1)}_{F_{k}({\bf d}^{(k)}(X_{2}))};_{k}F_{n+1}(X_{2})=$ $=1^{(n+1)}_{F_{k}(X_{1})};_{k}F_{n+1}(X_{2})=F_{n+1}(X_{1});_{k}F_{n+1}(X_{2})$ The case $dim(X_{2})=k$ is similar. So now assume that $dim(X_{1}),dim(X_{2})>k$. We shall use Lemma 8.7. Fix $a\in{\bf c}^{(k)}(X_{1})_{k}\cap\gamma(X_{1})$. So $a\in Sd(X)_{k}$. If $X_{1}=X^{\downarrow a}$ and $X_{2}=X^{\uparrow a}$ then we have $F(X)=F(X^{\downarrow a});_{k}F(X^{\uparrow a})=F(X_{1});_{k}F(X_{2}).$ If $a\in Sd(X_{1})_{k}$ then $\displaystyle F(X)=F(X^{\downarrow a});_{k}F(X^{\uparrow a})=$ $\displaystyle ind\;hyp\;II$ $\displaystyle=F(X^{\downarrow a});_{k}(F(X_{1}^{\uparrow a});_{k}F(X_{2}))=$ $\displaystyle=(F(X_{1}^{\downarrow a});_{k}F(X_{1}^{\uparrow a}));_{k}F(X_{2})=$ $\displaystyle ind\;hyp\;II$ $\displaystyle F(X_{1});_{k}F(X_{2})$ So in any case the composition is preserved. This ends the proof of the Lemma. $~{}~{}\Box$ For $n\in{\omega}$, we have a functor $(-)^{\sharp,n}:{\bf wFs}^{+/1}_{n}\longrightarrow{\bf Comma}^{+/1}_{n}$ such that, for $S$ in ${\bf wFs}^{+/1}_{n}$ $S^{\sharp,n}=(S_{n},S^{*}_{<n},[\delta],[\gamma])$ and for $f:S\rightarrow T$ in ${\bf wFs}^{+/1}_{n}$ we have $f^{\natural,n}=(f_{n},(f_{<n})^{*}).$ ###### Corollary 9.2 For every $n\in{\omega}$, the functor $(-)^{\sharp,n}$ is well defined, full, faithful, and it preserves existing pushouts. Moreover, for $S$ in ${\bf wFs}^{+/1}_{n}$ we have $S^{*}=\overline{S^{\sharp,n}}^{n}$. Proof. The functor $\overline{(-)}^{n}:{\bf Comma}^{+/1}_{n}\longrightarrow{{\bf Comp}^{+/1}_{n}}$ which is an equivalence of categories is described in the Appendix. Fullness and faithfulness of $(-)^{\sharp,n}$ is left for the reader. We shall show simultaneously that for every $n\in{\omega}$, both functors $(-)^{\sharp,n}:{\bf wFs}^{+/1}_{n}\longrightarrow{\bf Comma}^{+/1}_{n},\hskip 28.45274pt(-)^{*,n}:{\bf wFs}^{+/1}_{n}\longrightarrow{{\bf Comp}^{+/1}_{n}}$ preserve existing pushouts. For $n=0$ there is nothing to prove. For $n=1$ this is obvious. So assume that $n>1$ and that $(-)^{\sharp,n}$ preserves existing pushouts. Let $R\vphantom{T}$$\vphantom{R}T$$S\vphantom{S+_{R}T}$$\vphantom{S}S+_{R}T$ be a pushout in ${\bf wFs}^{+/1}_{n+1}$. Clearly its $n$-truncation is a pushout in ${\bf wFs}^{+/1}_{n}$. Hence by inductive hypothesis it is preserved by $(-)^{*,n}$. In dimension $n+1$, the functor $(-)^{\sharp,n+1}$ is an inclusion. Hence, in dimension $n+1$, this square is a pushout (of monos) in $Set$. So the whole square $R^{\sharp,n+1}\vphantom{T^{\sharp,n+1}}$$\vphantom{R^{\sharp,n+1}}T^{\sharp,n+1}$$S^{\sharp,n+1}\vphantom{(S+_{R}T)^{\sharp,n+1}}$$\vphantom{S^{\sharp,n+1}}(S+_{R}T)^{\sharp,n+1}$ is a pushout in ${\bf Comma}_{n+1}$, i.e. $(-)^{\sharp,n+1}$ preserves pushouts. As $(-)^{*,n+1}$ is a composition of $(-)^{\sharp,n+1}$ with an equivalence of categories it preserves the pushouts, as well. $~{}~{}\Box$ ###### Corollary 9.3 The functor $(-)^{*}:{\bf wFs}^{+/1}\longrightarrow{\bf Comp}^{+/1}$ is full and faithful and preserves special pushouts. Proof. This follows from the previous Corollary and the fact that the functor $\overline{(-)}^{n}:{\bf Comma}^{+/1}_{n}\longrightarrow{{\bf Comp}^{+/1}_{n}}$ (see Appendix) is an equivalence of categories. $~{}~{}\Box$ Let $P$ be a positive-to-one computad, $a$ a $k$-cell in $P$. A description of the cell $a$ is a pair $<T_{a},\tau_{a}:T_{a}^{*}\longrightarrow P>$ where $T_{a}$ is a positive face structure and $\tau_{a}$ is a computad map such that $\tau_{a}(T_{a})=a.$ In the remainder of this section we shall define some specific positive face structures that will be used later. First we define $\alpha^{n}$, for $n\in{\omega}$. We put $\alpha^{n}_{l}\;\;=\;\;\left\\{\begin{array}[]{ll}\emptyset&\mbox{ if $l>n$}\\\ \\{2n\\}&\mbox{ if $l=n$}\\\ \\{2l+1,\,2l\\}&\mbox{ if $0\leq l<n$}\end{array}\right.$ $d,\,c:\alpha^{n}_{l}\longrightarrow\alpha^{n}_{l-1}$ $d(x)=\\{2l-1\\}\;\;\;\;\;\;\;\;\;c(x)=2l-2$ for $x\in\alpha^{n}_{l}$, and $1\leq l\leq n$. For example $\alpha^{4}$ can be pictured as follows: ${\bf 1}$${\bf 0}$${\bf 3}$${\bf 2}$${\bf 5}$${\bf 4}$${\bf 7}$${\bf 6}$${\bf 8}$ i.e. $8$ is the unique cell of dimension $4$ in $\alpha^{4}$ that has $7$ as its domain and $6$ as its codomain, $7$ and $6$ have $5$ as its domain and $4$ as its codomain, and so on. Note that, for any $k\leq n$, we have ${\bf d}^{(k)}\alpha^{n}=\alpha^{k}={\bf c}^{(k)}\alpha^{n}.$ Let $n_{1}<n_{0},n_{2}$ and $n_{3}<n_{2},n_{4}$. We define the positive face structures $\alpha^{n_{0},n_{1},n_{2}}$ and $\alpha^{n_{0},n_{1},n_{2},n_{3},n_{4}}$ as the following colimits in ${\bf Fs}^{+/1}$: $\alpha^{n_{1}}\vphantom{\alpha^{n_{2}}}$$\vphantom{\alpha^{n_{1}}}\alpha^{n_{2}}$${\bf d}^{(n_{1})}_{\alpha^{n_{2}}}$$\alpha^{n_{0}},\vphantom{\alpha^{n_{0},n_{1},n_{2}}}$$\vphantom{\alpha^{n_{0}},}\alpha^{n_{0},n_{1},n_{2}}$$\kappa_{1}$${\bf c}^{(n_{1})}_{\alpha^{n_{0}}}$$\kappa_{2}$$\alpha^{n_{1}}\vphantom{\alpha^{n_{2}}}$$\vphantom{\alpha^{n_{1}}}\alpha^{n_{2}}$${\bf d}^{(n_{1})}_{\alpha^{n_{2}}}$$\alpha^{n_{0}},\vphantom{\alpha^{n_{0},n_{1},n_{2},n_{3},n_{4}}}$$\vphantom{\alpha^{n_{0}},}\alpha^{n_{0},n_{1},n_{2},n_{3},n_{4}}$$\kappa_{1}$${\bf c}^{(n_{1})}_{\alpha^{n_{0}}}$$\kappa_{2}$$\vphantom{\phantom{\alpha^{n_{2}}}}\alpha^{n_{3}}$${\bf c}^{(n_{3})}_{\alpha^{n_{2}}}$$\vphantom{\phantom{\alpha^{n_{0},n_{1},n_{2},n_{3},n_{4}}}}\alpha^{n_{4}}$$\kappa_{3}$${\bf d}^{(n_{3})}_{\alpha^{n_{4}}}$ We have ###### Proposition 9.4 The above colimits are preserved by the functor $(-)^{*}:{\bf Fs}^{+/1}\longrightarrow{\bf Comp}^{+/1}.$ Moreover for any ${\omega}$-category $C$ we have bijective correspondences ${{\omega}Cat}((\alpha^{n})^{*},C)=C_{n}$ ${{\omega}Cat}((\alpha^{n_{0},n_{1},n_{2}})^{*},C)=\\{(x,y)\in C_{n_{0}}\times C_{n_{2}}:c^{(n_{1})}(x)=d^{(n_{1})}(y)\\}$ ${{\omega}Cat}((\alpha^{n_{0},n_{1},n_{2},n_{3},n_{4}})^{*},C)=\hskip 227.62204pt$ $=\\{(x,y,z)\in C_{n_{0}}\times C_{n_{2}}\times C_{n_{4}}:c^{(n_{1})}(x)=d^{(n_{1})}(y)\;\;{\rm and}\;\;c^{(n_{3})}(y)=d^{(n_{3})}(z)\\}$ which are natural in $C$. Proof. As both positive face structures $\alpha^{n_{0},n_{1},n_{2}}$ and $\alpha^{n_{0},n_{1},n_{2},n_{3},n_{4}}$ are obtained vis special pushout (in the second case applied twice) these colimits are preserved by $(-)^{*}$. $~{}~{}\Box$ The essential image of the full and faithful functor $(-)^{*}:{\bf Fs}^{+/1}\longrightarrow{\bf Comp}^{+/1}$ is the category of the positive computypes and it will be denoted by ${{\cal C}types}^{+/1}$. Thus the categories ${\bf Fs}^{+/1}$ and ${{\cal C}types}^{+/1}$ are equivalent. The full image of the functor $(-)^{*}:{\bf Fs}^{+/1}\longrightarrow{{\omega}Cat}$ will be denoted by ${{\cal C}types}^{+/1}_{\omega}$. The objects of ${{\cal C}types}^{+/1}_{\omega}$ are ${\omega}$-categories isomorphic to those of form $S^{*}$ for $S$ being positive face structure and the morphism in ${{\cal C}types}^{+/1}_{\omega}$ are all ${\omega}$-functors. ## 10 The inner-outer factorization in ${{\cal C}types}^{+/1}_{\omega}$ Let $f:S^{*}\longrightarrow T^{*}$ be a morphism in ${{\cal C}types}^{+/1}_{\omega}$. We say that $f$ is outer111The names ’inner’ and ’outer’ are introduced in analogy with the morphism with the same name and role in the category of disks in [J]. if there is a map of face structures $g:S\longrightarrow T$ such that $g^{*}=f$. We say that $f$ is inner iff $f_{dim(S)}(S)=T$. From Corollary 9.3 we have ###### Lemma 10.1 An ${\omega}$-functor $f:S^{*}\longrightarrow T^{*}$ is outer iff it is a computad map. $~{}~{}\Box$ ###### Proposition 10.2 Let $f:S^{*}\longrightarrow T^{*}$ be an inner map, $dim(S)=dim(T)>0$. The maps ${\bf d}f:{\bf d}S\longrightarrow{\bf d}T$ and ${\bf c}f:{\bf c}S\longrightarrow{\bf c}T$, being the restrictions of $f$, are well defined, inner and the squares $({\bf d}S)^{*}\vphantom{S^{*}}$$\vphantom{({\bf d}S)^{*}}S^{*}$${\bf d}_{S}^{*}$$({\bf d}T)^{*}\vphantom{T^{*}}$$\vphantom{({\bf d}T)^{*}}T^{*}$${\bf d}_{T}^{*}$${\bf d}f$$f$ $\vphantom{\phantom{S^{*}}}({\bf c}S)^{*}$${\bf c}_{S}^{*}$ $\vphantom{\phantom{T^{*}}}({\bf c}T)^{*}$${\bf c}_{T}^{*}$${\bf c}f$ commute. Proof. So suppose that $f:S^{*}\rightarrow T^{*}$ is an inner map. So $f(S)=T$. Since $f$ is an ${\omega}$-functor we have $f({\bf d}S)={\bf d}f(S)={\bf d}T\;\;\;\;{\rm and}\;\;\;\;f({\bf c}S)={\bf c}f(S)={\bf c}T.$ This shows the proposition. $~{}~{}\Box$ We have ###### Proposition 10.3 The inner and outer morphisms form a factorization system in ${{\cal C}types}^{+/1}_{\omega}$. So any ${\omega}$-functor $f:S^{*}\longrightarrow T^{*}$ can be factored essentially uniquely by inner map $\stackrel{{\scriptstyle\bullet}}{{f}}$ followed by outer map $\stackrel{{\scriptstyle\circ}}{{f}}$: $S^{*}\vphantom{T^{*}}$$\vphantom{S^{*}}T^{*}$$f$$f(S)^{*}$$\stackrel{{\scriptstyle\bullet}}{{f}}$$\stackrel{{\scriptstyle\circ}}{{f}}$ Proof. This is almost tautological. $~{}~{}\Box$ ## 11 The terminal positive-to-one computad In this section we shall describe the terminal positive-to-one computad ${\cal T}$. The set of $n$-cell ${\cal T}_{n}$ consists of (isomorphisms classes of) positive face structures of dimension less than or equal to $n$. For $n>0$, the operations of domain and codomain $d^{\cal T},c^{\cal T}:{\cal T}_{n}\rightarrow{\cal T}_{n-1}$ are given, for $S\in{\cal T}_{n}$ by $d(S)=\left\\{\begin{array}[]{ll}S&\mbox{if $dim(S)<n$,}\\\ {\bf d}S&\mbox{if $dim(S)=n$,}\end{array}\right.$ and $c(S)=\left\\{\begin{array}[]{ll}S&\mbox{if $dim(S)<n$,}\\\ {\bf c}S&\mbox{if $dim(S)=n$.}\end{array}\right.$ and, for $S,S^{\prime}\in{\cal T}_{n}$ such that $c^{(k)}(S)=d^{(k)}(S^{\prime})$ the composition in ${\cal T}$ is just the pushout $S;_{k}S^{\prime}=S+_{k}S^{\prime}$ i.e. ${\bf c}^{(k)}S\vphantom{S^{\prime}}$$\vphantom{{\bf c}^{(k)}S}S^{\prime}$$S\vphantom{S+_{k}S^{\prime}}$$\vphantom{S}S+_{k}S^{\prime}$ The identity $id_{\cal T}:T_{n-1}\rightarrow{\cal T}_{n}$ is the inclusion map. The $n$-indets in ${\cal T}$ are the principal positive $n$-face structures. ###### Proposition 11.1 ${\cal T}$ just described is the terminal positive-to-one computad. Proof. The fact that ${\cal T}$ is an ${\omega}$-category is easy. The fact that ${\cal T}$ is free with free $n$-generators being principal $n$-face structures can be shown much like the freeness of $S^{*}$ before. The fact that ${\cal T}$ is terminal follows from the following observation: Observation. For every pair of parallel $n$-face structures $N$ and $B$ (i.e. ${\bf d}N={\bf d}B$ and ${\bf c}N={\bf c}B$) such that $N$ is normal and $B$ is principal there is a unique (up to an iso) principal $(n+1)$-face structure $N^{\bullet}$ such that ${\bf d}N^{\bullet}=N$ and ${\bf c}N^{\bullet}=B$. $~{}~{}\Box$ ###### Lemma 11.2 Let $S$ be a positive face structure and $!:S^{*}\longrightarrow{\cal T}$ the unique map from $S^{*}$ to ${\cal T}$. Then, for $T\in S^{*}_{k}$ we have $!_{k}(T)=T.$ Proof. The proof is by induction on $k\in{\omega}$ and the size of $T$ in $S^{*}_{k}$. For $k=0,1$ the lemma is obvious. Let $k>1$ and assume that lemma holds for $i<k$. If $dim(T)=l<k$ then, using the inductive hypothesis and the fact that $!$ is an ${\omega}$-functor, we have $!_{k}(T)=!_{k}(1^{(k)}_{T})=1^{(k)}_{!_{l}(T)}=1^{(k)}_{T}=T$ Suppose that $dim(T)=k$ and $T$ is principal. As $!$ is a computad map $!_{k}(T)$ is an indet, i.e. it is principal, as well. We have, using again the inductive hypothesis and the fact that $!$ is an ${\omega}$-functor, $d(!_{k}(T))=!_{k-1}({\bf d}T)={\bf d}T$ $c(!_{k}(T))=!_{k-1}({\bf c}T)={\bf c}T$ As $T$ is the only (up to a unique iso) positive face structure with the domain ${\bf d}T$ and the codomain ${\bf c}T$, it follows that $!_{k}(T)=T$, as required. Finally, suppose that $dim(T)=k$, $T$ is not principal, and for the positive face structures of size smaller than the size of $T$ the lemma holds. Thus there are $l\in{\omega}$ and $a\in Sd(T)_{l}$ so that $!_{k}(T)=!_{k}(T^{\downarrow a};_{l}T^{\uparrow a})=!_{k}(T^{\downarrow a});_{l}!_{k}(T^{\uparrow a})=T^{\downarrow a};_{l}T^{\uparrow a}=T,$ as required $~{}~{}\Box$ ## 12 A description of the positive-to-one computads In this section we shall describe all the cells in positive-to-one computads using positive face structures, in other words we shall describe in concrete terms the functor: $\overline{(-)}:{\bf Comma}^{+/1}_{n}\longrightarrow{{\bf Comp}^{+/1}_{n}}$ More precisely, the positive-to-one computads of dimension 1 (and all computads as well) are free computads over graphs and are well understood. So suppose that $n>1$, and we are given an object of ${\bf Comma}^{+/1}_{n}$, i.e. a quadruple $(|P|_{n},P,d,c)$ such that 1. 1. a positive-to-one $(n-1)$-computad $P$; 2. 2. a set $|P|_{n}$ with two functions $c:|P|_{n}\longrightarrow|P|_{n-1}$ and $d:|P|_{n}\longrightarrow P_{n-1}$ such that for $x\in|P|_{n}$, $cc(x)=cd(x)$ and $dc(x)=dd(x)$; we assume that $d(x)$ is not an identity for any $x\in|P|_{n}$. If the maps $d$ and $c$ in the object $(|P|_{n},P,d,c)$ are understood from the context we can abbreviate notation to $(|P|_{n},P)$. For a positive face structure $S$, with $dim(S)\leq n$, we denote by $S^{\sharp,n}$ the object $(S_{n},(S_{<n})^{*},[\delta],[\gamma])$ in ${\bf Comma}^{+/1}_{n}$. In fact, we have an obvious functor $(-)^{\sharp,n}:{\bf Fs}^{+/1}\longrightarrow{\bf Comma}^{+/1}_{n}$ such that $S\mapsto(S_{n},(S_{<n})^{*},[\delta],[\gamma])$ Any positive-to-one computad $P$ can be restricted to its part in ${\bf Comma}^{+/1}_{n}$. So we have an obvious forgetful functor $(-)^{\natural,n}:{\bf Comp}^{+/1}\longrightarrow{\bf Comma}^{+/1}_{n}$ such that $P\mapsto(|P|_{n},P_{<n},d,c)$ We shall describe the positive-to-one $n$-computad $\overline{P}$ whose $(n-1)$-truncation is $P$ and whose $n$-indets are $|P|_{n}$ with the domains and codomains given by $c$ and $d$. n-cells of $\overline{P}$. An $n$-cell in $\overline{P}_{n}$ is a(n equivalence class of) pair(s) $(S,f)$ where 1. 1. $S$ is a positive face structure, $dim(S)\leq n$; 2. 2. $f:(S_{n},(S_{<n})^{*},[\delta],[\gamma])\longrightarrow(|P|_{n},P,d,c)$ is a morphism in ${\bf Comma}^{+/1}_{n}$, i.e. $S^{*}_{n-1}\vphantom{P_{n-1}}$$\vphantom{S^{*}_{n-1}}P_{n-1}$$f_{n-1}$$S_{n}\vphantom{|P|_{n}}$$\vphantom{S_{n}}|P|_{n}$$|f|_{n}$${[}\delta{]}$${[}\gamma{]}$$d$$c$ commutes. We identify two pairs $(S,f)$, $(S^{\prime},f^{\prime})$ if there is an isomorphism $h:S\longrightarrow S^{\prime}$ such that the triangles of sets and of $(n-1)$-computads $S_{n}\vphantom{S^{\prime}_{n}}$$\vphantom{S_{n}}S^{\prime}_{n}$$h_{n}$$|P|_{n}$$f_{n}$$f^{\prime}_{n}$$(S_{<n})^{*}\vphantom{(S^{\prime}_{<n})^{*}}$$\vphantom{(S_{<n})^{*}}(S^{\prime}_{<n})^{*}$$(h_{<n})^{*}$$P$$f_{<n}$$f^{\prime}_{<n}$ commute. Clearly, such an $h$, if exists, is unique. Even if formally cells in $P_{n}$ are equivalence classes of triples we will work on triples themselves as if they were cells understanding that equality between such cells is an isomorphism in the sense defined above. Domains and codomains. The domain and codomain functions $d^{(k)},c^{(k)}:\overline{P}_{n}\longrightarrow\overline{P}_{k}$ are defined for an $n$-cell $(S,f)$ as follows: $d^{(k)}(S,f)=({\bf d}^{(k)}S,{\bf d}^{(k)}f)$ where, for $x\in({\bf d}^{(k)}S)_{k}$ $({\bf d}^{(k)}f)_{k}(x)=f_{k}([x])(x)$ (i.e. we take the sub-face structure $[x]$ of $S$, then value of $f$ on it, and then we evaluate the map in ${\bf Comma}^{+/1}_{n}$ on $x$ the only element of $[x]_{k}$), $({\bf d}^{(k)}f)_{l}=f_{l}$ for $l<k$; $c^{(k)}(S,f)=({\bf c}^{(k)}S,{\bf c}^{(k)}f)$ where, for $x\in({\bf c}^{(k)}S)_{k}$ $({\bf c}^{(k)}f)_{k}(x)=f_{k}([x])(x)$ and $({\bf d}^{(k)}f)_{l}=f_{l}$ for $l<k$, i.e. we calculate the $k$-th domain and $k$-th codomain of an $n$-cell $(S,f)$ by taking ${\bf d}^{(k)}$ and ${\bf c}^{(k)}$ of the domain $S$ of the cell $f$, respectively, and by restricting the maps $f$ accordingly. Identities. The identity function ${\bf i}:\overline{P}_{n-1}\longrightarrow\overline{P}_{n}$ is defined for an $(n-1)$-cell $((S,f)$ in $P_{n-1}$, as follows: ${\bf i}(S,f)=\left\\{\begin{array}[]{ll}(S,f)&\mbox{ if $dim(S)<n-1$,}\\\ (S,\overline{f})&\mbox{ if dim(S)=n-1}\end{array}\right.$ Note that $\overline{f}$ is the map ${{\bf Comp}^{+/1}_{n-1}}$ which is the value of the functor $\overline{(-)}$ on a map $f$ from ${\bf Comma}^{+/1}_{n-1}$. So it is in fact defined as ’the same $(n-1)$-cell’ but considered as an $n$-cell. Compositions. Suppose that $(S^{i},f^{i})$ are $n$-cells for $i=0,1$, such that $c^{(k)}(S^{0},f^{0})=d^{(k)}(S^{1},f^{1}).$ Then their composition is defined, via pushout in ${\bf Comma}^{+/1}_{n}$, as $(S^{0},f^{0});_{k}(S^{1},f^{1})=(S^{0}+_{k}S^{1},[f^{0},f^{1}])$ i.e. $((S^{0}+_{k}S^{1})_{\leq n-1})^{*}_{n-1}\vphantom{P_{n-1}}$$\vphantom{((S^{0}+_{k}S^{1})_{\leq n-1})^{*}_{n-1}}P_{n-1}$$[f^{0}_{n-1},f^{1}_{n-1}]$$S^{0}_{n}\sqcup\\!\\!\\!\\!\cdot\;S^{1}_{n}\vphantom{|P|_{n}}$$\vphantom{S^{0}_{n}\sqcup\\!\\!\\!\\!\cdot\;S^{1}_{n}}|P|_{n}$$[f^{0}_{n},f^{1}_{n}]$${[}\delta{]}$${[}\gamma{]}$$d$$c$ This ends the description of the computad $\overline{P}$. Now let $h:P\rightarrow Q$ be a morphism in ${\bf Comma}^{+/1}_{n}$, i.e. a function $h_{n}:|P|_{n}\longrightarrow|Q|_{n}$ and a $(n-1)$-computad morphism $h_{<n}:P_{<n}\longrightarrow Q_{<n}$ such that the square $P_{n-1}\vphantom{Q_{n-1}}$$\vphantom{P_{n-1}}Q_{n-1}$$h_{n-1}$$|P|_{n}\vphantom{|Q|_{n}}$$\vphantom{|P|_{n}}|Q|_{n}$$h_{n}$$d$$c$$d$$c$ commutes serially. We define $\bar{h}:\bar{P}\longrightarrow\bar{Q}$ by putting $\bar{h}_{k}=h_{k}$ for $k<n$, and for $(S,f)\in\bar{P}_{n}$, we put $\bar{h}(S,f)=(S,h\circ f).$ Notation. Let $x=(S,f)$ be a cell in $\bar{P}_{n}$ as above, and $a\in Sd(S)$. Then by $x^{\downarrow a}=(S^{\downarrow a},f^{\downarrow a})$ and $x^{\uparrow a}=(S^{\uparrow a},f^{\uparrow a})$ we denote the cells in $\bar{P}_{n}$ that are the obvious restriction $x$. Clearly, we have $c^{(k)}(x^{\downarrow a})=d^{(k)}((x^{\uparrow a})$ and that $x=x^{\downarrow a};_{k}x^{\uparrow a}$, where $k=dim(a)$. The following Proposition states several statements concerning the above construction. This includes that the above construction is correct. We have put all these statement together as we need to prove them together, that is by simultaneous induction. ###### Proposition 12.1 Let $n\in{\omega}$. We have 1. 1. Let $P$ be an object of ${\bf Comma}^{+/1}_{n}$. We define the function $\eta_{P}:|P|_{n}\longrightarrow\overline{P}_{n}$ as follows. Let $x\in|P|_{n}$. As $c(x)$ is an indet $d(x)$ is a normal cell of dimension $n-1$. Thus there is a unique description of the cell $d(x)$ $<T_{d(x)},\tau_{d(x)}:T_{d(x)}^{*}\longrightarrow P_{<n}>$ with $T_{d(x)}$ being normal positive face structure. Then we have a unique $n$-cell in $\bar{P}$: $\bar{x}=<T_{d(x)}^{\bullet},\;\;|\overline{\tau}_{x}|_{n}:\\{T_{d(x)}^{\bullet}\\}\rightarrow|P|_{n},\;\;(\overline{\tau}_{x})_{<n}:(T_{d(x)}^{\bullet})_{<n}^{*}\rightarrow P_{<n}>$ (note: $|T_{d(x)}^{\bullet}|_{n}=\\{T_{d(x)}^{\bullet}\\}$) such that $|\overline{\tau}_{x}|_{n}(T_{d(x)}^{\bullet})=x$ and $(\overline{\tau}_{x})_{n-1}(S)=\left\\{\begin{array}[]{ll}c(x)&\mbox{ if $S={\bf c}(T_{d(x)}^{\bullet})$}\\\ (\tau_{dx})_{n-1}(S)&\mbox{ if $S\subseteq T_{dx}$}\end{array}\right.$ and $(\overline{\tau}_{x})_{<(n-1)}=(\tau_{dx})_{<(n-1)}$. We put $\eta_{P}(x)=\bar{x}$. Then $\bar{P}$ is a positive-to-one computad with $\eta_{P}$ the inclusion of $n$-indeterminates. Moreover any positive-to-one $n$-computad $Q$ is equivalent to a computad $\bar{P}$, for some $P$ in ${\bf Comma}^{+/1}_{n}$. 2. 2. Let $P$ be an object of ${\bf Comma}^{+/1}_{n}$, $!:\overline{P}\longrightarrow{\cal T}$ the unique morphism into the terminal object ${\cal T}$ and $f:S^{\sharp,n}\rightarrow P$ a cell in $\overline{P}_{n}$. Then $!_{n}(f:S^{\sharp,n}\rightarrow P)=S.$ 3. 3. Let $h:P\rightarrow Q$ be an object of ${\bf Comma}^{+/1}_{n}$. Then $\bar{h}:\bar{P}\longrightarrow\bar{Q}$ is a computad morphism, 4. 4. Let $S$ be a positive face structure of dimension at most $n$. Moreover, for a morphism $f:S^{\sharp,n}\longrightarrow P$ in ${\bf Comma}^{+/1}_{n}$ we have that $\overline{f}_{k}(T)=f\circ(i_{T})^{\sharp,n}$ where $k\leq n$, $T\in S^{*}_{k}$ and $i_{T}:T\longrightarrow S$ is the inclusion. 5. 5. Let $S$ be a positive face structure of dimension $n$, $P$ positive-to-one computad, $g,h:S^{*}\longrightarrow P$ computad maps. Then $g=h\hskip 28.45274pt{\rm iff}\hskip 28.45274ptg_{n}(S)=h_{n}(S).$ 6. 6. Let $S$ be a positive face structure of dimension at most $n$, $P$ be an object in ${\bf Comma}^{+/1}_{n}$. Them we have a bijective correspondence $\begin{array}[]{c}f:S^{\sharp,n}\longrightarrow P\;\;\in{\bf Comma}^{+/1}_{n}\\\ \hline\cr\overline{f}:S^{*}\longrightarrow\overline{P}\;\;\in{{\bf Comp}^{+/1}_{n}}\end{array}$ such that, $\overline{f}_{n}(S)=f$, and for $g:S^{*}\longrightarrow\overline{P}$ we have $g=\overline{g_{n}(S)}$. 7. 7. The map $\kappa^{P}_{n}:\coprod_{S}{\bf Comp}(S^{*},\overline{P})\longrightarrow\overline{P}_{n}$ $g:S^{*}\rightarrow\overline{P}\;\;\;\mapsto\;\;\;g_{n}(S)$ where coproduct is taken over all (up to iso) positive face structures $S$ of dimension at most $n$, is a bijection. In other words, any cell in $\bar{P}$ has a unique description. Proof. Ad 1. We have to verify that $\overline{P}$ satisfy the laws of ${\omega}$-categories and that it is free in the appropriate sense. Laws ${\omega}$-categories are left for the reader. We shall show that $\overline{P}$ is free in the appropriate sense. Let $C$ be an ${\omega}$-category, $g_{<n}:P_{<n}\rightarrow C_{<n}$ and $(n-1)$-functor and $g_{n}:|P|_{n}\rightarrow C_{n}$ a function so that the diagram $P_{n-1}\vphantom{C_{n-1}}$$\vphantom{P_{n-1}}C_{n-1}$$g_{n-1}$$|P|_{n}\vphantom{C_{n}}$$\vphantom{|P|_{n}}C_{n}$$g_{n}$$d$$c$$d$$c$ commutes serially. We shall define an $n$-functor $\overline{g}:\overline{P}\rightarrow C$ extending $g$ and $g_{n}$. For $x=(S,f)\in\overline{P}_{n}$ we put $\overline{g}_{n}(x)=\left\\{\begin{array}[]{ll}1_{g_{n-1}\circ f_{n-1}(S)}&\mbox{if $dim(S)<n$,}\\\ g_{n}\circ f_{n}(m_{S})&\mbox{if $dim(S)=n$, $S$ is principal, $S_{n}=\\{m_{S}\\}$ }\\\ \overline{g}_{n}(x^{\downarrow a});_{k}\overline{g}_{n}(x^{\uparrow a})&\mbox{if $dim(S)=n$, $a\in Sd(S)_{k}$}\end{array}\right.$ We need to check that $\overline{g}$ is well defined, unique one that extends $g$, preserves domains, codomains, compositions and identities. All these calculations are similar, and they are very much like those in the proof of Proposition 9.1. We shall check, assuming that we already know that $\overline{g}$ is well defined, and preserves identities that compositions are preserved. So let $T$, $T_{1}$, $T_{2}$ be positive face structures such, that $T=T_{1}+_{k}T_{2}$. Since $\overline{g}$ preserves identities, we can restrict to the case $dim(T_{1}),dim(T_{2})>k$. Fix $a\in{\bf c}^{(k)}(T_{1})_{k}\cap\gamma(T_{1})$. So $a\in Sd(T)_{k}$. If $T_{1}=T^{\downarrow a}$ and $T_{2}=T^{\uparrow a}$ then we have $\overline{g}(T)=\overline{g}(T^{\downarrow a});_{k}\overline{g}(T^{\uparrow a})=\overline{g}(T_{1});_{k}\overline{g}(T_{2}).$ If $a\in Sd(T_{1})_{k}$ then $\displaystyle\overline{g}(T)=\overline{g}(T^{\downarrow a});_{k}\overline{g}(T^{\uparrow a})=$ $\displaystyle=\overline{g}(T^{\downarrow a});_{k}(\overline{g}(T_{1}^{\uparrow a});_{k}\overline{g}(T_{2}))=$ $\displaystyle=(\overline{g}(T_{1}^{\downarrow a});_{k}\overline{g}(T_{1}^{\uparrow a}));_{k}\overline{g}(T_{2})=$ $\displaystyle\overline{g}(T_{1});_{k}\overline{g}(T_{2})$ The remaining verifications are similar. Ad 2. Let $!:\overline{P}\longrightarrow{\cal T}$ be the unique computad map into the terminal object, $S$ a positive face structure such that $dim(S)=l\leq n$, $f:S^{\sharp,n}\longrightarrow P$ a cell in $\overline{P}_{n}$. If $l<n$ then by induction we have $!_{n}(f)=S$. If $l=n$ and $S$ is principal then we have, by induction $!_{n}(d(f):({\bf d}S)^{\sharp,n}\rightarrow P)={\bf d}S,\hskip 28.45274pt!_{n}(c(f):({\bf c}S)^{\sharp,n}\rightarrow P)={\bf c}S.$ As $f$ is an indet in $\overline{P}$, $!_{n}(f)$ is a principal positive face structure. But the only (up to an iso) principal positive face structure $B$ such that ${\bf d}B={\bf d}S,\hskip 28.45274pt{\bf d}B={\bf d}S$ is $S$ itself. Thus, in this case, $!_{n}(f)=S$. Now assume that $l=n$, and $S$ is not principal, and that for positive face structures $T$ of smaller size than $S$ the statement holds. Let $a\in Sd(S)_{k}$. We have $!_{n}(f)=!_{n}(f^{\downarrow a};_{k}f^{\uparrow a})=!_{n}(f^{\downarrow a});_{k}!_{n}(f^{\uparrow a})=S^{\downarrow a};_{k}S^{\uparrow a}=S$ where $f^{\downarrow a}=f\circ(\kappa^{\downarrow a})^{\sharp,n}$ and $f^{\uparrow a}=f\circ(\kappa^{\uparrow a})^{\sharp,n}$ and $\kappa^{\downarrow a}$ and $\kappa^{\uparrow a}$ are the maps as in the following pushout ${\bf c}^{(k)}S\vphantom{S^{\uparrow a}}$$\vphantom{{\bf c}^{(k)}S}S^{\uparrow a}$$S^{\downarrow a}\vphantom{S}$$\vphantom{S^{\downarrow a}}S$$\kappa^{\downarrow a}$$\kappa^{\uparrow a}$ Ad 3. The main thing is to show that $\overline{h}$ preserves compositions. This follows from the fact that the functor $(-)^{\sharp,n}:{\bf Fs}^{+/1}_{n}\longrightarrow{\bf Comma}^{+/1}_{n}$ preserves special pullbacks. Ad 4. This is an immediate consequence of 3. Ad 5. Let $S$ be a positive face structure $S$ of dimension at most $n$. To prove 5., we are going to use the description of the $n$-cells in positive-to- one computads given in 1. Moreover, note that by 3. and Lemma 11.2 we have that for $T\in S^{*}_{k}$, the value of $g$ at $T$ is a map in ${\bf Comma}^{+/1}_{k}$, such that $g_{k}(T):T^{\sharp,k}\longrightarrow P^{\natural,k}$, i.e. the domain of $g_{k}(T)$ is necessarily $T^{\sharp,k}$. The implication $\Rightarrow$ is obvious. So assume that $g,h:S^{*}\longrightarrow P$ are different computad maps. Then there is $k\leq n$ and $x\in S_{k}$ such that $g_{k}([x])\neq h_{k}([x])$. We shall show, by induction on size of $T$, that for any $T\in S^{*}_{l}$, such that $x\in T$ , we have $g_{k}(T)\neq h_{k}(T)$ (6) $T=[x]$ has the least size among those positive face structures that contain $x$. Clearly, (6) holds in this case by assumption. Suppose that (6) holds for all $U\in S^{*}_{l^{\prime}}$ whenever for $l^{\prime}<l$ and $x\in U$. Suppose that $T=[y]$ for some $y\in S_{l}$, and $x\in[y]$. Then either $x\in{\bf d}[y]$ or $x\in{\bf c}[y]$. In the former case we have, by inductive hypothesis, that $g_{k}({\bf d}T)\neq h_{k}({\bf d}T)$.Thus $d(g_{k}(T))=g_{k}({\bf d}T)\neq h_{k}({\bf d}T)=d(g_{k}(T))$ But then (6) holds as well. The later case ($x\in{\bf c}[y]$) is similar. Now suppose that $T$ is not principal $x\in T$ and that for $U$ of a smaller size with $x\in U$ the condition (6) holds. Let $a\in Sd(T)_{r}$. Then either $x\in T^{\downarrow a}$ or $x\in T^{\uparrow a}$. Both cases are similar, so we will consider the first one only. Thus, as $T^{\downarrow a}$ has a smaller size than $T$, by inductive hypothesis we have $g_{k}(T^{\downarrow a})\neq h_{k}(T^{\downarrow a})$ (7) As the compositions in $P$ are calculated via pushouts we have that $g_{l}(T^{\downarrow a});_{r}g_{l}(T^{\uparrow a})=[g_{l}(T^{\downarrow a}),g_{l}(T^{\uparrow a})]$ where $[g_{l}(T^{\downarrow a}),g_{l}(T^{\uparrow a})]$ is the unique morphism from the pushout as in the following diagram: $({\bf c}T^{\uparrow a})^{\sharp,l}\vphantom{(T^{\downarrow a})^{\sharp,l}}$$\vphantom{({\bf c}T^{\uparrow a})^{\sharp,l}}(T^{\downarrow a})^{\sharp,l}$$(T^{\uparrow a})^{\sharp,l}\vphantom{T^{\sharp,l}}$$\vphantom{(T^{\uparrow a})^{\sharp,l}}T^{\sharp,l}$$[g_{l}(T^{\downarrow a}),g_{l}(T^{\uparrow a})]$$P^{\natural,l}$$g_{l}(T^{\uparrow a})$$g_{l}(T^{\downarrow a})$ Similarly $h_{l}(T^{\downarrow a});_{r}h_{l}(T^{\uparrow a})=[h_{l}(T^{\downarrow a}),h_{l}(T^{\uparrow a})]$ As morphism form the pushout are equal if and only if their both components are equal we have $g_{l}(T)=g_{l}(T^{\downarrow a};_{r}T^{\uparrow a})=g_{l}(T^{\downarrow a});_{r}g_{l}(T^{\uparrow a})=$ $[g_{l}(T^{\downarrow a}),g_{l}(T^{\uparrow a})]\neq[h_{l}(T^{\downarrow a}),h_{l}(T^{\uparrow a})]=$ $=h_{l}(T^{\downarrow a});_{r}h_{l}(T^{\uparrow a})=h_{l}(T^{\downarrow a};_{r}T^{\uparrow a})=h_{l}(T)$ Thus (6) holds for all $T\in S^{*}$ such that $x\in T$. As $x\in S$, we get that $g_{n}(S)\neq h_{n}(S)$ as required. Ad 6. Fix a positive face structure $S$ of dimension at most $n$. Let $f:S^{\sharp,n}\longrightarrow P$ be a cell in $\overline{P}_{n}$. By 4, we have $\overline{f}_{n}(S)=f\circ(i_{S})^{\sharp,n}=f\circ(1_{S})^{\sharp,n}=f\circ(1_{S}^{\sharp,n})=f.$ Let $g:S^{*}\longrightarrow\overline{P}$ be a computad map. To show that $g=\overline{g_{n}(S)}$, by 5, it is enough to show that $(\overline{g_{n}(S)})_{n}(S)=g_{n}(S).$ Using 4 again, we have, $(\overline{g_{n}(S)})_{n}(S)=g_{n}(S)\circ(i_{S})^{\sharp,n}=$ $=g_{n}(S)\circ i_{S^{\sharp,n}}=g_{n}(S)\circ 1_{S^{\sharp,n}}=g_{n}(S).$ Thus, by 5, $(\overline{g_{n}(S)})=g$. Ad 7. It follows immediately from 6. $~{}~{}\Box$ From the Proposition 12.1.7 we know that each cell in a positive-to-one computad has (up to an isomorphism) a unique description. The following Proposition is a bit more specific. ###### Proposition 12.2 Let $P$ be a positive-to-one computad, $n\in{\omega}$, and $a\in P_{n}$. Let $T_{a}$ be $!^{P}_{n}(a)$ (where $!^{P}:P\longrightarrow{\cal T}$ is the unique morphism into the terminal computad). Then there is a unique computad map $\tau_{a}:T_{a}^{*}\longrightarrow P$ such that $(\tau_{a})_{n}(T_{a})=a$, i.e. each cell has an essentially unique description. Moreover, we have: 1. 1. for any $a\in P$ we have $\tau_{da}=d(\tau_{a})=\tau_{da}=\tau_{a}\circ({\bf d}_{T_{a}})^{*},\hskip 28.45274pt\tau_{c(a)}=c(\tau_{a})=\tau_{c(a)}=\tau_{a}\circ({\bf c}_{T_{a}})^{*},$ $\tau_{1_{a}}=\tau_{a}$ 2. 2. for any $a,b\in P$ such that $c^{(k)}(a)=d^{(k)}(b)$ we have $\tau_{a;_{k}b}=[\tau_{a},\tau_{b}]:T_{a}^{*}+_{{\bf c}^{(k)}T_{a}^{*}}T_{b}^{*}\longrightarrow P,$ 3. 3. for any positive face structure $S$, for any computad map $f:S^{*}\longrightarrow P$, $\overline{\tau_{f_{n}(S)}}=f.$ 4. 4. for any positive face structure $S$, any ${\omega}$-functor $f:S^{*}\longrightarrow P$ can be essentially uniquely factorized as $S^{*}\vphantom{P}$$\vphantom{S^{*}}P$$f$$T_{f(S)}^{*}$$f^{in}$$\tau_{f(S)}$ where $f^{in}$ is an inner map and $(\tau_{f(S)},T_{f(S)})$ is the description of the cell $f(S)$. Proof. Using the above description of the positive-to-one computad $P$ we have that $a:(T_{a})^{\sharp,n}\longrightarrow P^{\natural,n}$. We put $\tau_{a}=\overline{a}$. By Proposition 12.1 point 6, we have that $(\tau_{a})_{n}(T_{a})=\overline{a}_{n}(T_{a})=a$, as required. The uniqueness of $(T_{a},\tau_{a})$ follows from Proposition 12.1 point 5. The remaining part is left for the reader. $~{}~{}\Box$ ## 13 Positive-to-one computads form a presheaf category In this section we want to prove that the category ${\bf Comp}^{+/1}$ is equivalent to the presheaf category $Set^{({\bf pFs}^{+/1})^{op}}$. In fact, we will show that both categories are equivalent to the category $sPb(({\bf Fs}^{+/1})^{op},Set)$ of special pullbacks preserving functors from $({\bf Fs}^{+/1})^{op}$ to $Set$. First note that the inclusion functor ${\bf i}:{\bf pFs}^{+/1}\longrightarrow{\bf Fs}^{+/1}$ induces the adjunction $Ran_{\bf i}$${\bf i}^{*}$$Set^{{({\bf pFs}^{+/1})}^{op}}\vphantom{Set^{{({\bf Fs}^{+/1})}^{op}}}$$\vphantom{Set^{{({\bf pFs}^{+/1})}^{op}}}Set^{{({\bf Fs}^{+/1})}^{op}}$ with ${\bf i}^{*}$ is the functor of composing with ${\bf i}$ and $Ran_{\bf i}$ is a the right Kan extension along ${\bf i}$. Recall that for $F$ in $Set^{{({\bf pFs}^{+/1})}^{op}}$, $S$ in ${\bf Fs}^{+/1}$, it is defined as the following limit $(Ran_{\bf i}F)(S)=LimF\circ\Sigma^{S,op}$ where $\Sigma^{S,op}$ is the dual of the functor $\Sigma^{S}$ defined before Lemma 6.4. Note that as $({\bf pFs}^{+/1}\downarrow S)^{op}=S\downarrow({\bf pFs}^{+/1})^{op}$ we have $\Sigma^{S,op}:S\downarrow({\bf pFs}^{+/1})^{op}\longrightarrow({\bf pFs}^{+/1})^{op}.$ As ${\bf i}$ is full and faithful the right Kan extension $Ran_{\bf i}(F)$ is an extension. Therefore the counit of this adjunction $\varepsilon_{F}:(Ran_{\bf i}\,F)\circ{\bf i}\longrightarrow F$ is an isomorphism. The functor $Ran_{\bf i}F$ is so defined that it preserves special limits. Hence by Lemma 6.4 it preserves special pullbacks. It is easy to see, that for $G$ in $Set^{{({\bf Fs}^{+/1})}^{op}}$ the unit of adjunction $\eta_{G}:G\longrightarrow Ran_{\bf i}(G\circ{\bf i})$ is an isomorphism iff $G$ preserves special pullbacks. Thus we have ###### Proposition 13.1 The above adjunction restricts to the following equivalence of categories $Ran_{\bf i}$ ${\bf i}^{*}$$Set^{{({\bf pFs}^{+/1})}^{op}}\vphantom{sPb(({\bf Fs}^{+/1})^{op},Set)}$$\vphantom{Set^{{({\bf pFs}^{+/1})}^{op}}}sPb(({\bf Fs}^{+/1})^{op},Set)$ $~{}~{}\Box$ Now we will set up the adjunction $\widetilde{(-)}$$\widehat{(-)}={\bf Comp}^{+/1}((\simeq)^{*},-)$$sPb(({\bf Fs}^{+/1})^{op},Set)\vphantom{{\bf Comp}^{+/1}}$$\vphantom{sPb(({\bf Fs}^{+/1})^{op},Set)}{\bf Comp}^{+/1}$ which will turn out to be an equivalence of categories. The functor $\widehat{(-)}$ sends a positive-to-one computad $P$ to a functor $\widehat{P}={\bf Comp}^{+/1}((-)^{*},P):({\bf Fs}^{+/1})^{op}\longrightarrow Set$ $\widehat{(-)}$ is defined on morphism in the obvious way, by composition. We have ###### Lemma 13.2 Let $P$ be a positive-to-one computad. Then $\widehat{P}$ defined above is a special pullbacks preserving functor. Proof. This is an immediate consequence of the fact that the functor $(-)^{*}$ preserves special pushouts. $~{}~{}\Box$ Now suppose we have a special pullbacks preserving functor $F:({\bf Fs}^{+/1})^{op}\longrightarrow Set$. We shall define a positive-to-one computad $\widetilde{F}$. As $n$-cells of $\widetilde{F}$ we put $\widetilde{F}_{n}=\coprod_{S}F(S)$ where the coproduct is taken over all222 In fact, we think about such a coproduct $\coprod_{S}F(S)$ as if it were to be taken over sufficiently large (so that each isomorphism type of positive face structures is represented) set of positive face structures $S$ of dimension at most $n$. Then, if positive face structures $S$ and $S^{\prime}$ are isomorphic via (necessarily unique) isomorphism $h$, then the cells $x\in F(S)$ and $x^{\prime}\in F(S^{\prime})$ are considered equal iff $F(h)(x)=x^{\prime}$. (up to iso) positive face structures $S$ of dimension at most $n$. If $k\leq n$, the identity map $1^{(n)}:\widetilde{F}_{k}\longrightarrow\widetilde{F}_{n}$ is the obvious embedding induced by identity maps on the components of the coproducts. Now we shall describe the domains and codomains in $\widetilde{F}$. Let $S$ be a positive face structure of dimension at most $n$, $a\in F(S)\hookrightarrow\widetilde{F}_{n}$. We have in ${\bf Fs}^{+/1}$ the $k$-th domain and the $k$-th codomain morphisms: ${\bf d}^{(k)}S\vphantom{{\bf c}^{(k)}S}$$\vphantom{{\bf d}^{(k)}S}{\bf c}^{(k)}S$$S$${\bf d}^{(k)}_{S}$${\bf c}^{(k)}_{S}$ We put $d^{(k)}(a)=F({\bf d}^{(k)}_{S})(a)\in F({\bf d}^{(k)}S)\hookrightarrow\widetilde{F}_{k},$ $c^{(k)}(a)=F({\bf c}^{(k)}_{S})(a)\in F({\bf c}^{(k)}S)\hookrightarrow\widetilde{F}_{k}.$ Finally, we define the compositions in $\widetilde{F}$. Let $n_{1},n_{2}\in{\omega}$, $n=max(n_{1},n_{2})$, $k<min(n_{1},n_{2})$, and $a\in F(S)\hookrightarrow\widetilde{F}_{n_{1}}\hskip 28.45274ptb\in F(T)\hookrightarrow\widetilde{F}_{n_{2}},$ such that $c^{(k)}(a)=F({\bf c}^{(k)}_{S})(a)=F({\bf d}^{(k)}_{T})(b)=d^{(k)}(b).$ We shall define the cell $a;_{k}b\in\widetilde{F}_{n}$. We take a special pushout in ${\bf Fs}^{+/1}$: ${\bf c}^{(k)}S\vphantom{T}$$\vphantom{{\bf c}^{(k)}S}T$${\bf d}^{(k)}_{T}$$S\vphantom{S+_{k}T}$$\vphantom{S}S+_{k}T$$\kappa_{1}$${\bf c}^{(k)}_{S}$$\kappa_{2}$ As $F$ preserves special pullbacks (from $({\bf Fs}^{+/1})^{op}$) it follows that the square $F({{\bf c}^{(k)}S})\vphantom{F(T)}$$\vphantom{F({{\bf c}^{(k)}S})}F(T)$$F({\bf d}^{(k)}_{T})$$F(S)\vphantom{F(S+_{k}T)}$$\vphantom{F(S)}F(S+_{k}T)$$F(\kappa_{1})$$F({\bf c}^{(k)}_{S})$$F(\kappa_{2})$ is a pullback in $Set$. Thus there is a unique element $x\in F(S+_{k}T)\hookrightarrow\widetilde{F}_{n}$ such that $F(\kappa_{1})(x)=a,\hskip 28.45274ptF(\kappa_{1})(x)=b.$ We put $a;_{k}b=x.$ This ends the definition of $\widetilde{F}$. For a morphism $\alpha:F\longrightarrow G$ in $sPb(({\bf Fs}^{+/1})^{op},Set)$ we put $\widetilde{\alpha}=\\{\widetilde{\alpha}_{n}:\widetilde{F}_{n}\longrightarrow\widetilde{G}_{n}\\}_{n\in{\omega}}$ such that $\widetilde{\alpha}_{n}=\coprod_{S}\alpha_{S}:\widetilde{F}_{n}\longrightarrow\widetilde{G}_{n}$ where the coproduct is taken over all (up to iso) positive face structures $S$ of dimension at most $n$. This ends the definition of the functor $\widetilde{(-)}$. We have ###### Proposition 13.3 The functor $\widetilde{(-)}:sPb(({\bf Fs}^{+/1})^{op},Set)\longrightarrow{\bf Comp}^{+/1}$ is well defined. Proof. The verification that $\widetilde{(-)}$ is a functor into ${{\omega}Cat}$ is left for the reader. We shall verify that for any special pullbacks preserving functor $F:{{\bf Fs}^{+/1}}^{op}\longrightarrow Set$, $\widetilde{F}$ is a positive-to-one computad, whose $n$-indets are $|\widetilde{F}|_{n}=\coprod_{B\in{\bf pFs}^{+/1},dim(B)=n}F(B)\;\hookrightarrow\;\coprod_{S\in{\bf Fs}^{+/1},dim(S)\leq n}F(S)=\widetilde{F}_{n}.$ Let $P$ be the truncation of $\widetilde{F}$ in ${\bf Comma}^{+/1}_{n}$, i.e. $P=\widetilde{F}^{\natural,n}$. We shall show that $\widetilde{F}_{n}$ is in a bijective correspondence with $\overline{P}_{n}$ described in the previous section. We define a function $\varphi:\overline{P}_{n}\longrightarrow\widetilde{F}_{n}$ so that for a cell $f:S^{\sharp,n}\longrightarrow P$ in $\overline{P}_{n}$ we put $\varphi(f)=\left\\{\begin{array}[]{ll}1_{f_{n-1}(S)}&\mbox{if $dim(S)<n$,}\\\ f_{n}(m_{S})&\mbox{if $dim(S)=n$, $S$ principal, $S_{n}=\\{m_{S}\\}$}\\\ \varphi(f^{\downarrow a});_{k}\varphi(f^{\uparrow a})&\mbox{if $dim(S)=n$, $a\in Sd(S)_{k}$.}\end{array}\right.$ and the morphisms in $\varphi(f^{\downarrow a})$ and $\varphi(f^{\uparrow a})$ in ${\bf Comma}^{+/1}_{n}$ are obtained by compositions so that the diagram $S^{\sharp,n}$$(S^{\downarrow a})^{\sharp,n}$$(S^{\uparrow a})^{\sharp,n}$$\vphantom{}P$$f$$f^{\downarrow a}$$f^{\uparrow a}$ commutes. We need to verify, by induction on $n$, that $\varphi$ is well defined, bijective and that it preserves compositions, domains, and codomains. We shall only verify (partially) that $\varphi$ is well defined, i.e. the definition $\varphi$ for any non-principal positive face structure $S$ of dimension $n$ does not depend on the choice of the saddle point of $S$. Let $a,x\in Sd(S)$ so that $k=dim(x)<dim(a)=m$. Using Lemma 8.4 and the fact that $(-)^{\sharp,n}$ preserves special pushouts, we have $\displaystyle\varphi(f^{\downarrow a});_{m}\varphi(f^{\uparrow a})=$ $\displaystyle=(\varphi(f^{\downarrow a\downarrow x});_{k}\varphi(f^{\downarrow a\uparrow x}));_{m}(\varphi(f^{\uparrow a\downarrow x});_{k}\varphi(f^{\uparrow a\uparrow x}))=$ $\displaystyle=(\varphi(f^{\downarrow a\downarrow x});_{m}\varphi(f^{\uparrow a\downarrow x}));_{k}(\varphi(f^{\downarrow a\uparrow x});_{m}\varphi(f^{\uparrow a\uparrow x}))=$ $\displaystyle=(\varphi(f^{\downarrow x\downarrow a});_{m}\varphi(f^{\downarrow x\uparrow a}));_{k}(\varphi(f^{\uparrow x\downarrow a});_{m}\varphi(f^{\uparrow x\uparrow a}))=$ $\displaystyle=\varphi(f^{\downarrow x});_{m}\varphi(f^{\uparrow x})$ as required in this case. The reader can compare these calculations with the those, in the same case, of Proposition 9.1 ($F$ is replaces by $\varphi$ and $T$ is replaces by $f$). So there is no point to repeat the other calculations. $~{}~{}\Box$ For $P$ in ${\bf Comp}^{+/1}$ we define a computad map $\eta_{P}:P\longrightarrow\widetilde{\widehat{P}}$ so that for $x\in P_{n}$ we put $\eta_{P,n}(x)=\tau_{x}:T_{x}^{*}\rightarrow P.$ For $F$ in $sPb(({\bf Fs}^{+/1})^{op},Set)$ we define a natural transformation $\varepsilon_{F}:\widehat{\widetilde{F}}\longrightarrow F,$ such that, for a positive face structure $S$ of dimension $n$, $(\varepsilon_{F})_{S}:\widehat{\widetilde{F}}(S)\longrightarrow F(S)$ and $g:S^{*}\rightarrow\widetilde{F}\in\widehat{\widetilde{F}}(S)$ we put $(\varepsilon_{F})_{S}(g)=g_{n}(S).$ ###### Proposition 13.4 The functors $\widetilde{(-)}$ $\widehat{(-)}={\bf Comp}^{+/1}((\simeq)^{*},-)$$sPb(({\bf Fs}^{+/1})^{op},Set)\vphantom{{\bf Comp}^{+/1}}$$\vphantom{sPb(({\bf Fs}^{+/1})^{op},Set)}{\bf Comp}^{+/1}$ together with the natural transformations $\eta$ and $\varepsilon$ defined above form an adjunction ($\widehat{(-)}\dashv\widetilde{(-)}$). It establishes the equivalence of categories $sPb(({\bf Fs}^{+/1})^{op},Set)$ and ${\bf Comp}^{+/1}$. Proof. The fact that both $\eta$ and $\varepsilon$ are bijective on each component follows immediately from Proposition 12.1 point 6. So we shall verify the triangular equalities only. Let $P$ be a computad, and $F$ be a functor in $sPb(({\bf Fs}^{+/1})^{op},Set)$. We need to show that the triangles $\widehat{P}\vphantom{\widehat{P}}$$\vphantom{\widehat{P}}\widehat{P}$$1_{\widehat{P}}$$\widehat{\widetilde{\widehat{P}}}$$\widehat{\eta_{P}}$$\varepsilon_{\widehat{P}}$$\widetilde{F}\vphantom{\widetilde{F}}$$\vphantom{\widetilde{F}}\widetilde{F}$$1_{\widetilde{F}}$$\widetilde{\widehat{\widetilde{F}}}$$\eta_{\widetilde{F}}$$\widetilde{\varepsilon_{F}}$ commute. So let $f:S^{*}\rightarrow P\in\widehat{P}(S)$. Then, we have $\varepsilon_{\widehat{P}}\circ\widehat{\eta_{P}}(f)=\varepsilon_{\widehat{P}}(\eta_{P}\circ f)=(\eta_{P}\circ f)_{n}(S)=$ $=(\eta_{P})_{n}(f_{n}(S))=\tau_{f_{n}(S)}=f$ So let $x\in F(S)\longrightarrow\widetilde{F}_{n}$. Then we have $\widetilde{\varepsilon_{F}}\circ\eta_{\widetilde{F}}(x)=\widetilde{\varepsilon_{F}}(\tau_{x})=(\tau_{x})_{n}(1_{T_{x}})=x$ So both triangles commutes, as required. $~{}~{}\Box$ From Propositions 13.1 and 13.4 we get immediately ###### Corollary 13.5 The functor $\widehat{(-)}:{\bf Comp}^{+/1}\longrightarrow Set^{({\bf pFs}^{+/1})^{op}}$ such that for a positive-to-one computad $X$, $\widehat{X}={\bf Comp}^{+/1}((-)^{*},X):({\bf pFs}^{+/1})^{op}\longrightarrow Set$ is an equivalence of categories. ## 14 The principal pushouts Recall the positive face structure $\alpha^{n}$ from section 9. A total composition map is an inner ${\omega}$-functor whose domain is of form $(\alpha^{n})^{*}$, for some $n\in{\omega}$. If $S$ is a positive face structure of dimension $n$, then the total composition of $S$ (in fact $S^{*}$) is denoted by $\mu^{S^{*}}:\alpha^{n,*}\longrightarrow S^{*}.$ It is uniquely determined by the condition $\mu^{S^{*}}_{n}(\alpha^{n})=S$. We have ###### Proposition 14.1 Let $N$ be a normal positive face structure. With the notation as above, the square $\alpha^{n,*}\vphantom{\alpha^{n+1,*}}$$\vphantom{\alpha^{n,*}}\alpha^{n+1,*}$${\bf d}_{\alpha^{n+1}}^{*}$$N^{*}\vphantom{N^{\bullet,*}}$$\vphantom{N^{*}}N^{\bullet,*}$${\bf d}_{N^{\bullet}}^{*}$${\mbox{\boldmath$\mu$}}^{N^{*}}$${\mbox{\boldmath$\mu$}}^{N^{\bullet,*}}$ is a pushout in ${{\cal C}types}^{+/1}_{\omega}$. Such pushouts are called principal pushouts. Proof. This is an easy consequence of Proposition 7.3, particularly point 4. $~{}~{}\Box$ From the above proposition we immediately get ###### Corollary 14.2 If $n>0$ and $P$ is a principal positive face structure of dimension $n$ then the square $\alpha^{n-1,*}\vphantom{\alpha^{n,*}}$$\vphantom{\alpha^{n-1,*}}\alpha^{n,*}$${\bf d}_{\alpha^{n,*}}$${\bf d}P^{*}\vphantom{P^{*}}$$\vphantom{{\bf d}P^{*}}P^{*}$${\bf d}_{P}^{*}$${\mbox{\boldmath$\mu$}}^{{\bf d}P}$${\mbox{\boldmath$\mu$}}^{P}$ is a (principal) pushout in ${{\cal C}types}^{+/1}_{\omega}$. $~{}~{}\Box$ ###### Theorem 14.3 (V.Harnik) 333The original statement of V.Harnik is saying that the nerve functor from ${\omega}$-categories to (all) computads is monadic. However, in the present context the argument given by V.Harnik is directly proving the present statement, i.e. that the principal pullbacks are preserved whenever the special one’s are. This, statement is used to show that the category of ${\omega}$-categories is equivalent to the category of special pullback preserving functors from $({{\cal C}types}^{+/1})^{op}$ to $Set$, c.f. Corollary 15.2. From that statement, the monadicity of the nerve functor is an easy corollary, c.f. Theorem 16.7. In the remainder of this section the Harnik’s argument, adopted to the present context, is presented. Let $F:({{\cal C}types}^{+/1}_{\omega})^{op}\longrightarrow Set$ be a special pullback preserving functor. Then $F$ preserves the principal pullbacks as well. Theorem 14.3 is a special case of Lemma 14.6, for $k=n-1$. The proof of the above theorem will be divided into three Lemmas. Theorem 14.3 is a special case of Lemma 14.6, for $k=n-1$. Before we even formulate these Lemmas we need to introduce some constructions on positive face structures and define some ${\omega}$-functors between computypes. Introducing these constructs and notation for them we shall make some comments how they are going to be interpreted by special pullback preserving morphisms from $({{\cal C}types}^{+/1}_{\omega})^{op}$ to $Set$. Fix $k\leq n$, and a $P$ principal positive face structure of dimension $n$. We say that $P$ is $k$-globular iff ${\bf d}^{(l)}P$ is principal, for $k\leq l\leq n$, i.e. $\delta^{(l)}({\bf p}_{n})$ is a singleton, for $k\leq l\leq n$, where $P_{n}=\\{{\bf p}_{n}\\}$. The $k$-globularization $\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P$ of $P$ is the $k$-globular positive face structure of dimension $n$ defined as follows. Note that by Lemma 7.1 $P_{l}=\delta({\bf p}_{l+1})\cup\\{{\bf p}_{l}\\}$ for $0\leq l<n$. We put $\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P_{l}=\left\\{\begin{array}[]{ll}\\{{\bf q}_{l},{\bf p}_{l}\\}&\mbox{for $k\leq l<n$,}\\\ P_{l}&\mbox{otherwise.}\end{array}\right.$ For $x\in\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P$, $\gamma^{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P}(x)=\left\\{\begin{array}[]{ll}{\bf p}_{l-1}&\mbox{if $x={\bf q}_{l}$ for some $k\leq l<n$,}\\\ \gamma^{P}(x)&\mbox{otherwise.}\end{array}\right.$ and $\delta^{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P}(x)=\left\\{\begin{array}[]{ll}{\bf q}_{l}&\mbox{if $x\in\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P_{l+1}$ for some $k<l<n$,}\\\ \delta^{P}({\bf p}_{k})&\mbox{if $x\in\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P_{k}$,}\\\ \delta^{P}(x)&\mbox{otherwise.}\end{array}\right.$ Note that $\mbox{\scriptsize{\framebox(0.0,0.0)[bl]{$n$}}}P$ is $P$ itself. Thus the elements of the shape $\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$ are $k$-globularized versions of the elements of the shape $P^{*}$. As the following positive face structures ${\bf c}^{(k)}P\;\cong\;{\bf c}^{(k)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P\;\cong\;{\bf c}{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P\;\cong\;{\bf d}{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P\;\cong\;{\bf d}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P$ are isomorphic, we can form the following special pushouts ${\bf c}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$${\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P}$${\bf c}^{(k+1)}P^{*}\vphantom{{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k+1)}P^{*}}{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$$\kappa_{1}$${\bf c}^{*}_{{\bf c}^{(k+1)}P}$$\kappa_{2}$ and ${\bf c}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$${\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P}$${\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}\vphantom{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$$\kappa^{\prime}_{1}$${\bf c}^{*}_{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P}$$\kappa^{\prime}_{2}$ We describe in details the positive face structures we have just defined: $\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P,\;\;\;\;\;P^{\prime}={\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P,\;\;\;\;\;\mathrm{and}\;\;\;\;\;P^{\prime\prime}={\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P,$ for a $(k+1)$-globular positive face structures $P$, i.e. in case $P$ is equal to $\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P$. $\begin{array}[]{|c|c|c|c|c|}dim&P&\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P&P^{\prime}&P^{\prime\prime}\\\ \hline\cr n&\\{{\bf p}_{n}\\}&\\{{\bf p}_{n}\\}&\\{{\bf p}_{n}\\}&\\{{\bf p}_{n}\\}\\\ n\\!\\!-\\!\\!1&\\{{\bf q}_{n-1},{\bf p}_{n-1}\\}&\\{{\bf q}_{n-1},{\bf p}_{n-1}\\}&\\{{\bf q}_{n-1},{\bf p}_{n-1}\\}&\\{{\bf q}_{n-1},{\bf p}_{n-1}\\}\\\ \vdots&\vdots&\vdots&\vdots&\vdots\\\ k\\!\\!+\\!\\!1&\\{{\bf q}_{k\\!+\\!1},{\bf p}_{k\\!+\\!1}\\}&\\{{\bf q}_{k\\!+\\!1},{\bf p}_{k\\!+\\!1}\\}&\\{{\bf r}_{k\\!+\\!1},{\bf q}_{k\\!+\\!1},{\bf p}_{k\\!+\\!1}\\}&\\{{\bf r}_{k\\!+\\!1},{\bf q}_{k\\!+\\!1},{\bf p}_{k\\!+\\!1}\\}\\\ k&\delta({\bf p}_{k+1})\cup\\{{\bf p}_{k}\\}&\\{{\bf q}_{k},{\bf p}_{k}\\}&\delta({\bf p}_{k+1})\cup\\{{\bf q}_{k},{\bf p}_{k}\\}&\\{{\bf r}_{k},{\bf q}_{k},{\bf p}_{k}\\}\\\ \vdots&\vdots&\vdots&\vdots&\vdots\\\ l&P_{l}&P_{l}&P_{l}&P_{l}\end{array}$ $0\leq l<k$. The functions $\gamma$ and $\delta$ in both $P^{\prime}$ and $P^{\prime\prime}$ are easy to figure out. We give few less obvious values below. In $P^{\prime}$ $\delta^{P^{\prime}}({\bf r}_{k+1})=\delta^{P}({\bf p}_{k+1}),\;\;\delta^{P^{\prime}}({\bf q}_{k+1})=\delta^{P^{\prime}}({\bf p}_{k+1})={\bf q}_{k}$ $\gamma^{P^{\prime}}({\bf r}_{k+1})={\bf q}_{k},\;\;\gamma^{P^{\prime}}({\bf q}_{k+1})=\gamma^{P^{\prime}}({\bf p}_{k+1})={\bf p}_{k},$ and in $P^{\prime\prime}$ $\delta^{P^{\prime\prime}}({\bf r}_{k+1})={\bf r}_{k}.$ Now we shall define some ${\omega}$-functors between computypes. To describe their meaning let us fix a special pullback preserving functors from $F:({{\cal C}types}^{+/1}_{\omega})^{op}\longrightarrow Set$. The ${\omega}$-functors denoted by letter $\mu$ are interpreted as operation that ’globularize’ cells. We have two of them. The first one ${\mbox{\boldmath$\mu$}}^{S^{*}}:\alpha^{n,*}\longrightarrow S^{*}$ was already introduced at the beginning of this section for any positive face structure $S$. The second is the ${\omega}$-functor ${\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}}:\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}\longrightarrow\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$ such that ${\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}}(X)=\left\\{\begin{array}[]{ll}(X-\\{{\bf q}_{k}\\})\cup\delta({\bf p}_{k+1})&\mbox{if ${\bf q}_{k}\in X$}\\\ X&\mbox{otherwise.}\end{array}\right.$ for $X\in\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$, and The fact that these operations are interpreted as globularization of cells can be explained as follows. The function $F({\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}}):F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})\longrightarrow F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}),$ takes a $(k+1)$-globular $n$-cell $a\in F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})$ and returns a $k$-globular $n$-cell $F({\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}})(a)\in F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*})$. Intuitively, $F({\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}})$ is composing the $k$-domain of $a$ leaving the rest ’untouched’. So it is a ’one-step globularization’. On the other hand, the function $F({\mbox{\boldmath$\mu$}}^{S^{*}}):F(S^{*})\longrightarrow F(\alpha^{n,*})$ is taking an $n$-cell $b$ of an arbitrary shape $S^{*}$ and is returning a globular $n$-cell. This time $F({\mbox{\boldmath$\mu$}}^{S^{*}})$ is composing all the domains and codomains in the cell $b$ as much as possible, so that there is nothing left to be compose. This is the ’full globularization’. Instead of writing $F({\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}})(a)$ and $F({\mbox{\boldmath$\mu$}}^{S^{*}})(b)$ we write ${\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}}(a)$ and ${\mbox{\boldmath$\mu$}}^{S^{*}}(b)$ or even ${\mbox{\boldmath$\mu$}}(a)$ and ${\mbox{\boldmath$\mu$}}(b)$ if it does not lead to confusions. The same conventions will apply to the other operations that we introduce below. We need a separate notation for the ${\omega}$-functor ${\mbox{\boldmath$\mu$}}_{k}:{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}\longrightarrow{\bf c}^{(k+1)}P^{*}$ such that ${\mbox{\boldmath$\mu$}}_{k}(X)=\left\\{\begin{array}[]{ll}{\bf c}^{(k+1)}P&\mbox{if $X={\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P$}\\\ {\bf c}^{(k)}P&\mbox{if $X={\bf c}^{(k)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P$}\\\ X&\mbox{otherwise.}\end{array}\right.$ for $X\in{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$. It is a version of ${\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}}$. The ${\omega}$-functor ${\mbox{\boldmath$\iota$}}_{P}:P^{*}\longrightarrow{\bf d}P^{*}$ is given by ${\mbox{\boldmath$\iota$}}_{P}(X)=\left\\{\begin{array}[]{ll}{\bf d}P&\mbox{if ${\bf c}P\subseteq X$}\\\ X&\mbox{otherwise.}\end{array}\right.$ for $X\in P^{*}$. ${\mbox{\boldmath$\iota$}}_{P}$ is a kind of degeneracy map and it is interpreted as ’a kind of identity’. For a given $(n-1)$-cell $t$ of the shape ${\bf d}P^{*}$, ${\mbox{\boldmath$\iota$}}(t)$ is ’identity on $t$’ but with the codomain composed. This is why we will write ${\mbox{\boldmath$\iota$}}_{t}$ rather than ${\mbox{\boldmath$\iota$}}(t)$. The ${\omega}$-functor ${\mbox{\boldmath$\beta$}}_{k}:{\bf c}^{(k)}P\longrightarrow{\bf d}^{(k)}P$ such that ${\mbox{\boldmath$\beta$}}_{k}(X)=\left\\{\begin{array}[]{ll}{\bf d}^{(k)}P&\mbox{if $X={\bf c}^{(k)}P$}\\\ X&\mbox{otherwise.}\end{array}\right.$ for $X\in P^{*}$, is the operation of ’composition of all the cells at the top’ leaving the rest untouched. Clearly we have ${\mbox{\boldmath$\beta$}}_{n-1}={\bf d}_{P}^{*};{\mbox{\boldmath$\iota$}}_{P}$. The following two ${\omega}$-functors ${[{\mbox{\boldmath$\iota$}}_{{\bf c}^{(k+1)}P};{\bf d}^{(k),*}_{P},{\mbox{\boldmath$\mu$}}_{P}]}:{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}\longrightarrow P^{*}$ and ${[{\mbox{\boldmath$\iota$}}_{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P};{\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P},1_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P}]}:{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}\longrightarrow\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$ are defined as the unique ${\omega}$-functors making the following diagrams ${\bf c}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$${\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P}$${\bf c}^{(k+1)}P^{*}\vphantom{{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k+1)}P^{*}}{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$$\kappa_{1}$${\bf c}^{*}_{{\bf c}^{(k+1)}P}$$\kappa_{2}$$\vphantom{\phantom{{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$${[}{\mbox{\boldmath$\iota$}};{\bf d}^{(k)},{\mbox{\boldmath$\mu$}}{]}$${\bf d}^{(k)}P^{*}$${\mbox{\boldmath$\iota$}}_{{\bf c}^{(k+1)}P^{*}}$${\bf d}^{(k),*}_{P}$${\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}}$ and ${\bf c}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$${\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P}$${\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}\vphantom{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$$\kappa^{\prime}_{1}$${\bf c}^{*}_{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P}$$\kappa^{\prime}_{2}$$\vphantom{\phantom{{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$${[}{\mbox{\boldmath$\iota$}};{\bf d}^{(k)},1_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P^{*}}{]}$${\bf d}^{(k)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$${\mbox{\boldmath$\iota$}}_{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$${\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P}$$1_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P^{*}}$ commute. Finally, we introduce two maps that are a kind of a binary composition but with a wiskering. The first ${\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}:\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}\longrightarrow{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P$ is given by ${\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}(X)=\left\\{\begin{array}[]{ll}X\cup\\{{\bf r}_{k+1},{\bf q}_{k}\\}&\mbox{if $X\cap\\{{\bf q}_{k+1},{\bf p}_{k+1}\\}\neq\emptyset$}\\\ X&\mbox{otherwise.}\end{array}\right.$ for $X\in\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$. The other is ${\mbox{\boldmath$;$}}_{k}:\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}\longrightarrow{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P$ with the same defining formula, for $X\in\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$. In the following diagram all the morphisms that we introduced above are displayed. Most of the subscripts of the morphisms are suppressed for clarity of the picture. ${\bf c}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$${\bf d}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}}$$\vphantom{{\bf d}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$${\mbox{\boldmath$\beta$}}_{k}$${\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}}$${\bf d}^{(k),*}$${\bf d}^{(k),*}$${\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}\vphantom{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}}$$\vphantom{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}}{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$${\bf c}^{(k+1)}P^{*}\vphantom{{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}}$$\vphantom{{\bf c}^{(k+1)}P^{*}}{\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$$\kappa_{1}$${\mbox{\boldmath$\mu$}}_{k}+1$${\bf c}^{*}$${\bf d}^{*}$$\iota$$\kappa^{\prime}_{2}$${\mbox{\boldmath$;$}}_{k}$$[{\mbox{\boldmath$\iota$}};d^{*},1]$${\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}$$[{\mbox{\boldmath$\iota$}};d^{*},{\mbox{\boldmath$\mu$}}]$${\bf d}^{*}$$\iota$${\bf c}^{*}$$\kappa_{2}$${\mbox{\boldmath$\mu$}}_{k}$$\kappa^{\prime}_{1}$ The above cube contains two special pushouts mentioned above. The following Lemma describe some other commutations. ###### Lemma 14.4 With the notation as above we have: 1. 1. ${\bf c}^{*}_{{\bf c}^{(k+1)}P};\iota_{{\bf c}^{(k+1)}P^{*}}={\mbox{\boldmath$\beta$}}_{k}$, 2. 2. ${\bf d}^{*}_{{\bf c}^{(k+1)}P};\kappa_{1}={\bf d}^{(k),*}_{P};({\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}})$, 3. 3. ${\bf d}^{*}_{{\bf c}^{(k+1)}P};\iota_{{\bf c}^{(k+1)}P^{*}}=1_{{\bf d}^{(k)}P^{*}}$, 4. 4. ${\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}};({\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}})=({\mbox{\boldmath$;$}}_{k});({\mbox{\boldmath$\mu$}}_{k}+1_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P})$, 5. 5. $\iota_{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}};{\mbox{\boldmath$\beta$}}_{k}={\mbox{\boldmath$\mu$}}_{k};\iota_{{\bf c}^{(k+1)}P^{*}}$, 6. 6. $({\mbox{\boldmath$;$}}_{k});[{\mbox{\boldmath$\iota$}}_{{\bf c}^{(k+1)}\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P^{*}};{\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P},1_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P^{*}}]=1_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P^{*}}$, 7. 7. $({\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}});[{\mbox{\boldmath$\iota$}}_{{\bf c}^{(k+1)}P^{*}};{\bf d}^{(k),*}_{P},{\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}}]=1_{P^{*}}$. Proof. Routine check. $~{}~{}\Box$ ###### Lemma 14.5 Let $F:({{\cal C}types}^{+/1}_{\omega})^{op}\longrightarrow Set$ be a special pullback preserving functor, $P$ a principal positive face structure of dimension $n$. Then for any $0\leq k<n$, $F$ preserves the pullback in $({{\cal C}types}^{+/1})^{op}$ ${\bf c}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}}$$\vphantom{{\bf c}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*}$${\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P}$${\bf d}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}}$$\vphantom{{\bf d}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$${\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}P}$${\mbox{\boldmath$\beta$}}_{k}$${\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}}$ Proof. Let $F:({{\cal C}types}^{+/1}_{\omega})^{op}\longrightarrow Set$ be a special pullback preserving functor, $P$ a principal positive face structure of dimension $n$. We need to show that the square $F({\bf c}^{(k)}P^{*})\vphantom{F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*})}$$\vphantom{F({\bf c}^{(k)}P^{*})}F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P^{*})$$F({\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k}$}}}P})$$F({\bf d}^{(k)}P^{*})\vphantom{F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})}$$\vphantom{F({\bf d}^{(k)}P^{*})}F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})$$F({\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}P})$$F({\mbox{\boldmath$\beta$}}_{k})$$F({\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}})$ is a pullback in $Set$. Let us fix $t\in F({\bf d}^{(k)}P^{*})$ and $a\in F({\bf c}^{(k)}P^{*})$ such that ${\mbox{\boldmath$\beta$}}(t)={\bf d}^{(k)}(a)$. We will check that it is a pullback, by showing existence and uniqueness of an element $b\in F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})$ such that ${\bf d}^{(k)}(b)=t$ and ${\mbox{\boldmath$\mu$}}(b)=a$. Existence. By Lemma 14.4.1 we have ${\bf c}({\mbox{\boldmath$\iota$}}_{t})={\bf d}^{(k)}(a)$. Since $F$ preserves special pullbacks and ${\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P$ is a special pullback in $({{\cal C}types}^{+/1}_{\omega})^{op}$ we have an element $({\mbox{\boldmath$\iota$}}_{t},a)\in F({\bf c}^{(k+1)}P+_{{\bf c}^{(k)}P}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k$}}}P)$ such that $\kappa_{1}({\mbox{\boldmath$\iota$}}_{t},a)={\mbox{\boldmath$\iota$}}_{t}$ and $\kappa_{2}({\mbox{\boldmath$\iota$}}_{t},a)=a$. We shall show that the element ${\mbox{\boldmath$\iota$}}_{t}{\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}a\in F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})$ is the element we are looking for, i.e. ${\bf d}^{(k)}({\mbox{\boldmath$\iota$}}_{t}{\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}a)=t$ and ${\mbox{\boldmath$\mu$}}({\mbox{\boldmath$\iota$}}_{t}{\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}a)=a$. Using Lemma 14.4 2 and 3 we have ${\bf d}^{(k)}({\mbox{\boldmath$\iota$}}_{t}{\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}a)={\bf d}({\mbox{\boldmath$\iota$}}_{t})=t$ and using Lemma 14.4 4, 5, 6 and assumption ${\mbox{\boldmath$\beta$}}(t)={\bf d}^{(k)}(a)$ we have ${\mbox{\boldmath$\mu$}}({\mbox{\boldmath$\iota$}}_{t}{\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}a)={\mbox{\boldmath$\mu$}}({\mbox{\boldmath$\iota$}}_{t}){\mbox{\boldmath$;$}}_{k}a={\mbox{\boldmath$\iota$}}_{{\mbox{\boldmath$\beta$}}(t)}{\mbox{\boldmath$;$}}_{k}a={\mbox{\boldmath$\iota$}}_{{\bf d}^{(k)}(a)}{\mbox{\boldmath$;$}}_{k}a=a$ This ends the proof of existence. Uniqueness. Now suppose that we have two elements $b,b^{\prime}\in F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})$ such that ${\bf d}^{(k)}(b)=t={\bf d}^{(k)}(b^{\prime})$ and ${\mbox{\boldmath$\mu$}}(b)=a={\mbox{\boldmath$\mu$}}(b)$. Then using Lemma 14.4 7 and the assumption we have $b={\mbox{\boldmath$\iota$}}_{{\bf d}^{(k)}(b)}{\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}\,{\mbox{\boldmath$\mu$}}(b)={\mbox{\boldmath$\iota$}}_{{\bf d}^{(k)}(b^{\prime})}{\mbox{\boldmath$;$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}}\,{\mbox{\boldmath$\mu$}}(b^{\prime})=b^{\prime}.$ So the element with this property is unique. $~{}~{}\Box$ ###### Lemma 14.6 Let $F:({{\cal C}types}^{+/1}_{\omega})^{op}\longrightarrow Set$ be a special pullback preserving functor, $P$ a principal positive face structure of dimension $n$. Then for any $0\leq k<n$, $F$ preserves the following pullback in $({{\cal C}types}^{+/1}_{\omega})^{op}$ (8) $\alpha^{k,*}\vphantom{\alpha^{n,*}}$$\vphantom{\alpha^{k,*}}\alpha^{n,*}$${\bf d}^{(k),*}_{\alpha^{n}}$${\bf d}^{(k)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}}$$\vphantom{{\bf d}^{(k)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$${\bf d}^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}P}$${\mbox{\boldmath$\mu$}}^{{\bf d}^{(k)}P^{*}}$${\mbox{\boldmath$\mu$}}^{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}}$ Proof. The proof is by double induction, on dimension $n$ of the principal positive face structure $P$, and $k<n$. Note that if $k=0$ then, for any $n>0$, the vertical arrows in (8) are isomorphisms, so any functor form $({{\cal C}types}^{+/1}_{\omega})^{op}$ sends (8) to a pullback. This shows in particular that the Lemma holds for $n=1$. As we already mentioned, if $k=n-1$, the square (8) is an arbitrary special pushout. Thus, we assume that $F:({{\cal C}types}^{+/1}_{\omega})^{op}\longrightarrow Set$ is a special pullback preserving functor, and that $P$ a principal positive face structure of dimension $n$, $0\leq k<n$, $F$ preserves the pullback (8). Moreover, for $n^{\prime}<n$ and the principal positive face structure $Q$ of dimension $n^{\prime}$, $F$ preserves the principal pullback in $({{\cal C}types}^{+/1}_{\omega})^{op}$: $\alpha^{n^{\prime}-1,*}\vphantom{\alpha^{n^{\prime},*}}$$\vphantom{\alpha^{n^{\prime}-1,*}}\alpha^{n^{\prime},*}$${\bf d}_{\alpha^{n^{\prime}}}^{*}$${\bf d}Q^{*}\vphantom{Q^{*}}$$\vphantom{{\bf d}Q^{*}}Q^{*}$${\bf d}_{Q}^{*}$${\mbox{\boldmath$\mu$}}^{{\bf d}Q}$${\mbox{\boldmath$\mu$}}^{Q}$$\alpha^{n^{\prime}-1,*}\vphantom{\alpha^{n^{\prime},*}}$$\vphantom{\alpha^{n^{\prime}-1,*}}\alpha^{n^{\prime},*}$${\bf d}^{(n^{\prime}-1),*}_{\alpha^{n^{\prime}}}$${\bf d}^{(n^{\prime}-1)}Q^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$n^{\prime}$}}}Q^{*}}$$\vphantom{{\bf d}^{(n^{\prime}-1)}Q^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$n^{\prime}$}}}Q^{*}$${\bf d}^{(n^{\prime}-1),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$n^{\prime}$}}}Q}$${\mbox{\boldmath$\mu$}}^{{\bf d}^{(n^{\prime}-1)}Q^{*}}$${\mbox{\boldmath$\mu$}}^{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{n^{\prime}}$}}}Q^{*}}$$=$ We shall show that $F$ preserves the pullback $({{\cal C}types}^{+/1}_{\omega})^{op}$ (9) $\alpha^{k+1,*}\vphantom{\alpha^{n,*}}$$\vphantom{\alpha^{k+1,*}}\alpha^{n,*}$${\bf d}^{(k+1),*}_{\alpha^{n}}$${\bf d}^{(k+1)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*}}$$\vphantom{{\bf d}^{(k+1)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*}$${\bf d}^{(k+1),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!2}$}}}P}$${\mbox{\boldmath$\mu$}}^{{\bf d}^{(k+1)}P^{*}}$${\mbox{\boldmath$\mu$}}^{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*}}$ as well. In the following diagram (most of the subscripts and some superscripts were suppressed for clarity): $I$$II$$III$$\alpha^{k,*}\vphantom{\alpha^{n,*}}$$\vphantom{\alpha^{k,*}}\alpha^{n,*}$${\bf d}^{(k)}P^{*}\vphantom{}$$\mu$${\bf d}^{(k),*}$${\bf d}^{(k),*}$${\bf d}^{*}$$\mu$$\alpha^{k+1,*}\vphantom{\phantom{\alpha^{k,*}}}$${\bf d}^{*}$$\mu$${\bf d}^{(k+1),*}$$\mu$${\bf c}^{(k+1)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}}$$\vphantom{{\bf c}^{(k+1)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*}$${\bf d}^{(k+1)}P^{*}\vphantom{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*}}$$\vphantom{{\bf d}^{(k+1)}P^{*}}\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*}$${\bf d}^{(k+1),*}$${\mbox{\boldmath$\beta$}}_{k+1}$$\mu$${\bf d}^{(k+1),*}$ all the squares and triangles commute. Moreover, $F$ sends the squares $I$, $II$, $III$ to pullbacks in $Set$: $I$ by Lemma 14.5, $II$ by inductive hypothesis for $k$, $III$ by inductive hypothesis since $dim({\bf c}^{(k+1)}P)<n$. Let $f:X\longrightarrow F({\bf d}^{(k+1)}P^{*})$ and $g:X\longrightarrow F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})$ be functions such that $F({\mbox{\boldmath$\mu$}}^{{\bf d}^{(k+1)}P^{*}});f=F({\bf d}^{(k+1),*}_{\alpha^{n}});g$ Since $F$ applied to $II$ is a pullback in $Set$, and some squares and triangles in the above diagram commute, there is a unique function $h_{1}:X\longrightarrow F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P)$ such that $h_{1};F(d^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P})=f;F({\mbox{\boldmath$\beta$}}_{k+1});F({\bf d}^{*}_{{\bf c}^{(k+1)}P})\hskip 14.22636pt\mathrm{and}\hskip 14.22636pth_{1};F({\mbox{\boldmath$\mu$}}^{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}P^{*}})=g.$ (10) To get a unique function $h_{2}:X\longrightarrow F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*})$ such that $h_{2};F({\bf d}^{k+1,*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!2}$}}}P})=f\hskip 14.22636pt\mathrm{and}\hskip 14.22636pth_{2};F({\mbox{\boldmath$\mu$}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}})=h_{1}$ (11) we use the fact that $F$ sends $III$ to a pullback in $Set$. The application of $F$ to the diagram above will give the following diagram in $Set$, where we added the additional functions $f$, $g$, $h_{1}$, and $h_{2}$: $F(\alpha^{k,*})\vphantom{F(\alpha^{n,*})}$$\vphantom{F(\alpha^{k,*})}F(\alpha^{n,*})$$F({\bf d}^{(k)}P^{*})\vphantom{}$$F({\mbox{\boldmath$\mu$}})$$F({\bf d}^{(k),*})$$F({\bf d}^{(k),*})$$F({\bf d}^{*})$$F({\mbox{\boldmath$\mu$}})$$F(\alpha^{k+1,*})\vphantom{\phantom{F(\alpha^{k,*})}}$$F({\bf d}^{*})$$F({\bf d}^{(k+1),*})$$F({\mbox{\boldmath$\mu$}})$$F({\mbox{\boldmath$\mu$}})$$\vphantom{\phantom{F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*})}}X$$h_{2}$$h_{1}$$g$$f$$F({\bf c}^{(k+1)}P^{*})\vphantom{F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})}$$\vphantom{F({\bf c}^{(k+1)}P^{*})}F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P^{*})$$F({\bf d}^{(k+1)}P^{*})\vphantom{F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*})}$$\vphantom{F({\bf d}^{(k+1)}P^{*})}F(\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!2$}}}P^{*})$$F({\bf d}^{(k+1),*})$$F({\mbox{\boldmath$\beta$}}_{k+1})$$F({\mbox{\boldmath$\mu$}})$$F({\bf d}^{(k+1),*})$ Thus in order to verify that $f;F({\mbox{\boldmath$\beta$}}_{k+1})=h_{1};F({\bf d}^{(k+1),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P})$ and to get $h_{2}$ satisfying (11), it is enough to verify that $f;F({\mbox{\boldmath$\beta$}}_{k+1});F({\mbox{\boldmath$\mu$}}^{{\bf c}^{(k+1)}P^{*}})=h_{1};F({\bf d}^{(k+1),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P});F({\mbox{\boldmath$\mu$}}^{{\bf c}^{(k+1)}P^{*}})$ (12) $f;F({\mbox{\boldmath$\beta$}}_{k+1});F({\bf d}^{*}_{{\bf c}^{(k+1)}P})=h_{1};F({\bf d}^{(k+1),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}P});F({\bf d}^{*}_{{\bf c}^{(k+1)}P})$ (13) For (12), we have $f;F({\mbox{\boldmath$\beta$}}_{k+1});F({\mbox{\boldmath$\mu$}}^{{\bf c}^{(k+1)}P^{*}})=f;F({\mbox{\boldmath$\mu$}}^{{\bf d}^{(k+1)}P^{*}})=$ $=g;F({\bf d}^{(k+1),*}_{\alpha^{n}})=h_{1};F({\mbox{\boldmath$\mu$}}^{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}P^{*}});F({\bf d}^{(k+1),*}_{\alpha^{n}})=$ $=h_{1};F({\bf d}^{(k+1)^{*}}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}P});F({\mbox{\boldmath$\mu$}}^{{\bf c}^{(k+1)}P^{*}})$ and for (13), we have $f;F({\mbox{\boldmath$\beta$}}_{k+1});F({\bf d}^{*}_{{\bf c}^{(k+1)}P})=h_{1};F(d^{(k),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{$k\\!\\!+\\!\\!1$}}}P})=$ $=h_{1};F({\bf d}^{(k+1),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!1}$}}}P});F({\bf d}^{*}_{{\bf c}^{(k+1)}P}).$ By uniqueness of both $h_{1}$ and $h_{2}$, $h_{2}$ is the unique function such that $h_{2};{\bf d}^{(k+1),*}_{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!2}$}}}P}=f\hskip 14.22636pt\mathrm{and}\hskip 14.22636pth_{2};{\mbox{\boldmath$\mu$}}^{\mbox{\tiny{\framebox(0.0,0.0)[bl]{${}_{k\\!+\\!2}$}}}P^{*}}=g$ i.e. $F$ sends (9) to a pullback in $Set$, as required. $~{}~{}\Box$ ## 15 Yet another full nerve of the ${\omega}$-categories Let ${\cal S}$ denote the category of simple ${\omega}$-categories introduced in [MZ]. It was proved there that any simple ${\omega}$-category is isomorphic to one of form $(\alpha^{\vec{u}})^{*}$ for some ud-vector $\vec{u}$. In fact what we need here is that any simple ${\omega}$-category can be obtained from those of form $(\alpha^{n})^{*}$, with $n\in{\omega}$, via special pushouts. As every simple ${\omega}$-category is a positive computype, we have an inclusion functor ${\bf k}:{\cal S}\longrightarrow{{\cal C}types}^{+/1}_{\omega}.$ In [MZ] we have shown, that $sPb({\cal S}^{op},Set)$ the category of special pullbacks preserving functors from the dual of ${\cal S}$ to $Set$ is equivalent to the category ${\omega}$-categories. We have in fact an adjoint equivalence $\widetilde{(-)}$$\widehat{(-)}={{\omega}Cat}(\simeq,-)$${{\omega}Cat}\vphantom{sPb({\cal S}^{op},Set)}$$\vphantom{{{\omega}Cat}}sPb({\cal S}^{op},Set)$ where $\widehat{C}:{\cal S}^{op}\longrightarrow Set$ is given by $\widehat{C}(A)={{\omega}Cat}(A,C),$ where $A$ is a simple category. We shall show ###### Proposition 15.1 The adjunction $Ran_{\bf k}$ ${\bf k}^{*}$$Set^{{\cal S}^{op}}\vphantom{Set^{({{\cal C}types}^{+/1}_{\omega})^{op}}}$$\vphantom{Set^{{\cal S}^{op}}}Set^{({{\cal C}types}^{+/1}_{\omega})^{op}}$ restricts to an equivalence of categories. $Ran_{\bf k}$ ${\bf k}^{*}$$sPb({\cal S}^{op},Set)\vphantom{sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)}$$\vphantom{sPb({\cal S}^{op},Set)}sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ where $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ is the category of the special pullbacks preserving functors. Proof. First we shall describe the adjunction in details. The counit. Let $G$ be a functor in $sPb({\cal S}^{op},Set)$ , $A$ a simple ${\omega}$-category. We have a functor $({\bf k}\downarrow A)^{op}\vphantom{{\cal S}^{op}}$$\vphantom{({\bf k}\downarrow A)^{op}}{\cal S}^{op}$$\pi^{A}$$\vphantom{\phantom{{\cal S}^{op}}}Set$$G$ with the limit, say $(LimG\circ\pi^{A},\sigma^{A})$. Then the counit $(\varepsilon_{G})_{A}$ is $(\varepsilon_{G})_{A}:(Ran_{\bf k}(G)\circ{\bf k})(A)=Lim(G\circ\pi^{A})\vphantom{G(A)}$$\vphantom{(\varepsilon_{G})_{A}:(Ran_{\bf k}(G)\circ{\bf k})(A)=Lim(G\circ\pi^{A})}G(A)$$\sigma^{A}_{1_{A}}$ As ${\bf k}$ is full and faithful444This conditions translates to the fact that $1_{A}$ is the initial object in $({\bf k}\downarrow A)^{op}$ and therefore that we have an iso $\sigma^{A}_{1_{A}}:LimG\circ\pi^{A}\cong G\circ\pi^{A}(1_{A})=G(A)$., for any $G$, $\varepsilon_{G}$ is an iso. Thus $\varepsilon$ is an iso. The unit. Let $F$ be a functor in $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$, $T$ a positive face structure. We have a functor $({\bf k}\downarrow T^{*})^{op}\vphantom{{\cal S}^{op}}$$\vphantom{({\bf k}\downarrow T^{*})^{op}}{\cal S}^{op}$$\pi^{T^{*}}$$\vphantom{\phantom{{\cal S}^{op}}}{{\cal C}types}^{+/1}_{\omega}$${\bf k}$$\vphantom{\phantom{{{\cal C}types}^{+/1}_{\omega}}}Set$$F$ with the limit, say $(LimF\circ{\bf k}\circ\pi^{T^{*}},\sigma^{T^{*}})$. Then the unit $(\eta_{F})_{T^{*}}$ is the unique morphism into the limit: $F(T^{*})\vphantom{Ran_{\bf k}(F\circ{\bf k})(T^{*})}$$\vphantom{F(T^{*})}Ran_{\bf k}(F\circ{\bf k})(T^{*})$$(\eta_{F})_{T^{*}}$$\parallel$\- $F(A)\vphantom{F(B)}$$\vphantom{F(A)}F(B)$$F(f)$$LimF\circ{\bf k}\circ\pi^{T^{*}}$$\sigma_{a}^{T}$$\sigma_{b}^{T}$$F(a)$$F(b)$ where the triangle in ${{\cal C}types}^{+/1}_{\omega}$ $A\vphantom{B}$$\vphantom{A}B$$f$$T^{*}$$a$$b$ commutes. Note that, as $F$ preserves special pullbacks, and any simple category can be obtained from those of form $\alpha^{n}$ with $n\in{\omega}$, we can restrict the limiting cone $(LimF\circ{\bf k}\circ\pi^{T^{*}},\sigma^{T^{*}})$ to the objects of form $\alpha^{n}$, with $n\in{\omega}$. After this observation we shall prove by induction on size of a positive face structure $T$, that $(\eta_{F})_{T^{*}}$ is an iso. If $dim(T)\leq 1$ then $(\eta_{F})_{T^{*}}$ is obviously an iso. Suppose $T$ is not principal, i.e. we have $a\in Sd(T)$ for some $k\in{\omega}$. By inductive hypothesis the morphisms $(\eta_{F})_{(T^{\downarrow a})^{*}},\;\;\;(\eta_{F})_{{\bf c}^{(k)}(T^{\downarrow a})^{*}},\;\;\;(\eta_{F})_{(T^{\uparrow a})^{*}}$ are iso, and the square ${\bf c}^{(k)}(T^{\downarrow a})\vphantom{T^{\uparrow a}}$$\vphantom{{\bf c}^{(k)}(T^{\downarrow a})}T^{\uparrow a}$$T^{\downarrow a}\vphantom{T}$$\vphantom{T^{\downarrow a}}T$ is a special pushout (see Proposition 6.2) which is sent by $F$ to a pullback. Hence the morphism $(\eta_{F})_{T^{*}}=(\eta_{F})_{(T^{\downarrow a})^{*}}\times(\eta_{F})_{(T^{\uparrow a})^{*}}$ is indeed an iso in this case. If $T$ is principal and $T=(\alpha^{n})^{*}$ then the category $({\bf k}\downarrow(\alpha^{n})^{*})^{op}$ has the initial object $1_{(\alpha^{n})^{*}}$, so morphism $(\eta_{F})_{(\alpha^{n})^{*}}:F((\alpha^{n})^{*})\longrightarrow Ran_{\bf k}(F\circ{\bf k})((\alpha^{n})^{*})$ is an iso. Finally, let assume that $T(=P)$ is any principal positive face structure of dimension $n$. Thus, by Corollary 14.2, we have a principal pushout $(\alpha^{n-1})^{*}\vphantom{(\alpha^{n})^{*}}$$\vphantom{(\alpha^{n-1})^{*}}(\alpha^{n})^{*}$${\bf d}_{\alpha^{n}}^{*}$$({\bf d}P)^{*}\vphantom{P^{*}}$$\vphantom{({\bf d}P)^{*}}P^{*}$${\bf d}_{P}^{*}$${\mbox{\boldmath$\mu$}}^{{\bf d}P}$${\mbox{\boldmath$\mu$}}^{P}$ which, by Theorem 14.3, is preserved by $F$. By induction hypothesis the morphisms $(\eta_{F})_{(T^{\downarrow a})^{*}},\;\;\;(\eta_{F})_{{\bf c}^{(k)}(T^{\downarrow a})^{*}},\;\;\;(\eta_{F})_{(T^{\uparrow a})^{*}}$ are iso, so we have that the morphism $(\eta_{F})_{P^{*}}=(\eta_{F})_{({\bf d}P)^{*}}\times(\eta_{F})_{(\alpha^{n})^{*}}$ is iso, as well. $~{}~{}\Box$ ###### Corollary 15.2 We have a commuting triangle of adjoint equivalences ${{\omega}Cat}$$sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$$sPb({\cal S}^{op},Set)$$\widehat{(-)}$$\widetilde{(-)}$$\widehat{(-)}$$\widetilde{(-)}$${\bf k}^{*}$$Ran_{\bf k}$ In particular the categories ${{\omega}Cat}$ and $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ are equivalent. Proof. It is enough to show that in the above diagram ${\bf k}^{*}\circ\widehat{(-)}=\widehat{(-)}$. But this is obvious. $~{}~{}\Box$ ## 16 A monadic adjunction In this section we show that the inclusion functor ${\bf e}:{\bf Comp}^{+/1}\longrightarrow{{\omega}Cat}$ has a right adjoint which is monadic. First we will give an outline of the proof. Consider the following diagram of categories and functors ${{\omega}Cat}$$sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$${\bf Comp}^{+/1}$$sPb(({\bf Fs}^{+/1})^{op},Set)$$\widehat{(-)}$$\widetilde{(-)}$$\widehat{(-)}$$\widetilde{(-)}$$Lan_{\bf j}$${\bf j}^{*}$${\bf e}$ where ${\bf e}$ is just an inclusion of positive-to-one computads into ${\omega}$-categories and ${\bf j}=(-)^{*}:{\bf Fs}^{+/1}\longrightarrow{{\cal C}types}^{+/1}_{\omega}$ is an essentially surjective embedding. We have already shown (Proposition 13.4, Corollary 15.2) that the vertical morphisms constitute two adjoint equivalences. The proof that ${\bf e}$ has a right adjoint has two parts. First we will check that the functor $Lan_{\bf j}(F)$, the left Kan extension of a composition pullbacks preserving functor $F$ preserves the composition pullbacks. Then we shall check that the above square commutes, i.e. $\widehat{(-)}\circ{\bf e}=Lan_{\bf j}\circ\widehat{(-)}$. Thus reduces the problem to check whether ${\bf j}^{*}$ the left adjoint to $Lan_{\bf j}$ is monadic. ###### Lemma 16.1 The functor of the left Kan extension $Lan_{\bf j}:sPb(({\bf Fs}^{+/1})^{op},Set)\longrightarrow sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ is well defined, i.e. whenever $F:({\bf Fs}^{+/1})^{op}\longrightarrow Set$ preserves special pullbacks so does $Lan_{\bf j}(F):({{\cal C}types}^{+/1}_{\omega})^{op}\longrightarrow Set$. Moreover, the above functor $Lan_{\bf j}$ is the left adjoint to ${\bf j}^{*}:sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)\longrightarrow sPb(({\bf Fs}^{+/1})^{op},Set).$ Proof. Note that once we will prove the first part of the statement the part following ’moreover’ will follow immediately. Fix $F$ in $sPb(({\bf Fs}^{+/1})^{op},Set)$ for the whole proof. First we shall describe the left Kan extension along ${\bf j}$ in a more convenient then the usual way, c.f. [CWM]. Fix a positive face structure $S$. $Lan_{\bf j}(F)(S)$ is the colimit of the following functor ${\bf j}^{op}\downarrow S\vphantom{({\bf Fs}^{+/1})^{op}}$$\vphantom{{\bf j}^{op}\downarrow S}({\bf Fs}^{+/1})^{op}$$\pi^{S}$$\vphantom{\phantom{({\bf Fs}^{+/1})^{op}}}Set$$F$ i.e. $Lan_{\bf j}(F)(S)=(F\circ\pi^{S},\sigma^{F})$. Note however that if we have a map $f:a\longrightarrow b$ in ${\bf j}^{op}\downarrow S$ i.e. there is a commuting triangle $T_{1}^{*}\vphantom{T_{2}^{*}}$$\vphantom{T_{1}^{*}}T_{2}^{*}$$f^{*}$$S^{*}$$a$$b$ by Lemma 10.3 we can take the inner-outer factorizations of both $a=a^{\prime};(a^{\prime\prime})^{*}$ and $b=b^{\prime};(b^{\prime\prime})^{*}$, with $a^{\prime}$ and $b^{\prime}$ inner. Hence, again by Lemma 10.3, there is a morphism $f^{\prime}:a^{\prime}\longrightarrow b^{\prime}$ which must be an iso. In this way we get a commuting diagram $T_{1}^{*}\vphantom{T_{2}^{*}}$$\vphantom{T_{1}^{*}}T_{2}^{*}$$f^{*}$$T_{3}^{*}\vphantom{T_{4}^{*}}$$\vphantom{T_{3}^{*}}T_{4}^{*}$$(a^{\prime\prime})^{*}$$(b^{\prime\prime})^{*}$$S^{*}$$a$$b$$a^{\prime}$$b^{\prime}$$(f^{\prime})^{*}$ which correspond to the following part of the colimiting cocone $F(T_{1})\vphantom{F(T_{2})}$$\vphantom{F(T_{1})}F(T_{2})$$F(f)$$G(T_{3})\vphantom{F(T_{4})}$$\vphantom{G(T_{3})}F(T_{4})$$F(a^{\prime\prime})$$F(b^{\prime\prime})$$\sigma^{F}_{a}$$\sigma^{F}_{b}$$Lan_{\bf j}(F)(S)^{*}$$\sigma^{F}_{a^{\prime}}$$\sigma^{F}_{b^{\prime}}$$F(f^{\prime})$ Thus if there is a morphism between two objects in ${\bf j}^{op}\downarrow S$ $f:a\rightarrow b$, we have a commuting diagram $a\vphantom{b}$$\vphantom{a}b$$f$$a^{\prime}$$a^{\prime\prime}$$f^{\prime};b^{\prime\prime}$ with $a^{\prime}$ being the inner part of both $a$ and $b$. Otherwise there are no comparison maps. But this say, that in fact $Lan_{\bf j}(F)(S^{*})=\coprod_{a:S^{*}\rightarrow T^{*}\;{\rm inner}}\,F(T)\vphantom{F(T)}$$\vphantom{Lan_{\bf j}(F)(S^{*})=\coprod_{a:S^{*}\rightarrow T^{*}\;{\rm inner}}\,F(T)}F(T)$$\kappa^{S^{*}}_{a}$ where the coproduct is taken over all (up to iso) inner maps with the domain $S^{*}$, with the coprojections as shown. If $h:S_{1}^{*}\longrightarrow S_{2}^{*}$ is an ${\omega}$-functor and $a_{2}:S^{*}_{2}\rightarrow T^{*}_{2}$ is inner, by Lemma 10.3, we can form a diagram $T^{*}_{1}\vphantom{T^{*}_{2}}$$\vphantom{T^{*}_{1}}T^{*}_{2}$$(h^{\prime})^{*}$$S^{*}_{1}\vphantom{S^{*}_{2}}$$\vphantom{S^{*}_{1}}S^{*}_{2}$$h$$a_{1}$$a_{2}$ with $a_{1}$ inner and $(h^{\prime})^{*}$ outer. $Lan_{\bf j}(h)$ is so defined that, for any $h^{\prime}$, $a_{1}$, $a_{2}$ as above, the diagram $Lan_{\bf j}(F)(S_{1}^{*})=\coprod_{a_{1}:S_{1}^{*}\rightarrow T_{1}^{*}\,{\rm inner}}\,F(T_{1})\vphantom{F(T_{1})}$$\vphantom{Lan_{\bf j}(F)(S_{1}^{*})=\coprod_{a_{1}:S_{1}^{*}\rightarrow T_{1}^{*}\,{\rm inner}}\,F(T_{1})}F(T_{1})$$\kappa^{S_{1}^{*}}_{a_{1}}$$Lan_{\bf j}(F)(S_{2}^{*})=\coprod_{a_{2}:S_{2}^{*}\rightarrow T_{2}^{*}\,{\rm inner}}\,F(T_{2})\vphantom{F(T_{2})}$$\vphantom{Lan_{\bf j}(F)(S_{2}^{*})=\coprod_{a_{2}:S_{2}^{*}\rightarrow T_{2}^{*}\,{\rm inner}}\,F(T_{2})}F(T_{2})$$\kappa^{S_{2}^{*}}_{a_{2}}$$Lan_{\bf j}(F)(h)$$F(h^{\prime})$ commutes. This ends the description of the functor $Lan_{\bf j}$. We shall use this description to show that $Lan_{\bf j}(F)$ preserves the special pushouts. So assume that $S_{1}$ and $S_{2}$ are positive face structures such that ${\bf c}^{(k)}(S_{1})={\bf d}^{(k)}(S_{2})$, i.e. we have a pushout ${\bf c}^{(k)}(S_{1})\vphantom{S_{2}}$$\vphantom{{\bf c}^{(k)}(S_{1})}S_{2}$${\bf d}^{(k)}_{S_{2}}$$S_{1}\vphantom{S_{1}{;}_{k}S_{2}}$$\vphantom{S_{1}}S_{1}{;}_{k}S_{2}$$\kappa_{1}$${\bf c}^{(k)}_{S_{1}}$$\kappa_{2}$ in ${\bf Fs}^{+/1}$. We need to show that the square $Lan_{\bf j}(F)({\bf c}^{(k)}(S_{1}))\vphantom{Lan_{\bf j}(F)(S_{2})}$$\vphantom{Lan_{\bf j}(F)({\bf c}^{(k)}(S_{1}))}Lan_{\bf j}(F)(S_{2})$$Lan_{\bf j}(F)({\bf d}^{(k)}_{S_{2}})$$Lan_{\bf j}(F)(S_{1})\vphantom{Lan_{\bf j}(F)(S_{1}{;}_{k}S_{2})}$$\vphantom{Lan_{\bf j}(F)(S_{1})}Lan_{\bf j}(F)(S_{1}{;}_{k}S_{2})$$Lan_{\bf j}(F)(\kappa_{1})$$Lan_{\bf j}(F)({\bf c}^{(k)}_{S_{1}})$$Lan_{\bf j}(F)(\kappa_{2})$ is a pullback in $Set$, i.e. that the square $\coprod_{a:({\bf c}^{(k)}(S_{1}))^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)\vphantom{\coprod_{a:S_{2}^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)}$$\vphantom{\coprod_{a:({\bf c}^{(k)}(S_{1}))^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)}\coprod_{a:S_{2}^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)$$Lan_{\bf j}(F)({\bf d}^{(k)}_{S_{2}})$$\coprod_{a:S_{1}^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)\vphantom{\coprod_{a:(S_{1}{;}_{k}S_{2})^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)}$$\vphantom{\coprod_{a:S_{1}^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)}\coprod_{a:(S_{1}{;}_{k}S_{2})^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)$$Lan_{\bf j}(F)(\kappa_{1})$$Lan_{\bf j}(F)({\bf c}^{(k)}_{S_{1}})$$Lan_{\bf j}(F)(\kappa_{2})$ is a pullback in $Set$. So suppose we have $x_{1}\in F(T_{1})\vphantom{\coprod_{a:S_{1}^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)}$$\vphantom{x_{1}\in F(T_{1})}\coprod_{a:S_{1}^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)$$\kappa^{s^{*}}_{a_{1}}$$x_{2}\in F(T_{1})\vphantom{\coprod_{a:S_{1}^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)}$$\vphantom{x_{2}\in F(T_{1})}\coprod_{a:S_{1}^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)$$\kappa^{s^{*}}_{a_{2}}$ such that $Lan_{\bf j}(F)({\bf c}^{(k)}_{S_{1}})(x_{1})=Lan_{\bf j}(F)({\bf d}^{(k)}_{S_{2}})(x_{2})$ i.e. we have a commutative diagram in ${\bf Fs}^{+/1}$ $T_{1}^{*}\vphantom{T^{*}}$$\vphantom{T_{1}^{*}}T^{*}$$f_{1}^{*}$$S_{1}^{*}\vphantom{({\bf c}^{(k)}(S_{1}))^{*}=({\bf d}^{(k)}(S_{2}))^{*}}$$\vphantom{S_{1}^{*}}({\bf c}^{(k)}(S_{1}))^{*}=({\bf d}^{(k)}(S_{2}))^{*}$$({\bf c}^{(k)}_{S_{1}})^{*}$$a_{1}$$\vphantom{\phantom{T^{*}}}T_{2}^{*}$$f_{2}^{*}$$\vphantom{\phantom{({\bf c}^{(k)}(S_{1}))^{*}=({\bf d}^{(k)}(S_{2}))^{*}}}T_{2}^{*}$$({\bf d}^{(k)}_{S_{2}})^{*}$$a_{0}$$a_{2}$ such that $F(f_{1})(x_{1})=F(f_{2})(x_{2}).$ By Proposition 10.2, ${\bf c}^{(k)}a_{1}=a_{0}={\bf d}^{(k)}a_{2},\;\;\;f_{1}^{*}=({\bf c}^{(k)}_{T_{1}})^{*},\;\;\;f_{2}^{*}=({\bf d}^{(k)}_{T_{2}})^{*}$ and the square $T\vphantom{T_{2}}$$\vphantom{T}T_{2}$$f_{2}$$T_{1}\vphantom{T_{1}{;}_{k}T_{2}}$$\vphantom{T_{1}}T_{1}{;}_{k}T_{2}$$\kappa^{\prime}_{1}$$f_{1}$$\kappa^{\prime}_{2}$ is a special pushout. We have a commuting diagram $T^{*}\vphantom{T_{2}^{*}}$$\vphantom{T^{*}}T_{2}^{*}$${\bf c}^{(k)}(S_{1})^{*}\vphantom{S_{2}^{*}}$$\vphantom{{\bf c}^{(k)}(S_{1})^{*}}S_{2}^{*}$$a_{0}$$T_{1}^{*}\vphantom{(T_{1}{;}_{k}T_{2})^{*}}$$\vphantom{T_{1}^{*}}(T_{1}{;}_{k}T_{2})^{*}$$S_{1}^{*}\vphantom{(S_{1}{;}_{k}S_{2})^{*}}$$\vphantom{S_{1}^{*}}(S_{1}{;}_{k}S_{2})^{*}$$a_{1}+a_{2}$$a_{1}$$a_{2}$ where the bottom square is the above square, the top square is obvious. All the horizontal morphisms are outer. Since $a_{1}$ and $a_{2}$ are inner, $a_{1}(S_{1})=T_{1}$ and $a_{2}(S_{2})=T_{2}$ we have $(a_{1}+a_{2})(S_{1};_{k}S_{2})=a_{1}(S_{1});_{k}a_{2}(S_{2})=T_{1};_{k}T_{2}.$ i.e. $a_{1}+a_{2}:(S_{1}{;}_{k}S_{2})*\longrightarrow(T_{1}{;}_{k}T_{2})$ is inner, as well. So in fact all vertical morphisms in the above diagram are inner. Suppose we have another inner map $u$ and outer maps $\kappa^{\prime\prime}_{1}$, $\kappa^{\prime\prime}_{2}$ so that the squares $T_{1}^{*}\vphantom{U^{*}}$$\vphantom{T_{1}^{*}}U^{*}$${\kappa^{\prime\prime}_{1}}^{*}$$S_{1}^{*}\vphantom{(S_{1}{;}_{k}S_{2})^{*}}$$\vphantom{S_{1}^{*}}(S_{1}{;}_{k}S_{2})^{*}$$\kappa_{1}^{*}$$a_{1}$$u$$\vphantom{\phantom{U}}T_{2}^{*}$${\kappa^{\prime\prime}_{2}}^{*}$$\vphantom{\phantom{(S_{1}{;}_{k}S_{2})^{*}}}S_{2}^{*}$$\kappa_{2}^{*}$$a_{2}$ commute. A diagram chasing shows that $a_{0};f^{*}_{1};{\kappa^{\prime\prime}_{1}}^{*}=a_{0};f^{*}_{2};{\kappa^{\prime\prime}_{2}}^{*}$ As inner-outer factorization is essentially unique, it follows that $f^{*}_{1};{\kappa^{\prime\prime}_{1}}^{*}=f^{*}_{2};{\kappa^{\prime\prime}_{2}}^{*}$ By the universal property of the pushout $(T_{1}{;}_{k}T_{2})^{*}$ we have an ${\omega}$-functor $v:(T_{1}{;}_{k}T_{2})^{*}\longrightarrow U^{*}$ such that $\kappa^{\prime\prime}_{1}=\kappa^{\prime}_{1};u,\;\;\kappa^{\prime\prime}_{2}=\kappa^{\prime}_{2};u$ Then again by a diagram chasing we get $\kappa_{i};u=\kappa_{i};(a_{1}+a_{2});v$ for $i=1,2$. Hence by universal property of the pushout $(S_{1}{;}_{k}S_{2})^{*}$ we have that $u=(a_{1}+a_{2});v$. But both $u$ and $(a_{1}+a_{2})$ are inner so by uniqueness of factorization, see Lemma 10.3, $v$ must be an iso, as well. This means that if there is an $x\in\coprod_{a:(S_{1}{;}_{k}S_{2})^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)$ such that $Lan_{\bf j}(F)(\kappa_{1})(x)=x_{1},\;\;\;Lan_{\bf j}(F)(\kappa_{2})(x)=x_{2}$ it is necessary that $x$ belongs to the summand of the coproduct with the index $(a_{1}+a_{2})$, i.e. $x\in F(T_{1};_{k}T_{2})\vphantom{\coprod_{a:(S_{1};_{k}S_{2})^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)}$$\vphantom{x\in F(T_{1};_{k}T_{2})}\coprod_{a:(S_{1};_{k}S_{2})^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)$$\kappa^{(S_{1};_{k}S_{2})^{*}}_{(a_{1}+a_{2})}$ But $F$ sends special pushouts in ${\bf Fs}^{+/1}$ to pullbacks in $Set$ so the square $F(T)\vphantom{F(T_{2})}$$\vphantom{F(T)}F(T_{2})$$F(f_{2})$$F(T_{1})\vphantom{F(T_{1}{;}_{k}T_{2})}$$\vphantom{F(T_{1})}F(T_{1}{;}_{k}T_{2})$$F(\kappa^{\prime}_{1})$$F(f_{1})$$F(\kappa^{\prime}_{2})$ is a pullback in $Set$. Thus indeed there is a unique $x\in F(T_{1}{;}_{k}T_{2})`F(T)$ such that $F(\kappa^{\prime}_{i})(x)=x_{i}$ for $i=1,2$. This shows that $Lan_{\bf j}(F)$ preserves special pushouts. $~{}~{}\Box$ In the proof above we have described the left Kan extension $Lan_{\bf j}$ in a special way in terms coproducts. As it is a very important property the corollary below we restate this description explicitly, for the record. ###### Corollary 16.2 The functor $Lan_{\bf j}:sPb(({\bf Fs}^{+/1})^{op},Set)\longrightarrow sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ is defined for $F\in sPb(({\bf Fs}^{+/1})^{op},Set)$ as follows. For a positive face structure $S$ we have $Lan_{\bf j}(F)(S^{*})=\coprod_{a:S^{*}\rightarrow T^{*}\;{\rm inner}}\,F(T)\vphantom{F(T)}$$\vphantom{Lan_{\bf j}(F)(S^{*})=\coprod_{a:S^{*}\rightarrow T^{*}\;{\rm inner}}\,F(T)}F(T)$$\kappa^{S^{*}}_{a}$ where coproduct is taken over all up to iso inner maps in ${{\cal C}types}^{+/1}_{\omega}$ with the domain $S^{*}$, with the coprojections as shown. If $h:S_{1}^{*}\longrightarrow S_{2}^{*}$ is an ${\omega}$-functor and $a_{2}:S^{*}_{2}\rightarrow T^{*}_{2}$ is inner, by Lemma 10.3, we can form a diagram $T^{*}_{1}\vphantom{T^{*}_{2}}$$\vphantom{T^{*}_{1}}T^{*}_{2}$$(h^{\prime})^{*}$$S^{*}_{1}\vphantom{S^{*}_{2}}$$\vphantom{S^{*}_{1}}S^{*}_{2}$$h$$a_{1}$$a_{2}$ with $a_{1}$ inner and $h^{\prime}$ a face structures map, i.e. the map $(h^{\prime})^{*}$ is outer. $Lan_{\bf j}(h)$ is so defined that, for any $h$, $h^{\prime}$, $a_{1}$, $a_{2}$ as above, the diagram $Lan_{\bf j}(F)(S_{1}^{*})=\coprod_{a_{1}:S_{1}^{*}\rightarrow T_{1}^{*}\,{\rm inner}}\,F(T_{1})\vphantom{F(T_{1})}$$\vphantom{Lan_{\bf j}(F)(S_{1}^{*})=\coprod_{a_{1}:S_{1}^{*}\rightarrow T_{1}^{*}\,{\rm inner}}\,F(T_{1})}F(T_{1})$$\kappa^{S_{1}^{*}}_{a_{1}}$$Lan_{\bf j}(F)(S_{2}^{*})=\coprod_{a_{2}:S_{2}^{*}\rightarrow T_{2}^{*}\,{\rm inner}}\,F(T_{2})\vphantom{F(T_{2})}$$\vphantom{Lan_{\bf j}(F)(S_{2}^{*})=\coprod_{a_{2}:S_{2}^{*}\rightarrow T_{2}^{*}\,{\rm inner}}\,F(T_{2})}F(T_{2})$$\kappa^{S_{2}^{*}}_{a_{2}}$$Lan_{\bf j}(F)(h)$$F(h^{\prime})$ commutes.$~{}~{}\Box$ ###### Lemma 16.3 The following square ${{\omega}Cat}$$sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$${\bf Comp}^{+/1}$$sPb(({\bf Fs}^{+/1})^{op},Set)$$\widehat{(-)}$$\widehat{(-)}$$Lan_{\bf j}$${\bf e}$ commutes, up to an isomorphism. Proof. We shall define two natural transformations $\varphi$ and $\psi$ which are mutually inverse, i.e. for a positive-to-one computad $Q$ we define $Lan_{\bf j}\,({\bf Comp}^{+/1}((-)^{*},Q))\vphantom{{{\omega}Cat}((-)^{*},Q)}$$\vphantom{Lan_{\bf j}\,({\bf Comp}^{+/1}((-)^{*},Q))}{{\omega}Cat}((-)^{*},Q)$$\varphi_{Q}$$\psi_{Q}$ Let $a:S^{*}\longrightarrow T^{*}$ be an inner map and $f:T^{*}\rightarrow A$ be a computad map, i.e. $g$ is in the following coproduct $g\in{\bf Comp}^{+/1}(T^{*},Q)\vphantom{\coprod_{S^{*}\rightarrow R^{*}\,{\rm inner}}\,{\bf Comp}^{+/1}(R^{*},Q)}$$\vphantom{g\in{\bf Comp}^{+/1}(T^{*},Q)}\coprod_{S^{*}\rightarrow R^{*}\,{\rm inner}}\,{\bf Comp}^{+/1}(R^{*},Q)$$\kappa^{S^{*}}_{a}$ Then we put $\varphi_{Q}(g)=a;g.$ On the other hand, for an ${\omega}$-functor $f:S^{*}\rightarrow Q\in{{\omega}Cat}(S^{*},Q)$, by Proposition 12.2.4, we have a factorization $S^{*}\vphantom{Q}$$\vphantom{S^{*}}Q$$f$$T_{f(S)}^{*}$$f^{in}$$\tau_{f(S)}$ Then we put $\psi_{Q}(f)=\tau_{f(S)}\in{\bf Comp}^{+/1}(T_{f(S)}^{*},Q)\vphantom{\coprod_{S^{*}\rightarrow R^{*}\,{\rm inner}}\,{\bf Comp}^{+/1}(R^{*},Q)}$$\vphantom{\psi_{Q}(f)=\tau_{f(S)}\in{\bf Comp}^{+/1}(T_{f(S)}^{*},Q)}\coprod_{S^{*}\rightarrow R^{*}\,{\rm inner}}\,{\bf Comp}^{+/1}(R^{*},Q)$$\kappa^{S^{*}}_{f^{in}}$ The fact that these transformations are mutually inverse follows from the fact that the above factorization is essentially unique. The verifications that these transformations and natural is left for the reader. $~{}~{}\Box$ ###### Proposition 16.4 The functor $Lan_{\bf j}:sPb(({\bf Fs}^{+/1})^{op},Set)\longrightarrow sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ preserves connected limits. Proof. This is an easy consequence of Lemma 16.2, where $Lan_{\bf j}$ is described in terms of pushouts. For preservation of connected limits it is sufficient to show that wide pullbacks and equalizers are preserved. We shall sketch the preservation of the binary pullbacks leaving the details and other cases to the reader. So let $F\vphantom{H}$$\vphantom{F}H$$F\times_{H}G\vphantom{G}$$\vphantom{F\times_{H}G}G$ be a pullback in $sPb(({\bf Fs}^{+/1})^{op},Set)$. Then we have, for any positive face structure $S$, we have $Lan_{\bf j}(F\times_{H}G)(S^{*})=\coprod_{a:S^{*}\rightarrow T^{*}\,{\rm inner}}\,(F\times_{H}G)(T)\cong$ $\cong\coprod_{a:S^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)\times_{H(T)}G(T)\cong$ $\cong\coprod_{a:S^{*}\rightarrow T^{*}\,{\rm inner}}\,F(T)\times_{\coprod_{a:S^{*}\rightarrow T^{*}\,{\rm inner}}\,H(T)}\coprod_{a:S^{*}\rightarrow T^{*}\,{\rm inner}}\,G(T)\cong$ $\cong Lan_{\bf j}(F)(S^{*})\times_{Lan_{\bf j}(H)(S^{*})}Lan_{\bf j}(G)(S^{*})$ as required. $~{}~{}\Box$ From Propositions 13.4, 16.4, Lemma 16.3 and Corollary 15.2 we get immediately ###### Theorem 16.5 The embedding functor ${\bf e}:{\bf Comp}^{+/1}\longrightarrow{{\omega}Cat}$ preserves connected limits.$~{}~{}\Box$ We have ###### Theorem 16.6 The functor ${\bf j}^{*}:sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)\longrightarrow sPb(({\bf Fs}^{+/1})^{op},Set)$ is monadic. Proof. We are going to verify Beck’s conditions for monadicity. As ${\bf j}$ is essentially surjective ${\bf j}^{*}$ is conservative. By Lemma 16.1, the adjunction $Lan_{\bf j}\dashv j^{*}$ restricts to the above categories. So ${\bf j}^{*}$ has a left adjoint. It remains to show that $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ has coequalizers of ${\bf j}^{*}$-contractible coequalizer pairs and that ${\bf j}^{*}$ preserves them. To this aim, let assume that we have a parallel pair of morphisms in $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ $A\vphantom{B}$$\vphantom{A}B$$G$$F$ such that $A((-)^{*})\vphantom{B((-)^{*})}$$\vphantom{A((-)^{*})}B((-)^{*})$$G_{(-)^{*}}$$F_{(-)^{*}}$$t$$\vphantom{\phantom{B((-)^{*})}}Q$$q$$s$ is a split coequalizer in $sPb(({\bf Fs}^{+/1})^{op},Set)$, i.e. the following equations $s;q=1_{Q},\hskip 14.22636ptG_{(-)^{*}};q=F_{(-)^{*}};q,\hskip 14.22636ptt;F_{(-)^{*}}=1_{B((-)^{*})},\hskip 14.22636ptt;G_{(-)^{*}}=q;s$ hold. We are going to construct a special pullbacks preserving functor $C:({{\cal C}types}^{+/1}_{\omega})^{op}\longrightarrow Set$ and a natural transformation $H:B\longrightarrow C$ so that the diagram in $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ $A\vphantom{B}$$\vphantom{A}B$$G$$F$$\vphantom{\phantom{B}}C$$H$ is a coequalizer, and $H_{(-)^{*}}=q$. The functor $C$ on a morphism $f:T_{1}^{*}\longrightarrow T_{2}^{*}$ is defined as in the diagram $B(T_{1}^{*})\vphantom{B(T_{2}^{*})}$$\vphantom{B(T_{1}^{*})}B(T_{2}^{*})$$B(f)$$Q(T_{1})\vphantom{Q(T_{2})}$$\vphantom{Q(T_{1})}Q(T_{2})$$s_{T_{1}}$$q_{T_{2}}$$C(T_{1}^{*})\vphantom{C(T_{2}^{*})}$$\vphantom{C(T_{1}^{*})}C(T_{2}^{*})$$C(f)$$\|$$\|$ i.e. $C(T_{i})=Q(T_{i})$, for $i=1,2$ and $C(f)=s_{T_{1}};B(f);q_{T_{2}}$. The natural transformation $H$ is given by $H_{T^{*}}=q_{T}$ for $T\in{\bf Fs}^{+/1}$. It remains to verify that 1. 1. $C$ is a functor; 2. 2. $H$ is a natural transformation; 3. 3. $C((-)^{*})=Q$; 4. 4. $H_{(-)^{*}}=q$; 5. 5. $C$ preserves the special pullbacks; 6. 6. $H$ is a coequalizer. Ad 1. Let $T_{1}^{*}\vphantom{T_{2}^{*}}$$\vphantom{T_{1}^{*}}T_{2}^{*}$$f$$\vphantom{\phantom{T_{2}^{*}}}T_{3}^{*}$$g$ be a pair of morphisms in ${{\cal C}types}^{+/1}_{\omega}$. We calculate $C(f);C(g)=s_{T_{1}};B(f);q_{T_{2}};s_{T_{2}};B(g);q_{T_{3}}=$ $=s_{T_{1}};B(f);t_{T_{2}};G_{T_{2}^{*}};B(g);q_{T_{3}}=$ $=s_{T_{1}};B(f);t_{T_{2}};A(g);G_{T_{3}^{*}};q_{T_{3}}=$ $=s_{T_{1}};B(f);t_{T_{2}};A(g);F_{T_{3}^{*}};q_{T_{3}}=$ $=s_{T_{1}};B(f);t_{T_{2}};F_{T_{2}^{*}};B(g);q_{T_{3}}=$ $=s_{T_{1}};B(f);t_{T_{2}};F_{T_{2}^{*}};B(g);q_{T_{3}}=$ $=s_{T_{1}};B(f);B(g);q_{T_{3}}=$ $=s_{T_{1}};B(f;g);q_{T_{3}}=C(f;g)$ i.e. $C$ preserves compositions. If $T$ is a positive face structure, we also have $C(1_{T^{*}})=s_{T};B(1_{T^{*}});q_{T}=s_{T};q_{T}=1_{Q(T)}=1_{C(T^{*})}.$ i.e. $C$ preserves identities, as well. Ad 2. Let $f:T^{*}_{2}\longrightarrow T^{*}_{1}$ be a morphism in ${{\cal C}types}^{+/1}_{\omega}$. We have $B(f);H_{T^{*}_{2}}=B(f);q_{T_{2}}=$ $=t_{T_{1}};F(T^{*}_{1});B(f);q_{T_{2}}=$ $=t_{T_{1}};A(f);F(T^{*}_{2});q_{T_{2}}=$ $=t_{T_{1}};A(f);G(T^{*}_{2});q_{T_{2}}=$ $=t_{T_{1}};G(T^{*}_{1});B(f);q_{T_{2}}=$ $=q_{T_{1}};s_{T_{1}};B(f);q_{T_{2}}=$ $=q_{T_{1}};C(f)=H_{T^{*}_{1}};C(f)$ i.e. $H$ is a natural transformation. Ad 3. Let $f:T_{2}\longrightarrow T_{1}$ be a morphism in ${\bf Fs}^{+/1}$. Thus $q$ is natural with respect to $f$. So we have $C(f^{*})=s_{T_{1}};B(f^{*});q_{T_{2}}=s_{T_{1}};q_{T_{1}};Q(f)=1_{T_{1}};Q(f)=Q(f)$ i.e. $C_{(-)^{*}}=Q$. Note that we still don’t know that $C$ is in $spPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$. Ad 4. $H_{(-)^{*}}=q$ holds by definition. Ad 5. Since special pullbacks involve only the outer morphisms (i.e. those that comes from ${\bf Fs}^{+/1}$), and $Q$ preserves special pullbacks so does $C$. Ad 6. Finally, we shall show that $H$ is a coequalizer. Let $p:B\longrightarrow Z$ be a natural transformation in $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$ such that $pF=pG$. We put $k=s;p:C\longrightarrow Z$ , so that we have a diagram $A\vphantom{B}$$\vphantom{A}B$$G$$F$$\vphantom{\phantom{B}}C$$H$$Z$$k=s;p$$p$ We need to verify that $k$ is a natural transformation in $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$, such that $p=H;k$. Then, the uniqueness of $k$ will follows from the fact that $q$ is a split epi. Let $f:T^{*}_{2}\longrightarrow T^{*}_{1}$ be a morphism in ${{\cal C}types}^{+/1}_{\omega}$. Then $C(f);k_{T^{*}_{2}}=s_{T_{1}};B(f);q_{T_{2}};k_{T^{*}_{2}}=$ $=s_{T_{1}};B(f);q_{T_{2}};s_{T_{2}};p_{T^{*}_{2}}=$ $=s_{T_{1}};B(f);t_{T_{2}};G_{T_{2}^{*}};p_{T^{*}_{2}}=$ $=s_{T_{1}};B(f);t_{T_{2}};F_{T_{2}^{*}};p_{T^{*}_{2}}=$ $=s_{T_{1}};B(f);p_{T^{*}_{2}}=$ $=s_{T_{1}};p_{T^{*}_{1}};D(f)=k_{T^{*}_{1}};D(f)$ i.e. $k$ a natural transformation and hence $H$ is indeed a coequalizer of $F$ and $G$ in $sPb(({{\cal C}types}^{+/1}_{\omega})^{op},Set)$, as required. $~{}~{}\Box$ Combining the above theorem with Corollaries 13.5 and 15.2 we get ###### Theorem 16.7 The nerve functor $\widehat{(-)}:{{\omega}Cat}\longrightarrow sPb(({\bf Fs}^{+/1})^{op},Set)$ sending the ${\omega}$-category $C$ to the presheaf ${{\omega}Cat}((-)^{*},C):({\bf Fs}^{+/1})^{op}\longrightarrow Set$ is monadic. ## 17 Appendix A definition of the positive-to-one computads and the comma categories The notion of a computad was introduced by Ross Street. We repeat this definition for a subcategory ${\bf Comp}^{+/1}$ of the category of all computads that have indeterminates of a special shape, namely their codomains are again indeterminates and their domains are not identities. We use this opportunity to introduce the notation used in the paper. In order to define ${\bf Comp}^{+/1}$ we define two sequences of categories ${{\bf Comp}^{+/1}_{n}}$ and ${\bf Comma}^{+/1}_{n}$. 1. 1. For $n=0$, the categories ${{\bf Comp}^{+/1}_{n}}$, ${\bf Comma}^{+/1}_{n}$ are just $Set$, and the functor $\overline{(-)}^{n}:{\bf Comma}^{+/1}_{n}\longrightarrow{{\bf Comp}^{+/1}_{n}}$ is the identity. 2. 2. For $n=1$, the category ${\bf Comma}^{+/1}_{n}$ is the category of graphs (i.e. 1-graphs) and ${{\bf Comp}^{+/1}_{n}}$ is the category of free ${\omega}$-categories over graphs with morphisms being the functors sending indets (=indeterminates=generators) to indets. 3. 3. Let $n\geq 1$. We define the following functor $\pi^{+1}_{n}:{{\bf Comp}^{+/1}_{n}}\longrightarrow Set$ such that $\pi^{+1}_{n}(P)=\\{(a,b):a\in(P_{n}-\iota(P_{n-1}),\;b\in|P|_{n},\;d(a)=d(b),\;c(a)=c(b)\\}$ i.e. $\pi^{+1}_{n}(P)$ consists of those parallel pairs $(a,b)$ of $n$-cells of $P$ such that $a$ is not an identity and $b$ is an indet. On morphisms $\pi^{+1}_{n}$ is defined in the obvious way. We define ${\bf Comma}^{+/1}_{n+1}$ to be equal to the comma category $Set\downarrow\pi^{+1}_{n}$. So we have a diagram ${{\bf Comp}^{+/1}_{n}}\vphantom{Set}$$\vphantom{{{\bf Comp}^{+/1}_{n}}}Set$$\pi^{+1}_{n}$${\bf Comma}^{+/1}_{n+1}$$(-)_{\leq n}$$|-|_{n+1}$$\mu\Downarrow$ 4. 4. For $n\geq 1$, we can define also a functor $\pi_{n}:{n{\bf Cat}}\longrightarrow Set$ such that $\pi_{n}(C)=\\{(a,b):a,b\in C_{n},\;d(a)=d(b),\;c(a)=c(b)\\}$ i.e. $\pi_{n}(C)$ consists of all parallel pairs $(a,b)$ of $n$-cells of the $n$-category $C$. We define ${\bf Comma}_{n+1}$ to be equal to the comma category $Set\downarrow\pi_{n}$. We often denote objects of ${\bf Comma}_{n+1}$ as quadruples $C=(|C|_{n+1},C_{\leq n},d,c)$, where $C_{\leq n}$ is an $n$-category, $|C|_{n+1}$ is a set and $(d,c):|C|_{n+1}\longrightarrow\pi_{n}(C_{\leq n})$ is a function. Clearly, we have a forgetful functor ${\cal U}_{n+1}:(n+1){\bf Cat}\longrightarrow{\bf Comma}_{n+1}$ such that for an $(n+1)$-category $A$ we have ${\cal U}_{n+1}(A)=(A_{n+1},A_{\leq n},d,c)$ which forgets the structure of compositions and identities at the top level. This functor has a left adjoint ${\cal F}_{n+1}:{\bf Comma}_{n+1}\longrightarrow(n+1){\bf Cat}$ The category ${\cal F}_{n+1}(|B|_{n+1},B_{\leq n},d,c)$ is said to be a free extension of the $n$-category $B$ by the indets $|B|_{n}$. The category of positive-to-one $(n+1)$-computads ${{\bf Comp}^{+/1}_{n+1}}$ is a subcategory of $(n+1){\bf Cat}$ whose objects are free extensions of positive-to-one $n$-computads. The morphisms in ${{\bf Comp}^{+/1}_{n+1}}$ are $(n+1)$-functors that sends indets to indets. Thus the functor ${\cal F}_{n+1}$ restricts to an equivalence of categories ${\cal F}^{+1}_{n+1}:{\bf Comma}^{+/1}_{n+1}\longrightarrow{{\bf Comp}^{+/1}_{n+1}},$ its essential inverse will be denoted by $\|-\|_{n+1}:{{\bf Comp}^{+/1}_{n+1}}\longrightarrow{\bf Comma}^{+/1}_{n+1}.$ Thus for an $(n+1)$-computad $P$ we have $\|P\|_{n}=(P_{\leq n},|P|_{n+1},d,c)$. 5. 5. The category ${\bf Comp}^{+/1}$ is the category of such ${\omega}$-categories $P$, that for every $n\in{\omega}$, $P_{\leq n}$ is a positive-to-one $n$-computad, and whose morphisms are ${\omega}$-functors sending indets to indets. For $n\in{\omega}$, we have functors $|-|_{n}:{\bf Comp}^{+/1}\longrightarrow Set$ associating to computads their $n$-indets, i.e. $f:A\longrightarrow B\mapsto|f|_{n}:|A|_{n}\longrightarrow|B|_{n},$ they all preserve colimits. Moreover we have a functor $|-|:{\bf Comp}^{+/1}\longrightarrow Set$ associating to computads all their indets, i.e. $f:A\longrightarrow B\mapsto|f|:|A|\longrightarrow|B|,$ where $|A|=\coprod_{n\in{\omega}}|A|_{n}.$ It also preserves colimits and moreover it is faithful. 6. 6. We have a functor $(-)_{\leq n}:{{\omega}Cat}\longrightarrow{n{\bf Cat}}$ such that $f:A\longrightarrow B\mapsto f_{\leq k}:A_{\leq k}\longrightarrow B_{\leq k}$ with $k\in{\omega}$, it preserves limits and colimits. ## References * [BD] J. Baez, J. Dolan, Higher-dimensional algebra II: n-Categories and the algebra of opetopes. Advances in Math. 135 (1998), pp. 145-206. * [B] M. Batanin, Monoidal globular categories as a natural environment for the theory of weak n-categories. Advances in Math. 136 (1998), 39-103. * [Be] C. Berger, A Cellular Nerve for Higher Categories. Adv. in Mathematics 169, (2002), pp. 118-175. * [HMP] C. Hermida, M. Makkai, J. Power, On weak higher dimensional categories, I Parts 1,2,3, J. Pure and Applied Alg. 153 (2000), pp. 221-246, 157 (2001), pp. 247-277, 166 (2002), pp. 83-104. * [J] A. Joyal, Disks, Duality and $\Theta$-categories. Preprint, (1997). * [CWM] S. MacLane, Categories for Working Mathematician, Springer-Verlag, New York, (1971). * [SGL] S. MacLane, I.Moerdijk, Sheaves in Geometry and Logic: a first introduction to topos theory, Springer-Verlag, New York, (1992). * [MM] M.Makkai, The multitopic omega-category of all multitopic omega-categories. Preprint (1999). * [MZ] M. Makkai and M. Zawadowski, Disks and duality. TAC 8(7), 2001, pp. 114-243. ## Index * category * ${\bf Comma}^{+/1}_{n}$, 74 * ${\bf Comp}^{+/1}$, 74 * ${{\bf Comp}^{+/1}_{n}}$, 74 * ${{\cal C}types}^{+/1}$, 42 * ${{\cal C}types}^{+/1}_{\omega}$, 42 * free extension of -, 75 * ${\bf Fs}^{+/1}$, 8 * ${\bf Hg}^{+1}$, 6 * c${\bf nFs}^{+/1}$, 8 * ${{\omega}Cat}$, 2 * ${\bf pFs}^{+/1}$, 8 * ${\bf wFs}^{+/1}$, 8 * cell * description of a -, 40 * colimit * special, 26 * composition * total - map, 55 * decomposition * proper, 34 * description of a cell, 40 * disjointness, 3, 8 * face * internal, 7 * saddle -, 34 * unary, 7 * face structure * $n$\- - , 8 * positive * normal -, 8 * principal -, 8 * positive -, 7 * weak positive -, 8 * free extension, 75 * globular * k-, 56 * globularity, 3, 8 * globularization * k-, 56 * hypergraph * positive -, 6 * positive - morphism, 6 * linearity * $\delta$\- -, 8 * $\gamma$\- -, 8 * pencil -, 8 * map * inner -, 42 * outer -, 42 * total composition -, 55 * morphism * positive hypergraph -, 6 * order * lower, 3, 7 * upper, 3, 7 * path * in $X$, 10 * lower, 7 * maximal, 13 * upper, 7 * pencil * linearity, 3, 8 * positive * computype, 42 * principal pushout, 55 * pushout * principal, 55 * special - , 26 * simple extension, 30 * size * of positive face structure, 8 * special * colimit, 26 * pushout, 26 * strictness, 3, 8

arxiv-papers

2024-09-04T02:48:53.094173

0708.2697

DAMTP-2007-77 Epsilon Expansion for Multicritical Fixed Points and Exact Renormalisation Group Equations J. O’Dwyer111jpo23@damtp.cam.ac.uk and H. Osborn222ho@damtp.cam.ac.uk Department of Applied Mathematics and Theoretical Physics, Wilberforce Road, Cambridge, CB3 0WA, England The Polchinski version of the exact renormalisation group equations is applied to multicritical fixed points, which are present for dimensions between two and four, for scalar theories using both the local potential approximation and its extension, the derivative expansion. The results are compared with the epsilon expansion by showing that the non linear differential equations may be linearised at each multicritical point and the epsilon expansion treated as a perturbative expansion. The results for critical exponents are compared with corresponding epsilon expansion results from standard perturbation theory. The results provide a test for the validity of the local potential approximation and also the derivative expansion. An alternative truncation of the exact RG equation leads to equations which are similar to those found in the derivative expansion but which gives correct results for critical exponents to order $\varepsilon$ and also for the field anomalous dimension to order $\varepsilon^{2}$. An exact marginal operator for the full RG equations is also constructed. PACS:11.10.-z, 11.10.Gh, 64.60.Fr, 64.60Ak, 64.60.Kw, 68.35.Rh Keywords:Epsilon Expansion, Exact Renormalisation Group, Multicritical points. ## 1 Introduction A fundamental development in quantum field theory was understanding the role of the renormalisation scale induced by the presence of a cut off, or any other regularisation ensuring finiteness, and the associated flow of the couplings of the theory under changes of scale. The RG flow equations therefore reflect the essential arbitrariness of the renormalisation scale. Nevertheless the global nature of the renormalisation flows in the space of couplings and the various fixed points that are present are crucial properties of any particular quantum field theory of physical interest, although in general their analysis is beyond the scope of conventional perturbation theory. Since the time of Wilson [1, 2, 3, 4] various exact RG equations have been formulated which in principle transcend perturbation theory and allow the determination of fixed points and also the critical exponents that determine the flow of the couplings in the neighbourhood of fixed points, for recent reviews see [5, 6, 7, 8] and for a critical discussion [9]. For theories involving just scalar fields, when a cut off function is introduced in the quadratic part of the action, these have been extensively explored. At a rigorous level they may be used to provide an alternative proof of the renormalisability of such theories [4, 10]. On the other hand outside the perturbative domain it is necessary to resort to approximations when the functional differential equations for the RG flow of the effective action, which are in principle exact, are reduced to non linear coupled differential equations which may then be analysed numerically. The simplest approximation is when the effective action, in general a nonlocal functional of local fields, is restricted to a function just of the scalar field without any derivatives, the local potential approximation (LPA) [11]. Beyond the LPA it is possible to consider a derivative expansion to second and potentially higher orders in the number of derivatives. However these approximations are essentially uncontrolled. The resulting equations depend in detail on the form of the cut off function and it is unclear whether there is any systematic procedure for improving, in principle, order by order the accuracy of results for critical exponents which should be independent of the particular form of the cut off. Despite such difficulties the numerical results are often impressive and are in good agreement with other methods of determining critical exponents for appropriate statistical field theories in three dimensions. The LPA is applicable to various different versions of the exact renormalisation group. In general the resulting equations are inequivalent but the LPA for the Polchinski equation [4] with scalar fields, where the cut off dependence can be removed by simple rescalings and so is absent from calculated critical exponents, the results are identical to the LPA ERG equations for the one particle irreducible generating function with a particular smooth cut off function [12, 13]. Expanding the action as an integral over local functions of the fields with increasing numbers of derivatives then at other than zeroth order there is an intrinsic dependence on the cut off in the resulting truncated equations which cannot be removed by redefinitions. For the Polchinski equation this involves at each order just a finite set of parameters which are essentially arbitrary. Nevertheless the basic LPA, yielding a simple nonlinear differential flow equation for a potential $V(\phi)$, encapsulates the essential fixed point structure of such scalar theories. As the dimension $d$ is reduced a new fixed point is generated whenever the operator $\phi^{2n}$, for $n=2,3,\dots$, becomes marginal. In the neighbourhood of each fixed point the flow equations determine various critical exponents which may be compared with results from other calculational methods. A not yet fully realised goal is whether it is possible to improve the LPA, while restricting to just a tractable finite set of coupled partial differential equations but with a systematic prescription for the determination of any parameters present, so as to ensure that results for critical exponents should be quantitatively improved, closer to the results of the particular quantum field theory, for all fixed points. As a possible procedure for understanding how far the LPA and its extensions are valid we consider here the connection with the $\varepsilon$-expansion. As originally shown by Wilson and Fisher [14] this provides a method whereby standard quantum field theory calculations of $\beta$-functions and related anomalous dimensions as a loop expansion in $d=4-\varepsilon$ dimensions may be applied to determine critical exponents for $d=3$ as an asymptotic power expansion in $\varepsilon$. For an extensive discussion in the context of standard quantum field theory see [15]. The $\varepsilon$-expansion can also be obtained directly from exact RG equations, as was the case historically, since for $\varepsilon\to 0$ the equations become linear and the non linear terms may be treated perturbatively. An interesting question is then the extent to which the $\varepsilon$-expansion results are compatible with those from the LPA. Although this has been considered previously we here attempt a systematic discussion in relation to the Polchinski RG equation. Initially this is applied for just the LPA itself but we also consider derivative expansion extensions to see whether any improvements in the domain of joint validity is feasible. A similar discussion for $2<d<4$ is undertaken for the hierarchical RG in [16]. An alternative approximation for the exact RG flow equations is to consider expanding the effective action in terms of translation invariant functions of the basic fields which are eigenfunctions of the linearised RG flow functional differential operator which are referred to as scaling fields [17]. The non linear part of the RG flow equation may then be expanded in this basis. This gives a set of coupled equations which in the simplest approximation is equivalent to the LPA and at the next order is very similar to the derivative approximation. However in this approach the dependence on the cut off function is more controlled and in the $\varepsilon$-expansion it is possible to get the correct result for the critical exponent $\eta$ at order ${\rm O}(\varepsilon^{2})$, unlike in the usual derivative expansion. In this paper in Section 2 we first consider standard perturbative calculations, with the aid of the background field method, for determining critical exponents in the $\varepsilon$-expansion at all multicritical points for a single scalar field. This is applied both for scalar operators with no and also two derivatives. Some higher order results, which involve multi-loop calculations, are obtained in Appendix A. Although the methods used are very different from exact RG calculations they provide results which are useful comparison for later approximations. In Section 3 we consider the LPA. It is shown how at ${\rm O}(\varepsilon)$ the solution for each multicritical point is a single Hermite polynomial, whose coefficient is determined by the nonlinear terms, and at ${\rm O}(\varepsilon^{2})$ it is just a finite sum. In Section 4 the results are worked out in more detail for the first three critical points and graphical comparisons are made between the approximate analytic solution and numerical solutions for various $d$. In 5 the corresponding critical exponents, within the LPA, are found at ${\rm O}(\varepsilon)$ and also ${\rm O}(\varepsilon^{2})$, where they disagree with the perturbative results. The LPA is well known to be of restricted validity, it requires that the critical exponent $\eta$, which is essentially the anomalous dimension of the elementary scalar field, is zero. The derivative expansion attempts to overcome these limitations and we consider this in the context of the $\varepsilon$-expansion in Section 6. The coupled equations now depend on two cut off function dependent constants $A,B$ but they now allow $\eta$ to be determined. The solutions in terms of Hermite polynomials may also be extended to this case with some modifications. Following this in 7 we use these results to determine critical exponents at ${\rm O}(\varepsilon)$ for two classes of scalar operators. For one class the results are the same as in the LPA case and agree with perturbation theory, for the other set of operators which involve derivatives the calculated exponents depend on $A,B$. The scaling field approach based on the exact RG flow equation is considered in Section 8. A similar truncation to the derivative expansion is possible leading to equations which also may be solved simply in the $\varepsilon$-expansion. In this case the dependence on the cut off function resides in various integrals. In special cases these are independent of the precise cut off function and they then determine universal results for critical exponents to ${\rm O}(\varepsilon)$ and also $\eta$ to ${\rm O}(\varepsilon^{2})$. The relevant integrals are discussed in Appendix C where the cut off function independent values are shown to be related to logarithmic divergences in two vertex Feynman integrals. In Section 9 the resulting equations are recast as coupled differential equations which are very similar, although different in detail, to those arising in the derivative expansion. Some more general remarks are contained in a conclusion. In Appendix D we obtain some exact results for perturbations of the full RG flow equations and show how to construct an exact marginal operator. The existence of such an operator, leading to a line of equivalent fixed points, ensures that the RG equations determine $\eta$. ## 2 Perturbation Calculations We here discuss for the purposes of comparison a conventional quantum field theory calculation of critical exponents at multicritical fixed points in the $\varepsilon$-expansion. We initially consider just the basic Lagrangian ${\cal L}={\textstyle\frac{1}{2}}(\partial\phi)^{2}+V(\phi)\,.$ (2.1) To determine the counterterms to ensure a finite theory it is sufficient as usual to consider connected one particle irreducible graphs. We adopt as the basic propagator $G_{0}(x)=\frac{1}{4\pi}\,\frac{\Gamma(\nu)}{\pi^{\nu}}\,\frac{1}{(x^{2})^{\nu}}\,,\qquad\nu={\textstyle\frac{1}{2}}d-1\,,$ (2.2) satisfying $-\partial^{2}G_{0}(x)=\delta^{d}(x)$, and also use a background field approach, following similar methods used for four dimensional theories in [18], where $\phi=\varphi+f\,,$ (2.3) with $f$ the quantum field. Only vacuum graphs are then necessary and since with dimensional regularisation $G_{0}(x)|_{x=0}=0$ no graphs with lines involving a single vertex need be included. At lowest order for the one particle irreducible functional $W$ we have $W_{1}=\sum_{r\geq 2}\frac{1}{2r!}\int{\rm d}^{d}x_{1}\,{\rm d}^{d}x_{2}\,V^{(r)}(\varphi_{1})\,G_{0}(x_{12})^{r}\,V^{(r)}(\varphi_{2})\,,\qquad x_{ij}=x_{i}-x_{j}\,,\ \varphi_{i}=\varphi(x_{i})\,.$ (2.4) To evaluate this we note that $\int{\rm d}^{d}x\,e^{ik\cdot x}\,G_{0}(x)^{r}=\frac{1}{(4\pi)^{r}}\,\frac{\Gamma(\nu)^{r}}{\Gamma(r\nu)}\ \Gamma\big{(}1-(r-1)\nu\big{)}\,\bigg{(}\frac{k^{2}}{4\pi}\bigg{)}^{(r-1)\nu-1}\,.$ (2.5) This has a pole whenever $(r-1)\nu=1,2,\dots$ so that writing $d=d_{n}-\varepsilon\,,\qquad d_{n}=\frac{2n}{n-1}\,,$ (2.6) it is easy to see from (2.5) that for $\varepsilon\to 0$ $\displaystyle G_{0}(x)^{n}\sim{}$ $\displaystyle\frac{2}{\varepsilon}\,\frac{1}{(4\pi)^{n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,\delta^{d}(x)\,,$ (2.7a) $\displaystyle G_{0}(x)^{2n-1}\sim{}$ $\displaystyle\frac{1}{\varepsilon}\,\frac{1}{(4\pi)^{2n}}\,\frac{n-1}{n}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\,\partial^{2}\delta^{d}(x)\,.$ (2.7b) From (2.4), assuming $r\leq 2n$, the necessary counterterms are then $\displaystyle{\cal L}_{\rm{c.t.}1}={}$ $\displaystyle\frac{1}{\varepsilon}\,\frac{1}{(4\pi)^{n}}\,\frac{1}{n!}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,V^{(n)}(\phi)^{2}$ $\displaystyle{}-\frac{1}{\varepsilon}\,\frac{1}{(4\pi)^{2n}}\,\frac{n-1}{(2n)!}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\,V^{(2n)}(\phi)^{2}(\partial\phi)^{2}\,,$ (2.8) where the two terms arise from (2.7a) and (2.7b) at $n-1$ and $2n-2$ loops respectively. The theory is therefore renormalisable for $V(\phi)$ a polynomial of degree $2n$. For $\mu$ a regularisation scale and $\mu^{-\varepsilon}({\cal L}+{\cal L}_{\rm{c.t.}})={\cal L}_{0}={\textstyle\frac{1}{2}}(\partial\phi_{0})^{2}+V_{0}(\phi_{0})$ the usual perturbative $\beta$-functions and anomalous dimensions may be defined by $\mu\frac{{\rm d}}{{\rm d}\mu}{\cal L}\Big{|}_{{\cal L}_{0}}=-{\hat{\gamma}}_{\phi}\,\phi\frac{\partial}{\partial\phi}{\cal L}+{\hat{\beta}}^{V}\\!(\phi)\,,$ (2.9) where ${\hat{\gamma}}_{\phi}=-{\textstyle\frac{1}{2}}\varepsilon+\gamma_{\phi}\,,\qquad{\hat{\beta}}^{V}\\!(\phi)=-\varepsilon\big{(}{\textstyle\frac{1}{2}}\phi V^{\prime}(\phi)-V(\phi)\big{)}+{\beta}^{V}\\!(\phi)\,.$ (2.10) At lowest order (2.8) then gives $\beta_{1}^{V}\\!(\phi)=\frac{1}{(4\pi)^{n}}\,\frac{n-1}{n!}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,V^{(n)}(\phi)^{2}\,.$ (2.11) Assuming $V(\phi)=\frac{1}{(2n)!}\,g\,\phi^{2n}+\sum_{k}\frac{1}{k!}\,g_{k}\,\phi^{k}\,,$ (2.12) then $\gamma_{\phi}=\gamma_{\phi}(g)$ and $\displaystyle{\hat{\beta}}^{V}(\phi)={}$ $\displaystyle\frac{1}{(2n)!}\,{\hat{\beta}}^{g}(g)\,\phi^{2n}+\sum_{k}\frac{1}{k!}\,{\hat{\gamma}}_{k}(g)\,g_{k}\,\phi^{k}+{\rm O}(g_{k}g_{k^{\prime}})\,$ $\displaystyle{\hat{\beta}}^{g}(g)={}$ $\displaystyle-\varepsilon\,(n-1)g+\beta^{g}(g)\,,\qquad{\hat{\gamma}}_{k}(g)=-\varepsilon\,{\textstyle\frac{1}{2}}(k-2)+\gamma_{k}(g)\,.$ (2.13) In general ${\hat{\gamma}}_{1}(g)=-{\hat{\gamma}}_{\phi}(g)$. Applying (2.10) and (2.11) to ${\rm O}(g)$ we then have $\gamma_{k,1}(g)=\frac{2g}{(4\pi)^{n}}\,\frac{n-1}{n!^{2}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,\frac{k!}{(k-n)!}\,.$ (2.14) As usual in the $\varepsilon$-expansion, there is a fixed point such that ${\hat{\beta}}^{g}(g_{*})=0$ with $g_{*}$ expressible perturbatively as a power series in $\varepsilon$. At lowest order it is easy to see from (2.11) and (2.13) that $\frac{g_{*}}{(4\pi)^{n}}\,\frac{(2n)!}{n!^{3}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}=\varepsilon\,,$ (2.15) and the result (2.14) then gives ${\hat{\gamma}}_{k}(g_{*})=-{\textstyle\frac{1}{2}}(k-2)\varepsilon+2(n-1)\,\frac{n!}{(2n)!}\,\frac{k!}{(k-n)!}\,\varepsilon+{\rm O}(\varepsilon^{2})\,.$ (2.16) We may also use (2.8) to determine the anomalous dimension $\gamma_{\phi}(g)$ of the field $\phi$ which is non zero at $2(n-1)$ loops, $\gamma_{\phi,1}(g)=\frac{g^{2}}{(4\pi)^{2n}}\,\frac{2(n-1)^{2}}{(2n)!}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\,.$ (2.17) At the fixed point the critical exponent $\eta=2\gamma_{\phi}(g_{*})$ so that from (2.17) $\eta=4(n-1)^{2}\,\frac{n!^{6}}{(2n)!^{3}}\,\varepsilon^{2}+{\rm O}(\varepsilon^{3})\,.$ (2.18) The $\varepsilon$-expansion at multicritical points using standard quantum field theory was considered in [19] who obtained (2.15) and (2.18). In three dimensions, corresponding to $n=3$, results equivalent to (2.14) were obtained in [20]. The local operators in the basic quantum field include also those with derivatives as well as just $\phi^{k}$. To extend the above discussion we consider in addition to (2.1) ${\cal L}^{Z}={\textstyle\frac{1}{2}}Z(\phi)(\partial\phi)^{2}\,,\qquad Z(\phi)=\sum_{k\geq 2n}\frac{1}{(k-2n)!}\,h_{k}\,\phi^{k-2n}\,,$ (2.19) where we keep only graphs which involve one $Z$-vertex. At the same order as (2.4) we have $\displaystyle W^{Z}_{1}=\sum_{r\geq 2}\frac{1}{2r!}$ $\displaystyle\int{\rm d}^{d}x_{1}\,{\rm d}^{d}x_{2}\,\Big{\\{}Z^{(r)}(\varphi_{1})(\partial\varphi_{1})^{2}\,G_{0}(x_{12})^{r}$ $\displaystyle{}+2r\,Z^{(r-1)}(\varphi_{1})\,\partial\varphi_{1}\cdot\partial G_{0}(x_{12})G_{0}(x_{12})^{r-1}$ $\displaystyle{}+r(r-1)\,Z^{(r-2)}(\varphi_{1})\,\big{(}\partial G_{0}(x_{12})\big{)}^{2}G_{0}(x_{12})^{r-2}\Big{\\}}V^{(r)}(\varphi_{2})\,.$ (2.20) Using (2.7a) with $\displaystyle\partial G_{0}(x)\,G_{0}(x)^{n-1}\sim{}$ $\displaystyle\frac{2}{\varepsilon}\,\frac{1}{n}\,\frac{1}{(4\pi)^{n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,\partial\,\delta^{d}(x)\,,$ $\displaystyle\big{(}\partial G_{0}(x)\big{)}^{2}\,G_{0}(x)^{n-2}\sim{}$ $\displaystyle\frac{2}{\varepsilon}\,\frac{1}{n(n-1)}\,\frac{1}{(4\pi)^{n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,\partial^{2}\delta^{d}(x)\,,$ (2.21) shows that in addition to (2.8) the counterterms are $\displaystyle{\cal L}^{Z}_{\rm{c.t.}1}={}$ $\displaystyle\frac{1}{\varepsilon}\,\frac{1}{(4\pi)^{n}}\,\frac{1}{n!}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,\Big{(}Z^{(n)}(\phi)V^{(n)}(\phi)+Z^{(n-1)}(\phi)V^{(n+1)}(\phi)\Big{)}(\partial\phi)^{2}\,.$ (2.22) With $\mu^{-\varepsilon}({\cal L}^{Z}+{\cal L}^{Z}_{\rm{c.t.}})={\cal L}_{0}^{Z}$ the corresponding $\beta$-function is given by $\mu\frac{{\rm d}}{{\rm d}\mu}{\cal L}^{Z}\Big{|}_{{\cal L}_{0}+{\cal L}_{0}^{Z}}=-{\hat{\gamma}}_{\phi}\,\phi\frac{\partial}{\partial\phi}{\cal L}^{Z}+{\textstyle\frac{1}{2}}\,{\hat{\beta}}^{Z}(\phi)(\partial\phi)^{2}\,,\qquad{\hat{\beta}}^{Z}(\phi)=-\varepsilon{\textstyle\frac{1}{2}}\phi Z^{\prime}(\phi)+{\beta}^{Z}(\phi)\,,$ (2.23) which can be expanded as $\displaystyle{\hat{\beta}}^{Z}(\phi)={}$ $\displaystyle\sum_{k\geq 2n}\frac{1}{(k-2n)!}\,\big{(}{\hat{\gamma}}^{hh}_{k}(g)\,h_{k}+{\gamma}^{hg}_{k}(g)\,g_{k}\big{)}\,\phi^{k-2n}\,,$ $\displaystyle{\hat{\gamma}}^{hh}_{k}(g)=-\varepsilon\,{\textstyle\frac{1}{2}}(k-2n)+\gamma^{hh}_{k}(g)\,.$ (2.24) We must also extend (2.13) to include mixing effects if $k\geq 2n$ to the form ${\hat{\beta}}^{V}(\phi)=\frac{1}{(2n)!}\,{\hat{\beta}}^{g}(g)\,\phi^{2n}+\sum_{k}\frac{1}{k!}\,\big{(}{\hat{\gamma}}^{gg}_{k}(g)\,g_{k}+{\gamma}^{gh}_{k}(g)\,h_{k}\big{)}\,\phi^{k}\,,\quad{\hat{\gamma}}^{gg}_{k}(g)={\hat{\gamma}}_{k}(g)\,.$ (2.25) The terms in ${\cal L}^{Z}$ involving $g_{2n},h_{2n}$ may be absorbed in ${\cal L}$ by a redefinition of $\phi,g$ giving for this special case $\displaystyle{\hat{\gamma}}^{gg}_{2n}(g)={}$ $\displaystyle{\hat{\beta}}^{g}{}^{\prime}(g)-2ng\,{\hat{\gamma}}_{\phi}{}^{\prime}(g)\,,\quad\gamma^{gh}_{2n}(g)=-ng\,{\hat{\gamma}}^{gg}_{2n}(g)+n{\hat{\beta}}^{g}(g)\,,$ $\displaystyle\gamma^{hg}_{2n}(g)={}$ $\displaystyle-2{\hat{\gamma}}_{\phi}{}^{\prime}(g)\,,\hskip 52.63777pt{\hat{\gamma}}^{hh}_{2n}(g)=2ng\,{\hat{\gamma}}_{\phi}{}^{\prime}(g)\,.$ (2.26) The anomalous dimensions of operators at the fixed point are then given by the eigenvalues $\omega_{k,1},\,\omega_{k,2}$ of the matrix $\left(\begin{array}[]{cc}{\hat{\gamma}}^{gg}_{k}(g_{*})&{\gamma}^{gh}_{k}(g_{*})\\\ \vskip 3.0pt\cr{\gamma}^{hg}_{k}(g_{*})&{\hat{\gamma}}^{hh}_{k}(g_{*})\\\ \end{array}\right)\,,$ (2.27) for $k=2n,\dots,4n-3$. It is easy to see from (2.26) that $\omega_{2n,1}={\hat{\beta}}^{g}{}^{\prime}(g_{*}),\ \omega_{2n,2}=0$. More generally, using the equations of motion, $\omega_{k,2}=\omega_{k-2n+1,1}-{\textstyle\frac{1}{2}}\varepsilon+{\textstyle\frac{1}{2}}\eta$. Since $\beta^{Z}_{1}=(n-1)\varepsilon{\cal L}^{Z}_{\rm{c.t.}1}$ then from (2.22) we have at lowest order ${\gamma}^{hh}_{k,1}(g)=\frac{2g}{(4\pi)^{n}}\,\frac{n-1}{n!^{2}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\frac{(k-2n+1)!}{(k-3n+1)!}\,,$ (2.28) and ${\gamma}^{gg}_{k,1}(g)={\gamma}_{k,1}(g)$ as in (2.14). For the off diagonal parts of the anomalous dimension matrix we may note that (2.8) implies a $2(n-1)$-loop contribution to $\beta^{Z}$, $\Delta\beta^{Z}_{1}(\phi)=-\frac{1}{(4\pi)^{2n}}\,\frac{4(n-1)^{2}}{(2n)!}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\,V^{(2n)}(\phi)^{2}\,,$ (2.29) and also we have an additional term at $(n-1)$-loops to that in (2.11), $\Delta\beta_{1}^{V}(\phi)=-\frac{1}{(4\pi)^{n}}\,\frac{1}{(n-2)!}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,V^{(n)}(\phi)^{2}Z(\phi)\,.$ (2.30) Thus the lowest order non zero off diagonal terms are $\displaystyle{\gamma}^{gh}_{k,1}(g)={}$ $\displaystyle-\frac{g^{2}}{(4\pi)^{n}}\,\frac{1}{(n-2)!n!^{2}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,\frac{k!}{(k-2n)!}\,,$ $\displaystyle{\gamma}^{hg}_{k,1}(g)={}$ $\displaystyle-\frac{g}{(4\pi)^{2n}}\,\frac{8(n-1)^{2}}{(2n)!}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\,.$ (2.31) Thus $\gamma^{gh}_{k,1}(g_{*})={\rm O}(\varepsilon^{2})$ and ${\gamma}^{hg}_{k,1}(g_{*})={\rm O}(\varepsilon)$ so that the lowest order eigenvalues of (2.27) are given by (2.16) for $\omega_{k,1}$ and $\omega_{k,2}={\hat{\gamma}}^{hh}_{k}(g_{*})+{\rm O}(\varepsilon^{2})=-{\textstyle\frac{1}{2}}(k-2n)\varepsilon+2(n-1)\,\frac{n!}{(2n)!}\,\frac{(k-2n+1)!}{(k-3n+1)!}\,\varepsilon+{\rm O}(\varepsilon^{2})\,.$ (2.32) Further perturbative results for $\beta^{V}$ and anomalous dimensions are obtained in Appendix A. ## 3 Local Potential Approximation In the LPA, the Polchinski RG equation in $d$ dimensions may be reduced by an appropriate rescaling to the following renormalisation flow for a potential $V(\phi,t)$, $\dot{V}(\phi,t)=V^{\prime\prime}(\phi,t)-V^{\prime}(\phi,t)^{2}+dV(\phi,t)-{\textstyle\frac{1}{2}}(d-2)\,\phi V^{\prime}(\phi,t)\,,$ (3.1) where $t=-\log\Lambda$ and $\Lambda$ is a cut off scale. At a fixed point, $V(\phi,t)\to V_{*}(\phi)$ which solves $V_{*}^{\prime\prime}(\phi)-V_{*}^{\prime}(\phi)^{2}+dV_{*}(\phi)-{\textstyle\frac{1}{2}}(d-2)\,\phi V_{*}^{\prime}(\phi)=0\,.$ (3.2) This equation has been extensively analysed, both numerically and analytically [21, 22]. There are two trivial solutions $V_{*}=0$, the Gaussian fixed point, and $V_{*}(\phi)={\textstyle\frac{1}{2}}\phi^{2}-\frac{1}{d}$, the so called high temperature fixed point. For non trivial solutions even in $\phi$, with $V_{*}{\\!}^{\prime}(0)=0$, and bounded below, it is necessary to fine tune $V_{*}(0)$ to ensure that there are no singularities for all $\phi$. Such solutions appear whenever $d$ is reduced below $2n/(n-1)$ for $n=2,3,\dots$. For our purposes it is convenient to consider a further rescaling by defining $x={\textstyle\frac{1}{2}}(d-2)^{\frac{1}{2}}\phi\,,\qquad v(x,t)=V(\phi,t)\,,\ \ v_{*}(x)=V_{*}(\phi)\,.$ (3.3) Then (3.2) becomes $v_{*}^{\prime\prime}-2xv_{*}^{\prime}+\frac{4d}{d-2}\,v_{*}=v_{*}^{\prime 2}\,,$ (3.4) where $v_{*}^{\prime}=\frac{{\rm d}v_{*}}{{\rm d}x}$. We first consider the linearised form of (3.4), $\frac{{\rm d}^{2}v}{{\rm d}x^{2}}-2x\frac{{\rm d}v}{{\rm d}x}+\frac{4d}{d-2}\,v=0\,,$ (3.5) which becomes a valid approximation when $v(x)$ is small. Global solutions for all $x$ which are bounded by a power for large $x$ are only possible for Hermite polynomials $H_{n}(x)$, $n=0,1,2,\dots$ which satisfy $(D+n)H_{n}=0\,,\qquad D={\textstyle\frac{1}{2}}\frac{{\rm d}^{2}}{{\rm d}x^{2}}-x\frac{{\rm d}}{{\rm d}x}\,.$ (3.6) Hence (3.5) has solutions with the appropriate behaviour $v(x)=cH_{n}(x)$ (3.7) only when $d$ is restricted to $n=\frac{2d}{d-2}.$ (3.8) With the further requirement that $v(x)$ be bounded below, we must restrict to $n$ to be even and $c>0$. Relabelling $n\rightarrow 2n$ we then have solutions of the linearised equation $v(x)=cH_{2n}(x)\quad\hbox{for}\quad d=d_{n}\,,$ (3.9) with $d_{n}$ defined in (2.6). This may then used as a starting point for the analysis of the full non linear fixed point equation. For the non-linear equation (3.4) we consider the case when $d$ is close to the value in (3.9), where the linearised solution holds, and may be written as in (2.6). We therefore seek solutions as $\varepsilon\to 0$ of the form $v_{*}(x)=v_{n}(x)+{\rm O}(\varepsilon^{2})\,,\qquad v_{n}(x)=c_{n}\varepsilon H_{2n}(x)\,.$ (3.10) The right hand side of (3.4) is clearly ${\rm O}(\varepsilon^{2})$. Writing $\frac{4d}{d-2}=4n+2(n-1)^{2}\varepsilon+(n-1)^{3}\varepsilon^{2}+{\rm O}(\varepsilon^{3}).$ (3.11) the leading ${\rm O}(\varepsilon)$ terms on the left hand side of (3.4) are then absent. To ${\rm O}(\varepsilon^{2})$ we may determine $c_{n}$ in (3.10) by noting that the differential operator $D+2n$, which is hermitian with respect to the measure ${\rm d}x\,e^{-x^{2}}$, generates only functions orthogonal to $H_{2n}$ so we must require at this order $2(n-1)^{2}\varepsilon\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{2n}(x)v_{n}(x)=\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{2n}(x)v_{n}{\\!}^{\prime}(x)^{2}\,.$ (3.12) Both sides may be evaluated using the integrals for Hermite polynomials $N_{k}\equiv\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}{H_{k}(x)}^{2}=2^{k}\pi^{\frac{1}{2}}\,k!\,,$ (3.13) and also for three $H_{n}$’s, $\displaystyle G_{klm}\equiv{}$ $\displaystyle\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}{H_{k}(x)}{H_{l}(x)}{H_{m}}(x)=2^{s}\pi^{\frac{1}{2}}\frac{k!\,l!\,m!}{(s-k)!(s-l)!(s-m)!}\,,$ $\displaystyle{}s={\textstyle\frac{1}{2}}(k+l+m)\,,\qquad k,l,m\leq s\,,$ (3.14) with $s$ required to be an integer. On the right hand side of (3.12) $\displaystyle\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{2n}(x)v_{n}{\\!}^{\prime}(x)^{2}$ $\displaystyle=c_{n}{\\!}^{2}\varepsilon^{2}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{2n}(x){H_{2n}{\\!}^{\prime}(x)}^{2}$ $\displaystyle=-{\textstyle\frac{1}{2}}c_{n}{\\!}^{2}\varepsilon^{2}\int_{-\infty}^{\infty}{\rm d}x\,{H_{2n}(x)}^{2}\frac{{\rm d}}{{\rm d}x}\big{(}e^{-x^{2}}H_{2n}{\\!\\!}^{\prime}(x)\big{)}$ $\displaystyle=2nc_{n}{\\!}^{2}\varepsilon^{2}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,{H_{2n}(x)}^{3}=2nc_{n}{\\!}^{2}\varepsilon^{2}G_{2n\,2n\,2n}\,,$ (3.15) using standard identities. Hence (3.12) determines a value for $c_{n}$ $2(n-1)^{2}c_{n}N_{2n}=2nc_{n}{\\!}^{2}\,G_{2n\,2n\,2n}\quad\Rightarrow\quad c_{n}=\frac{(n-1)^{2}}{2^{n}n}\,\frac{n!^{3}}{(2n)!^{2}}\,.$ (3.16) Since $c_{n}>0$ then for relevant solutions in (3.10) we must have $\varepsilon>0$. We may now extend the solution (3.10) to ${\rm O}(\varepsilon^{2})$ by assuming the form $v_{*}(x)=c_{n}\varepsilon H_{2n}(x)+c_{n}\varepsilon^{2}\sum_{m}a_{m}H_{m}(x)+{\rm O}(\varepsilon^{3})\,.$ (3.17) Inserting in (3.4) with (3.11) and keeping only terms which are ${\rm O}(\varepsilon^{2})$ gives $(n-1)^{2}H_{2n}(x)+\sum_{m}(4n-2m)a_{m}H_{m}(x)=c_{n}H_{2n}{\\!}^{\prime}(x)^{2}\,,$ (3.18) and hence using the orthogonality properties of Hermite polynomials we may determine $a_{m}$ for $m\neq 2n$, $N_{m}(4n-2m)a_{m}=c_{n}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{m}(x)H_{2n}{\\!}^{\prime}(x)^{2}\,.$ (3.19) The integral on the right hand side may be calculated by using $H_{n}{\\!}^{\prime}=2nH_{n-1}$ or with the following judicious integrations by parts $\displaystyle\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{m}{H_{2n}{\\!}^{\prime}}^{2}$ $\displaystyle=-\frac{1}{2}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{m}{\\!}^{\prime}\frac{{\rm d}}{{\rm d}x}\big{(}H_{2n}{\\!}^{2}\big{)}-\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,H_{m}H_{2n}\frac{{\rm d}}{{\rm d}x}\big{(}e^{-x^{2}}H_{2n}{\\!}^{\prime}\big{)}$ $\displaystyle=\frac{1}{2}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,\frac{{\rm d}}{{\rm d}x}\big{(}e^{-x^{2}}H_{m}{\\!}^{\prime}\big{)}H_{2n}{\\!}^{2}+4n\int_{-\infty}^{\infty}{\rm d}x\,e^{-x^{2}}H_{m}H_{2n}{\\!}^{2}$ $\displaystyle=(4n-m)\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{m}H_{2n}{\\!}^{2}$ $\displaystyle=(4n-m)\,G_{m\,2n\,2n}\,,$ (3.20) which is zero unless $m$ is even. Hence taking $m=2p$, $\displaystyle a_{2p}$ $\displaystyle=\frac{c_{n}}{N_{2p}}\,\frac{2n-p}{2(n-p)}\,G_{2p\,2n\,2n}$ $\displaystyle=2^{n-p-1}\frac{(n-1)^{2}}{n}\,\frac{2n-p}{n-p}\,\frac{n!^{3}}{p!^{2}(2n-p)!}\,,\quad p=0,1,\dots 2n-1,\,p\neq n\,.$ (3.21) Finally we compute $a_{2n}$ by using (3.17) in (3.4) and imposing orthogonality with $H_{2n}$ to ${\rm O}(\varepsilon^{3})$, just as $c_{n}$ was determined in (3.12). $\displaystyle(n-1)^{2}a_{2n}N_{2n}\ $ $\displaystyle+\ {\textstyle\frac{1}{2}}(n-1)^{3}N_{2n}$ $\displaystyle=c_{n}\sum_{p}a_{2p}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{2n}H_{2n}{\\!}^{\prime}H_{2p}{\\!}^{\prime}$ $\displaystyle=2c_{n}\sum_{p}pa_{2p}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{2n}{\\!}^{2}H_{2p}$ $\displaystyle=2c_{n}\sum_{p\neq n}pa_{2p}G_{2n\,2n\,2p}+2nc_{n}a_{2n}G_{2n\,2n\,2n}.$ (3.22) Using (3.16), (3.22) becomes $\displaystyle(n-1)^{2}a_{2n}N_{2n}={\textstyle\frac{1}{2}}(n-1)^{3}N_{2n}-2c_{n}\sum_{p\neq n}pa_{2p}G_{2n\,2n\,2p},$ (3.23) and so $a_{2n}=\frac{1}{2}(n-1)-\frac{2c_{n}}{(n-1)^{2}}\sum_{p\neq n}p\,a_{2p}\,2^{p}\frac{(2n)!(2p)!}{p!^{2}(2n-p)!},$ (3.24) where we may use (3.16) and (3.21). Since $H_{n}(x)\sim 2^{n}x^{n}$ for large $x$ then for $\varepsilon x^{2}={\rm O}(1)$ the next to leading terms in (3.17) are comparable with the leading $H_{2n}(x)$ term so the $\varepsilon$ expansion for $v_{*}(x)$ breaks down. From (3.21) $a_{2p}<0$ for $p>n$ so that the result given by (3.17), which is a polynomial of degree $4n-2$, is negative for sufficiently large $x$. For the exact solution to (3.4) the nonlinear terms play a crucial role for large $x$ and we have $v_{*}(x)\sim\frac{2}{d-2}x^{2}$. ## 4 Applications in Particular Cases The results obtained for $v_{*}$ obtained above are here considered in more detail for $n=2,3,4$ and compared with results of numerical calculations. In particular $V_{*}(\phi)$ was calculated numerically in the LPA in [23] for these cases and various values of the dimension $d$. The basic approximation from (3.17) is $v_{*}(x)=c_{n}\varepsilon H_{2n}(x)+c_{n}\varepsilon^{2}\sum_{p=0}^{2n-1}{a}_{2p}H_{2p}(x),$ (4.1) where $a_{2p}$ is given by (3.21) and (3.24). The first case of interest is $n=2$, corresponding to the Wilson-Fisher fixed point. From (3.21) we have $k_{2}=\frac{1}{4!^{2}}\,,\qquad{a}_{2p}=\frac{4-p}{2-p}\,\frac{2^{3-p}}{p!^{2}(4-p)!}\,,\quad p=0,1,3\,,$ (4.2) and hence ${a}_{0}=\frac{2}{3}\,,\quad{a}_{2}=2\,,\quad{a}_{6}=-\frac{1}{36}\,.$ (4.3) From these, and (3.24), we have that ${a}_{4}=\frac{3}{2}\,.$ (4.4) Using the following results for Hermite polynomials, $\displaystyle H_{2n}(0)=(-1)^{n}\frac{(2n)!}{n!}\,,\qquad H^{\prime\prime}_{n}(0)=-2nH_{n}(0)\,,$ (4.5) we may then obtain from (4.1) in this case $v_{*}(0)=\frac{1}{48}\varepsilon\left(1+\frac{3}{2}\varepsilon\right)\,,\qquad v^{\prime\prime}_{*}(0)=-\frac{1}{6}\varepsilon\left(1+\frac{7}{4}\varepsilon\right)\,.$ (4.6) As a consequence of (3.3) for the original $V_{*}(\phi)$ we have $V_{*}(0)=v_{*}(0)$ and $\displaystyle V_{*}^{\prime\prime}(0)$ $\displaystyle={\textstyle\frac{1}{4}}(d-2)v_{*}^{\prime\prime}(0)$ $\displaystyle=-\frac{\varepsilon}{12}\left(1+\frac{5\varepsilon}{4}\right)+{\rm O}(\varepsilon^{3})\,.$ (4.7) This results satisfy the consistency check $\frac{V_{*}^{\prime\prime}(0)}{V_{*}(0)}=-4\left(1-\frac{1}{4}\varepsilon\right)=-d\,,$ (4.8) as follows directly from (3.2). As remarked earlier for solutions of the fixed point equation without singularities it is necessary to fine tune $V_{*}^{\prime\prime}(0)$, or equivalently $V_{*}(0)$. In Table 1 we compare the results from (4.7) to ${\rm O}(\varepsilon^{2})$ with those from numerical calculation, as contained in [23], for various $d$. The detailed form of the approximate solution in comparison with numerical results is shown in Figure 1 for various $d$. For small $\varepsilon$ the agreement is good. $d$ | Numerical | $\varepsilon$-expansion at ${\rm O}(\varepsilon^{2})$ ---|---|--- 4 | $0$ | $0\qquad$ 3.9 | $-0.009$ | $-0.009\qquad$ 3.8 | $-0.021$ | $-0.021\qquad$ 3.7 | $-0.035$ | $-0.034\qquad$ 3.6 | $-0.051$ | $-0.050\qquad$ 3.5 | $-0.070$ | $-0.068\qquad$ 3.4 | $-0.092$ | $-0.088\qquad$ 3.3 | $-0.119$ | $-0.109\qquad$ 3.2 | $-0.149$ | $-0.133\qquad$ 3.1 | $-0.186$ | $-0.159\qquad$ 3.0 | $-0.229$ | $-0.188\qquad$ Table 1: Comparison of ERG numerical and analytical results for $V_{*}^{\prime\prime}(0)$ for the case $n=2$. Figure 1: Graphs for $v_{*}(x)$ for $n=2$ from $\varepsilon$-expansion at ${\rm O}(\varepsilon^{2})$ and numerical solution with $v_{*}(0)=-V_{*}^{\prime\prime}(0)/d$ from Table 1 with $d=3.9$, $3.8$, $3.6$, $3.4$, $3.2$, $3.0$. Similarly we consider the multi-critical fixed points obtained for $n=3$ and $n=4$, which correspond to $\varepsilon$-expansions for dimensions $d=3$ and $d=\frac{8}{3}$ respectively. When $n=3$ the results are $k_{3}=\frac{1}{5!^{2}}\,,\qquad{a}_{2p}=\frac{6-p}{3-p}\,\frac{9.2^{7-p}}{p!^{2}(6-p)!}\,,\quad p=0,1,2,4,5\,.$ (4.9) Hence ${a}_{0}=\frac{16}{5}\,,\quad{a}_{2}=12\,,\quad{a}_{4}=12\,,\quad{a}_{8}=-\frac{1}{8}\,,\quad{a}_{10}=-\frac{1}{800}\,.$ (4.10) From these, we can determine from (3.24) ${a}_{6}=\frac{173}{20}\,.$ (4.11) At the origin we then obtain $v_{*}(0)=-\frac{1}{120}\varepsilon\left(1+\frac{1087}{120}\varepsilon^{2}\right)\,,\qquad v^{\prime\prime}_{*}(0)=\frac{1}{10}\varepsilon\left(1+\frac{389}{40}\varepsilon\right)\,,$ (4.12) and hence $V_{*}^{\prime\prime}(0)=\frac{1}{40}\varepsilon\left(1+\frac{349}{40}\varepsilon\right)\,.$ (4.13) It is easy to verify that $V_{*}^{\prime\prime}(0)/V_{*}(0)=-d$ to this order again. In Table 2 the result (4.13) is compared with the numerical results of [23] in this case and in Figure 2 a graphical comparison of $\varepsilon$-expansion and numerical solutions is made. $d$ | Numerical | $\varepsilon$-expansion at ${\rm O}(\varepsilon^{2})$ ---|---|--- 3 | 0 | $0\qquad$ 2.9 | 0.005 | $0.005\qquad$ 2.8 | 0.015 | $0.014\qquad$ 2.7 | 0.032 | $0.027\qquad$ 2.6 | 0.062 | $0.045\qquad$ 2.5 | 0.108 | $0.067\qquad$ Table 2: Comparison of ERG numerical and analytical results for $V_{*}^{\prime\prime}(0)$ for the $n=3$ fixed point. Figure 2: $n=3$ $v_{*}(x)$ from $\varepsilon$-expansion at ${\rm O}(\varepsilon^{2})$ and numerical solution for $d=2.9$, $2.8$ and $2.6$. Finally, we consider the $n=4$ case. Following the same programme as above, the coefficients in the expansion are $\displaystyle{a}_{0}=$ $\displaystyle\frac{432}{35}\,,\quad{a}_{2}=\frac{288}{5}\,,\quad{a}_{4}=\frac{324}{5}\,,\quad{a}_{6}=36\,,$ $\displaystyle{a}_{10}=$ $\displaystyle-\frac{27}{100}\,,\quad{a}_{12}=-\frac{3}{800}\,,\quad{a}_{14}=-\frac{1}{39200}\,,$ (4.14) and $a_{8}=\frac{23904}{875}\,.$ (4.15) The fixed point solution at order $\varepsilon^{2}$ then determines $V_{*}^{\prime\prime}(0)=-\frac{3}{560}\varepsilon\left(1+\frac{99441}{3500}\varepsilon\right)\,.$ (4.16) Table 3 compares this result with [23] and with some corresponding graphs exhibited in Figure 3. $d$ | Numerical | $\varepsilon$-expansion at ${\rm O}(\varepsilon)$ ---|---|--- $\frac{8}{3}$ | 0 | $0\qquad$ 2.6 | $-0.001$ | $-0.001\qquad$ 2.5 | $-0.008$ | $-0.005\qquad$ 2.4 | $-0.028$ | $-0.012\qquad$ 2.3 | $-0.082$ | $-0.022\qquad$ 2.2 | $-0.226$ | $-0.035\qquad$ Table 3: Comparison of ERG numerical and analytical results for $V_{*}^{\prime\prime}(0)$ for the $n=4$ fixed point. Figure 3: $n=4$ $v_{*}(x)$ from $\varepsilon$-expansion at ${\rm O}(\varepsilon^{2})$ and numerical solution for $d=2.6$, $2.5$ and $2.4$. ## 5 Critical Exponents Having computed $v_{*}(x)$ to order $\varepsilon^{2}$ for the fixed points below each critical dimension $d=2n/(n-1)$, we now consider the RG flow near these fixed points and compute certain critical exponents. In the local potential approximation, the ERG flow is given by (3.1) and with the change of variables as in (3.3) we now have the following RG flow equation for $v(x,t)$ $\frac{4}{d-2}\big{(}\dot{v}(x,t)-dv(x,t)\big{)}=v^{\prime\prime}(x,t)-2xv^{\prime}(x,t)-v^{\prime}(x,t)^{2}\,.$ (5.1) In the neighbourhood of a fixed point $v(x,t)=v_{*}(x)+e^{\lambda t}f(x)\,,$ (5.2) where $f(x)$ therefore satisfies the following linear eigenvalue equation $\frac{4}{d-2}(\lambda-d)=f^{\prime\prime}-2xf^{\prime}-2v^{\prime}_{*}f^{\prime}\,.$ (5.3) This may be rewritten in the form $-Df(x)+v_{*}{\\!}^{\prime}(x)f^{\prime}(x)=\hat{\lambda}f(x)\,,\qquad\hat{\lambda}=-\frac{2}{d-2}(\lambda-d)\,,$ (5.4) with $D$ the differential operator in (3.6). In the case of the Gaussian fixed point $v_{*}=0$, the eigenvalues are $\hat{\lambda}_{k}=k$ and the associated eigenfunctions $f_{k}$ are just Hermite polynomials, so that $f_{k}(x)=H_{k}(x)\,,\qquad\lambda_{k}=d-{\textstyle\frac{1}{2}}k(d-2)\,,\qquad k=0,1,2,\dots\,.$ (5.5) These correspond to the operators $\phi^{k}$ where $\phi$ has dimensions ${\textstyle\frac{1}{2}}(d-2)$. For the high temperature fixed point then in (5.3) $v_{*}^{\prime}(x)=\frac{4}{d-2}x$ and the eigenfunctions are again Hermite polynomials with rescaled variable and $\lambda_{k}=d-{\textstyle\frac{1}{2}}k(d+2)$. For non trivial critical points we require $v_{*}$ to be a non singular solution of (3.4) For dimension $d$ as in (2.6) we may then consider a perturbation expansion in $\varepsilon$, so that $\displaystyle f_{k}(x)$ $\displaystyle=H_{k}(x)+{\rm O}(\varepsilon)\,,$ $\displaystyle\hat{\lambda}_{k}$ $\displaystyle=k+\varepsilon\hat{\lambda}_{k}{\\!}^{(1)}+\varepsilon^{2}\hat{\lambda}_{k}{\\!}^{(2)}+{\rm O}(\varepsilon^{3})\,.$ (5.6) To ${\rm O}(\varepsilon)$ it is sufficient to take in (5.4) $v_{*}(x)=c_{n}\varepsilon H_{2n}(x)\,.$ (5.7) To extract $\hat{\lambda}_{k}{\\!}^{(1)}$ we use standard first-order perturbation theory using the basis of eigenfunctions of $D$, $\displaystyle N_{k}\hat{\lambda}_{k}{\\!}^{(1)}$ $\displaystyle=c_{n}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{k}(x)H_{2n}{\\!}^{\prime}(x)H_{k}{\\!}^{\prime}(x)$ $\displaystyle=-{\textstyle\frac{1}{2}}c_{n}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,H_{k}(x)^{2}\frac{{\rm d}}{{\rm d}x}\left(e^{-x^{2}}H_{2n}{\\!}^{\prime}(x)\right)$ $\displaystyle=2nc_{n}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{k}(x)^{2}H_{2n}(x)$ $\displaystyle=2nc_{n}\,G_{kk\,2n}\,.$ (5.8) This gives $\hat{\lambda}_{k}{\\!}^{(1)}=2(n-1)^{2}\frac{n!}{(2n)!}\frac{k!}{(k-n)!}\,,$ (5.9) and in terms of the exponents $\lambda_{k}$, we have $\displaystyle\lambda_{k}$ $\displaystyle=d-{\textstyle\frac{1}{2}}(d-2)\hat{\lambda}_{k}$ $\displaystyle=\frac{2n-k}{n-1}+\varepsilon\left(\frac{1}{2}k-1-2(n-1)\frac{n!}{(2n)!}\frac{k!}{(k-n)!}\right)+{\rm O}(\varepsilon^{2})\,.$ (5.10) The results (5.9) or equivalently (5.10) were found in the beginning of the RG analysis of critical points in [24], using an approximation to the Wegner- Houghton RG equation, see also [17], and (up to misprints) in [25], using the LPA for the Wegner-Houghton equation. They are identical with the perturbative result (2.18). For higher order calculations it is convenient to modify the eigenvalue equation in (5.4) by considering the transformed differential operator $\Delta=e^{-v_{*}(x)}\Big{(}-D+v_{*}{\\!}^{\prime}(x)\frac{{\rm d}}{{\rm d}x}\Big{)}e^{v_{*}(x)}=-D+\frac{2d}{d-2}\,v_{*}(x)\,,\\\ $ (5.11) using that $v_{*}(x)$ satisfies the fixed point equation (3.4). It is obvious that the eigenvalue equation $\Delta{\hat{f}}={\hat{\lambda}}{\hat{f}}\,,$ (5.12) is equivalent to (5.4) and furthermore the operator $\Delta$ is hermitian111We may also note $e^{-\frac{1}{2}x^{2}}\Delta\,e^{\frac{1}{2}x^{2}}$ is a Scrödinger operator with potential $U(x)={\textstyle\frac{1}{2}}x^{2}-{\textstyle\frac{1}{2}}+\frac{2d}{d-2}\,v_{*}(x)$. Asymptotically $U(x)\sim{\textstyle\frac{1}{2}}\frac{(d+2)^{2}}{(d-2)^{2}}\,x^{2}$. with respect to the measure ${\rm d}x\,e^{-x^{2}}$. To ${\rm O}(\varepsilon)$ it is easy to see that, with $v_{*}$ given by (5.7) and using the expansion (3.11), that the eigenvalues are the same as (5.9). To ${\rm O}(\varepsilon^{2})$ and using (4.1) second-order perturbation theory gives $\displaystyle\hat{\lambda}_{k}{\\!}^{(2)}$ $\displaystyle=(n-1)^{2}\frac{c_{n}}{N_{k}}G_{2n\,kk}+\frac{2nc_{n}}{N_{k}}\sum_{p=0}^{2n-1}a_{2p}G_{2p\,kk}-(2nc_{n})^{2}\sum_{m\neq k}\frac{1}{N_{m}N_{k}}\frac{G_{2n\,mk}^{\,2}}{m-k}\,,$ (5.13) where the first two terms arise from the ${\rm O}(\varepsilon^{2})$ terms in the operator itself, and the final term is the usual second order perturbation expression for a perturbative potential $2nc_{n}\varepsilon H_{2n}$. Substituting the expressions for $a_{2n}$ and $a_{2p}(p\neq n)$, (3.24) and (3.21) respectively, the expression above becomes $\displaystyle\hat{\lambda}_{k}{\\!}^{(2)}$ $\displaystyle=\left((n-1)^{2}+n(n-1)\right)\frac{c_{n}}{N_{k}}G_{2n\,kk}-\frac{4n}{(n-1)^{2}}\frac{c_{n}}{N_{2n}}\sum_{p\neq n}\frac{p(2n-p)}{2(n-p)}\frac{G_{2n\,2n\,2p}^{\,2}}{N_{2p}}$ $\displaystyle\ \ +2nc_{n}^{2}\sum_{p\neq n}\frac{2n-p}{2(n-p)}\frac{G_{2p\,kk}G_{2n\,2n\,2p}}{N_{2p}N_{k}}-(2nc_{n})^{2}\sum_{m\neq k}\frac{1}{N_{m}N_{k}}\frac{G_{2n\,mk}^{\,2}}{m-k}\,.$ (5.14) Finally, substituting for $N_{m}$ from (3.13), $G_{klm}$ from (3.14) and $c_{n}$ from (3.16), we have $\displaystyle\hat{\lambda}_{k}{\\!}^{(2)}$ $\displaystyle=(n-1)^{3}(2n-1)\frac{1}{n}\frac{n!k!}{(2n)!(k-n)!}$ $\displaystyle\ \ -2(n-1)^{4}\frac{1}{n^{2}}\frac{n!^{7}}{(2n)!^{2}}\frac{k!}{(k-n)!}\sum_{p\neq n}\frac{p(2n-p)}{n-p}\frac{(2p)!}{p!^{4}(2n-p)!^{2}}$ $\displaystyle\ \ +(n-1)^{4}\frac{1}{n}\frac{n!^{6}}{(2n)!^{2}}k!\sum_{p\neq n}\frac{2n-p}{n-p}\frac{(2p)!}{p!^{4}(k-p)!(2n-p)!}$ $\displaystyle\ \ -2(n-1)^{4}\frac{n!^{6}}{(2n)!^{2}}k!\sum_{s\neq n}\frac{1}{n-s}\frac{(2n+k-2s)!}{(k-s)!^{2}s!^{2}(2n-s)!^{2}}\,,$ (5.15) where $2s=2n+k-m$. While this expression is somewhat complicated, for any particular choices of $n$ it can be simplified to a polynomial in $k$ of order $2n-1$. Here we give the results for the first two fixed points $\displaystyle\hat{\lambda}_{k}{\\!}^{(2)}\big{|}_{n=2}=$ $\displaystyle-\frac{1}{12}k(k-1)(k-4)\,,$ (5.16a) $\displaystyle\hat{\lambda}_{k}{\\!}^{(2)}\big{|}_{n=3}=$ $\displaystyle-\frac{1}{600}k(k-1)(13k^{3}-17k^{2}-424k+800)\,.$ (5.16b) A similar, but not identical result to (5.16a) was obtained in [25], again using the LPA for the Wegner-Houghton RG equation rather than the Polchinski equation. Having computed the eigenvalues $\hat{\lambda}$ to ${\rm O}(\varepsilon^{2})$, we may now extend (5.10) to calculate the corresponding critical exponents to ${\rm O}(\varepsilon^{2})$. The exponents $\lambda_{k}$ are given in terms of the eigenvalues as $\displaystyle\lambda_{k}$ $\displaystyle=\frac{2n-k}{n-1}+\varepsilon\left(\frac{k}{2}-1-2(n-1)\frac{n!}{(2n)!}\frac{k!}{(k-n)!}\right)$ $\displaystyle\ \ +\varepsilon^{2}\left((n-1)^{2}\frac{n!}{(2n)!}\frac{k!}{(k-n)!}-\frac{1}{n-1}\hat{\lambda}_{k}{\\!}^{(2)}\right)+{\rm O}(\varepsilon^{3})\,.$ (5.17) This gives $\displaystyle\lambda_{k}\big{|}_{n=2}={}$ $\displaystyle 4-k+\varepsilon\big{(}{\textstyle\frac{1}{2}}k-1-\textstyle{\frac{1}{6}}k(k-1)\big{)}+\varepsilon^{2}\,\textstyle{\frac{1}{12}}k(k-1)(k-3)+{\rm O}(\varepsilon^{3})\,,$ $\displaystyle\lambda_{k}\big{|}_{n=3}={}$ $\displaystyle{\textstyle\frac{1}{2}}(6-k)+\varepsilon\big{(}{\textstyle\frac{1}{2}}k-1-\textstyle{\frac{1}{30}}k(k-1)(k-2)\big{)}$ $\displaystyle{}+\varepsilon^{2}\,\textstyle{\frac{1}{1200}}\,k(k-1)(k-5)(13k^{2}+48k-144)+{\rm O}(\varepsilon^{3})\,.$ (5.18) ### 5.1 Exact Exponents The results (5.9) and (5.16a), (5.16b) show that ${\hat{\lambda}}_{k}=0$ for $k=0,1$ to ${\rm O}(\varepsilon^{2})$, at least for $n=2,3$. For general $n$ from (5.15) $\displaystyle\hat{\lambda}_{1}{\\!}^{(2)}$ $\displaystyle=(n-1)^{4}\frac{1}{n}\frac{n!^{6}}{(2n)!^{2}}\sum_{p=0}^{1}\frac{2n-p}{n-p}\frac{(2p)!}{p!^{4}(1-p)!(2n-p)!}$ $\displaystyle\ \ \ -2(n-1)^{4}\frac{n!^{6}}{(2n)!^{2}}\sum_{s=0}^{1}\frac{1}{n-s}\frac{(2n+1-2s)!}{(1-s)!^{2}s!^{2}(2n-s)!^{2}}$ $\displaystyle=2(n-1)^{4}\frac{n!^{6}}{(2n)!^{2}}\frac{1}{n(n-1)(2n)!}\left((n-1)(-2n)+2n(2n-1)-2n^{2}\right)$ $\displaystyle=0\,.$ (5.19) These results follow in general since it is possible to find exact eigenfunctions for $k=0,1$ in (5.4). Firstly we have the trivial case, $f_{0}=1\,,\qquad\hat{\lambda}_{0}=0\,.$ (5.20) For $k=1$ we also have $f_{1}(x)=v^{\prime}_{*}(x)-\frac{4x}{d-2}\,,\qquad\hat{\lambda}_{1}=1\,.$ (5.21) To verify this we first obtain from (3.4) $Dv_{*}{\\!}^{\prime}+\frac{d+2}{d-2}\,v_{*}{\\!}^{\prime}=v_{*}{\\!}^{\prime}v_{*}{\\!}^{\prime\prime}\,,$ (5.22) since then $Df_{1}(x)-v_{*}{\\!}^{\prime}(x)f_{1}(x)=-v_{*}{\\!}^{\prime}(x)+\frac{4x}{d-2}=-f_{1}(x)\,.$ (5.23) The vanishing of $\hat{\lambda}_{1}$ reflects that there is no anomalous dimension for the field $\phi$ in the LPA. Additionally from (5.22) we have another exact eigenfunction and eigenvalue, $f_{2n-1}(x)=v^{\prime}_{*}(x)\,,\qquad\hat{\lambda}_{2n-1}=\frac{d+2}{d-2}\,.$ (5.24) Expanding the exact eigenvalue $\hat{\lambda}_{2n-1}$ in powers of $\varepsilon$ gives $\hat{\lambda}_{2n-1}=2n-1+(n-1)^{2}\varepsilon+{\textstyle\frac{1}{2}}(n-1)^{3}\varepsilon^{2}+\dots\,,$ (5.25) which is consistent with (5.9) since $\hat{\lambda}^{(1)}_{2n-1}=(n-1)^{2}\,.$ (5.26) It is also in accord with the second-order result, (5.15). Relabelling $s=2n-p$, $\displaystyle\hat{\lambda}_{2n-1}^{(2)}$ $\displaystyle=(n-1)^{3}(2n-1)\frac{1}{n}\frac{n!(2n-1)!}{(2n)!(n-1)!}$ $\displaystyle\ \ \ -2(n-1)^{4}\frac{1}{n^{2}}\frac{n!^{7}}{(2n)!^{2}}\frac{(2n-1)!}{(n-1)!}\sum_{p\neq n}\frac{p(2n-p)}{n-p}\frac{(2p)!}{p!^{4}(2n-p)!^{2}}$ $\displaystyle\ \ \ +(n-1)^{4}\frac{1}{n}\frac{n!^{6}}{(2n)!^{2}}(2n-1)!\sum_{p\neq n}\frac{2n-p}{n-p}\frac{(2p)!}{p!^{4}(2n-1-p)!(2n-p)!}$ $\displaystyle\ \ \ +2(n-1)^{4}\frac{n!^{6}}{(2n)!^{2}}(2n-1)!\sum_{p\neq n}\frac{1}{n-p}\frac{(2p-1)!}{(p-1)!^{2}p!^{2}(2n-p)!^{2}}$ $\displaystyle=\frac{1}{2}(n-1)^{3}+\frac{(n-1)^{4}}{2n^{2}}\frac{n!^{6}}{(2n)!}\sum_{p=0}^{2n}(4n-3p)\frac{(2p)!}{p!^{4}(2n-p)!^{2}}\,.$ (5.27) Provided the final sum is identically zero for all $n$ this agrees with the exact result. We demonstrate that the sum vanishes in Appendix B. In this case the relevant operator $\phi^{2n-1}\propto\partial^{2}\phi$ by the equations of motion and has dimension ${\textstyle\frac{1}{2}}(d+2)$. ## 6 Beyond the Local Potential Approximation Although the LPA captures the essential features of the fixed point structure in the space of couplings for scalar theories it neglects the momentum dependence of vertices and is therefore limited in terms of calculating critical exponents quantitatively, the anomalous dimension $\eta$ of the scalar field is undetermined and set to zero. Although at the Wilson-Fisher fixed point $\eta$ is small and in the $\varepsilon$-expansion $\eta={\rm O}(\varepsilon^{2})$ it is of course necessary to take $\eta$ into account in more systematic treatments. To this end a natural extension of the LPA is to assume a solution of the exact RG flow equations which is expressible in terms of a local functional of the fields and their derivatives, the derivative expansion [26], for a recent discussion see [27]. In the derivative expansion there is a necessary dependence on the form of the cut off function but in application to the Polchinski RG equation, with terms quadratic in derivatives, there are two constants, see [28, 29]. At the second order in the derivative expansion, the Polchinski equation may be reduced [28] to a following pair of coupled ODEs, extending (3.1), for a potential $V(\phi,t)$ and also the coefficient $Z(\phi,t)$ of ${\textstyle\frac{1}{2}}(\partial\phi)^{2}$ in the derivative expansion, $\displaystyle\dot{V}$ $\displaystyle=V^{\prime\prime}-{\textstyle\frac{1}{2}}(d-2+\eta)\phi V^{\prime}+dV-V^{\prime 2}+AZ\,,$ (6.1a) $\displaystyle\dot{Z}$ $\displaystyle=Z^{\prime\prime}-{\textstyle\frac{1}{2}}(d-2+\eta)\phi Z^{\prime}-2V^{\prime}Z^{\prime}-4V^{\prime\prime}Z-\eta Z-\eta+2BV^{\prime\prime 2}\,,$ (6.1b) where $A$ and $B$ are the two cut off dependent constants which cannot be eliminated by any rescaling (essentially the same equations were obtained from the Wilson RG equations in [30]). As earlier it is convenient in our discussion to introduce a rescaled variable $x={\textstyle\frac{1}{2}}(d-2+\eta)^{\frac{1}{2}}\phi\,,\qquad v(x,t)=V(\phi,t)\,,\quad z(x,t)=Z(\phi,t),$ (6.2) so that the coupled equations (6.1a), (6.1b) become $\displaystyle\frac{2}{d-2+\eta}\,\dot{v}$ $\displaystyle=(D+K)v-{\textstyle\frac{1}{2}}\,v^{\prime 2}+\tilde{A}z\,,$ (6.3a) $\displaystyle\frac{2}{d-2+\eta}\,\dot{z}$ $\displaystyle=(D-L)z-v^{\prime}z^{\prime}-2v^{\prime\prime}z-L+{\textstyle\frac{1}{2}}\tilde{B}\,v^{\prime\prime 2}\,,$ (6.3b) where $K=\frac{2d}{d-2+\eta}\,,\quad L=\frac{2\eta}{d-2+\eta}\,,\qquad\tilde{A}=\frac{2}{d-2+\eta}A\,,\quad\tilde{B}={\textstyle\frac{1}{2}}(d-2+\eta)B\,.$ (6.4) At a fixed point, $v(x,t)\to v_{*}(x),\,z(x,t)\to z_{*}(x)$ which satisfy the equations $\displaystyle(D+K)v_{*}$ $\displaystyle={\textstyle\frac{1}{2}}\,v_{*}{\\!}^{\prime 2}-\tilde{A}z_{*}\,,\quad$ (6.5a) $\displaystyle(D-L)z_{*}-v_{*}{\\!}^{\prime}z_{*}{\\!}^{\prime}-2v_{*}{\\!}^{\prime\prime}z_{*}$ $\displaystyle=L-{\textstyle\frac{1}{2}}\tilde{B}\,v_{*}{\\!}^{\prime\prime 2}\,.$ (6.5b) Assuming (2.6) then in an $\varepsilon$-expansion, assuming $A,B$ are $O(1)$, a consistent solution is obtained by requiring $v_{*}(x)={\rm O}(\varepsilon)$ and $\eta,z_{*}(x)={\rm O}(\varepsilon^{2})$. To lowest order then $v_{*}\to\varepsilon c_{n}H_{2n}$ as in (3.10). With this result then (6.5b) becomes $Dz_{*}=(n-1)\eta-\varepsilon^{2}{\textstyle\frac{1}{2}}\tilde{B}c_{n}{\\!}^{2}H_{2n}{\\!\\!}^{\prime\prime 2}+{\rm O}(\varepsilon^{3})\,.$ (6.6) This determines $\eta=\varepsilon^{2}\,\eta^{(2)}+{\rm O}(\varepsilon^{3})$ since the left hand side of (6.6) is orthogonal to 1 so that $(n-1)\eta^{(2)}N_{0}={\textstyle\frac{1}{2}}\tilde{B}c_{n}{\\!}^{2}\int^{\infty}_{-\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{2n}{\\!\\!}^{\prime\prime 2}\,,$ (6.7) which implies $\eta^{(2)}=2^{2n+2}\frac{n(2n-1)}{n-1}(2n)!\,\tilde{B}c_{n}{\\!}^{2}\,.$ (6.8) We consider this further later but first we may then use (6.6) to determine $z_{*}(x)$ to lowest order by using the expansion $z_{*}=\varepsilon^{2}c_{n}{\\!}^{2}\sum_{p}b_{2p}H_{2p}+{\rm O}(\varepsilon^{3})\,.$ (6.9) Since $DH_{2p}=-4pH_{2p}$ and with $H_{2n}{\\!\\!}^{\prime\prime}=8n(2n-1)H_{2n-2}$ we may obtain $\displaystyle b_{2p}={}$ $\displaystyle 16\tilde{B}\,n^{2}(2n-1)^{2}\frac{G_{2p\,2n-2\,2n-2}}{pN_{2p}}$ $\displaystyle={}$ $\displaystyle 2^{2n-p}\tilde{B}\frac{(2n)!^{2}}{p(2n-2-p)!p!^{2}}\,,\quad p=1,\dots,2n-2\,.$ (6.10) We note that $b_{0}$ is not constrained by (6.6). Nevertheless $b_{0}$ may be determined by imposing $z_{*}(0)=0$ [28]. This ensures that (6.5a), (6.5b) have well defined solutions for all $x$ only for a specific choice of $\eta$ and $v_{*}(0)$. With these results (6.5a) becomes $(D+2n)v_{*}+(n-1)^{2}\varepsilon v_{*}=\varepsilon^{2}c_{n}{\\!}^{2}\bigg{(}H_{2n}{\\!\\!}^{\prime 2}-{\tilde{A}}\sum_{p=0}^{2n-2}b_{2p}H_{2p}\bigg{)}+{\rm O}(\varepsilon^{3})\,,$ (6.11) where $v_{*}$ is expressible as in (4.1) and $b_{2p}$ is given by (6.10). Just as in (3.12) and (3.16) this determines normalisation coefficient $c_{n}$, $(n-1)^{2}c_{n}N_{2n}=c_{n}{\\!}^{2}\big{(}nG_{2n\,2n\,2n}-{\tilde{A}}N_{2n}b_{2n}\big{)}\quad\Rightarrow\quad c_{n}\Big{(}1-\frac{n-1}{n}\,{\tilde{A}}{\tilde{B}}\Big{)}=\frac{(n-1)^{2}}{2^{n}n}\,\frac{n!^{3}}{(2n)!^{2}}\,.$ (6.12) Furthermore in (4.1) the expansion coefficients are also determined by (6.11) $\displaystyle a_{2p}$ $\displaystyle=c_{n}\frac{1}{2(n-p)}\ \Big{(}(2n-p)\,\frac{G_{2p\,2n\,2n}}{N_{2p}}-{\tilde{A}}\,b_{2p}\Big{)}$ $\displaystyle=c_{n}\,\frac{2^{2n-p}}{2(n-p)}\frac{(2n)!^{2}}{p!^{2}(2n-1-p)!}\Big{(}1-\frac{2n-1-p}{p}\,{\tilde{A}}{\tilde{B}}\Big{)}\,,\quad p\neq n\,.$ (6.13) except for $a_{2n}$. To obtain this it is necessary to extend (6.11) to ${\rm O}(\varepsilon^{3})$. Using the result (6.12) the expression (6.8) becomes $\eta^{(2)}=4(n-1)^{2}\frac{n!^{6}}{(2n!)^{3}}\,\frac{n(n-1)(2n-1){\tilde{B}}}{\big{(}n-(n-1){\tilde{A}}{\tilde{B}}\big{)}^{2}}\,.$ (6.14) The perturbative result (2.18) requires that the final factor, depending on $A,B$, should be one. ## 7 Critical Exponents with the Derivative Expansion We now consider small departures from the fixed point of the form $\displaystyle v(x,t)$ $\displaystyle=v_{*}(x)+e^{\lambda t}f(x)\,,$ $\displaystyle z(x,t)$ $\displaystyle=z_{*}(x)+e^{\lambda t}g(x)\,.$ (7.1) Substituting this form into (6.3a), (6.3b) we find the eigenvalue equations $\displaystyle-(D+K)f+v_{*}{\\!}^{\prime}f^{\prime}-\tilde{A}\,g={}$ $\displaystyle\hat{\lambda}f\,,$ (7.2a) $\displaystyle-(D-L)g+v_{*}{\\!}^{\prime}g^{\prime}+2v_{*}{\\!}^{\prime\prime}g+z_{*}{\\!}^{\prime}f^{\prime}+2z_{*}f^{\prime\prime}-\tilde{B}\,v_{*}{\\!}^{\prime\prime}f^{\prime\prime}={}$ $\displaystyle\hat{\lambda}f\,,$ (7.2b) with $D$ as in (3.6) and where $\lambda$ is related to $\hat{\lambda}$ according to (this differs from (5.4)). $\lambda=-{\textstyle\frac{1}{2}}(d-2+\eta)\hat{\lambda}\,.$ (7.3) Defining the vector $F=\left(\begin{array}[]{c}f\\\ g\\\ \end{array}\right)\,,$ (7.4) then (7.2a), (7.2b) may be written more compactly in matrix form as $\Delta F=\hat{\lambda}F\,.$ (7.5) The operator $\Delta$ is not hermitian so for later convenience we also consider the dual equation defining the dual eigenvectors ${\tilde{\Delta}}\tilde{F}=\hat{\lambda}\tilde{F}\,,$ (7.6) where ${\tilde{\Delta}}$ is the adjoint of $\Delta$ with a scalar product defined $\int{\rm d}x\,e^{-x^{2}}\,F_{1}{\\!}^{T}F_{2}$. Just as in Section 5 there are exact eigenfunctions and eigenvalues [28] $\displaystyle F={}$ $\displaystyle\left(\begin{array}[]{c}1\\\ 0\\\ \end{array}\right)\,,\qquad\qquad\qquad\quad\ \ \hat{\lambda}=-K\,,\qquad\ \ \lambda=d\,,$ (7.7c) $\displaystyle F={}$ $\displaystyle\left(\begin{array}[]{c}v_{*}{\\!}^{\prime}-(K-2)x\\\ z_{*}{\\!}^{\prime}\\\ \end{array}\right)\,,\qquad\hat{\lambda}=-K+1\,,\quad\lambda={\textstyle\frac{1}{2}}(d+2-\eta)\,,$ (7.7f) $\displaystyle F={}$ $\displaystyle\left(\begin{array}[]{c}v_{*}{\\!}^{\prime}\\\ z_{*}{\\!}^{\prime}\\\ \end{array}\right)\,,\qquad\qquad\qquad\quad\,\hat{\lambda}=-1\,,\qquad\quad\,\lambda={\textstyle\frac{1}{2}}(d-2+\eta)\,.$ (7.7i) The first corresponds to the identity operator and at the $n$-th multicritical point the second and third to the operators $\phi$ and $\phi^{2n-1}$. In the $\varepsilon$-expansion $\displaystyle\Delta=\Delta^{(0)}+\varepsilon\Delta^{(1)}+\dots\,,$ (7.8) and correspondingly in (7.5) $\hat{\lambda}=\hat{\lambda}^{(0)}+\varepsilon\hat{\lambda}^{(1)}+\dots\,,\qquad F=F^{(0)}+\varepsilon F^{(1)}+\dots\,,$ (7.9) and similarly for $\tilde{F}$. At zeroth order in $\varepsilon$, $\Delta^{(0)}=\left(\begin{array}[]{cc}-D-2n&-\tilde{A}\\\ 0&-D\\\ \end{array}\right)\,.$ (7.10) There are two sets of eigenfunctions, which are easily obtained $F^{(0)}_{k,1}=\left(\begin{array}[]{c}1\\\ 0\\\ \end{array}\right)H_{k}\,,\qquad\tilde{F}^{(0)}_{k,1}=\left(\begin{array}[]{c}1\\\ \frac{\tilde{A}}{2n}\\\ \end{array}\right)H_{k}\,,\qquad k=0,1,2,\dots\,,$ (7.11) and $F^{(0)}_{k,2}=\left(\begin{array}[]{c}-\frac{\tilde{A}}{2n}\\\ 1\\\ \end{array}\right)H_{k-2n}\,,\qquad\tilde{F}^{(0)}_{k,2}=\left(\begin{array}[]{c}0\\\ 1\\\ \end{array}\right)H_{k-2n}\,,\qquad k=2n,2n+1,\dots\,.$ (7.12) The eigenvalues in both cases are $\displaystyle\hat{\lambda}_{k,1}^{(0)}=\hat{\lambda}_{k,2}^{(0)}=k-2n\,,$ (7.13) which are thus two-fold degenerate for $k\geq 2n$. In general in terms of the perturbative anomalous dimensions at the fixed point discussed in Section 2 $\lambda_{k,1}=\frac{2n-k}{n-1}-\omega_{k,1}\,,\qquad\lambda_{k,2}=\frac{2n-k}{n-1}-\omega_{k,2}\,.$ (7.14) ### 7.1 Eigenvalues at ${\rm O}(\varepsilon)$ We now use first-order perturbation theory to compute $\hat{\lambda}^{(1)}_{k,1}$ and $\hat{\lambda}^{(1)}_{k,2}$. In the expansion (7.8) at ${\rm O}(\varepsilon)$ $\Delta^{(1)}=\left(\begin{array}[]{cc}-(n-1)^{2}+c_{n}H_{2n}{\\!}^{\prime}\frac{{\rm d}}{{\rm d}x}&0\\\ -\tilde{B}\,c_{n}H_{2n}{\\!\\!}^{\prime\prime}\frac{{\rm d}^{2}}{{\rm d}x^{2}}&c_{n}H_{2n}{\\!\\!}^{\prime}\frac{{\rm d}}{{\rm d}x}+2c_{n}H_{2n}{\\!\\!}^{\prime\prime}\\\ \end{array}\right)\,.$ (7.15) The ${\rm O}(\varepsilon)$ results are determined in terms of the $2\times 2$ matrix ${\cal M}_{k}$ defined by ${\cal M}_{k,ij}=\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}\tilde{F}^{(0)\,T}_{k,i}\,\Delta^{(1)}F^{(0)}_{k,j}\,.$ (7.16) For $k<2n$ there is a single eigenfunction so that the perturbation theory result is just $\displaystyle N_{k}\hat{\lambda}^{(1)}_{k,1}$ $\displaystyle={\cal M}_{k,11}$ $\displaystyle=-(n-1)^{2}N_{k}+c_{n}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{k}H_{2n}{\\!\\!}^{\prime}H_{k}{\\!}^{\prime}-c_{n}\frac{1}{2n}\tilde{A}\tilde{B}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{k}H_{2n}{\\!\\!}^{\prime\prime}H_{k}{\\!}^{\prime\prime}\,.$ (7.17) Hence, using the formulae for $N_{k}$, $G_{klm}$ in (3.13), (3.14) we obtain $\displaystyle\hat{\lambda}^{(1)}_{k,1}={}$ $\displaystyle-(n-1)^{2}+c_{n}\,2^{n+1}\frac{(2n)!k!}{n!^{2}(k-n)!}\big{(}n-(n-1)\tilde{A}\tilde{B}(n-1)\big{)}$ $\displaystyle={}$ $\displaystyle-(n-1)^{2}+2(n-1)^{2}\frac{n!}{(2n)!}\frac{k!}{(k-n)!}\,,$ (7.18) where $c_{n}$ is determined by (6.12) so that the dependence on $\tilde{A},\tilde{B}$ disappears. This result is essentially identical to (5.9) and is in accord with the exact results (7.7c), (7.7f), (7.7i) which correspond to $k=0,1,2n-1$. For $k\geq 2n$ there is a two-fold degeneracy, and so we use degenerate perturbation theory. In particular, for non-trivial first-order eigenfunctions we require that the first-order perturbations to the eigenvalues solve the following characteristic equation $\displaystyle\det\big{[}\lambda^{(1)}_{k}{\cal N}_{k}-{\cal M}_{k}\big{]}=0\,,$ (7.19) where the ${\cal N}_{k}$ is the diagonal matrix ${\cal N}_{k}=\left(\begin{array}[]{cc}N_{k}&0\\\ 0&N_{k-2n}\\\ \end{array}\right)\,,$ (7.20) More explicitly, the elements of $\cal M$ are given by (7.17) and $\displaystyle{\cal M}_{k,22}$ $\displaystyle=c_{n}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{k-2n}^{\vphantom{T}}\left(H^{\prime}_{2n}H^{\prime}_{k-2n}+2H^{\prime\prime}_{2n}H_{k-2n}+\frac{1}{2n}\tilde{A}\tilde{B}\,H^{\prime\prime}_{2n}H^{\prime\prime}_{k-2n}\right)\,,$ (7.21a) $\displaystyle{\cal M}_{k,21}$ $\displaystyle=-\tilde{B}\,c_{n}\int_{-\infty}^{\infty}\\!\\!\\!\\!{\rm d}x\,e^{-x^{2}}H_{k-2n}^{\vphantom{T}}H^{\prime\prime}_{2n}H^{\prime\prime}_{k},$ (7.21b) $\displaystyle{\cal M}_{k,12}$ $\displaystyle=c_{n}\frac{1}{2n}\tilde{A}\int_{-\infty}^{\infty}{\rm d}x\,e^{-x^{2}}H_{k}^{\vphantom{T}}\left(2H^{\prime\prime}_{2n}H_{k-2n}^{\vphantom{T}}+\frac{1}{2n}\tilde{A}\tilde{B}H^{\prime\prime}_{2n}H^{\prime\prime}_{k-2n}\right)$ $\displaystyle=0\,,$ (7.21c) where in (7.21c) we use that $H_{k}$ is orthogonal to polynomials of degree $<k$. Hence since the matrix ${\cal M}_{k}$ is lower triangular the eigenvalues solving (7.19) are just $\hat{\lambda}^{(1)}_{1,k}$ as in (7.18) and $\displaystyle\hat{\lambda}^{(1)}_{k,2}={}$ $\displaystyle\frac{{\cal M}_{k,22}}{N_{k-2n}}=c_{n}\,2^{n+1}\frac{(2n)!(k-2n)!}{n!^{2}(k-3n+1)!}\left(n(k-n+1)+\tilde{A}\tilde{B}(n-1)(k-3n+1)\right)\,.$ (7.22) From (7.3) $\lambda=-(\frac{1}{n-1}-{\textstyle\frac{1}{2}}\varepsilon){\hat{\lambda}}+{\rm O}(\varepsilon^{2})$ so to ${\rm O}(\varepsilon)$ we then have $\displaystyle\lambda_{k,1}={}$ $\displaystyle\frac{2n-k}{n-1}+\varepsilon\left[\frac{1}{2}k-1-2(n-1)\frac{n!\,k!}{(2n)!\,(k-n)!}\right]\,,$ (7.23a) $\displaystyle\lambda_{k,2}={}$ $\displaystyle\frac{2n-k}{n-1}$ $\displaystyle\\!{}+\varepsilon\left[\frac{1}{2}k-n-2(n-1)\frac{n!\,(k-2n)!}{(2n)!\,(k-3n+1)!}\frac{n(k-n+1)+(n-1)\tilde{A}\tilde{B}(k-3n+1)}{n-(n-1)\tilde{A}\tilde{B}}\right]\,.$ (7.23b) The result (7.23a) matches the perturbative result in (2.16) but (7.23b) depends on $AB$ so cannot agree with (2.32) in general. ## 8 Expansion of Exact RG Equation An alternative approach, which nevertheless has many similarities although some crucial differences to the derivative expansion, is to expand the full renormalisation group equation which contains linear and quadratic terms in terms of translation invariant operators, or scaling fields, which are exact eigen-solutions for its linearised part. This has been applied in [17] and [31], [32] to the Wilson RG equation. Here we apply similar methods more simply to the Polchinski equation for the effective action $S[\varphi,t]$ for a scalar field $\varphi$. This takes the form, after convenient rescalings by the cut off scale $\Lambda$, $t=-\ln\Lambda$, $\displaystyle\frac{\partial}{\partial t}S={}$ $\displaystyle\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,\big{(}{\textstyle\frac{1}{2}}d+1-{\textstyle\frac{1}{2}}\eta+p\cdot\partial_{p}\big{)}\tilde{\varphi}(p)\frac{\delta S}{\delta\tilde{\varphi}(p)}$ $\displaystyle{}-\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,K^{\prime}(p^{2})\bigg{(}\frac{\delta^{2}S}{\delta\tilde{\varphi}(p)\delta\tilde{\varphi}(-p)}-\frac{\delta S}{\delta\tilde{\varphi}(p)}\,\frac{\delta S}{\delta\tilde{\varphi}(-p)}\bigg{)}$ $\displaystyle{}-{\textstyle\frac{1}{2}}\eta\,\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,K(p^{2})^{-1}p^{2}\tilde{\varphi}(p)\tilde{\varphi}(-p)+C\,,$ (8.1) where $K(p^{2}/\Lambda^{2})$ is a cut off function and $\tilde{\varphi}$ is the Fourier transform of $\varphi$, $\delta\tilde{\varphi}(p)/\delta\tilde{\varphi}(p^{\prime})=(2\pi)^{d}\delta^{d}(p-p^{\prime})$. $C$ is an additional constant independent of $\varphi$, in general it is irrelevant and may be neglected222The standard derivation gives $C=\delta^{d}(0)\int{\rm d}^{d}p\,K^{\prime}(p^{2})K(p^{2})^{-1}p^{2}$ with $(2\pi)^{d}\delta^{d}(0)$ the overall volume.. Apart from $K(0)=1$ and sufficient rapid fall off for large $p^{2}$ no restriction on the cut off function $K(p^{2})$ is imposed here. The Gaussian fixed point corresponds to $S=0$ and also when the anomalous scale dimension for $\varphi$, $\eta=0$. The appearance of $\eta$ in (8.1) arises by assuming that $\varphi$ varies with the cut off $\Lambda$ with an anomalous dimension ${\textstyle\frac{1}{2}}\eta$. As will be apparent later the additional term proportional to $\eta$ in (8.1) is necessary for consistent RG flow solutions. The derivative expansion is obtained directly by approximating (8.1) by assuming $\tilde{\varphi}(p)$ is expanded as $(2\pi)^{d}(\phi-i\partial\phi\cdot\partial_{p})\delta^{d}(p)$ and requiring $S[\varphi,t]\to\int{\rm d}^{d}x\big{(}V(\phi,t)+{\textstyle\frac{1}{2}}Z(\phi,t)(\partial\phi)^{2}\big{)}$. The starting point of the discussion in this Section requires solutions of $\big{(}{\cal D}_{1}+{\cal D}_{2}\big{)}{\cal O}=\lambda{\cal O}\,,$ (8.2) for ${\cal D}_{1}+{\cal D}_{2}$ the differential operator defined by the linear part of (8.1), $\displaystyle{\cal D}_{1}={}$ $\displaystyle\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,\big{(}{\textstyle\frac{1}{2}}d+1+p\cdot\partial_{p}\big{)}\tilde{\varphi}(p)\,\frac{\delta}{\delta\tilde{\varphi}(p)}\,,$ $\displaystyle{\cal D}_{2}={}$ $\displaystyle-\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,K^{\prime}(p^{2})\frac{\delta^{2}}{\delta\tilde{\varphi}(p)\delta\tilde{\varphi}(-p)}\,.$ (8.3) The eigenvalue equation ${\cal D}_{1}{\hat{\cal O}}=\lambda{\hat{\cal O}}\,,$ (8.4) is easily solved in terms of local translational invariant operators by ${\hat{\cal O}}=\frac{1}{(2\pi)^{d(k-1)}}\int\prod_{i=1}^{k}{\rm d}^{d}p_{i}\,\tilde{\varphi}(p_{i})\ \delta^{d}\big{(}{\textstyle\sum_{i}}p_{i}\big{)}\,O(p_{1},\dots,p_{k})\,,\quad\lambda=d-k({\textstyle\frac{1}{2}}d-1)-r\,,$ (8.5) where $O(p_{1},\dots,p_{k})$ is a scalar symmetric homogeneous polynomial of degree $r$. We then obtain corresponding solutions of (8.2) in the form ${\cal O}=e^{-{\cal Y}}{\hat{\cal O}}\,,$ (8.6) where ${\cal Y}$ is defined by $[{\cal D}_{1},{\cal Y}]={\cal D}_{2}\,,$ (8.7) ensuring that ${\cal O}$ as defined by (8.6) has the same eigenvalue as $\hat{\cal O}$. It is easy to solve (8.7) giving ${\cal Y}={\textstyle\frac{1}{2}}\,\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\ G(p)\,\frac{\delta^{2}}{\delta\tilde{\varphi}(p)\delta\tilde{\varphi}(-p)}\,,\qquad G(p)=\frac{K(p^{2})}{p^{2}}\,,$ (8.8) so that in (8.6) $e^{-{\cal Y}}$ essentially generates normal ordering. For comparison with the derivative expansion earlier we consider just operators with $r=0,2$ for which a convenient basis is $\displaystyle{\cal O}_{k0}={}$ $\displaystyle e^{-{\cal Y}}\frac{1}{(2\pi)^{d(k-1)}}\int\prod_{i=1}^{k}{\rm d}^{d}p_{i}\,\tilde{\varphi}(p_{i})\ \delta^{d}\big{(}p_{(k)}\big{)}\,,\qquad p_{(k)}={\textstyle\sum_{i=1}^{k}}\,p_{i}\,,$ $\displaystyle{\cal O}_{k2}={}$ $\displaystyle\frac{1}{k(k-1)}\,e^{-{\cal Y}}\,\frac{1}{(2\pi)^{d(k-1)}}\int\prod_{i=1}^{k}{\rm d}^{d}p_{i}\,\tilde{\varphi}(p_{i})\ \delta^{d}\big{(}p_{(k)}\big{)}\,\sum_{i=1}^{k}p_{k}{\\!}^{2}\,,\quad k=2,3,\dots\,.$ (8.9) The result for ${\cal O}_{k0}$ may be expressed in terms of Hermite polynomial as a consequence of the identity $e^{-\frac{1}{4}\,\frac{{\rm d}^{2}}{{\rm d}x^{2}}}(2x)^{n}=H_{n}(x)\,.$ (8.10) The nonlinear term in (8.1) may be evaluated in terms operators ${\cal O}$ and ${\cal O}^{\prime}$, respectively of degree $l$ and $m$ in $\tilde{\varphi}$ and as given by (8.5) and (8.6), by considering [17] $\displaystyle e^{{\cal Y}}\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,K^{\prime}(p^{2})\,\frac{\delta{\cal O}}{\delta\tilde{\varphi}(p)}\,\frac{\delta{\cal O}^{\prime}}{\delta\tilde{\varphi}(-p)}$ $\displaystyle{}=\exp\bigg{(}\frac{1}{(2\pi)^{d}}\\!\int\\!{\rm d}^{d}p\,G(p)\,\frac{\delta^{2}}{\delta\tilde{\varphi}(p)\delta\tilde{\varphi}^{\prime}(-p)}\bigg{)}\,lm\frac{1}{(2\pi)^{d(l+m-3)}}\\!\int\prod_{i=1}^{l-1}{\rm d}^{d}p_{i}\,\tilde{\varphi}(p_{i})\\!\int\prod_{j=1}^{m-1}{\rm d}^{d}q_{j}\,\tilde{\varphi}^{\prime}(q_{j})$ $\displaystyle{}\\!\quad\times\delta^{d}\big{(}p_{(l-1)}+q_{(m-1)}\big{)}\ O\big{(}p_{1},\dots,p_{l-1},-p_{(l-1)}\big{)}\,O^{\prime}\big{(}q_{1},\dots,q_{m-1},-q_{(m-1)}\big{)}\Big{|}_{\tilde{\varphi}^{\prime}=\tilde{\varphi}}\,K^{\prime}\big{(}p_{(l-1)}{\\!\\!}^{2}\big{)}$ $\displaystyle{}=\sum_{n\geq 0}\frac{l!\,m!}{n!(l-1-n)!(m-1-n)!}\,\frac{1}{(2\pi)^{d(l+m-3-2n)}}\int\prod_{i=1}^{l-1-n}\\!\\!\\!{\rm d}^{d}p_{i}\,\tilde{\varphi}(p_{i})\,\int\prod_{j=1}^{m-1-n}\\!\\!\\!{\rm d}^{d}q_{j}\,\tilde{\varphi}(q_{j})\,$ $\displaystyle{}\qquad\times\delta^{d}\big{(}p_{(l-1-n)}+q_{(m-1-n)}\big{)}\,F_{n}(p_{1},\dots,p_{l-1-n};q_{1},\dots,q_{m-1-n})\,,$ (8.11) where, with $p_{(l-1-n)}+q_{(m-1-n)}=0$, $\displaystyle{}F_{n}(p_{1},\dots,p_{l-1-n};q_{1},\dots,q_{m-1-n})$ $\displaystyle{}=\frac{1}{(2\pi)^{dn}}\int\prod_{h=1}^{n}{\rm d}^{d}r_{h}\,G(r_{h})\,O\big{(}p_{1},\dots,p_{l-1-n},r_{n},\dots,r_{1},-p_{(l-1-n)}-r_{(n)}\big{)}$ $\displaystyle\quad\quad{}\times O^{\prime}\big{(}q_{1},\dots,q_{m-1-n},-r_{n},\dots,-r_{1},-q_{(m-1-n)}-r_{(n)}\big{)}\,K^{\prime}\big{(}(p_{(l-1-n)}+r_{(n)})^{2}\big{)}\,.$ (8.12) For the cases of interest here $\displaystyle\hskip-14.22636pt{\cal O}={\cal O}_{l0}\,,\quad{\cal O}^{\prime}={\cal O}_{m0}\,,$ $\displaystyle\hskip-14.22636ptF_{n}(p_{1},\dots,p_{l-1-n};q_{1},\dots,q_{m-1-n})=\rho_{n}\big{(}p_{(l-1-n)}{\\!}^{2}\big{)}\,,$ $\displaystyle\hskip-14.22636pt{\cal O}={\cal O}_{l0}\,,\quad{\cal O}^{\prime}={\cal O}_{m2}\,,$ $\displaystyle\hskip-14.22636ptF_{n}(p_{1},\dots,p_{l-1-n};q_{1},\dots,q_{m-1-n})={\textstyle{\sum_{j=1}^{m-1-n}\\!q_{j}{\\!}^{2}}}\,\rho_{n}\big{(}p_{(l-1-n)}{\\!}^{2}\big{)}+\tau_{n}\big{(}p_{(l-1-n)}{\\!}^{2}\big{)}\,,$ (8.13) with $\displaystyle\rho_{n}(p^{2})={}$ $\displaystyle\frac{1}{(2\pi)^{dn}}\int\prod_{h=1}^{n}{\rm d}^{d}r_{h}\,G(r_{h})\,K^{\prime}\big{(}(p+r_{(n)})^{2}\big{)}\,,$ (8.14a) $\displaystyle\tau_{n}(p^{2})={}$ $\displaystyle\frac{1}{(2\pi)^{dn}}\int\prod_{h=1}^{n}{\rm d}^{d}r_{h}\,G(r_{h})\,\big{(}n\,r_{n}{\\!}^{2}+(p+r_{(n)})^{2}\big{)}\,K^{\prime}\big{(}(p+r_{(n)})^{2}\big{)}\,.$ (8.14b) With the aid of (8.11) and (8.13) we may then write $\displaystyle\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,K^{\prime}(p^{2})\,\frac{\delta{\cal O}_{l0}}{\delta\tilde{\varphi}(p)}\frac{\delta{\cal O}_{m0}}{\delta\tilde{\varphi}(-p)}=\sum_{k}\big{(}{\cal O}_{k0}\,C_{klm}+{\cal O}_{k2}\,{\tilde{C}}_{klm}+\dots\big{)}\,,$ $\displaystyle\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,K^{\prime}(p^{2})\,\frac{\delta{\cal O}_{l0}}{\delta\tilde{\varphi}(p)}\frac{\delta{\cal O}_{m2}}{\delta\tilde{\varphi}(-p)}=\sum_{k}\big{(}{\cal O}_{k0}\,D_{klm}+{\cal O}_{k2}\,{\tilde{D}}_{klm}+\dots\big{)}\,,$ (8.15) for $\displaystyle C_{klm}={}$ $\displaystyle\frac{l!\,m!}{(s-1-k)!(s-l)!(s-m)!}\,\rho_{s-1-k}(0)\,,$ $\displaystyle{\tilde{C}}_{klm}={}$ $\displaystyle\frac{l!\,m!}{(s-1-k)!(s-1-l)!(s-1-m)!}\,\rho^{\prime}{\\!}_{s-1-k}(0)\,,$ $\displaystyle D_{klm}={}$ $\displaystyle\frac{l!\,(m-2)!}{(s-1-k)!(s-l)!(s-m)!}\,\tau_{s-1-k}(0)\,,$ $\displaystyle{\tilde{D}}_{klm}={}$ $\displaystyle\frac{l!\,(m-2)!}{(s-1-k)!(s-1-l)!(s-m)!}\,\big{(}(k-1)\rho_{s-1-k}(0)+\tau^{\prime}{\\!}_{s-1-k}(0)\big{)}\,,$ (8.16) which are non zero so long as $s={\textstyle\frac{1}{2}}(k+l+m)=k+1,k+2,\dots$. The truncation of the full RG equation (8.1) corresponding to the derivative expansion as considered in Section 6 is obtained by writing $S=\textstyle{\sum_{k}}\big{(}a_{k}\,{\cal O}_{k0}+b_{k}\,{\cal O}_{k2}\big{)}\,,$ (8.17) and then reducing (8.1) to $\displaystyle{\dot{a}}_{k}={}$ $\displaystyle\big{(}d-{\textstyle\frac{1}{2}}(d-2+\eta)k\big{)}\,a_{k}+\textstyle{\sum_{l,m}\big{(}C_{klm}\,a_{l}a_{m}+2D_{klm}\,a_{l}b_{m}}\big{)}\,,$ (8.18a) $\displaystyle{\dot{b}}_{k}={}$ $\displaystyle\big{(}d-2-{\textstyle\frac{1}{2}}(d-2+\eta)k\big{)}\,b_{k}-{\textstyle\frac{1}{2}}\eta\,\delta_{k2}+\textstyle{\sum_{l,m}\big{(}{\tilde{C}}_{klm}\,a_{l}a_{m}+2{\tilde{D}}_{klm}\,a_{l}b_{m}}\big{)}\,.$ (8.18b) These equations are arbitrary up to the rescalings $\displaystyle a_{k}\to a_{k}/\alpha_{k}\,,\ b_{k}\to b_{k}/\beta_{k}\,,\quad C_{klm}\to{}$ $\displaystyle C_{klm}\alpha_{l}\alpha_{m}/\alpha_{k}\,,\quad D_{klm}\to D_{klm}\alpha_{l}\beta_{m}/\alpha_{k}\,,$ $\displaystyle{\tilde{C}}_{klm}\to{}$ $\displaystyle{\tilde{C}}_{klm}\alpha_{l}\alpha_{m}/\beta_{k}\,,\quad{\tilde{D}}_{klm}\to{\tilde{D}}_{klm}\alpha_{l}\beta_{m}/\beta_{k}\,,$ (8.19) for any $\alpha_{k},\beta_{k}$ so long as $\beta_{2}=1$ because of the inhomogeneous $\eta$ term in (8.18b). Deferring further discussion to later we may impose $b_{2}=0$. Letting $\alpha_{k}=u^{\frac{1}{2}k-1}v$, $\beta_{k}=u^{\frac{1}{2}k-1}$ then these rescalings correspond to changes in the cut off dependent functions in (8.16) of the form $\rho_{n}(0)\to u^{n}v\,\rho_{n}(0)\,,\quad\rho^{\prime}{\\!}_{n}(0)\to u^{n}v^{2}\,\rho^{\prime}{\\!}_{n}(0)\,,\quad\tau_{n}(0)\to u^{n}\,\tau_{n}(0)\,,\quad\tau^{\prime}{\\!}_{n}(0)\to u^{n}v\,\tau^{\prime}{\\!}_{n}(0)\,.$ (8.20) Critical exponents calculated from (8.18a),(8.18b) must be independent of such transformations. The analysis of (8.18a),(8.18b) in the $\varepsilon$-expansion, with $d$ as in (2.6), is very similar to the previous discussion in the context of the derivative expansion. At a fixed point a consistent solution is obtained with $a_{*k}=\varepsilon\,a_{*}\delta_{k\,2n}+\varepsilon^{2}\,a_{*k}{\\!}^{(2)}+{\rm O}(\varepsilon^{3})\,,\quad b_{*k}=\varepsilon^{2}\,b_{*k}{\\!}^{(2)}+{\rm O}(\varepsilon^{3})\,,\qquad\eta=\varepsilon^{2}\,\eta^{(2)}+{\rm O}(\varepsilon^{3})\,,$ (8.21) where $\displaystyle a_{*}={}$ $\displaystyle-\frac{n-1}{C_{2n\,2n\,2n}}\,,\qquad\quad\eta^{(2)}=2a_{*}{\\!}^{2}{\tilde{C}}_{2\,2n\,2n}\,,$ $\displaystyle b_{*2p}{\\!}^{(2)}={}$ $\displaystyle\frac{n-1}{2(p-1)}\,a_{*}{\\!}^{2}{\tilde{C}}_{2p\,2n\,2n}\,,\ p\neq 1\,,\qquad a_{*2p}{\\!}^{(2)}=\frac{n-1}{2(p-n)}\,a_{*}{\\!}^{2}{C}_{2p\,2n\,2n}\,,\ p\neq n\,,$ $\displaystyle a_{*2n}{\\!}^{(2)}={}$ $\displaystyle-\frac{2a_{*}}{n-1}\Big{(}{\textstyle{\sum_{p\neq n}}}C_{2n\,2n\,2p}\,a_{*2p}{\\!}^{(2)}+{\textstyle{\sum_{p\neq 1}}}D_{2n\,2n\,2p}\,b_{*2p}{\\!}^{(2)}\Big{)}\,.$ (8.22) Hence with (8.16) this gives $\eta^{(2)}=4(n-1)^{2}\frac{n!^{6}}{(2n)!^{3}}\,\frac{2n-1}{n}\,\frac{\rho^{\prime}{\\!}_{2n-1}(0)}{\rho_{n-1}(0)^{2}}\,.$ (8.23) In general $\rho_{n}$, and also $\tau_{n}$, as defined in (8.14a) and (8.14b) depend on the cut off function but as shown in Appendix C there are certain universal quantities which are related to logarithmic divergences. In particular $\frac{\rho^{\prime}{\\!}_{2n-1}(0)}{\rho_{n-1}(0)^{2}}\bigg{|}_{d=d_{n}}=\frac{n}{2n-1}=\frac{d_{n}}{d_{n}+2}\,,$ (8.24) so that (8.23) is in exact accord with (2.18). In the neighbourhood of a fixed point the solutions of (8.18a), (8.18b) are written as $a_{k}(t)=a_{*k}+e^{\lambda t}f_{k}\,,\qquad b_{k}(t)=b_{*k}+e^{\lambda t}g_{k+2(n-1)}\,,$ (8.25) for $f_{k},g_{k+2(n-1)}$ small. The critical exponents are then determined by the linear eigenvalue equation $\frac{2n-k}{n-1}\,\left(\begin{array}[]{cc}f_{k}\\\ g_{k}\\\ \end{array}\right)+\sum_{l}\left(\begin{array}[]{cc}{\cal M}_{11,kl}&{\cal M}_{12,kl}\\\ {\cal M}_{21,kl}&{\cal M}_{22,kl}\\\ \end{array}\right)\left(\begin{array}[]{cc}f_{l}\\\ g_{l}\\\ \end{array}\right)=\lambda\left(\begin{array}[]{cc}f_{k}\\\ g_{k}\\\ \end{array}\right)\,,$ (8.26) for $\displaystyle{\cal M}_{11,kl}={}$ $\displaystyle\big{(}\varepsilon({\textstyle\frac{1}{2}}k-1)-{\textstyle\frac{1}{2}}k\eta\big{)}\delta_{kl}+2{\textstyle{\sum_{m}}}\big{(}C_{klm}\,a_{*m}+D_{klm}\,b_{*m}\big{)}\,,$ $\displaystyle{\cal M}_{22,kl}={}$ $\displaystyle\big{(}\varepsilon({\textstyle\frac{1}{2}}k-n)-({\textstyle\frac{1}{2}}k-n+1)\eta\big{)}\delta_{kl}+2{\textstyle{\sum_{m}}}{\tilde{D}}_{k-2(n-1)\ m\ l-2(n-1)}\,a_{*m}^{\vphantom{g}}\,,$ $\displaystyle{\cal M}_{12,kl}={}$ $\displaystyle 2{\textstyle{\sum_{m}}}D_{km\ l-2(n-1)}\,a_{*m}^{\vphantom{g}}\,,$ $\displaystyle{\cal M}_{21,kl}={}$ $\displaystyle 2{\textstyle{\sum_{m}}}\big{(}{\tilde{C}}_{k-2(n-1)\ lm}\,a_{*m}^{\vphantom{g}}+{\tilde{D}}_{k-2(n-1)\ lm}\,b_{*m}^{\vphantom{g}}\big{)}\,.$ (8.27) To first order in $\varepsilon$ we may let $a_{*m}\to\varepsilon\,a_{*}\delta_{m\,2n}$ and $b_{*m}\to 0$. Since $D_{k\,2n\,k-2(n-1)}=\frac{2n}{k+1-2n}\,\tau_{0}(0)=0\,,$ (8.28) as from (8.14b) $\tau_{0}(p^{2})=p^{2}K^{\prime}(p^{2})$, then ${\cal M}_{12,kl}$ is ${\rm O}(\varepsilon^{2})$, whereas otherwise ${\cal M}_{11,kl},{\cal M}_{21,kl}$ and ${\cal M}_{22,kl}$ are ${\rm O}(\varepsilon)$. Hence the first order critical exponents from (8.26) are $\displaystyle\lambda_{1,k}{\\!}^{(1)}={}$ $\displaystyle{\textstyle\frac{1}{2}}k-1+2a_{*}\,C_{kk\,2n}={\textstyle\frac{1}{2}}k-1-2(n-1)\frac{n!\,k!}{(2n)!\,(k-n)!}\,,$ $\displaystyle\lambda_{2,k}{\\!}^{(1)}={}$ $\displaystyle{\textstyle\frac{1}{2}}k-n+2a_{*}\,{\tilde{D}}_{k-2(n-1)\ 2n\ k-(2(n-1)}$ $\displaystyle={}$ $\displaystyle{\textstyle\frac{1}{2}}k-n-2(n-1)\frac{n!\,(k-2n)!}{(2n)!\,(k-3n+1)!}\bigg{(}k-2n+1+n\,\frac{\tau^{\prime}{\\!}_{n-1}(0)}{\rho_{n-1}(0)}\bigg{)}\,.$ (8.29) Just as in (8.24) $\tau^{\prime}{\\!}_{n-1}(0)\big{|}_{d=d_{n}}=0\,,$ (8.30) so that (8.29) is in exact agreement with the perturbative results (2.16) and (2.32). ## 9 Modified Derivative Expansion The results of Section 8 can be rewritten in a form which is close to the derivative expansion. To achieve this it is necessary to make specific choices of the cut off dependent quantities which appear in (8.16) and which are arbitrary up to the freedom exhibited in (8.20). It is crucial of course that the cut-off function independent results in (8.24) and (8.30), as well as $\tau_{0}(0)=0$, should be satisfied. To this end we choose $\rho_{n}(0)=-1\,,\qquad\rho^{\prime}{\\!}_{n}(0)=\frac{d}{d+2}\,,\qquad\tau_{n}(0)={\cal A}n\,,\qquad\tau^{\prime}{\\!}_{n}(0)=0\,,$ (9.1) where ${\cal A}$ is arbitrary. With these choices, and with (3.13) and (3.14), then (8.16) gives $\displaystyle C_{klm}={}$ $\displaystyle-2^{\frac{1}{2}(k-l-m)+1}\,\frac{1}{N_{k}}\,lm\,G_{k\,l-1\,m-1}\,,$ $\displaystyle{\tilde{C}}_{k+2\,lm}={}$ $\displaystyle\frac{d}{d+2}\,2^{\frac{1}{2}(k-l-m)+2}\,\frac{1}{N_{k}}\,l(l-1)m(m-1)\,G_{k\,l-2\,m-2}\,,$ $\displaystyle D_{kl\,m+2}={}$ $\displaystyle{\cal A}\,2^{\frac{1}{2}(k-l-m)+1}\,\frac{1}{N_{k}}\,l(l-1)G_{k\,l-2\,m}\,,$ $\displaystyle{\tilde{D}}_{k+2\,l\,m+2}={}$ $\displaystyle-2^{\frac{1}{2}(k-l-m)+1}\,\frac{1}{N_{k}}\,\big{(}lm\,G_{k\,l-1\,m-1}+l(l-1)\,G_{k\,l-2\,m}\big{)}\,.$ (9.2) If we now define ${\tilde{v}}(x)=\sum_{k}a_{k}\,2^{-\frac{1}{2}k}H_{k}(x)\,,\qquad{\tilde{z}}(x)=\sum_{k}b_{k+2}\,2^{-\frac{1}{2}k}H_{k}(x)\,,$ (9.3) then the truncated RG equations (8.18a) and (8.18b) are equivalent, subject to requiring non singular solutions for all $x$, to the coupled differential equations $\displaystyle{\dot{\tilde{v}}}={}$ $\displaystyle\big{(}d+{\textstyle\frac{1}{2}}(d-2+\eta)D\big{)}\tilde{v}-{\textstyle\frac{1}{2}}\,\tilde{v}^{\prime}{}^{2}+{\cal A}\,\tilde{v}^{\prime\prime}\tilde{z}\,,$ (9.4a) $\displaystyle{\dot{\tilde{z}}}={}$ $\displaystyle\big{(}-\eta+{\textstyle\frac{1}{2}}(d-2+\eta)D\big{)}\tilde{z}-\tilde{v}^{\prime}\tilde{z}^{\prime}-\tilde{v}^{\prime\prime}\tilde{z}-{\textstyle\frac{1}{2}}\eta+\frac{d}{4(d+2)}\,\tilde{v}^{\prime\prime}{}^{2}\,.$ (9.4b) With the redefinitions $v=\frac{2}{d-2+\eta}\,\tilde{v}\,,\qquad z=2\tilde{z}\,,$ (9.5) and, with $K,L$ defined as in (6.4), then these become $\displaystyle\frac{2}{d-2+\eta}\,{\dot{v}}={}$ $\displaystyle(D+K)v-{\textstyle\frac{1}{2}}\,v^{\prime}{}^{2}+{\tilde{\cal A}}\,v^{\prime\prime}z\,,$ (9.6a) $\displaystyle\frac{2}{d-2+\eta}\,{\dot{z}}={}$ $\displaystyle(D-L)z-v^{\prime}z^{\prime}-v^{\prime\prime}z-L+{\textstyle\frac{1}{2}}{\tilde{\cal B}}\,v^{\prime\prime}{}^{2}\,,$ (9.6b) for ${\tilde{\cal A}}=\frac{2}{d-2+\eta}\,{\cal A}\,,\qquad{\tilde{\cal B}}={\textstyle\frac{1}{2}}(d-2+\eta)\,\frac{d}{d+2}\,.$ (9.7) The results (9.6a) and (9.6b) are very similar to (6.3a) and (6.3b), although there is no linear $z$ term in (9.6a) and the coefficient ${\tilde{\cal B}}$ is determined in (9.6b). As a consequence the coefficient $c_{n}$ in the leading order solution (3.10) is unchanged from (3.16). Furthermore following the same discussion as in Sections 6 and 7 gives the correct values for $\eta$ to ${\rm O}(\varepsilon^{2})$. Thus instead of (7.10) $\Delta^{(0)}=\left(\begin{array}[]{cc}-D-2n&0\\\ 0&-D\\\ \end{array}\right)\,,$ (9.8) and replacing (7.15) $\Delta^{(1)}=\left(\begin{array}[]{cc}-(n-1)^{2}+c_{n}H_{2n}{\\!}^{\prime}\frac{{\rm d}}{{\rm d}x}&-{\tilde{\cal A}}\,c_{n}H_{2n}{\\!\\!}^{\prime\prime}\\\ -\tilde{{\cal B}}\,c_{n}H_{2n}{\\!\\!}^{\prime\prime}\frac{{\rm d}^{2}}{{\rm d}x^{2}}&c_{n}H_{2n}{\\!\\!}^{\prime}\frac{{\rm d}}{{\rm d}x}+c_{n}H_{2n}{\\!\\!}^{\prime\prime}\\\ \end{array}\right)\,.$ (9.9) It is then easy to see that this ensures the correct ${\rm O}(\varepsilon)$ result instead of (7.23b) as well as preserving (7.23a). In fact it is easy to verify that (7.7c), (7.7f) and (7.7i) still give exact eigenfunctions and eigenvalues for the linearised perturbations (9.6a) and (9.6b) about fixed points $v_{*},z_{*}$. These considerations may ensure that (9.6a) and (9.6b) have a greater chance of predictive success when they are analysed without using the $\varepsilon$-expansion. Of course setting $z,\eta$ to zero in (9.6a) reduces it to just the LPA. ## 10 Conclusion The status of the derivative expansion for exact RG flow equations is not entirely clear. In some respects it may be similar to effective field theories describing the large distance or low energy aspects of more fundamental theories. Having identified the relevant degrees of freedom and appropriate symmetries an effective lagrangian is constructed in terms of all symmetric scalars formed from the basic fields up to some scale dimension so as to reproduce physical amplitudes as far as contributions of the form $(E/\Lambda)^{p}$ for some $p$ where $E$ is a physical energy scale and $\Lambda$ a cut off [33]. The couplings which appear in the effective lagrangian can in principle be determined by matching the predictions of the effective theory with the fundamental theory for some specific physical amplitude. In a somewhat similar fashion a derivative expansion generates terms in the differential flow equations whose coefficients appear to depend on the cut off function and so are essentially arbitrary. A possible resolution is to match the results to those coming from the $\varepsilon$-expansion for $\varepsilon\to 0$, although the approximate flow equations may then be used for general $d$. The results in this paper essentially show how this can be achieved to ${\rm O}(\varepsilon)$ and to include thereby the universal aspects of all two vertex Feynman graphs. It would be interesting although non trivial to extend this to three vertex graphs. The results in Appendix A show how at this order various transcendental numbers arise which make achieving this for different multicritical points simultaneously hard to achieve. An important issue in using exact RG equations is to determine which solutions are physically relevant and give results independent of the particular RG equation or the detailed cut off function as an infra-red fixed point is approached and we may take $\Lambda\to\infty$. This question becomes more significant in approximation schemes when the symmetries of the original exact RG equation are no longer maintained and spurious solutions and critical exponents may be generated. The Polchinski RG equations for a fixed point action $S_{*}[\varphi]$ has an exactly marginal operator with zero critical exponent. This ensures that there is in general a line of physically equivalent fixed points $S_{*}[\varphi,a]$ depending on a parameter $a$. The exact marginal operator, which is constructed in detail in Appendix D, corresponds to an infinitesimal change in the scale of the field $\varphi$ under which the functional integral is invariant (conventionally the kinetic term in the action may be normalised to one but this is not essential, the physical couplings need only be redefined appropriately), for a further discussion see [29]. In the perturbative context the presence of such a marginal operator was demonstrated after (2.27) and is a property of the ${\rm O}(\varepsilon)$ results in (2.32) for $k=2n$. The presence of the irrelevant gauge parameter $a$ is in general necessary for the RG equations to determine $\eta$. If the symmetry under rescaling of the fields were to be maintained in a derivative expansion it would imply that critical exponents should be independent of $z_{*}(0)$ [34]. In the derivative expansion results obtained here in (6.5a), (6.5b), or the corresponding equations from (9.6a), (9.6b), there is a relation in general between $z_{*}(0)$ and $\eta$ so that $\eta$ is not determined unless $z_{*}(0)$ is fixed. At lowest order in $\varepsilon$ as in (6.6) the dependence on $z_{*}(0)$ disappears. Imposing $z_{*}(0)=0$ makes the equations well defined but is not a necessary requirement in general. The marginal operator constructed in Appendix D involves an integration over $\tilde{\varphi}(q)$ for all $q$ and so approximations such as the derivative approximation emphasising low $q$ fail to maintain the exact zero value for the critical exponent. Nevertheless preserving as far as possible the presence of a marginal operator is then a potential further constraint on solutions of exact RG equations in the derivative approximation [32] which may be used to restrict cut off dependence. Finally we note that the LPA has desirable features which are absent in any straightforward fashion in the derivative expansion. As shown in (5.11) the operator determining critical exponents for the LPA can be recast in self adjoint form. Related to this is the fact that for the LPA it is possible to construct a $C$-function from which the RG flow equations can be obtained [35] and that the equations can be written as a gradient flow [25, 36]. Whether this is true more generally remains to be demonstrated. ## Appendix A Further Perturbative Calculations The results in Section 2 may be extended to the the next order ${\rm O}(V^{3})$ in a similar fashion and we obtain some results here. The relevant contributions to $W$ are, with the same notation as (2.4), given by $\displaystyle W_{2}={}$ $\displaystyle-\frac{1}{6}\sum_{r,s,t\geq 1}\frac{1}{r!s!t!}\int{\rm d}^{d}x_{1}\,{\rm d}^{d}x_{2}\,{\rm d}^{d}x_{3}\,V^{(r+t)}(\varphi_{1})V^{(r+s)}(\varphi_{2})V^{(s+t)}(\varphi_{3})$ $\displaystyle\hskip 170.71652pt{}\times G_{0}(x_{12})^{r}G_{0}(x_{23})^{s}G_{0}(x_{31})^{t}$ $\displaystyle{}-\frac{1}{2}\sum_{r,s\geq 2}\frac{1}{r!s!}\int{\rm d}^{d}x_{1}\,{\rm d}^{d}x_{2}\,{\rm d}^{d}x_{3}\,V^{(r)}(\varphi_{1})V^{(r+s)}(\varphi_{2})V^{(s)}(\varphi_{3})\,G_{0}(x_{12})^{r}G_{0}(x_{23})^{s}$ $\displaystyle{}+\sum_{r\geq 2}\frac{1}{r!}\int{\rm d}^{d}x_{1}\,{\rm d}^{d}x_{2}\,V^{(r)}(\varphi_{1})\,G_{0}(x_{12})^{r}\,V^{(r)}_{{\rm{c.t.}1}}(\varphi_{2})\,,$ (A.1) where $V_{{\rm{c.t.}1}}$ is determined by the first term in (2.8). This removes subdivergencies arising in (A.1) for $r,s,t=n$ and we restrict $r,s,t<2n-1$ so that no further subtractions are necessary. The divergencies coming from the second term in (A.1) are easily obtained since if ${\cal R}$ is the usual operation defining a finite part, so that from (2.7a) ${\cal R}\big{(}G_{0}(x)^{n}\big{)}=G_{0}(x)^{n}-\frac{2}{\varepsilon}\,\frac{1}{(4\pi)^{n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{n-1}\,\delta^{d}(x)\,,$ (A.2) then in this term the finite part is given by ${\cal R}\big{(}G_{0}(x_{12})^{n}G_{0}(x_{23})^{n}\big{)}={\cal R}\big{(}G_{0}(x_{12})^{n}\big{)}\,{\cal R}\big{(}G_{0}(x_{23})^{n}\big{)}\,,$ (A.3) so that the divergent pole terms are given by $G_{0}(x_{12})^{n}G_{0}(x_{23})^{n}-{\cal R}\big{(}G_{0}(x_{12})^{n}G_{0}(x_{23})^{n}\big{)}$. For the first term in (A.1) there is an overall divergence for $d=d_{n}$ when $r+s+t=2n$. To analyse this we make use of the Mellin-Barnes representation [37] $\displaystyle\int{\rm d}^{d}x_{1}\,{\rm d}^{d}x_{2}\,e^{ik_{1}\cdot x_{1}+k_{2}\cdot x_{2}}\,G_{0}(x_{12})^{r}G_{0}(x_{2})^{s}G_{0}(x_{1})^{t}$ $\displaystyle{}=\frac{1}{(4\pi)^{r+s+t}}\,\frac{\Gamma(\nu)^{r+s+t}}{\Gamma(r\nu)\Gamma(s\nu)\Gamma(t\nu)}\,\frac{1}{\Gamma((r+s+t-1)\nu-1)}\,J_{rst}\bigg{(}\frac{k_{1}^{2}}{k_{3}^{2}},\frac{k_{2}^{2}}{k_{3}^{2}}\bigg{)}\,\bigg{(}\frac{k_{3}^{2}}{4\pi}\bigg{)}^{(r+s+t-2)\nu-2},$ $\displaystyle J_{rst}(u,v)=\frac{1}{(2\pi i)^{2}}\int_{\gamma-i\infty}^{\gamma+i\infty}\\!\\!\\!\\!\\!\\!\\!\\!{\rm d}y\int_{\gamma^{\prime}-i\infty}^{\gamma^{\prime}+i\infty}\\!\\!\\!\\!\\!\\!\\!\\!{\rm d}z\ \Gamma(-y)\,\Gamma(-z)$ $\displaystyle\hskip 85.35826pt{}\times\Gamma\big{(}(r+s-1)\nu-1-y\big{)}\,\Gamma\big{(}(r+t-1)\nu-1-z\big{)}$ $\displaystyle\hskip 85.35826pt{}\times\Gamma\big{(}y+z+1-(r-1)\nu\big{)}\,\Gamma\big{(}y+z+2-(r+s+t-2)\nu\big{)}\,u^{y}v^{z}\,,$ (A.4) where $k_{3}=-k_{1}-k_{2}$ and $\gamma,\gamma^{\prime}$ are chosen that the poles in $y,z$ are on the opposite side of the contours from those in $y+z$. The functions $J_{rst}(u,v)$ satisfy various symmetry relations, in particular $J_{rst}(u,v)=v^{(r+s+t-2)\nu-2}J_{srt}(u/v,1/v)=u^{(r+s+t-2)\nu-2}J_{tsr}(1/u,v/u)=J_{rts}(v,u)\,,$ (A.5) which are necessary to ensure that (A.4) is symmetric under permutations of $k_{1},k_{2},k_{3}$ and also $s,t,r$. For $d=d_{n}-\varepsilon$ as in (2.6), the poles in $\varepsilon$, reflecting divergences of relevance here, arise only from the residues of the poles at $y,z=0$. For $r,s,t\neq n,\ r+s+t=2n$ there is then a simple $\varepsilon$-pole arising from $\Gamma\big{(}2-2(n-1)\nu\big{)}$ since $\nu=\frac{1}{n-1}-{\textstyle\frac{1}{2}}\varepsilon$ which gives $\displaystyle G_{0}(x_{12})^{r}G_{0}(x_{23})^{s}G_{0}(x_{31})^{t}$ $\displaystyle\,\big{|}_{r+s+t=2n,r,s,t\neq n}\sim\frac{1}{\varepsilon}\,\frac{1}{(4\pi)^{2n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-1}K_{rst}\,\delta^{d}(x_{12})\delta^{d}(x_{13})\,,$ $\displaystyle K_{rst}=\frac{\Gamma\big{(}\frac{n-r}{n-1}\big{)}\,\Gamma\big{(}\frac{n-s}{n-1}\big{)}\,\Gamma\big{(}\frac{n-t}{n-1}\big{)}}{\Gamma\big{(}\frac{r}{n-1}\big{)}\,\Gamma\big{(}\frac{s}{n-1}\big{)}\,\Gamma\big{(}\frac{t}{n-1}\big{)}}\,.$ (A.6) For $r=n,\ s+t=n$ the $y,z=0$ residues in (A.4) have a double pole in $\varepsilon$ from $\Gamma\big{(}1-(n-1)\nu\big{)}$ as well as $\Gamma\big{(}2-(2n-2)\nu\big{)}$. Expanding in $\varepsilon$ gives $\displaystyle{\cal R}\big{(}G_{0}(x_{12})^{n}\big{)}\,$ $\displaystyle G_{0}(x_{23})^{s}G_{0}(x_{31})^{t}\,\big{|}_{s+t=n}$ $\displaystyle\sim{}$ $\displaystyle\frac{1}{\varepsilon^{2}}\,\frac{1}{(4\pi)^{2n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\big{(}-2+(n-1)^{2}\varepsilon+(n-1)L_{st}\,\varepsilon\big{)}\,\delta^{d}(x_{12})\delta^{d}(x_{13})\,,$ $\displaystyle L_{st}=\psi\Big{(}\frac{1}{n-1}\Big{)}-\psi\Big{(}\frac{s}{n-1}\Big{)}-\psi\Big{(}\frac{t}{n-1}\Big{)}+\psi(1)\,,\ s+t=n\,.$ (A.7) With the aid of (A.6) and (A.7) then the pole terms in (A.1) require $\displaystyle V_{\rm{c.t.}2}(\phi)={}$ $\displaystyle-\frac{1}{\varepsilon}\,\frac{1}{(4\pi)^{2n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-1}\,\frac{1}{6}\sum_{\genfrac{}{}{0.0pt}{}{r,s,t\geq 1,r,s,t\neq n}{r+s+t=2n}}\frac{K_{rst}}{r!s!t!}\ V^{(r+t)}(\phi)V^{(r+s)}(\phi)V^{(s+t)}(\phi)$ $\displaystyle+\frac{1}{\varepsilon^{2}}\,\frac{1}{(4\pi)^{2n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\,\frac{1}{n!}\sum_{\genfrac{}{}{0.0pt}{}{s,t\geq 1}{s+t=n}}\frac{1}{s!t!}\,\big{(}1-{\textstyle\frac{1}{2}}(n-1)^{2}\varepsilon-{\textstyle\frac{1}{2}}(n-1)L_{st}\big{)}$ $\displaystyle\hskip 184.9429pt{}\times V^{(n)}(\phi)V^{(n+s)}(\phi)V^{(n+t)}(\phi)$ $\displaystyle+\frac{2}{\varepsilon^{2}}\,\frac{1}{(4\pi)^{2n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\,\frac{1}{n!^{2}}\,V^{(n)}(\phi)^{2}\,V^{(2n)}(\phi)\,.$ (A.8) The double poles are in accord with standard RG equations from (2.9) $\bigg{(}\varepsilon-{\hat{\beta}}^{V}\\!\cdot\frac{\partial}{\partial V}+{\hat{\gamma}}_{\phi}\,\phi\frac{\partial}{\partial\phi}\bigg{)}\big{(}V(\phi)+V_{\rm{c.t.}}(\phi)\big{)}=0\,,$ (A.9) since we have at this order with (2.10) $\beta_{2}^{V}(\phi)-\gamma_{\phi,1}\,\phi V^{\prime}(\phi)+\beta_{1}^{V}\\!\cdot\frac{\partial}{\partial V}\,V_{\rm{c.t.}1}(\phi)=2(n-1)\varepsilon\,V_{\rm{c.t.}2}(\phi)\,.$ (A.10) From (A.8) we then obtain $\displaystyle\beta^{V}_{2}(\phi)={}$ $\displaystyle-\frac{1}{3}(n-1)\,\frac{1}{(4\pi)^{2n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-1}\\!\\!\\!\\!\sum_{\genfrac{}{}{0.0pt}{}{r,s,t\geq 1,r,s,t\neq n}{r+s+t=2n}}\frac{K_{rst}}{r!s!t!}\ V^{(r+t)}(\phi)V^{(r+s)}(\phi)V^{(s+t)}(\phi)$ $\displaystyle-(n-1)^{2}\,\frac{1}{(4\pi)^{2n}}\,\Gamma\Big{(}\frac{1}{n-1}\Big{)}^{2n-2}\,\frac{1}{n!}\sum_{\genfrac{}{}{0.0pt}{}{s,t\geq 1}{s+t=n}}\frac{1}{s!t!}\,\big{(}n-1+L_{st}\big{)}$ $\displaystyle\hskip 199.16928pt{}\times V^{(n)}(\phi)V^{(n+s)}(\phi)V^{(n+t)}(\phi)$ $\displaystyle+\gamma_{\phi,1}\,\phi V^{\prime}(\phi)\,.$ (A.11) For general $n$ it is not straightforward to analyse (A.11) further so we content ourselves for the simplest cases of $n=2,3$ which give $\beta_{2}^{V}(\phi)\big{|}_{n=2}=-{\textstyle\frac{1}{2}}\,\frac{1}{(4\pi)^{4}}\,V^{(2)}(\phi)V^{(3)}(\phi)^{2}+\gamma_{\phi,1}\,\phi V^{\prime}(\phi)\,,\quad\gamma_{\phi,1}={\frac{1}{12}}\,\frac{g^{2}}{(4\pi)^{4}}\,,$ (A.12) and $\displaystyle\beta_{2}^{V}(\phi)\big{|}_{n=3}={}$ $\displaystyle\frac{\pi^{2}}{(4\pi)^{6}}\Big{(}{\textstyle\frac{1}{6}}V^{(2)}(\phi)V^{(5)}(\phi)^{2}-{\textstyle\frac{1}{12}}\pi^{2}V^{(4)}(\phi)^{3}-4V^{(3)}(\phi)V^{(4)}(\phi)V^{(5)}(\phi)\Big{)}$ $\displaystyle{}+\gamma_{\phi,1}\,\phi V^{\prime}(\phi)\,,\qquad\qquad\gamma_{\phi,1}={\frac{1}{90}}\,\frac{g^{2}\pi^{2}}{(4\pi)^{6}}\,.$ (A.13) Using (A.12) we may obtain ${\rm O}(\varepsilon^{2})$ corrections to (2.15) and (2.16) for $n=2$ $\displaystyle\frac{3g_{*}}{(4\pi)^{2}}={}$ $\displaystyle\varepsilon+{\textstyle\frac{2}{3}}\,\varepsilon^{2}-2\eta\,,\qquad\eta={\textstyle\frac{1}{54}}\,\varepsilon^{2}+{\rm O}(\varepsilon^{3})\,,$ $\displaystyle{\hat{\gamma}}_{k}(g_{*})={}$ $\displaystyle-{\textstyle\frac{1}{2}}(k-2)\,\varepsilon+{\textstyle\frac{1}{2}}k\eta+\textstyle{\frac{1}{6}}k(k-1)(\varepsilon-2\eta)-\textstyle{\frac{1}{18}}k(k-1)(k-3)\,\varepsilon^{2}+{\rm O}(\varepsilon^{3})\,.$ (A.14) This agrees with standard results for $k=1,2,3$. Furthermore for $n=3$ $\displaystyle\frac{20}{3}\,\frac{\pi g_{*}}{(4\pi)^{3}}=2\varepsilon+{\textstyle\frac{9}{20}}\big{(}47+{\textstyle\frac{3}{2}}\pi^{2}\big{)}\,\varepsilon^{2}-3\eta\,,\qquad\eta={\textstyle\frac{1}{500}}\,\varepsilon^{2}+{\rm O}(\varepsilon^{3})\,,$ $\displaystyle{\hat{\gamma}}_{k}(g_{*})=-{\textstyle\frac{1}{2}}(k-2)\,\varepsilon+{\textstyle\frac{1}{2}}k\eta+\textstyle{\frac{1}{30}}k(k-1)(k-2)\big{(}\varepsilon-{\textstyle\frac{3}{2}}\eta\big{)}$ $\displaystyle\hskip 39.83368pt{}-\textstyle{\frac{3}{100}}k(k-1)(k-5)\big{(}k(k-2)-\textstyle{\frac{1}{24}}k(k-4)+\textstyle{\frac{3}{16}}\pi^{2}(k-2)\big{)}\,\varepsilon^{2}+{\rm O}(\varepsilon^{3})\,.$ (A.15) When $k\geq 2n$ the ${\rm O}(\varepsilon^{2})$ results are modified due to mixing effects. Without computing the ${\rm O}(\varepsilon^{2})$ terms in ${\hat{\gamma}}^{hh}_{k}(g_{*})$ in the matrix (2.27) we have for one eigenvalue $\omega_{1,k}={\hat{\gamma}}_{k}(g_{*})+\Delta\omega_{1,k}+{\rm O}(\varepsilon^{3})\,,\qquad\Delta\omega_{1,k}=\frac{{\gamma}^{gh}_{k,1}(g_{*})\,{\gamma}^{hg}_{k,1}(g_{*})}{{\hat{\gamma}}^{gg}_{k,1}(g_{*})-{\hat{\gamma}}^{hh}_{k,1}(g_{*})}\,.$ (A.16) This gives $\displaystyle\Delta\omega_{1,k}\big{|}_{n=2}={}$ $\displaystyle\textstyle{\frac{1}{6}}k(k-1)(k-2)\,\eta\,,\qquad\qquad\ \ k\geq 4\,,$ $\displaystyle\Delta\omega_{1,k}\big{|}_{n=3}={}$ $\displaystyle\textstyle{\frac{1}{30}}k(k-1)(k-3)(k-4)\,\eta\,,\quad\,k\geq 6\,.$ (A.17) with $\eta$ as in (A.14) and (A.15). We may note that in both cases the results are consistent with $\omega_{1,2n}={\hat{\beta}}^{g}{}^{\prime}(g_{*})$. By considering the residues in (A.4) at $y=1,z=0$ and $y=0,z=1$, and requiring $r+s+t=3n-1$, we may also determine directly higher order contributions to $\gamma_{\phi}$ although it is then necessary to include an additional counterterm for when $r,s,t=2n-1$ in (A.1). Such results are omitted as they are irrelevant in the context of this paper. ## Appendix B Verification of Vanishing of a Sum In the discussion of critical exponents in Section 5 consistency required that the sum appearing in (5.27) $S=\sum_{p=0}^{n}(2n-3p)\frac{(2p)!}{p!^{4}(n-p)!^{2}}\,,$ (B.1) should vanish. Although in the case where it arises here $n$ is even $S=0$ for any $n$. To show this directly we note that $S=\frac{1}{n!^{2}}\sum_{p=0}^{n}(2n-3p)\binom{2p}{p}\binom{n}{p}^{2}\,,$ (B.2) where we may write $\sum_{r=0}^{p}\binom{p}{r}^{2}=\binom{2p}{p}\,.$ (B.3) Hence $\displaystyle S={}$ $\displaystyle\frac{1}{n!^{2}}\sum_{p=0}^{n}\sum_{r=0}^{p}(2n-3p)\binom{p}{r}^{2}\binom{n}{p}^{2}$ $\displaystyle={}$ $\displaystyle\sum_{p=0}^{n}\sum_{r=0}^{p}(2n-3p)\frac{1}{r!^{2}(p-r)!^{2}(n-p)!^{2}}$ $\displaystyle={}$ $\displaystyle\sum_{s=0}^{n}\sum_{r=0}^{n-s}(2n-3s-3r)\frac{1}{r!^{2}s!^{2}(n-s-r)!^{2}}\,,$ (B.4) where $s=p-r$. Then, setting $t=n-s-r$, $\displaystyle S=$ $\displaystyle\sum_{\begin{subarray}{c}r,s,t\geq 0\\\ r+s+t=n\end{subarray}}(2t-r-s)\frac{1}{r!^{2}s!^{2}t!^{2}}\,,$ (B.5) from which it follows using symmetry of the $r,s,t$-sums that $S=0$. This then implies (8.30). ## Appendix C Integrals and Cut Off Function Dependence In the discussion in section 8 the dependence on the cut off function was reduced to particular integrals such as appeared in (8.14a) and (8.14b). In general the presence of an arbitrary cut off function $K(p^{2})$, constrained only by $K(0)=1$ and rapid fall off for large $p^{2}$, ensures that they can take any value but in special cases the integrals are identical with the logarithmically divergent part of standard Feynman integrals and so they have a universal form independent of any particular $K(p^{2})$. Reinstating the cut off in the propagator which appears in (8.8) so that $G_{\Lambda}(p)=\frac{K(p^{2}/\Lambda^{2})}{p^{2}}\,,$ (C.1) then (8.14a) can be written as $\displaystyle\Lambda^{m(d-2)-2}\rho_{m}\big{(}p^{2}/\Lambda^{2}\big{)}={}$ $\displaystyle\frac{1}{\Lambda^{2}}\,\frac{1}{(2\pi)^{dm}}\\!\int\prod_{h=1}^{m}{\rm d}^{d}r_{h}\,G_{\Lambda}(r_{h})\,K^{\prime}\big{(}(p+r_{(m)})^{2}/\Lambda^{2}\big{)}$ $\displaystyle={}$ $\displaystyle-\frac{1}{2(m+1)}\,\Lambda\frac{\partial}{\partial\Lambda}\,\frac{1}{(2\pi)^{dm}}\\!\int\prod_{h=1}^{m}{\rm d}^{d}r_{h}\,G_{\Lambda}(r_{h})\,G_{\Lambda}(p+r_{(m)})\,.$ (C.2) The logarithmic divergencies present in the product of $m+1$ propagators for $d=d_{n}$ as in (2.6) when $m(d_{n}-2)=2+2l,\ l=0,1,\dots$ then generate a cut off independent result for $\rho_{m}^{(l)}(0)$. To obtain the detailed coefficients, following [17], the momentum space convolution integrals in (8.14a), (8.14b) are expressed in terms of the $x$-space propagator ${\tilde{G}}(x)=\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,e^{-ip\cdot x}\,G(p)\,.$ (C.3) and, using $(1+{\textstyle\frac{1}{2}}p\cdot\partial_{p})G(p)=K^{\prime}(p^{2})$, $-{\textstyle\frac{1}{2}}\big{(}d-2+x\cdot\partial_{x}\big{)}\,{\tilde{G}}(x)=\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,e^{-ip\cdot x}K^{\prime}(p^{2})\,.$ (C.4) Then (8.14a) becomes $\rho_{m}(p^{2})=-\frac{1}{2}\int{\rm d}^{d}x\,e^{ip\cdot x}\,{\tilde{G}}(x)^{m}\,\big{(}d-2+x\cdot\partial_{x}\big{)}\,{\tilde{G}}(x)\,.$ (C.5) This gives for the Taylor expansion coefficients at $p^{2}=0$ $\rho^{(l)}_{m}(0)=-e_{l}\,\frac{1}{2}\int{\rm d}^{d}x\,(x^{2})^{l}\,{\tilde{G}}(x)^{m}\,\big{(}d-2+x\cdot\partial_{x}\big{)}\,{\tilde{G}}(x)\,,\quad e_{l}=\frac{(-1)^{l}}{2^{2l}l!({\textstyle\frac{1}{2}}d)_{l}}\,,$ (C.6) or with $r^{2}=x^{2}$ $\rho^{(l)}_{m}(0)=-e_{l}\,\frac{S_{d}}{2(m+1)}\int_{0}^{\infty}\\!\\!\\!\\!{\rm d}r\,r^{d+2l-1}\big{(}(m+1)(d-2)+r\partial_{r}\big{)}\,{\tilde{G}}(x)^{m+1}\,,$ (C.7) for $S_{d}={2\pi^{\frac{1}{2}}}/{\Gamma({\textstyle\frac{1}{2}}d)}$. When $d+2l=(m+1)(d-2)$ the integrand is a total derivative and using ${\tilde{G}}(x)\sim\frac{1}{(d-2)S_{d}}\,\frac{1}{r^{d-2}}\quad\hbox{as}\quad r\to\infty\,,$ (C.8) then there is only a surface term for large $r$ giving $\rho^{(l)}_{(l+1)(n-1)}(0)\big{|}_{d=d_{n}}=-(-1)^{l}e_{l}\,\frac{n-1}{4(n+l(n-1))}\,\big{(}(d_{n}-2)S_{d_{n}}\big{)}^{-(l+1)(n-1)}\,.$ (C.9) This result directly implies (8.24). For $l=0,1$ the coefficients obtained in (C.9) correspond exactly to the pole terms in dimensional regularisation in (2.7a), (2.7b). In a similar vein from (8.14b) $\displaystyle\tau_{m}(p^{2})=\frac{1}{2}\int{\rm d}^{d}x\,e^{ip\cdot x}\,\Big{(}{}$ $\displaystyle m\,{\tilde{G}}(x)^{m-1}\partial^{2}{\tilde{G}}(x)\,\big{(}d-2+x\cdot\partial_{x}\big{)}\,{\tilde{G}}(x)$ $\displaystyle{}+{\tilde{G}}(x)^{m}\big{(}d+x\cdot\partial_{x}\big{)}\,\partial^{2}{\tilde{G}}(x)\Big{)}\,,$ (C.10) and $\tau^{(l)}_{m}(0)=e_{l}\,\frac{1}{2}S_{d}\int_{0}^{\infty}\\!\\!\\!\\!{\rm d}r\,r^{d+2l-1}\big{(}m(d-2)+d+r\partial_{r}\big{)}\,{\tilde{G}}(x)^{m}\partial^{2}{\tilde{G}}(x)\,.$ (C.11) In this case the condition for a the integrand to be a total derivative is $m(d-2)=2l$ but there is no correspond surface term as $\partial^{2}{\tilde{G}}(x)$ vanishes more rapidly than $r^{-d}$ as $r\to\infty$ and therefore $\tau^{(l)}_{l(n-1)}(0)\big{|}_{d=d_{n}}=0\,.$ (C.12) ## Appendix D Perturbations of Exact RG Flow Equations We here discuss perturbations of the exact RG flow equations in (8.1) which may be written, neglecting $C$, more succinctly in the form $\frac{\partial}{\partial t}S=({\cal D}_{1}+{\cal D}_{2})S+S*S-\eta\,\varphi\cdot K^{-1}\varphi\,,$ (D.1) where $\displaystyle S*S={}$ $\displaystyle\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,K^{\prime}(p^{2})\,\frac{\delta S}{\delta\tilde{\varphi}(p)}\,\frac{\delta S}{\delta\tilde{\varphi}(-p)}\,,$ $\displaystyle\varphi\cdot K^{-1}\varphi={}$ $\displaystyle{\textstyle\frac{1}{2}}\,\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,K(p^{2})^{-1}p^{2}\tilde{\varphi}(p)\tilde{\varphi}(-p)\,,$ (D.2) and using the definitions (8.3) save that now ${\cal D}_{1}=\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,\big{(}{\textstyle\frac{1}{2}}d+1-{\textstyle\frac{1}{2}}\eta+p\cdot\partial_{p}\big{)}\tilde{\varphi}(p)\,\frac{\delta}{\delta\tilde{\varphi}(p)}\,.$ (D.3) For a small variation $\delta S$ $\frac{\partial}{\partial t}\delta S=({\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S})\delta S\,,\qquad{\cal D}_{S}=\frac{2}{(2\pi)^{d}}\int{\rm d}^{d}p\,K^{\prime}(p^{2})\,\frac{\delta S}{\delta\tilde{\varphi}(-p)}\,\frac{\delta}{\delta\tilde{\varphi}(p)}\,.$ (D.4) At a fixed point $S\to S_{*}$ with ${\dot{S}}_{*}=0$. The critical exponents are then defined by $\big{(}{\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S_{*}}\big{)}{\cal O}=\lambda{\cal O}\,,$ (D.5) for ${\cal O}$ the corresponding eigen-operator. From (D.1) and the definitions for ${\cal D}_{1},{\cal D}_{2},{\cal D}_{S_{*}}$ we easily obtain $\displaystyle\big{(}{\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S_{*}}\big{)}\tilde{\varphi}(q)={}$ $\displaystyle\big{(}{\textstyle\frac{1}{2}}d+1-{\textstyle\frac{1}{2}}\eta+q\cdot\partial_{q}\big{)}\tilde{\varphi}(q)+2K^{\prime}(q^{2})\,\frac{\delta S_{*}}{\delta\tilde{\varphi}(-q)}\,,$ $\displaystyle\big{(}{\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S_{*}}\big{)}\frac{\delta S_{*}}{\delta\tilde{\varphi}(-q)}={}$ $\displaystyle\big{(}{\textstyle\frac{1}{2}}d-1+{\textstyle\frac{1}{2}}\eta+q\cdot\partial_{q}\big{)}\frac{\delta S_{*}}{\delta\tilde{\varphi}(-q)}+\eta\,K(q^{2})^{-1}q^{2}\tilde{\varphi}(q)\,.$ (D.6) Hence there are two exact solutions of (D.5) $\displaystyle{\cal O}={}$ $\displaystyle\tilde{\varphi}(0)+\frac{2K^{\prime}(0)}{2-\eta}\,\frac{\delta S_{*}}{\delta\tilde{\varphi}(0)}\,,\qquad\lambda={\textstyle\frac{1}{2}}d+1-{\textstyle\frac{1}{2}}\eta\,,$ (D.7a) $\displaystyle{\cal O}={}$ $\displaystyle\frac{\delta S_{*}}{\delta\tilde{\varphi}(0)}\,,\hskip 93.89418pt\lambda={\textstyle\frac{1}{2}}d-1+{\textstyle\frac{1}{2}}\eta\,.$ (D.7b) These are identical with the results obtained in (7.7f) and (7.7i) using the derivative expansion. More generally we consider solutions of (D.5) which may be expressed as ${\cal O}_{\Psi}={\cal O}_{\Psi,1}+{\cal O}_{\Psi,2}\,,$ (D.8) where ${\cal O}_{\Psi,1}=\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}q\,\bigg{(}\Psi(q)\frac{\delta S_{*}}{\delta\tilde{\varphi}(q)}-\frac{\delta\Psi(q)}{\delta\tilde{\varphi}(q)}\bigg{)}\,,\quad{\cal O}_{\Psi,2}=\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}q\,K(q^{2})^{-1}q^{2}\,\Psi(q)\tilde{\varphi}(-q)\,.$ (D.9) For operators of this form a perturbation $\epsilon{\cal O}_{\Psi}$ may be removed by a redefinition of $\varphi$ in the basic functional integral $Z=\int{\rm d}[\varphi]\,e^{-\varphi\cdot K^{-1}\varphi-S_{*}[\varphi]}$ so that $Z$ is invariant. Such operators are termed redundant [17]. The operator in (D.7b) is of this form by taking $\Psi(q)\to(2\pi)^{d}\delta^{d}(q)$. For ${\cal O}_{\Psi,1}$ using ${\cal D}_{S}\frac{\delta S}{\delta\tilde{\varphi}(q)}=\frac{\delta}{\delta\tilde{\varphi}(q)}(S*S)\,,\quad\Big{[}{\cal D}_{1},\frac{\delta}{\delta\tilde{\varphi}(q)}\Big{]}=\big{(}q\cdot\partial_{q}+{\textstyle\frac{1}{2}}d-1+{\textstyle\frac{1}{2}}\eta\big{)}\frac{\delta}{\delta\tilde{\varphi}(q)}\,.$ (D.10) we have $\displaystyle({\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S_{*}}){\cal O}_{\Psi,1}={}$ $\displaystyle{\cal O}_{\Psi_{1},1}+\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}q\,\Psi(q)\,\frac{\delta}{\delta\tilde{\varphi}(q)}\big{(}({\cal D}_{1}+{\cal D}_{2})S_{*}+S_{*}*S_{*}\big{)}$ $\displaystyle\Psi_{1}(q)={}$ $\displaystyle\big{(}{\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S}-q\cdot\partial_{q}-{\textstyle\frac{1}{2}}d-1+{\textstyle\frac{1}{2}}\eta\big{)}\Psi(q)\,.$ (D.11) For ${\cal O}_{\Psi,2}$ $\displaystyle({\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S_{*}}){\cal O}_{\Psi,2}={\cal O}_{\Psi_{2},2}+{\cal O}_{\Psi_{3},1}\,,$ $\displaystyle\Psi_{2}(q)=\Psi_{1}(q)-\eta\Psi(q)+\Psi_{3}(q)\,,\qquad\Psi_{3}(q)=2K(q^{2})^{-1}K^{\prime}(q^{2})q^{2}\Psi(q)\,.$ (D.12) Hence using the equation for $S_{*}$ $\displaystyle({\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S_{*}}){\cal O}_{\Psi}={\cal O}_{\Psi^{\prime}}\,,\qquad\Psi^{\prime}(q)=\Psi_{1}(q)+\Psi_{3}(q)\,.$ (D.13) The result (D.13) demonstrates that the operators $\\{{\cal O}_{\Psi}\\}$ form a closed subspace under RG flow near a fixed point. If ${\cal O}(q)$ is a local operator satisfying the generalisation of (D.5) $\big{(}{\cal D}_{1}+{\cal D}_{2}+{\cal D}_{S_{*}}\big{)}{\cal O}(q)=\big{(}q\cdot\partial_{q}+\lambda_{\cal O}\big{)}{\cal O}(q)\,,$ (D.14) then taking $\Psi(q)=(q^{2})^{m}K(q^{2}){\cal O}(q)\,,$ (D.15) gives an eigen-operator ${\cal O}_{\Psi}$ with $\lambda=\lambda_{\cal O}-{\textstyle\frac{1}{2}}d-1+{\textstyle\frac{1}{2}}\eta-2m\,.$ (D.16) If $m$ were arbitrary the eigenvalue could take any value but for locality we require $m$ to be an integer. The operator in (D.7a) may be extended to all $q$ by considering ${\cal O}(q)=a(q^{2})\,\tilde{\varphi}(q)+b(q^{2})\,\frac{\delta S_{*}}{\delta\tilde{\varphi}(-q)}\,,$ (D.17) where $a(0)=1,\,b(0)=K^{\prime}(0)/(1-{\textstyle\frac{1}{2}}\eta)$. Imposing (D.14) with $\lambda_{\cal O}={\textstyle\frac{1}{2}}d+1-{\textstyle\frac{1}{2}}\eta$ gives $a^{\prime}(x)={\textstyle\frac{1}{2}}\eta K(x)^{-1}b(x)$, $K^{\prime}(x)a(x)-xb^{\prime}(x)=(1-{\textstyle\frac{1}{2}}\eta)b(x)$ which have the solutions, assuming $\eta<2$, $a(x)=\frac{1+xb(x)}{K(x)}\,,\qquad b(x)=x^{\frac{1}{2}\eta-1}K(x)\int_{0}^{x}\\!u^{-\frac{1}{2}\eta}\,\frac{K^{\prime}(u)}{K(u)^{2}}\ {\rm d}u\,.$ (D.18) When $\eta=0$, $a(x)=1,\ b(x)=(K(x)-1)/x$. With these results and using (D.17) and (D.15), with $m=0$, in (D.9) and (D.8) gives an exactly marginal eigen-operator with $\lambda=0$. Integrating these marginal deformations generates solutions $S_{*}[\varphi,a]$ for some parameter $a$ representing a line of equivalent fixed points. The various formulae may be verified with the Gaussian solution $S_{*}[\varphi,a]=-{\textstyle\frac{1}{2}}\,\frac{1}{(2\pi)^{d}}\int{\rm d}^{d}p\,\frac{p^{2}}{K(p^{2})+a}\,\tilde{\varphi}(p)\tilde{\varphi}(-p)\,.$ (D.19) More generally assuming the eigen-operators corresponding to $\lambda_{k,1}$ ${\cal O}_{k,1}$ may be extended to ${\cal O}_{k,1}(q)$ satisfying (D.14) then this construction determines ${\cal O}_{k+2n-1,2}$ so that from (D.16) $\lambda_{k,2}=\lambda_{k-2n+1,1}-{\textstyle\frac{1}{2}}d-1+{\textstyle\frac{1}{2}}\eta\,.$ (D.20) This is compatible with the ${\rm O}(\varepsilon)$ perturbative results and also the modified derivative expansion calculations described here. Acknowledgements J. O’D. is grateful to C. Bervillier and T. Morris for useful correspondence. ## References * [1] K.G. Wilson, “Renormalization Group and Critical Phenomena. 1. Renormalization Group and the Kadanoff Scaling Picture,” Phys. Rev. B4 (1971) 3174. * [2] F.J. Wegner and A. Houghton, “Renormalization Group Equation for Critical Phenomena,” Phys. Rev. A8 (1972) 401. * [3] K.G. Wilson and J.B. Kogut, “The Renormalization Group and the Epsilon Expansion,” Phys. Rept. 12 (1974) 75. * [4] J. Polchinski, “Renormalization and Effective Lagrangians,” Nucl. Phys. B231 (1984) 269. * [5] C. Bagnuls and C. Bervillier, “Exact Renormalization Group Equations. An Introductory Review,” Phys. Rep. 348 (2001) 91, hep-th/0002034. * [6] J. Berges, N. Tetradis and C. Wetterich, “Nonperturbative renormalization flow in quantum field theory and statistical physics,” Phys. Rept. 363 (2002) 223, hep-ph/0005122. * [7] J.M. Pawlowski, “Aspects of the Functional Renormalization Group,” hep-th/0512261. * [8] B. Delamotte, “An introduction to the nonperturbative renormalization group,” cond-mat/0702365. * [9] B. Delamotte, D. Mouhanna and M. Tissier, “Nonperturbative renormalization group approach to frustrated magnets,” Phys. Rev. B69 (2004) 134413, cond-mat/0309101. * [10] V.F. Muller, “Perturbative renormalization by flow equations,” Rev. Math. Phys. 15 (2003) 491, hep-th/0208211. * [11] A. Hasenfratz and P. Hasenfratz, “Renormalization Group Study of Scalar Field Theories,” Nucl. Phys. B270 (1986) 687. * [12] D.F. Litim, “Universality and the Renormalisation Group,” JHEP, 0507 (2005) 005, hep-th/0503096. * [13] T.R. Morris, “Equivalence of Local Potential Approximations,” JHEP, JHEP 0507 (2005) 027, hep-th/0503161. * [14] K.G. Wilson and M.E. Fisher, “Critical Exponents in 3.99 Dimensions,” Phys. Rev. Lett. 28 (1972) 240. * [15] H. Kleinert and V. Schulte-Frohlinde, “Critical Properties of $\phi^{4}$-Theories,” World Scientific, Singapore (2001). * [16] K. Pinn, A. Pordt and C. Wieczerkowski, “Computation of Hierarchical Renormalization-Group Fixed Points and Their $\varepsilon$-Expansions,” J. Statistical Physics 77 (1994) 977, hep-lat/9402020. * [17] F.J. Wegner in “Phase Transitions and Critical Phenomena, Vol. 6,” ed. C. Domb and M.S. Green, Academic Press, New York (1976). * [18] I. Jack and H. Osborn, “Two-loop Background Field Calculations for Arbitrary Background Fields,” Nucl. Phys. B207 (1982) 474. * [19] C. Itzykson and J.M. Drouffe, _Statistical Field Theory_ , Vol.1, appendix Chapter 5, (CUP, Cambridge, 1989). * [20] G.J. Huish and D.J. Toms, “Renormalization of Interacting Scalar Field Theory in Three-dimensional Curved Spacetime,” Phys. Rev. D49 (1994) 6767. * [21] G. Felder, “Renormalization Group in the Local Potential Approximation,” Comm. Math. Phys. 111 (1987) 101. * [22] P.C. Lima, “Renormalization Fixed Points in the Local Potential Approximation for $d\geq 3$,” Comm. Math. Phys. 170 (1995) 529. * [23] C.S.F. Harvey-Fros, “The Local Potential Approximation of the Renormalization Group,” University of Southampton Ph.D. Thesis (1999), hep-th/0108018. * [24] J.F. Nicholl, T.S. Chang and H.E. Stanley, “Approximate Renormalization Group Based on the Wegner-Houghton Differential Generator,” Phys. Rev. Lett. 33 (1974) 540, (E) 1525. * [25] P.E. Haagensen et al, “Gradient Flows from an Approximation to the Exact Renormalization Group,” Phys. Lett. B323 (1994) 330, hep-th/9310032. * [26] T.R. Morris, “Derivative Expansion of the Exact Renormalization Group,” Phys. Lett. B329 (1994) 241. * [27] C. Bervillier, “Status of the derivative expansion in the ERGE,” talk at the 3rd International Conference on the Exact Renormalization Group, Lefkada, Greece (2006), http://www.cc.uoa.gr/%7Epapost/Bervillier.pdf. * [28] R.D. Ball, P.E. Haagensen, J.I. Latorre and E. Moreno, “Scheme Independence and the Exact Renormalization Group,” Phys. Lett. B347 (1995) 80, hep-th/9411122. * [29] J. Commellas, “Polchinski equation, reparameterization invariance and the derivative expansion,” Nucl. Phys. B509 (1999) 662, hep-th/9705129. * [30] G.R. Golner, “Nonperturbative Renormalization Group Calculations for Continuum Spin Systems,” Phys. Rev. B33 (1986) 7863. * [31] E.K. Riedel, G.R. Golner and K.E. Newman, “Scaling Field Representation of Wilson’s Exact Renormalization-Group Equation,” Ann. Phys. 161 (1985) 178. * [32] K.E. Newman and E.K. Riedel, “Critical exponents by the scaling field method: The isotropic $N$-vector model in three dimensions,” Phys. Rev B30 (1984) 6615. * [33] H. Georgi, “Effective Field Theory,” Ann. Rev. Nucl. Part. Sci. 43 (1993) 209. * [34] T.R. Morris, “Properties of derivative expansion approximations to the renormalization group,” Int. J. Mod. Phys. B12 (1998) 1343, hep-th/9610012. * [35] J. Generowicz, C. Harvey-Fros and T.R. Morris, “C Function Representation of the Local Potential Approximation,” Phys. Lett. B407 (1997) 27, hep-th/9705088. * [36] G. Zumbach, “The Renormalization group in the local potential approximation and its applications to the O(n) model,” Nucl. Phys. B413 (1994) 771. * [37] A.I. Davydychev and J.B. Tausk, “A magic connection between massive and massless diagrams,” Phys. Rev. D53 (1996) 7381, hep-ph/9504431.

arxiv-papers

2024-09-04T02:48:53.111859

0709.0078

# Holding strategies in a bus-route model Scott A. Hill sahill@mailaps.org Department of Physics, Southern Methodist University ###### Abstract A major source of delays in public transportation is the clustering instability, which causes late buses to become progressively later while the buses trailing it become progressively earlier. In this paper, we study this instability and how to neutralize it using the common practices of holding and schedule slack. Starting with an on-time route, we delay one or more buses at a single stop, and determine how these delays grow over time. We compare the effects of two different types of holding on the stability of the system, and briefly investigate how our results change with the use of timepoints. ## I Introduction The typical bus or train system is intended to provide regular, periodic service, so that each stop has an arrival every $n$ minutes. In practice, however, public transportation often suffers from maddening inconsistencies, which discourages potential passengers from depending on it for their daily commute. Given the environmental, economic, and political impact of increased gasoline consumption, it is important for us to understand these delays and reduce their frequency, if possible. One cause of these delays is an inherent instabilityWelding in the dynamics of a bus route: when a single bus (bus A) is delayed, it has to pick up more passengers which delays it further, while the bus following it (bus B) has fewer passengers to pick up, and runs faster. In many cases, bus A is unable to recover and, on busy routes, bus B may even catch up to bus A, forming a _cluster_. This is not always a problem: for an evening outbound route, where most passengers board at the beginning of the route, it does not matter that the service is irregular so long as the passengers are gotten to their destinations quickly. However, in most circumstances, this instability results in longer, unpredictable passenger waits. Bus dispatchers can counter this behavior with _holding_ and _slack_. Buses which are running too fast (such as bus B above) are _held_ at a stop so that they do not catch up to the bus in front. In _schedule-based holding_ , buses are prevented from leaving a stop until a scheduled departure time, while in _headway-based holding_ , buses are prevented from leaving until the preceding bus is far enough away. Most bus services implement holding only at a few _time points_ along the route, while light rail services, which typically stop at every station anyway, can implement holding at every stop. If holding is only implemented when buses are running more quickly than usual, then it is only partially effective in preventing clusters: it may keep bus B from catching up to bus A, but it does not allow bus A to recover from its delay. However, if most buses are held, then _not_ holding a delayed bus may be enough to allow it to recover. This is done by introducing _slack_ into the schedule; that is, by allowing more time for buses to travel from stop to stop than is ordinarily necessary, so that the typical bus is held at every stop. The clustering instability was first mentioned in the engineering literature by WeldingWelding in 1957; since then, it has been covered extensively in the engineering literatureNewellPotts ; Newell76 ; Engineering , although not to the point of exhaustion. A few physicists, inspired perhaps by the much larger field of traffic studyHelbing , have turned their attention to modelling bus routesOLoan ; Krbalek ; Nagatani ; Nagatani-other ; Hill , using a number of methods (cellular automata, linear stability analysis, and chaos theory). The engineering literature has primarily focused on the statistical properties of multiple delays along an entire bus route. In this paper we take a more “microscopic” approach, by investigating the deterministic propagation of delays which occur at a single stop, on a bus route that is otherwise running on time. Using a model which is similar to others in the literatureNewellPotts ; Nagatani ; Hill , we calculate the maximum delay that a bus can recover from (a quantity we call the “buffer”) for a given amount of slack, and how this buffer varies when two or more consecutive buses are delayed at the same stop. We show that headway-based holding is typically better in recovering from random delays, although schedule-based holding allows buses to recover more quickly when the delays are shorter. We mostly assume that buses are held at every stop; however, we end by briefly showing the effect that timepoints have on the buffer of a single bus. ## II Model Figure 1: How buses and stops are indexed in this model. Consider a series of irregularly spaced bus stops, labelled with the index $s$, and a series of buses labelled with the index $b$. Each bus visits each stop in increasing order, and each stop is visited by each bus in increasing order (Fig. 1). We define $t_{b,s}$ to be the time (in minutes) at which bus $b$ departs stop $s$. Before bus $b$ can depart stop $s$, it must do three things: 1. 1. depart stop $s-1$, at time $t_{b,s-1}$; 2. 2. drive from stop $s-1$ to stop $s$, which takes time $T_{s}$ ; and 3. 3. pick up passengers at stop $s$. As a simplification, we ignore the time it takes for passengers to get off the bus: this corresponds to the situation in which the number of alighting passengers is either very small, or in which passengers can alight through a rear door at the same time other passengers are boarding. The time it takes to pick up passengers at stop $s$ is equal to the time it takes each passenger to board (the _unit boarding time_), multiplied by the number of passengers waiting at the stop. This number of passengers is, in turn, equal to the wait since the last bus ($t_{b,s}-t_{b-1,s}$) divided by the time it takes each passenger to arrive (the _interarrival time_ , the reciprocal of the arrival frequency). Thus, the time it takes to board passengers is $\hbox{boarding time}=\mu(t_{b,s}-t_{b-1,s}),$ (1) with the _passenger constant_ $\mu>0$ defined as $\mu={\hbox{unit boarding time}\over\hbox{interarrival time}}.$ (2) We can now specify a recursion relation for $t_{b,s}$: $t_{b,s}=t_{b,s-1}+T_{s}+\mu(t_{b,s}-t_{b-1,s}).$ (3) This can be simplified with a change of variable $\displaystyle t_{b,s}\to t_{b,s}+\sum_{i=0}^{s}T_{i}$, which eliminates $T_{s}$ from the equation entirely. If we then solve Eq. 3 for $t_{b,s}$ (and $T_{s}$ removed), we have $t_{b,s}=(1+\mu^{\prime})t_{b,s-1}-\mu^{\prime}t_{b-1,s},$ (4) where we define $\mu^{\prime}={\mu\over 1-\mu}.$ (5) To implement _holding_ , we must define a _schedule function_ $S_{b,s}$, which specifies the time at which bus $b$ _should_ depart from stop $s$. For simplicity, we assume a homogeneous schedule in which buses are evenly spaced: $S_{b,s}=b\Delta+s(\mu\Delta+\sigma).$ (6) The parameter $\Delta$ is the time between successive buses at a particular stop, $\mu\Delta$ is the time it takes a particular bus to travel from stop to stop under normal conditions, and $\sigma$ is the _slack_ built into the schedule111Note that $S_{b,s}$ would include an additional term $\sum_{k}T_{k}$, if we hadn’t normalized $T_{s}$ out of our equations.. If there is no slack ($\sigma=0$), then $t_{b,s}=S_{b,s}$ is a solution to Eq. 4. We implement _holding_ by specifying some _earliest departure time_ $t^{\min}_{b,s}$; if the bus is ready to leave before $t^{\min}$, it is held so that it leaves at $t^{\min}$. We do this by rewriting Eq. 4 as the conditional $t_{b,s}=\max\cases{(1+\mu^{\prime})t_{b,s-1}-\mu^{\prime}t_{b-1,s}\cr t^{\min}_{b,s}\cr}.$ (7) Note that our model will hold buses at _every_ stop, rather than at the less frequent _time points_ seen in most bus systems. It is convenient to work with the _delay_ of a bus at any given stop, rather than its departure time. We define the _unnormalized delay_ of bus $b$ at stop $s$ to be $\ell_{b,s}=t_{b,s}-S_{b,s},$ (8) and the _normalized delay_ (or, simply, the _delay_) to be $d_{b,s}={\mu\over\sigma}\ell_{b,s}.$ (9) Delay is measured relative to the schedule function Eq. 6; early buses have $d<0$. We rewrite Eq. 7 in terms of delay, to arrive at our main _dynamic equation_ $d_{b,s}=\max\cases{(1+\mu^{\prime})d_{b,s-1}-\mu^{\prime}d_{b-1,s}-\mu^{\prime}\cr cd_{b-1,s}\cr}$ (10) where $cd_{b-1,s}$ is the _minimum delay_ of bus $b$ at stop $s$. If $c=0$, then buses are never allowed to be early (and the delay $d_{b,s}$ is never negative), and we have schedule-based holding. If $c=1$, then buses are never allowed to be earlier than the bus preceding them ($d_{b,s}\geq d_{b-1,s}$), and we have headway-based holding. When the second of the two conditions in Eq. 10 is larger, we say that “holding has been triggered”. In the calculations that follow, it will be necessary to solve several recursion relations of the form $x_{s}=zx_{s-1}+Az^{s}+\gamma;$ (11) this has the solution $x_{s}={\gamma\over 1-z}+\left(x_{0}-{\gamma\over 1-z}\right)z^{s}+Asz^{s}.$ (12) ## III Solutions We now consider a case in which, following a series of on-time buses ($d_{b,s}=0$ for $s<0$ and $b<1$), one or more buses, starting with bus $b=1$, are delayed at stop $s=0$. ### III.1 The First Bus Given our assumption that $d_{b,s}=0$ for $b<1$, the dynamic equation (Eq. 10) for the first bus is $d_{1,s+1}=\max\left[(1+\mu^{\prime})d_{1,s}-\mu^{\prime},0\right].$ (13) If the bus is early or on-time at any stop $s_{0}$ ($d_{1,s_{0}}\leq 0$), it will be on-time from then on ($d_{1,s>s_{0}}=0$). As long as $d_{1,s}>0$, however, Eq. 13 has the solution (cf Eq. 12) $d_{1,s}=1-(1+\mu^{\prime})^{s}(1-d_{1,0})$ (14) which grows exponentially towards positive or negative infinity, depending on the sign of $1-d_{1,0}$. Thus there are two types of results (as shown in Fig. 2): 1. (a) If $d_{1,0}<1$, then the delay of the bus decreases until it is on time ($d_{1,s_{0}}\leq 0$), after which the bus remains on time. We say that such a bus is _recovering_. The bus has fully recovered when $s>s_{0}\equiv{-\ln(1-d_{1,0})\over\ln(1+\mu^{\prime})}.$ (15) Note that bus routes with larger numbers of passengers (larger values of $\mu^{\prime}$) recover more quickly (Fig. 2). A special case of this is when $s_{0}<1$, in which case the bus _recovers instantly_ ; this occurs when $d_{1,0}<\mu$. 2. (b) If $d_{1,0}>1$, then the delay of the bus increases exponentially; we say that such a bus is _unrecoverable_. A single unrecoverable bus delays all following buses: either the bus behind it catches up to it (if $c=0$), forming a cluster, or else it also becomes exponentially late (if $c=1$); in either case, the homogeneous behavior has broken down irrevocably. Figure 2: The normalized delay of the first bus, calculated from Eq. 13. In Fig. 2a, the passenger constant is fixed at $\mu^{\prime}=0.1$, while the initial delay is varied: examples of unrecoverable, recovering, and instantly recovering buses are shown. In Fig. 2b, the initial delay is fixed at $d_{1,0}=0.9$, while the passenger constant is varied. Notice that the bus recovers faster when the passenger constant is greater. We define the _buffer_ $\beta$ of a bus to be the largest delay $d_{b,0}$ that it is able to recover from; the buffer of the first bus is thus $\beta_{1}=1$. This is a normalized quantity based on the normalized delay $d_{b,s}$. In terms of the _unnormalized delay_ $\ell_{b,s}$ (Eq. 8), the bus recovers instantly when $\ell_{b,s}<\sigma$, because the bus need only eliminate part or all of its slack. The bus’s unnormalized buffer is $\sigma/\mu$: as NewellNewell76 explains it, if the slack per stop $\sigma$ is larger than the time $\mu\ell_{b,s}$ it takes to board the additional passengers due to the delay, then the bus will eventually recover. ### III.2 The Second Bus If the first bus is late, it will pick up some of the passengers originally destined for the second bus, and so the second bus will run faster. Therefore, if the second bus starts out on-time, it will have no trouble remaining on- time, so long as the first bus is not unrecoverable. It also stands to reason that if the second bus is delayed at stop $s=0$, then it will recover more quickly, and have a larger buffer $\beta$, than if the first bus were not also running late. The dynamic equation (Eq. 10) for the second bus is $d_{2,s}=\max\cases{(1+\mu^{\prime})d_{2,s-1}-\mu^{\prime}(d_{1,s}+1)\cr cd_{1,s}\cr},$ (16) If the first bus is unrecoverable, then the second bus will be unrecoverable as well, so we need only consider the case where $d_{1,0}<1$. If $d_{1,0}<\mu$, then the first bus recovers instantly, and the behavior of the second bus can be described using Section III.1. Thus we only need consider the case where ${\mu^{\prime}\over 1+\mu^{\prime}}<d_{1,0}<1$ (note that ${\mu^{\prime}\over 1+\mu^{\prime}}=\mu$). There are two cases to be considered: 1. (a) If the holding condition in Eq. 16 is triggered at some stop $s-1$, then $d_{2,s-1}=cd_{1,s-1}$. Now as long as the first bus hasn’t recovered, Eq. 14 can be rewritten as $d_{1,s-1}=(d_{1,s}+\mu^{\prime})/(1+\mu^{\prime})$ (17) and so $d_{2,s}=\max\cases{cd_{1,s}-\mu^{\prime}d_{1,s}-\mu(1-c)\cr cd_{1,s}}.$ (18) Clearly the upper term is smaller than the lower term, and so the holding condition is triggered at stop $s$ as well. Therefore, once the holding condition has been triggered for bus 2, it will continue to be triggered from there on out. If $c=0$, then the second bus recovers immediately; if $c=1$, then the second bus recovers along with the first bus. In either case, however, once the holding condition is triggered, both buses recover. 2. (b) If the second bus’s holding condition is never triggered, and the first bus hasn’t recovered yet (i.e. $s<s_{0}$), we can substitute Eq. 14 into Eq. 16 to get $d_{2,s}=(1+\mu^{\prime})d_{2,s-1}+\mu^{\prime}(1+\mu^{\prime})^{s}(1-d_{1,0})-2\mu^{\prime},$ (19) which, according to Eq. 12, has the solution $d_{2,s}=2+\left[\mu^{\prime}(1-d_{1,0})s-(2-d_{2,0})\right](1+\mu^{\prime})^{s}.$ (20) Once $s>s_{0}$, $d_{1,s}=0$ and the second bus follows the “first bus” pattern; therefore, the second bus will ultimately recover only if $d_{2,s_{0}}<1$. Since $d_{2,s_{0}}=2-{\mu^{\prime}\ln(1-d_{1,0})\over\ln(1+\mu^{\prime})}-{2-d_{2,0}\over 1-d_{1,0}},$ (21) the inequality $d_{2,s_{0}}<1$ holds if $d_{2,0}<\beta_{2}$ where $\beta_{2}=1+d_{1,0}+{\mu^{\prime}\over\ln(1+\mu^{\prime})}(1-d_{1,0})\ln(1-d_{1,0}).$ (22) Figure 3: A diagram showing the buffer of the second bus as a function of the first bus’s initial delay $d_{1,0}$. One can interpret this as a phase diagram: in the unshaded region, both buses recover, while in the shaded regions one or both buses are unrecoverable. The buffer depends on the passenger constant $\mu^{\prime}$, being smaller for larger passenger constants, but never going below 1. This figure is independent of the holding strategy $c$. Figure 3 shows a graph of the buffer of the second bus as a function of the first bus’s initial delay. The buffer’s curve marks the transition between the case in which both buses recover (below), and the phase in which one or both buses are unrecoverable (above and to the right): we see that when the first bus is very late (but recovering), the second bus can recover from a much larger delay. As the passenger constant $\mu^{\prime}$ increases, the region where both buses recover becomes smaller; this is because, for larger passenger constants, the first bus recovers more quickly (as seen in Fig. 2) and so the second bus does not get as much benefit from the first bus’s delay. ### III.3 The Third Bus Figure 4: Figure 4a shows the buffer of the third bus as a function of the delay of the second bus, and for two values of the delay of the first bus ($t_{1,0}=0.4$ and $t_{1,0}=0.8$); all three buses recover in the region underneath the corresponding buffer curve. Clearly the buffer of the third bus depends on the holding strategy $c$. Figure 4b shows how the three buses behave at the X marked in (a), with $t_{1,0}=0.8$. For schedule-based holding ($c=0$), the third bus is unrecoverable, while for headway-based holding ($c=1$), all three buses recover. The passenger constant is $\mu^{\prime}=0.1$ throughout. Although the recovery of the first two buses is independent of the holding strategy $c$, the same is not true for subsequent delayed buses. Fig. 4a–b shows the buffer of the third bus ($b=3$) as it depends on the initial delay of the first two buses; this clearly depends on the holding strategy. Fig. 4c shows why this difference exists: if schedule-based holding is in place, then the second bus recovers quickly from a small initial delay, and the third bus loses the benefit of following a late bus. Note how, when $c=0$, the third bus’s buffer approaches $1$ (the buffer of a bus following an undelayed bus) when $d_{2,0}$ approaches zero. When $c=1$ and $d_{2,0}$ is small and $d_{1,0}$ is large, on the other hand, the second bus can only recover as quickly as the first bus, giving the third bus a larger buffer. In a sense, the delay of a recovering bus is a _resource_ which makes the trailing buses more resistant to delay. ### III.4 Many Buses To expand our model to the many-bus limit in a manageable way, we consider the case where all buses are delayed by the same amount $\tau$ at stop $s=0$ (i.e. $d_{b,0}=\tau$ for all $b\geq 1$); this might be caused by construction or traffic on one particular street. If $\tau>\beta_{1}=1$, the first bus, and eventually all buses, will be unrecoverable; thus we only consider the case when $\tau<1$. Any bus with $d_{b,0}<1$ will ultimately recover, so our concern is not whether this situation recovers, but how quickly it recovers. For $c=0$ (Fig. 5), the solutions $d_{b,s}$ approach (in an alternating manner) the steady-state normalized solution $d_{b,s}=\tau-s\mu$ as $b$ increases. This solution can be better understood by looking at the corresponding unnormalized solution $\ell_{b,s}={\sigma\over\mu}\tau-s\sigma$: buses make up time by eliminating slack from the schedule, becoming $\sigma$ minutes earlier at each stop until they have recovered. Figure 5: The solution of Eq. 10 when $d_{b,0}=\tau$ and $c=0$; buses alternate above and below the steady-state solution, shown as a solid black line here. For $c=1$, we can prove that $d_{b,s}=d_{1,s}$ for all $s$: > Proof: by induction over $b$ and $s$. The base case $b=1$ is automatic, > while the base case $s=0$ is true because $d_{b,0}=d_{1,0}=\tau$. Suppose > that $d_{b^{\prime},s^{\prime}}=d_{1,s}$ for all $b^{\prime}<b$ and > $s^{\prime}<s$. Then Eq. 10 becomes > > > $d_{b,s}=\max\cases{(1+\mu^{\prime})d_{1,s-1}-\mu^{\prime}d_{1,s}-\mu^{\prime}\cr > d_{1,s}\cr}.$ (23) > > If $d_{1,s}=0$ and $d_{1,s-1}=0$, then $d_{b,s}=0$ and the hypothesis is > satisfied. If $d_{1,s-1}\neq 0$ but $d_{1,s}=0$, then > $(1+\mu^{\prime})d_{1,s-1}<\mu^{\prime}$ (according to Eq. 13), which means > that the first case in Eq. 23 is negative, and so $d_{b,s}=0=d_{1,s}$ and > the hypothesis is again satisfied. If $d_{1,s}\neq 0$ and $d_{1,s-1}\neq 0$ > then, according to Eq. 13, $(1+\mu^{\prime})d_{1,s-1}=d_{1,s}+\mu^{\prime}$, > and so > > $d_{b,s}=\max\cases{d_{1,s}+\mu^{\prime}-\mu^{\prime}d_{1,s}-\mu^{\prime}\cr > d_{1,s}\cr},$ (24) > > and since $(1-\mu^{\prime})d_{1,s}<d_{1,s}$, the second condition applies > and $d_{b,s}=d_{1,s}$. q.e.d. The steady-state solutions are thus $\lim_{b\to\infty}d_{b,s}=\cases{\tau-s{\mu^{\prime}\over 1+\mu^{\prime}}&$c=0$\cr 1-(1+\mu^{\prime})^{s}(1-\tau)&$c=1$\cr}.$ (25) It can be shown that, as long as $d_{b,1}>0$ for both cases, the $c=0$ case in Eq. 25 reaches zero faster than the $c=1$ case: this means that schedule-based holding allows the buses to recover more quickly than headway-based holding. ## IV Timepoints So far we have considered the situation where holding occurs at every stop. This may be a realistic description of a light-rail system, where trains stop at every stop anyway. However, a typical bus route may have a stop at every intersection, and a model which forces buses to stop and wait at each of these is unrealistic. In reality, most bus systems designate a few stops as _timepoints_ , and only implement holding there. To model this, we assume that every $N$th stop is a timepoint, and evaluate Eq. 10 for $b=1$, with the stipulation that the holding condition only applies when $s\equiv 0\bmod N$. As we vary $N$, we keep the amount of slack _per stop_ , which we call $\sigma$, constant; the amount of slack _per timepoint_ (that is, the amount of time a bus might actually have to wait while being held) is then $N\sigma$. A computer simulation calculates the (dimensionless) buffer $\beta$ of a “first bus” by determining the initial delay $d_{1,0}$ for which the system moves from recovering ($d_{1,1000}<10$) and unrecoverable ($d_{1,1000}>10$: $10$ and $1000$ are both arbitrarily chosen). Figure 6: The amount of slack necessary per _stop_ for a given amount of buffer $B$, both measured in minutes, as the number of stops between timepoints increases. The inset shows the amount of slack necessary per _timepoint_ (that is, $N\sigma/B$). Data points were found using computer simulation. If we let $B$ be the number of minutes corresponding to the dimensionless buffer $\beta$, then according to Eq. 9, $B={\sigma\over\mu}\beta\Longrightarrow{\sigma\over B}={\beta\over\mu}.$ (26) Figure 6 shows how this ratio $\sigma/B$ depends on the passenger constant $\mu^{\prime}$ and the timepoint spacing $N$; for example, we find that for a busy bus route with $N=16$ and $\mu^{\prime}=0.1$, the ratio $\sigma/B=0.21$, so the route must have 0.42 minutes’ slack per stop, or a 6.7 minutes’ slack per timepoint, to allow a bus to recover from a two-minute delay. For small passenger constants, the amount of slack per stop remains fairly constant, as if the holding from each stop can be “saved up” until the next timepoint. For larger passenger rates, however, holding becomes much less effective if the timepoints are spread too far apart. ## V Discussion When a single bus is delayed, it is able to recover if its delay is no larger than $\beta=1$ (or $B=\sigma/\mu$ minutes), which we call the bus’s buffer; the buffer is even larger for buses that trail already-delayed buses. If the number of passengers increases (such as during rush hour), the buffer will be reduced unless steps are taken to keep it steady, either by increasing slack, or by decreasing the unit boarding time (Eq. 2); this is part of the reason that bus routes are generally more unreliable during peak hours. Since slack increases the time it takes to complete a route, one would like to find strategies that make it as small as possible: for example, it might be optimal to keep the slack proportional to the passenger arrival rate, keeping the buffer $\beta=\mu/\sigma$ constant, so that busier stops get more slack. Alternatively, giving more slack to the stops immediately preceding or following the busier stops might be more advantageous; further study is required in this case. When it comes to choosing a holding strategy, both schedule and headway-based holding have advantages depending on the circumstances. When a series of buses experience the same delay at a given stop, schedule-based holding can allow the buses to recover more quickly as long as the delay is no larger than the buffer. However, headway-based holding may be more appropriate in scenarios with larger, random delays, as a very delayed bus will have a greater probability of trailing a slightly delayed bus, and have a better opportunity to recover. One interesting, possibly counterintuitive result of our analysis is that buses on busier routes recover more quickly from small delays. The passenger constant seems to set the “timescale” of the route’s behavior. Our next step will be to more fully investigate the effect of timepoints on these results. We have already shown that the necessary slack per stop increases dramatically when we only implement holding at every $N$th stop. Future work will determine the exact nature of this relationship, and the effect that timepoints have on the behavior and buffers of subsequent buses. The model presented here is a “microscopic” model, dealing on the effects of delays at a single stop; how these single-stop delays interact with one another over the course of an entire bus route is also something worth investigating. Although engineers have spent a lot of time investigating this “macroscopic” regime, there may yet be basic truths which can be uncovered by a physicist’s perspective. We thank Dr. Peter Furth of Northeastern University for useful conversations and his engineering perspective. ## References * (1) P.I. Welding. Operational Research 8: 133-142 (1957). * (2) G.F. Newell and R.B. Potts. Proc. Sec. Conf. Australian Road Research Board 2: 388-393 (1964). * (3) G.F. Newell. _Statistical Mechanics and Statistical Methods in Theory and Application_ , Uzi Landman, ed. 645-667. * (4) A recent review of the holding problem in the engineering literature is given by M.D. Hickman. Transportation Science 35: 215-237 (2001). * (5) D. Helbing. Rev. Mod. Phys 73: 1067 (2001). * (6) O.J. O’Loan, M.R. Evans, and M.E. Cates. Phys. Rev. E 58: 1404 (1998). * (7) M. Krbalek and P. Seba. J. Phys. A 36: L7 (2003) * (8) T. Nagatani. Physica A 296: 320 (2001). * (9) T. Nagatani has written a number of papers on buses, focusing recently on chaotic behavior that emerges in shuttle bus routes. See, for example, T. Nagatani. Physica A 319: 568-578 (2003). * (10) S.A. Hill. Physica A 328: 261 (2003).

arxiv-papers

2024-09-04T02:48:53.141785

0709.4546

On harmonic quasiconformal quasi-isometries M. Mateljević and M. Vuorinen FILE: main.tex, printed on: 2024-8-27, 16.07 Abstract. The purpose of this paper is to explore conditions which guarantee Lipschitz-continuity of harmonic maps w.r.t. quasihyperbolic metrics. For instance, we prove that harmonic quasiconformal maps are Lipschitz w.r.t. quasihyperbolic metrics. 2000 Mathematics Subject Classification. Primary 30C85. Secondary 30C65. Keywords. harmonic mapping, Lipschitz mapping, quasiconformal mapping, quasihyperbolic metric. ## 1 Introduction Let $G\subset{\mathbb{R}}^{2}$ be a domain and let $f:G\to{\mathbb{R}}^{2},f=(f_{1},f_{2}),$ be a harmonic mapping. This means that $f$ is a map from $G$ into ${\mathbb{R}}^{2}$ and both $f_{1}$ and $f_{2}$ are harmonic functions, i.e. solutions of the two-dimensional Laplace equation $\Delta u=0\,.$ (1.1) The Cauchy-Riemann equations, which characterize analytic functions, no longer hold for harmonic mappings and therefore these mappings are not analytic. Intensive studies during the past two decades show that much of the classical function theory can be generalized to harmonic mappings (see the recent book of Duren [9] and the survey of Bshouty and Hengartner [7]). The purpose of this paper is to continue the study of the subclass of quasiconformal and harmonic mappings, introduced by Martio in [31] and further studied for example in [32, 33, 34, 37, 38, 16, 2, 3, 19, 20, 21, 17]. The above definition of a harmonic mapping extends in a natural way to the case of vector-valued mappings $f:G\to{\mathbb{R}}^{n},f=(f_{1},\dots,f_{n}),$ defined on a domain $G\subset{\mathbb{R}}^{n},n\geq 2.$ We first recall the classical Schwarz lemma for the unit disk ${\mathbb{D}}=\\{z\in{\mathbb{C}}:|z|<1\\}\,$: 1.2. Lemma. Let $f:{\mathbb{D}}\to{\mathbb{D}}$ be an analytic function with $f(0)=0\,.$ Then $|f(z)|\leq|z|$, $z\in{\mathbb{D}}\,.$ For the case of harmonic mappings this lemma has the following counterpart. 1.3. Lemma. ([15], [9, p. 77]) Let $f:{\mathbb{D}}\to{\mathbb{D}}$ be a harmonic mapping with $f(0)=0\,.$ Then $|f(z)|\leq\frac{4}{\pi}{\mathrm{tan}}^{-1}|z|$ and this inequality is sharp for each point $z\in{\mathbb{D}}\,.$ The classical Schwarz lemma is one of the cornerstones of geometric function theory and it also has a counterpart for quasiconformal maps ([1, 26, 41, 45]). Both for analytic functions and for quasiconformal mappings it has a form that is conformally invariant under conformal automorphisms of ${\mathbb{D}}\,.$ In the case of harmonic mappings this invariance is no longer true. In general, if $\varphi:{\mathbb{D}}\to{\mathbb{D}}$ is a conformal automorphism and $f:{\mathbb{D}}\to{\mathbb{D}}$ is harmonic, then $\varphi\circ f$ is harmonic only in exceptional cases. Therefore one expects that harmonic mappings from the disk into a strip domain behave quite differently from harmonic mappings from the disk into a half-plane and that new phenomena will be discovered in the study of harmonic maps. For instance, it follows from Lemma 1 that holomorphic functions in plane do not increase hyperbolic distances. In general, planar harmonic mappings do not enjoy this property. On the other hand, we shall give here an additional hypothesis under which the situation will change, in the plane as well as in higher dimensions. It turns out that the local uniform boundedness property, which we are going to define, has an important role in our study. For a domain $G\subset{\mathbb{R}}^{n},n\geq 2,x,y\in G,$ let $r_{G}(x,y)=\frac{|x-y|}{\mathrm{min}\\{{d}(x),{d}(y)\\}}\,\textrm{where}\,d(x)=d(x,\partial G)\equiv\inf\\{|z-x|:z\in\partial G\\}\,.$ If the domain $G$ is understood from the context, we write $r$ instead $r_{G}$. This quantity is used, for instance, in the study of quasiconformal and quasiregular mappings, cf. [45]. It is a basic fact that [43, Theorem 18.1] for $n\geq 2,K\geq 1,c_{2}>0$ there exists $c_{1}\in(0,1)$ such that whenever $f:G\to fG$ is a $K$-quasiconformal mapping with $G,fG\subset{\mathbb{R}}^{n}$ then $x,y\in G$ and $r_{G}(x,y)\leq c_{1}$ imply $r_{fG}(f(x),f(y))\leq c_{2}.$ We call this property the local uniform boundedness of $f$ with respect to $r_{G}\,.$ Note that quasiconformal mappings satisfy the local uniform boundedness property and so do quasiregular mappings under appropriate conditions; it is known that one to one mappings satisfying the local uniform boundedness property may not be quasiconformal. We also consider a weaker form of this property and say that $f:G\to fG$ with $G,fG\subset{\mathbb{R}}^{n}$ satisfies the weak uniform boundedness property on $G$ (with respect to $r_{G}\,$) if there is a constant $c>0$ such that $r_{G}(x,y)\leq 1/2$ implies $r_{fG}(f(x),f(y))\leq c\,.$ Univalent harmonic mappings fail to satisfy the weak uniform boundedness property as a rule, see Example 2.3 below. We show that if $f:G\to fG$ is harmonic then $f$ is Lipschitz w.r.t. quasihyperbolic metrics on $G$ and $fG$ if and only if it satisfies the weak uniform boundedness property; see Theorem 2.5. The proof is based on a higher dimensional version of the Schwarz lemma: harmonic maps satisfy the inequality (2.15) below. An inspection of the proof of Theorem 2.5 shows that the class of harmonic mappings can be replaced by $OC^{1}$ class defined by (3.1) (see Section 3 below) and it leads to generalizations of the result; see Theorem 3. Another interesting application is Theorem 2.5 which shows that if $f$ is a harmonic $K$-quasiregular map such that the boundary of the image is a continuum containing at least two points, then it is Lipschitz. In Subsection 2.5, we study conditions under which a qc mapping is quasi isometry with respect to the corresponding quasihyperbolic metrics; see Theorems 2.6 and 2.6. In particular, using a quasiconformal analogue of Koebe’s theorem, cf. [4], we give a simple proof of the following result, cf. [30, 33]: if $D$ and $D^{\prime}$ are proper domains in $\mathbb{R}^{2}$ and $h:D\rightarrow D^{\prime}$ is $K$-qc and harmonic, then it is bi-Lipschitz with respect to quasihyperbolic metrics on $D$ and $D^{\prime}$. The results in this paper may be generalized into various directions. One direction is to consider weak continuous solutions of the $p$-Laplace equation $div(|\nabla u|^{p-2}\nabla u)=0,\quad 1<p<\infty,$ so called $p$-harmonic functions. Note that $2$-harmonic functions in the above sense are harmonic in the usual sense. It seems that the case of the upper half space is of particular interest, cf. [37, 33, 16, 3]. In Subsection 2.6, using Theorem 3.1 [23] we prove that if $h$ is a quasiconformal $p$-harmonic mapping of the upper half space $\mathbb{H}^{n}$ onto itself and $h(\infty)=\infty$, then $h$ is quasi- isometry with respect to both the Euclidean and the Poincaré distance. ## 2 Lipschitz property of harmonic maps w.r.t. quasihyperbolic metrics ### 2.1 Hyperbolic type metrics Let $B^{n}(x,r)=\\{z\in\mathbb{R}^{n}:|z-x|<r\\},S^{n-1}(x,r)=\partial B^{n}(x,r)$ and let $\mathbb{B}^{n},S^{n-1}$ stand for the unit ball and the unit sphere in $\mathbb{R}^{n}$, respectively. Sometimes we write $\mathbb{D}$ instead of $\mathbb{B}^{2}$. For a domain $G\subset\mathbb{R}^{n}$ let $\rho:G\to(0,\infty)$ be a continuous function. We say that $\rho$ is a weight function or a metric density if for every locally rectifiable curve $\gamma$ in $G,$ the integral $l_{\rho}(\gamma)=\int_{\gamma}\rho(x)ds$ exists. In this case we call $l_{\rho}(\gamma)$ the $\rho$-length of $\gamma.$ A metric density defines a metric $d_{\rho}:G\times G\to(0,\infty)$ as follows. For $a,b\in G,$ let $d_{\rho}(a,b)=\inf_{\gamma}l_{\rho}(\gamma)$ where the infimum is taken over all locally rectifiable curves in $G$ joining $a$ and $b.$ For a fixed $a,b\in G\,,$ suppose that there exists a $d_{\rho}$-length minimizing curve $\gamma:[0,1]\to G$ with $\gamma(0)=a,\gamma(1)=b$ such that $d_{\rho}(a,b)=l_{\rho}(\gamma|[0,t])+l_{\rho}(\gamma|[t,1])$ for all $t\in[0,1]\,.$ Then $\gamma$ is called a geodesic segment joining $a$ and $b\,.$ It is an easy exercise to check that $d_{\rho}$ satisfies the axioms of a metric. For instance, the hyperbolic (or Poincaré) metric of the unit ball $\mathbb{B}^{n}$ and the upper half space $\mathbb{H}^{n}=\\{x\in\mathbb{R}^{n}:x_{n}>0\\}$ are defined in terms of the densities $\rho(x)=2/(1-|x|^{2})$ and $\rho(x)=1/x_{n}\,,$ respectively. It is a classical fact that in both cases the length-minimizing curves, geodesics, exist and that they are circular arcs orthogonal to the boundary [6]. In both cases we have even explicit formulas for the distances: $\sinh\frac{\rho_{\mathbb{B}^{n}}(x,y)}{2}=\frac{|x-y|}{\sqrt{(1-|x|^{2})(1-|y|^{2})}}\,,\quad x,y\in\mathbb{B}^{n}\,,$ (2.1) and $\cosh{\rho_{\mathbb{H}^{n}}(x,y)}=1+\frac{|x-y|^{2}}{2x_{n}y_{n}}\,,\quad x,y\in\mathbb{H}^{n}\,.$ (2.2) Because the hyperbolic metric is invariant under conformal mappings, we may define the hyperbolic metric in any simply connected plane domain by using the Riemann mapping theorem, see for example [24]. The Schwarz lemma may now be formulated by stating that an analytic function from a simply connected domain into another simply connected domain is a contraction mapping, i.e. the hyperbolic distance between the images of two points is at most the hyperbolic distance between the points. The hyperbolic metric is often the natural metric in classical function theory. For the modern mapping theory, which also considers dimensions $n\geq 3\,,$ we do not have a Riemann mapping theorem and therefore it is natural to look for counterparts of the hyperbolic metric. So called hyperbolic type metrics have been the subject of many recent papers. Perhaps the most important of these metrics are the quasihyperbolic metric $k_{G}$ and the distance ratio metric $j_{G}$ of a domain $G\subset{\mathbb{R}^{n}}\,.$ They are defined as follows. 2.3. The quasihyperbolic and distance ratio metrics . Let $G\subset\mathbb{R}^{n}$ be a domain. The quasihyperbolic metric $k_{G}$ is a particular case of the metric $d_{\rho}$ when $\rho(x)=\frac{1}{d(x,\partial G)}$ (see [13, 12, 45]). It was proved in [12] that for given $x,y\in G\,,$ there exists a geodesic segment of length $k_{G}(x,y)$ joining them. The distance ratio metric is defined for $x,y\in G$ by setting $j_{G}(x,y)=\log(1+r_{G}(x,y))=\log(1+\frac{|x-y|}{\mathrm{min}\\{{d}(x),{d}(y)\\}})\,$ where $r_{G}$ is as in the Introduction. It is clear that $j_{G}(x,y)\leq r_{G}(x,y)\,.$ Some applications of these metrics are reviewed in [46]. The recent PhD theses [27], [24], [29] study the quasihyperbolic geometry or use it as a tool. 2.4. Lemma. ([13], [45, (3.4), Lemma 3.7]) Let $G$ be a proper subdomain of $\mathbb{R}^{n}\,.$ 1. (a) If $x,y\in G$ and $|y-x|\leq d(x)/2$, then $k_{G}(x,y)\leq 2j_{G}(x,y)\,.$ 2. (b) For $x,y\in G$ we have $k_{G}(x,y)\geq j_{G}(x,y)\geq\log\left(1+\frac{|y-x|}{d(x)}\right)$. ### 2.2 Quasiconformal and quasiregular maps 2.5. Maps of class ACL and ACLn . For each integer $k=1,...,n$ we denote $R_{k}^{n-1}=\\{x\in R^{n}:x_{k}=0\\}$. The orthogonal projection $P_{k}:\mathbb{R}^{n}\to\mathbb{R}_{k}^{n-1}$, is given by $P_{k}x=x-x_{k}\,e_{k}\,.$ Let $I=\\{x\in\mathbb{R}^{n}:a_{k}\leq x_{k}\leq b_{k}\\}$ be a closed $n$-interval. A mapping $f:I\rightarrow\mathbb{R}^{m}$ is said to be absolutely continuous on lines (ACL) if $f$ is continuous and if $f$ is absolutely continuous on almost every line segment in $I$, parallel to the coordinate axes. More precisely, if $E_{k}$ is the set of all $x\in P_{k}I$ such that the function $t\mapsto u(x+te_{k})$ is not absolutely continuous on $[a_{k},b_{k}]$, then $m_{n-1}(E_{k})=0$ for all $1\leq k\leq n$. If $\Omega$ is an open set in $\mathbb{R}^{n}$, a mapping $f:\Omega\rightarrow\mathbb{R}^{m}$ is absolutely continuous if $f|I$ is ACL for every closed interval $I\subset\Omega$. If $\Omega$ and $\Omega^{\prime}$ are domains in $\mathbb{\overline{R}}^{n}$, a homeomorphism $f:\Omega\rightarrow\Omega^{\prime}$ is called ACL if $f|\Omega\setminus\\{\infty,f^{-1}(\infty)\\}$ is ACL. If $f:\Omega\rightarrow\mathbb{R}^{m}$ is ACL, then the partial derivatives of $f$ exist a.e. in $\Omega$, and they are Borel functions. We say that $f$ is ACLn if the partials are locally integrable. 2.6. Quasiregular mappings . Let $G\subset\mathbb{R}^{n}$ be a domain. A mapping $f:G\to\mathbb{R}^{n}$ is said to be quasiregular (qr) if $f$ is ACLn and if there exists a constant $K\geq 1$ such that $|f^{\prime}(x)|^{n}\leq KJ_{f}(x)\;,\;\;|f^{\prime}(x)|=\max_{|h|=1}|f^{\prime}(x)h|\;,$ a.e. in $G$. Here $f^{\prime}(x)$ denotes the formal derivative of $f$ at $x\,,$ The smallest $K\geq 1$ for which this inequality is true is called the outer dilatation of $f$ and denoted by $K\,_{O}(f)$. If $f$ is quasiregular, then the smallest $K\geq 1$ for which the inequality $J_{f}(x)\leq K\,l(f^{\prime}(x))^{n}\;,\;\;l(f^{\prime}(x))=\min_{|h|=1}|f^{\prime}(x)h|\;,$ holds a.e. in $G$ is called the inner dilatation of $f$ and denoted by $K\,_{I}(f)$. The maximal dilatation of $f$ is the number $K(f)=\max\\{\,K\,_{I}(f),\,K\,_{O}(f)\,\\}$. If $K(f)\leq K$, then $f$ is said to be $K$–quasiregular ($K$–qr). If $f$ is not quasiregular, we set $K\,_{O}(f)=K\,_{I}(f)=K(f)=\infty\,.$ Let $\Omega_{1}$ and $\Omega_{2}$ be domains in $\mathbb{R}^{n}$ and fix $K\geq 1\,.$ We say that a homeomorphism $f:\Omega_{1}\rightarrow\Omega_{2}$ is a $K$-quasiconformal (qc) mapping if it is $K$-qr and injective. Some of the standard references for qc and qr mappings are [11], [26], [43], and [45]. These mappings generalize the classes of conformal maps and analytic functions to Euclidean spaces. The Kühnau handbook [25] contains several reviews dealing with qc maps. It should be noted that various definitions for qc maps are studied in [43]. The above definition of $K$-quasiconformality is equivalent to the definition based on moduli of curve families in [43, p. 42]. It is well-known that qr maps are differentiable a.e., satisfy condition (N) i.e. map sets of measure zero (w.r.t. Lebesgue’s $n$-dimensional measure) onto sets of measure zero. The inverse mapping of a $K$-qc mapping is also $K$-qc. The composition of a $K_{1}$-qc and of a $K_{2}$-qc map is a $K_{1}K_{2}$-qc map if it is defined. ### 2.3 Examples We first show that, as a rule, univalent harmonic mappings fail to satisfy the local uniform boundedness property. 2.7. Example . The univalent harmonic mapping $f:{\mathbb{H}}^{2}\to f({\mathbb{H}}^{2})\,,$ $f(z)=\mathrm{arg}\,z+i\,\mathrm{Im}z,$ fails to satisfy the local uniform boundedness property with respect to $r_{{\mathbb{H}}^{2}}\,.$ Let $z_{1}=\rho e^{i\pi/4}$, $z_{2}=\rho e^{i3\pi/4}$, $w_{1}=f(z_{1})$ and $w_{2}=f(z_{2})$. Then $r_{{\mathbb{H}}^{2}}(z_{1},z_{2})=2$ and $r_{f{\mathbb{H}}^{2}}(w_{1},w_{2})=\frac{\pi}{\sqrt{2}\,\rho}$ if $\rho$ is small enough and we see that $f$ does not satisfy the local uniform boundedness property. 2.8. Example. The univalent harmonic mapping $f:{\mathbb{H}}^{2}\to{\mathbb{H}}^{2}\,,$ $f(z)=\mathrm{Re}\,z\,\mathrm{Im}\,z+i\,\mathrm{Im}z,$ fails to satisfy the local uniform boundedness property with respect to $r_{{\mathbb{H}}^{2}}\,.$ For a harmonic mapping $f(z)=h(z)+\overline{g(z)}\,,$ we introduce the following notation $\lambda_{f}(z)=|h^{\prime}(z)|-|g^{\prime}(z)|\,,\quad\Lambda_{f}(z)=|h^{\prime}(z)|+|g^{\prime}(z)|\,\,\,{\rm and}\quad\nu(z)=g^{\prime}(z)/f^{\prime}(z).$ The following Proposition shows that a one to one harmonic function satisfying the local uniform boundedness property need not be quasiconformal. 2.9. Proposition . The function $f(z)={\mathrm{log}}(|z|^{2})+2iy$ is a univalent harmonic mapping and satisfies the local uniform boundedness property, but $f$ is not quasiconformal on $V=\\{z:x>1,0<y<1\\}$. Proof. It is clear that $f$ is harmonic in $\Pi^{+}=\\{z:\,{\rm Re\,}z>0\\}$. Next $f(z)=h(z)+\overline{g(z)}$, where $h(z)={\mathrm{log}}z+z$ and $g(z)={\mathrm{log}}z-z$. Since $h^{\prime}(z)=1+1/z$ and $g^{\prime}(z)=-1+1/z$, we have $|\nu(z)|<1$ for $z\in\Pi^{+}$. Moreover, $f$ is quasiconformal on every compact subset $D\subset\Pi^{+}$ and $\lambda_{f}$, $\Lambda_{f}$ are bounded from above and below on $D$. Therefore $f$ is a quasi-isometry on $D$ and by Theorem 2.5 below, $f$ satisfies the local uniform boundedness property on $D$. From now on we consider the restriction of $f$ to $V=\\{z=x+iy:x>1,0<y<1\\}$. Then $fV=\\{w=(u,v):u>\log(1+v^{2}/4),0<v<2\\}\,.$ We are going to show that: 1. $\bullet$ $f$ satisfies the local uniform boundedness property, but $f$ is not quasiconformal on $V$ . We see that $f$ is not quasiconformal on $V\,,$ because $|\nu(z)|\rightarrow 1$ as $z\rightarrow\infty,z\in V\,.$ For $s>1$, define $V_{s}=\\{z:1<x<s,0<y<1\\}$. Note that $f$ is qc on $V_{s}$ and therefore $f$ satisfies the property of local uniform boundedness on $V_{s}$ for every $s>1$. We consider separately two cases. Case A. $z\in V_{4}\,.$ If $r>1$ is big enough, then $d(z,\partial V_{r})=d(z,\partial V)$ and $d\big{(}f(z),\partial f(V_{r})\big{)}=d(f(z),\partial f(V))$ and therefore $f$ satisfies the property of local uniform boundedness on $V_{4}$ with respect to $r_{V}$. Case B. It remains to prove that $f$ satisfies the property of local uniform boundedness on $V\setminus V_{4}$ with respect to $r_{V}$. Observe first that for $z,z_{1}\in V$ and $|z_{1}|\geq|z|\geq 1\,,$ we have the estimate $\log\left(\frac{|z_{1}|}{|z|}\right)\leq\frac{|z_{1}|}{|z|}-1\leq|z_{1}-z|,$ and therefore for $z,z_{1}\in V$ $|f(z_{1})-f(z)|\leq 4|z_{1}-z|.$ (2.10) We write $\partial V=[1,1+i]\cup A\cup B\,;A=\\{(x,0):x\geq 1\\}\,,B=\\{(x,1):x\geq 1\\}\,.$ Then $\partial(fV)=f(\partial V)\subset f[1,1+i]\cup(fA)\cup(fB)$ and by the definition of $f$ we see that $fA=\\{(x,0):x\geq 0\\}\,,\quad fB=\\{(x,2):x\geq\log 2\\},\quad f[1,1+i]\subset[0,\log 2]\times[0,2]\,.$ Clearly for $w\in fV$ $d(w)=\min\\{d(w,fA),d(w,fB),d(w,f[1,1+i])\\}\,,$ and for ${\rm Re}w>1+\log{2}$ and $w\in fV\,,$ we find $d(w)=\min\\{d(w,fA),d(w,fB)\\}\,.$ (2.11) For $z\in V\setminus V_{4}$ we have ${\rm Re}f(z)\geq\log(16)>1+\log{2}$ and therefore, in view of the definition of $f$, (2.11) yields $d(f(z))=2d(z)$. This together with (2.10) shows that $f$ satisfies the property of local uniform boundedness on $V\setminus V_{4}$. $\Box$ ### 2.4 Higher dimensional version of Schwarz lemma Before giving a proof of the higher dimensional version of the Schwarz lemma for harmonic maps we first establish some notation. Suppose that $h:\overline{B}^{n}(a,r)\rightarrow\mathbb{R}^{n}$ is a continuous vector-valued function, harmonic on $B^{n}(a,r)$, and let $M_{a}^{*}=\sup\\{|h(y)-h(a)|:y\in S^{n-1}(a,r)\\}.$ Let $h=(h^{1},h^{2},\dots,h^{n})$. A modification of the estimate in [14, Equation (2.31)] gives $r|\nabla h^{k}(a)|\leq nM_{a}^{*}\,,\quad k=1,\dots,n.$ We next extend this result to the case of vector valued functions. See also [8] and [5, Theorem 6.16]. 2.12. Lemma. Suppose that $h:{\overline{B}^{n}(a,r)}\rightarrow\mathbb{R}^{n}$ is a continuous mapping, harmonic in ${B}^{n}(a,r)$. Then $r|h^{\prime}(a)|\leq nM_{a}^{*}\,.$ (2.13) Proof. Without loss of generality, we may suppose that $a=0$ and $h(0)=0$. Let $K(x,y)=K_{y}(x)=\frac{r^{2}-|x|^{2}}{n\omega_{n}\,r|x-y|^{n}},$ where $\omega_{n}$ is the volume of the unit ball $\mathbb{B}^{n}$ in ${\mathbb{R}}^{n}$. Then $h(x)=\int_{S^{n-1}(0,r)}K(x,t)h(t)d\sigma,\quad x\in{B}^{n}(0,r),$ where $d\sigma$ is the $(n-1)$-dimensional surface measure on $S^{n-1}(0,r)$. A simple calculation yields $\frac{\partial}{\partial x_{j}}K(x,\xi)=\frac{1}{n\omega_{n}\,r}\left(\frac{-2x_{j}}{|x-\xi|^{n}}-n(r^{2}-|x|^{2})\frac{x_{j}-\xi_{j}}{|x-\xi|^{n+2}}\right).$ Hence, for $1\leq j\leq n$, we have $\frac{\partial}{\partial x_{j}}K(0,\xi)=\frac{\xi_{j}}{\omega_{n}\,r^{n+1}}\,.$ Let $\eta\in S^{n-1}$ be a unit vector and $|\xi|=r$. For given $\xi$, it is convenient to write $K_{\xi}(x)=K(x,\xi)$ and consider $K_{\xi}$ as a function of $x$. Then $K_{\xi}^{\prime}(0)\eta=\frac{1}{\omega_{n}\,r^{n+1}}(\xi,\eta)\,.$ Since $|(\xi,\eta)|\leq|\xi||\eta|=r$, we see that $|K_{\xi}^{\prime}(0)\eta|\leq\frac{1}{\omega_{n}\,r^{n}}\,,\textrm{ and therefore\,}\quad|\nabla K^{\xi}(0)|\leq\frac{1}{\omega_{n}r^{n}}\,.$ This last inequality yields $|h^{\prime}(0)({\eta})|\leq\int_{{S^{n-1}(a,r)}}|\nabla K^{y}(0)|\,|h(y)|\,d\sigma(y)\leq\frac{M_{0}^{*}\,n\omega_{n}r^{n-1}}{\omega_{n}r^{n}}=\frac{M_{0}^{*}n}{r}$ and the proof is complete. $\Box$ Let $G\subset\mathbb{R}^{n},$ be a domain, let $h:G\to{\mathbb{R}}^{n}$ be continuous. For $x\in G\,$ let $B_{x}=B^{n}(x,\frac{1}{4}d(x))$ and $M_{x}=\omega_{h}(x)=\sup\\{|h(y)-h(x)|:y\in B_{x}\\}.$ (2.14) If $h$ is a harmonic mapping, then the inequality (2.13) yields $\frac{1}{4}d(x)|h^{\prime}(x)|\leq n\,\omega_{h}(x),\quad x\in G\,.$ (2.15) We also refer to (2.15) as the inner gradient estimate. ### 2.5 Harmonic quasiconformal quasi-isometries For our purpose it is convenient to have the following lemma. 2.16. Lemma. Let $G$ and $G^{\prime}$ be two domains in $\mathbb{R}^{n}$, and let $\sigma$ and $\rho$ be two continuous metric densities on $G$ and $G^{\prime}$, respectively, which define the elements of length $ds=\sigma(z)|dz|$ and $ds=\rho(w)|dw|$, respectively; and suppose that $f:G\rightarrow G^{\prime}$, is a $C^{1}$-mapping. a) If there is a positive constant $c_{1}$ such that $\rho(f(z))\,|f^{\prime}(z)|\leq c_{1}\,\sigma(z),$ $z\in G\,,$ then $d_{\rho}(f(z_{2}),f(z_{1}))\leq c_{1}\,d_{\sigma}(z_{2},z_{1})$, $z_{1},z_{2}\in G$. b) If $f(G)=G^{\prime}$ and there is a positive constant $c_{2}$ such that $\rho(f(z))\,l(f^{\prime}(z))\geq c_{2}\,\sigma(z),$ $z\in G\,,$ then $d_{\rho}(f(z_{2}),f(z_{1}))\geq c_{2}\,d_{\sigma}(z_{2},z_{1})$, $z_{1},z_{2}\in G$ The proof of this result is straightforward and it is left to the reader as an exercise. 2.17. Pseudo-isometry and a quasi-isometry . Let $f$ be a map from a metric space $(M,d_{M})$ into another metric space $(N,d_{N})$. * • We say that $f$ is a pseudo-isometry if there exist two positive constants $a$ and $b$ such that for all $x,y\in M$, $a^{-1}d_{M}(x,y)-b\leq d_{N}(f(x),f(y))\leq ad_{M}(x,y).$ * • We say that $f$ is a quasi-isometry or a bi-Lipschitz mapping if there exists a positive constant $a\geq 1$ such that for all $x,y\in M$, $a^{-1}d_{M}(x,y)\leq d_{N}(f(x),f(y))\leq ad_{M}(x,y).$ For the convenience of the reader we begin our discusssion for the unit disk case. 2.18. Theorem. Suppose that $h:{\mathbb{D}}\to{\mathbb{R}}^{2}$ is harmonic and satisfies the weak uniform boundedness property. 1. (c) Then $h:({\mathbb{D}},k_{\mathbb{D}})\to(h({\mathbb{D}}),k_{h({\mathbb{D}})})$ is Lipschitz. 2. (d) If, in addition, $h$ is a qc mapping, then $h:({\mathbb{D}},k_{\mathbb{D}})\to(h({\mathbb{D}}),k_{h({\mathbb{D}})})$ is a quasi-isometry. Proof. The part (d) is proved in [33]. For the proof of part (c) fix $x\in\mathbb{D}$ and $y\in B_{x}=B(x,\frac{1}{4}d(x))$. Then $d(y)\geq\frac{3}{4}d(x)$ and therefore $r(x,y)<1/2$. By the hypotheses $|h(y)-h(x)|\leq c\,d(h(x))$. The Schwarz lemma, applied to $B_{x}\,,$ yields in view of (2.14) $\frac{1}{4}d(x)|h^{\prime}(x)|\leq 2M_{x}\leq 2c\,d(h(x))$ The proof of part (c) follows from Lemma 2.5. $\Box$ A similar proof applies for higher dimensions; the following result is a generalization of the part $(c)$ of Theorem 2.5 . 2.19. Theorem. Suppose that $G$ is a proper subdomain of $\mathbb{R}^{n}$ and $h:G\to\mathbb{R}^{n}$ is a harmonic mapping. Then the following conditions are equivalent 1. (1) $h$ satisfies the weak uniform boundedness property. 2. (2) $h:({G},k_{G})\to(h({G}),k_{h({G})})$ is Lipschitz. Proof. Let us prove that $(1)$ implies $(2)$. By the hypothesis $(1)$ $f$ satisfies the weak uniform boundedness property: for every $x\in G$ and $t\in B_{x}$ $|f(t)-f(x)|\leq c_{2}\,d(f(x))\,.$ (2.20) This inequality together with Lemma 2.4 gives $d(x)|f^{\prime}(x)|\leq c_{3}\,d(f(x))$ for every $x\in G$. Now an application of Lemma 2.5 shows that $(1)$ implies $(2)$. It remains to prove that $(2)$ implies $(1)$. Suppose that $f$ is Lipschitz with the multiplicative constant $c_{2}$. Fix $x,y\in G$ with $r_{G}(x,y)\leq 1/2$. Then $|y-x|\leq d(x)/2$ and therefore by Lemma 2.1 $k_{G}(x,y)\leq 2j_{G}(x,y)\leq 2\,r_{G}(x,y)\leq 1.$ Hence $k_{G^{\prime}}(fx,fy)\leq c_{2}$. Since $j_{G^{\prime}}(fx,fy)\leq k_{G^{\prime}}(fx,fy)\leq c_{2}$, we find $j_{G^{\prime}}(fx,fy)=\log(1+r_{G^{\prime}}(fx,fy))\leq c_{2}$ and therefore $r_{G^{\prime}}(fx,fy)\leq e^{c_{2}}-1$. $\Box$ Since $f^{-1}$ is qc, an application of [12, Theorem 3] to $f^{-1}$ and Theorem 2.5 give the following corollary: 2.21. Corollary. Suppose that $G$ is a proper subdomain of $\mathbb{R}^{n}$, $h:G\rightarrow hG$ is harmonic and $K$-qc. Then $h:({G},k_{G})\to(h({G}),k_{h({G})})$ is a pseudo-isometry. In [45, Example 11.4] (see also [44, Example 3.10]), it is shown that the analytic function $f:{\mathbb{D}}\to G,G={\mathbb{D}}\setminus\\{0\\},f(z)=\exp((z+1)/(z-1))\,,$ $f({\mathbb{D}})=G,$ fails to be uniformly continuous as a map $f:({\mathbb{D}},k_{{\mathbb{D}}})\to(G,k_{G})\,.$ Therefore bounded analytic functions do not satisfy the weak uniform boundedness property in general. The situation will be different for instance if the boundary of the image domain is a continuum containing at least two points. Note that if $k_{G}$ is replaced by the hyperbolic metric $\lambda_{G}$ of $G,$ then $f:({\mathbb{D}},k_{{\mathbb{D}}})\to(G,\lambda_{G})\,$ is Lipschitz. 2.22. Theorem. Suppose that $G\subset\mathbb{R}^{n}$, $f:G\rightarrow\mathbb{R}^{n}$ is $K$-qr and $G^{\prime}=f(G)$. Let $\partial G^{\prime}$ be a continuum containing at least two distinct points. If $f$ is a harmonic mapping, then $f:({G},k_{G})\to({G^{\prime}},k_{G^{\prime}})$ is Lipschitz. Proof. Fix $x\in G$ and let $B_{x}=B^{n}(x,d(x)/4)$. If $|y-x|\leq d(x)/4$, then $d(y)\geq 3d(x)/4$ and therefore, $r_{G}(y,x)\leq\frac{4}{3}\frac{|y-x|}{d(x)}.$ Because $j_{G}(x,y)=\log(1+r_{G}(x,y))\leq r_{G}(x,y)$, using Lemma 2.1(a), we find $k_{G}(y,x)\leq 2\,j_{G}(y,x)\leq 2/3<1.$ By [45, Theorem 12.21] there exists a constant $c_{2}>0$ depending only on $n$ and $K$ such that $k_{G^{\prime}}(fy,fx)\leq c_{2}\max\\{k_{G}(y,x)^{\alpha},k_{G}(y,x)\\},\alpha=K^{1/(1-n)},$ and hence, using Lemma 2.1(b) and $k_{G}(y,x)\leq 1$, we see that $|fy-fx|\leq e^{c_{2}}d(fx),\quad\mbox{i.e.}\quad M_{x}=\omega_{f}(x)\leq e^{c_{2}}\,d(fx).$ (2.23) By (2.15) applied to $B_{x}=B^{n}(x,d(x)/4)$, we have $\frac{1}{4}d(x)|f^{\prime}(x)|\leq 2M_{x}$ and therefore using the inequality (2.23), we have $\frac{1}{4}d(x)|f^{\prime}(x)|\leq 2\,c\,d(f(x)),$ where $c=e^{c_{2}}$; and the proof follows from Lemma 2.5. $\Box$ The first author has asked the following Question (cf. [33]: Suppose that $G\subset\mathbb{R}^{n}$ is a proper subdomain, $f:G\rightarrow\mathbb{R}^{n}$ is harmonic $K$-qc and $G^{\prime}=f(G)$. Determine whether $f$ is a quasi- isometry w.r.t. quasihyperbolic metrics on $G$ and $G^{\prime}$. This is true for $n=2$ (see Theorem 2.6 below). It seems that one can modify the proof of Proposition 4.6 in [42] and show that this is true for the unit ball if $n\geq 3$ and $K<2^{n-1}$, cf. also [20]. ### 2.6 Quasi-isometry in planar case Astala and Gehring [4] proved a quasiconformal analogue of Koebe’s theorem, stated here as Theorem 2.6. These concern the quantity $a_{f}(x)=a_{f,G}(x):={\rm exp}\left(\frac{1}{n|B_{x}|}\int_{B_{x}}{\textrm{log}}J_{f}(z)dz\right),x\in G,$ associated with a quasiconformal mapping $f:G\rightarrow f(G)\subset\mathbb{R}^{n}$; here $J_{f}$ is the Jacobian of f ; while $B_{x}$ stands for the ball $B(x;d(x,\partial G)$; and $|B_{x}|$ for its volume. 2.24. Theorem[4]. Suppose that G and $G^{\prime}$ are domains in $R^{n}$: If $f:G\rightarrow G^{\prime}$ is $K$\- quasiconformal, then $\frac{1}{c}\frac{d(fx,\partial G^{\prime})}{d(x,\partial G)}\leq a_{f,G}(x)\leq c\frac{d(fx,\partial G^{\prime})}{d(x,\partial G)},\quad x\in G,$ where $c$ is a constant which depends only on $K$ and $n$. Let $\Omega\in\mathbb{R}^{n}$ and $\mathbb{R}^{+}=[0,\infty)$. If $f,g:\Omega\rightarrow\mathbb{R}^{+}$ and there is a positive constant $c$ such that $\frac{1}{c}\,g(x)\leq f(x)\leq c\,g(x)\,,\quad x\in\Omega\,,$ we write $f\approx g$ on $\Omega$. Our next result concerns the quantity $E_{f,G}(x):=\frac{1}{|B_{x}|}\int_{B_{x}}J_{f}(z)dz\,,x\in G,$ associated with a quasiconformal mapping $f:G\rightarrow f(G)\subset\mathbb{R}^{n}$; here $J_{f}$ is the Jacobian of $f$ ; while $B_{x}$ stands for the ball $B(x,d(x,\partial G)/2$ and $|B_{x}|$ for its volume. Define $A_{f,G}=\sqrt[n]{E_{f,G}}\,.$ 2.25. Theorem. Suppose $f:\Omega\rightarrow\Omega^{\prime}$ is a $C^{1}$ qc homeomorphism. The following conditions are equivalent: a)$f$ is bi-Lipschitz with respect to quasihyperbolic metrics on $\Omega$ and $\Omega^{\prime}~\,,$ b) $\sqrt[n]{J_{f}}\approx d_{*}/d\,,$ c) $\sqrt[n]{J_{f}}\approx a_{f}\,,$ d) $\sqrt[n]{J_{f}}\approx A_{f}$, where $d(x)=d(x,\partial\Omega)$ and $d_{*}(x)=d(f(x),\partial\Omega^{\prime})$. Proof. It is known that a) is equivalent to b) (see for example [36]). In [36], using Gehring’s result on the distortion property of qc maps (see [10], p.383; [43], p.63), the first author gives short proofs of a new version of quasiconformal analogue of Koebe’s theorem; it is proved that $A_{f}\approx d_{*}/d$. By Theorem 2.6, $a_{f}\approx d_{*}/d$ and therefore b) is equivalent to c). The rest of the proof is straightforward. $\Box$ If $\Omega$ is planar domain and $f$ a harmonic qc map, then we proved that the condition d) holds. The next theorem is a short proof of a recent result of V. Manojlovic [30],see also [33]. 2.26. Theorem. Suppose D and $D^{\prime}$ are proper domains in $\mathbb{R}^{2}$. If $h:D\rightarrow D^{\prime}$ is $K$\- qc and harmonic, then it is bi-Lipschitz with respect to quasihyperbolic metrics on D and $D^{\prime}$. Proof. Without loss of generality, we may suppose that $h$ is preserving orientation. Let $z\in D$ and $h=f+\overline{g}$ be a local representation of $h$ on $B_{z}$, where $f$ and $g$ are analytic functions on $B_{z}$, $\Lambda_{h}(z)=|f^{\prime}(z)|+|g^{\prime}(z)|$, $\lambda_{h}(z)=|f^{\prime}(z)|-|g^{\prime}(z)|$ and $k=\frac{K-1}{K+1}$. Since $h$ is $K$\- qc, we see that $(1-k^{2})|f^{\prime}|^{2}\leq J_{h}\leq K|f^{\prime}|^{2}$ (2.27) on $B_{z}$ and since ${\textrm{log}}\,|f^{\prime}(\zeta)|$ is harmonic, ${\textrm{ log}}\,|f^{\prime}(z)|=\frac{1}{2|B_{z}|}\int_{B_{z}}\textrm{ log}\,|f^{\prime}(\zeta)|^{2}d\xi\,d\eta\,.$ Hence, using the right hand side of (2.27), we find $\displaystyle{\textrm{log}}\,a_{h,D}(z)\leq\frac{1}{2}\,{\textrm{ log}}K+\frac{1}{2|B_{z}|}\int_{B_{z}}{\textrm{ log}}\,|f^{\prime}(\zeta)|^{2}d\xi\,d\eta$ (2.28) $\displaystyle={\textrm{ log}}\,\sqrt{K}\,|f^{\prime}(z)|\,.$ (2.29) Hence, $a_{h,D}(z)\leq\sqrt{K}\,|f^{\prime}(z)|$ and in a similar way using the left hand side of (2.27), we have $\sqrt{1-k^{2}}\,|f^{\prime}(z)|\leq a_{h,D}(z)\,.$ Now, an application of the Astala-Gehring result gives $\Lambda_{h}(z)\asymp\frac{d(hz,\partial D^{\prime})}{d(z,\partial D}\asymp\lambda_{h}(z)\,.$ This pointwise result, combined with Lemma 2.5 (integration along curves), easily gives $k_{D^{\prime}}(h(z_{1}),h(z_{2}))\asymp k_{D}(z_{1},z_{2}),\quad z_{1},z_{2}\in D\,.$ $\Box$ Note that in [30] the proof makes use of the interesting fact that ${\log}\frac{1}{J_{h}}$ is a subharmonic function; but we do not use it here. Define $m_{f}(x,r)=\min\\{|f(x^{\prime})-f(x)|:|x^{\prime}-x|=r\\}$. Suppose that $G$ and $G^{\prime}$ are domains in $\mathbb{R}^{n}$: If $f:G\rightarrow G^{\prime}$ is $K$\- quasiconformal; by the distortion property we find $m_{f}(x,r)\geq a(x)r^{1/\alpha}$. Hence, as in [20] and [36] , we get: 2.30. Lemma. If $f\in C^{1,1}$ is a $K-$ quasiconformal mapping defined in a domain $\Omega\subset\mathbb{R}^{n}$ $(n\geq 3)$, then $J_{f}(x)>0,\quad x\in\Omega$ provided that $K<2^{n-1}$. The constant $2^{n-1}$ is sharp. 2.31. Theorem. Under the hypothesis of the lemma, if $\overline{G}\subset\Omega$, then $f$ is bi-Lipschitz with respect to Euclidean and quasihyperbolic metrics on $G$ and $G^{\prime}=f(G).$ Proof. Since $\overline{G}$ is compact $J_{f}$ attains minimum on $\overline{G}$ at a point $x_{0}\in\overline{G}$. By Lemma 2.6, $m_{0}=J_{f}>0$ and therefore since $f\in C^{1,1}$ is a $K-$ quasiconformal, we conclude that functions $|f_{x_{k}}|$, $1\leq k\leq n$ are bounded from above and below on $\overline{G}$; hence $f$ is bi-Lipschitz with respect to Euclidean metric on $G$. By Theorem 2.6, we find $a_{f,G}\approx d_{*}/d$, where $d(x)=d(x,\partial G)$ and $d_{*}(x)=d(f(x),\partial G^{\prime})$. Since we have here $\sqrt[n]{J_{f}}\approx a_{f}$, we find $\sqrt[n]{J_{f}}\approx d_{*}/d$ on $G$. An application of Theorem 2.6 completes the proof. $\Box$ ### 2.7 The upper half space ${\mathbb{H}}^{n}.$ Let $\mathbb{H}^{n}$ denote the half-space in $\mathbb{R}^{n}$. If $D$ is a domain in $\mathbb{R}^{n}$, by $QCH(D)$ we denote the set of Euclidean harmonic quasiconformal mappings of $D$ onto itself. In particular if $x\in\mathbb{R}^{3}$, we use notation $x=(x_{1},x_{2},x_{3})$ and we denote by $\partial_{x_{k}}f=f^{\prime}_{x_{k}}$ the partial derivative of $f$ with respect to $x_{k}$ . A fundamental solution in space $\mathbb{R}^{3}$ of the Laplace equation is $\frac{1}{|x|}$. Let $U_{0}=\frac{1}{|x+e_{3}|}$, where $e_{3}=(0,0,1)$. Define $h(x)=(x_{1}+\varepsilon_{1}U_{0},x_{2}+\varepsilon_{2}U_{0},x_{3}).$ It is easy to verify that $h\in QCH(\mathbb{H}^{3})$ for small values of $\varepsilon_{1}$ and $\varepsilon_{2}$. Using the Herglotz representation of a nonnegative harmonic function $u$ (see Theorem 7.24 and Corollary 6.36 [5]), one can get: Lemma A. If $u$ is a nonnegative harmonic function on a half space $\mathbb{H}^{n}$, continuous up to the boundary with $u=0$ on $\mathbb{H}^{n}$, then $u$ is (affine) linear. In [33], the first author has outlined a proof of the following result: Theorem A. If $h$ is a quasiconformal harmonic mapping of the upper half space $\mathbb{H}^{n}$ onto itself and $h(\infty)=\infty$, then $h$ is quasi- isometry with respect to both the Euclidean and the Poincaré distance. Note that the outline of proof in [33] can be justified by Lemma A. We show that the analog statement of this result holds for $p$-harmonic vector functions (solutions of $p$-Laplacian equations) using the mentioned result obtained in the paper [23], stated here as: Theorem B. If $u$ is a nonnegative $p$-harmonic function on a half space $\mathbb{H}^{n}$, continuous up to the boundary with $u=0$ on $\mathbb{H}^{n}$, then $u$ is (affine) linear. 2.32. Theorem. If $h$ is a quasiconformal $p$-harmonic mapping of the upper half space $\mathbb{H}^{n}$ onto itself and $h(\infty)=\infty$, then both $h:(\mathbb{H}^{n},|\cdot|)\to(\mathbb{H}^{n},|\cdot|)$ and $h:(\mathbb{H}^{n},\rho_{\mathbb{H}^{n}})\to(\mathbb{H}^{n},\rho_{\mathbb{H}^{n}})$ are bi-Lipschitz where $\rho=\rho_{\mathbb{H}^{n}}$ is the $Poincar\acute{e}$ distance. Since $2$-harmonic mapping are Euclidean harmonic this result includes Theorem A. Proof. It suffices to deal with the case $n=3$ as the proof for the general case is similar. Let $h=(h_{1},h_{2},h_{3})$. By Theorem B, we get $h_{3}(x)=ax_{3}$, where $a$ is a positive constant. Without loss of generality we may suppose that $a=1$. Since $h_{3}(x)=x_{3}$, we have $\partial_{x_{3}}h_{3}(x)=1$, and therefore $|h^{\prime}_{x_{3}}(x)|\geq 1$. In a similar way, $|g^{\prime}_{x_{3}}(x)|\geq 1$, where $g=h^{-1}$. Hence, there exists a constant $c=c(K)$, $|h^{\prime}(x)|\leq c\quad{\rm and}\quad 1/c\leq l(h^{\prime}(x))\,.$ Therefore partial derivatives of $h$ and $h^{-1}$ are bounded from above; and, in particular, $h$ is Euclidean bi-Lipschitz. Since $h_{3}(x)=x_{3}$, $\frac{|h^{\prime}(x)|}{h_{3}(x)}\leq\frac{c}{x_{3}};$ and hence, by Lemma 2.5, $\rho(h(a),h(b))\leq c\rho(a,b)\,.$ $\Box$ ## 3 Pseudo-isometry and $OC^{1}(G)$ In this section, we give a sufficient condition for a qc mapping $f:G\rightarrow f(G)$ to be a pseudo-isometry w.r.t. quasihyperbolic metrics on $G$ and $f(G)$. First we adopt the following notation. If $V$ is a subset of $\mathbb{R}^{n}$ and $u:V\rightarrow\mathbb{R}^{m}$, we define ${\rm osc}_{V}u=\sup\\{|u(x)-u(y)|\,:x,y\in V\\}\,.$ Suppose that $G\subset\mathbb{R}^{n}$ and $B_{x}=B(x,d(x)/2)$. Let $OC^{1}(G)$ denote the class of $f\in C^{1}(G)$ such that $d(x)|f^{\prime}(x)|\leq c_{1}\,{\rm osc}_{B_{x}}f$ (3.1) for every $x\in G$. Similarly, let $SC^{1}(G)$ be the class of functions $f\in C^{1}(G)$ such that $|f^{\prime}(x)|\leq ar^{-1}\,\omega_{f}(x,r)\quad{\rm for\,all}\,\,B^{n}(x,r)\subset G,$ (3.2) where $\omega_{f}(x,r)=\sup\\{|f(y)-f(x)|:\,y\in B^{n}(x,r)\\}$. The proof of Theorem 2.5 gives the following more general result: 3.3. Theorem. Suppose that $G\subset\mathbb{R}^{n}$, $f:G\rightarrow G^{\prime}$, $f\in OC^{1}(G)$ and it satisfies the weak property of uniform boundedness with a constant $c$ on $G$. Then 1. (e) $f:({G},k_{G})\to({G^{\prime}},k_{G^{\prime}})$ is Lipschitz. 2. (f) In addition, if $f$ is $K$-qc, then $f$ is pseudo-isometry w.r.t. quasihyperbolic metrics on $G$ and $f(G)$. Proof. By the hypothesis $f$ satisfies the weak property of uniform boundedness: $|f(t)-f(x)|\leq c_{2}\,d(f(x)$ for every $t\in B_{x}$, that is ${\rm osc}_{B_{x}}f\leq c_{2}\,d(f(x))$ (3.4) for every $x\in G$. This inequality together with (3.1) gives $d(x)|f^{\prime}(x)|\leq c_{3}\,d(f(x))$. Now an application of Lemma 2.5 gives part (e). Since $f^{-1}$ is qc, an application of [12, Theorem 3] on $f^{-1}$ gives part (f). $\Box$ In order to apply the above method we introduce subclasses of $OC^{1}(G)$ (see, for example, below (3.5)). Let $f:G\rightarrow G^{\prime}$ be a $C^{2}$ function and $B_{x}=B(x,d(x)/2).$ We denote by $OC^{2}(G)$ the class of functions which satisfy the following condition: $\sup_{B_{x}}d^{2}(x)|\Delta f(x)|\leq c\,\,{\rm osc}_{B_{x}}f$ (3.5) for every $x\in G$. If $f\in OC^{2}(G)$, then by Theorem 3.9 in [14], applied to $\Omega=B_{x}$, $\sup_{t\in B_{x}}d(t)|f^{\prime}(t)|\leq C(\sup_{t\in B_{x}}|f(t)-f(x)|+\sup_{t\in B_{x}}d^{2}(t)|\Delta f(t)|)$ and hence by (3.5) $d(x)|f^{\prime}(x)|\leq c_{1}\,{\rm osc}_{B_{x}}f$ (3.6) for every $x\in G$ and therefore $OC^{2}(G)\subset OC^{1}(G)$. Now the following result follows from the previous theorem. 3.7. Corollary. Suppose that $G\subset\mathbb{R}^{n}$ is a proper subdomain, $f:G\rightarrow G^{\prime}$ is $K$-qc and $f$ satisfies the condition (3.5). Then $f:({G},k_{G})\to({G^{\prime}},k_{G^{\prime}})$ is Lipschitz. We will now give some examples of classes of functions to which Theorem 3 is applicable. Let $SC^{2}(G)$ denote the class of $f\in C^{2}(G)$ such that $|\Delta f(x)|\leq ar^{-1}\sup\\{|f^{\prime}(y)|:\,y\in B^{n}(x,r)\\},$ for all $B^{n}(x,r)\subset G$, where $a$ is a positive constant. Note that the class $SC^{2}(G)$ contains every function for which $d(x)|\Delta f(x)|\leq a|f^{\prime}(x)|$, $x\in G$. It is clear that $SC^{1}(G)\subset OC^{1}(G)$ and by the mean value theorem, $OC^{2}(G)\subset SC^{2}(G)$. For example, in [39] it is proved that $SC^{2}(G)\subset SC^{1}(G)$ and that the class $SC^{2}(G)$ contains harmonic functions, eigenfunctions of the ordinary Laplacian if $G$ is bounded, eigenfunctions of the hyperbolic Laplacian if $G={\mathbb{B}}^{n}$ and thus our results are applicable for instance to these classes. Acknowledgements. This work was initiated during the visit of the first author to the Universities of Helsinki and Turku in October 2005 and continued during the visit of the second author to Belgrade in December 2006. The authors’ research was supported by the grant no. 8118525 the Academy of Finland. The first author was partially supported by MNTRS, Serbia, Grant No. 144 020. The authors are indebted to the referee for valuable comments and to O. Martio and S. Ponnusamy for interesting discussions on this paper. ## References * [1] L. V. Ahlfors: Conformal Invariants: Topics in Geometric Function Theory, McGraw-Hill, New York, 1973. * [2] M. Arsenović, V. Kojić and M. Mateljević: On Lipschitz continuity of harmonic quasiregural mappings on the unit ball in $R^{n}$, Ann. Acad. Sci. Fenn. Math. Vol. 33, (2008), 315–318. * [3] M. Arsenović, V. Manojlović, and M. Mateljević,: Lipschitz-type spaces and harmonic mappings in the space, Ann. Acad. Sci. Fenn. 35 (2010), 1–9. * [4] K. Astala and F. W. Gehring: Quasiconformal analogues of theorems of Koebe and Hardy-Littlewood, Mich.Math.J. 32 (1985) 99-107. * [5] S. Axler, P. Bourdon and W. Ramey: Harmonic function theory, Springer-Verlag, New York 1992. * [6] A. F. Beardon: The geometry of discrete groups, Graduate Texts in Math. Vol 91, Springer Verlag, Berlin – Heidelberg – New York, 1982. * [7] D. Bshouty and W. Hengartner: _Univalent harmonic mappings in the plane,_ In: Handbook of Complex Analysis: Geometric Function Theory, Vol. 2, (2005), 479–506, Edited by R. Kühnau (ISBN: 0-444-51547-X), Elsevier. * [8] B. Burgeth: A Schwarz lemma for harmonic and hyperbolic-harmonic functions in higher dimensions, Manuscripta Math. 77 (1992), 283–291. * [9] P. Duren: _Harmonic mappings in the plane_ , Cambridge University Press, 2004. * [10] F.W. Gehring: Rings and quasiconformal mappings in space, Trans. Amer. Math. Soc. 103 , 1962, 353–393. * [11] F.W. Gehring: Quasiconformal mappings in Euclidean spaces. Handbook of complex analysis: geometric function theory. Vol. 2, 1–29, Ed. by R. Kühnau, Elsevier, Amsterdam, 2005. * [12] F.W. Gehring and B.G. Osgood: Uniform domains and the quasi-hyperbolic metric, J. Anal. Math. 36(1979), 50–74. * [13] F. W. Gehring and B. P. Palka: Quasiconformally homogeneous domains, J. Anal. Math. 30 (1976), 172–199. * [14] D. Gilbarg and N. Trudinger: Elliptic Partial Differential Equation of Second Order, Second Edition, 1983. * [15] E. Heinz: On one-to-one harmonic mappings, Pacific J. Math. 9(1959), 101–105. * [16] D. Kalaj: Quasiconformal and harmonic mappings between Jordan domains, Math. Z. 260:2(2008), 237–252. * [17] D. Kalaj: Harmonic quasiconformal mappings and Lipschitz spaces, Ann. Acad. Sci. Fenn. Math. Math. 34:2 (2009), 475-485. * [18] D. Kalaj and M. Mateljević: Inner estimate and quasiconformal harmonic maps between smooth domains, J. Anal. Math. 100, 117–132, (2006). * [19] D.Kalaj and M. Mateljević: Quasiconformal and harmonic mappings between smooth Jordan domains, Novi Sad J. Math. Vol. 38 (2008), 147–156. * [20] D. Kalaj and M. Mateljević: Harmonic quasiconformal self-mappings and Möbius transformations of the unit ball , to appear in Pacific J. Math. * [21] D. Kalaj and M. Pavlović: Boundary correspondence under quasiconformal harmonic diffeomorphisms of a half-plane, Ann. Acad. Sci. Fenn. Math. 30 (2005), no. 1, 159–165. * [22] L. Keen and N. Lakic: Hyperbolic geometry from a local viewpoint. London Mathematical Society Student Texts, 68. Cambridge University Press, Cambridge, 2007. * [23] T. Kilpeläinen, H. Shahgholian and X. Zhong: Growth estimates through scaling for quasilinear partial differential equations, Ann. Acad. Sci. Fenn. Math. 32 (2007), 595–599. * [24] R. Klén: On hyperbolic type metrics, Ann. Acad. Sci. Fenn. Math. Diss. 152 (2009), 1–49. * [25] R. Kühnau, ed.: Handbook of complex analysis: geometric function theory, Vol. 1-2. Elsevier Science B.V., Amsterdam, 2002, 2005. * [26] O. Lehto and K. I. Virtanen: Quasiconformal Mappings in the Plane, 2nd ed., Grundlehren Math. Wiss., Band 126, Springer-Verlag, New York, 1973. * [27] H. Lindén: Quasihyperbolic geodesics and uniformity in elementary domains, Ann. Acad. Sci. Fenn. Math. Diss. 146 (2005), 1–50. * [28] A. Lyzzaik: Local properties of Light Harmonic Mappings, Canadian J. Math. 44(1)(1992), 135–153. * [29] V. Manojlović: Moduli of Continuity of Quasiregular Mappings , 2008, arXiv.math 0808.3241. * [30] V. Manojlovic: Bi-Lipschicity of quasiconformal harmonic mappings in the plane, Filomat, 23 (2009), 85–89. * [31] O.Martio: On harmonic quasiconformal mappings, Ann. Acad. Sci. Fenn. Ser. A I No. 425 (1968) 3-10. * [32] M. Mateljević: A version of Bloch theorem for quasiregular harmonic mappings, Rev. Roum. Math. Pures Appl. 47(2002), 705–707. * [33] M. Mateljević: Distortion of harmonic functions and harmonic quasiconformal quasi-isometry, Rev. Roum. Math. Pures Appl. 51(2006), 711–722. * [34] M. Mateljević: Quasiconformal and quasiregular harmonic analogues of Koebe’s theorem and applications, Ann. Acad. Sci. Fenn. 32(2007), 301–315. * [35] M. Mateljević: On quasiconformal harmonic mappings, unpublished manuscript, 2006\. * [36] M. Mateljević: Lipschitz-type spaces,Quasiconformal and Quasiregular harmonic mappings and Applications, unpublished manuscript, 2008. * [37] M. Mateljević and M. Knezević: On the quasi-isometries of harmonic quasi-conformal mappings, J. Math. Anal. Appl. 334(2007), 404–413. * [38] D. Partyka and K. Sakan: On bi-Lipschitz type inequalities for quasiconformal harmonic mappings, Ann. Acad. Sci. Fenn. Math. 32 (2007), 579–594. * [39] M. Pavlović: On subharmonic behaviour of smooth functions, Mat. Vesnik 48(1996), 15–21. * [40] M. Pavlović: Boundary correspondence under harmonic quasiconformal homeomorfisms of the unit disc, Ann. Acad. Sci. Fenn., Vol 27, 2002, 365–372. * [41] S. L. Qiu, M. K. Vamanamurthy, and M. Vuorinen: Some inequalities for the Hersch-Pfluger distortion function, J. Math. Anal. Appl. Vol. 4 (1999), 2, 115-139 * [42] L. Tam and T. Wan: On quasiconformal harmonic maps, Pacific J. Math. 182(1998), 359–383. * [43] J.Väisälä: Lectures on n-Dimensional Quasiconformal Mappings, Lecture Notes in Math. 229, Springer-Verlag, 1971. * [44] M. Vuorinen: _Conformal invariants and quasiregular mappings._ J. Anal. Math. 45(1985), 69–115. * [45] M. Vuorinen: Conformal Geometry and Quasiregular Mappings, Lecture Notes in Math. 1319, Springer-Verlag, Berlin–New York, 1988. * [46] M. Vuorinen: _Metrics and quasiregular mappings._ Proc. Int. Workshop on Quasiconformal Mappings and their Applications, IIT Madras, Dec 27, 2005 - Jan 1, 2006, ed. by S. Ponnusamy, T. Sugawa and M. Vuorinen, _Quasiconformal Mappings and their Applications_ , Narosa Publishing House, 291–325, New Delhi, India, 2007. M. Mateljević Faculty of Mathematics University of Belgrade Studentski trg 16 11000 Beograd Serbia miodrag’at’matf.bg.ac.yu M. Vuorinen Department of Mathematics FIN-20014 University of Turku FINLAND vuorinen’at’utu.fi

arxiv-papers

2024-09-04T02:48:53.188386

0710.1304

Balanced Metrics and Noncommutative Kähler Geometry Balanced Metrics and Noncommutative Kähler Geometry\star\star$\star$This paper is a contribution to the Special Issue “Noncommutative Spaces and Fields”. The full collection is available at http://www.emis.de/journals/SIGMA/noncommutative.html Sergio LUKIĆ S. Lukić Department of Physics and Astronomy, Rutgers University, Piscataway, NJ 08855-0849, USA lukic@physics.rutgers.edu Received March 01, 2010, in final form August 02, 2010; Published online August 27, 2010 In this paper we show how Einstein metrics are naturally described using the quantization of the algebra of functions $C^{\infty}(M)$ on a Kähler manifold $M$. In this setup one interprets $M$ as the phase space itself, equipped with the Poisson brackets inherited from the Kähler 2-form. We compare the geometric quantization framework with several deformation quantization approaches. We find that the _balanced metrics_ appear naturally as a result of requiring the vacuum energy to be the constant function on the moduli space of _semiclassical vacua_. In the classical limit these metrics become Kähler–Einstein (when $M$ admits such metrics). Finally, we sketch several applications of this formalism, such as explicit constructions of special Lagrangian submanifolds in compact Calabi–Yau manifolds. balanced metrics; geometric quantization; Kähler–Einstein 14J32; 32Q15; 32Q20; 53C25; 53D50 ## 1 Introduction Noncommutative deformations of Kähler geometry exhibit some extraordinary features, similar to those that appear in the description of $n$ quantum harmonic oscillators by the noncommutative phase space $\mathbb{C}^{n}$. Noncommutative geometry in Calabi–Yau compactifications is expected to play a special role when the $B$-field is turned on [22, 16], in the formulation of M(atrix) theory [4, 8, 15], and in the large $N$ limit of probe D0-branes [14]. Also, as we show below, one can use the geometric quantization approach to noncommutative geometry to determine111More precisely, what one can determine are certain metrics, known as balanced metrics, which obey the equations of motion in the classical limit. important objects in string theory compactifications, which allow the computation of the exact form of the Lagrangian in the four dimensional effective field theory [9, 10, 11, 12]. In this paper we show how the notion of _balanced metrics_ appears naturally in the framework of Kähler quantization/noncommutative geometry. In the geometric quantization formalism the balanced metric appears as a consequence of requiring the norm of the coherent states to be constant; these states are parameterized by the Kodaira’s embedding of the Kähler manifold $M$ into the projectivized quantum Hilbert space. In the Reshetikhin–Takhtajan approach to the deformation quantization of $M$ [21], the balanced metric appears as a consequence of requiring the unit of the quantized algebra of functions $C^{\infty}(M)[[\hbar]]$ to be the constant function ${\bf 1}\colon M\to 1\in{\mathbb{R}}$. Finally, we quantize the phase space with constant classical Hamiltonian on $M$ using the path integral formalism; here, one considers a different class of semiclassical vacuum states which are not coherent states, and uses them to define222By requiring the vacuum energy density to be the constant function on the semiclassical vacua, labeled by the points $x\in M$. a generalization of the balanced metrics (which also become Einstein metrics in the _classical limit_). The organization of the paper is as follows. In Section 2 we recall some basic facts of geometric quantization [3, 13, 23], show how the balanced metrics appear naturally in this framework, and sketch how differential geometric objects, such as Kähler–Einstein metrics or Lagrangian submanifolds, can be described in this language. In Section 3 we summarize the work of Reshetikhin–Takhtajan, and show how the constant function is the unit element of their quantized algebra of functions if and only if the metric on $M$ is balanced. Finally, in Section 4 we consider a different set of semiclassical vacuum states in path integral quantization and use them to define _generalized balanced metrics_ , which differ slightly from the balanced metrics in geometric quantization. ## 2 Geometric quantization Classical mechanics and geometric quantization have a beautiful formulation using the language of symplectic geometry, vector bundles, and operator algebras [3, 13, 23]. In this language, symplectic manifolds $M$ are interpreted as phase spaces, and spaces of smooth functions $C^{\infty}(M)$ as the corresponding classical observables. Kähler quantization is understood far better than quantization on general symplectic manifolds; for this reason we only consider Kähler manifolds (which are symplectic manifolds endowed with a compatible complex structure). $(M,{\cal L}^{\otimes\kappa})$ denotes a polarized Kähler manifold $M$ with a very ample hermitian line bundle333In other words, ${\cal L}^{\otimes\kappa}$ is an element of the Kähler cone associated to $M$. ${\cal L}^{\otimes\kappa}$, and $\kappa\in{\mathbb{Z}}_{+}$ a positive integer. For technical reasons, we consider $M$ to be compact and simply connected. We work with a trivialization of ${\cal L}|_{U}\to U$, where $U\subset M$ is an open subset; we define $K(\phi,\bar{\phi})$ to be the associated analytic Kähler potential and $e^{-\kappa K(\phi,\bar{\phi})}$ the hermitian metric on ${\cal L}^{\otimes\kappa}\to M$. If ${\rm dim}_{\mathbb{C}}\,M=n$ and $\\{\phi_{i}\\}_{0<i\leq n}$ is a local holomorphic coordinate chart for the open subset $U\subset M$, we can write the Kähler metric on $M$ and its compatible symplectic form as $\displaystyle i\kappa g_{i\bar{\jmath}}d\phi^{i}\otimes d\bar{\phi}^{\bar{\jmath}}=\kappa\omega_{i\bar{\jmath}}d\phi^{i}\wedge d\bar{\phi}^{\bar{\jmath}}=i\kappa{\partial\over\partial\phi^{i}}{\partial\over\partial\bar{\phi}^{\bar{\jmath}}}K(\phi,\bar{\phi})d\phi^{i}\otimes d\bar{\phi}^{\bar{\jmath}}.$ Classically, the space $(C^{\infty}(M),\omega)$ of observables has, in addition to a Lie algebra structure defined by the Poisson bracket $\\{f,g\\}_{PB}=\omega^{i\bar{\jmath}}(\partial_{i}f\bar{\partial}_{\bar{\jmath}}g-\partial_{i}g\bar{\partial}_{\bar{\jmath}}f),\qquad f,g\in C^{\infty}(M),$ the structure of a commutative algebra under pointwise multiplication, $(fg)(x)=f(x)g(x)=(gf)(x).$ Quantization can be understood as a non-commutative deformation of $C^{\infty}(M)$ parameterized by $\hbar$, with commutativity recovered when $\hbar=0$. We will discuss the formalism of deformation quantization in the next section, although generally speaking, quantization refers to an assignment $T\colon f\to T(f)$ of classical observables to operators on some Hilbert space ${\cal H}$. When $M$ is compact, the Hilbert space will be finite-dimensional with dimension $\dim{\cal H}={{\rm vol}\,M\over\hbar^{n}}+O(\hbar^{1-n})$. The assignment $T$ must satisfy the following requirements: * • Linearity, $T(af+g)=aT(f)+T(g)$, $\forall\,a\in\mathbb{C}$, $f,g\in C^{\infty}(M)$. * • Constant map $1$ is mapped to the identity operator ${\rm Id}$, $T(1)={\rm Id}$. * • If $f$ is a real function, $T(f)$ is a hermitian operator. * • In the limit $\hbar\to 0$, the Poisson algebra is recovered $[T(f),T(g)]=i\hbar T(\\{f,g\\}_{PB})+O(\hbar^{2})$. In geometric quantization the positive line bundle ${\cal L}^{\otimes\kappa}$ is known as _prequantum line bundle_. The prequantum line bundle is endowed with a unitary connection whose curvature is the symplectic form $\kappa\omega$ (which is quantized, i.e., $\omega\in H^{2}(M,{\mathbb{Z}})$). The _prequantum Hilbert space_ is the space of $L^{2}$ sections $L^{2}({\cal L}^{\otimes\kappa},M)=\left\\{s\in\Omega^{0}({\cal L}^{\otimes\kappa})\colon\ \int_{M}h^{\kappa}\langle s,\bar{s}\rangle{\omega^{n}\over n!}<\infty\right\\},$ where $h^{\kappa}$ is the compatible hermitian metric on ${\cal L}^{\otimes\kappa}$. The Hilbert space is merely a subspace of $L^{2}({\cal L}^{\otimes\kappa},M)$, defined with the choice of a polarization on $M$. In the case of Kähler polarization, the split of the tangent space in holomorphic and anti-holomorphic directions, $TM=TM^{(1,0)}\oplus TM^{(0,1)}$, defines a Dolbeault operator on ${\cal L}^{\otimes\kappa}$, $\bar{\partial}\colon\Omega^{(0)}({\cal L}^{\otimes\kappa})\to\Omega^{(0,1)}({\cal L}^{\otimes\kappa})$. The Hilbert space ${\cal H}_{\kappa}$ is only the kernel of $\bar{\partial}$, i.e., the space of holomorphic sections $H^{0}(M,{\cal L}^{\otimes\kappa})$. As a final remark, the quantization map $T$ is not uniquely defined; there are different assignments of smooth functions on $M$ to matrices on ${\cal H}_{\kappa}$ that obey the same requirements stated above, giving rise to equivalent classical limits. For simplicity, we mention only the most standard ones [5]: * • _The Toeplitz map:_ $T(f)_{\alpha\bar{\beta}}=\int_{M}f(z,\bar{z})s_{\alpha}(z)\bar{s}_{\bar{\beta}}(\bar{z})h^{\kappa}(z,\bar{z}){\omega(z,\bar{z})^{n}\over n!},$ with $s_{\alpha}$ a basis of sections for ${\cal H}_{\kappa}$ and $s_{\alpha}(z)$ the corresponding evaluation of $s_{\alpha}$ at $z\in U\subset M$. * • _The geometric quantization map:_ $Q(f)=iT\left(f-{1\over 2}\Delta f\right)$, with $\Delta$ the corresponding Laplacian on $M$. We will work only with completely degenerated Hamiltonian systems (i.e. a constant Hamiltonian function on $M$); therefore the choice of quantization map will not be important. Rather we will study the semiclassical limit of the corresponding quantized system by determining the semiclassical vacuum states. ### 2.1 Coherent states and balanced metrics As we described above, the geometric quantization picture is characterized by the prequantum line bundle, ${\cal L}^{\otimes\kappa}\to M$, a holomorphic line bundle on $M$ which is endowed with a $U(1)$ connection with Kähler 2-forms $\kappa\omega$. As the positive integer $\kappa$ always appears multiplying the symplectic form, one can interpret $\kappa^{-1}=\hbar$ as a discretized Planck’s constant. Thus, according to this convention, the semiclassical appears in the limit $\kappa\to\infty$. In the local trivialization $U\subset M$, where $K(\phi,\bar{\phi})$ is the Kähler potential and $e^{-\kappa K(\phi,\bar{\phi})}$ the hermitian metric on ${\cal L}^{\otimes\kappa}|_{U}$, one can set the compatible Dolbeault operator to be locally trivial and write the covariant derivative as $\widetilde{\nabla}=d\phi^{i}(\partial_{i}-\kappa\partial_{i}K)+d\bar{\phi}^{\bar{\imath}}\bar{\partial}_{\bar{\imath}},$ where $K$ is the yet undetermined analytic Kähler potential on ${\cal L}$. One can also determine the associated unitary connection up to a $U(1)$ gauge transformation, $\nabla=d\phi^{i}(\partial_{i}+A_{i})+d\bar{\phi}^{\bar{\imath}}(\bar{\partial}_{\bar{\imath}}-A_{i}^{\dagger}),$ with $A_{i}=\sqrt{h^{-\kappa}}\partial_{i}\sqrt{h^{\kappa}}$, and $h=\exp(-K(\phi,\bar{\phi}))$. As explained above, the Hilbert space ${\cal H}_{\kappa}$ corresponds to the kernel of the covariant half-derivative $\nabla^{(0,1)}\colon\Omega^{(0)}({\cal L})\to\Omega^{(0,1)}({\cal L})$, which are the holomorphic sections of ${\cal L}^{\otimes\kappa}$ ${\cal H}_{\kappa}=H^{0}(M,{\cal L}^{\otimes\kappa})={\rm span}_{\mathbb{C}}\left\\{|s_{\alpha}\rangle\right\\}_{\alpha=1}^{N}.$ The dimension of the quantum Hilbert space is $N=\dim{\cal H}_{\kappa}={1\over n!}\int_{M}c_{1}({\cal L})^{n}\kappa^{n}+{1\over 2(n-1)!}\int_{M}c_{1}({\cal L})^{n-1}c_{1}(M)\kappa^{n-1}+O\big{(}\kappa^{n-2}\big{)}.$ We identify $|s_{\alpha}\rangle$ as the basis elements of ${\cal H}_{\kappa}$. The _coherent state_ localized at $x\in M$ can be defined (see [20]) on the trivialization ${\cal L}^{\otimes\kappa}|_{U}\to U\subset M$ as the ray in $\mathbb{P}{\cal H}_{\kappa}$ generated by $|\tilde{\Omega}_{x}\rangle=\sum_{\alpha}s_{\alpha}(x)\exp(-\kappa K(x,\bar{x})/2)|s_{\alpha}\rangle\in{\cal H}_{\kappa},$ where $s_{\alpha}(x)\exp(-\kappa K(x,\bar{x})/2)$ is the evaluation of the holomorphic section $|s_{\alpha}\rangle$ at the point $x\in U\subset M$, in the trivialization ${\cal L}^{\otimes\kappa}|_{U}$. The coherent states are a supercomplete basis of ${\cal H}_{\kappa}$, and obey the Parseval identity $\displaystyle\langle\zeta|\xi\rangle=\int_{M\hookrightarrow\mathbb{P}{\cal H}_{\kappa}}\langle\zeta|\tilde{\Omega}_{x}\rangle\langle\tilde{\Omega}_{x}|\xi\rangle{\omega^{n}(x,\bar{x})\over n!},\qquad\forall\,\zeta,\xi\in{\cal H}_{\kappa}.$ (1) These points in $\mathbb{P}{\cal H}_{\kappa}$ are independent of the trivialization, and they have the property of being localized at $x\in M$ with minimal quantum uncertainty. The _distortion function_ , diagonal of the Bergman kernel, or expected value of the identity at $x$, $\rho(x,\bar{x})$ is defined as $\displaystyle\rho=\langle\tilde{\Omega}_{x}|\tilde{\Omega}_{x}\rangle=\sum_{\alpha,\beta}\bar{s}_{\bar{\alpha}}(\bar{x})s_{\beta}(x)\exp\left(-\kappa K(x,\bar{x})\right)\langle s_{\alpha}|s_{\beta}\rangle,$ (2) which measures the relative normalization of the coherent states located at different points of $M$. Imposing $\rho(x,\bar{x})=\langle\tilde{\Omega}_{x}|\tilde{\Omega}_{x}\rangle=\mathrm{const}$, constrains the Kähler potential $K(x,\bar{x})$ to be a Fubini–Study Kähler potential: $\displaystyle K(x,\bar{x})={1\over\kappa}\log\left(\sum_{\alpha,\beta}\bar{s}_{\bar{\alpha}}(\bar{x})s_{\beta}(x)\langle s_{\alpha}|s_{\beta}\rangle\right).$ (3) One of the most important ingredients in the quantization procedure is the definition of the quantization map, $T:C^{\infty}(M)\to{\rm Herm}({\cal H}_{\kappa})$. This maps classical observables, i.e. smooth real functions on the phase space $X$, to quantum observables, i.e., self-adjoint operators on the Hilbert space ${\cal H}_{\kappa}$. If we work with an orthonormal basis $\langle s_{\beta}|s_{\alpha}\rangle=\delta_{\beta\alpha}$, the quantization condition $T(1_{M})=\mathrm{Id}\in{\cal H}_{\kappa}\otimes{\cal H}_{\kappa}^{\ast}$ implies that the embedding of the coherent states satisfies the _balanced_ condition [10], $\displaystyle\delta_{\alpha\beta}=\langle s_{\alpha}|s_{\beta}\rangle=\sum_{x}\langle s_{\alpha}|\tilde{\Omega}_{x}\rangle\langle\tilde{\Omega}_{x}|s_{\beta}\rangle=\int_{M}{\bar{s}_{\alpha}(\bar{x})s_{\beta}(x)\over\sum_{\gamma}|s_{\gamma}(x)|^{2}}{\omega(x,\bar{x})^{n}\over n!};$ (4) here, we have used the Parseval identity (1), and the Liouville’s volume form on the phase space $M$, which can be written as ${1\over n!}\omega(z,\bar{z})^{n}={1\over n!}\left[\bar{\partial}_{\bar{\jmath}}\partial_{i}K(z,\bar{z})dz^{i}\wedge d\bar{z}^{\bar{\jmath}}\right]^{n}.$ In summary, in the geometric quantization of an algebraic Kähler manifold, the homogeneity of the distortion function $\langle\tilde{\Omega}_{x}|\tilde{\Omega}_{x}\rangle$ and the mapping of the constant function on $M$ to the identity operator ${\rm Id}\colon{\cal H}_{\kappa}\to{\cal H}_{\kappa}$, determines a unique metric on $M$ known as _balanced metric_. In the semiclassical limit, $\kappa\to\infty$, this sequence of balanced metrics approaches the Kähler–Einstein metric (if it exists) as sketched below (see [9, 12]). ### 2.2 Emergence of classical geometry For every $\kappa$, the _balanced metric_ has just been defined as result of requiring $\langle\tilde{\Omega}_{x}|\tilde{\Omega}_{x}\rangle$ to be the constant function on $M$. In the semiclassical limit, $\kappa\to\infty$, we can expand the distortion function in inverse powers of $\kappa$ (see [24]) $\langle\tilde{\Omega}_{x,\kappa}|\tilde{\Omega}_{x,\kappa}\rangle\sim 1+{1\over 2\kappa}R+O\big{(}\kappa^{-2}\big{)}+\cdots,$ and therefore the sequence of balanced metrics will converge to a metric of constant scalar curvature at $\kappa=\infty$. For a Calabi–Yau manifold this is equivalent to a Ricci flat Kähler metric. It is interesting to note that if the identity matrix is identified with the quantum Hamiltonian, and the coherent states with the semiclassical states, the _balanced metric_ can also be defined as the metric that yields a constant semiclassical vacuum energy $\langle\tilde{\Omega}_{x,\kappa}|\tilde{\Omega}_{x,\kappa}\rangle$, as a function of $x\in M$ and fixed $\kappa$. Other geometrical elements that one can recover naturally are the Lagrangian submanifolds with respect the Kähler–Einstein symplectic form. In the Kähler $n$-fold $(M,\omega)$, the level sets of $n$ commuting functions $(f_{1},f_{2},\ldots,f_{n})$ under the Poisson bracket $\\{f_{a},f_{b}\\}_{PB}=\omega^{i\bar{\jmath}}(\partial_{i}f_{a}\bar{\partial}_{\bar{\jmath}}f_{b}-\partial_{i}f_{b}\bar{\partial}_{\bar{\jmath}}f_{a})=0,\qquad\forall\,a,b,$ define a foliation by Lagrangian submanifolds. One can recover such commutation relations as the classical limit of $n$ commuting self-adjoint operators on the Hilbert space ${\cal H}_{\kappa}$ [5]: $\langle\tilde{\Omega}_{x,\kappa}|[\hat{f}_{a},\hat{f}_{b}]|\tilde{\Omega}_{x,\kappa}\rangle\sim{i\over\kappa}\\{f_{a},f_{b}\\}_{PB}+O(\kappa^{-2}),$ with $\langle\tilde{\Omega}_{x,\kappa}|\hat{f}_{a}|\tilde{\Omega}_{x,\kappa}\rangle\to f_{a}(x)$, and $|\tilde{\Omega}_{x,\kappa}\rangle$ the coherent state peaked at $x\in X$. Thus, one can approximate Lagrangian submanifolds by using $n$-tuples of commuting matrices for large enough $\kappa$. One can impose further conditions, i.e. $\mathrm{Im}(\Omega)|_{\rm SLag}=0$, in order to describe special Lagrangian submanifolds. More precisely, we define the quantum operator $\displaystyle{\cal I}_{\bar{\alpha}\beta\bar{\alpha}_{1}\beta_{1}\dots\bar{\alpha}_{n}\beta_{n}}={1\over 2i}\int_{M}{\omega^{n}\over n!}\bar{s}_{\bar{\alpha}}s_{\beta}{\rm e}^{-\kappa K}\Big{(}\Omega_{i_{1}\dots i_{n}}\partial^{i_{1}}\big{(}\bar{s}_{\bar{\alpha}_{1}}s_{\beta_{1}}{\rm e}^{-\kappa K}\big{)}\cdots\partial^{i_{n}}\big{(}\bar{s}_{\bar{\alpha}_{n}}s_{\beta_{n}}{\rm e}^{-\kappa K}\big{)}$ $\displaystyle\phantom{{\cal I}_{\bar{\alpha}\beta\bar{\alpha}_{1}\beta_{1}\dots\bar{\alpha}_{n}\beta_{n}}=}{}-\overline{\Omega}_{\bar{\imath}_{1}\dots\bar{\imath}_{n}}\partial^{\bar{\imath}_{1}}\big{(}\bar{s}_{\bar{\alpha}_{1}}s_{\beta_{1}}{\rm e}^{-\kappa K}\big{)}\cdots\partial^{\bar{\imath}_{n}}\big{(}\bar{s}_{\bar{\alpha}_{n}}s_{\beta_{n}}{\rm e}^{-\kappa K}\big{)}\Big{)},$ with $\partial^{i}=g^{i\bar{\jmath}}\bar{\partial}_{\bar{\jmath}}$ and $\partial^{\bar{\imath}}=g^{\bar{\imath}j}\partial_{j}$. If ${\rm Herm}({\cal H}_{\kappa})$ is the space of hermitian matrices in ${\cal H}_{\kappa}$ and ${\rm Comm}(\oplus^{n}{\rm Herm}({\cal H}_{\kappa}))$ is the space of $n$ mutually commuting tuples of hermitian matrices in ${\cal H}_{\kappa}$, we can write the map as ${\cal I}\colon{\rm Comm}(\oplus^{n}{\rm Herm}({\cal H}_{\kappa}))\to{\rm Herm}({\cal H}_{\kappa})$. Therefore, one can use the kernel of ${\cal I}$ to approximate special Lagrangian submanifolds as the level sets of the $n$ functions “$\langle\tilde{\Omega}_{x,\kappa}|{\rm ker}({\cal I})|\tilde{\Omega}_{x,\kappa}\rangle$”. Also, one can generalize this quantum system by coupling the particle to a rank $r$ holomorphic vector bundle $V\to M$. We will not give many details of this generalization here, although we will say a few words. For instance, the system can be interpreted as a particle endowed with certain $U(r)$-charge. The associated quantum Hilbert space is $H^{0}(M,V\otimes{\cal L}^{\otimes\kappa})$. One can also define an analogous set of coherent states and an associated distortion function. In the semiclassical limit, requiring the generalized distortion function to be constant as a function of $M$ gives rise to generalized balanced metrics, and therefore, to hermite-Yang–Mills metrics on $V\to M$ when $\kappa^{-1}=0$ [11]. Finally, as a technical comment, the _balanced metric_ equations (4) and (3) can be explicitly solved for finite $\kappa$, and its solutions used to approximate Ricci-flat metrics and hermitian Yang–Mills connections. A method to solve them involves the concepts of T-map and algebraic Monte-Carlo integration [10, 12], which can be applied whenever one has enough analytical control on the Kodaira’s embeddings $M\hookrightarrow\mathbb{P}H^{0}(M,{\cal L}^{\otimes\kappa})$. We leave the problem of developing technical methods for constructing special Lagrangian submanifolds and other geometric objects for future work. In the following sections we will show how the concept of _balanced metric_ appears naturally in other frameworks for quantization (Berezin’s star product and path integral quantization), and thus gives rise to Kähler–Einstein metrics in the classical limit. ## 3 Berezin’s star product Instead of quantizing the space of observables by introducing a Hilbert space of states, ${\cal H}_{\kappa}$, and its corresponding space of quantum observables (i.e., the hermitian matrices), one can understand quantization as a noncommutative deformation of the geometry of $M$. In the deformation quantization approach to noncommutative geometry, the ordinary algebra of functions $C^{\infty}(M)$ is replaced by the noncommutative $\star$ algebra $C^{\infty}(M)[[\kappa^{-1}]]$, which reflects the operator algebra of hermitian operators on ${\cal H}_{\kappa}$. The $\star$ product of two elements in $C^{\infty}(M)[[\kappa^{-1}]]$ is defined through formal series expansions in powers of $\kappa^{-1}$, such that, $[f,g]:=f\star g-g\star f=i\kappa^{-1}\\{f,g\\}_{PB}+O\big{(}\kappa^{-2}\big{)}.$ The explicit form of the algebra is not unique [17], in the same way that the quantization of a classical system is not unique. Here, we will first explore the Reshetikhin–Takhtajan star product in Kähler geometry [21]. To describe this algebra, we first introduce the diagonal of the Bergman kernel and the Calabi’s diastatic function. Using the notation introduced above, the diagonal of the Bergman kernel can be written as $e(z,\bar{z})=\sum_{\alpha}\bar{s}_{\bar{\alpha}}(\bar{z})s_{\alpha}(z)\exp(-\kappa K(z,\bar{z})),$ which coincides with the distortion function defined in (2). The Calabi function is simply $\phi(z,\bar{z};v,\bar{v})=K(z,\bar{v})+K(v,\bar{z})-K(z,\bar{z})-K(v,\bar{v}).$ Note that $e(z,\bar{z})$ and $\phi(z,\bar{z};v,\bar{v})$ are invariant under Kähler transformations $K\to K+f+\bar{f}$. Using the Berezin’s formula, one can define a non-normalized product given by $(f\bullet g)(z,\bar{z}):=\int_{M}f(z,\bar{v})g(v,\bar{z})\exp\left(-\kappa\phi(z,\bar{z};v,\bar{v})\right){\omega^{n}\over n!},$ which can be used to introduce the normalized product $(f\star g)(z,\bar{z})=\int_{M}f(z,\bar{v})g(v,\bar{z}){e(z,\bar{v})e(v,\bar{z})\over e(z,\bar{z})}\exp\left(-\kappa\phi(z,\bar{z};v,\bar{v})\right){\omega^{n}\over n!}.$ The Calabi’s diastatic function is defined in some neighborhood of the diagonal $M\times M$, and the point $v=z$ is a critical point of the Calabi function considered as a function of $v$ and $\bar{v}$; the Laplace expansion of $e^{-\kappa\phi}$ at the critical point $v=z$ yields a formal power series in $\kappa^{-1}$. As it is shown in [21], one can determine naturally the $\bullet$ product as a combinatoric expansion in powers of $\kappa^{-1}$, derived from the Laplace expansion of the diastatic function. Therefore, one has to compute the $\bullet$ product in order to determine the normalized $\star$ product. The unit element of their noncommutative deformation $C^{\infty}(M)[[\hbar]]$ given by the $\bullet$ product is the diagonal of the Bergman kernel $e(z,\bar{z})$. Therefore, as a corollary, if the unit element of the Reshetikhin and Takhtajan algebra is constant, i.e. the corresponding Bergman kernel has a constant diagonal, the metric on $M$ is balanced. This shows how balanced metrics become natural objects in deformation quantization. One can study this phenomenon further in the path integral quantization formalism. ## 4 Completely degenerated quantum systems In this section we compute the quantum vacuum energy density associated to a constant Hamiltonian function on $M$, in the semiclassical limit, $\hbar=\kappa^{-1}\to 0$. In the geometric quantization framework, the quantum Hamiltonian associated with the classical Hamiltonian function ${\bf 1}\colon M\to 1$, is the identity operator ${\rm Id}$; if we identify the semiclassical vacuum states with the coherent states, the semiclassical vacuum energy density will be proportional to $\langle\tilde{\Omega}_{x}|\tilde{\Omega}_{x}\rangle$. However, in the path integral approach, the Hamiltonian is set to be zero, and the choice of vacuum is not necessarily the same as the identification “coherent state” = “vacuum state.” By subtracting the classical energy density to $\langle\tilde{\Omega}_{x}|\tilde{\Omega}_{x}\rangle$, we will compare the path integral approach and the geometric quantization approach, and find that the leading term in $\kappa^{-1}$ is the same, though the first sub-leading correction is not. This means that requiring the semiclassical vacuum energies (in both quantization frameworks) to be constant, yields metrics on $M$ that become Kähler–Einstein in the classical limit. The fact that the sub-leading corrections are different only affects higher corrections to the aforementioned metrics when $\kappa^{-1}$, though small, it is not zero. ### 4.1 Vacuum energy in geometric quantization The system is completely degenerated when the Hamiltonian function is constant; each point in the phase space is a _classical vacuum state_ and the quantum Hilbert space becomes the space of _quantum vacua_. On the geometric quantization side, one identifies the quantum Hilbert space ${\cal H}_{\kappa}$ with the space of holomorphic sections $H^{0}(M,{\cal L}^{\otimes k})$. The natural candidate to be the semiclassical quantum vacuum state peaked at $x$ is the coherent state introduced by Rawnsley [20], and denoted by $|\tilde{\Omega}_{x}\rangle$. We construct $|\tilde{\Omega}_{x}\rangle$ as follows: first, we choose an orthonormal basis of holomorphic sections, $\\{s_{\alpha}\in H^{0}({\cal L}^{\otimes k})\\}_{1}^{N(\kappa)}$ with $\langle s_{\alpha}|s_{\beta}\rangle=\delta_{\bar{\alpha}\beta}=\int_{M}\bar{s}_{\bar{\alpha}}(\bar{x})s_{\beta}(x){\rm e}^{-\kappa K(x,\bar{x})}{\omega^{n}(x,\bar{x})\over n!},$ where $s_{\alpha}(x){\rm e}^{-\kappa K(x,\bar{x})/2}$ is the complex number associated with the evaluation of the holomorphic section $s_{\alpha}$ at $x\in U$, and defined in the trivialization ${\cal L}^{\otimes\kappa}|_{U}\simeq\mathbb{C}\times U$. Second and lastly, one can define the _coherent state_ peaked at $x\in M$ as the ray in $\mathbb{P}{\cal H}_{\kappa}$ generated by $|\tilde{\Omega}_{x}\rangle:=\sum_{\alpha=1}^{N(\kappa)}s_{\alpha}(x){\rm exp}\left(-\kappa K(\bar{x},x)/2\right)|s_{\alpha}\rangle\in{\cal H}_{\kappa},\quad x\in U\subset M,$ and one can easily show how such a ray is independent of the choice of trivialization of the line bundle. The set of coherent states $\\{|\tilde{\Omega}_{x}\rangle\\}_{x\in M}$ is a supercomplete system of vectors in ${\cal H}_{\kappa}$, parametrized by the points of $M$. It also defines an embedding of $M$ into $\mathbb{P}{\cal H}_{\kappa}$, and implies the Parseval identity (1). This allows the definition of an embedding of the space of quantum observables in ${\cal H}_{\kappa}$ (i.e., the self-adjoint matrices in ${\rm Herm}({\cal H}_{\kappa})\subset{\cal H}_{\kappa}^{\ast}\otimes{\cal H}_{\kappa}$) into the space of classical observables $C^{\infty}(M)$, according to the formula $\hat{f}\mapsto\langle\tilde{\Omega}_{\circ}|\hat{f}|\tilde{\Omega}_{\circ}\rangle=f\in C^{\infty}(M),\qquad\hat{f}=\hat{f}^{\dagger},$ where $\circ$ denotes the pre-image of $f$ in $M$. The function $f$ is called a _covariant_ symbol of the matrix $\hat{f}$. A function $\check{f}\in C^{\infty}(M)$ such that the matrix $\hat{f}$ is representable as $\hat{f}=\int_{M}|\tilde{\Omega}_{x}\rangle\otimes\langle\tilde{\Omega}_{x}|{\check{f}\omega^{n}(x,\bar{x})\over n!},$ is called _contravariant_ symbol of the matrix $\hat{f}$. The map $\check{f}\mapsto\hat{f}$ is also known as _Toeplitz map_ , $T\colon C^{\infty}(M)\to{\rm Herm}({\cal H}_{\kappa})\subset{\cal H}_{\kappa}^{\ast}\otimes{\cal H}_{\kappa}$. It is important to stress that the identification between the _quantum vacuum state_ $|\Omega_{x}\rangle$ localized at $x\in M$ and the _coherent state_ $|\tilde{\Omega}_{x}\rangle$ is, in somehow, made arbitrarily. Such an identification is motivated by the fact that the coherent state is peaked at $x$ and localized within a neighborhood $V_{x}\subset M$ with minimal quantum uncertainty. More precisely, the coherent state $|\tilde{\Omega}_{x,\kappa}\rangle$ satisfies $\displaystyle{1\over\langle\tilde{\Omega}_{x,\kappa}|\tilde{\Omega}_{x,\kappa}\rangle}\int_{V_{x}\subset M}\langle\tilde{\Omega}_{x,\kappa}|\tilde{\Omega}_{y,\kappa}\rangle\langle\tilde{\Omega}_{y,\kappa}|\tilde{\Omega}_{x,\kappa}\rangle{\omega^{n}(y,\bar{y})\over n!}\sim 1,$ (5) with $\int_{V_{x}}\omega^{n}/\int_{M}\omega^{n}\sim{1\over N}$. Such an identification is correct as a first approximation in $\kappa^{-1}$, although the $O(\kappa^{-2})$ terms are not universal and depend on the choice of vacuum state $|\Omega_{x,\kappa}\rangle$. The $O(\kappa^{-2})$ corrections are important at the time of computing correlation functions of the type $\displaystyle\langle\Omega_{x,\kappa}|\hat{f}_{1}\hat{f}_{2}\cdots\hat{f}_{m}|\Omega_{x,\kappa}\rangle$ (6) in the limit $\kappa\\!\to\\!\infty$, as power series in $\kappa^{-1}$ of the covariant symbols $f_{1},f_{2},\ldots,f_{m}\\!\in\\!C^{\infty}(M)[[\kappa^{-1}]]$. For instance, every identification of the vacuum state with peaked states that obey equation (5), gives rise to the same semiclassical limit $\langle\Omega_{x,\kappa}|[\hat{f},\hat{g}]|\Omega_{x,\kappa}\rangle=i\kappa^{-1}\omega^{i\bar{\jmath}}\left(\partial_{i}f(x,\bar{x})\bar{\partial}_{\bar{\jmath}}g(x,\bar{x})-\partial_{i}g(x,\bar{x})\bar{\partial}_{\bar{\jmath}}f(x,\bar{x})\right)+O\big{(}\kappa^{-2}\big{)},$ with $f(x,\bar{x}):=\langle\Omega_{x,\kappa}|\hat{f}|\Omega_{x,\kappa}\rangle$ and $g(x,\bar{x}):=\langle\Omega_{x,\kappa}|\hat{g}|\Omega_{x,\kappa}\rangle$, [5]. However, the higher corrections $O(\kappa^{-2})$ will depend on the choice of vacuum state. To compute the semiclassical limit of the correlators (6) beyond $O(\kappa^{-1})$ is a difficult task which involves hard analysis; see [19] for the most recent results. For simplicity, we study only the vacuum expectation value of the identity operator using the naïve vacuum state $|\tilde{\Omega}_{x,\kappa}\rangle$. Z. Lu computed the lower order terms in powers of $\kappa^{-1}$ of the squared norm of the naïve vacuum state, [19], $\displaystyle\langle\tilde{\Omega}_{x,\kappa}|\tilde{\Omega}_{x,\kappa}\rangle=1+{1\over 2\kappa}R+{1\over 3\kappa^{2}}\left(\Delta R+{1\over 8}\left(|\mathrm{Riemann}|^{2}-4|\mathrm{Ricci}|^{2}+3R^{2}\right)\right)+O\big{(}\kappa^{-3}\big{)}.$ Computing the asymptotic series of the vacuum energy on the path integral side involves perturbative expansions of Feynman vacuum diagrams. As the classical energy density is set to be zero in the path integral formalism, one should compare the path integral result to the “renormalized” geometric quantized vacuum energy $\displaystyle E_{0}(x)=\langle\tilde{\Omega}_{x,\kappa}|\tilde{\Omega}_{x,\kappa}\rangle-1$ $\displaystyle\phantom{E_{0}(x)}{}={1\over 2\kappa}R+{1\over 3\kappa^{2}}\left(\Delta R+{1\over 8}\left(|\mathrm{Riemann}|^{2}-4|\mathrm{Ricci}|^{2}+3R^{2}\right)\right)+O\big{(}\kappa^{-3}\big{)},$ (7) which is zero at $\kappa=\infty$. ### 4.2 Path integral derivation of the vacuum energy On the _path integral quantization_ side, Cattaneo and Felder [7] give a prescription for computing correlation functions for quantized observables $f,g\in C^{\infty}(M)[[\kappa^{-1}]]$, by evaluating path integrals perturbatively as formal expansions in powers of $\kappa^{-1}$. In such perturbative expansion one considers perturbations around the constant map, i.e., the solution of the equations of motion or _classical vacuum state_ $\Phi_{0}\colon{\mathbb{R}}\mapsto x\in M$. For simplicity, we choose a local coordinate chart around $x$, $\\{\phi_{i}\\}_{0<i\leq n}$ given by the Kähler–Riemann normal coordinates [2]. For a vanishing Hamiltonian, the phase- space action associated with perturbations $\Phi$ around the classical vacuum state $\Phi_{0}$ is the line integral of the $U(1)$-connection on the prequantum bundle ${\cal L}^{\otimes k}$, along the path $\Phi$ in $M$ $\displaystyle S[\Phi]=i\int_{-\infty}^{\infty}dt\bar{\partial}_{\bar{\jmath}}K(\phi,\bar{\phi}){d\bar{\phi}^{\bar{\jmath}}\over dt},$ (8) with $\Phi\in{\rm Maps}({\mathbb{R}}\to M)$ and444We denote by ${\rm Maps}({\mathbb{R}}\to M)$ the functional space of maps of the real line ${\mathbb{R}}$ to the phase space $M$. $\Phi(\pm\infty)=x$, implying $\phi(\pm\infty)=0$ in the local Kähler–Riemann normal coordinate system around $x$. The functional integration of fluctuations around the classical vacuum $x\in M$ defines a semiclassical quantum vacuum state that we denote as $|\Omega_{x,\kappa}\rangle$, although we don’t know how to describe it as an element of the Hilbert space, ${\cal H}_{\kappa}$. The prescription for computing the correlation functions that appear in deformation quantization [7, 18] is given by the path integral on the phase- space variables $\displaystyle\langle\Omega_{x}|\hat{f}\hat{g}|\Omega_{x}\rangle:=f\star g(x,\bar{x})=\int_{{\rm Maps}({\mathbb{R}}\to M)|\Phi(\pm\infty)=x}d\Phi\,f(\Phi(1))g(\Phi(0))\exp\left(i\kappa S[\Phi]\right),$ (9) with $f,g\in C^{\infty}(M)[[\kappa^{-1}]]$. In Riemann normal coordinates we can write the action (8) as follows $\displaystyle i\kappa S[\Phi]=-\kappa\int_{{\mathbb{R}}}dt\bar{\partial}_{\bar{\jmath}}K(\phi,\bar{\phi}){d\bar{\phi}^{\bar{\jmath}}\over dt}=-\kappa\int_{{\mathbb{R}}}dt\Bigg{(}g_{i\bar{\jmath}}(x,\bar{x})\phi^{i}\dot{\bar{\phi}}^{\bar{\jmath}}+{1\over 2}R_{i\bar{\jmath}k\bar{l}}\phi^{i}\bar{\phi}^{\bar{\jmath}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}$ $\displaystyle\phantom{i\kappa S[\Phi]=}{}+{1\over 6}D_{m}R_{i\bar{\jmath}k\bar{l}}\phi^{m}\phi^{i}\bar{\phi}^{\bar{\jmath}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}+{1\over 6}\bar{D}_{\bar{m}}R_{i\bar{\jmath}k\bar{l}}\bar{\phi}^{\bar{m}}\phi^{i}\bar{\phi}^{\bar{\jmath}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}+{1\over 12}D_{n}D_{m}R_{i\bar{\jmath}k\bar{l}}\phi^{n}\phi^{m}\phi^{i}\bar{\phi}^{\bar{\jmath}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}$ $\displaystyle\phantom{i\kappa S[\Phi]=}{}+{1\over 12}\bar{D}_{\bar{n}}\bar{D}_{\bar{m}}R_{i\bar{\jmath}k\bar{l}}\bar{\phi}^{\bar{n}}\bar{\phi}^{\bar{m}}\phi^{i}\bar{\phi}^{\bar{\jmath}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}+{1\over 12}\bar{D}_{(\bar{n}}D_{m}R_{i\bar{\jmath}k\bar{l})}\phi^{m}\bar{\phi}^{\bar{n}}\phi^{i}\bar{\phi}^{\bar{\jmath}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}$ $\displaystyle\phantom{i\kappa S[\Phi]=}{}+{1\over 4}g^{o\bar{r}}R_{o(\bar{\jmath}m\bar{l}}R_{i\bar{n}k)\bar{r}}\phi^{m}\bar{\phi}^{\bar{n}}\phi^{i}\bar{\phi}^{\bar{\jmath}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}+O\big{(}\phi^{7}\big{)}\cdots\Bigg{)},$ where $\dot{\phi}=d\phi/dt$, and the parentheses enclosing indices indicate the completely symmetric part of such indices; we sum over repeated indices. Still, the measure $d\Phi$ in the functional integration (9) also depends of the phase-space coordinate field $\phi$. Hence, in this choice of coordinates $\displaystyle{\rm e}^{-iE_{0}^{\prime}(x)\delta(0)}=\\!\int_{{\rm Maps}({\mathbb{R}}\to M)|\Phi(\pm\infty)=x}\\!\prod_{i=1}^{n}d\phi^{i}\bar{d}\bar{\phi}^{\bar{i}}\det\omega\left(\phi,\bar{\phi}\right)\exp\\!\left(\\!-\kappa\\!\int_{{\mathbb{R}}}dt\bar{\partial}_{\bar{\jmath}}K(\phi,\bar{\phi}){d\bar{\phi}^{\bar{\jmath}}\over dt}\right),\\!\\!\\!\\!$ (10) where $\det\omega$ is the determinant of the Kähler form $\omega_{i\bar{\jmath}}$, and $E_{0}^{\prime}(x)$ is the quantum vacuum energy density, depending on the choice of semiclassical vacuum labeled as $x\in M$. Therefore, as $\omega$ depends on the integration variables, we can introduce an anti-commuting auxiliary field $\lambda$ to write the path integral using a standard gaussian measure $\displaystyle{\rm e}^{-iE_{0}^{\prime}(x)\delta(0)}=\int\prod_{i=1}^{n}d\phi^{i}\bar{d}\bar{\phi}^{\bar{i}}d\lambda^{i}\bar{d}\bar{\lambda}^{\bar{i}}\exp\left(-\kappa\int_{{\mathbb{R}}}dt\bar{\partial}_{\bar{\jmath}}K(\phi,\bar{\phi}){d\bar{\phi}^{\bar{\jmath}}\over dt}+\int_{{\mathbb{R}}}dt\,\omega(\phi,\bar{\phi})_{i\bar{\jmath}}\lambda^{i}\bar{\lambda}^{\bar{\jmath}}\right),$ where the functional integral of the auxiliary field $\lambda$ obeys the rules of the Grassmann integration. One can expand the action for the auxiliary field, in powers of the field $\phi$, to find out the interactions between auxiliary field and phase-space coordinate field, $\displaystyle\int_{{\mathbb{R}}}dt\,\omega(\phi,\bar{\phi})_{i\bar{\jmath}}\lambda^{i}\bar{\lambda}^{\bar{\jmath}}=i\int_{{\mathbb{R}}}dt\Bigg{(}g_{i\bar{\jmath}}\lambda^{i}\bar{\lambda}^{\bar{\jmath}}+R_{i\bar{\jmath}k\bar{l}}\lambda^{i}\bar{\lambda}^{\bar{\jmath}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}+{1\over 12}\bar{D}_{(\bar{n}}D_{m}R_{i\bar{\jmath}k\bar{l})}\lambda^{i}\bar{\lambda}^{\bar{\jmath}}\phi^{m}\bar{\phi}^{\bar{n}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}$ $\displaystyle\hphantom{\int_{{\mathbb{R}}}dt\,\omega(\phi,\bar{\phi})_{i\bar{\jmath}}\lambda^{i}\bar{\lambda}^{\bar{\jmath}}=}{}+{1\over 4}g^{o\bar{r}}R_{o(\bar{\jmath}m\bar{l}}R_{i\bar{n}k)\bar{r}}\lambda^{i}\bar{\lambda}^{\bar{\jmath}}\phi^{m}\bar{\phi}^{\bar{n}}\phi^{k}\dot{\bar{\phi}}^{\bar{l}}+\cdots\Bigg{)},$ where the coefficients $g_{i\bar{\jmath}}$, $R_{i\bar{\jmath}k\bar{l}}$, etc., are evaluated at $x\in M$. Therefore, we can evaluate $E_{0}^{\prime}(x)$ perturbatively as an expansion of the path integral in powers of $\kappa^{-1}$, in the limit $\kappa\to\infty$. If we write the Fourier transform of the field to the momentum variables as $\displaystyle\hat{\phi}(p)={1\over 2\pi}\int_{{\mathbb{R}}}dt\,\exp(ipt)\phi(t),\qquad\phi(\pm\infty)=0,$ $\displaystyle\hat{\lambda}(p)={1\over 2\pi}\int_{{\mathbb{R}}}dt\,\exp(ipt)\lambda(t),\qquad\lambda(\pm\infty)=0,$ the propagators in the momentum space are Figure 1: Propagators for the phase-space coordinate field (left), and the auxiliary field (right). In order to compute (10), we perform a perturbative expansion in powers of $\kappa^{-1}=\hbar$. We can compute the vacuum energy $E_{0}^{\prime}(x)$, by simply summing the connected vacuum (or bubble) diagrams, and dividing by the total length of ${\mathbb{R}}$. Thus, a vacuum diagram with $L$ _loops_ contributes to $E_{0}^{\prime}(x)$ a term proportional to $\kappa^{1-L}=\kappa^{V-P}$, with $V$ the _number of vertices_ and $P$ the _number of propagators_. Therefore, to determine $E_{0}^{\prime}(x)$ up to order $\kappa^{-2}$, we have to sum the connected diagrams depicted in Figs. 4–10. Figure 2: Interaction vertices for the phase-space field. $p_{\bar{l}}$ denotes the momentum carried by the particle which propagates along the $\bar{l}$-leg. Figure 3: Interaction vertices for the auxiliary field with the phase-space field. As we want only to evaluate diagrams up to order $\kappa^{-2}$, we only need to consider a few interactions; by drawing the diagrams, one realizes that the only vertices that appear are the ones depicted in Figs. 2 and 3. The asymptotic expansion of $E_{0}^{\prime}(x)$ in powers of $\kappa^{-1}$ can be expressed as $\displaystyle- iE_{0}^{\prime}(x)\delta(0)=\sum_{\Gamma\in\mathfrak{G}_{2}}{1\over\kappa|{\rm Aut}(\Gamma)|}D_{\Gamma}(x)+\sum_{\Gamma\in\mathfrak{G}_{3}}{1\over\kappa^{2}|{\rm Aut}(\Gamma)|}D_{\Gamma}(x)+O\big{(}\kappa^{-3}\big{)},$ (11) where $\mathfrak{G}_{L}$ is the set of bubble diagrams with $L$ loops, ${\rm Aut}(\Gamma)$ is the subgroup of the group of automorphisms of $\Gamma$ that maps vertices to vertices of the same type and oriented propagators to oriented propagators of the same type (which start and end at the same vertices), $|{\rm Aut}(\Gamma)|=\\#{\rm Aut}(\Gamma)$ is also known as _symmetry factor_ , and $D_{\Gamma}(x)$ is the evaluation of the Feynman diagram. Figure 4: Two-loop vacuum diagrams, (2.i) left and (2.ii) right. The evaluation of each diagram $D_{\Gamma}(x)$ follows from the Feynman rules in momentum space: to each line we associate its corresponding propagator (Fig. 1), to each vertex we associate its corresponding numerical factor (Figs. 2 and 3), we impose momentum conservation at each vertex and integrate over each undetermined momentum $\int_{{\mathbb{R}}}{dp\over 2\pi}$. There are two types of integrals that appear in the evaluation of bubble diagrams $\lim_{\varepsilon\to 0^{+}}{1\over 2\pi}\int_{{\mathbb{R}}}{idp\over p-i\varepsilon}={1\over 2}\qquad{\rm and}\qquad\delta(0)={1\over 2\pi}\int_{{\mathbb{R}}}dp.$ Each vacuum diagram is proportional to the Dirac delta $\delta(0)$, or the “total length” of ${\mathbb{R}}$, because the calculation in the momentum space yields the total vacuum energy in ${\mathbb{R}}$. As we are just interested in the vacuum energy _density_ , we will divide out by infinite total length of the ($0+1$)-spacetime, ${\mathbb{R}}$. Thus, in order to determine $E_{0}^{\prime}(x)$ up to three loops, we use equation (11). The evaluation of the two- and three-loop diagrams gives rise to the following: for two-loops, * • (2.i) in Fig. 4 $D_{\rm(2.i)}(x)=-iR_{i\bar{\jmath}k\bar{l}}g^{i\bar{\jmath}}g^{k\bar{l}}{1\over(2\pi)^{2}}\int_{{\mathbb{R}}}dp_{1}{ip_{1}\over p_{1}-i\varepsilon}\int_{{\mathbb{R}}}dp_{2}{i\over p_{2}-i\varepsilon}=\delta(0){1\over 2}R.$ * • (2.ii) in Fig. 4 $D_{\rm(2.ii)}(x)=iR_{i\bar{\jmath}k\bar{l}}{1\over(2\pi)^{2}}\int_{{\mathbb{R}}}dp_{1}(-i)g^{k\bar{l}}\int_{{\mathbb{R}}}dp_{2}{ig^{i\bar{\jmath}}\over p_{2}-i\varepsilon}=\delta(0)\frac{1}{2}R.$ Figure 5: Three-loop vacuum diagram (3.i). Figure 6: Three-loop vacuum diagram (3.ii). For three-loop diagrams, * • (3.i) in Fig. 5 $\displaystyle D_{\rm(3.i)}=(-i)^{2}R_{i\bar{\jmath}k\bar{l}}g^{i\bar{j}}R_{m\bar{n}o\bar{p}}g^{m\bar{n}}g^{k\bar{p}}g^{o\bar{l}}\int_{{\mathbb{R}}^{3}}{dp_{1}dp_{2}dp_{3}\over(2\pi)^{3}}{i\over p_{1}-i\varepsilon}\left({i\over p_{2}-i\varepsilon}\right)^{2}p_{2}{ip_{3}\over p_{3}-i\varepsilon}$ $\displaystyle\phantom{D_{\rm(3.i)}}{}=\delta(0){1\over 4}|{\rm Ricci}|^{2}.$ * • (3.ii) in Fig. 6 $\displaystyle D_{\rm(3.ii)}=(-i)^{2}R_{i\bar{\jmath}k\bar{l}}R^{i\bar{\jmath}k\bar{l}}\int_{{\mathbb{R}}^{4}}{dp_{1}dp_{2}dp_{3}dp_{4}\over(2\pi)^{4}}\delta(p_{1}+p_{3}-p_{2}-p_{4})p_{1}p_{2}$ $\displaystyle\phantom{D_{\rm(3.ii)}=}{}\times{i\over p_{1}-i\varepsilon}{i\over p_{2}-i\varepsilon}{i\over p_{3}-i\varepsilon}{i\over p_{4}-i\varepsilon}$ $\displaystyle\phantom{D_{\rm(3.ii)}}{}=-|{\rm Riemann}|^{2}\int_{{\mathbb{R}}^{3}}{dp_{1}dp_{2}dp_{3}\over(2\pi)^{3}}{1\over(p_{3}-i\varepsilon)[(p_{1}+p_{3}-p_{2})-i\varepsilon]}$ $\displaystyle\phantom{D_{\rm(3.ii)}}{}=-|{\rm Riemann}|^{2}\int_{{\mathbb{R}}^{3}}{dl_{1}dl_{2}dl_{3}\over(2\pi)^{3}}{i\over(l_{1}-i\varepsilon)(l_{2}-i\varepsilon)}=\delta(0){1\over 4}|{\rm Riemann}|^{2},$ where $l_{1}=p_{1}+p_{2}-p_{3}$, $l_{2}=p_{3}$ and $l_{3}=p_{2}$. Figure 7: Three-loop vacuum diagram (3.iii). * • (3.iii) in Fig. 7 $\displaystyle D_{\rm(3.iii)}={-i\over 4}\left(\Delta R+3\left({2\over 3}|{\rm Ricci}|^{2}+{1\over 3}|{\rm Riemann}|^{2}\right)\right)$ $\displaystyle\phantom{D_{\rm(3.iii)}=}{}\times\lim_{\varepsilon\to 0^{+}}\int_{{\mathbb{R}}^{3}}{dp_{1}dp_{2}dp_{3}\over(2\pi)^{3}}{i\over p_{1}-i\varepsilon}{i\over p_{2}-i\varepsilon}{ip_{3}\over p_{3}}$ $\displaystyle\phantom{D_{\rm(3.iii)}}{}=\delta(0){1\over 16}\left(\Delta R+2|{\rm Ricci}|^{2}+|{\rm Riemann}|^{2}\right).$ * • (3.iv) in Fig. 8. Similarly to (3.i) $D_{\rm(3.iv)}=\delta(0){1\over 4}|{\rm Ricci}|^{2}.$ * • (3.v) in Fig. 8 $D_{\rm(3.v)}=\delta(0){1\over 4}|{\rm Ricci}|^{2}.$ Figure 8: Three-loop vacuum diagrams with auxiliary field, (3.iv) left and (3.v) right. Figure 9: Three-loop vacuum diagram with auxiliary field (3.vi). Figure 10: Three-loop vacuum diagram with auxiliary field (3.vii). * • (3.vi) in Fig. 9. Similarly to (3.ii), $\displaystyle D_{\rm(3.vi)}=(iR_{i\bar{m}k\bar{o}})(iR_{n\bar{\jmath}p\bar{l}})\big{(}-ig^{i\bar{\jmath}}\big{)}\big{(}-ig^{k\bar{l}}\big{)}\int_{{\mathbb{R}}^{3}}{dp_{1}dp_{2}dp_{3}\over(2\pi)^{3}}{ig^{n\bar{m}}\over p_{1}-i\varepsilon}{ig^{p\bar{o}}\over p_{2}-i\varepsilon}$ $\displaystyle\phantom{D_{\rm(3.vi)}}{}=\delta(0){1\over 4}|{\rm Riemann}|^{2}.$ * • (3.vii) in Fig. 10. Similarly to (3.iii), $\displaystyle D_{\rm(3.vii)}=i\left(\Delta R+3\left({2\over 3}|{\rm Ricci}|^{2}+{1\over 3}|{\rm Riemann}|^{2}\right)\right)\int_{{\mathbb{R}}^{3}}{dp_{1}dp_{2}dp_{3}\over(2\pi)^{3}}{i\over p_{1}-i\varepsilon}{i\over p_{2}-i\varepsilon}$ $\displaystyle\phantom{D_{\rm(3.vii)}}{}=\delta(0){1\over 4}\left(\Delta R+2|{\rm Ricci}|^{2}+|{\rm Riemann}|^{2}\right).$ Finally, including the _symmetry factors_ of each diagram, and summing them as in equation (11), yields the vacuum energy density associated to the semiclassical vacuum state localized at $x\in M$, $\displaystyle E_{0}^{\prime}(x)={1\over 2\kappa}R+{1\over 96\kappa^{2}}\left(5\Delta R+42|{\rm Ricci}|^{2}+17|{\rm Riemann}|^{2}\right)+O\left({1\over\kappa^{3}}\right).$ (12) Thus, comparing equation (12) with the equivalent result in geometric quantization (7), yields different vacuum energy densities $E_{0}(x)\neq E_{0}^{\prime}(x)$, despite the fact that the leading terms are identical. The corollary is an interesting one: fixing $\kappa$ and requiring the quantum vacuum energy density to be constant on the quantum moduli space of semiclassical vacua is equivalent to endowing $M$ with a “generalized balanced metric”. This generalized notion of balanced metric gives rise to the same Kähler–Einstein metrics in the classical limit, which shows that the emergence of Kähler–Einstein metrics in the classical limit is generic for a broad choice of semiclassical vacuum states. ## 5 Conclusion We have shown how the Kähler–Einstein metrics appear naturally in the classical limit of Kähler quantization. In geometric quantization, identifying semiclassical vacuum states with coherent states allows us to define balanced metrics as those metrics which yield constant semiclassical vacuum energy (for constant classical Hamiltonian). In the Berezin’s approach to deformation quantization, the unit element of the noncommutative algebra $C^{\infty}(M)[[\kappa^{-1}]]$ is the constant function, if and only if the metric is balanced. Also in path integral quantization, requiring the semiclassical vacuum energy to be constant yields a metric that is Kähler–Einstein in the classical limit. Strictly speaking, the metrics that appear in path integral quantization are not balanced. This is due to a different choice of vacuum states in the path integral formalism; thus, for each choice of moduli spaces of semiclassical vacua one can define different _generalized balanced metrics_. It would be interesting to study the properties exhibited by this general class of metrics. For instance, it is especially interesting to understand how introducing quantum corrections to the Kähler potential deforms the moduli of semiclassical vacua [17]. Another interesting problem would be to understand balanced metrics in vector bundles within the framework of Kähler quantization. Also, one could explicitly construct special Lagrangian submanifolds in Calabi–Yau threefolds, and give a _geometric quantization_ formulation of the Bressler–Soilbeman conjecture [6] (which conjectures a correspondence of the Fukaya category with a certain category of holonomic modules over the quantized algebra of functions). A final motivation for future research comes from the fact that the geometric objects explored in this paper appear in the large volume limit of string theory compactifications. We have shown how these objects can be explicitly constructed in the semiclassical limit of geometric quantization; one would expect that different areas of string theory, such as Matrix theory, black holes, and Calabi–Yau compactification theory [4, 8, 14, 15], where the quantized algebra of functions plays a special role, could be understood better through a deeper study of the ideas explored here. ### Acknowledgements It is a pleasure to thank T. Banks, E. Diaconescu, M. Douglas, R. Karp, S. Klevtsov, and specially the author’s advisor G. Moore, for valuable discussions. We would like to thank as well G. Moore and G. Torroba for their comments on the manuscript, and J. Nannarone for kind encouragement and support. This work was supported by DOE grant DE-FG02-96ER40949. ## References * [1] * [2] Alvarez-Gaumé L., Freedman D.Z., Mukhi S., The background field method and the ultraviolet structure of the supersymmetric nonlinear $\sigma$-model, Ann. Physics 134 (1981), 85–109. * [3] Axelrod S., Della Pietra S., Witten E., Geometric quantization of Chern–Simons gauge theory, J. Differential Geom. 33 (1991), 787–902. * [4] Banks T., Fischler W., Shenker S.H., Susskind L., M theory as a matrix model: a conjecture, Phys. Rev. D 55 (1997), 5112–5128, hep-th/9610043. * [5] Bordemann M., Meinrenken E., Schlichenmaier M., Toeplitz quantization of Kähler manifolds and $gl(N)$, $N\to\infty$ limits, Comm. Math. Phys. 165 (1994), 281–296, hep-th/9309134. * [6] Bressler P., Soibelman Y., Mirror symmetry and deformation quantization, hep-th/0202128. * [7] Cattaneo A.S., Felder G., A path integral approach to the Kontsevich quantization formula, Comm. Math. Phys. 212 (2000), 591–611, math.QA/9902090. * [8] Cornalba L., Taylor W., Holomorphic curves from matrices, Nuclear Phys. B 536 (1998), 513–552, hep-th/9807060. * [9] Donaldson S.K., Scalar curvature and projective embeddings. I, J. Differential Geom. 59 (2001), 479–522. * [10] Donaldson S.K., Some numerical results in complex differential geometry, Pure Appl. Math. Q. 5 (2009), 571–618, math.DG/0512625. * [11] Douglas M.R., Karp R.L., Lukic S., Reinbacher R., Numerical solution to the hermitian Yang–Mills equation on the Fermat quintic, J. High Energy Phys. 2007 (2007), no. 12, 083, 24 pages, hep-th/0606261. * [12] Douglas M.R., Karp R.L., Lukic S., Reinbacher R., Numerical Calabi–Yau metrics, J. Math. Phys. 49 (2008), 032302, 19 pages, hep-th/0612075. * [13] Elitzur S., Moore G.W., Schwimmer A., Seiberg N., Remarks on the canonical quantization of the Chern–Simons–Witten theory, Nuclear Phys. B 326 (1989), 108–134. * [14] Gaiotto D., Simons A., Strominger A., Yin X., D0-branes in black hole attractors, J. High Energy Phys. 2006 (2006), no. 3, 019, 24 pages, hep-th/0412179. * [15] Kachru S., Lawrence A.E., Silverstein E., On the matrix description of Calabi–Yau compactifications, Phys. Rev. Lett. 80 (1998), 2996–2999, hep-th/9712223. * [16] Kapustin A., Topological strings on noncommutative manifolds, Int. J. Geom. Methods Mod. Phys. 1 (2004), 49–81, hep-th/0310057. * [17] Karabegov A.V., Deformation quantizations with separation of variables on a Kähler manifold, Comm. Math. Phys. 180 (1996), 745–755, hep-th/9508013. * [18] Kontsevich M., Deformation quantization of Poisson manifolds. I, Lett. Math. Phys. 66 (2003), 157–216, q-alg/9709040. * [19] Lu Z., On the lower order terms of the asymptotic expansion of Tian–Yau–Zelditch, Amer. J. Math. 122 (2000), 235–273, math.DG/9811126. * [20] Rawnsley J.H., Coherent states and Kähler manifolds, Quart. J. Math. Oxford Ser. (2) 28 (1977), no. 112, 403–415. * [21] Reshetikhin N., Takhtajan L.A., Deformation quantization of Kähler manifolds, in L.D. Faddeev’s Seminar on Mathematical Physics, Amer. Math. Soc. Transl. Ser. 2, Vol. 201, Amer. Math. Soc., Providence, RI, 2000, 257–276, math.QA/9907171. * [22] Seiberg N., Witten E., String theory and noncommutative geometry, J. High Energy Phys. 1999 (1999), no. 9, 032, 93 pages, hep-th/9908142. * [23] Souriau J.M., Structure of dynamical systems. A symplectic view of physics, Progress in Mathematics, Vol. 149, Birkhäuser Boston, Inc., Boston, MA, 1997. * [24] Zelditch S., Szegö kernels and a theorem of Tian, Internat. Math. Res. Notices 1998 (1998), no. 6, 317–331, math-ph/0002009.

arxiv-papers

2024-09-04T02:48:53.211902

0710.2156

# COLLABORATIVE OLAP WITH TAG CLOUDS Web 2.0 OLAP Formalism and Experimental Evaluation Kamel Aouiche, Daniel Lemire and Robert Godin Université du Québec à Montréal, 100 Sherbrooke West, Montreal, Canada kamel.aouiche@gmail.com, lemire@acm.org, godin.robert@uqam.ca ## 1 INTRODUCTION The Web 2.0, or Social Web, is about making available social software applications on the Web in an unrestricted manner. Enabling a wide range of distributed individuals to collaborate on data analysis tasks may lead to significant productivity gains [Heer et al., 2007, Wattenberg and Kriss, 2006]. Several companies, like SocialText and IBM, are offering Web 2.0 solutions dedicated to enterprise needs. The data visualization Web sites Many Eyes [IBM, 2007] and Swivel [Swivel, Inc, 2007] have become part of the Web 2.0 landscape: over 1 million data sets were uploaded to Swivel in less than 3 months [Butler, 2007]. These Web 2.0 data visualization sites use traditional pie charts and histograms, but also tag clouds. Tag clouds are a form of histogram which can represent the amplitude of over a hundred items by varying the font size. The use of hyperlinks makes tag clouds naturally interactive. Tag clouds are used by many Web 2.0 sites such as Flickr, del.icio.us and Technorati. Increasingly, e-Commerce sites such as Amazon or O’Reilly Media, are using tag clouds to help their users navigate through aggregated data. Meanwhile, OLAP (On-Line Analytical Processing) [Codd, 1993] is a dominant paradigm in Business Intelligence (BI). OLAP allows domain experts to navigate through aggregated data in a multidimensional data model. Standard operations include drill-down, roll-up, dice, and slice. The data cube [Gray et al., 1996] model provides well-defined semantics and performance optimization strategies. However, OLAP requires much effort from database administrators even after the data has been cleaned, tuned and loaded: schemas must be designed in collaboration with users having fast changing needs and requirements [Body et al., 2002, Morzy and Wrembel, 2004]. Vendors such as Spotfire, Business Objects and QlikTech have reacted by proposing a new class of tools allowing end-user to customize their applications and to limit the need for centralized schema crafting [Havenstein, 2003]. OLAP itself has never been formally defined though rules have been proposed to recognize an OLAP application [Codd, 1993]. In a similar manner, we propose rules to recognize Web 2.0 OLAP applications (see also Table 1): 1. 1. Data and schemas are provided autonomously by users. 2. 2. It is available as a Web application. 3. 3. It supports complete online interaction over aggregated multidimensional data. 4. 4. Users are encouraged to collaborate. Tag clouds are well suited for Web 2.0 OLAP. They are flexible: a tag cloud can represent a dozen or hundred different amplitudes. And they are accessible: the only requirement is a browser that can display different font sizes. We describe a tag-cloud formalism, as an instance of Web 2.0 OLAP. Since we implemented a prototype, technical issues will be discussed regarding application design. In particular, we used iceberg cubes [Carey and Kossmann, 1997] to generate tag clouds online when the data and schema are provided extemporaneously. Because tag clouds are meant to convey a general impression, presenting approximate measures and clustering is sufficient: we propose specific metrics to measure the quality of tag-cloud approximations. We conclude the paper with experimental results on real and synthetic data sets. Table 1: Conventional OLAP versus Web 2.0 OLAP Conventional OLAP | Web 2.0 OLAP ---|--- recurring needs | ephemeral projects predefined schemas | spontaneous schemas centralized design | user initiative histograms | tag clouds plots and reports | iframes, wikis, blogs access control | social networking ## 2 RELATED WORK There are decentralized models [Taylor and Ives, 2006] and systems [Green et al., 2007] to support collaborative data sharing without a single schema. According to Wu et al., it is difficult to navigate an OLAP schema without help; they have proposed a keyword-driven OLAP model [Wu et al., 2007]. There are several OLAP visualization techniques including the Cube Presentation Model (CPM) [Maniatis et al., 2005], Multiple Correspondence Analysis (MCA) [Ben Messaoud et al., 2006] and other interactive systems [Techapichetvanich and Datta, 2005]. Tag clouds have been popularized by the Web site Flickr launched in 2004. Several optimization opportunities exist: similar tags can be clustered together [Kaser and Lemire, 2007], tags can be pruned automatically [Hassan- Montero and Herrero-Solana, 2006] or by user intervention [Millen et al., 2006], tags can be indexed [Millen et al., 2006], and so on. Tag clouds can be adapted to spatio-temporal data [Russell, 2006, Jaffe et al., 2006]. ## 3 OLAP FORMALISM ### 3.1 Conventional OLAP Formalism Most OLAP engines rely on a data cube [Gray et al., 1996]. A data cube $\mathcal{C}$ contains a non empty set of $d$ dimensions $\mathcal{D}=\\{D_{i}\\}_{1\leq i\leq d}$ and a non empty set of measures $\mathcal{M}$. Data cubes are usually derived from a _fact table_ (see Table 2) where each dimension and measure is a column and all rows (or facts) have disjoint dimension tuples. Figure 1(a) gives tridimensional representation of the data cube. Table 2: Fact table example Dimensions | Measures ---|--- location | time | salesman | product | cost | profit Montreal | March | John | shoe | 100$ | 10 $ Montreal | December | Smith | shoe | 150$ | 30 $ Quebec | December | Smith | dress | 175$ | 45 $ Ontario | April | Kate | dress | 90$ | 10 $ Paris | March | John | shoe | 100$ | 20 $ Paris | March | Marc | table | 120$ | 10 $ Paris | June | Martin | shoe | 120$ | 5 $ Lyon | April | Claude | dress | 90$ | 10 $ New York | October | Joe | chair | 100$ | 10 $ New York | May | Joe | chair | 90$ | 10 $ Detroit | April | Jim | dress | 90$ | 10 $ (a) OLAP data cube (b) Tag-cloud data cube (c) OLAP roll-up (d) Tag-cloud roll-up (e) OLAP dice (f) Tag-cloud dice (g) OLAP slice (h) Tag-cloud slice Figure 1: Conventional OLAP operations vs. tag-cloud OLAP operations Measures can be aggregated using several operators such as average, max, min, sum, and count. All of these measures and dimensions are typically prespecified in a database schema. Database administrators preaggregate views to accelerate queries. The data cube supports the following operations: * • A _slice_ specifies that you are only interested in some attribute values of a given dimension. For example, one may want to focus on one specific product (see Figure 1(g)). Similarly, a _dice_ selects ranges of attribute values (see Figure 1(e)). * • A _roll-up_ aggregates the measures on coarser attribute values. For example, from the sales given for every store, a user may want to see the sales aggregated per country (see Figure 1(c)). A _drill-down_ is the reverse operation: from the sales per country, one may want to explore the sales per store in one country. The various specific multidimensional views in Figure 1 are called _cuboids_. ### 3.2 Tag-Cloud OLAP Formalism A Web 2.0 OLAP application should be supported by a flexible formalism that can adapt a wide range of data loaded by users. Processing time must be reasonable and batch processing should be avoided. Unlike in conventional data cubes, we do not expect that most dimensions have explicit hierarchies when they are loaded: instead, users can specify how the data is laid out (see Section 5). As a related issue, the dimensions are not orthogonal in general: there might be a “City” dimension as a well as “Climate Zone” dimension. It is up to the user to organize the cities per climate zone or per country. ###### Definition 1 (Tag) A tag is a term or phrase describing an object with corresponding non-negative weights determining its relative importance. Hence, a tag is made of a triplet (term, object, weight). As an example, a picture may have been attributed the tags “dog” (12 times) and “cat” (20 times). In a Business Intelligence context, a tag may describe the current state of a business. For example, the tags “USA” (16,000$) and “Canada” (8,000$) describe the sales of a given product by a given salesman. We can aggregate several attribute values, such as “Canada” and “March,” into a single term, such as “Canada–March.” A tag composed of $k$ attribute values is called a $k$-tag. Figure 1(b) shows a tag cloud representation of Table 2 using 3-tags. Each tag $T$ is represented visually using a font size, font color, background color, area or motif, depending on its measure values. ### 3.3 Tag-Cloud Operations In our system, users can upload data, select a data set, and define a schema by choosing dimensions (see Figure 2). Then, users can apply various operations on the data using a menu bar. On the one hand, OLAP operations such as slice, dice, roll-up and drill-down generate new tag clouds and new cuboids from existing cuboids. Figures 1(d), 1(f) and 1(h), show the results of a roll-up, a dice, and a slice as tag clouds. On the other hand, we can apply some operations on an existing tag cloud: sort by either the weights or the terms of tags, remove some tags, remove lesser weighted tags, and so on. We estimate that a tag cloud should not have more than 150 tags. Figure 2: User-driven schema design Tag-cloud layout has measurable benefits when trying to convey a general impression [Rivadeneira et al., 2007]. Hence, we wish to optimize the visual arrangement of tags. Chen et al. propose the computation of similarity measures between cuboids to help users explore data [Chen et al., 2000]: we apply this idea to define similarities between tags. First of all, users are asked to provide one or several dimensions they want to use to cluster the tags. Choosing the “Country” dimension would mean that the user wants the tags rearranged by countries so that “Montreal–April” and “Toronto–March” are nearby (see Figure 3). The clustering dimensions selected by the user together with the tag-cloud dimensions form a cuboid: in our example, we have the dimensions “Country,” “City,” and “Time.” Since a tag contains a set of attribute values, it has a corresponding _subcuboid_ defined by slicing the cuboid. Figure 3: Choosing similarity dimensions Several similarity measures can be applied between subcuboids: Jaccard, Euclidean distance, cosine similarity, Tanimoto similarity, Pearson correlation, Hamming distance, and so on. Which similarity measure is best depends on the application at hand, so advanced users should be given a choice. Commonly, similarity measures take up values in the interval $[-1,1]$. Similarity measures are expected to be reflexive ($f(\texttt{a},\texttt{a})=1$), symmetric ($f(\texttt{a},\texttt{b})=f(\texttt{b},\texttt{a})$) and transitive: if a is similar to b, and b is similar to c, then a is also similar to c. Recall that given two vectors $v$ and $w$, the cosine similarity measure is defined as $\cos(v,w)=\sum_{i}v_{i}w_{i}/\sqrt{\sum_{i}v_{i}^{2}\sum_{i}w_{i}^{2}}=v/|v|\cdot w/|w|$. The Tanimoto similarity is given by $\sum_{i}v_{i}w_{i}/(\sum_{i}v_{i}^{2}+\sum_{i}w_{i}^{2}-\sum_{i}v_{i}w_{i})$; it becomes the Jaccard similarity when the vectors have binary values. Both of these measures are reflexive, symmetric and transitive. Specifically, the cosine similarity is transitive by this inequality: $\cos(v,z)\geq\cos(w,z)-\sqrt{1-\cos(v,w)^{2}}$. To generalize the formulas from vectors to cuboids, it suffices to replace the single summation by one summation per dimension. Figure 4 shows an example of tag-cloud reordering to cluster similar tags. In this example, the “City–Product” tags were compared according to the “Country” dimension. The result is that the tags are clustered by countries. Figure 4: Tag-cloud reordering based on similarity ## 4 FAST COMPUTATION Because only a moderate number of tags can be displayed, the computation of tag clouds is a form of top-$k$ query: given any user-specified range of cells, we seek the top-$k$ cells having the largest measures. There is a little hope of answering such queries in near constant-time with respect to the number of facts without an index or a buffer. Indeed, finding all and only the elements with frequency exceeding a given frequency threshold [Cormode and Muthukrishnan, 2005] or merely finding the most frequent element [Alon et al., 1996] requires $\Omega(m)$ bits where $m$ is the number of distinct items. Various efficient techniques have been proposed for the related range max problem [Chazelle, 1988, Poon, 2003], but they do not necessarily generalize. Instead, for the range top-$k$ problem, we can partition sparse data cubes into customized data structures to speed up queries by an order of magnitude [Luo et al., 2001, Loh et al., 2002a, Loh et al., 2002b]. We can also answer range top-$k$ queries using RD-trees [Chung et al., 2007] or R-trees [Seokjin et al., 2005]. In tag clouds, precision is not required and accuracy is less important; only the most significant tags are typically needed. Further, if all tags have similar weights, then any subset of tag may form an acceptable tag cloud. A strategy to speed up top-$k$ queries is to transform them into comparatively easier iceberg queries [Carey and Kossmann, 1997]. For example, in computing the top-10 ($k=10$) best vendors, one could start by finding all vendors with a rating above 4/5. If there are at least 10 such vendors, then sorting this smaller list is enough. If not, one can restart the query, seeking vendors with a rating above 3/5. Given a histogram or selectivity estimates, we can reduce the number of expected iceberg queries [Donjerkovic and Ramakrishnan, 1999]. Unfortunately, this approach is not necessarily applicable to multidimensional data since even computing iceberg aggregates once for each query may be prohibitive. However, iceberg cuboids can still be put to good use. That is, one materializes the iceberg of a cuboid, small enough to fit in main memory, from which the tag clouds are computed. Intuitively, a cuboid representing the largest measures is likely to provide reasonable tag clouds. Users mostly notice tags with large font sizes [Rivadeneira et al., 2007]. A good approximation captures the tags having significantly larger weights. To determine whether a tag cloud has such significant tags, we can compute the _entropy_. ###### Definition 2 (Entropy of a tag cloud) Let $T\in\mathcal{T}$ be a tag from a tag cloud $\mathcal{T}$, then $\textrm{entropy}(\mathcal{T})=-\sum_{T\in\mathcal{T}}p(T)log(p(T))$ where $p(T)=\frac{\textrm{weight}(T)}{\sum_{x\in\mathcal{T}}\textrm{weight}(x)}$. The entropy quantifies the disparity of weights between tags. The lower the entropy, the more interesting the corresponding tag cloud is. Indeed, tag clouds with uniform tag weights have maximal entropy and are visually not very informative (see Figure 5). Figure 5: Example of non informative tag cloud We can measure the quality of a low-entropy tag cloud by measuring false positives and negatives: false positive happens when a tag has been falsely added to a tag cloud whereas a false negative occurs when a tag is missing. These measures of error assume that we limit the number of tags to a moderately small number. We use the following quality indexes; index values are in $[0,1]$ and a value of 0 is ideal; they are not applicable to high- entropy tag clouds. ###### Definition 3 Given approximate and exact tag clouds $A$ and $E$, the false-positive and false-negative indexes are $\frac{\max_{t\in A,t\not\in E} \textrm{weight}(t)}{\max_{t\in A} \textrm{weight}(t)}$ and $\frac{\max_{t\in E,t\not\in A} \textrm{weight}(t)}{\max_{t\in E} \textrm{weight}(t)}$. ## 5 TAG-CLOUD DRAWING While we can ensure some level of device-independent displays on the Web, by using images or plugins, text display in HTML may vary substantially from browser to another. There is no common set of font browsers are required to support, and Web standards do not dictate line-breaking algorithms or other typographical issues. It is not practical to simulate the browser on a server. Meanwhile, if we wish to remain accessible and to abide by open standards, producing HTML and ECMAScript is the favorite option. Given tag-cloud data, the tag-cloud drawing problem is to optimally display the tags, generally using HTML, so that some desirable properties are met, including the following: (1) the screen space usage is minimized; (2) when applicable, similar tags are clustered together. Typically, the width of the tag cloud is fixed, but its height can vary. For practical reasons, we do not wish for the server to send all of the data to the browser, including a possibly large number of similarity measures between tags. Hence, some of the tag-cloud drawing computations must be server-bound. There are two possible architectures. The first scenario is a browser-aware approach [Kaser and Lemire, 2007]: given the tag-cloud data provided by the server, the browser sends back to the server some display- specific data, such as the box dimensions of various tags using different font sizes. The server then sends back an optimized tag cloud. The second approach is browser-oblivious: the server optimizes the display of the tag cloud without any knowledge of the browser by passing simple display hints. The browser can then execute a final and inexpensive display optimization. While browser-oblivious optimization is necessarily limited, it has reduced latency and it is easily cacheable. Browser-oblivious optimization can take many forms. For example, we could send classes of tags and instruct the browser to display them on separate lines [Hassan-Montero and Herrero-Solana, 2006]. In our system, tags are sent to the browser as an ordered list, using the convention that successive tags are similar and should appear nearby. Given a similarity measure $w$ between tags, we want to minimize $\sum_{p,q}w(p,q)d(p,q)$ where $d(p,q)$ is a distance function between the two tags in the list and the sum is over all tags. Ideally, $d(p,q)$ should be the physical distance between the tags as they appear in the browser; we model this distance with the index distance: if tag $a$ appears at index $i$ in the list and tag $b$ appears at index $j$, their distance is the integer $|i-j|$. This optimization problem is an instance of the NP-complete minimum linear arrangement (MLA) problem: an optimal linear arrangement of a graph $G=(V,E)$, is a map $f$ from $V$ onto $\\{1,2,\ldots,N\\}$ minimizing $\sum_{u,v\in V}|f(u)-f(v)|$. ###### Proposition 1 The browser-oblivious tag-cloud optimization problem is NP-Complete. There is an O($\sqrt{\log n}\log\log n$)-approximation for the MLA problem [Feige and Lee, 2007] in some instances. However, for our generic purposes, the greedy Nearest Neighbor (NN) algorithm might suffice: insert any tag in an empty list, then repeatedly append a tag most similar to the latest tag in the list, until all tags have been inserted. It runs in O($n^{2}$) time where $n$ is the number of tags. Another heuristic for the MLA problem is the pairwise exchange Monte Carlo (PWMC) method [Bhasker and Sahni, 1987]: after applying NN, you repeatedly consider the exchange of two tags chosen at random, permuting them if it reduces the MLA cost. Another Monte Carlo (MC) heuristic begins with the application of NN [Johnson et al., 2004]: cut the list into two blocks at a random location, test if exchanging the two blocks reduces the MLA cost, if so proceed; repeat. Additional display hints can be inserted in this list. For example, if two tags must absolutely be very close to each other, a glued token could be inserted. Also, if two tags can be permuted freely in the list, then a permutable token could be inserted: the list could take the form of a PQ tree [Booth and Lueker, 1976]. ## 6 EXPERIMENTS Throughout these experiments, we used the Java version 1.6.0_02 from Sun Microsystems Inc. on an Apple MacPro machine with 2 Dual-Core Intel Xeon processors running at 2.66 GHz and 2 GiB of RAM. ### 6.1 Iceberg-Based Computation To validate the generation of tag clouds from icebergs, we have run tests over the US Income 2000 data set [Hettich and Bay, 2000] (42 dimensions and about $2\times 10^{5}$ facts) as well as a synthetic data set (18 dimensions and $2\times 10^{4}$ facts) provided by Swivel (http://www.swivel.com/data_sets/show/1002247). Figure 6 shows that while some tag-cloud computations require several minutes, iceberg-based computations can be much faster. Figure 6: Computing tag clouds from original data vs. icebergs: iceberg limit value set at 150 and tag-cloud size is 9 (US Income 2000). From each data set, we generated a 4-dimensional data cube. We used the COUNT function to aggregate data. Tag clouds were computed from each data cube using the iceberg approximation with different values of _limit_ : the number of facts retained. We also implemented exact computations using temporary tables. We specified different values for tag-cloud size, limiting the maximum number of tags. For each iceberg limit value and tag-cloud size, we computed the entropy of the tag cloud, the false-positive and false-negative indexes, and processing time for both of iceberg approximation and exact computation. We plotted in Figure 7 the false-positive and false-negative indexes as a function of the relative entropy (entropy/$\log(\textrm{tag-cloud size})$) using various iceberg limit values (150, 600, 1200, 4800, and 19600) and various tag-cloud sizes (50, 100, 150, and 200), for a total of 20 tag clouds per dimension. The Y axis is in a logarithmic scale. Points having their indexes equal to zero are not displayed. As discussed in Section 4, false- positive and false-negative indexes should be low when the entropy is low. We verify that for low-entropy values ($<\frac{3}{4}\log(\textrm{tag-cloud size})$), the indexes are always close to zero which indicates a good approximation. Meanwhile, small iceberg cuboids can be processed much faster. (a) Swivel (b) US Income 2000 Figure 7: False-negative and false-positive indexes (0 is best, 1 is worst), values under 0.0001 are not included (a) Displaying dimension “Givenname” and clustering by “State” (Swivel) (b) Displaying dimension “HHDFMX” and clustering by “ARACE” (US Income 2000) Figure 8: MLA costs for two examples: the PWMC heuristic was applied using 10, 100 and 1000 random exchanges. ### 6.2 Similarity Computation Using our two data sets, we tested the NN, PWMC, and MC heuristics using both the cosine and the Tanimoto similarity measures. From data cubes made of all available dimensions, we used all possible 1-tag clouds, using successively all other dimensions as clustering dimension for a total of $2\times(18\times 17+42\times 41)=4056$ layout optimizations. The iceberg limit value was set at 150. The MC heuristic never fared better than NN, even when considering a very large number of random block permutations: we rejected this heuristic as ineffective. However, as Figure 8 shows, the PWMC heuristic can sometimes significantly outperform NN when a large number (1000) of tag exchanges are considered, but it only outperforms NN by more than 20% in less than 5% of all layout optimizations. Meanwhile, PWMC can be several order of magnitudes slower than NN: NN is 10 times faster than PWMC with 100 exchanges and 70 times faster than PWMC with 1000 exchanges. Computing the similarity function over an iceberg cuboid was moderately expensive (0.07 s) for a small iceberg cuboid (limit set to 150 cells): the exact computation of the similarity function can dwarf the cost of the heuristics (NN and PWMC) over a moderately large data set. Informal tests suggest that NN computed over a small iceberg cuboid provides significant visual layouts. ## 7 CONCLUSION According to our experimental results, precomputing a single iceberg cuboid per data cube allows to generate adequate approximate tag clouds online. Combined with modern Web technologies such as AJAX and JSON, it provides a responsive application. However, we plan to make more precise the relationship between iceberg cubes, entropy, dimension sizes, and our quality indexes. Yet another approach to compute tag clouds quickly may be to use a bitmap index [O’Neil and Quass, 1997]. While we built a Web 2.0 with support for numerous collaborations features such as permalinks, tag-cloud embeddings with iframe elements, we still need to experiment with live users. Our approach to multidimensional tag clouds has been to rely on $k$-tags. However, this approach might not be appropriate when a dimension has a linear flow such as time or latitude. A more appropriate approach is to allow the use of a slider [Russell, 2006] tying several tag clouds, each one corresponding to a given attribute value. ## ACKNOWLEDGMENTS The second author is supported by NSERC grant 261437 and FQRNT grant 112381. The third author is supported by NSERC grant OGP0009184 and FQRNT grant PR-119731. The authors wish to thank Owen Kaser from UNB for his contributions. ## REFERENCES * Alon et al., 1996 Alon, N., Matias, Y., and Szegedy, M. (1996). The space complexity of approximating the frequency moments. In STOC ’96, pages 20–29. * Ben Messaoud et al., 2006 Ben Messaoud, R., Boussaid, O., and Loudcher Rabaséda, S. (2006). Efficient multidimensional data representations based on multiple correspondence analysis. In KDD’06, pages 662–667. * Bhasker and Sahni, 1987 Bhasker, J. and Sahni, S. (1987). Optimal linear arrangement of circuit components. J. VLSI Comp. Syst., 2(1):87–109. * Body et al., 2002 Body, M., Miquel, M., Bédard, Y., and Tchounikine, A. (2002). A multidimensional and multiversion structure for OLAP applications. In DOLAP ’02, pages 1–6. * Booth and Lueker, 1976 Booth, K. S. and Lueker, G. S. (1976). Testing for the consecutive ones property, interval graphs, and graph planarity using PQ-tree algorithms. Journal of Computer and System Sciences, 13:335–379. * Butler, 2007 Butler, D. (2007). Data sharing: the next generation. Nature, 446(7131):1–10. * Carey and Kossmann, 1997 Carey, M. J. and Kossmann, D. (1997). On saying “enough already!” in SQL. In SIGMOD’97, pages 219–230. * Chazelle, 1988 Chazelle, B. (1988). A functional approach to data structures and its use in multidimensional searching. SIAM J. Comput., 17(3):427–462. * Chen et al., 2000 Chen, Q., Dayal, U., and Hsu, M. (2000). OLAP-based data mining for business intelligence applications in telecommunications and e-commerce. In DNIS ’00, pages 1–19. * Chung et al., 2007 Chung, Y., Yang, W., and Kim, M. (2007). An efficient, robust method for processing of partial top-k/bottom-k queries using the RD-tree in OLAP. Decision Support Systems, 43(2):313–321. * Codd, 1993 Codd, E. (1993). Providing OLAP (on-line analytical processing) to user-analysis: an IT mandate. Technical report, E.F. Codd and Associates. * Cormode and Muthukrishnan, 2005 Cormode, G. and Muthukrishnan, S. (2005). What’s hot and what’s not: tracking most frequent items dynamically. ACM Trans. Database Syst., 30(1):249–278. * Donjerkovic and Ramakrishnan, 1999 Donjerkovic, D. and Ramakrishnan, R. (1999). Probabilistic optimization of top n queries. In VLDB’99, pages 411–422. * Feige and Lee, 2007 Feige, U. and Lee, J. R. (2007). An improved approximation ratio for the minimum linear arrangement problem. Inf. Process. Lett., 101(1):26–29. * Gray et al., 1996 Gray, J., Bosworth, A., Layman, A., and Pirahesh, H. (1996). Data cube: A relational aggregation operator generalizing group-by, cross-tab, and sub-total. In ICDE ’96, pages 152–159. * Green et al., 2007 Green, T. J., Karvounarakis, G., Taylor, N. E., Biton, O., Ives, Z. G., and Tannen, V. (2007). ORCHESTRA: facilitating collaborative data sharing. In SIGMOD ’07, pages 1131–1133, New York, NY, USA. ACM. * Hassan-Montero and Herrero-Solana, 2006 Hassan-Montero, Y. and Herrero-Solana, V. (2006). Improving tag-clouds as visual information retrieval interfaces. In InSciT’06. * Havenstein, 2003 Havenstein, H. (2003). BI vendors seek to tap end-user power: New class of tools built to reap user knowledge for customizing analytic applications. InfoWorld, 22:20–21. * Heer et al., 2007 Heer, J., Viégas, F. B., and Wattenberg, M. (2007). Voyagers and voyeurs: supporting asynchronous collaborative information visualization. In CHI ’07, pages 1029–1038. * Hettich and Bay, 2000 Hettich, S. and Bay, S. D. (2000). The UCI KDD archive. http://kdd.ics.uci.edu. [Online; accessed 21/12/2007]. * IBM, 2007 IBM (2007). Many Eyes. http://services.alphaworks.ibm.com/manyeyes/. [Online; accessed 7-6-2007]. * Jaffe et al., 2006 Jaffe, A., Naaman, M., Tassa, T., and Davis, M. (2006). Generating summaries and visualization for large collections of geo-referenced photographs. In MIR ’06, pages 89–98. * Johnson et al., 2004 Johnson, D., Krishnan, S., Chhugani, J., Kumar, S., and Venkatasubramanian, S. (2004). Compressing large boolean matrices using reordering techniques. In VLDB’04, pages 13–23. * Kaser and Lemire, 2007 Kaser, O. and Lemire, D. (2007). Tag-cloud drawing: Algorithms for cloud visualization. In WWW 2007 – Tagging and Metadata for Social Information Organization. * Loh et al., 2002a Loh, Z., Ling, T., Ang, C., and Lee, S. (2002a). Adaptive method for range top-k queries in OLAP data cubes. In DEXA’02, pages 648–657. * Loh et al., 2002b Loh, Z. X., Ling, T. W., Ang, C. H., and Lee, S. Y. (2002b). Analysis of pre-computed partition top method for range top-k queries in OLAP data cubes. In CIKM’02, pages 60–67. * Luo et al., 2001 Luo, Z., Ling, T., Ang, C., Lee, S., and Cui, B. (2001). Range top/bottom k queries in OLAP sparse data cubes. In DEXA’01, pages 678–687. * Maniatis et al., 2005 Maniatis, A., Vassiliadis, P., Skiadopoulos, S., Vassiliou, Y., Mavrogonatos, G., and Michalarias, I. (2005). A presentation model & non-traditional visualization for OLAP. International Journal of Data Warehousing and Mining, 1:1–36. * Millen et al., 2006 Millen, D. R., Feinberg, J., and Kerr, B. (2006). Dogear: Social bookmarking in the enterprise. In CHI ’06, pages 111–120. * Morzy and Wrembel, 2004 Morzy, T. and Wrembel, R. (2004). On querying versions of multiversion data warehouse. In DOLAP ’04, pages 92–101. * O’Neil and Quass, 1997 O’Neil, P. and Quass, D. (1997). Improved query performance with variant indexes. In SIGMOD ’97, pages 38–49. * Poon, 2003 Poon, C. (2003). Dynamic orthogonal range queries in OLAP. Theoretical Computer Science, 296(3):487–510. * Rivadeneira et al., 2007 Rivadeneira, A. W., Gruen, D. M., Muller, M. J., and Millen, D. R. (2007). Getting our head in the clouds: toward evaluation studies of tagclouds. In CHI’07, pages 995–998. * Russell, 2006 Russell, T. (2006). cloudalicious: folksonomy over time. In JCDL’06, pages 364–364. * Seokjin et al., 2005 Seokjin, H., Moon, B., and Sukho, L. (2005). Efficient execution of range top-k queries in aggregate r-trees. IEICE – Transactions on Information and Systems, E88-D(11):2544–2554. * Swivel, Inc, 2007 Swivel, Inc (2007). Swivel. http://www.swivel.com. [Online; accessed 7-6-2007]. * Taylor and Ives, 2006 Taylor, N. E. and Ives, Z. G. (2006). Reconciling while tolerating disagreement in collaborative data sharing. In SIGMOD ’06, pages 13–24, New York, NY, USA. ACM. * Techapichetvanich and Datta, 2005 Techapichetvanich, K. and Datta, A. (2005). Interactive visualization for OLAP. In ICCSA ’05, pages 206–214. * Wattenberg and Kriss, 2006 Wattenberg, M. and Kriss, J. (2006). Designing for social data analysis. IEEE Transactions on Visualization and Computer Graphics, 12(4):549–557. * Wu et al., 2007 Wu, P., Sismanis, Y., and Reinwald, B. (2007). Towards keyword-driven analytical processing. In SIGMOD ’07, pages 617–628.

arxiv-papers

2024-09-04T02:48:53.226094

0710.3794

# Curve complexes are rigid Kasra Rafi rafi@math.uchicago.edu and Saul Schleimer s.schleimer@warwick.ac.uk ###### Abstract. Any quasi-isometry of the complex of curves is bounded distance from a simplicial automorphism. As a consequence, the quasi-isometry type of the curve complex determines the homeomorphism type of the surface. This work is in the public domain. ## 1\. Introduction The curve complex of a surface was introduced into the study of Teichmüller space by Harvey [11] as an analogue of the Tits building of a symmetric space. Since then the curve complex has played a key role in many areas of geometric topology such as the classification of infinite volume hyperbolic three- manifolds, the study of the cohomology of mapping class groups, the geometry of Teichmüller space, and the combinatorics of Heegaard splittings. Our motivation is the work of Masur and Minsky [22, 23] which focuses on the coarse geometric structure of the curve complex, the mapping class group, and other combinatorial moduli spaces. It is a sign of the richness of low- dimensional topology that the geometric structure of such objects is not well understood. Suppose that $S=S_{{\sf g},{\sf b}}$ is an orientable, connected, compact surface with genus ${\sf g}$ and ${\sf b}$ boundary components. Define the complexity of $S$ to be $\xi(S)=3{\sf g}-3+{\sf b}$. Let $\mathcal{C}(S)$ be the curve complex of $S$. Our main theorem is: Suppose that $\xi(S)\geq 2$. Then every quasi-isometry of $\mathcal{C}(S)$ is bounded distance from a simplicial automorphism of $\mathcal{C}(S)$. Before discussing the sharpness of Theorem 7.1 recall the definition of $\operatorname{QI}(\mathcal{X})$. This is the group of quasi-isometries of a geodesic metric space $\mathcal{X}$, modulo an equivalence relation; quasi- isometries $f$ and $g$ are equivalent if and only if there is a constant ${\sf d}$ so that for every $x\in\mathcal{X}$ we have $d_{\mathcal{X}}(f(x),g(x))\leq{\sf d}$. Define $\operatorname{Aut}(\mathcal{C}(S))$ to be the group of simplicial automorphisms of $\mathcal{C}(S)$; notice that these are always isometries. From Theorem 7.1 deduce: ###### Corollary 1.1. Suppose that $\xi(S)\geq 2$. Then the natural map $\operatorname{Aut}(\mathcal{C}(S))\to\operatorname{QI}(\mathcal{C}(S))$ is an isomorphism. ###### Proof. The map is always an injection. To see this recall Ivanov’s Theorem [13, 19, 21]: if $\xi(S)\geq 2$ then every $f\in\operatorname{Aut}(\mathcal{C}(S))$ is induced by some homeomorphism of $S$, called $f_{S}$. (When $\xi(S)=2$ every automorphism of $\mathcal{C}(S)$ is induced by some homeomorphism of $S_{0,5}$; see [21].) Suppose that $f\in\operatorname{Aut}(\mathcal{C}(S))$ is not the identity element. Then there is some curve $a$ with $f_{S}(a)$ not isotopic to $a$. Consider the action of $f_{S}$ on $\mathcal{PML}(S)$. There is a small neighborhood of $a$ in $\mathcal{PML}(S)$, say $U$, so that $f_{S}(U)\cap U=\emptyset$. Since ending laminations are dense, $f_{S}$ moves some ending lamination of $S$. By Klarreich’s Theorem (see Theorem 2.3 below), we deduce that $f$ moves some point of $\partial\mathcal{C}(S)$. Finally, any isometry of a Gromov hyperbolic space moving a point of the boundary is nontrivial in the quasi-isometry group. On the other hand, Theorem 7.1 implies that the homomorphism $\operatorname{Aut}(\mathcal{C}(S))\to\operatorname{QI}(\mathcal{C}(S))$ is a surjection. ∎ Note that Theorem 7.1 and Corollary 1.1 are sharp. If $S$ is a sphere, disk, or pair of pants then the complex of curves is empty. If $S$ is an annulus then, following [23], the complex $\mathcal{C}(S)$ is quasi-isometric to $\mathbb{Z}$ (see below) and the conclusion of Theorem 7.1 does not hold. If $S$ is a torus, four-holed sphere or once-holed torus then the curve complex is a copy of the Farey graph. Thus $\mathcal{C}(S)$ is quasi-isometric to $T_{\infty}$, the countably infinite valence tree [4]. Hence $\operatorname{QI}(\mathcal{C}(S))$ is uncountable while $\operatorname{Aut}(\mathcal{C}(S))=\operatorname{PGL}(2,\mathbb{Z})$ is countable. Thus, for these surfaces the conclusion of Theorem 7.1 does not hold. We now give an application of Corollary 1.1: ###### Theorem 1.2. Suppose that $S$ and $\Sigma$ are surfaces with $\mathcal{C}(S)$ quasi- isometric to $\mathcal{C}(\Sigma)$. Then either * • $S$ and $\Sigma$ are homeomorphic, * • $\\{S,\Sigma\\}=\\{S_{0,6},S_{2}\\}$, * • $\\{S,\Sigma\\}=\\{S_{0,5},S_{1,2}\\}$, * • $\\{S,\Sigma\\}\subset\\{S_{0,4},S_{1},S_{1,1}\\}$, or * • $\\{S,\Sigma\\}\subset\\{\mathbb{S},\mathbb{D},S_{0,3}\\}$. Thus, two curve complexes are quasi-isometric if and only if they are isomorphic. To prove Theorem 1.2 we require Theorem 7.1 and the following folk theorem: Suppose that $S$ and $\Sigma$ are compact, connected, orientable surfaces with $\mathcal{MCG}(S)$ isomorphic to $\mathcal{MCG}(\Sigma)$. Then either * • $S$ and $\Sigma$ are homeomorphic, * • $\\{S,\Sigma\\}=\\{S_{1},S_{1,1}\\}$, or * • $\\{S,\Sigma\\}=\\{\mathbb{S},\mathbb{D}\\}$. Apparently no proof of Theorem A.1 appears in the literature. In Appendix A we discuss previous work (Remark A.2) and, for completeness, give a proof of Theorem A.1. ###### Proof of Theorem 1.2. For brevity, we restrict to the case where $\xi(S)$ and $\xi(\Sigma)$ are at least four. By Corollary 1.1 the automorphism groups of $\mathcal{C}(S)$ and $\mathcal{C}(\Sigma)$ are isomorphic. Ivanov’s Theorem [13, 19, 21] tells us that the simplicial automorphism group is isomorphic to the mapping class group. Finally, it follows from Theorem A.1 that such surfaces are characterized, up to homeomorphism, by their mapping class groups. ∎ ### Outline of the paper The proof of Theorem 7.1 has the following ingredients. A pair of ending laminations is cobounded if the projections of this pair to any strict subsurface of $S$ are uniformly close to each other in the complex of curves of that subsurface (see Definition 2.8). Suppose that $\xi(S)\geq 2$ and suppose that $\phi\colon\mathcal{C}(S)\to\mathcal{C}(\Sigma)$ is a quasi-isometric embedding. Then the induced map on boundaries preserves the coboundedness of ending laminations. Theorem 5.2 is important in its own right and may have other applications. For example, it may be helpful in classifying quasi-isometric embeddings of one curve complex into another. (See [26].) The proof of Theorem 5.2 uses the following theorem in an essential way: ###### Theorem 1.3 (Gabai [8]). Suppose that $\xi(S)\geq 2$. Then $\partial\mathcal{C}(S)$ is connected. ∎ ###### Remark 1.4. Leininger and the second author [20] previously gave a quite different proof of Theorem 1.3 in the cases where $S$ has genus at least four, or where $S$ has genus at least two and non-empty boundary. Note that Gabai’s Theorem is sharp; $\partial\mathcal{C}(S)$ is not connected when $S$ is an annulus, torus, once-holed torus or four-holed sphere. Let $\mathcal{M}(S)$ denote the marking complex of the surface $S$. We show that a marking on $S$ can be coarsely described by a pair of cobounded ending laminations and a curve in $\mathcal{C}(S)$. Theorem 5.2 implies that a quasi- isometric embedding of $\mathcal{C}(S)$ into $\mathcal{C}(\Sigma)$ induces a map from $\mathcal{M}(S)$ to $\mathcal{M}(\Sigma)$. Suppose that $\xi(S)\geq 2$ and $\phi\colon\mathcal{C}(S)\to\mathcal{C}(\Sigma)$ is a ${\sf q}$–quasi- isometric embedding. Then $\phi$ induces a coarse Lipschitz map $\Phi\colon\mathcal{M}(S)\to\mathcal{M}(\Sigma)$ so that the diagram $\begin{CD}\mathcal{M}(S)@>{\Phi}>{}>\mathcal{M}(\Sigma)\\\ @V{}V{p}V@V{}V{\pi}V\\\ \mathcal{C}(S)@>{\phi}>{}>\mathcal{C}(\Sigma)\end{CD}$ commutes up to an additive error. Furthermore, if $\phi$ is a quasi-isometry then so is $\Phi$. As the final step of the proof of Theorem 7.1 we turn to a recent theorem of Behrstock, Kleiner, Minsky and Mosher [1]. See also [10]. ###### Theorem 1.5. Suppose that $\xi(S)\geq 2$ and $S\not=S_{1,2}$. Then every quasi-isometry of $\mathcal{M}(S)$ is bounded distance from the action of a homeomorphism of $S$. ∎ So, if $f\colon\mathcal{C}(S)\to\mathcal{C}(S)$ is a quasi-isometry then Theorem 6.1 gives a quasi-isometry $F$ of marking complexes. This and Theorem 1.5 imply Theorem 7.1 except when $S=S_{1,2}$. But the curve complexes $\mathcal{C}(S_{0,5})$ and $\mathcal{C}(S_{1,2})$ are identical. Therefore to prove Theorem 7.1 for $\mathcal{C}(S_{1,2})$ it suffices to prove it for $\mathcal{C}(S_{0,5})$. ### Acknowledgements This paper was sparked by a question of Slava Matveyev. We thank Dan Margalit for useful conversations. ## 2\. Background ### Hyperbolic spaces A geodesic metric space $\mathcal{X}$ is Gromov hyperbolic if there is a hyperbolicity constant $\delta\geq 0$ so that every triangle is $\delta$–slim: for every triple of vertices $x,y,z\in\mathcal{X}$ and every triple of geodesics $[x,y],[y,z],[z,x]$ the $\delta$–neighborhood of $[x,y]\cup[y,z]$ contains $[z,x]$. Suppose that $(\mathcal{X},d_{\mathcal{X}})$ and $(\mathcal{Y},d_{\mathcal{Y}})$ are geodesic metric spaces and $f\colon\mathcal{X}\to\mathcal{Y}$ is a map. Then $f$ is ${\sf q}$–coarsely Lipschitz if for all $x,y\in\mathcal{X}$ we have $d_{\mathcal{Y}}(x^{\prime},y^{\prime})\leq{\sf q}\,d_{\mathcal{X}}(x,y)+{\sf q}$ where $x^{\prime}=f(x)$ and $y^{\prime}=f(y)$. If, in addition, $d_{\mathcal{X}}(x,y)\leq{\sf q}\,d_{\mathcal{Y}}(x^{\prime},y^{\prime})+{\sf q}$ then $f$ is a ${\sf q}$–quasi-isometric embedding. Two maps $f,g\colon\mathcal{X}\to\mathcal{Y}$ are ${\sf d}$–close if for all $x\in\mathcal{X}$ we have $d_{\mathcal{Y}}(f(x),g(x))\leq{\sf d}.$ If $f\colon\mathcal{X}\to\mathcal{Y}$ and $g\colon\mathcal{Y}\to\mathcal{X}$ are ${\sf q}$–coarsely Lipschitz and also $f\circ g$ and $g\circ f$ are ${\sf q}$–close to identity maps then $f$ and $g$ are ${\sf q}$–quasi-isometries. A quasi-isometric embedding of an interval $[s,t]\subset\mathbb{Z}$, with the usual metric, is called a quasi-geodesic. In hyperbolic spaces quasi-geodesics are stable: ###### Lemma 2.1. Suppose that $(\mathcal{X},d_{\mathcal{X}})$ has hyperbolicity constant $\delta$ and that $f\colon[s,t]\to\mathcal{X}$ is a ${\sf q}$–quasi-geodesic. Then there is a constant ${\sf M}_{\mathcal{X}}={\sf M}(\delta,{\sf q})$ so that for any $[p,q]\subset[s,t]$ the image $f([p,q])$ and any geodesic $[f(p),f(q)]$ have Hausdorff distance at most ${\sf M}_{\mathcal{X}}$ in $\mathcal{X}$. ∎ See [6] for further background on hyperbolic spaces. ### Curve Complexes Let $S=S_{{\sf g},{\sf b}}$, as before. Define the vertex set of the curve complex, $\mathcal{C}(S)$, to be the set of simple closed curves in $S$ that are essential and non-peripheral, considered up to isotopy. When the complexity $\xi(S)$ is at least two, distinct vertices $a,b\in\mathcal{C}(S)$ are connected by an edge if they have disjoint representatives. When $\xi(S)=1$ vertices are connected by an edge if there are representatives with geometric intersection exactly one for the torus and once-holed torus or exactly two for the four-holed sphere. This gives the Farey graph. When $S$ is an annulus the vertices are essential embedded arcs, considered up to isotopy fixing the boundary pointwise. Vertices are connected by an edge if there are representatives with disjoint interiors. For any vertices $a,b\in\mathcal{C}(S)$ define the distance $d_{S}(a,b)$ to be the minimal number of edges appearing in an edge path between $a$ and $b$. ###### Theorem 2.2 (Masur-Minsky [22]). The complex of curves $\mathcal{C}(S)$ is Gromov hyperbolic. ∎ We use $\delta_{S}$ to denote the hyperbolicity constant of $\mathcal{C}(S)$. ### Boundary of the curve complex Let $\partial\mathcal{C}(S)$ be the Gromov boundary of $\mathcal{C}(S)$. This is the space of quasi-geodesic rays in $\mathcal{C}(S)$ modulo equivalence: two rays are equivalent if and only if their images have bounded Hausdorff distance. Recall that $\mathcal{PML}(S)$ is the projectivized space of measured laminations on $S$. A measured lamination $\ell$ is filling if every component $S{\smallsetminus}\ell$ is a disk or a boundary-parallel annulus. Take $\mathcal{FL}(S)\subset\mathcal{PML}(S)$ to be the set of filling laminations with the subspace topology. Define $\mathcal{EL}(S)$, the space of ending laminations, to be the quotient of $\mathcal{FL}(S)$ obtained by forgetting the measures. See [17] for an expansive discussion of laminations. ###### Theorem 2.3 (Klarreich [18]). There is a mapping class group equivariant homeomorphism between $\partial\mathcal{C}(S)$ and $\mathcal{EL}(S)$. ∎ We define $\overline{\mathcal{C}(S)}=\mathcal{C}(S)\cup\partial\mathcal{C}(S)$. ### Subsurface projection Suppose that $Z\subset S$ is an essential subsurface: $Z$ is embedded, every component of $\partial Z$ is essential in $S$, and $Z$ is not a boundary- parallel annulus nor a pair of pants. An essential subsurface $Z\subset S$ is strict if $Z$ is not homeomorphic to $S$. A lamination $b$ cuts a subsurface $Z$ if every isotopy representative of $b$ intersects $Z$. If $b$ does not cut $Z$ then $b$ misses $Z$. Suppose now that $a,b\in\overline{\mathcal{C}(S)}$ both cut a strict subsurface $Z$. Define the subsurface projection distance $d_{Z}(a,b)$ as follows: isotope $a$ with respect to $\partial Z$ to realize the geometric intersection number. Surger the arcs of $a\cap Z$ to obtain $\pi_{Z}(a)$, a finite set of vertices in $\mathcal{C}(Z)$. Notice that $\pi_{Z}(a)$ has uniformly bounded diameter in $\mathcal{C}(Z)$ independent of $a$, $Z$ or $S$. Define $d_{Z}(a,b)=\operatorname{diam}_{Z}\big{(}\pi_{Z}(a)\cup\pi_{Z}(b)\big{)}.$ We now recall the Lipschitz Projection Lemma [23, Lemma 2.3]: ###### Lemma 2.4 (Masur-Minsky). Suppose that $\\{a_{i}\\}_{i=0}^{N}\subset\mathcal{C}(S)$ is a path where every vertex cuts $Z\subset S$. Then $d_{Z}(a_{0},a_{N})\leq 2N$. ∎ For geodesics, the much stronger Bounded Geodesic Image Theorem holds [23, 25]: ###### Theorem 2.5. There is a constant ${\sf c_{0}}={\sf c_{0}}(S)$ with the following property. For any strict subsurface $Z$ and any points $a,b\in\overline{\mathcal{C}(S)}$, if every vertex of the geodesic $[a,b]$ cuts $Z$ then $d_{Z}(a,b)<{\sf c_{0}}$. ∎ ### Marking complex We now discuss the marking complex, following Masur and Minsky [23]. A complete clean marking $m$ is a pants decomposition $\operatorname{base}(m)$ of $S$ together with a transversal $t_{a}$ for each element $a\in\operatorname{base}(m)$. To define $t_{a}$, let $X_{a}$ be the non-pants component of $S{\smallsetminus}(\operatorname{base}(m){\smallsetminus}\\{a\\})$. Then any vertex of $\mathcal{C}(X_{a})$ not equal to $a$ and meeting $a$ minimally may serve as a transversal $t_{a}$. Notice that diameter of $m$ in $\mathcal{C}(S)$ is at most $2$. Masur and Minsky also define elementary moves on markings. The set of markings and these moves define the marking complex, $\mathcal{M}(S)$: a locally finite graph quasi-isometric to the mapping class group. The projection map $p\colon\mathcal{M}(S)\to\mathcal{C}(S)$, sending $m$ to any element of $\operatorname{base}(m)$, is coarsely mapping class group equivariant. We now record, from [23], the Elementary Move Projection Lemma: ###### Lemma 2.6. If $m$ and $m^{\prime}$ differ by an elementary move then for any essential subsurface $Z\subseteq S$, we have $d_{Z}(m,m^{\prime})\leq 4$. ∎ A converse follows from the distance estimate [23]. ###### Lemma 2.7. For every constant ${\sf c}$ there is a bound ${\sf e}={\sf e}({\sf c},S)$ with the following property. If $d_{Z}(m,m^{\prime})\leq{\sf c}$ for every essential subsurface $Z\subseteq S$ then $d_{\mathcal{M}}(m,m^{\prime})\leq{\sf e}$. ∎ ### 2.1. Tight geodesics The curve complex is locally infinite. Generally, there are infinitely many geodesics connecting a given pair of points in $\mathcal{C}(S)$. In [23] the notion of a tight geodesic is introduced. This is a technical hypothesis which provides a certain kind of local finiteness. Lemma 2.9 below is the only property of tight geodesics used in this paper. ###### Definition 2.8. A pair of curves, markings or laminations $a,b$ are ${\sf c}$–cobounded if $d_{Z}(a,b)\leq{\sf c}$ for all strict subsurfaces $Z\subset S$ cut by both $a$ and $b$. Minsky shows [25, Lemma 5.14] that if $a,b\in\overline{\mathcal{C}(S)}$ then there is a tight geodesic $[a,b]\subset\mathcal{C}(S)$ connecting them. All geodesics from here on are assumed to be tight. ###### Lemma 2.9 (Minsky). There is a constant ${\sf c_{1}}={\sf c_{1}}(S)$ with the following property. Suppose that $(a,b)$ is a ${\sf c}$–cobounded pair in $\overline{\mathcal{C}(S)}$ and $c\in[a,b]$ is a vertex of a tight geodesic. Then the pairs $(a,c)$ and $(c,b)$ are $({\sf c}+{\sf c_{1}})$–cobounded. ∎ ## 3\. Extension Lemmas We now examine how points of $\mathcal{C}(S)$ can be connected to infinity. ###### Lemma 3.1 (Completion). There is a constant ${\sf c_{2}}={\sf c_{2}}(S)$ with the following property. Suppose that $b\in\mathcal{C}(S)$ and $\ell\in\overline{\mathcal{C}(S)}$. Suppose that the pair $(b,\ell)$ is ${\sf c}$–cobounded. Then there is a marking $m$ so that $b\in\operatorname{base}(m)$ and $(m,\ell)$ are $({\sf c}+{\sf c_{2}})$–cobounded. ∎ The existence of the marking $m$ follows from the construction preceding [3, Lemma 6.1]. ###### Lemma 3.2 (Extension past a point). Suppose that $a,z\in\mathcal{C}(S)$ with $a\neq z$. Then there is a point $\ell\in\partial\mathcal{C}(S)$ so that the vertex $a$ lies in the one- neighborhood of $[z,\ell]$. ###### Proof. Let $k\in\partial\mathcal{C}(S)$ be any lamination. Let $Y$ be a component of $S{\smallsetminus}a$ that meets $z$. Pick any mapping class $\phi$ with support in $Y$ and with translation distance at least $(2{\sf c_{0}}+2)$ in $\mathcal{C}(Y)$. We have either $d_{Y}(z,k)\geq{\sf c_{0}}\quad\text{or}\quad d_{Y}(z,\phi(k))\geq{\sf c_{0}}.$ By Theorem 2.5, at least one of the geodesics $[z,k]$ or $[z,\phi(k)]$ passes through the one-neighborhood of $a$. ∎ ###### Proposition 3.3 (Extension past a marking). There is a constant ${\sf c_{3}}={\sf c_{3}}(S)$ such that if $m$ is a marking on $S$, then there are laminations $k$ and $\ell$ such that the pairs $(k,\ell)$, $(k,m)$ and $(m,\ell)$ are ${\sf c_{3}}$–cobounded and $[k,\ell]$ passes through the one-neighborhood of $m$. ###### Proof. There are only finitely many markings up to the action of the mapping class group. Fix a class of markings and pick a representative $m$. We will find a pseudo-Anosov map with stable and unstable laminations $k$ and $\ell$ such that $[k,\ell]$ passes through the one-neighborhood of $m$. This suffices to prove the proposition: there is a constant ${\sf c_{3}}(m)$ large enough so that the pairs $(k,\ell)$, $(k,m)$ and $(m,\ell)$ are ${\sf c_{3}}(m)$–cobounded. The same constant works for every marking in the orbit $\mathcal{MCG}(S)\cdot m$, by conjugation. We can now take ${\sf c_{3}}$ to be the maximum of the ${\sf c_{3}}(m)$ as $m$ ranges over the finitely many points of the quotient $\mathcal{M}(S)/\mathcal{MCG}(S)$. So choose any pseudo-Anosov map $\phi^{\prime}$ with stable and unstable laminations $k^{\prime}$ and $\ell^{\prime}$. Choose any point $b^{\prime}\in[k^{\prime},\ell^{\prime}]$. We may conjugate $\phi^{\prime}$ to $\phi$, sending $(k^{\prime},\ell^{\prime},b^{\prime})$ to $(k,\ell,b)$, so that $b$ is disjoint from some curve $a\in\operatorname{base}(m)$. This finishes the proof. ∎ ## 4\. The shell is connected Let $\mathcal{B}(z,{\sf r})$ be the ball of radius ${\sf r}$ about $z\in\mathcal{C}(S)$. The difference of concentric balls is called a shell. ###### Proposition 4.1. Suppose that $\xi(S)\geq 2$ and ${\sf d}\geq\max\\{\delta_{S},1\\}$. Then, for any ${\sf r}\geq 0$, the shell $\mathcal{B}(z,{\sf r}+2{\sf d}){\smallsetminus}\mathcal{B}(z,{\sf r}-1)$ is connected. Below we will only need the corollary that $\mathcal{C}(S){\smallsetminus}\mathcal{B}(z,{\sf r}-1)$ is connected. However, the shell has other interesting geometric properties. We hope to return to this subject in a future paper. One difficulty in the proof of Proposition 4.1 lies in pushing points of the inner boundary into the interior of the shell. To deal with this we use the fact that $\mathcal{C}(S)$ has no dead ends. ###### Lemma 4.2. Fix vertices $z,a\in\mathcal{C}(S)$. Suppose $d_{S}(z,a)={\sf r}$. Then there is a vertex $a^{\prime}\in\mathcal{C}(S)$ with $d_{S}(a,a^{\prime})\leq 2$ and $d_{S}(z,a^{\prime})={\sf r}+1$. ∎ Note that this implies that any geodesic $[a,a^{\prime}]$ lies outside of $\mathcal{B}(z,{\sf r}-1)$. For a proof of Lemma 4.2, see Proposition 3.1 of [27]. ###### Proof of Proposition 4.1. For any $z\in\mathcal{C}(S)$ and any geodesic or geodesic segement $[a,b]\subset\overline{\mathcal{C}(S)}$ define $d_{S}(z,[a,b])=\min\\{d_{S}(z,c)\mathbin{\mid}c\in[a,b]\\}$. Define a product on $\overline{\mathcal{C}(S)}$ by: $\langle a,b\rangle_{z}=\inf\big{\\{}d_{S}(z,[a,b])\big{\\}}$ where the infimum ranges over all geodesics $[a,b]$. For every $k\in\partial\mathcal{C}(S)$ let $U(k)=\\{\ell\in\partial\mathcal{C}(S)\mathbin{\mid}\langle k,\ell\rangle_{z}>{\sf r}+2{\sf d}\\}.$ The set $U(k)$ is a neighborhood of $k$ by the definition of the topology on the boundary [9]. Notice that if $\ell\in U(k)$ then $k\in U(\ell)$. Consider the set $V(k)$ of all $\ell\in\partial\mathcal{C}(S)$ so that there is a finite sequence $k=k_{0},k_{1},\ldots,k_{N}=\ell$ with $k_{i+1}\in U(k_{i})$ for all $i$. Now, if $\ell\in V(k)$ then $U(\ell)\subset V(k)$; thus $V(k)$ is open. If $\ell$ is a limit point of $V(k)$ then there is a sequence $\ell_{i}\in V(k)$ entering every neighborhood of $\ell$. So there is some $i$ where $\ell_{i}\in U(\ell)$. Thus $\ell\in U(\ell_{i})\subset V(k)$ and we find that $V(k)$ is closed. Finally, as $\partial\mathcal{C}(S)$ is connected (Theorem 1.3), $V(k)=\partial\mathcal{C}(S)$. Let $a^{\prime},b^{\prime}$ be any vertices in the shell $\mathcal{B}(z,{\sf r}+2{\sf d}){\smallsetminus}\mathcal{B}(z,{\sf r}-1)$. We connect $a^{\prime}$, via a path in the shell, to a vertex $a$ so that $d_{S}(z,a)={\sf r}+{\sf d}$. This is always possible: points far from $z$ may be pushed inward along geodesics and points near $z$ may be pushed outward by Lemma 4.2. Similarly connect $b^{\prime}$ to $b$. By Lemma 3.2 there are points $k,\ell\in\partial\mathcal{C}(S)$ so that there are geodesic rays $[z,k]$ and $[z,\ell]$ within distance one of $a$ and $b$ respectively. Connect $k$ to $\ell$ by a chain of points $\\{k_{i}\\}$ in $V(k)$, as above. Define $a_{i}\in[z,k_{i}]$ so that $d_{S}(z,a_{i})={\sf r}+{\sf d}$. Connect $a$ to $a_{0}$ via a path of length at most $2$. Notice that $d_{S}(a_{i},[k_{i},k_{i+1}])>{\sf d}\geq\delta$. As triangles are slim, the vertex $a_{i}$ is $\delta$–close to $[z,k_{i+1}]$. Thus $a_{i}$ and $a_{i+1}$ may be connected inside of the shell via a path of length at most $2\delta$. ∎ ## 5\. Image of a cobounded geodesic is cobounded We begin with a simple lemma: ###### Lemma 5.1. For every ${\sf c}$ and ${\sf r}$ there is a constant ${\sf K}$ with the following property. Let $[a,b]\subset\mathcal{C}(S)$ be a geodesic segment of length $2{\sf r}$ with $(a,b)$ being ${\sf c}$–cobounded. Let $z$ be the midpoint. Then there is a path $P$ of length at most ${\sf K}$ connecting $a$ to $b$ outside of $\mathcal{B}(z,{\sf r}-1)$. ###### Proof. There are only finitely many such triples $(a,z,b)$, up the action of the mapping class group. (This is because there are only finitely many hierarchies having total length less than a given upper bound; see [23]). The conclusion now follows from the connectedness of the shell (Proposition 4.1). ∎ Note that any quasi-isometric embedding $\phi\colon\mathcal{C}(S)\to\mathcal{C}(\Sigma)$ extends to a one-to-one continuous map from $\partial\mathcal{C}(S)$ to $\partial\mathcal{C}(\Sigma)$. ###### Theorem 5.2. There is a function ${\sf H}\colon\mathbb{N}\to\mathbb{N}$, depending only on ${\sf q}$ and the topology of $S$ and $\Sigma$, with the following property. Suppose $(k,\ell)$ is a pair of ${\sf c}$–cobounded laminations and $\phi\colon\mathcal{C}(S)\to\mathcal{C}(\Sigma)$ is a ${\sf q}$–quasi- isometric embedding. Then $\kappa=\phi(k)$ and $\lambda=\phi(\ell)$ are ${\sf H}({\sf c})$–cobounded ###### Proof. For every strict subsurface $\Omega\subset\Sigma$ we must bound $d_{\Omega}(\kappa,\lambda)$ from above. Now, if $d_{\Sigma}(\partial\Omega,[\kappa,\lambda])\geq 2$ then by the Bounded Geodesic Image Theorem (2.5) we find $d_{\Omega}(\kappa,\lambda)\leq{\sf c}_{0}={\sf c}_{0}(\Sigma)$ and we are done. 2pt $\mathcal{C}(S)$ [Bl] at 144.5 149.6 $\mathcal{C}(\Sigma)$ [Bl] at 349.8 149.6 $k$ [Br] at 35.1 154.6 $\kappa$ [Br] at 237.4 153.6 $\ell$ [t] at 85.8 0 $\lambda$ [t] at 308.7 1 $a$ [Br] at 69.8 94.4 $\alpha$ [r] at 264.5 109.4 $b$ [tr] at 84.8 39.6 $\beta$ [r] at 292.6 43.7 $z$ [r] at 78.8 68.8 $\phi(z)$ [t] at 255 62.2 $\partial\Omega$ [l] at 296.6 81.3 $\begin{array}[]{c}\includegraphics[scale=.8]{cobounded2}\end{array}$ Figure 1. Points outside of an ${\sf r}$–ball about $z$ are sent by $\phi$ outside of an $({\sf q}+{\sf M}_{\Sigma}+2)$–ball about $\partial\Omega$. Now suppose $d_{\Sigma}(\partial\Omega,[\kappa,\lambda])\leq 1$. Note that $[\kappa,\lambda]$ lies in the ${\sf M}$–neighborhood of $\phi([k,\ell])$, where ${\sf M}={\sf M}_{\Sigma}$ is provided by Lemma 2.1. Choose a vertex $z\in[k,\ell]$ so that $d_{\Sigma}(\phi(z),\partial\Omega)\leq{\sf M}+1$. Set ${\sf r}={\sf q}({\sf q}+2{\sf M}+3)+{\sf q}$. Thus $\displaystyle d_{S}(y,z)\geq{\sf r}$ $\displaystyle\implies d_{\Sigma}(\phi(y),\phi(z))\geq{\sf q}+2{\sf M}+3$ $\displaystyle\implies d_{\Sigma}(\phi(y),\partial\Omega)\geq{\sf q}+{\sf M}+2.$ Let $a$ and $b$ be the intersections of $[k,\ell]$ with $\partial\mathcal{B}(z,{\sf r})$, chosen so that $[k,a]$ and $[b,\ell]$ meet $\mathcal{B}(z,{\sf r})$ at the vertices $a$ and $b$ only. Connect $a$ to $b$ via a path $P$ of length ${\sf K}$, outside of $\mathcal{B}(z,{\sf r}-1)$, as provided by Lemma 5.1. Let $\alpha=\phi(a)$ and $\beta=\phi(b)$. Now, any consecutive vertices of $P$ are mapped by $\phi$ to vertices of $\mathcal{C}(\Sigma)$ that are at distance at most $2{\sf q}$. Connecting these by geodesic segments gives a path $\Pi$ from $\alpha$ to $\beta$. Note that $\Pi$ has length at most $2{\sf q}{\sf K}$. Since every vertex of $\phi(P)$ is $({\sf q}+{\sf M}+2)$–far from $\partial\Omega$ every vertex of $\Pi$ is $({\sf M}+2)$–far from $\partial\Omega$. So every vertex of $\Pi$ cuts $\Omega$. It follows that $d_{\Omega}(\alpha,\beta)\leq 4{\sf q}{\sf K}$, by Lemma 2.4. All that remains is to bound $d_{\Omega}(\kappa,\alpha)$ and $d_{\Omega}(\beta,\lambda)$. It suffices, by the Bounded Geodesic Image Theorem, to show that every vertex of $[\kappa,\alpha]$ cuts $\Omega$. The same will hold for $[\beta,\lambda]$. Every vertex of $[\kappa,\alpha]$ is ${\sf M}$–close to a vertex of $\phi([k,a])$. But each of these is $({\sf q}+{\sf M}+2)$–far from $\partial\Omega$. This completes the proof. ∎ ## 6\. The induced map on markings In this section, given a quasi-isometric embedding of one curve complex into another we construct a coarsely Lipschitz map between the associated marking complexes. Let $\mathcal{M}(S)$ and $\mathcal{M}(\Sigma)$ be the marking complexes of $S$ and $\Sigma$ respectively. Let $p\colon\mathcal{M}(S)\to\mathcal{C}(S)$ and $\pi\colon\mathcal{M}(\Sigma)\to\mathcal{C}(\Sigma)$ be maps that send a marking to some curve in that marking. ###### Theorem 6.1. Suppose that $\xi(S)\geq 2$ and $\phi\colon\mathcal{C}(S)\to\mathcal{C}(\Sigma)$ is a ${\sf q}$–quasi- isometric embedding. Then $\phi$ induces a coarse Lipschitz map $\Phi\colon\mathcal{M}(S)\to\mathcal{M}(\Sigma)$ so that the diagram $\begin{CD}\mathcal{M}(S)@>{\Phi}>{}>\mathcal{M}(\Sigma)\\\ @V{}V{p}V@V{}V{\pi}V\\\ \mathcal{C}(S)@>{\phi}>{}>\mathcal{C}(\Sigma)\end{CD}$ commutes up to an additive error. Furthermore, if $\phi$ is a quasi-isometry then so is $\Phi$. ###### Proof. For a marking $m$ and laminations $k$ and $\ell$, we say the triple $(m,k,\ell)$ is ${\sf c}$–admissible if * • $d_{S}\big{(}m,[k,\ell]\big{)}\leq 3$ and * • the pairs $(k,m)$, $(m,\ell)$ and $(k,\ell)$ are ${\sf c}$–cobounded. For ${\sf c}$ large enough and for every marking $m$, Proposition 3.3 shows that there exists a ${\sf c}$-admissible triple $(m,k,\ell)$. Given a ${\sf c}$-admissible triple $(m,k,\ell)$ we will now construct a triple $(\mu,\kappa,\lambda)$ for $\Sigma$. Let $\alpha$ be any curve in $\phi(m)\subset\mathcal{C}(\Sigma)$, $\kappa=\phi(k)$ and $\lambda=\phi(\ell)$. Note that (6.2) $d_{\Sigma}\big{(}\alpha,[\kappa,\lambda]\big{)}\leq 4{\sf q}+{\sf M}_{\Sigma},$ by the stability of quasi-geodesics (Lemma 2.1). Also $(\kappa,\lambda)$ is a ${\sf H}({\sf c})$–cobounded pair, by Theorem 5.2. Let $\beta$ be a closest point projection of $\alpha$ to the geodesic $[\kappa,\lambda]$. By Lemma 2.9, the pair $(\beta,\kappa)$ is $({\sf H}({\sf c})+{\sf c_{1}})$–cobounded. Using Lemma 3.1, there is a marking $\mu$ so that $\beta\in\operatorname{base}(\mu)$ and $(\mu,\kappa)$ are $({\sf H}({\sf c})+{\sf c_{1}}+{\sf c_{2}})$–cobounded. Therefore, for ${\sf C}=2{\sf H}({\sf c})+{\sf c_{1}}+{\sf c_{2}}$ the triple $(\mu,\kappa,\lambda)$ is ${\sf C}$–admissible. Define $\Phi(m)$ to be equal to $\mu$. 2pt $\mathcal{C}(\Sigma)$ [Bl] at 139.3 163.4 $\lambda$ [Br] at 13.3 134.8 $\lambda^{\prime}$ [tr] at 15.3 39.4 $\kappa$ [Bl] at 173.1 118.2 $\kappa^{\prime}$ [tl] at 168.4 50.9 $\alpha$ [l] at 96.9 115.9 $\alpha^{\prime}$ [l] at 79.8 50.4 $\mu$ [bl] at 96.6 92.6 $\mu^{\prime}$ [tl] at 79.8 75.3 $\begin{array}[]{c}\includegraphics[scale=1]{markings2}\end{array}$ Figure 2. Markings $\mu$ and $\mu^{\prime}$ are bounded apart. We now prove $\Phi$ is coarsely well-defined and coarsely Lipschitz. Suppose that $m$ and $m^{\prime}$ differ by at most one elementary move and the triples $(m,k,\ell)$ and $(m^{\prime},k^{\prime},\ell^{\prime})$ are ${\sf c}$-admissible. Let $(\mu,\kappa,\lambda)$ and $(\mu^{\prime},\kappa^{\prime},\lambda^{\prime})$ be any corresponding ${\sf C}$–admissible triples in $\Sigma$, as constructed above. (See Figure 2.) We must show that there is a uniform bound on the distance between $\mu$ and $\mu^{\prime}$ in the marking graph. By Lemma 2.7, it suffices to prove: ###### Claim. For every subsurface $\Omega\subseteq\Sigma$, $d_{\Omega}(\mu,\mu^{\prime})=O(1)$. Now, Lemma 2.6 gives $d_{S}(m,m^{\prime})\leq 4$. Deduce $d_{\Sigma}(\phi(m),\phi(m^{\prime}))\leq 5{\sf q}.$ Therefore, $\displaystyle d_{\Sigma}(\mu,\mu^{\prime})$ $\displaystyle\leq d_{\Sigma}\big{(}\mu,\phi(m)\big{)}+d_{\Sigma}\big{(}\phi(m),\phi(m^{\prime})\big{)}+d_{\Sigma}\big{(}\phi(m^{\prime}),\mu^{\prime})\big{)}$ $\displaystyle\leq 2(7{\sf q}+{\sf M}_{\Sigma}+2)+5{\sf q}.$ On the other hand, for any strict subsurface $\Omega\subset\Sigma$, we have $d_{\Omega}(\mu,\mu^{\prime})\leq d_{\Omega}(\mu,\kappa)+d_{\Omega}(\kappa,\kappa^{\prime})+d_{\Omega}(\kappa^{\prime},\mu^{\prime}).$ The first and third terms on the right are bounded by ${\sf C}$. By Theorem 5.2, the second term is bounded by ${\sf H}(2{\sf c}+4)$. This is because, for every strict subsurface $Y\subset S$, $d_{Y}(k,k^{\prime})\leq d_{Y}(k,m)+d_{Y}(m,m^{\prime})+d_{Y}(m^{\prime},k^{\prime})\leq 2{\sf c}+4.$ This proves the claim; thus $\Phi$ is coarsely well-defined and coarsely Lipschitz. Now assume that $\phi$ is a quasi-isometry with inverse $f\colon\mathcal{C}(\Sigma)\to\mathcal{C}(S)$. Let $\Phi$ and $F$ be the associated maps between marking complexes. We must show that $F\circ\Phi$ is close to the identity map on $\mathcal{C}(S)$. Fix $m\in\mathcal{M}(S)$ and define $\mu=\Phi(m)$, $m^{\prime}=F(\mu)$. For any admissible triple $(m,k,\ell)$ we have, as above, an admissible triple $(\mu,\kappa,\ell)$. Since $f(\kappa)=k$ and $f(\lambda)=\ell$ it follows that $(m^{\prime},k,\ell)$ is also admissible for a somewhat larger constant. As above, by Lemma 2.7 it suffices to show that $d_{Z}(m,m^{\prime})=O(1)$ for every essential subsurface $Z\subset S$. Since $f\circ\phi$ is close to the identity, commutivity up to additive error implies that $d_{S}(m,m^{\prime})$ is bounded. Since $(m,k)$ and $(m^{\prime},k)$ are cobounded, the triangle inequality implies that $d_{Z}(m,m^{\prime})=O(1)$ for strict subsurfaces $Z\subset S$. This completes the proof. ∎ ## 7\. Rigidity of the curve complex ###### Theorem 7.1. Suppose that $\xi(S)\geq 2$. Then every quasi-isometry of $\mathcal{C}(S)$ is bounded distance from a simplicial automorphism of $\mathcal{C}(S)$. ###### Proof. Let $f\colon\mathcal{C}(S)\to\mathcal{C}(S)$ be a ${\sf q}$-quasi-isometry. By Theorem 6.1 there is a ${\sf Q}$-quasi-isometry $F\colon\mathcal{M}(S)\to\mathcal{M}(S)$ associated to $f$. By Theorem 1.5 the action of $F$ is uniformly close to the induced action of some homeomorphism $G\colon S\to S$. That is, (7.2) $d_{\mathcal{M}}\big{(}F(m),G(m)\big{)}=O(1).$ Let $g\colon\mathcal{C}(S)\to\mathcal{C}(S)$ be the simplicial automorphism induced by $G$. We need to show that $f$ and $g$ are equal in $\operatorname{QI}(\mathcal{C}(S))$. Fix a curve $a\in\mathcal{C}(S)$. We must show the distance $d_{S}(f(a),g(a))$ is bounded by a constant independent of the curve $a$. Choose a marking $m$ containing $a$ as a base curve. Note that $d_{S}(a,p(m))\leq 2$, thus $d_{S}\big{(}f(a),f(p(m))\big{)}\leq 3{\sf q}.$ By Theorem 6.1, for every marking $m\in\mathcal{M}(S)$, $\displaystyle d_{S}\big{(}f(p(m)),p(F(m))\big{)}=O(1).$ From Equation 7.2 and Lemma 2.6 we have $d_{S}\big{(}p(F(m)),p(G(m))\big{)}=O(1).$ Also, $g(a)$ is a base curve of $G(m)$, hence $d_{S}\big{(}p\big{(}G(m)\big{)},g(a)\big{)}\leq 2.$ These four equations imply that $d_{S}\big{(}f(a),g(a)\big{)}=O(1).$ This finishes the proof. ∎ ## Appendix A Classifying mapping class groups For any compact, connected, orientable surface $S$ let $\mathcal{MCG}(S)$ be the extended mapping class group: the group of homeomorphisms of $S$, considered up to isotopy. We will use Ivanov’s characterization of Dehn twists via algebraic twist subgroups [12], the action of $\mathcal{MCG}(S)$ on $\mathcal{PML}(S)$, and the concept of a bracelet [2] to give a detailed proof of: ###### Theorem A.1. Suppose that $S$ and $\Sigma$ are compact, connected, orientable surfaces with $\mathcal{MCG}(S)$ isomorphic to $\mathcal{MCG}(\Sigma)$. Then either * • $S$ and $\Sigma$ are homeomorphic, * • $\\{S,\Sigma\\}=\\{S_{1},S_{1,1}\\}$, or * • $\\{S,\Sigma\\}=\\{\mathbb{S},\mathbb{D}\\}$. By the classification of surfaces, $S$ is determined up to homeomorphism by the two numbers ${\sf g}=\operatorname{genus}(S)$ and ${\sf b}=|\partial S|$. So to prove Theorem A.1 it suffices prove that $\xi(S)=3{\sf g}-3+{\sf b}$, the complexity of $S$, and ${\sf g}$, the genus, are algebraic: determined by the isomorphism type of $\mathcal{MCG}(S)$. ###### Remark A.2. Theorem A.1 is a well-known folk-theorem. A version of Theorem A.1, for pure mapping class groups, is implicitly contained in [12] and was known to N. Ivanov as early as the fall of 1983 [14]. Additionally, Ivanov and McCarthy [16] prove Theorem A.1 when ${\sf g}\geq 1$ (see also [15]). There is also a “folk proof” of Theorem A.1 relying on the fact that the rank and virtual cohomological dimension are algebraic and give two independent linear equations in the unknowns ${\sf g}$ and ${\sf b}$. However, the formula for the vcd changes when ${\sf g}=0$ and when ${\sf b}=0$. Thus there are two infinite families of pairs of surfaces which are not distinguished by these invariants. These difficult pairs can be differentiated by carefully considering torsion elements in the associated mapping class groups. We prefer the somewhat lighter proof of Theorem A.1 given here. The rank of a group is the size of a minimal generating set. The algebraic rank of a group $G$, $\operatorname{rank}(G)$, is the maximum of the ranks of free abelian subgroups $H<G$. Now, the algebraic rank of $\mathcal{MCG}(S)$ is equal to $\xi(S)$, when $\xi(S)\geq 1$ (Birman-Lubotzky-McCarthy [5]). When $\xi(S)\leq 0$ the algebraic rank is zero or one. So when $\xi(S)\geq 1$ the complexity is algebraically determined. There are only finitely many surfaces having $\xi(S)<1$; we now dispose of these and a few other special cases. ### Low complexity A Dehn twist along an essential, non-peripheral curve in an orientable surface has infinite order in the mapping class group. Thus, the only surfaces where the mapping class group has algebraic rank zero are the sphere, disk, annulus, and pants. We may compute these and other low complexity mapping class groups using the Alexander method [7]. For the sphere and the disk we find $\mathcal{MCG}(\mathbb{S}),~{}\mathcal{MCG}(\mathbb{D})\cong\mathbb{Z}_{2},$ where $\mathbb{Z}_{2}$ is the group of order two generated by a reflection. For the annulus and pants we find $\mathcal{MCG}(\mathbb{A})\cong K_{4}\quad\text{and}\quad\mathcal{MCG}(S_{0,3})\cong\mathbb{Z}_{2}\times\Sigma_{3},$ where $K_{4}$ is the Klein $4$–group and $\Sigma_{3}$ is the symmetric group acting on the boundary of $S_{0,3}$. Here, in addition to reflections, there is the permutation action of $\mathcal{MCG}(S)$ on $\partial S$. The surfaces with algebraic rank equal to one are the torus, once-holed torus, and four-holed sphere. For the torus and the once-holed torus we find $\mathcal{MCG}(\mathbb{T}),~{}\mathcal{MCG}(S_{1,1})\cong\operatorname{GL}(2,\mathbb{Z}).$ The first isomorphism is classical [28, Section 6.4]. The second has a similar proof: the pair of curves meeting once is replaced by a pair of disjoint arcs cutting $S_{1,1}$ into a disk. Now we compute the mapping class group of the four-holed sphere: $\mathcal{MCG}(S_{0,4})\cong K_{4}\rtimes\operatorname{PGL}(2,\mathbb{Z}).$ The isomorphism arises from the surjective action of $\mathcal{MCG}(S_{0,4})$ on the Farey graph. If $\phi$ lies in the kernel then $\phi$ fixes each of the slopes $\\{0,1,\infty\\}$ setwise. Examining the induced action of $\phi$ on these slopes and their intersections shows that $\phi$ is either the identity or one of the three “fake” hyperelliptic involutions. It follows that $\mathcal{MCG}(S_{0,4})$ has no center, and so distinguishes $S_{0,4}$ from $S_{1}$ and $S_{1,1}$. All other compact, connected, orientable surfaces have algebraic rank equal to their complexity and greater than one (again, see [5]). Among these $S_{1,2}$ and $S_{2}$ are the only ones with mapping class group having nontrivial center [7]. As the complexities of $S_{1,2}$ and $S_{2}$ differ, their mapping class groups distinguish them from each other and from surfaces with equal complexity. This disposes of all surfaces of complexity at most three except for telling $S_{0,6}$ apart from $S_{1,3}$. We defer this delicate point to the end of the appendix. ### Characterizing twists Recall that if $a\subset S$ is a essential non-peripheral simple closed curve then $T_{a}$ is the Dehn twist about $a$. We call $a$ the support of the Dehn twist. If $a$ is a separating curve and one component of $S{\smallsetminus}a$ is a pair of pants then $a$ is a pants curve. In this case there is a half- twist, $T_{a}^{1/2}$, about $a$. (When $a$ cuts off a pants on both sides then the two possible half-twists differ by a fake hyperelliptic.) We say that two elements $f,g\in\mathcal{MCG}(S)$ braid if $f$ and $g$ are conjugate and satisfy $fgf=gfg$. From [24, Lemma 4.3] and [16, Theorem 3.15] we have: ###### Lemma A.3. Two powers of twists commute if and only if their supporting curves are disjoint. Two twists braid if and only if their supporting curves meet exactly once. ∎ The proof generalizes to half-twists along pants curves: ###### Lemma A.4. Two half-twists commute if and only if the underlying curves are disjoint. Two half-twists braid if and only if the underlying curves meet exactly twice. ∎ Now, closely following Ivanov [12, Section 2], we essay an algebraic characterization of twists on nonseparating curves inside of $\mathcal{MCG}(S)$. Define a subgroup $H<\mathcal{MCG}(S)$ to be an algebraic twist subgroup if it has the following properties. * • $H=\langle g_{1},\ldots,g_{k}\rangle$ is a free abelian group of rank $k=\xi(S)$, * • for all $i,j$ the generators $g_{i},g_{j}$ are conjugate inside of $\mathcal{MCG}(S)$, * • for all $i$ and $n$ the center of the centralizer, $Z(C(g_{i}^{n}))$, is cyclic, and * • for all $i$, the generator $g_{i}$ is not a proper power in $C(H)$. We have: ###### Theorem A.5. Suppose that $\xi(S)\geq 2$ and $\mathcal{MCG}(S)$ has trivial center. Suppose that $H=\langle g_{1},\ldots,g_{k}\rangle$ is an algebraic twist group. Then the elements $g_{i}$ are all either twists on nonseparating curves or are all half-twists on pants curves. Furthermore, the underlying curves for the $g_{i}$ form a pants decomposition of $S$. ###### Proof. By work of Birman-Lubotzky-McCarthy [5] (see also [24]) we know that there is a power $m$ so that each element $f_{i}=g_{i}^{m}$ is either a power of Dehn twist or a pseudo-Anosov supported in a subsurface of complexity one. Suppose that $f_{i}$ is pseudo-Anosov with support in $Y\subset S$; then the center of the centralizer $Z(C(f_{i}))$ contains the group generated by $f_{i}$ and all twists along curves in $\partial Y$. However, this group is not cyclic, a contradiction. Deduce instead that each $f_{i}$ is the power of a Dehn twist. Let $b_{i}$ be the support of $f_{i}$. Since the $f_{i}$ commute with each other and are not equal it follows that the $b_{i}$ are disjoint and are not isotopic. Thus the $b_{i}$ form a pants decomposition. Since the $g_{i}$ are conjugate the same holds for the $f_{i}$. Thus all the $b_{i}$ have the same topological type. If $S$ has positive genus then it follows that all of the $b_{i}$ are nonseparating. If $S$ is planar then it follows that $S=S_{0,5}$ or $S_{0,6}$ and all of the $b_{i}$ are pants curves. Now fix attention on any $h\in C(H)$, the centralizer of $H$ in $\mathcal{MCG}(S)$. We will show that $h$ preserves each curve $b_{i}$. First recall that, for any index $i$ and any $n\in\mathbb{Z}$, the element $h$ commutes with $f_{i}^{n}$. Let $\ell\in\mathcal{PMF}(S)$ be any filling lamination. We have $\displaystyle f_{i}^{n}(\ell)\to b_{i}\quad\text{as}\quad n\to\infty,$ so $\displaystyle h\circ f_{i}^{n}(\ell)\to h(b_{i})\quad\text{and}\quad f_{i}^{n}\circ h(\ell)\to b_{i}.$ It follows that $h(b_{i})=b_{i}$ for all $i$. Fix attention on any $b_{\ell}\in\\{b_{i}\\}$. Suppose first that the two sides of $b_{\ell}$ lie in a single pair of pants, $P$. If $\partial P$ meets $b_{\ell}$ and no other pants curve then $S=S_{1,1}$, a contradiction. If $\partial P$ meets only $b_{\ell}$ and $b_{k}$ then $b_{\ell}$ is nonseparating and $b_{k}$ is separating, a contradiction. We deduce that $b_{\ell}$ meets two pants, $P$ and $P^{\prime}$. Now, if $h$ interchanges $P$ and $P^{\prime}$ then $S$ is in fact the union of $P$ and $P^{\prime}$; as $\xi(S)>1$ it follows that $S=S_{1,2}$ or $S_{2}$. However, in both cases the mapping class group has non-trivial center, contrary to hypothesis. We next consider the possibility that that $h$ fixes $P$ setwise. So $h|P$ is an element of $\mathcal{MCG}(P)$. If $h|P$ is orientation reversing then so is $h$; thus $h$ conjugates $f_{\ell}$ to $f_{\ell}^{-1}$, a contradiction. If $h|P$ permutes the components of $\partial P{\smallsetminus}b_{\ell}$ then either $b_{\ell}$ cuts off a copy of $S_{0,3}$ or $S_{1,1}$ from $S$; in the latter case we have $b_{i}$’s of differing types, a contradiction. To summarize: $h$ is orientation preserving, after an isotopy $h$ preserves each of the $b_{i}$, and $h$ preserves every component of $S{\smallsetminus}\\{b_{i}\\}$. Furthermore, when restricted to any such component $P$, the element $h$ is either isotopic to the identity or to a half twist. The latter occurs only when $P\cap\partial S=\delta_{+}\cup\delta_{-}$, with $h(\delta_{\pm})=\delta_{\mp}$. So, if the $b_{i}$ are nonseparating then $h$ is isotopic to the identity map on $S{\smallsetminus}\\{b_{i}\\}$. It follow that $h$ is a product of Dehn twists on the $b_{i}$. In particular this holds for each of the $g_{i}$ and we deduce that $T_{b_{i}}$, the Dehn twist on $b_{i}$, is an element of $C(H)$. Now, since $Z(C(g_{i}))=\mathbb{Z}$ we deduce that the support of $g_{i}$ is a single curve. By the above $g_{i}$ is a power of $T_{b_{i}}$; since $g_{i}$ is primitive in $C(H)$ we find $g_{i}=T_{b_{i}}$, as desired. The other possibility is that the $b_{i}$ are all pants curves. It follows that $S=S_{0,5}$ or $S_{0,6}$. Here $h$ is the identity on the unique pants component of $S{\smallsetminus}\\{b_{i}\\}$ meeting fewer than two components of $\partial S$. On the others, $h$ is either the identity or of order two. So $h$ is a product of half-twists on the $\\{b_{i}\\}$. As in the previous paragraph, this implies that $g_{i}=T_{a}^{1/2}$. ∎ ### Bracelets We now recall a pretty definition from [2]. Suppose $g$ is a twist or a half- twist. A bracelet around $g$ is a set of mapping classes $\\{f_{i}\\}$ so that * • every $f_{i}$ braids with $g$ (and so is conjugate to $g$), * • if $i\neq j$ then $f_{i}\neq f_{j}$ and $[f_{i},f_{j}]=1$, and * • no $f_{i}$ is equal to $g$. Note that bracelets are algebraically defined. The _bracelet number_ of $g$ is the maximal size of a bracelet around $g$. ###### Claim A.6. A half-twist on a pants curve has bracelet number at most two. ###### Proof. If two half-twists braid, then they intersect twice (Lemma A.4). Thus, the pants they cut off share a curve of $\partial S$. If two half-twists commute, the pants are disjoint (again, Lemma A.4). Therefore, there are at most two commuting half-twists braiding with a given half-twist. ∎ On the other hand: ###### Claim A.7. Suppose that $S\neq\mathbb{T}$. Then a twist on a nonseparating curve has bracelet number $2{\sf g}-2+{\sf b}$. ###### Proof. Suppose that $a$ is a non-separating curve with associated Dehn twist. Let $\\{b_{i}\\}$ be curves underlying the twists in the bracelet. Each $b_{i}$ meets $a$ exactly once (Lemma A.3). Also, each $b_{i}$ is disjoint from the others and not parallel to any of the others (again, Lemma A.3). Thus the $b_{i}$ cut $S$ into a collection of surfaces $\\{X_{j}\\}$. For each $j$ let $a_{j}=a\cap X_{j}$ be the remains of $a$ in $X_{j}$. Note that $a_{j}\neq\emptyset$. Each component of $\partial X_{j}$ either meets exactly one endpoint of exactly one arc of $a_{j}$ or is a boundary component of $S$. Now: * • if $X_{j}$ has genus, * • if $|\partial X_{j}|>4$, or * • if $|\partial X_{j}|=4$ and $a_{j}$ is a single arc, then there is a non-peripherial curve in $X_{j}$ meeting $a_{j}$ transversely in a single point. However, this contradicts the maximality of $\\{b_{i}\\}$. Thus every $X_{j}$ is planar and has at most four boundary components. If $X_{j}$ is an annulus then $S$ is a torus, violating our assumption. If $X_{j}$ is a pants then $a_{j}$ is a single arc. If $X_{j}=S_{0,4}$ then $a_{j}$ is a pair of arcs. An Euler characteristic computation finishes the proof. ∎ ### Proving the theorem Suppose now that $\xi(S)\geq 4$. By Theorem A.5 any basis element of any algebraic twist group is a Dehn twist on a nonseparating curve. Here the bracelet number is $2{\sf g}-2+{\sf b}$. Thus $\xi(S)$ minus the bracelet number is ${\sf g}-1$. This, together with the fact that $\xi(S)$ agrees with the algebraic rank, gives an algebraic characterization of ${\sf g}$. When $\xi(S)\leq 2$ the discussion of low complexity surfaces proves the theorem. The same is true when $\xi(S)=3$ and $\mathcal{MCG}(S)$ has nontrivial center. The only surfaces remaining are $S_{0,6}$ and $S_{1,3}$. In $\mathcal{MCG}(S_{0,6})$ every basis element of every algebraic twist group has bracelet number two. In $\mathcal{MCG}(S_{1,3})$ every basis element of every algebraic twist group has bracelet number three. So Theorem A.1 is proved. ∎ ## References * [1] Jason Behrstock, Bruce Kleiner, Yair Minsky, and Lee Mosher. Geometry and rigidity of mapping class groups. arXiv:0801.2006. * [2] Jason Behrstock and Dan Margalit. Curve complexes and finite index subgroups of mapping class groups. Geom. Dedicata, 118:71–85, 2006. arXiv:math/0504328. * [3] Jason A. Behrstock. Asymptotic geometry of the mapping class group and Teichmüller space. Geom. Topol., 10:1523–1578 (electronic), 2006. arXiv:math/0502367. * [4] Gregory C. Bell and Koji Fujiwara. The asymptotic dimension of a curve graph is finite. J. Lond. Math. Soc. (2), 77(1):33–50, 2008. arXiv:math/0509216. * [5] Joan S. Birman, Alex Lubotzky, and John McCarthy. Abelian and solvable subgroups of the mapping class groups. Duke Math. J., 50(4):1107–1120, 1983. http://www.math.columbia.edu/$\sim$jb/papers.html. * [6] Martin R. Bridson and André Haefliger. Metric spaces of non-positive curvature. Springer-Verlag, Berlin, 1999. * [7] Benson Farb and Dan Margalit. A primer on mapping class groups, 2010. http://www.math.utah.edu/$\sim$margalit/primer/. * [8] David Gabai. Almost filling laminations and the connectivity of ending lamination space. Geom. Topol., 13(2):1017–1041, 2009. arXiv:0808.2080. * [9] Mikhael Gromov. Hyperbolic groups. In Essays in group theory, pages 75–263. Springer, New York, 1987\. * [10] Ursula Hamenstaedt. Geometry of the mapping class groups III: Quasi-isometric rigidity. arXiv:math/0512429. * [11] Willam J. Harvey. Boundary structure of the modular group. In Riemann surfaces and related topics: Proceedings of the 1978 Stony Brook Conference (State Univ. New York, Stony Brook, N.Y., 1978), pages 245–251, Princeton, N.J., 1981. Princeton Univ. Press. * [12] N. V. Ivanov. Automorphisms of Teichmüller modular groups. In Topology and geometry—Rohlin Seminar, volume 1346 of Lecture Notes in Math., pages 199–270. Springer, Berlin, 1988. * [13] Nikolai V. Ivanov. Mapping class groups. In Handbook of geometric topology, pages 523–633. North-Holland, Amsterdam, 2002. * [14] Nikolai V. Ivanov. 2007\. Personal communication. * [15] Nikolai V. Ivanov and John D. McCarthy. On injective homomorphisms between Teichmüller modular groups. 1995\. http://www.mth.msu.edu/$\sim$ivanov/i.ps. * [16] Nikolai V. Ivanov and John D. McCarthy. On injective homomorphisms between Teichmüller modular groups. I. Invent. Math., 135(2):425–486, 1999. * [17] Michael Kapovich. Hyperbolic manifolds and discrete groups. Birkhäuser Boston Inc., Boston, MA, 2001. * [18] Erica Klarreich. The boundary at infinity of the curve complex and the relative Teichmüller space. http://nasw.org/users/klarreich/research.htm. * [19] Mustafa Korkmaz. Automorphisms of complexes of curves on punctured spheres and on punctured tori. Topology Appl., 95(2):85–111, 1999. * [20] Christopher J. Leininger and Saul Schleimer. Connectivity of the space of ending laminations. Duke Math. J., 150(3):533–575, 2009. arXiv:0801.3058. * [21] Feng Luo. Automorphisms of the complex of curves. Topology, 39(2):283–298, 2000. arXiv:math/9904020. * [22] Howard A. Masur and Yair N. Minsky. Geometry of the complex of curves. I. Hyperbolicity. Invent. Math., 138(1):103–149, 1999. arXiv:math/9804098. * [23] Howard A. Masur and Yair N. Minsky. Geometry of the complex of curves. II. Hierarchical structure. Geom. Funct. Anal., 10(4):902–974, 2000. arXiv:math/9807150. * [24] John D. McCarthy. Automorphisms of surface mapping class groups. A recent theorem of N. Ivanov. Invent. Math., 84(1):49–71, 1986. * [25] Yair N. Minsky. The classification of Kleinian surface groups, I: Models and bounds. arXiv:math/0302208. * [26] Kasra Rafi and Saul Schleimer. Covers and the curve complex. Geom. Topol., 13(4):2141–2162, 2009. arXiv:math/0701719. * [27] Saul Schleimer. The end of the curve complex. To appear. arXiv:math/0608505. * [28] John Stillwell. Classical topology and combinatorial group theory, volume 72 of Graduate Texts in Mathematics. Springer-Verlag, New York, second edition, 1993.

arxiv-papers

2024-09-04T02:48:53.247661

0710.3901

# Simple, Linear-time Modular Decomposition(Extended Abstract)111Marc Tedder and Derek Corneil thank NSERC of Canada for partially funding this research; Michel Habib and Christophe Paul received funding from the French ANR project “Graph Decomposition and Algorithms” (GRAAL). Marc Tedder111Department of Computer Science, University of Toronto; {mtedder,dgc}@cs.toronto.edu, Derek Corneil111Department of Computer Science, University of Toronto; {mtedder,dgc}@cs.toronto.edu, Michel Habib222LIAFA and the University of Paris 7 - Denis Diderot; habib@liafa.jussieu.fr, Christophe Paul333CNRS - LIRMM, Univ. Montpellier II France; christophe.paul@lirmm.fr; part of this research was conducted while on sabbatical in the School of Computer Science at the University of McGill, Montréal, Canada. ###### Abstract Modular decomposition is fundamental for many important problems in algorithmic graph theory including transitive orientation, the recognition of several classes of graphs, and certain combinatorial optimization problems. Accordingly, there has been a drive towards a practical, linear-time algorithm for the problem. Despite considerable effort, such an algorithm has remained elusive. The linear-time algorithms to date are impractical and of mainly theoretical interest. In this paper we present the first simple, linear-time algorithm to compute the modular decomposition tree of an undirected graph. ## 1 Introduction A natural operation to perform on a graph $G$ is to take one of its vertices, say $v$, and replace it with another graph $G^{\prime}$, making $v$’s neighbours universal to the vertices of $G^{\prime}$. Modular decomposition is interested in the inverse operation: finding a set of vertices sharing the same neighbours outside the set – that is, finding a _module_ – and contracting this module into a single vertex. A graph’s modules form a partitive family [2], and as such, define a decomposition scheme for the graph with an associated decomposition tree composed of the graph’s _strong modules_ – those that don’t overlap other modules. To compute this _modular decomposition tree_ is to compute the _modular decomposition_ (and vice versa); and with its succinct representation of a graph’s structure, its computation is often a first-step in many algorithms. Indeed, since Gallai first noticed its importance to comparability graphs [11], modular decomposition has been established as a fundamental tool in algorithmic graph theory. All efficient transitive orientation algorithms make essential use of modular decomposition (e.g., [17]). It is frequently employed in recognizing different families of graphs, including interval graphs [18], permutation graphs [24], and cographs [3]. Furthermore, restricted versions of many combinatorial optimization problems can be efficiently solved using modular decomposition (e.g., [8]). While the papers [18, 19, 20] provide older surveys of its numerous applications, new uses continue to be found, such as in the areas of graph drawing [23] and bioinformatics [10]. Not surprisingly, the problem of computing the modular decomposition has received considerable attention. Much like planarity testing and interval graph recognition, the importance of the problem has bent efforts toward a simple and efficient solution. The first polynomial-time algorithm appeared in the early 1970’s and ran in time $O(n^{4})$ [5]. Incremental improvements were made over the years – [13, 21], for example – culminating in 1994 with the first linear-time algorithms, developed independently by McConnell and Spinrad [16], and Cournier and Habib [4]. These are unfortunately so complex as to be viewed primarily as theoretical contributions, with Spinrad himself hoping they would be supplanted by something simpler (pg. 149, [25]). Subsequent algorithms, though, have fallen short, either failing to achieve linear-time or appealing to sophisticated data-structure techniques in doing so. The attempts made by [17] and [7] are illustrative. Both adopt an approach pioneered by Ehrenfeucht et. al. [9], later improved upon by Dahlhaus [6]. The idea is to pick an arbitrary vertex, say $x$, and recursively compute the modular decomposition tree for its neighbourhood, $N(x)$, and its non- neighbourhood, $\overline{N(x)}$. Any strong module not containing $x$ must be a module of either $G[N(x)]$ or $G[\overline{N(x)}]$, and therefore can be extracted from their recursively computed modular decomposition trees. Once extracted, these can then be used to compute the strong modules containing $x$. The two types of modules are then assembled to form the tree. Although this approach is conceptually simple, [17] only managed an $O(n+m\log{n})$ implementation, while [7] required advanced union-find data structures and complicated charging arguments to achieve linear-time. The difficult step in the recursive approach is the computation of the strong modules containing $x$ and their incorporation into the tree; in other words, the explicit construction of the tree. Capelle and Habib [1] responded by proposing the use of _factorizing permutations_ , a permutation of the vertices in which the strong modules appear consecutively. They suggested that a factorizing permutation be computed in place of the tree; if the tree is required it can be derived from the permutation once, at the end of the algorithm, using the linear-time procedure in [1]. But how to compute the permutation? The linear-time algorithm claimed in [12] contains an error that kills its simplicity, and the algorithm of [15] has a $\log{n}$-factor. It seemed factorizing permutations merely traded one bottleneck for another. The real problem with the two approaches is that they were applied in isolation. This paper shows that the two are truly complementary: the recursively computed trees facilitate the computation of a factorizing permutation, which in turn facilitates the computation of the modular decomposition tree. By unifying the approaches in this way we produce an elegant, linear-time algorithm for the modular decomposition, thus realizing a long-standing goal in the area. We combine the best aspects of the two methods, maintaining the conceptual simplicity of both. This allows a straightforward proof of correctness. The only data-structure employed is an ordered list of trees, and on these, only elementary traversals are required. Moreover, to produce the factorizing permutation from the recursively computed trees, we introduce a procedure that generalizes partition refinement [22] from sets to trees. This and other ideas we develop here can also be applied to the transitive orientation problem: the authors are confident of having developed the first simple, linear-time transitive orientation algorithm by doing just this [26]. ### 1.1 Preliminaries All graphs in this paper are simple and undirected. Connected components will simply be referred to as _components_ , while the connected components of the complement will be referred to as _co-components_. We will talk often of an _ordered list of trees_ , which will sometimes be referred to as an _ordered forest_. When we speak of them as defining an ordering of the graph’s vertices, we mean a pre-ordering of the leaves of each tree in order. Note that sometimes a set of vertices will be referred to as a “tree”. We do this to streamline the exposition; our intent will become clear. The modular decomposition tree will occasionally be referred to as the _MD tree_. The MD tree can be recursively defined as follows: the root of the tree corresponds to the entire graph; if the graph is disconnected, the root is called _parallel_ and its children are the MD trees of its components; if the graph’s complement is disconnected, the root is called _series_ and its children are the MD trees of the co-components; in all other cases the root is called _prime_ 555This definition of prime differs somewhat from that which normally appears in the literature., and its children are the MD trees of the graph’s maximal modules. Recall that the nodes in this tree are the graph’s strong modules, which are those that don’t _overlap_ others. ### 1.2 Outline of the Paper The rest of the paper breaks down into four sections. The first provides an overview of the algorithm, explaining its operation and how this contributes to its correctness and the ultimate construction of the MD tree. In the next section we specify the algorithm in detail and sketch the proof of its correctness. An analysis of the algorithm’s running time follows. The paper concludes with a discussion of our contributions. The appendix contains an example and some omitted proofs. ## 2 Overview of the Algorithm ### 2.1 Recursion The algorithm begins in a familiar way, selecting an arbitrary vertex, $x$, called the _pivot_ , and placing its neighbourhood to its left and its non- neighbourhood to its right, giving us the ordered list of trees, $N(x),x,\overline{N(x)}$. Next, the modular decomposition tree for $G[N(x)]$ is recursively computed. As this occurs, the neighbours of $N(x)$ in $\overline{N(x)}$ are “pulled forward” so that afterwards we have the ordered list of trees, $T(N(x)),x,\overline{N_{A}(x)},\overline{N_{N}(x)}$, where $T(N(x))$ is the modular decomposition tree for $G[N(x)]$, and $\overline{N_{A}(x)}$ is the subset of $\overline{N(x)}$ with at least one neighbour in $N(x)$. The algorithm then recursively computes the modular decomposition tree for $\overline{N_{A}(x)}$, pulling its neighbours in $\overline{N_{N}(x)}$ forward in a similar fashion. And so on. Eventually we arrive at the following ordered list of trees: $\underbrace{T(N_{0})}_{N(x)},x,\underbrace{T(N_{1}),\ldots,T(N_{k})}_{\overline{N(x)}},$ (1) where the $N_{i}$’s correspond to the distance layers in a breadth-first- search begun from $x$, and the $T(N_{i})$’s are their modular decomposition trees. The rest of this paper assumes that the graph is connected and thus each vertex in $N_{i}$ has an edge to $N_{i-1}$ (or $x$ in the case of $N_{0}$). When the graph is disconnected, the $N_{i}$’s up to $N_{k-1}$ along with $x$ form one of its connected components. In this case the algorithm builds the MD tree for this component as described below, then unifies the result with $T(N_{k})$ under a common root labeled parallel. This adds a constant amount of work to each stage. Each stage is defined by a pivot, and vertices are only pivots once, so this work is consistent with linear-time. ### 2.2 Refinement We wish to transform the above ordered list of trees into a factorizing permutation that will help build the modular decomposition tree. We begin doing so by refining the trees using the active edges: ###### Definition 2.1. An edge is _active_ if it is incident to $x$ or if its endpoints are in different $N_{i}$’s. Refinement is a natural generalization of partition refinement from sets to trees, or equivalently, from sets to multi-sets. We process each vertex in turn and use its incident active edges to refine the trees other than its own. The process amounts to a simple recursive marking procedure and is specified in detail in section 3.1. To see how refinement moves us toward a factorizing permutation, first consider a strong module not containing $x$, say $M$. Notice that for some $N_{i}$, we have $M\subseteq N_{i}$, with $M$ a module of $G[N_{i}]$. A theorem of [19] says that either $M$ is a strong module in $G[N_{i}]$, and thus an internal node in $T(N_{i})$, or it is the union of siblings in $T(N_{i})$. In the former case, the node corresponding to $M$ will be unaffected by refinement; in the latter case, refinement will group the siblings under a new internal node inserted into $T(N_{i})$. Thus: ###### Lemma 2.2 (Proved in section 3.1). The strong modules not containing $x$ appear consecutively after refinement. We are not so fortunate with the strong modules containing $x$, although refinement does get them close to appearing consecutively. As described above, refinement groups siblings under new nodes. When those siblings are at depth-1, however, instead of making that new node the child of the siblings’ former parent, the new node is made the root of its own tree in our ordered list – the siblings’ old tree is effectively split. The intuition here comes from the special role played by the (co-)components of the $G[N_{i}]$’s and their placement within the $T(N_{i})$’s: ###### Proposition 2.3 (Proved in the appendix). If $C$ is a co-component of $G[N_{0}]$ and $M$ is a strong module containing $x$, then either $C\subset M$ or $C\cap M=\emptyset$. Similarly for $C$ a component of $G[N_{i}],i>0$. Of course, the (co-)components are either the roots of the $T(N_{i})$’s or are the nodes at depth-1. The result of this splitting is that the strong modules containing $x$ will be “bound” by the trees in our ordered forest: ###### Lemma 2.4 (Proved in section 3.1). Let $T_{k},\ldots,T_{1},x,T^{\prime}_{1},\ldots,T^{\prime}_{\ell}$ be the ordered forest resulting from refinement, and let $M$ be a strong module containing $x$. Then there are _bounding trees_ $T_{i}$ and $T^{\prime}_{j}$ such that, $M\supset T_{i-1}\cup\cdots\cup T_{1}\cup\\{x\\}\cup T^{\prime}_{1}\cup\cdots\cup T^{\prime}_{j-1}$ and $M\subseteq T_{i}\cup\cdots\cup T_{1}\cup\\{x\\}\cup T^{\prime}_{1}\cup\cdots\cup T^{\prime}_{j}$. ### 2.3 Promotion When siblings are grouped under a new node during refinement it is because a vertex in a different tree is adjacent to them but not their other siblings. The siblings’ former parent cannot therefore be a module; this is also true of all their ancestors. Refinement accounts for this by marking these nodes for deletion. When refinement has finished, the nodes without marked children will correspond to the strong modules not containing $x$. Promotion is the process of deleting all the marked nodes with marked children – internal nodes are “promoted” upward as their ancestors are deleted – leaving only the strong modules not containing $x$. The real benefit of promotion however is that it gives us the desired factorizing permutation. The strong modules not containing $x$ are left intact and are therefore consecutive. But now the strong modules containing $x$ will also be consecutive: as nodes are deleted from these modules’ bounding trees, the ones that remain and that are in the module are placed next to the rest of the module. Section 3.2 details the procedure, a simple depth-first traversal of our ordered forest. So with nothing more than elementary traversals of our ordered forest we arrive at a factorizing permutation: ###### Lemma 2.5 (Proved in section 3.2). The ordered forest that results from promotion provides a factorizing permutation. ### 2.4 Assembly In fact, promotion gives us much more than a factorizing permutation: we have an ordered list of trees whose nodes (excepting $x$) correspond to the strong modules not containing $x$; moreover, each of these strong modules is itself properly decomposed (their parts were originally in their respective $T(N_{i})$’s, and neither refinement nor promotion changes this). What remains, then, is to identify the strong modules containing $x$, determine the trees in our list constituting them, then use this information to assemble the modular decomposition tree. This was the part that proved difficult for the previous recursive algorithms. Our factorizing permutation makes it easy. With a factorizing permutation we know the strong modules containing $x$ are nested: $[\cdots[\cdots[\cdots x\cdots]\cdots]\cdots].$ Since our ordered forest consists of the strong modules not containing $x$, no tree in it overlaps these brackets. So to build the MD tree, it suffices to insert the brackets between the trees in our list: once this is done, a node is made for each pair of brackets and a “spine” for the MD tree is built; to this we merely affix the trees in our list according to the placement of the brackets. A simple greedy algorithm, described in section 3.3, inserts the brackets. In this way the modular decomposition tree is built with minimal effort. ## 3 Details and Correctness ### 3.1 Refinement The refinement process described in the overview is given by algorithm 1. foreach _vertex $v$_ do Let $\alpha(v)$ be its incident active edges; Refine the list of trees using $\alpha(v)$ according to algorithm 2, such that: if _$v$ is to $x$’s left_ then refine using left splits, and when a node is marked, mark it with “left”; else if _$v$ is to $x$’s right and refines a tree to $x$’s left_ then refine using left splits, and when a node is marked, mark it with “left”; else if _$v$ is to $x$’s right and refines a tree to $x$’s right_ then refine using right splits, and when a node is marked, mark it with “right”; end end Algorithm 1 Refinement of the ordered list of trees (1) by the active edges Let $T_{1},\ldots,T_{k}$ be the maximal subtrees in the forest whose leaves are all in $X$; Let $P_{1},\ldots,P_{\ell}$ be the set of parents of the $T_{i}$’s; foreach _non-prime $P_{i}$_ do Let $A$ be the set of $P_{i}$’s children amongst the $T_{j}$’s, and $B$ its remaining children; Let $T_{a}$ either be the single tree in $A$ or the tree formed by unifying the trees in $A$ under a common root, and define $T_{b}$ symmetrically; Assign $P_{i}$’s label to $T_{a}$ and $T_{b}$; if _$P_{i}$ is a root_ then Replace $P_{i}$ in the forest with either $T_{a},T_{b}$ (left split) or $T_{b},T_{a}$ (right split) else Replace the children of $P_{i}$ with $T_{a}$ and $T_{b}$; end Mark the roots of $T_{a}$ and $T_{b}$ as well as all their ancestors; end foreach _prime $P_{i}$_ do Mark $P_{i}$ as well as all of its children and all of its ancestors; end Algorithm 2 Refinement of an ordered list of trees by the set $X$ Below we sketch the proof of lemmas 2.2 and 2.4. For the former we actually prove something slightly stronger from which lemma 2.2 follows immediately: ###### Lemma 3.1. The nodes in the ordered list of trees resulting from refinement that do not have marked children correspond exactly to the strong modules containing $x$. ###### Proof. [Sketch] Let $M$ be a strong module not containing $x$. As stated in the overview, $M$ must be entirely contained in some $N_{i}$, and it must be a module of $G[N_{i}]$. A theorem of [19] guarantees that $M$ is either a node in $T(N_{i})$ or the union of children, say $c_{1},\ldots,c_{k}$, of a series or parallel node in $T(N_{i})$. Appealing to algorithm 1, we see that in the former case it remains a node throughout refinement and none of its children are ever marked, since each vertex outside $T(N_{i})$ is either universal to, or isolated from, the node. Algorithm 1 also makes clear that in the latter case the children will remain siblings throughout refinement, and will not be marked at any time, since, again, each refining vertex is either universal to them or isolated from them. So for contradiction, assume that after refinement the $c_{i}$’s have a sibling $c$ different from them. Inspecting algorithm 1, we see that $c$ must have been a sibling of the $c_{i}$’s in $T(N_{i})$, and that $c$ and the $c_{i}$’s must have the same set of neighbours outside $N_{i}$. Hence, $c\cup c_{1}$ is a module overlapping $c_{1}\cup\cdots\cup c_{k}$, contradicting the latter being strong. For the converse, consider a node $N$ without any marked children, and suppose $N$ was formed from the refinement of $T(N_{i})$. Clearly, the vertices of $N$ have the same neighbours outside $T(N_{i})$. By algorithm 1, if $N$ is prime, it existed in $T(N_{i})$ and so has the same neighbours within $T(N_{i})$. This is also true when $N$ is not prime, since its children must have been children of the same non-prime node in $T(N_{i})$. Hence, each node with unmarked children is a module. If the node existed in $T(N_{i})$ then it is clearly strong. If it is new, a simple case analysis shows that no other module can overlap it, since two overlapping modules must be a module themselves. ∎ ###### Proof. [Sketch of lemma 2.4] We prove this by induction on the number of vertices refining. Prior to refinement we have the ordered list of trees $T(N_{0}),x,T(N_{1}),\ldots,T(N_{k})$. In the appendix we show that if $M\cap N_{i}\neq\emptyset$ for some $i>1$, then $M=V$. Thus, the lemma holds prior to refinement since $T(N_{0})$ and either $T(N_{1})$ or $T(N_{k})$ can be taken as the bounding trees. So suppose there are such bounding trees $T_{i}$ and $T^{\prime}_{j}$ after some number of vertices have refined; now consider what happens after the next vertex refines. Clearly we need only focus on $T_{i}$ and $T^{\prime}_{j}$; we’ll argue the case for $T_{i}$, with the case for $T^{\prime}_{j}$ being similar. Now, if $T_{i}$ is not split we are done, so assume $T_{i}$ is split and replaced by the trees $T_{A},T_{B}$ in order. Let $v$ be the vertex doing the refining and observe that $v$ is universal to the leaves of $T_{A}$ and not universal to the leaves of $T_{B}$; additionally, we must have $v\in N_{1}$. If $v\in M$ as well, then $v$ is universal to the portion of $T_{i}$ outside $M$ and hence we take $T_{A}$ as the new left-bounding tree. If $v\notin M$, then it is isolated from the portion of $T_{i}$ in $M$, and so we take $T_{B}$ as the new left-bounding tree in this case. ∎ ### 3.2 Promotion The promotion process is given by algorithm 3. Below we sketch the proof of lemma 2.5. The key here is that refinement distinguishes between nodes marked “left” and “right” and promotion handles these cases differently. while _there is a root $r$ with a child $c$ both marked by “left”_ do Remove from $r$ the subtree rooted at $c$ and place it just before $r$; end while _there is a root $r$ with a child $c$ both marked by “right”_ do Remove from $r$ the subtree rooted at $c$ and place it just after $r$; end Delete all marked roots in the forest with one child, replacing them with that child; Delete all marked roots in the forest with no children; Remove all marks; Algorithm 3 The promotion algorithm ###### Proof. [Sketch of lemma 2.5] By lemma 3.1 and inspection of algorithm 3, we see that the strong modules not containing $x$ will appear consecutively after promotion. Let $M$ be a strong module containing $x$. Let $T_{i}$ and $T^{\prime}_{j}$ be the bounding trees provided by lemma 2.4. It suffices to show that promotion deletes nodes in such a way as to place the portions of $T_{i}$ and $T_{j}^{\prime}$ that are in $M$ next to the other vertices in $M$. We’ll focus on $T^{\prime}_{j}$, with the case for $T_{i}$ following similarly. In the proof of lemma 2.4 we observed that if $M\cap N_{i}\neq\emptyset$ for some $i>1$, then $M=V$. As such, we’ll assume $T^{\prime}_{j}$ is composed of vertices in $N_{1}$. If $T^{\prime}_{j}$ only contains vertices in $M$, then clearly we are done since promotion does not rearrange trees in our ordered list. So assume $T^{\prime}_{j}$ contains some vertices in $M$ and some outside $M$. By proposition 2.3, this means it contains vertices in at least two different components of $G[N_{1}]$, say $C$ and $C^{\prime}$ with $C\subset M$ and $C^{\prime}\cap M=\emptyset$. Now, $C$ and $C^{\prime}$ were siblings at depth-1 in $T(N_{1})$, and by assumption, some portion of each remains in the same tree after refinement. Appealing to algorithm 1, we see that this is only possible if all vertices in $C$ and $C^{\prime}$ remain in the same tree after refinement; that is, $C$ and $C^{\prime}$ must still be siblings after refinement, which means they remained siblings throughout refinement. If both $C$ and $C^{\prime}$ share the same neighbours outside $N_{1}$, then $C\cup C^{\prime}$ is a module overlapping $M$, contradicting $M$ being strong. It follows that at least one of $C$ and $C^{\prime}$ is marked “left” or “right” (or both). We now consider the cases: _Case 1_ : Assume $C^{\prime}$ is marked by “left”. This means a vertex in $N_{0}$ is adjacent to some but not all vertices in $C^{\prime}$; let $v$ be the first such vertex. Note that $v\notin M$ if it is adjacent to some of $C^{\prime}$; thus, $v$ is universal to $C$. But we remarked above that $C$ and $C^{\prime}$ had the same parents throughout refinement; so at the time $v$ refined, it would have split $C$ away from $C^{\prime}$, contradicting their being siblings afterwards. This case is therefore impossible. _Case 2_ : Assume $C^{\prime}$ is marked by “right”. Observe that no vertex in $C$ can be adjacent to a vertex in $N_{i}$, $i>1$, since such vertices are outside $M$ and not adjacent to $x$. Thus $C$ cannot be marked by “right”. Thus, promotion places the vertices of $C^{\prime}$ to the right of those in $C$. _Case 3_ : Assume $C^{\prime}$ is not marked by a split. Then $C$ must be marked by a split, as argued above, and as seen in case 2, it must be a left- split that marks it. Thus, promotion places the vertices of $C$ to the left of those of $C^{\prime}$. In all cases, promotion puts $C$ to the left of $C^{\prime}$. Since $C$ and $C^{\prime}$ were chosen arbitrarily, we can conclude that the vertices of $M$ appear consecutively. ∎ ### 3.3 Assembly After promotion we are left with an ordered list of trees representing a factorizing permutation. As we explained in the overview, the problem of constructing the modular decomposition tree reduces to placing brackets between these trees in a way that delineates the strong modules containing $x$. We can actually simplify things even further. Recall from the end of section 2.2 that the (co-)components of the $G[N_{i}]$’s appear consecutively prior to refinement. A look at algorithms 1 and 3 confirms that this holds after promotion as well. Our ordered list of trees can therefore be viewed as an ordered list of (co-)components: $C_{\kappa},\ldots,C_{1},x,C_{1}^{\prime},\ldots,C_{\lambda}^{\prime}$, where the $C_{i}$’s correspond to the co-components of $G[N_{0}]$, and the $C^{\prime}_{i}$’s correspond to the components of the $G[N_{j}]$’s, $j>0$. Proposition 2.3 allows us to place the brackets between these instead. A simple greedy procedure based on the following lemma does this easily: ###### Lemma 3.2 (Proved in the appendix.). Let $M$ be the smallest strong module containing $x$. Then $M$ satisfies one of the following three conditions: 1. (i) $M$ is the maximally contiguous module containing $x$ and no $C^{\prime}_{i}$ (in which case $M$ is series); 2. (ii) $M$ is the maximally contiguous module containing $x$ and no $C_{i}$, and only $C^{\prime}_{j}$’s in $N_{1}$ with no edge to their right (in which case $M$ is parallel); 3. (iii) $M$ is the minimally contiguous module containing $x$ and at least $C_{1}$ and $C_{1}^{\prime}$ (in which case $M$ is prime). To use the lemma we first determine if any vertex in a $C^{\prime}_{i}$ in $N_{1}$ has a neighbour in a $N_{j}$, $j>1$ (as required by (ii) above). Next, we determine the $\mu$-values of the (co-)components: for $C_{i}$ this is defined as follows: let $C^{\prime}_{j}$ be the co-component with smallest index such that $C^{\prime}_{\lambda},\ldots,C^{\prime}_{j}$ are all isolated from $C_{i}$, then $\mu(C_{i})$ is $x$ if $j=1$ and $C^{\prime}_{j-1}$ otherwise; the $\mu$-values for the $C^{\prime}_{i}$’s are defined symmetrically. These $\mu$-values help the procedure determine when a module is formed. Given this information, the procedure can follow the lemma directly, first trying for a series module, then a parallel module if this fails, and finally a prime module failing this. Series and parallel modules are attempted by comparing the $\mu$-values of the (co-)components against $x$ and maximally adding those for which the two are equal. Prime modules are formed by first adding $C_{1}$ and $C^{\prime}_{1}$, and then iteratively applying the following rule: once a $C_{i}$ is added, so too must be $C^{\prime}_{1},\ldots,\mu(C_{i})$ (symmetrically for a $C^{\prime}_{i}$ being added), stopping when the rule can no longer be applied. Once a module is found, brackets are placed accordingly and the process begins anew, treating the just formed module as though it were $x$. ## 4 Running Time and Implementation ### 4.1 Recursion In order to effect the partitioning required of the recursion, we need to traverse the pivot’s adjacency list in its entirety. However, each vertex is a pivot exactly once during the algorithm, so this is consistent with linear- time. We will need to isolate the incident active edges of each vertex so that refinement, promotion, and assembly can be performed efficiently; this can be done during the recursion. Initially we assume all vertices are marked as _unvisited_ and that each has associated with it an empty list denoted by $\alpha$ (which will be used to store the incident active edges). As pivots are chosen during the recursion they are marked as _visited_. When a pivot’s adjacency list is traversed, it is appended to the $\alpha$-list of all its visited neighbours. Thus, after recursion the $\alpha$-lists of each vertex in $N_{i}$ will correspond to their incident active edges to $N_{i+1}$. The rest of their active edges can then be added by traversing the $\alpha$-list of each vertex, and appending vertices to the other $\alpha$-lists in the obvious way. At the end of each stage the $\alpha$-lists must be cleared to satisfy our induction hypothesis. We can thus assume that the active edges at each stage can be isolated at the cost of work proportional to their number. Notice that each edge is active precisely once during the algorithm, so this effort is consistent with linear-time overall. ### 4.2 Refinement A simple recursive marking procedure finds the maximal subtrees required by algorithm 2. All nodes in our trees have at least two children, so the sizes of these subtrees are linear in the number of their leaves, which is equal to the number of incident active edges of the vertex refining. Notice that each vertex has at least one incident active edge. Thus, finding these trees (and the constant amount of work required afterward) is proportional to the number of active edges at each stage and so is consistent with linear-time. The children of a prime node need only be marked once, and the ancestors of a node need only be marked twice (once each for “left” and “right”). The time for this marking is therefore proportional to the size of our ordered forest, which is linear in the number of its leaves, which is linear in the number of active edges (since each leaf has at least one active edge), and hence consistent with linear-time overall. ### 4.3 Promotion If we implement promotion in a depth-first manner, we see that it requires no more than a single traversal of our ordered forest, which as just observed, is consistent with linear-time. ### 4.4 Assembly Identifying the (co-)components requires at most two traversals of the forest: one prior to refinement to mark them and one after promotion to retrieve them. Determining if a $C^{\prime}_{i}$ has an edge to its right needs only a traversal of each vertex’s $\alpha$-list. Computing the $\mu$-values of the (co-)components can be accomplished by processing each vertex in order and traversing its $\alpha$-list. All this work is therefore consistent with linear-time. The placement of the brackets amounts to a single traversal of the list of (co-)components, each of which contains an active edge, and so is consistent with linear-time. The final assembly of the tree can be done merely by traversing our ordered forest, and is therefore consistent with linear-time. ## 5 Conclusion Like other algorithmic problems of comparable importance, research in modular decomposition has focused on finding a simple, efficient algorithm for its computation. This paper finally provides such an algorithm. There have been many previous attempts, but all have either failed to achieve linear-time or were complicated to the point of being impractical. Our algorithm suffers from no such shortcomings. Its elegance derives from unifying two existing approaches, utilizing the best elements from each. The unification is effected through the introduction of a new refinement technique which generalizes partition refinement from sets to trees. To our knowledge, no similar type of procedure has so far been formalized. With so many applications for traditional partition refinement (see, e.g., [14]), the authors are hopeful this tree refinement will find further application in the near future, especially given the breakthrough it proves to be here. Already, it and other ideas from this paper have been applied to the transitive orientation problem, with the authors confident of having achieved the first simple, linear-time algorithm for transitively orienting a graph. ## References * [1] C. Capelle, M. Habib, and F. de Montgolfier. Graph decompositions and factorizing permutations. Discrete Mathematics and Theoretical Computer Science, 5:55–70, 2002. * [2] M. Chein, M. Habib, and M.C. Maurer. Partitive hypergraphs. Discrete Mathematics, 37:35–50, 1981. * [3] D.G. Corneil, Y. Perl, and L.K. Stewart. A linear recognition algorithm for cographs. SIAM Journal of Computing, 14:926–934, 1985. * [4] A. Cournier and M. Habib. A new linear algorithm of modular decomposition. In Trees in algebra and programming (CAAP), volume 787 of Lecture Notes in Computer Science, pages 68–84, 1994. * [5] D.D. Cowan, L.O. James, and R.G. Stanton. Graph decomposition for undirected graphs. In 3rd S-E Conference on Combinatorics, Graph Theory and Computing, Utilitas Math, pages 281–290, 1972. * [6] E. Dahlhaus. Efficient parallel algorithms for cographs and distance hereditary graphs. Discrete Applied Mathematics, 57:29–54, 1995. * [7] E. Dahlhaus, J. Gustedt, and R.M. McConnell. Efficient and practical algorithm for sequential modular decomposition algorithm. Journal of Algorithms, 41(2):360–387, 2001. * [8] Celina M. H. de Figueiredo and Frédéric Maffray. Optimizing bull-free perfect graphs. SIAM J. Discret. Math., 18(2):226–240, 2005. * [9] A. Ehrenfeucht, H.N. Gabow, R.M. McConnell, and S.L. Sullivan. An ${O}(n^{2})$ divide-and-conquer algorithm for the prime tree decomposition of two-structures and modular decomposition of graphs. Journal of Algorithms, 16:283–294, 1994. * [10] J. Gagneur, R. Krause, T. Bouwmeester, and G. Casari. Modular decomposition of protein-protein interaction networks. Genome Biology, 5(8):R57, 2004. * [11] T. Gallai. Transitiv orientierbare graphen. Acta Math. Acad. Sci. Hungar., 18:25–66, 1967. * [12] M. Habib, F. de Montgolfier, and C. Paul. A simple linear-time modular decomposition algorithm for graphs, using order extension. In Scandinavian Workshop on Algorithm Theory (SWAT), volume 3111 of Lecture Notes in Computer Science, pages 187–198, 2004. * [13] M. Habib and M.C. Maurer. On the $x$-join decomposition of undirected graphs. Discrete Applied Mathematics, 1:201–207, 1979. * [14] M. Habib, R.M. McConnell, C. Paul, and L. Viennot. Lex-bfs and partition refinement, with applications to transitive orientation, interval graph recognition and consecutive ones testing. Theoretical Computer Science, 234:59–84, 2000. * [15] M. Habib, C. Paul, and L. Viennot. A synthesis on partition refinement: a useful routine for strings, graphs, boolean matrices and automata. In 15th Symposium on Theoretical Aspect of Computer Science (STACS), volume 1373 of Lecture Notes in Computer Science, pages 25–38, 1998. * [16] R.M. McConnell and J. Spinrad. Linear-time modular decomposition and efficient transitive orientation of comparability graphs. In 5th Annual ACM-SIAM Symposium on Discrete Algorithms (SODA), pages 536–545, 1994. * [17] R.M. McConnell and J. Spinrad. Ordered vertex partitioning. Discrete Mathematics and Theoretical Computer Science, 4:45–60, 2000. * [18] R.H. Mohring. Algorithmic aspects of comparability graphs and interval graphs. In I. Rival, editor, Graphs and Orders, pages 41–101. D. Reidel, Boston, 1985. * [19] R.H. Mohring. Algorithmic aspects of the substitution decomposition in optimization over relations, set systems and boolean functions. Annals of Operations Research, 4:195–225, 1985. * [20] R.H. Mohring and F.J. Radermacher. Substitution decomposition for discrete structures and connections with cominatorial optimization. Annals of Discrete Mathematics, 19:257–356, 1984. * [21] J.H. Muller and J. Spinrad. Incremental modular decomposition. Journal of the ACM, 36(1):1–19, 1989. * [22] R. Paige and R.E. Tarjan. Three partition refinement algorithms. SIAM J. Comput., 16(6):973–989, 1987. * [23] C. Papadopoulos and C. Voglis. Drawing graphs using modular decomposition. In Patrick Healy and Nikola S. Nikolov, editors, Graph Drawing, Limerick, Ireland, September 12-14, 2005, pages pp. 343–354. Springer, 2006\. * [24] A. Pnueli, S. Even, and A. Lempel. Transitive orientation of graphs and identification of permutation graphs. Canad. J. Math., 23:160–175, 1971. * [25] J. Spinrad. Efficient graph representation, volume 19 of Fields Institute Monographs. American Mathematical Society, 2003. * [26] M. Tedder, D.G. Corneil, M. Habib, and C. Paul. Simple, Linear-time Transitive Orientation. (In preparation). ## Appendix ### Omitted Proofs We prove proposition 2.3 from the overview: Proposition 2.3. _If $C$ is a co-component of $G[N_{0}]$ and $M$ is a strong module containing $x$, then either $C\subset M$ or $C\cap M=\emptyset$. Similarly for $C$ a component of $G[N_{i}],i>0$._ ###### Proof. Let $C$ be a co-component of $G[N_{0}]$ and $M$ a strong module containing $x$. Obviously, $M-C\neq\emptyset$ because of $x$. Suppose for contradiction that $C\cap M\neq\emptyset$ and $C-M\neq\emptyset$, and say $C_{1}=C\cap M$ and $C_{2}=C-C_{1}$. Since $x\in M$ and $C\subset N(x)$, we must have a join between $C_{1}$ and $C_{2}$. But then this contradicts $C$ being a co- component of $G[N_{0}]$. The case where $C$ is a component of some $G[N_{i}],i>0$ is exactly symmetric. ∎ The following fact was used in the proof of lemmas 2.4 and 2.5: ###### Proposition 5.1. If $M$ is a strong module containing $x$, and $M\cap N_{i}\neq\emptyset$ for some $i>1$, then $M=V$. ###### Proof. Let $u$ be a vertex in $M\cap N_{i}$, $i>1$. First we show that $N(x)\subset M$. Suppose for contradiction there is some $v\in N(x)-M$. Note that $v$ must be universal to $M$, which is impossible since it cannot be adjacent to $u$. Next we show $N_{1}\subset M$. Suppose for contradiction there is a $q\in N_{1}-M$. Note that $q$ must be isolated from $M$. But this is impossible as $q\in N_{1}$ and therefore has at least one neighbour in $N(x)\subset M$. Thus $N_{1}\subset M$. We simply need to progressively apply a symmetric argument to show that $N_{2}\subset M,N_{3}\subset M,\ldots$. ∎ We now prove lemma 3.2: Lemma 3.2. _Let $M$ be the smallest strong module containing $x$. Then $M$ satisfies one of the following three conditions:_ 1. _(i)_ _$M$ is the maximally contiguous module containing $x$ and no $C^{\prime}_{i}$ (in which case $M$ is series);_ 2. _(ii)_ _$M$ is the maximally contiguous module containing $x$ and no $C_{i}$, and only $C^{\prime}_{j}$’s in $N_{1}$ with no edge to their right (in which case $M$ is parallel);_ 3. _(iii)_ _$M$ is the minimally contiguous module containing $x$ and at least $C_{1}$ and $C_{1}^{\prime}$ (in which case $M$ is prime)._ ###### Proof. By proposition 2.3, each (co-)component is either entirely contained in $M$ or entirely outside $M$. Hence, we need only consider modules including $x$ and formed by including or excluding these (co-)components. Of course, the (co)-components constituting $M$ must be contiguous because our ordering represents a factorizing permutation. We first show that if $M$ is series, it cannot contain any $C^{\prime}_{i}$. So assume for contradiction that $M$ contains a $C^{\prime}_{i}$. We can therefore assume that $M$ contains $C^{\prime}_{1}$, since our ordering is a factorizing permutation. However, because $M$ is the smallest strong module containing $x$, and $M$ is series, we must have $x$ as a co-component in the graph induced on $M$. In other words, $x$ must be universal to the vertices in the graph induced on $M$. With $C^{\prime}_{1}$ in $M$, this is of course impossible. So if $M$ is series, it can contain no $C^{\prime}_{i}$. We now must show that $M$ must be the maximally contiguous module with this property. Let $M^{\prime}$ be the maximally contiguous module containing no $C^{\prime}_{i}$. Note that by the maximality of $M^{\prime}$ and $M$ being strong, we must have $M^{\prime}\supseteq M$. Assume for contradiction that $M^{\prime}\neq M$. Then there must be a $C_{i}\in M^{\prime}-M$. Of course, we must also have some $C_{j}\in M$. Observe that $C_{i}\cup C_{j}$ is a module. Moreover, it is a module that overlaps $M$ (recall that $x\in M$ as well), contradicting $M$ being strong. Thus, we must have $M=M^{\prime}$, and therefore $M$ is the maximally contiguous module not containing any $C^{\prime}_{i}$. Now consider the case where $M$ is parallel. First we show that $M$ cannot contain any $C^{\prime}_{i}$ in a $N_{j},j>1$. In this case, proposition 5.1 says that $M=V$. If $M$ is to be parallel the graph induced on it must be disconnected, which is impossible since we assumed in section 2.1 that the graphs in this paper were all connected. To show that $M$ cannot contain any $C^{\prime}_{i}$ from $N_{1}$ that contains an edge to its right, say to $v\in N_{j},j>1$, assume otherwise for contradiction. We must then have $v\in M$ since it is adjacent to some vertex in $C^{\prime}_{i}$ but not to $x$. But then the component of $v$ must be added to $M$, by proposition 2.3, which we just saw is impossible. We next need to show that when $M$ is parallel it is the maximally contiguous module containing only components of $G[N_{1}]$ without edges to their right. We saw above that it can only contain components of $G[N_{1}]$ without edges to their right, and earlier that it must be contiguous, so we need only show that it is maximally so. Let $M^{\prime}$ be the maximally contiguous module only containing $C^{\prime}_{i}$’s in $N_{1}$ with no edge to their right. Observe that since $M$ is strong and $M^{\prime}$ maximal, we must have $M^{\prime}\supseteq M$. Assume for contradiction that $M^{\prime}\neq M$. Thus, there is a $C^{\prime}_{i}\in M^{\prime}-M$, and also a $C^{\prime}_{j}\in M$. Notice that $C^{\prime}_{i}\cup C^{\prime}_{j}$ is a module that overlaps $M$ (recall that $x\in M$ as well), contradicting it being a strong module. So now assume $M$ is prime. In this case the graph induced by $M$ cannot be disconnected, nor can its complement be disconnected. As such, $M$ cannot consist entirely of $C^{\prime}_{i}$’s, nor can it consist entirely of $C_{i}$’s. Because this is a factorizing permutation, we must then have $C_{1}\subset M$ and $C^{\prime}_{1}\subset M$. Hence, we need only show that $M$ is the minimally contiguous module containing $C_{1}$ and $C^{\prime}_{1}$. Let $M^{\prime}$ be the minimally contiguous module containing $C_{1}$ and $C^{\prime}_{1}$. Since $M$ is strong and $M^{\prime}$ minimal, we must have $M^{\prime}\subseteq M$. Assume for contradiction that $M^{\prime}\neq M$. Recall the theorem of [19] employed earlier, saying that when a module is not strong, it is the union of (strong) siblings in the MD tree. These siblings must be descendants of $M$ in the MD tree. Moreover, $x$ must be a descendant of one of these siblings. But if $M$ is the smallest strong module containing $x$, we must have $M$ as the parent of $x$ in the MD tree, which gives us the desired contradiction. ∎ ### An Example A graph $G$ is described in figure 1 by the modular decomposition tree pictured therein. In it, prime nodes are labeled by the graph their children induce, while series nodes are labeled by $1$ and parallel nodes by $0$, following the cograph convention. We demonstrate how our algorithm operates when input $G$. Assume $x$ is the vertex chosen to start the algorithm. In this case, $N(x)=N_{0}=\\{c,d,e,a\\}$, $N_{1}=\\{f,g,h,i,b,j,k,\ell,m,n,p,q\\}$, and $N_{2}=\\{r\\}$. Figure 2 displays the modular decomposition trees recursively computed: $T(N_{0}),T(N_{1})$, and $T(N_{2})$. We use $\alpha(u)$ to denote the list of incident active edges of the vertex $u$. The active edges in our example are summarized in table 1. Using the active edges, the algorithm refines each tree in the forest; the result is displayed in figure 3. The shading in the diagram corresponds to the marks on the nodes: horizontal shading for “left” marks, vertical shading for “right ”marks, and cross-hatched shading when a node has both “left” and “right” marks. Promotion is applied to these marked nodes, with the result being figure 4. Read the trees of figure 4 from left-to-right and label them $T_{2},T_{1},T^{\prime}_{1},T^{\prime}_{2},T^{\prime}_{3},T^{\prime}_{4},T^{\prime}_{5},T^{\prime}_{6}$. We are interested in the following ordered list of trees: $T_{2},T_{1},x,T^{\prime}_{1},T^{\prime}_{2},T^{\prime}_{3},T^{\prime}_{4},T^{\prime}_{5},T^{\prime}_{6}$ (2) We now rephrase this list in terms of the (co-)components of the $G[N_{i}]$’s as we described in the paper. From figure 2, we see that $G[N_{0}]$ is series, and has co-components $\\{a\\}$ and $\\{c,d,e\\}$. We also see that $G[N_{1}]$ is parallel, and has the components $\\{b\\},\\{j\\},\\{i,g,h,f\\},\\{k,\ell\\}$, and $\\{q,m,n,p\\}$. Lastly, the figure tells us that $\\{r\\}$ is the only component of $G[N_{2}]$. Observe from figure 4 that each of these (co-)components appears consecutively after promotion. Reading these from left-to-right we can view the list of (2) as: $C_{2},C_{1},x,C^{\prime}_{1},C^{\prime}_{2},C^{\prime}_{3},C^{\prime}_{4},C^{\prime}_{5},C^{\prime}_{6},$ where $C_{2}=\\{a\\},C_{1}=\\{c,d,e\\},C^{\prime}_{1}=\\{g,h,f,i\\},C^{\prime}_{2}=\\{b\\},C^{\prime}_{3}=\\{k,\ell\\},C^{\prime}_{4}=\\{j\\},C^{\prime}_{5}=\\{q,m,n,p\\}$, and $C^{\prime}_{6}=\\{r\\}$. The algorithm must now insert brackets between these (co-)components in such a way as to delineate the strong modules containing $x$. To do this we first determine which of the $C^{\prime}_{i}$’s have edges to their right. In this case only $C^{\prime}_{5}=\\{q,m,n,p\\}$ does, by virtue of $q$ being adjacent to $r$. Next we must calculate the $\mu$-values for each (co-)component, as was described in the paper. These values are summarized in table 2. We can now proceed to introduce the brackets. The first set of brackets will correspond to the smallest strong module containing $x$. We first try a series module by comparing $\mu(C_{1})$ with $x$. Notice that they are not equal, so a series module cannot be formed. Next, a parallel module is attempted by comparing $\mu(C^{\prime}_{1})$ with $x$. Once more, they are not equal, and so a parallel module cannot be formed. We now know that $M$ must be prime and include $C_{1}$ and $C^{\prime}_{1}$. Therefore, $C^{\prime}_{1},\ldots,\mu(C_{1})=C^{\prime}_{1}$ must be included as well (it already is), and so must be $C_{1},\ldots,\mu(C^{\prime}_{1})=C_{1}$ as well (it already is). Thus, $C_{1}\cup\\{x\\}\cup C^{\prime}_{1}$ represents the minimal contiguous module containing $x$ and $C_{1}$ and $C^{\prime}_{1}$. We have therefore found the smallest strong module containing $x$. We bracket it accordingly and move on: $C_{2},[C_{1},x,C^{\prime}_{1}],C^{\prime}_{2},C^{\prime}_{3},C^{\prime}_{4},C^{\prime}_{5},C^{\prime}_{6}.$ We once more try for a series module, but this time compare $\mu(C_{2})$ with $C^{\prime}_{1}$ (the last component in the previous module). These are not equal so a series module cannot be formed. So we try for a parallel module, comparing $\mu(C^{\prime}_{2})$ with $C_{1}$ (the last co-component in the previous module). Here they are equal, so perhaps a parallel module can be formed. We must also check that $C^{\prime}_{2}$ does not have an edge to its right, which it doesn’t (recall that only $C^{\prime}_{5}$ does), and so a parallel module can in fact be formed. We now maximally add components in the same way. Doing so allows us to add $C^{\prime}_{3}$ and $C^{\prime}_{4}$. Although, $\mu(C^{\prime}_{5})$ also equals $C_{1}$, it cannot be added because it has an edge to its right. We bracket this module accordingly and move on: $C_{2},[[C_{1},x,C^{\prime}_{1}],C^{\prime}_{2},C^{\prime}_{3},C^{\prime}_{4}],C^{\prime}_{5},C^{\prime}_{6}.$ Again, we first try for a series module. Here $\mu(C_{2})$ does not equal $C^{\prime}_{4}$ so none can be formed. No parallel module can be formed because $C^{\prime}_{5}$ has an edge to its right. Thus, the module is prime and must include both $C_{2}$ and $C^{\prime}_{5}$. It thus also includes $C^{\prime}_{1},\ldots,\mu(C_{2})=C^{\prime}_{6}$, and so we know this module corresponds to the entire graph. We create the necessary brackets: $[C_{2},[[C_{1},x,C^{\prime}_{1}],C^{\prime}_{2},C^{\prime}_{3},C^{\prime}_{4}],C^{\prime}_{5},C^{\prime}_{6}].$ Based on the above bracketing we can now construct the tree according to the procedure outlined in the paper, the result of which is clearly the tree of figure 1. $e$$1$$k$$0$$1$$m$$n$$p$$q$$r$$a$$0$$i$$b$$j$$\ell$$0$$1$$1$$d$$x$$g$$h$$f$$c$ Figure 1: The modular decomposition tree for a graph $G$. Prime nodes are labeled by the graphs their children induce. Series nodes are labeled by $1$ while parallel nodes are labeled by $0$, as per the cograph convention. $c$$1$$1$$a$$e$$d$$0$ $b$$i$$g$$h$$j$$m$$n$$p$$1$$0$$1$$1$$0$$k$$\ell$$f$$q$ $r$ Figure 2: The recursively computed modular decomposition trees when $x$ is chosen to start the algorithm on graph $G$: (a) $T(N_{0})$; (b) $T(N_{1})$; (c) $T(N_{2})$. vertex $u$ | $\alpha(u)$ ---|--- $a$ | $x,b,j,f,g,h,i,k,\ell,m,n,p,q$ $c,d,e$ | $x,i$ $b,j,f,g,h,k,\ell,m,n,p$ | $a$ $i$ | $a,c,d,e$ $q$ | $a,r$ $r$ | $q$ Table 1: The active edges for the graph $G$ after $x$ is chosen to start the algorithm. $0$$1$$e$$d$$c$$a$ $i$$j$$0$$1$$k$$\ell$$b$$g$$h$$f$$1$$1$$n$$p$$m$$0$$1$$1$$q$ $r$ Figure 3: The trees of figure 2 after the active edges (table 1) have refined them. Horizontal shading represents a node marked by a left split; vertical shading represents a node marked by a right split; cross-hatched shading represents a node marked by both left and right splits. $a$$1$$e$$d$$0$$c$ $q$$g$$h$$1$$f$$0$$1$$k$$\ell$$b$$j$$n$$p$$m$$0$$1$$i$ $r$ Figure 4: The trees of figure 3 after promotion. (Co-)Component | $\mu(C)$ ---|--- $C_{2}$ | $C^{\prime}_{5}$ $C_{1}$ | $C^{\prime}_{1}$ $C^{\prime}_{1}$ | $C_{1}$ $C^{\prime}_{2}$ | $C_{1}$ $C^{\prime}_{3}$ | $C_{1}$ $C^{\prime}_{4}$ | $C_{1}$ $C^{\prime}_{5}$ | $C_{1}$ $C^{\prime}_{6}$ | $C_{2}$ Table 2: The $\mu$-values.

arxiv-papers

2024-09-04T02:48:53.255227

0710.4061

# How to quantify quantum entanglement degree Yong Zhou China University of Petroleum, College of Physics Science and Technology, 739 Bei’yi Road, Dongying Shandong 257061, China ###### Abstract Since the density matrix possesses the essential property of linear decomposition and linear superposition of quantum states, it only can be used to describe quantum states. So the entanglement contained in a density matrix is just the quantum entanglement. This means a bipartite state described by a density matrix contains quantum entanglement, unless the density matrix can be expressed by a direct product of the subsystem’s density matrices $\rho_{AB}=\rho_{A}\otimes\rho_{B}$. Thus the quantum entanglement degree between A and B systems can be determined by the expanding coefficients according to A or B system’s orthonormal bases of the density matrix $\rho_{AB}$. ###### pacs: 03.67.Mn, 03.65.Ud Quantum entanglement is a very important phenomenon since it is proposed by Einstein, Podolsky and Rosen c1 , Schrodinger c2 and Bell c3 . The nonlocality and the coherence of quantum entangled systems makes them have important applications in quantum information, such as in quantum teleportation c4 (for experimental realization see c5 ), quantum cryptographic schemes c6 , dense coding c7 and quantum parallel computation c8 . The quantifying of the quantum entanglement degree of bipartite pure states has been well done c9 , but how to define and quantify other complex systems’ quantum entanglement degree is still a very hard problem c10 . Since any bipartite quantum system has a corresponding (reduced) density matrix to describe it, we can study the bipartite quantum entanglement degree through its density matrix. It’s well known that the density matrix is composed of the state vectors of linear Hilbert space, so the density matrix has the property of linear decomposition and linear superposition which is the unique property of quantum states, therefore a density matrix cannot be used to describe a classical state, it only can be used to describe quantum state. Speaking concretely, there isn’t quantum correlation between classical states, so a superposition of several classical states cannot create new state through linear decomposition and recombination. On the contrary, if a system is described by the density matrix $\rho=\sum_{i}p_{i}|\psi_{i}\rangle\langle\psi_{i}|$, the system’s state isn’t limited to the range $\\{|\psi_{i}\rangle\\}$. The measuring result of the system can be any linear combination state $\psi^{{}^{\prime}}$ of the states $\\{|\psi_{i}\rangle\\}$ which makes $tr(\rho|\psi^{{}^{\prime}}\rangle\langle\psi^{{}^{\prime}}|)$ nonzero. This is because the states $\\{|\psi_{i}\rangle\\}$ in the density matrix $\rho$ can be freely linearly decomposed and recombined to create new states. This means the states in a density matrix are quantum entangled together. So the entanglement present in a density matrix is just the quantum entanglement. This means that only the density matrix which can be expressed by a direct product of the subsystem’s density matrices $\rho_{AB}=\rho_{A}\otimes\rho_{B}$ doesn’t contain quantum entanglement. Any other density matrix, including the following form c11 : $\rho_{AB}\,=\,\sum_{i}p_{i}\,\rho^{A}_{i}\otimes\rho^{B}_{i}\,,$ (1) where $p_{i}$ satisfy $\sum_{i}p_{i}=1$ and $\rho^{A}_{i}$ and $\rho^{B}_{i}$ are the density matrices of the two subsystems, contains quantum entanglement. Obviously this conclusion contradicts the point of the scientists who considered the entanglement in Eq.(1) is classical entanglement c10 ; c11 ; c12 . As we have carefully discussed above, the state in Eq.(1) is a quantum state in which the states $\\{\rho^{A}_{i}\otimes\rho^{B}_{i}\\}$ get entangled with each other through quantum superposition, thus has quantum entanglement. Besides this point there are two other evidences to prove our conclusion. Firstly consider the three systems: A, B and C, if A gets quantum entangled with BC system group through a pure state: $|\psi_{ABC}\rangle\,=\,\sum_{i}\sqrt{p_{i}}|A_{i}B_{i}C_{i}\rangle\,,$ (2) where $\\{|A_{i}\rangle\\}$, $\\{|B_{i}\rangle\\}$ and $\\{|C_{i}\rangle\\}$ are the orthonormal bases of the corresponding systems, the density matrices describing A and B entanglement and A and C entanglement are: $\rho_{AB}\,=\,\sum_{i}p_{i}|A_{i}B_{i}\rangle\langle A_{i}B_{i}|\,,\hskip 28.45274pt\rho_{AC}\,=\,\sum_{i}p_{i}|A_{i}C_{i}\rangle\langle A_{i}C_{i}|\,.$ (3) If the point of the references c10 ; c11 ; c12 is right, there isn’t quantum entanglement between A and B systems and A and C systems. But this conclusion contradicts a logical fact: if A gets entangled with BC system group, A must get entangled with at least one of B and C systems, otherwise A will get not entangled with BC system group. It is a logical fact that speaks for itself. Obviously only accepting the point that Eq.(1) contains quantum entanglement can resolve this question. Another evidence can be found in quantum teleportation. Suppose B system is far away from A system and A and C systems are in a same location, and A, B and C systems are in the following quantum states: $\rho_{AB}\,=\,\frac{1}{2}(|A_{1}B_{1}\rangle\langle A_{1}B_{1}|+|A_{2}B_{2}\rangle\langle A_{2}B_{2}|)\,,\hskip 28.45274pt|\psi_{C}\rangle\,=\,c_{1}|C_{1}\rangle+c_{2}|C_{2}\rangle\,.$ (4) Perform an union measurement of A and C systems to obtain one of the following possible states: $\displaystyle|\psi_{AC1}\rangle\,=\,\frac{1}{\sqrt{2}}(|A_{1}C_{1}\rangle+|A_{2}C_{2}\rangle)\,,\hskip 28.45274pt|\psi_{AC2}\rangle\,=\,\frac{1}{\sqrt{2}}(|A_{1}C_{1}\rangle-|A_{2}C_{2}\rangle)\,,$ $\displaystyle|\psi_{AC3}\rangle\,=\,\frac{1}{\sqrt{2}}(|A_{1}C_{2}\rangle+|A_{2}C_{1}\rangle)\,,\hskip 28.45274pt|\psi_{AC4}\rangle\,=\,\frac{1}{\sqrt{2}}(|A_{1}C_{2}\rangle-|A_{2}C_{1}\rangle)\,,$ (5) accordingly B system will change its state to one of the following possible states: $\displaystyle\rho_{B1}\,=\,|c_{1}|^{2}|B_{1}\rangle\langle B_{1}|+|c_{2}|^{2}|B_{2}\rangle\langle B_{2}|\,,\hskip 28.45274pt\rho_{B2}\,=\,\rho_{B1}\,,$ $\displaystyle\rho_{B3}\,=\,|c_{2}|^{2}|B_{1}\rangle\langle B_{1}|+|c_{1}|^{2}|B_{2}\rangle\langle B_{2}|\,,\hskip 28.45274pt\rho_{B4}\,=\,\rho_{B3}\,.$ (6) From Eqs.(6) we find B system is still in quantum state, and B system state contains part of the information of C system state–the module squares of the expanding coefficients of C system state. On the other hand, if use the well accepted quantum entangled state $|\psi_{AB}\rangle\,=\,\frac{1}{\sqrt{2}}(|A_{1}B_{1}\rangle+|A_{2}B_{2}\rangle)\,,$ (7) to replace the AB state in Eqs.(4) to perform the above experiment, we will obtain one of the following possible B system states according to the possible measuring results of Eqs.(5): $\displaystyle\rho_{B1}\,=\,|c_{1}|^{2}|B_{1}\rangle\langle B_{1}|+c_{1}c_{2}^{*}|B_{1}\rangle\langle B_{2}|+c_{2}c_{1}^{*}|B_{2}\rangle\langle B_{1}|+|c_{2}|^{2}|B_{2}\rangle\langle B_{2}|\,,$ $\displaystyle\rho_{B2}\,=\,|c_{1}|^{2}|B_{1}\rangle\langle B_{1}|-c_{1}c_{2}^{*}|B_{1}\rangle\langle B_{2}|-c_{2}c_{1}^{*}|B_{2}\rangle\langle B_{1}|+|c_{2}|^{2}|B_{2}\rangle\langle B_{2}|\,,$ $\displaystyle\rho_{B3}\,=\,|c_{2}|^{2}|B_{1}\rangle\langle B_{1}|+c_{2}c_{1}^{*}|B_{1}\rangle\langle B_{2}|+c_{1}c_{2}^{*}|B_{2}\rangle\langle B_{1}|+|c_{1}|^{2}|B_{2}\rangle\langle B_{2}|\,,$ $\displaystyle\rho_{B4}\,=\,|c_{2}|^{2}|B_{1}\rangle\langle B_{1}|-c_{2}c_{1}^{*}|B_{1}\rangle\langle B_{2}|-c_{1}c_{2}^{*}|B_{2}\rangle\langle B_{1}|+|c_{1}|^{2}|B_{2}\rangle\langle B_{2}|\,.$ (8) Compare this result with Eqs.(6) we find the B system states in Eqs.(8) contain more information of C system state–the relative phase of the expanding coefficients $c_{1}$ and $c_{2}$. Therefore the quantum information channel $|\psi_{AB}\rangle\langle\psi_{AB}|$ in Eq.(7) is larger than the quantum information channel $\rho_{AB}$ in Eqs.(4). However we cannot conclude the information channel $\rho_{AB}$ in Eqs.(4) doesn’t have the ability to transport quantum information. Since $\rho_{AB}$ can transport quantum information, it must contain a certain kind of quantum entanglement between A and B systems. From the above discussion it is obvious the quantum entanglement degree in Eq.(7) is larger than that in Eqs.(4). Compare the two equations it is believed that the increase of the quantum entanglement degree of Eq.(7) comes from the two more terms $\frac{1}{2}|A_{1}\rangle\langle A_{2}|B_{1}\rangle\langle B_{2}|+\frac{1}{2}|A_{2}\rangle\langle A_{1}|B_{2}\rangle\langle B_{1}|$ in the density matrix of Eq.(7) and the states of A and B systems in the two more terms are different from those in the other terms. In general it can be concluded that the quantum entanglement degree of A and B systems can be determined by the expanding coefficients according to A or B system’s orthonormal bases of the density matrix $\rho_{AB}$. If A system’s orthonormal bases are $\\{|A_{i}\rangle,i=1...n\\}$, and B system’s orthonormal bases are $\\{|B_{i}\rangle,i=1...m\\}$, expand the density matrix $\rho_{AB}$ according to A system’s orthonormal bases: $\rho_{AB}\,=\,\sum_{i,i^{\prime}=1}^{n}\rho_{ii^{\prime}}|A_{i}\rangle\langle A_{i^{\prime}}|\,,$ (9) where the expanding coefficients $\\{\rho_{ii^{\prime}}\\}$ can be written as follows: $\rho_{ii^{\prime}}\,=\,\sum_{j,j^{\prime}=1}^{m}p_{ii^{\prime},jj^{\prime}}|B_{j}\rangle\langle B_{j^{\prime}}|\,.$ (10) If all of $\\{\rho_{ii^{\prime}}\\}$ are same, A and B systems will not get quantum entangled with each other, so the more different the different $\\{\rho_{ii^{\prime}}\\}$ are, the more large the quantum entanglement degree between A and B systems is. The difference between different $\\{\rho_{ii^{\prime}}\\}$ can be determined by the following quantity $X_{i_{1}i_{1}^{\prime},i_{2}i_{2}^{\prime}}\,=\,tr(\rho_{i_{1}i_{1}^{\prime}}\rho_{i_{2}i_{2}^{\prime}}^{\dagger})\,=\,\sum_{j,j^{\prime}}^{m}p_{i_{1}i_{1}^{\prime},jj^{\prime}}p^{*}_{i_{2}i_{2}^{\prime},jj^{\prime}}\,.$ (11) The matrix $X$ is a Hermitian matrix, so it can be diagonalized by an unitary matrix. This is equivalent to linear recombining the B system’s states $\\{\rho_{ii^{\prime}}\\}$ to a set of orthonormal bases. The rank of matrix $X$, i.e. the number of nonzero diagonal elements of the diagonalized matrix $X$, is just the number of the orthonormal bases of the B system’s states $\\{\rho_{ii^{\prime}}\\}$. Obviously the more large the rank of matrix $X$ is, the more large the quantum entanglement degree between A and B systems is. But it is difficult to use matrix $X$ to quantify quantum entanglement degree, since the values of the matrix elements of $X$ reflect not only the probability of the different states $\\{\rho_{ii^{\prime}}\\}$ in the density matrix $\rho_{AB}$ and the difference between the different states $\\{\rho_{ii^{\prime}}\\}$, but also the difference between pure states and mixed states (the trace of the square of pure state density matrix is 1, but the trace of the square of mixed state density matrix is less than 1), and the latter is nothing to do with the quantum entanglement degree of A and B systems. Up to the present we still haven’t found a suitable method to quantify the quantum entanglement degree of a general quantum state. In general, we discuss how to determine the quantum entanglement degree of a general quantum state described by a density matrix. We clarify the important facts that density matrices can only be used to describe quantum states and the entanglement contained in density matrices is just quantum entanglement. Under this point most of the quantum separable states people regarded as before, such as the bipartite state in Eq.(1), are in fact quantum entangled states, unless the density matrix of it can be expressed by a direct product of the subsystem’s density matrices. So the quantum entanglement degree of A and B systems can be determined through the expanding coefficients according to A or B system’s orthonormal bases of the density matrix $\rho_{AB}$ (see Eqs.(9, 10)). Obviously the different expanding coefficients $\\{\rho_{ii^{\prime}}\\}$ are more different, the quantum entanglement degree is more large. The difference between different expanding coefficients can be quantified by the Hermitian matrix $X$ in Eq.(11). Unfortunately, it is difficult to use matrix $X$ to quantify quantum entanglement degree, since the values of the matrix elements of $X$ reflect not only the probability of the different states $\\{\rho_{ii^{\prime}}\\}$ in the density matrix $\rho_{AB}$ and the difference between the different states $\\{\rho_{ii^{\prime}}\\}$, but also the difference between pure states and mixed states, and the latter is nothing to do with the quantum entanglement degree of A and B systems. Therefore a method of quantifying the quantum entanglement degree of a general quantum state is absent up to the present. However we should realize that the physics is a practical science and a mathematic definition without practical application isn’t the thing we need, so it is always not too late for us to quantify quantum entanglement degree until we find the practical application of it. Acknowledgments The author thanks Prof. Hong-bo Zhu and Doctor Pei-lin Lang for the useful discussions with them. ## References * (1) A. Einstein, B. Podolsky and N. Rosen, Phys. Rev. 47 (1935) 777. * (2) E. Schrodinger, Naturwissenschaften 23 (1935) 807. * (3) J.S. Bell, Physics (N.Y.) 1 (1964) 195. * (4) C.H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and W.K. Wootters, Phys. Rev. Lett. 70 (1993) 1895; S. Albeverio and S.M. Fei, Phys. Lett. A 276 (2000) 8; G.M. D Ariano, P.Lo Presti, M.F. Sacchi, Phys. Lett. A 272 (2000) 32; S. Albeverio and S.M. Fei and W.L. Yang, Phys. Rev. A 66 (2002) 012301. * (5) D. Bouwmeester, J.-W. Pan, K. Mattle, M. Elbl, H. Weinfurter and A. Zeilinger, Nature (London) 390 (1997) 575; M. A. Nielsen, E. Knill and R. Laflamme, Nature 396 (1998) 52; Jian-Wei Pan, Matthew Daniell, Sara Gasparoni, Gregor Weihs, Anton Zeilinger, Phys. Rev. Lett. 86 (2001) 4435; Thomas Jennewein, Gregor Weihs, Jian-Wei Pan, Anton Zeilinger, Phys. Rev. Lett. 88 (2001) 017903; Qiang Zhang, Alexander Goebel, Claudia Wagenknecht, Yu-Ao Chen, Bo Zhao, Tao Yang, Alois Mair, Joerg Schmiedmayer, Jian-Wei Pan, Nature Physics 2 (2006) 678. * (6) A. K. Ekert, Phys. Rev. Lett. 67 (1991) 661; D. Deutsch, A. K. Ekert, R. Jozsa, C. Macchiavello, S. Popescu and A. Sanpera, Phys. Rev. Lett. 77 (1996) 2818; C.A. Fuchs, N. Gisin, R.B. Griffiths, C-S. Niu, and A. Peres, Phys. Rev. A 56 (1997) 1163. * (7) C.H. Bennett and S.J. Wiesner, Phys. Rev. Lett. 69 (1992) 2881. * (8) D.P. DiVincenzo, Science 270 (1995) 255; M.A. Nielsen and I.L. Chuang, Quantum Computation and Quantum Information, Cambridge University Press, Cambridge, 2000. * (9) A. Peres, Quantum Theory: Concepts and Methods , Kluwer Academic Publishers (1995). * (10) C. H. Bennett, D. P. DiVincenzo, J. A. Smolin, and W. K. Wootters, Phys. Rev. A 54 (1996) 3824; V. Vedral, M.B. Plenio, M.A. Rippin, P. L. Knight, Phys. Rev. Lett. 78 (1997) 2275; C. H. Bennett, D. P. DiVincenzo, T. Mor, P. W. Shor, J. A. Smolin and B. M. Terhal, Phys. Rev. Lett. 82 (1999) 5385; W. Dur, J. I. Cirac and R. Tarrach, Phys. Rev. Lett. 83 (1999) 3562; P. Horodecki, M. Lewenstein, G. Vidal and I. Cirac, Phys. Rev. A 62 (2000) 032310; S. Karnas and M. Lewenstein, Phys. Rev. A 64 (2001) 042313; K. Chen and L.A. Wu, Phys. Lett. A 306 (2002) 14; S.M. Fei, J. Jost, X.Q. Li-Jost and G.F. Wang, Phys. Lett. A 310 (2003) 333; O. Rudolph, Phys. Rev. A 67 (2003) 032312. * (11) R. F. Werner, Phys. Rev. A 40 (1989) 4277. * (12) Asher Peres, Phys. Rev. Lett. 77 (1996) 1413; M. Horodecki, P. Horodecki and R. Horodecki, Springer Tracts in Mod. Phy. 173 (2001) 151; B.M. Terhal, Theor. Comput. Sci. 287 (2002) 313; K. Chen, S. Albeverio and S.M. Fei, Phys. Rev. Lett. 95 (2005) 210501.

arxiv-papers

2024-09-04T02:48:53.263261

0710.5646

# On the Hopf Algebra of Rooted Trees Shouchuan Zhang, Jieqiong He, Peng Wang Department of Mathematics, Hunan University Changsha 410082, P.R. China ###### Abstract We find a formula to compute the number of the generators, which generate the $n$-filtered space of Hopf algebra of rooted trees, i.e. the number of equivalent classes of rooted trees with weight $n$. Applying Hopf algebra of rooted trees, we show that the analogue of Andruskiewitsch and Schneider’s Conjecture is not true. The Hopf algebra of rooted trees and the enveloping algebra of the Lie algebra of rooted trees are two important examples of Hopf algebras. We give their representation and show that they have not any nonzero integrals. We structure their graded Drinfeld doubles and show that they are local quasitriangular Hopf algebras. Mathematics Subject Classification: 16W30, Keywords: Rooted tree, Hopf algebra, Representation. ## 0 Introduction In [CK, BK, Kr1, Kr2], the Hopf algebra of rooted trees ${\cal H}_{R}$ was introduced for renormalization theory. Paper [Fo] classified the finite dimensional comodules over ${\cal H}_{R}$. It is well-known that Hopf algebra ${\cal H}_{R}$ of rooted trees is the algebra of polynomials over $\mathbb{Q}$, whose indeterminate elements are equivalent classes of rooted trees. That is, ${\cal H}_{R}=\mathbb{Q}[{\cal R}T]$ as algebras, where ${\cal R}T$ is the set of equivalent classes of all rooted trees (see section 1). Therefore, it is necessary to find a formula to compute the number of equivalent classes of rooted trees with weight $n$. Fortunately, we complete this in this paper. The Hopf algebra of rooted trees and the enveloping algebra of the Lie algebra of rooted trees are two important examples of Hopf algebras. We give their representation by means of path algebras. In particular, primitive elements of ${\cal H}_{R}$ are extraordinary. In this paper we use this special structure to show that the analogue of Andruskiewitsch and Schneider’s Conjecture [AS2, Conjecture 1.4] is not true. That is, there is an infinite dimensional pointed Hopf algebra, which is not generated by its coradical and its skew-primitive elements. That is, this conjecture is not true when the condition which $H$ is finite dimensional is omitted. Paper [ZGZ] pointed out that we can systematically structure the solutions of Yang-Baxter equations by means of local quasitriangular Hopf algebras. In this paper we structure the graded Drinfeld double of Hopf algebra of rooted trees and show that it is a local quasitriangular Hopf algebra. ## Preliminaries We will use notations of [CK] and [Fo]. We call rooted tree t a connected and simply-connected finite set of oriented edges and vertices such that there is one distinguished vertex with no incoming edge; this vertex is called the root of $t$. The $weight$ of $t$ is the number of its vertices. The fertility of a vertex $v$ of a tree $t$ is the number of edges outgoing from $v$. A $ladder$ is a rooted tree such that every vertex has fertility less than or equal to 1. There is a unique ladder of weight $i$; we denote it by $l_{i}$. Let $\mathbb{Z}$, $\mathbb{Z}^{+}$ and $\mathbb{N}$ denote the sets of all integers, all positive integers and all non-negative integers, respectively. $k=\mathbb{Q}$ denotes rational number field. Let ${\cal R}T$ denote the set of equivalent classes of all rooted trees (see section 1). $H_{\cal R}$ and ${\cal L}^{1}$ denote the Hopf algebra and Lie algebra in [CK] and [Fo, Section 2], called the Hopf algebra of rooted tree and the Lie algebra of rooted tree, respectively. $H_{ladd}$ denotes the Hopf subalgebra of $H_{\cal R}$ generated by all ladders $l_{i}^{\prime}s.$ We define the algebra $H_{\cal R}$ as the algebra of polynomials over $\mathbb{Q}$ in ${\cal R}T$. The monomials of $H_{\cal R}$ will be called $forests$. It is often useful to think of the unit $1$ of $H_{\cal R}$ as an empty forest. Let $M,N$ be two forests of ${\cal H}_{R}$. We define: $M\top N=\left\\{\begin{array}[]{cc}\frac{1}{weight(N)}\sum\mbox{forests obtained by appending M to every node of N }&\mbox{if }N\neq 1\\\ 0&\mbox{if }N=1.\end{array}\right.$ We can extend $.\top.$ to a bilinear map from ${\cal H}_{R}\times{\cal H}_{R}$ into ${\cal H}_{R}$. ## 1 The number of rooted trees In this section we give a formula to compute the number of equivalent classes of rooted trees with weight $n$. A sub-tree $u$ of rooted tree $t$ is said to be of the height $r$ if the distant from root of $t$ to root of $u$ is $r$, i.e. there exist exactly $r$ arrows from root of $t$ to root of $u$. Assume that $u$ and $u^{\prime}$ are two sub-trees of rooted tree $t$ with the same height. If we change the position of $u$ and $u^{\prime}$ in $t$, we get a new rooted tree $t^{\prime}$, said that $t^{\prime}$ is obtained by a elementary transformation from $t$. Two rooted trees $t$ and $t^{\prime}$ are called equivalent if $t^{\prime}$ can be obtained by finite elementary transformations from $t$, written as $t\sim t^{\prime}$. Note that $t\sim t$ for any rooted tree $t$ since we can choice $u=u^{\prime}$. For example, $\sim$,$\sim$ Figure 1: the first and second rooted trees are equivalent; the third and fourth rooted trees are equivalent. Obviously, $``\sim"$ is an equivalent relation in all rooted trees. Let ${\cal R}T$ be the set of equivalent classes of all rooted trees and ${\cal R}T_{n}$ be the set of equivalent classes of all rooted trees with weight $n$. Define $a(n):={\rm card}\ {\cal R}T_{n}$, i.e. the number of elements in ${\cal R}T_{n}$. ###### Theorem 1.1. $\displaystyle a(n+1)=\sum_{1\lambda_{1}+2\lambda_{2}+\cdots+n\lambda_{n}=n}C^{\lambda_{1}}_{a(1)+\lambda_{1}-1}C^{\lambda_{2}}_{a(2)+\lambda_{2}-1}\cdots C^{\lambda_{n}}_{a(n)+\lambda_{n}-1}.$ (1) Proof. Let $t$ be a rooted tree with weight $n+1$. Assume that there exists $\lambda_{i}$ subtrees of $t$ with height $1$ and weight $i$ for $i=1,2,\cdots,n$. Then $1\lambda_{1}+2\lambda_{2}+\cdots+n\lambda_{n}=n$ and we have the formula (1) by combination theory. $\Box$ Remark: The indeterminate elements of algebra ${\cal H}_{R}$ of polynomials are the equivalent classes of rooted trees, instead of rooted trees. Otherwise, $B_{+}$ in [CK, map (45)] is not a map. ## 2 The representation of the Hopf algebra and the Lie algebra of rooted trees In this section we give the representation of the Hopf algebra and the Lie algebra of rooted trees. Let us recall the representation of a quiver. Let $G=\\{e\\}$ be a trivial group and $Q$ the quiver over $G$ with arrow set $Q_{1}$. Obviously, $Q$ is a Hopf quiver. Let $kQ^{a}$ denote the path algebra of quiver $Q$. It is clear that the free algebra generated by $Q_{1}$ is isomorphic to the path algebra $kQ^{a}$ as algebras. Let $V$ be a vector space and $f=\\{f_{t}\\}_{t\in Q_{1}}$ with $f_{t}\in End_{k}V$ for any $t\in Q_{1}$. We define a map $\alpha_{f}$ from $kQ^{a}\otimes V$ to $V$ as follows: $\displaystyle\alpha_{f}(t_{1}t_{2}\cdots t_{n}\otimes v)=t_{1}t_{2}\cdots t_{n}\cdot v:=f_{t_{1}}f_{t_{2}}\cdots f_{t_{n}}(v)$ (2) for any $v\in V,t_{1},t_{2},\cdots,t_{n}\in Q_{1}$. It is clear that $(V,\alpha_{f})$ is a $kQ^{a}$-module, called a quiver module. Assume that $\rho$ is a subset of $kQ^{a}$ and $(\rho)$ denotes the ideal generated by $\rho$. Let $(V,\alpha_{f})$ be a quiver module. For any element $\sigma\in kQ^{a}$, we write $f_{\sigma}:V\rightarrow V$ by sending $x$ to $\sigma\cdot x$ for any $x\in V$. Obviously, if $(V,\alpha_{f})$ is a quiver module with $f_{\sigma}=0$ for any $\sigma\in\rho$, then $(V,\alpha_{f})$ is a $kQ^{a}/(\rho)$ -module with $\bar{a}\cdot m=a\cdot m$ for any $a\in kQ^{a},m\in M$. This is called a factor quiver module. Note every quiver module over $kQ^{a}$ can be view as a factor quiver module over $kQ^{a}/(\rho)$ when $\rho$ is empty. Assume that $(V,\alpha_{f})$ and $(V^{\prime},\alpha_{f^{\prime}})$ are two factor quiver modules. If $h:(V,\alpha_{f})\rightarrow(V^{\prime},\alpha_{f^{\prime}})$ a homomorphism of $kQ^{a}/(\rho)$-modules, then $hf_{t}=f_{t}^{\prime}h$ for any $t\in Q_{1}.$ Conversely, if $h$ is a $k$-linear map from $V$ to $V^{\prime}$ with $hf_{t}=f_{t}^{\prime}h$ for any $t\in Q_{1},$ then $h$ is homomorphism of $kQ^{a}/(\rho)$-modules. ###### Lemma 2.1. (see [ZZ, Theorem 2.9] or [ARS]) (i) $(V,\alpha)$ is a left $kQ^{a}$-module if and only if there exist $f_{t}\in End_{k}(V)$ for any $t\in Q_{1}$ such that $\alpha=\alpha_{f}$. That is to say every $kQ^{a}$-module is a quiver module. (ii) $(V,\alpha)$ is a left $kQ^{a}/(\rho)$-module if and only if there exist $f_{t}\in End_{k}(V)$ for any $t\in Q_{1}$ such that $\alpha=\alpha_{f}$ and $f_{\sigma}=0$ for any $\sigma\in\rho$. That is to say every $kQ^{a}/(\rho)$-module is a factor quiver module. Obviously, ${\cal H}_{R}\cong kQ^{a}/(\rho)$ as algebras with $Q_{1}={\cal R}T$, $\rho:=\\{tt^{\prime}-t^{\prime}t\mid t,t^{\prime}\in{\cal R}T\\}$. Using Lemma 2.1 we have the following: ###### Theorem 2.2. $(V,\alpha)$ is a left ${\cal H}_{R}$-module if and only if there exists $f_{t}\in End_{k}(V)$ for any $t\in{\cal R}T$ such that $\alpha=\alpha_{f}$ with $f=\\{f_{t}\\}_{t\in{\cal R}T}$ and $f_{t}f_{t^{\prime}}=f_{t^{\prime}}f_{t}$ for any $t,t^{\prime}\in{\cal R}T\\}$. Explicitly, in the cases above, $\displaystyle\alpha_{f}(t\otimes v)=t\cdot v:=f_{t}(v)$ (3) for any $v\in V,$ $t\in{\cal R}T$. Let $\\{p_{i}\mid i\in\mathbb{Z}^{+}\\}$ be a homogeneous basis of $Prim({\cal H}_{R})$, the set of all primitive elements. By the proof of [Fo, Proposition 8.1], $\\{1\\}\cup(\cup_{n=1}^{\infty}\\{p_{i_{1}}\top p_{i_{2}}\top\cdots\top p_{i_{n}}\mid(i_{1},i_{2},\cdots,i_{n})\in(\mathbb{Z}^{+})^{n}\\})$ is a basis of ${\cal H}_{R}$. Let $\\{1\\}\cup(\cup_{n=1}^{\infty}\\{\xi_{\bf i}\mid{\bf i}=(i_{1},i_{2},\cdots,i_{n})\in(\mathbb{Z}^{+})^{n}\\})$ be its dual basis in ${\cal H}_{R}^{g}$ with $<\xi_{\bf j},p_{i_{1}}\top p_{i_{2}}\top\cdots\top p_{i_{n}}>=\delta_{{\bf i}{\bf j}}$ for any ${\bf i},{\bf j}\in\cup_{n=1}^{\infty}(\mathbb{Z}^{+})^{n}\cup\\{0\\}$, where $p_{i_{1}}\top p_{i_{2}}\top\cdots\top p_{i_{n}}$ and $\xi_{\bf j}$ denote 1 when ${\bf j}={\bf i}=0$. Let $l_{(i_{1},i_{2},\cdots,i_{n})}\in U({\cal L}^{1})$ with $\Psi(l_{(i_{1},i_{2},\cdots,i_{n})})=\xi_{(i_{1},i_{2},\cdots,i_{n})}$ for any ${\bf i}\in\cup_{n=1}^{\infty}(\mathbb{Z}^{+})^{n}\cup\\{0\\}\\}$, where $\Psi$ is the same as in [Fo, Corollary 3.3]. Therefore, $\\{l_{\bf i}\mid{\bf i}\in\cup_{n=1}^{\infty}(\mathbb{Z}^{+})^{n}\cup\\{0\\}\\}$ is a basis of $U({\cal L}^{1})$ and $U({\cal L}^{1})$ is the free algebra generated by $\\{l_{\bf i}\mid{\bf i}\in\cup_{n=1}^{\infty}(\mathbb{Z}^{+})^{n}\cup\\{0\\}\\}$. ###### Theorem 2.3. $(V,\alpha)$ is a left $U({\cal L}^{1})$-module ( or ${\cal L}^{1}$-module ) if and only if there exists $f_{l_{\bf i}}\in End_{k}(V)$ for any ${\bf i}\in\cup_{n=1}^{\infty}(\mathbb{Z}^{+})^{n}\cup\\{0\\}$ such that $\alpha=\alpha_{f}$ with $f=\\{f_{l_{\bf i}}\\}_{{\bf i}\in(\cup_{n=1}^{\infty}(\mathbb{Z}^{+})^{n})\cup\\{0\\}}$. Explicitly, in the cases above, $\displaystyle\alpha_{f}(l_{\bf i}\otimes v)=l_{\bf i}\cdot v:=f_{l_{\bf i}}(v)$ (4) for any $v\in V,$ ${\bf i}\in(\cup_{i=1}^{\infty}(\mathbb{Z}^{+})^{i})\cup\\{0\\}$. ## 3 Some properties about Hopf algebras In this section we give some properties of the Hopf algebra of rooted trees. ### 3.1 Integrals ###### Proposition 3.1. ${\cal H}_{R}$, $H_{ladd}$ and $U({\cal L}^{1})$ have not any nonzero integrals. Proof. Let $l_{1}$ denote the rooted tree with weight $(l_{1})=1$. It is clear that subalgebra $k[l_{1}]$ of ${\cal H}_{R}$ generated by $l_{1}$ is a Hopf subalgebra of ${\cal H}_{R}$. By [DNR, Proposition 5.6.11], $k[l_{1}]$ has not any nonzero integral. Therefore, it follows from [DNR, Corollary 5.3.3] that ${\cal H}_{R}$ has not any nonzero integral. Similarly, we can show the other. $\Box$ ### 3.2 Andruskiewitsch and Schneider’s Conjecture ###### Proposition 3.2. $H_{ladd}$ can not be generated by primitive elements of $H_{ladd}$ as algebras. Proof. By [Fo, Proposition 9.3 and Theorem 9.5], $l_{2}$ can not generated by set $\\{P_{i}\mid i=1,2,\cdots,\cdots\\}$, where $P_{i}$ is defined in [Fo, Proposition 9.3]. $\Box$ N. Andruskiewitsch and H. J. Schneider in [AS2, Conjecture 1.4] gave a conjecture: every finite dimensional pointed Hopf algebra can be generated by its coradical and its skew-primitive elements. By Proposition 3.2, analogue of this conjecture for infinite dimensional pointed Hopf algebra does not hold. Furthermore, $H_{ladd}$ and ${\cal H}_{R}$ are not Nichols algebras over trivial group since $(H_{ladd})_{(1)}\not=Prim(H_{ladd})$ and $({\cal H}_{R})_{(1)}\not=Prim({\cal H}_{R})$. They also are not strictly graded coalgebras (see [Sw, P 232]). ### 3.3 Hopf subalgebra A Hopf algebra $H$ is called trivial, if dim $H=1$. ###### Proposition 3.3. (i) If $H$ is a nontrivial pointed irreducible Hopf algebra, then $H$ has not any finite dimensional nontrivial Hopf subalgebra; (ii) ${\cal H}_{R}$ and $U({\cal L}^{1})$ have not any finite dimensional nontrivial Hopf subalgebra. Proof. (i) Assume that $A$ is a Hopf subalgebra of $H$ with $A\not=H_{0}=k1_{H}$. Obviously, $A_{0}=k1_{H}$ and $A_{1}\not=A_{0}$, so there exists a nonzero primitive element $x\in A_{1}$. By [ZZC, Lemma 3.5] or [AS1, Lemma 3.3], set $\\{x^{i}\mid i=1,2,\cdots\\}$ is linearly independent in $A$, so $A$ is infinite dimensional. (ii) It immediately follows from (i). $\Box$ ### 3.4 Graded duality By [Fo, Theorem 3.1], there is a bilinear form $<,>$ on ${\cal U}({\cal L}^{1})\times{\cal H}_{R}$ such that $\displaystyle<1,h>$ $\displaystyle=$ $\displaystyle\varepsilon(h),\ \ <Z_{t},h>=(\frac{\partial}{\partial t}h)\mid_{t=0},$ $\displaystyle\mbox{ and }<Z_{t}Z_{t^{\prime}},h>$ $\displaystyle=$ $\displaystyle<Z_{t}\otimes Z_{t^{\prime}},\Delta(h)>$ for any $h\in{\cal H}_{R}$, $t,t^{\prime}\in{\cal R}T.$ ###### Lemma 3.4. (see [Sw, Section 11.2]) Assume that $H=\oplus_{i=0}^{\infty}H_{(i)}$ is a local finite graded Hopf algebra(i.e. dim $H_{(i)}<\infty$ for any $i$). Define $H^{g}:=\oplus_{i=0}^{\infty}(H_{(i)})^{*}$, called graded dual of $H$. Then $H^{g}$ is a graded Hopf algebra and $H\cong(H^{g})^{g}$ as graded Hopf algebra. Proof. It is clear that $\oplus_{i=0}^{\infty}(H_{(i)})^{*}\subseteq H^{\circ}.$ Now we show that $H^{g}=\oplus_{i=0}^{\infty}(H_{(i)})^{*}$ is a graded Hopf algebra. Considering [Ni, Pro.1.5.1], we have to prove only that it is a graded bialgebra. However, it is easy. Therefore, $(H_{(i)})^{*}=(H^{g})_{(i)}$ for $i=0,1,2,\cdots.$ Finally, we show that $\sigma_{H}$ is a graded Hopf algebra isomorphism from $H$ to $H^{gg}$, where $<\sigma_{H}(h),f>=<f,h>$ for any $f\in H^{*},h\in H$. Indeed, we have to show only that $\sigma_{H}(H)=H^{gg}.$ For any $h\in H_{(i)}$, $f\in H_{(j)}^{*}=(H^{g})_{(j)}$ with $i\not=j,$ we have that $<\sigma_{H}(h),f>=<f,h>=0,$ so $\sigma_{H}(h)\in(H^{gg})_{(i)}$. That implies $\sigma_{H}(H)=H^{gg}.$ $\Box$ ###### Corollary 3.5. ${\cal H}_{R}^{gg}\cong{\cal H}_{R}$ and $U({\cal L}^{1})^{gg}\cong U({\cal L}^{1})$ as graded Hopf algebras. Recall [Fo, Corollary 3.3], $\Phi:\left\\{\begin{array}[]{l l}{\cal H}_{R}\rightarrow U({\cal L}^{1})^{g}\\\ h\mapsto<.,h>\end{array}\right.$ and $\Psi:\left\\{\begin{array}[]{l l}U({\cal L}^{1})\rightarrow({\cal H}_{R})^{g}\\\ l\mapsto<l,.>\end{array}\right.$ are a coalgebra isomorphism and an algebra isomorphism, respectively. However, we have ###### Theorem 3.6. $\Phi$ is not algebraic and $\Psi$ is not coalgebraic. Proof. Let $t$ and $t^{\prime}$ are two rooted trees, which are in different equivalent classes. It is clear that $\Phi(tt^{\prime})(Z_{t})=t^{\prime}$ and $<(\Phi(t)*\Phi(t^{\prime})),Z_{t}>=0$, so $\Phi(tt^{\prime})\not=\Phi(t)*\Phi(t^{\prime})$. That is, $\Phi$ is not algebraic. Similarly, $\Psi$ is not coalgebraic. $\Box$ ## 4 The graded Drinfeld double In this section we structure the graded Drinfeld double of Hopf algebra of rooted trees and show that it is a local quasitriangular Hopf algebra. Let $V=\oplus_{i=0}^{\infty}V_{(i)}$ be a local finite graded vector space and $V_{n}:=\sum_{i=0}^{n}V_{(i)}$ for any $n\in\mathbb{N}$. Let $ev_{V_{(n)}}:=d_{V_{(n)}}$ and $coev_{V_{(n)}}:=b_{V_{(n)}}$ denote the evaluation and coevaluation of $V_{(n)}$, respectively. If we denote by $\\{e^{(n)}_{1},e^{(n)}_{2},\cdots,e^{(n)}_{r_{n}}\\}$ a basis of $V_{(n)}$ and $\\{f^{(n)}_{1},f^{(n)}_{2},\cdots,f^{(n)}_{r_{n}}\\}$ a its dual basis in $(V_{(n)})^{*}$ , then $\displaystyle b_{V_{(n)}}:=\sum_{i=1}^{r_{n}}(e^{(n)}_{i}\otimes f^{(n)}_{i})\ \ \hbox{ and }b_{V_{n}}:=\sum_{i=0}^{n}b_{V_{(i)}}$ (5) are coevaluations of $V_{(n)}$ and $V_{n}$, respectively, for any $n\in\mathbb{N}$. Let $C_{U,V}$ denote the flip from $U\otimes V$ to $V\otimes U$ by sending $u\otimes v$ to $v\otimes u$ for any $u\in U$, $v\in V.$ ###### Lemma 4.1. Let $H=\oplus_{i=0}^{\infty}H_{(i)}$ be a local finite graded Hopf algebra. Under notation above, set $A=(H^{g})^{cop},$ $\tau=ev_{H}C_{H,H}$, $P_{n}=b_{H_{n}}$ and $R_{n}=[P_{n}]$ $=1\otimes P_{n}\otimes 1=\sum_{m=0}^{n}\sum_{i=1}^{r_{m}}1\otimes e_{i}^{(m)}\otimes f_{i}^{(m)}\otimes 1.$ Then $(D(H),\\{R_{n}\\})$ is a local quasitriangular Hopf algebra, where $D(H)=A\bowtie_{\tau}H,$ called graded Drinfeld double of $H$. Proof. It follows from [ZGZ, Lemma 3.4] and [ZGZ, Lemma 3.6]. $\Box$ Let $\\{p_{i}^{(n)}\mid 1\leq i\leq r_{n}\\}$ be a basis of $(Prim({\cal H}_{R}))_{(n)}$ with $n\in\mathbb{Z}^{+}$. By the proof of [Fo, Proposition 8.1], $\\{p_{i_{1}}^{(j_{1})}\top p_{i_{2}}^{(j_{2})}\top\cdots\top p_{i_{s}}^{(j_{s})}\mid$ $j_{1}+j_{2}+\cdots+j_{s}=n$, $j_{1},j_{2},\cdots,j_{s}\in\mathbb{Z}^{+}$; $1\leq i_{1}\leq r_{j_{1}},\cdots,1\leq i_{s}\leq r_{j_{s}}$; $s\in\mathbb{Z}^{+}\\}$ is a basis of $({\cal H}_{R})_{(n)}$. Let $e_{j_{1},j_{2},\cdots,j_{s};i_{1},i_{2},\cdots,i_{s}}^{(n)}:=p_{i_{1}}^{(j_{1})}\top p_{i_{2}}^{(j_{2})}\top\cdots\top p_{i_{s}}^{(j_{s})}$. Therefore, $\\{e^{(n)}_{j_{1},\cdots,j_{s};i_{1},\cdots,i_{s}}\mid$ $j_{1}+j_{2}+\cdots+j_{s}=n$, $j_{1},j_{2},\cdots,j_{s}\in\mathbb{Z}^{+}$; $1\leq i_{1}\leq r_{j_{1}},\cdots,1\leq i_{s}\leq r_{j_{s}}$; $s\in\mathbb{Z}^{+}\\}$ is a basis of $({\cal H}_{R})_{(n)}$. Let $\\{f^{(n)}_{j_{1},\cdots,j_{s};i_{1},\cdots,i_{s}}\mid$ $j_{1}+j_{2}+\cdots+j_{s}=n$, $j_{1},j_{2},\cdots,j_{s}\in\mathbb{N}$; $1\leq i_{1}\leq r_{j_{1}},\cdots,1\leq i_{s}\leq r_{j_{s}}$; $s\in\mathbb{Z}^{+}\\}$ is its dual basis in $({\cal H}_{R})^{g}_{(n)}$. Applying these basis and Lemma above we have ###### Theorem 4.2. Let $H={\cal H}_{R}$ $=\oplus_{i=0}^{\infty}H_{(i)}$. Then $(D(H),\\{R_{n}\\})$ is a local quasitriangular Hopf algebra with $R_{n}=[P_{n}]$ $=1\otimes P_{n}\otimes 1=$ $1\otimes 1\otimes 1\otimes 1+\sum_{m=1}^{n}(\sum\\{1\otimes e^{(m)}_{j_{1},\cdots,j_{s};i_{1},\cdots,i_{s}}\otimes f^{(m)}_{j_{1},\cdots,j_{s};i_{1},\cdots,i_{s}}\otimes 1\mid$ $j_{1}+j_{2}+\cdots+j_{s}=m$, $j_{1},j_{2},\cdots,j_{s}\in\mathbb{N}$; $1\leq i_{1}\leq r_{j_{1}},\cdots,1\leq i_{s}\leq r_{j_{s}}$; $s\in\mathbb{Z}^{+}\\})$. Furthermore $D(H)=A\bowtie_{\tau}H=A\\#_{\beta}H$ (i.e. (right) smash product), i.e. $(a\bowtie h)(b\bowtie g)=\sum_{(b)}ab_{1}\otimes\beta(h,b_{2})g$ and $\beta(h,a)=\sum_{(a),(h)}<a_{1},h_{1}><a_{2},S(h_{3})>h_{2}$ for any $a,b\in A$, $h,g\in H.$ ###### Theorem 4.3. Let $H=U({\cal L}^{1})=\oplus_{i=0}^{\infty}H_{(i)}$. Then $(D(H),\\{R_{n}\\})$ is a local quasitriangular Hopf algebra. Furthermore, $D(H)=A\bowtie_{\tau}H=A\\#H$ (i.e. smash product), i.e. $(a\bowtie h)(b\bowtie g)=\sum_{(b)}a\alpha(h_{1},b)\otimes h_{2}g$ and $\alpha(h,a)=\sum_{(a),(h)}<a_{1},h_{1}><a_{3},S(h_{2})>a_{2}$ for any $a,b\in A$, $h,g\in H.$ ## 5 Appendix A rooted tree $t$ is called an $r$-branch tree if the fertility of every vertex of $t$ is less than or equal to $r$. Let $a_{r}(n)$ denote the number of equivalent classes of $r$-branch rooted trees with weight $n$. We can show the following by the method similar to the proof of Theorem 1.1. ###### Theorem 5.1. $\displaystyle a_{r}(n+1)$ $\displaystyle=$ $\displaystyle\sum\\{C^{\lambda_{1}}_{a_{r}(1)+\lambda_{1}-1}C^{\lambda_{2}}_{a_{r}(2)+\lambda_{2}-1}\cdots C^{\lambda_{n}}_{a_{r}(n)+\lambda_{n}-1}\ \mid\ $ (6) $\displaystyle 1\lambda_{1}+2\lambda_{2}+\cdots+n\lambda_{n}=n;\ \ {\rm card}\ \\{i\mid\ \lambda_{i}\not=0\\}\leq r\\}.$ Acknowledgement : The first author thanks the Chern institute of mathematics for hospitality. ## References * [ARS] M. Auslander, I. Reiten and S.O. Smal$\phi$, _Representation theory of Artin algebras_ , Cambridge University Press, 1995. * [AS1] N. Andruskiewitsch and H. J. Schneider,_Lifting of quantum linear spaces and pointed Hopf algebras of order $p^{3}$,_ J. Algebra, 208(1998), 645-691. * [AS2] N. Andruskiewitsch and H. J. Schneider, _Finite quantum groups and Cartan matrices_ , Adv. Math. 154 (2000), 1–45. * [BK] D. J. Broadhurst, D. Kreimer, _Renormalization automated by Hopf algebra_ , hep-th/9810087. * [CK] A. Connes, D. Kreimer, _Hopf Algebras, Renormalization and Noncommutative Geometry_ , Communications in Mathematical Physics, 199(1998), 203-242. * [DNR] S.Dascalescu, C.Nastasecu and S. Raianu, _Hopf algebras: an introduction_ , Marcel Dekker Inc. , 2001. * [Fo] L. Foissy, _Finite dimensional comodules over the Hopf algebra of rooted trees,_ Journal of Algebra, 255(2002) 1, 89-120. See also math.QA/0105210. * [Kr1] D. Kreimer, _On the Hopf algebra structure of pertubative quantum field theories_ , q-alg/9707029. * [Kr2] D. Kreimer, _On Overlapping Divergences_ , (1999), hep-th/9810022. * [Ni] W. Nichols, _Bialgebras of type one,_ Commun. Alg. 6 (1978), 1521–1552. * [Sw] M. E. Sweedler, _Hopf Algebras_ , Benjamin, New York, 1969. * [ZZ] Shouchuan Zhang, Yaozhong Zhang, _Structures and Representations of Generalized Path Algebras,_ Algebras and Representation Theory, 10(2007)2,117-134. Also in math.RA/0402188. * [ZGZ] Shouchuan Zhang, Mark D. Gould, Yao-Zhong Zhang, _Local Quasitriangular Hopf Algebras,_ math.QA/0601541 * [ZZC] Shouchuan Zhang, Yaozhong Zhang, Huixiang Chen, _Classification of PM Quiver Hopf Algebras,_ to appear Journal of Algebra and Its Applications. Also in math. QA/0410150.

arxiv-papers

2024-09-04T02:48:53.282156

0710.5926

# Mod $2$ cohomology of $2$-local finite groups of low rank Shizuo Kaji Department of Mathematics Kyoto University Kyoto 606-8502 Japan kaji@math.kyoto-u.ac.jp ###### Abstract. We determine the mod $2$ cohomology over the Steenrod algebra $\mathcal{A}_{2}$ of the classifying space of a free loop group $LG$ for $G=Spin(7)$, $Spin(8)$, $Spin(9)$, $F_{4}$, and $DI(4)$. Then we show that it is isomorphic as algebras over $\mathcal{A}_{2}$ to the mod $2$ cohomology of the classifying space of a certain $2$-local finite group of type $G$. ###### Key words and phrases: mod 2 cohomology, free loop groups, 2-local finite groups ###### 2000 Mathematics Subject Classification: Primary 55R35; Secondary 55S10. The first author is partially supported by the Grant-in-Aid for JSPS Fellows 182641. ## 1\. Introduction In [Ku], Kuribayashi considered the cohomology of the free loop space $LX$ over a space $X$ by developing a tool called the module derivation, which is a map from the cohomology of $X$ to that of $LX$ with degree $-1$ having some nice properties. A free loop space $LX$ can be considered as the homotopy fixed points space of the identity map of $X$. In addition, for a prime $p$, the mod $p$ homotopy type of the classifying space of a certain finite group also occurs as the Bousfield and Kan $p$-completion of the homotopy fixed points space of the self-map of the classifying space $BG$ of a compact Lie group $G$, namely the unstable Adams operation ([F]). From this point of view, Kishimoto and Kono ([KK]) generalized Kuribayashi’s method to calculate the cohomology of the homotopy fixed points space of a self-map $\phi$ of $X$, which they call the twisted loop space $\mathbb{L}_{\phi}X$. When the cohomology of $BG$ is polynomial algebra, the calculation of the cohomology of the homotopy fixed points space sometimes reduces to an easy computation using their method. In this note, we give actual computations for the mod $2$ cohomology over the Steenrod algebra $\mathcal{A}_{2}$ of the classifying spaces of free loop groups $LG$ of $G$ and $2$-local finite groups ([BM]) of type $G$ with $G=Spin(7),Spin(8),Spin(9),F_{4}$, and $DI(4)$ the finite loop space at prime $2$ constructed by Dwyer and Wilkerson ([DW]). And we have the following Theorem: ###### Theorem 1.1. We have the following isomorphisms of algebras over the Steenrod algebra $\mathcal{A}_{2}$. $\displaystyle H^{*}(BLSpin(n);\mathbb{Z}/2)$ $\displaystyle\cong H^{*}(Spin_{n}(q);\mathbb{Z}/2)\quad(n=7,8,9)$ $\displaystyle H^{*}(BLF_{4};\mathbb{Z}/2)$ $\displaystyle\cong H^{*}(F_{4}(q);\mathbb{Z}/2)$ $\displaystyle H^{*}(BLDI(4);\mathbb{Z}/2)$ $\displaystyle\cong H^{*}(BSol(q);\mathbb{Z}/2),$ where $q$ is an odd prime power. Note that $H^{*}(BLG;\mathbb{Z}/2)=H^{*}(LBG;\mathbb{Z}/2)$ (see for example [Ku, §2]). For $G=Spin(7)$, $Spin(8)$, $Spin(9)$, $F_{4}$, and $DI(4)$, the explicit computations for $H^{*}(LBG;\mathbb{Z}/2)$ are given in the sections §3, §4 and §5. Acknowledgment. We would like to thank Prof. Akira Kono for various suggestions. ## 2\. Main tool Here we summarize the result of [KK] necessary for our purpose. Let $\phi$ be a based self-map of a based space $X$. The twisted loop space $\mathbb{L}_{\phi}X$ of $X$ is defined in the following pull-back diagram: $\textstyle{\mathbb{L}_{\phi}X\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{X^{[0,1]}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{e_{0}\times e_{1}}$$\textstyle{X\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{1\times\phi}$$\textstyle{X\times X}$ where $e_{i}\ (i=0,1)$ is the evaluation at $i$. In other words, $\mathbb{L}_{\phi}X$ is a space of all continuous maps $l$ from the interval $[0,1]$ to $X$ which satisfy $l(0)=\phi(l(1))$. The twisted tube $\mathbb{T}_{\phi}X$ of $X$ is defined by $\mathbb{T}_{\phi}X=\frac{[0,1]\times X}{(0,x)\simeq(1,\phi(x))}$ and there is a canonical inclusion $\iota:X\hookrightarrow\mathbb{T}_{\phi}X$, where $\iota(x)=(0,x)$. ###### Remark 1. When $\phi$ is the identity map, then $\mathbb{L}_{\phi}X$ is merely the free loop space $LX$ and $\mathbb{T}_{\phi}X=S^{1}\times X$. The cohomology of $\mathbb{T}_{\phi}X$ and $X$ is related by the Wang exact sequence (A) $\cdots H^{n-1}(X;R)\xrightarrow{1-\phi^{*}}H^{n-1}(X;R)\xrightarrow{\delta}H^{n}(\mathbb{T}_{\phi}X;R)\xrightarrow{\iota^{*}}H^{n}(X;R)\xrightarrow{1-\phi^{*}}H^{n}(X;R)\cdots,$ where $R$ is any commutative ring. In particular, this exact sequence splits off to the short exact sequences when $H^{*}(\phi;R)$ is the identity map. Let $ev$ be the evaluation map $\displaystyle S^{1}\times LX$ $\displaystyle\to$ $\displaystyle X$ $\displaystyle(t,l)$ $\displaystyle\mapsto$ $\displaystyle l(t).$ Then for any commutative ring $R$, a map $\sigma_{X}:H^{*}(X;R)\to H^{*-1}(LX;R)$ is defined by the following equation: $ev^{*}(x)=s\otimes\sigma_{X}(x)+1\otimes x,\quad(x\in H^{*}(X;R)),$ where $s\in H^{1}(S^{1};R)$ is a generator. On the other hand, we define a map $in:\mathbb{L}_{\phi}X\to L\mathbb{T}_{\phi}X$ by $\displaystyle\mathbb{L}_{\phi}X$ $\displaystyle\to$ $\displaystyle L\mathbb{T}_{\phi}X$ $\displaystyle l$ $\displaystyle\mapsto$ $\displaystyle t\mapsto(t,l(t)).$ Then the twisted cohomology suspension is defined by the following composition $\hat{\sigma}_{\phi}:H^{*}(\mathbb{T}_{\phi}X;R)\xrightarrow{\sigma_{\mathbb{T}_{\phi}X}}H^{*-1}(L\mathbb{T}_{\phi}X;R))\xrightarrow{in^{*}}H^{*-1}(\mathbb{L}_{\phi}X;R).$ Moreover, if we have a section $r:H^{*}(X;R)\to H^{*}(\mathbb{T}_{\phi}X;R)$ of $\iota^{*}$, we can define another map $\tilde{\sigma}_{\phi}=r\circ\hat{\sigma}_{\phi}:H^{*}(X;R)\to H^{*-1}(\mathbb{L}_{\phi}X;R)$. ###### Remark 2. When $\phi=Id$ the identity map, we can take the section $r=\pi^{*}$, where $\pi:\mathbb{T}_{\phi}X=S^{1}\times X\to X$ is the projection. Then $\tilde{\sigma}_{\phi}=\sigma_{X}$ and $\tilde{\sigma}_{\phi}$ also coincides with Kuribayashi’s module derivation $\mathcal{D}_{X}$ defined in [Ku]. The map $\tilde{\sigma}_{\phi}$ together with the Wang sequence above relates the cohomology of $X$ to that of $\mathbb{L}_{\phi}X$. We consider the following conditions: $(*)\left\\{\begin{array}[]{cl}\mbox{(i)}&H^{*}(X;\mathbb{Z}/2)\mbox{ is a polynomial algebra }\mathbb{Z}/2[x_{1},x_{2},\ldots,x_{l}],\\\ \mbox{(ii)}&H^{*}(\phi;\mathbb{Z}/2)\mbox{ is the identity map},\\\ \mbox{(iii)}&H^{n}(\phi;\mathbb{Z}/4)\mbox{ is the identity map for all odd $n$ and $n\equiv 0\mod 4$.}\end{array}\right.$ Then the result of [KK] specializes to the following Proposition. ###### Proposition 2.1 (Kishimoto-Kono). Assume that (i) and (ii) in the conditions (*) are satisfied. Suppose that there is a section $r:H^{*}(X;\mathbb{Z}/2)\to H^{*}(\mathbb{T}_{\phi}X;\mathbb{Z}/2)$ of $\iota^{*}$, which commutes with the Steenrod operations. Then we have 1. (1) $H^{*}(\mathbb{L}_{\phi}X;\mathbb{Z}/2)=\mathbb{Z}/2[e^{*}(x_{1}),e^{*}(x_{2}),\ldots,e^{*}(x_{l})]\otimes\Delta(\tilde{\sigma}_{\phi}(x_{1}),\tilde{\sigma}_{\phi}(x_{2}),\ldots,\tilde{\sigma}_{\phi}(x_{l}))$, where $e:\mathbb{L}_{\phi}X\to X$ is the evaluation at $0$. 2. (2) $\tilde{\sigma}_{\phi}(xy)=\tilde{\sigma}_{\phi}(x)e^{*}(y)+e^{*}(x)\tilde{\sigma}_{\phi}(y)$ for $x,y\in H^{*}(X;\mathbb{Z}/2)$ 3. (3) $\tilde{\sigma}_{\phi}$ commutes with the Steenrod operations. In the rest of this note, we restrict ourselves to the case when $X=BG$, where $G$ is either $Spin(7)$, $Spin(8)$, $Spin(9)$, $F_{4}$, or $DI(4)$. When $\phi$ is the identity map, $\mathbb{L}_{\phi}BG$ is merely the free loop space $LBG$, which is homotopy equivalent to $BLG$ (see for example [Ku, §2]). Then the conditions (*) is trivially satisfied. Moreover, we can take $\pi^{*}$ as a section $r$ of $\iota^{*}$ which commutes with the Steenrod operations, where $\pi:S^{1}\times X\to X$ is the projection. Hence we can calculate $H^{*}(BLG;\mathbb{Z}/2)$ by above Proposition. For $G=Spin(7),Spin(8),Spin(9),F_{4}$ and a odd prime power $q$, there is a self-map $\psi^{q}$ of $BG$ called the unstable Adams operation of degree $q$ ([W]), where $H^{2r}(\psi^{q};\mathbb{Q})$ is multiplication by $q^{r}$. When $\phi=\psi^{q}$, the conditions (*) is satisfied. The Bousfield and Kan $2$-completion ([BK]) of $\mathbb{L}_{\phi}BG$ is known to be homotopy equivalent to that of the classifying space of a finite Chevalley group of type $G(q)$ ([F]). Hence we can calculate $H^{*}(G(q);\mathbb{Z}/2)$ by above Proposition. For $G=DI(4)$ and a odd prime power $q$, Notbohm ([N]) showed that there is a self-map $\psi^{q}$ of $BDI(4)$ also called the unstable Adams operation of degree $q$ ([N]), where $H^{2r}(\psi^{q};\mathbb{Q}_{2}^{\wedge})$ is multiplication by $q^{r}$. When $\phi=\psi^{q}$, the conditions (*) is satisfied. Using this map, Benson ([B]) defined the classifying space $BSol(q)$ of an exotic $2$-local finite group as $L_{\psi^{q}}BDI(4)$ which can be regarded as the “classifying space” of Solomon’s non-existent finite group ([S]). Hence we can calculate $H^{*}(BSol(q);\mathbb{Z}/2)$ by above Proposition. To sum up, Theorem 1.1 reduces to the computation of $H^{*}(\mathbb{L}_{\phi}BG;\mathbb{Z}/2)$, where $\phi$ is the identity map or $\psi^{q}$. In the following sections, our main observation is to construct a section $r:H^{*}(BG;\mathbb{Z}/2)\to H^{*}(\mathbb{T}_{\phi}BG;\mathbb{Z}/2)$ which commutes with Steenrod operations when $\phi=\psi^{q}$ to show the following: ###### Theorem 2.1. Let $G=Spin(7),Spin(8),Spin(9),F_{4}$ or $DI(4)$. Then $H^{*}(LBG;\mathbb{Z}/2)\simeq H^{*}(\mathbb{L}_{\psi^{q}}BG;\mathbb{Z}/2)$ as algebras over the Steenrod algebra $\mathcal{A}_{2}$, where $\psi^{q}$ is the unstable Adams operation of degree an odd prime power $q$. ## 3\. Computations for $G=Spin(7),Spin(8),Spin(9)$ The mod $2$ cohomology over $\mathcal{A}_{2}$ of $BSpin(7),BSpin(8)$ and $BSpin(9)$ are well known ([Q, Ko]). Since we rely on these results, we recall them here. $H^{*}(BSpin(7);\mathbb{Z}/2)=\mathbb{Z}[w_{4},w_{6},w_{7},w_{8}]$ and the action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rcccc}&w_{4}&w_{6}&w_{7}&w_{8}\\\ Sq^{1}&0&w_{7}&0&0\\\ Sq^{2}&w_{6}&0&0&0\\\ Sq^{4}&w_{4}^{2}&w_{4}w_{6}&w_{4}w_{7}&w_{4}w_{8}.\\\ \end{array}$ $H^{*}(BSpin(8);\mathbb{Z}/2)=\mathbb{Z}[w_{4},w_{6},w_{7},w_{8},e_{8}]$ and the action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rccccc}&w_{4}&w_{6}&w_{7}&w_{8}&e_{8}\\\ Sq^{1}&0&w_{7}&0&0&0\\\ Sq^{2}&w_{6}&0&0&0&0\\\ Sq^{4}&w_{4}^{2}&w_{4}w_{6}&w_{4}w_{7}&w_{4}w_{8}&w_{4}e_{8}.\\\ \end{array}$ $H^{*}(BSpin(9);\mathbb{Z}/2)=\mathbb{Z}[w_{4},w_{6},w_{7},w_{8},e_{16}]$ and the action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rccccc}&w_{4}&w_{6}&w_{7}&w_{8}&e_{16}\\\ Sq^{1}&0&w_{7}&0&0&0\\\ Sq^{2}&w_{6}&0&0&0&0\\\ Sq^{4}&w_{4}^{2}&w_{4}w_{6}&w_{4}w_{7}&w_{4}w_{8}&0\\\ Sq^{8}&0&0&0&w_{8}^{2}&w_{8}e_{16}+w_{4}^{2}e_{16}.\\\ \end{array}$ Based on these results, we compute the mod $2$ cohomology of classifying spaces of free loop groups $LG$ for $G=Spin(7),Spin(8),Spin(9)$. ###### Proposition 3.1. $H^{*}(LBSpin(7);\mathbb{Z}/2)=\mathbb{Z}/2[v_{4},v_{6},v_{7},v_{8},y_{3},y_{5},y_{7}]/I\quad(|v_{i}|=i,|y_{i}|=i)$, where $I$ is the ideal generated by $\\{y_{5}^{2}+y_{3}^{2}v_{4}+y_{3}v_{7},y_{3}^{4}+y_{3}^{2}v_{6}+y_{5}v_{7},y_{7}^{2}+y_{3}^{2}v_{8}+y_{7}v_{7}\\}.$ The action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rccccccc}&v_{4}&v_{6}&v_{7}&v_{8}&y_{3}&y_{5}&y_{7}\\\ Sq^{1}&0&v_{7}&0&0&0&y_{3}^{2}&0\\\ Sq^{2}&v_{6}&0&0&0&y_{5}&0&0\\\ Sq^{4}&v_{4}^{2}&v_{4}v_{6}&v_{4}v_{7}&v_{4}v_{8}&0&y_{3}v_{6}+y_{5}v_{4}&y_{3}v_{8}+y_{7}v_{4}.\\\ \end{array}$ ###### Proof. We apply Proposition 2.1 when $\phi=Id$ the identity map. Since $\mathbb{L}_{Id}BG=LBG$ and $\mathbb{T}_{\phi}X=S^{1}\times X$, we can use Proposition 2.1 with $r=\pi^{*}$, where $\pi:S^{1}\times X\to X$ is the projection. We take $v_{i}=e^{*}(w_{i}),\ y_{i-1}=\tilde{\sigma}_{Id}(w_{i})\quad(i=4,6,7,8)$. Then by Proposition 2.1 (1), we have $H^{*}(LBSpin(7);\mathbb{Z}/2)=\mathbb{Z}/2[v_{4},v_{6},v_{7},v_{8}]\otimes\Delta[y_{3},y_{5},y_{6},y_{7}]$. By Proposition 2.1 (2) and (3), the action of $\mathcal{A}_{2}$ on the generator $v_{i}\ (i=4,6,7,8)$ is obvious, and on $y_{i}\ (i=3,5,7)$ we calculate as follows: $\displaystyle Sq^{1}y_{3}=Sq^{1}\tilde{\sigma}_{Id}(w_{4})=\tilde{\sigma}_{Id}(Sq^{1}w_{4})=0$ $\displaystyle Sq^{2}y_{3}=\tilde{\sigma}_{Id}(Sq^{2}w_{4})=\tilde{\sigma}_{Id}(w_{6})=y_{5}$ $\displaystyle Sq^{1}y_{5}=\tilde{\sigma}_{Id}(Sq^{1}w_{6})=y_{6}$ $\displaystyle Sq^{2}y_{5}=\tilde{\sigma}_{Id}(Sq^{2}w_{6})=0$ $\displaystyle Sq^{4}y_{5}=\tilde{\sigma}_{Id}(Sq^{4}w_{6})=\tilde{\sigma}_{Id}(w_{4}w_{6})=\tilde{\sigma}_{Id}(w_{4})e^{*}(w_{6})+\tilde{\sigma}_{Id}(w_{6})e^{*}(w_{4})=y_{3}v_{6}+y_{5}v_{4}$ $\displaystyle Sq^{1}y_{7}=\tilde{\sigma}_{Id}(Sq^{1}w_{8})=0$ $\displaystyle Sq^{2}y_{7}=\tilde{\sigma}_{Id}(Sq^{2}w_{8})=0$ $\displaystyle Sq^{4}y_{7}=\tilde{\sigma}_{Id}(Sq^{4}w_{8})=\tilde{\sigma}_{Id}(w_{4}w_{8})=y_{3}v_{8}+y_{7}v_{4}.$ On the other hand, with the aid of the Adem relations, we can determine the ring structure as follows: $\displaystyle y_{3}^{2}$ $\displaystyle=Sq^{3}y_{3}=Sq^{1}Sq^{2}y_{3}=Sq^{1}y_{5}=y_{6}$ $\displaystyle y_{5}^{2}$ $\displaystyle=Sq^{5}y_{5}=Sq^{1}Sq^{4}y_{5}=Sq^{1}(y_{5}v_{4}+y_{3}v_{6})=y_{6}v_{4}+y_{3}v_{7}=y_{3}^{2}v_{4}+y_{3}v_{7}$ $\displaystyle y_{7}^{2}$ $\displaystyle=Sq^{7}y_{7}=Sq^{1}Sq^{2}Sq^{4}y_{7}=Sq^{1}Sq^{2}(y_{3}v_{8}+y_{7}v_{4})=Sq^{1}(y_{5}v_{8}+y_{7}v_{6})=y_{3}^{2}v_{8}+y_{7}v_{7}$ $\displaystyle y_{3}^{4}$ $\displaystyle=y_{6}^{2}=Sq^{6}y_{6}=(Sq^{2}Sq^{4}+Sq^{5}Sq^{1})y_{6}=\tilde{\sigma}_{\phi}((Sq^{2}Sq^{4}+Sq^{5}Sq^{1})w_{7})$ $\displaystyle=\tilde{\sigma}_{\phi}(Sq^{2}w_{4}w_{7})=\tilde{\sigma}_{\phi}(w_{6}w_{7})=y_{5}v_{7}+v_{6}y_{3}^{2}.$ ∎ Now we proceed to the computation of the mod $2$ cohomology algebra over $\mathcal{A}_{2}$ of a finite Chevalley group of type $Spin_{7}(q)$ for an odd prime power $q$. The Bousfield-Kan $2$-completion of the classifying space of $Spin_{7}(q)$ is shown to be homotopy equivalent to the Bousfield-Kan $2$-completion of $\mathbb{L}_{\psi^{q}}BSpin(7)$ by Friedlander ([F]) where $\psi^{q}$ is the unstable Adams operation of degree $q$. Thus $H^{*}(\mathbb{L}_{\psi^{q}}BSpin(7);\mathbb{Z}/2)\cong H^{*}(Spin_{7}(q);\mathbb{Z}/2)$ as algebras over the Steenrod algebra $\mathcal{A}_{2}$. ###### Proposition 3.2. For $\phi=\psi^{q}$ the unstable Adams operation of degree an odd prime power $q$, $H^{*}(\mathbb{L}_{\phi}BSpin(7);\mathbb{Z}/2)$ is isomorphic to $H^{*}(LBSpin(7);\mathbb{Z}/2)$ as algebras over $\mathcal{A}_{2}$. ###### Proof. By Proposition 2.1, we only have to construct a section $r$ of $\iota^{*}:H^{*}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/2)\to H^{*}(BSpin(7);\mathbb{Z}/2)$ which commutes with the Steenrod operations. To do so, we carefully choose an element of $(\iota^{*})^{-1}(w_{i})\subset H^{i}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/2)$ for each generator $w_{i}$ of $H^{i}(BSpin(7);\mathbb{Z}/2)$ so that the action of $\mathcal{A}_{2}$ on it is compatible with that on $w_{i}$. As mentioned in the first section, now the conditions (*) are satisfied and the Wang sequences (A) for $R=\mathbb{Z}/2$ and $\mathbb{Z}/4$ split to the short exact sequences $0\to H^{*-1}(BSpin(7);\mathbb{Z}/2)\xrightarrow{\delta}H^{*}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/2)\xrightarrow{\iota^{*}}H^{*}(BSpin(7);\mathbb{Z}/2)\to 0,$ $0\to H^{n-1}(BSpin(7);\mathbb{Z}/4)\xrightarrow{\delta}H^{n}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/4)\xrightarrow{\iota^{*}}H^{n-1}(BSpin(7);\mathbb{Z}/4)\to 0,$ where $n\equiv 0\mod 4$. Let $u_{4}\in H^{4}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/2)\simeq\mathbb{Z}/2$ be a generator and we put $u_{6}=Sq^{2}u_{4}$, $u_{7}=Sq^{1}u_{6}$. By the Wang sequence for $\mathbb{Z}/2$, we have that $H^{8}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/2)$ is generated by $\delta(w_{7})$ and any element $u^{\prime}_{8}\in(\iota^{*})^{-1}(w_{8})$. Note that $w_{8}$ is the mod $2$ reduction of a generator of $H^{8}(BSpin(7);\mathbb{Z})$. Thus from the Wang sequence for $\mathbb{Z}/4$, we have that $u^{\prime}_{8}$ is in the image under $\rho$, where $\rho$ is the mod $2$ reduction map in the following Bockstein sequence $\to H^{8}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/2)\to H^{8}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/4)\xrightarrow{\rho}H^{8}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/2)$ $\xrightarrow{Sq^{1}}H^{9}(\mathbb{T}_{\phi}BSpin(7);\mathbb{Z}/2)\to\cdots$ . Therefore we have $Sq^{1}(u^{\prime}_{8})=0$. Furthermore, using the Wang sequence for $\mathbb{Z}/2$, we have $Sq^{2}u^{\prime}_{8}=0$ since $H^{9}(BSpin(7);\mathbb{Z}/2)=0$ and $\iota^{*}(Sq^{2}u^{\prime}_{8})=Sq^{2}\iota^{*}(u^{\prime}_{8})=Sq^{2}w_{8}=0$. Now we want to replace $u^{\prime}_{8}$ with the one compatible with the action of $Sq^{4}$ on $w_{8}$, without changing the action of $Sq^{i}$ for $i<8$. Since $H^{11}(BSpin(7);\mathbb{Z}/2))\cong\mathbb{Z}/2$ is generated by $w_{4}w_{7}$ and $\iota^{*}(Sq^{4}u^{\prime}_{8})=Sq^{4}w_{8}=w_{4}w_{8}$, we have $Sq^{4}u^{\prime}_{8}=u_{4}u^{\prime}_{8}+\epsilon\delta(w_{4}w_{7})$, where $\epsilon=0$ or $1$. We put $u_{8}=u^{\prime}_{8}+\epsilon\delta(w_{7})$. Since $\delta(w_{4}w_{7})=\delta(Sq^{4}w_{7})=Sq^{4}\delta(w_{7})$, we have $Sq^{4}(u_{8})=u_{4}u_{8}$. Since $Sq^{i}\delta(w_{7})=\delta(Sq^{i}w_{7})=0,\ (i=1,2)$, we have $Sq^{i}(u_{8})=0,\ (i=1,2)$. Take $r$ to be the ring homomorphism defined by $r(w_{i})=u_{i}\ (i=4,6,7,8)$, then $r$ is a section of $\iota^{*}$ which commutes with the Steenrod algebra $\mathcal{A}_{2}$. ∎ Now we proceed to the case when $G=Spin(8)$. ###### Proposition 3.3. $H^{*}(LBSpin(8);\mathbb{Z}/2)=\mathbb{Z}/2[v_{4},v_{6},v_{7},v_{8},f_{8},y_{3},y_{5},y_{7},z_{7}]/I$ $(|v_{i}|=i,|y_{i}|=i,|v_{8}|=8,|z_{7}|=7)$, where $I$ is the ideal generated by $\\{y_{5}^{2}+y_{3}^{2}v_{4}+y_{3}v_{7},y_{3}^{4}+y_{3}^{2}v_{6}+y_{5}v_{7},y_{7}^{2}+y_{3}^{2}v_{8}+y_{7}v_{7},z_{7}^{2}+y_{3}^{2}f_{8}+z_{7}v_{7}\\}.$ The action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rccccccccc}&v_{4}&v_{6}&v_{7}&v_{8}&f_{8}&y_{3}&y_{5}&y_{7}&z_{7}\\\ Sq^{1}&0&v_{7}&0&0&0&0&y_{3}^{2}&0&0\\\ Sq^{2}&v_{6}&0&0&0&0&y_{5}&0&0&0\\\ Sq^{4}&v_{4}^{2}&v_{4}v_{6}&v_{4}v_{7}&v_{4}v_{8}&v_{4}f_{8}&0&y_{3}v_{6}+y_{5}v_{4}&y_{3}v_{8}+y_{7}v_{4}&y_{3}f_{8}+z_{7}v_{4}.\\\ \end{array}$ ###### Proof. Completely parallel to the case of $Spin(7)$ since the generator $e_{8}\in H^{8}(BSpin(8);\mathbb{Z}/2)$ looks same as $w_{8}$. ∎ A finite Chevalley group of type $Spin_{8}(q)$ has the mod $2$ cohomology algebra over $\mathcal{A}_{2}$ isomorphic to $H^{*}(\mathbb{L}_{\psi^{q}}BSpin(8);\mathbb{Z}/2)$, where $q$ is an odd prime power. And we have ###### Proposition 3.4. For $\phi=\psi^{q}$ the unstable Adams operation of degree an odd prime power $q$, $H^{*}(\mathbb{L}_{\phi}BSpin(8);\mathbb{Z}/2)$ is isomorphic to $H^{*}(LBSpin(8);\mathbb{Z}/2)$ as algebras over $\mathcal{A}_{2}$. ###### Proof. We can construct a section $r$ completely parallel to the case of $BSpin(7)$. ∎ Now we proceed to the case when $G=Spin(9)$. ###### Proposition 3.5. $H^{*}(LBSpin(9);\mathbb{Z}/2)=\mathbb{Z}/2[v_{4},v_{6},v_{7},v_{8},f_{16},y_{3},y_{5},y_{7},z_{15}]/I$ $(|v_{i}|=i,|y_{i}|=i,|f_{16}|=16,|z_{16}|=16)$, where $I$ is the ideal generated by $\\{y_{5}^{2}+y_{3}v_{7}+v_{4}y_{3}^{2},y_{3}^{4}+y_{3}^{2}v_{6}+y_{5}v_{7},y_{7}^{2}+y_{3}^{2}v_{8}+y_{7}v_{7},z_{15}^{2}+v_{7}v_{8}z_{15}+v_{7}y_{7}f_{16}+y_{3}^{2}v_{8}f_{16}\\}.$ The action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rccccccccc}&v_{4}&v_{6}&v_{7}&v_{8}&f_{16}&y_{3}&y_{5}&y_{7}&z_{15}\\\ Sq^{1}&0&v_{7}&0&0&0&0&y_{3}^{2}&0&0\\\ Sq^{2}&v_{6}&0&0&0&0&y_{5}&0&0&0\\\ Sq^{4}&v_{4}^{2}&v_{4}v_{6}&v_{4}v_{7}&v_{4}v_{8}&0&0&y_{3}v_{6}+y_{5}v_{4}&y_{3}v_{8}+y_{7}v_{4}&0\\\ Sq^{8}&0&0&0&v_{8}^{2}&v_{8}f_{16}+v_{4}^{2}f_{16}&0&0&0&J_{1}\end{array}$ where $J_{1}=y_{7}f_{16}+v_{8}z_{15}+v_{4}^{2}z_{15}.$ ###### Proof. In dimensions lower than $9$, the calculation is completely same as in the case of $BSpin(7)$. We take $f_{16}=e^{*}(e_{16}),\ z_{15}=\tilde{\sigma}_{Id}(e_{16})$. Then we have only to calculate the following: $\displaystyle Sq^{8}z_{15}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{8}e_{16})=\tilde{\sigma}_{Id}(w_{8}e_{16}+w_{4}^{2}e_{16})=y_{7}f_{16}+v_{8}z_{15}+v_{4}^{2}z_{15}$ $\displaystyle z_{15}^{2}$ $\displaystyle=Sq^{15}z_{15}=\tilde{\sigma}_{Id}(Sq^{15}e_{16})=\tilde{\sigma}_{Id}(Sq^{7}Sq^{8}e_{16})=\tilde{\sigma}_{Id}(Sq^{7}(w_{8}e_{16}+w_{4}^{2}e_{16}))$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{3}Sq^{4}(w_{8})e_{16}+Sq^{3}Sq^{4}(w_{4}^{2})e_{16})=\tilde{\sigma}_{Id}(Sq^{1}Sq^{2}(w_{4}w_{8})e_{16}+Sq^{3}(w_{6}^{2})e_{16})$ $\displaystyle=\tilde{\sigma}_{Id}(w_{7}w_{8}e_{16})=v_{7}v_{8}\tilde{\sigma}_{Id}(e_{16})+\tilde{\sigma}_{Id}(w_{7}w_{8})f_{16}=v_{7}v_{8}z_{15}+v_{7}y_{7}f_{16}+y_{3}^{2}v_{8}f_{16}.$ ∎ A finite Chevalley group of type $Spin_{9}(q)$ has the mod $2$ cohomology algebra over $\mathcal{A}_{2}$ isomorphic to $H^{*}(\mathbb{L}_{\phi}BSpin(9);\mathbb{Z}/2)$, where $\phi=\psi^{q}_{BSpin(9)}$ and $q$ is an odd prime power. And we have ###### Proposition 3.6. For $\phi=\psi^{q}_{BSpin(9)}$ the unstable Adams operation of degree an odd prime power $q$, $H^{*}(\mathbb{L}_{\phi}BSpin(9);\mathbb{Z}/2)$ is isomorphic to $H^{*}(LBSpin(9);\mathbb{Z}/2)$ as algebras over $\mathcal{A}_{2}$. ###### Proof. In dimensions lower than $9$, we can construct a section $r_{BSpin(9)}$ completely parallel to the case of $BSpin(7)$, namely $r_{BSpin(9)}(w_{i})=u_{i}\ (i=4,6,7,8)$. Now the conditions (*) are satisfied and the Wang sequences (A) for $R=\mathbb{Z}/2$ and $\mathbb{Z}/4$ split to the short exact sequences $0\to H^{*-1}(BSpin(9);\mathbb{Z}/2)\xrightarrow{\delta}H^{*}(\mathbb{T}_{\phi}BSpin(9);\mathbb{Z}/2)\xrightarrow{\iota^{*}}H^{*}(BSpin(9);\mathbb{Z}/2)\to 0,$ $0\to H^{n-1}(BSpin(9);\mathbb{Z}/4)\xrightarrow{\delta}H^{n}(\mathbb{T}_{\phi}BSpin(9);\mathbb{Z}/4)\xrightarrow{\iota^{*}}H^{n-1}(BSpin(9);\mathbb{Z}/4)\to 0,$ where $n\equiv 0\mod 4$. Using the Wang sequence for $\mathbb{Z}/4$ and the Bockstein sequence, we can choose an element $h^{\prime}_{16}\in\ker(Sq^{1})\subset(\iota^{*})^{-1}(e_{16})\subset H^{16}(\mathbb{T}_{\phi}BSpin(9);\mathbb{Z}/2)$ by the same observation for $u_{8}$ in the proof of $BSpin(7)$. Then by the Wang sequence for $\mathbb{Z}/2$, we have $Sq^{2}h^{\prime}_{16}=\epsilon_{1}\delta(w_{4}w_{6}w_{7})$ since $H^{17}(BSpin(9);\mathbb{Z}/2)\simeq\mathbb{Z}/2$ is generated by $w_{4}w_{6}w_{7}$ and $\iota^{*}(Sq^{2}h^{\prime}_{16})=Sq^{2}e_{16}=0$. Then $Sq^{2}Sq^{2}h^{\prime}_{16}=\epsilon_{1}\delta(Sq^{2}(w_{4}w_{6}w_{7}))=\epsilon_{1}\delta(w_{6}^{2}w_{7})$. Since $Sq^{2}Sq^{2}=Sq^{3}Sq^{1}$ by Adem relation and $Sq^{1}h^{\prime}_{16}=0$, $\epsilon_{1}$ must be $0$. Similarly we have $Sq^{4}h^{\prime}_{16}=\epsilon_{2}\delta(w_{4}^{3}w_{7})+\epsilon_{3}\delta(w_{6}^{2}w_{7})+\epsilon_{4}\delta(w_{4}w_{7}w_{8})$. Then we have $Sq^{4}Sq^{4}h^{\prime}_{16}=\epsilon_{2}\delta(w_{4}w_{6}^{2}w_{7})+\epsilon_{3}\delta(w_{4}w_{6}^{2}w_{7})+\epsilon_{4}\delta(w_{4}^{2}w_{7}w_{8})$. By Adem relation we have $Sq^{4}Sq^{4}h^{\prime}_{16}=(Sq^{7}Sq^{1}+Sq^{6}Sq^{2})h^{\prime}_{16}=0$. Therefore we have $\epsilon_{2}=\epsilon_{3},\epsilon_{4}=0$. Put $h_{16}=h^{\prime}_{16}+\epsilon_{2}\delta(w_{4}^{2}w_{7})$, then we have $Sq^{4}h_{16}=0$ since $Sq^{4}(w_{4}^{2}w_{7})=w_{4}^{3}w_{7}+w_{6}^{2}w_{7}$. Note that we also have $Sq^{i}h_{16}=0\ (i=1,2)$ since $Sq^{i}(w_{4}^{2}w_{7})=0\ (i=1,2)$. Similarly we have $Sq^{8}h_{16}=u_{8}h_{16}+u_{4}^{2}h_{16}+\epsilon_{5}\delta(w_{4}^{4}w_{7})+\epsilon_{6}\delta(w_{4}^{2}w_{7}w_{8})+\epsilon_{7}\delta(w_{4}w_{6}^{2}w_{7})+\epsilon_{8}\delta(w_{7}w_{8}^{2})+\epsilon_{9}\delta(w_{7}e_{16})$. By Adem relation $Sq^{8}Sq^{8}h_{16}=0$ and we have $\epsilon_{5}=\epsilon_{7}=\epsilon_{9}=0,\epsilon_{6}=\epsilon_{8}$. Replacing $h_{16}$ by $h_{16}+\epsilon_{6}\delta(w_{7}w_{8})$ we have $Sq^{8}h_{16}=u_{8}h_{16}+u_{4}^{2}h_{16}$ and $Sq^{i}h_{16}=0\ (i<8)$. Define $r_{BSpin(9)}(e_{16})=h_{16}$, then $r_{BSpin(9)}$ is a section of $\iota^{*}$ which commutes with the Steenrod operations. ∎ ## 4\. Computations for $G=F_{4}$ The same method applies for the case of simply connected, simple exceptional compact Lie group $F_{4}$. We first recall the mod $2$ cohomology of $BF_{4}$. Denote by $i$ the classifying map of the canonical inclusion $Spin(9)\hookrightarrow F_{4}$. Kono determined the mod $2$ cohomology algebra over $\mathcal{A}_{2}$ in [Ko] as follows: $H^{*}(BF_{4};\mathbb{Z}/2)=\mathbb{Z}[x_{4},x_{6},x_{7},x_{16},x_{24}],$ where $i^{*}(x_{4})=w_{4},i^{*}(x_{6})=w_{6},i^{*}(x_{7})=w_{7},i^{*}(x_{16})=e_{16}+w_{8}^{2},i^{*}(x_{24})=w_{8}e_{16}$ and the action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rcccccc}&x_{4}&x_{6}&x_{7}&x_{16}&x_{24}\\\ Sq^{1}&0&x_{7}&0&0&0\\\ Sq^{2}&x_{6}&0&0&0&0\\\ Sq^{4}&x_{4}^{2}&x_{4}x_{6}&x_{4}x_{7}&0&x_{4}x_{24}\\\ Sq^{8}&0&0&0&x_{24}+x_{4}^{2}x_{16}&x_{4}^{2}x_{24}\\\ Sq^{16}&0&0&0&x_{16}^{2}&x_{16}x_{24}+x_{4}x_{6}^{2}x_{24}.\end{array}$ Then we can compute the mod $2$ cohomology of classifying space of the free loop group $LF_{4}$ just as in the same manner in the previous section. ###### Proposition 4.1. $H^{*}(LBF_{4};\mathbb{Z}/2)=\mathbb{Z}/2[v_{4},v_{6},v_{7},v_{16},v_{24},y_{3},y_{5},y_{15},y_{23}]/I$ $(|v_{i}|=i,|y_{i}|=i)$, where $I$ is the ideal generated by $\\{y_{5}^{2}+y_{3}v_{7}+v_{4}y_{3}^{2},y_{3}^{4}+v_{6}y_{3}^{2}+y_{5}v_{7},y_{15}^{2}+v_{7}y_{23}+v_{24}y_{3}^{2},y_{23}^{2}+y_{3}^{2}v_{16}v_{24}+v_{7}v_{24}y_{15}+v_{7}v_{16}y_{23}\\}.$ The action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rccccc}&v_{4}&v_{6}&v_{7}&v_{16}&v_{24}\\\ Sq^{1}&0&v_{7}&0&0&0\\\ Sq^{2}&v_{6}&0&0&0&0\\\ Sq^{4}&v_{4}^{2}&v_{4}v_{6}&v_{4}v_{7}&0&v_{4}v_{24}\\\ Sq^{8}&0&0&0&v_{24}+v_{4}^{2}v_{16}&v_{4}^{2}v_{24}\\\ Sq^{16}&0&0&0&v_{16}^{2}&v_{16}v_{24}+v_{4}v_{6}^{2}v_{24}\\\ \hline\cr\\\ &y_{3}&y_{5}&y_{15}&y_{23}\\\ Sq^{1}&0&y_{3}^{2}&0&0\\\ Sq^{2}&y_{5}&0&0&0\\\ Sq^{4}&0&y_{3}v_{6}+v_{4}y_{5}&0&y_{3}v_{24}+v_{4}y_{23}\\\ Sq^{8}&0&0&y_{23}+v_{4}^{2}y_{15}&v_{4}^{2}y_{23}\\\ Sq^{16}&0&0&0&J_{2}\end{array}$ where $J_{2}=v_{24}y_{15}+v_{16}y_{23}+y_{3}v_{6}^{2}v_{24}+v_{4}v_{6}^{2}y_{23}$. ###### Proof. We take $v_{i}=e^{*}(x_{i}),\ y_{i-1}=\tilde{\sigma}_{Id}(x_{i})\quad(i=4,6,7,16,24)$. Then In dimensions lower than $9$, the calculation is completely parallel to the the case of $BSpin(9)$. And the rest are as follows: $\displaystyle Sq^{1}y_{15}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{1}x_{16})=0$ $\displaystyle Sq^{2}y_{15}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{2}x_{16})=0$ $\displaystyle Sq^{4}y_{15}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{4}x_{16})=0$ $\displaystyle Sq^{8}y_{15}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{8}x_{16})=\tilde{\sigma}_{Id}(x_{24}+x_{4}^{2}x_{16})=y_{23}+v_{4}^{2}y_{15}$ $\displaystyle Sq^{1}y_{23}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{1}x_{24})=0$ $\displaystyle Sq^{2}y_{23}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{2}x_{24})=0$ $\displaystyle Sq^{4}y_{23}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{4}x_{24})=\tilde{\sigma}_{Id}(x_{4}x_{24})=y_{3}v_{24}+v_{4}y_{23}$ $\displaystyle Sq^{8}y_{23}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{8}x_{24})=\tilde{\sigma}_{Id}(x_{4}^{2}x_{24})=v_{4}^{2}y_{23}$ $\displaystyle Sq^{16}y_{23}$ $\displaystyle=\tilde{\sigma}_{Id}(Sq^{16}x_{24})=\tilde{\sigma}_{Id}(x_{16}x_{24}+x_{4}x_{6}^{2}x_{24})=y_{15}v_{24}+v_{16}y_{23}+y_{3}v_{6}^{2}v_{24}+v_{4}v_{6}^{2}y_{23}$ $\displaystyle y_{15}^{2}$ $\displaystyle=Sq^{15}y_{15}=\tilde{\sigma}_{Id}(Sq^{15}x_{16})=\tilde{\sigma}_{Id}(Sq^{7}Sq^{8}x_{16})=\tilde{\sigma}_{Id}(Sq^{7}x_{24})=\tilde{\sigma}_{Id}(x_{7}x_{24})=v_{7}y_{23}+y_{3}^{2}v_{24}$ $\displaystyle y_{23}^{2}$ $\displaystyle=Sq^{23}y_{23}=\tilde{\sigma}_{Id}(Sq^{23}x_{24})=\tilde{\sigma}_{Id}(Sq^{7}Sq^{16}x_{24})=\tilde{\sigma}_{Id}(Sq^{7}(x_{16}x_{24}+x_{4}x_{6}^{2}x_{24}))$ $\displaystyle\tilde{\sigma}_{Id}(x_{16}Sq^{7}(x_{24})+Sq^{7}Sq^{4}(x_{6}^{2}x_{24})))=\tilde{\sigma}_{Id}(x_{7}x_{16}x_{24})=y_{3}^{2}v_{16}v_{24}+v_{7}y_{15}v_{24}+v_{7}v_{16}y_{23}.$ ∎ A finite Chevalley group of type $F_{4}(q)$ has the mod $2$ cohomology algebra over $\mathcal{A}_{2}$ isomorphic to $H^{*}(\mathbb{L}_{\phi}BF_{4};\mathbb{Z}/2)$, where $\phi=\psi^{q}_{BF_{4}}$ and $q$ is an odd prime power. And we have ###### Proposition 4.2. For $\phi=\psi^{q}_{BF_{4}}$ the unstable Adams operation of degree an odd prime power $q$, $H^{*}(\mathbb{L}_{\phi}BF_{4};\mathbb{Z}/2)$ is isomorphic to $H^{*}(LBF_{4};\mathbb{Z}/2)$ as algebras over $\mathcal{A}_{2}$. ###### Proof. By [JMO] the following diagram is homotopy commutative $\lx@xy@svg{\hbox{\raise 0.0pt\hbox{\kern 25.8925pt\hbox{\ignorespaces\ignorespaces\ignorespaces\hbox{\vtop{\kern 0.0pt\offinterlineskip\halign{\entry@#!@&&\entry@@#!@\cr&\\\&\crcr}}}\ignorespaces{\hbox{\kern-25.8925pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{BSpin(9)^{\wedge}_{2}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 31.64806pt\raise 7.11111pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.61111pt\hbox{$\scriptstyle{\psi^{q}_{BSpin(9)}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 49.8925pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 0.0pt\raise-20.77777pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-2.30556pt\hbox{$\scriptstyle{(i)^{\wedge}_{2}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 0.0pt\raise-29.49998pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 49.8925pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{BSpin(9)^{\wedge}_{2}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 75.785pt\raise-20.77777pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-2.30556pt\hbox{$\scriptstyle{(i)^{\wedge}_{2}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 75.785pt\raise-29.49998pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern-18.1087pt\raise-41.55554pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{(BF_{4})^{\wedge}_{2}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces{\hbox{\kern 31.64806pt\raise-34.49554pt\hbox{{}\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\hbox{\hbox{\kern 0.0pt\raise-1.66223pt\hbox{$\scriptstyle{\psi^{q}_{BF_{4}}}$}}}\kern 3.0pt}}}}}}\ignorespaces{\hbox{\kern 57.6763pt\raise-41.55554pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 57.6763pt\raise-41.55554pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{(BF_{4})^{\wedge}_{2}}$}}}}}}}\ignorespaces}}}}\ignorespaces,$ where $X^{\wedge}_{2}$ is the Bousfield-Kan $2$-completion of $X$. By the naturality of the construction of the twisted tube, there is a map $\mathbb{T}_{\phi}(i)^{\wedge}_{2}:\mathbb{T}_{\phi}BSpin(9)^{\wedge}_{2}\to\mathbb{T}_{\phi}(BF_{4})^{\wedge}_{2}$, which makes the following diagram commute: $\textstyle{0\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{H^{*-1}(BSpin(9);\mathbb{Z}/2)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{H^{*}(\mathbb{T}_{\phi}BSpin(9);\mathbb{Z}/2)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\iota^{*}_{BSpin(9)}}$$\textstyle{H^{*}(BSpin(9);\mathbb{Z}/2)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{0}$$\textstyle{0\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{H^{*-1}(BF_{4};\mathbb{Z}/2)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{i^{*}}$$\textstyle{H^{*}(\mathbb{T}_{\phi}BF_{4};\mathbb{Z}/2)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\iota^{*}_{BF_{4}}}$$\scriptstyle{(\mathbb{T}_{\phi}(i)^{\wedge}_{2})^{*}}$$\textstyle{H^{*}(BF_{4};\mathbb{Z}/2)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{i^{*}}$$\textstyle{0,}$ where the horizontal lines are the Wang sequences. Since $i^{*}$ is injective for degrees less than $48$, so is $(\mathbb{T}_{\phi}(i)^{\wedge}_{2})^{*}$. Then we can define a section $r_{BF_{4}}$ of $\iota^{*}_{BF_{4}}$ as $r_{BF_{4}}(x_{i})=((\mathbb{T}_{\phi}(i)^{\wedge}_{2})^{*})^{-1}\circ r_{BSpin(9)}\circ i^{*}(x_{i}),\quad(i=4,6,7,16,24),$ where $r_{BSpin(9)}$ is the section of $\iota^{*}_{BSpin(9)}$ constructed in the previous section. For the commutativity with the Steenrod operations, we have only to consider the degrees less than or equal to $|Sq^{16}x_{24}|=40$, thus we have that $r$ commutes with the Steenrod algebra. ∎ ## 5\. Computations for $G=DI(4)$ In [DW], Dwyer and Wilkerson constructed a finite loop space $DI(4)$, whose classifying space $BDI(4)$ has the mod $2$ cohomology isomorphic to the mod $2$ Dickson invariant of rank $4$, that is, $H^{*}(BDI(4);\mathbb{Z}/2)=\mathbb{Z}/2[x_{8},x_{12},x_{14},x_{15}]$, where $|x_{j}|=j$. The action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rcccc}&x_{8}&x_{12}&x_{14}&x_{15}\\\ Sq^{1}&0&0&x_{15}&0\\\ Sq^{2}&0&x_{14}&0&0\\\ Sq^{4}&x_{12}&0&0&0\\\ Sq^{8}&x_{8}^{2}&x_{8}x_{12}&x_{8}x_{14}&x_{8}x_{15}.\\\ \end{array}$ In ([B]), Benson defined the classifying space $BSol(q)$ of an exotic $2$-local finite group as $L_{\psi^{q}}BDI(4)$, where $q$ is an odd prime power and $\psi^{q}$ is the unstable Adams operation of degree $q$ constructed in [N]. Recently in [G], Grbić calculated the mod 2 cohomology of $BSol(q)$ over $\mathcal{A}_{2}$ by using the Eilenberg-Moore spectral sequence. Here we confirm it by the same method in previous sections. To do so, we first calculate the mod $2$ cohomology of the free loop space $LBDI(4)$. ###### Proposition 5.1. $H^{*}(LBDI(4);\mathbb{Z}/2)=\mathbb{Z}/2[v_{8},v_{12},v_{14},v_{15},y_{7},y_{11},y_{13}]/I$ $(|v_{i}|=i,|y_{i}|=i)$, where $I$ is the ideal generated by $\\{y_{7}^{4}+y_{13}v_{15}+v_{14}y_{7}^{2},y_{11}^{2}+v_{7}y_{15}+v_{8}y_{7}^{2},y_{13}^{2}+y_{11}v_{15}+v_{12}y_{7}^{2}\\}.$ The action of $\mathcal{A}_{2}$ is determined by: $\begin{array}[]{rccccccccc}&v_{8}&v_{12}&v_{14}&v_{15}&y_{7}&y_{11}&y_{13}\\\ Sq^{1}&0&0&v_{15}&0&0&0&y_{7}^{2}\\\ Sq^{2}&0&v_{14}&0&0&0&y_{13}&0\\\ Sq^{4}&v_{12}&0&0&0&y_{11}&0&0&\\\ Sq^{8}&v_{8}^{2}&v_{8}v_{12}&v_{8}v_{14}&v_{8}v_{15}&0&v_{8}y_{11}+y_{7}v_{12}&v_{8}y_{13}+y_{7}v_{14}.\\\ \end{array}$ ###### Remark 3. Kuribayashi has also this result in [Ku]. ###### Proof. We take $v_{i}=e^{*}(x_{i}),\ y_{i-1}=\tilde{\sigma}_{Id}(x_{i})\quad(i=8,12,14,15)$. Just as in the previous calculations, we have $\displaystyle y_{7}^{2}=Sq^{7}y_{7}=\tilde{\sigma}_{Id}(Sq^{7}x_{8})=\tilde{\sigma}_{Id}(x_{15})=y_{14}$ $\displaystyle Sq^{1}y_{i}=\tilde{\sigma}_{Id}(Sq^{1}x_{i+1})=0\ (i=7,11)$ $\displaystyle Sq^{1}y_{13}=\tilde{\sigma}_{Id}(Sq^{1}x_{14})=\tilde{\sigma}_{Id}(x_{15})=y_{14}=y_{7}^{2}$ $\displaystyle Sq^{2}y_{i}=\tilde{\sigma}_{Id}(Sq^{2}x_{i+1})=0\ (i=7,13)$ $\displaystyle Sq^{2}y_{11}=\tilde{\sigma}_{Id}(Sq^{2}x_{12})=\tilde{\sigma}_{Id}(x_{14})=y_{13}$ $\displaystyle Sq^{4}y_{i}=\tilde{\sigma}_{Id}(Sq^{4}x_{i+1})=0\ (i=11,13)$ $\displaystyle Sq^{4}y_{7}=\tilde{\sigma}_{Id}(Sq^{4}x_{8})=\tilde{\sigma}_{Id}(x_{12})=y_{11}$ $\displaystyle Sq^{8}y_{7}=\tilde{\sigma}_{Id}(Sq^{8}x_{8})=0$ $\displaystyle Sq^{8}y_{11}=\tilde{\sigma}_{Id}(Sq^{8}x_{12})=\tilde{\sigma}_{Id}(x_{8}x_{12})=y_{7}v_{12}+v_{8}y_{11}$ $\displaystyle Sq^{8}y_{13}=\tilde{\sigma}_{Id}(Sq^{8}x_{14})=\tilde{\sigma}_{Id}(x_{8}x_{14})=y_{7}v_{14}+v_{8}y_{13}$ $\displaystyle y_{11}^{2}=Sq^{11}y_{11}=\tilde{\sigma}_{Id}(Sq^{11}x_{12})=\tilde{\sigma}_{Id}(Sq^{1}Sq^{2}Sq^{8}v_{12})=\tilde{\sigma}_{Id}(x_{8}x_{15})=v_{8}y_{7}^{2}+y_{7}v_{15}$ $\displaystyle y_{13}^{2}=Sq^{13}y_{13}=\tilde{\sigma}_{Id}(Sq^{13}v_{14})=\tilde{\sigma}_{Id}((Sq^{5}Sq^{8}+Sq^{11}Sq^{2})x_{14})$ $\displaystyle\qquad=\tilde{\sigma}_{Id}(Sq^{5}x_{8}x_{14})=\tilde{\sigma}_{Id}(x_{12}x_{15})=y_{11}v_{15}+v_{12}y_{7}^{2}$ $\displaystyle y_{7}^{4}=y_{14}^{2}=Sq^{14}y_{14}=\tilde{\sigma}_{Id}(Sq^{14}x_{15})=\tilde{\sigma}_{Id}(x_{14}x_{15})=y_{13}v_{15}+v_{14}y_{7}^{2}.$ ∎ Now we proceed to show that the mod 2 cohomology of $BSol(q)=L_{\psi^{q}}BDI(4)$ over $\mathcal{A}_{2}$ is isomorphic to that of $LBDI(4)$. ###### Proposition 5.2. For $\phi=\psi^{q}$ the unstable Adams operation of degree an odd prime power $q$, $H^{*}(\mathbb{L}_{\phi}BDI(4);\mathbb{Z}/2)$ is isomorphic to $H^{*}(LBDI(4);\mathbb{Z}/2)$ as algebras over $\mathcal{A}_{2}$. ###### Proof. Using the Wang sequence for $\mathbb{Z}/4$ and the Bockstein sequence, we can choose an element $u_{8}\in\ker(Sq^{1})\cap\ker(Sq^{1}Sq^{4})\subset(\iota^{*})^{-1}(x_{8})$ by the same observation in the proof of Proposition 3.2. Put $u_{12}=Sq^{4}u_{8},u_{14}=Sq^{2}u_{12}$ and $u_{15}=Sq^{1}u_{14}$. Then we have $Sq^{1}u_{i}=0\ (i=8,12,15)$. And by dimensional reason, we have $Sq^{2}u_{8}=0$. Therefore $Sq^{4}u_{12}=Sq^{4}Sq^{4}u_{8}=(Sq^{7}Sq^{1}+Sq^{6}Sq^{2}+Sq^{5}Sq^{3})u_{8}=0$. Since $H^{19}(BDI(4);\mathbb{Z}/2)=0$ and $\iota^{*}(Sq^{8}u_{12})=x_{8}x_{12}$, using the Wang sequence for $\mathbb{Z}/2$ we have $Sq^{8}u_{12}=u_{8}u_{12}$. Other operations are calculated as follows: $\displaystyle Sq^{2}u_{14}=Sq^{2}Sq^{2}u_{12}=0$ $\displaystyle Sq^{4}u_{14}=Sq^{4}Sq^{6}u_{8}=Sq^{2}Sq^{8}u_{8}=0$ $\displaystyle Sq^{8}u_{14}=Sq^{8}Sq^{2}u_{12}=(Sq^{4}Sq^{6}+Sq^{2}Sq^{8})u_{12}=u_{8}u_{14}$ $\displaystyle Sq^{2}u_{15}=Sq^{2}Sq^{7}u_{8}=Sq^{9}u_{8}=0$ $\displaystyle Sq^{4}u_{15}=Sq^{4}Sq^{7}u_{8}=Sq^{11}u_{8}=0$ $\displaystyle Sq^{8}u_{15}=Sq^{8}Sq^{1}u_{14}=(Sq^{9}+Sq^{2}Sq^{7})u_{14}=Sq^{1}Sq^{8}u_{14}=u_{8}u_{15}.$ Hence we can construct a section $r$ by $x_{i}\mapsto u_{i}$. ∎ ## References * [B] D. Benson, Cohomology of sporadic groups, finite loop spaces, and the Dickson invariants, Geometry and cohomology in group theory, London Math. Soc. Lecture notes ser. 252, Cambridge Univ. Press (1998), 10 – 23. * [BK] A.K.Bousfield and D.M.Kan, Homotopy limits, completions and localizations, Lecture Notes in Mathematics, Vol. 304, Springer-Verlag, Berlin-New York, 1972. * [BM] C. Broto and J. Møller, Homotopy finite Chevalley versions of p-compact groups, preprint. * [DW] W.Dwyer and C.Wilkerson, A new finite loop space at prime two, J. Amer. Math. Soc. 6 (1993), 37–64. * [F] E.M.Friedlander, Étale homotopy of simplicial schemes, Ann. of Math. Studies 104, Princeton Univ. Press, Princeton, 1963. * [G] J.Grbić, The cohomology of exotic 2-local finite groups, Manuscripta Math. 120 (2006), no. 3, 307–318. * [JMO] S.Jackowski, J.McClure and B.Oliver, Self-homotopy equivalences of classifying spaces of compact connected Lie groups, Fund. Math. 147 (1995), no. 2, 99–126. * [KK] D.Kishimoto and A.Kono, Cohomology of free and twisted loop spaces, preprint. * [Ko] A.Kono, On the 2-rank of compact connected Lie groups, J. Math. Kyoto Univ. 17 (1977), no. 1, 1–18. * [KK2] A.Kono and K.Kozima, The adjoint action of the Dwyer-Wilkerson $H$-space on its loop space, J. Math. Kyoto Univ. 35 (1995), no. 1, 53–62. * [Ku] K.Kuribayashi, Module derivations and the adjoint action of a finite loop space, J. Math. Kyoto Univ. 39 (1999), no. 1, 67–85. * [N] D.Notbohm, On the 2-compact group $DI(4)$, J.Reine Angew. Math. 555(2003), 163–185. * [Q] D.Quillen, The mod $2$ cohomology rings of extra-special $2$-groups and the spinor groups, Math. Ann. 194 (1971), 197–212. * [S] R.Solomon, Finite groups with Sylow 2-subgroups of type .3, J. Algebra 28 (1974), 182 – 198. * [VV] A.Vavpetič and A.Viruel, On the homotopy type of the classifying space of the exceptional Lie group $F_{4}$, Manuscripta Math. 107 (2002), no. 4, 521–540. * [W] C.Wilkerson, Self-maps of classifying spaces, Localization in group theory and homotopy theory, and related topics, Lecture Notes in Math., Vol. 418, Springer, Berlin, 1974, 150–157.

arxiv-papers

2024-09-04T02:48:53.290136

0711.0637

Rep. Prog. Phys. 70 (2007) 2067-2148 # Symmetry breaking and quantum correlations in finite systems: Studies of quantum dots and ultracold Bose gases and related nuclear and chemical methods Constantine Yannouleas and Uzi Landman School of Physics, Georgia Institute of Technology, Atlanta, Georgia 30332-0430, USA Constantine.Yannouleas@physics.gatech.edu Uzi.Landman@physics.gatech.edu ###### Abstract Investigations of emergent symmetry breaking phenomena occurring in small finite-size systems are reviewed, with a focus on the strongly correlated regime of electrons in two-dimensional semicoductor quantum dots and trapped ultracold bosonic atoms in harmonic traps. Throughout the review we emphasize universal aspects and similarities of symmetry breaking found in these systems, as well as in more traditional fields like nuclear physics and quantum chemistry, which are characterized by very different interparticle forces. A unified description of strongly correlated phenomena in finite systems of repelling particles (whether fermions or bosons) is presented through the development of a two-step method of symmetry breaking at the unrestricted Hartree-Fock level and of subsequent symmetry restoration via post Hartree-Fock projection techniques. Quantitative and qualitative aspects of the two-step method are treated and validated by exact diagonalization calculations. Strongly-correlated phenomena emerging from symmetry breaking include: (I) Chemical bonding, dissociation, and entanglement (at zero and finite magnetic fields) in quantum dot molecules and in pinned electron molecular dimers formed within a single anisotropic quantum dot, with potential technological applications to solid-state quantum-computing devices. (II) Electron crystallization, with particle localization on the vertices of concentric polygonal rings, and formation of rotating electron molecules (REMs) in circular quantum dots. Such electron molecules exhibit ro- vibrational excitation spectra, in analogy with natural molecules. (III) At high magnetic fields, the REMs are described by parameter-free analytic wave functions, which are an alternative to the Laughlin and composite-fermion approaches, offering a new point of view of the fractional quantum Hall regime in quantum dots (with possible implications for the thermodynamic limit). (IV) Crystalline phases of strongly repelling bosons. In rotating traps and in analogy with the REMs, such repelling bosons form rotating boson molecules (RBMs). For a small number of bosons, the RBMs are energetically favored compared to the Gross-Pitaevskii solutions describing vortex formation. We discuss the present status concerning experimental signatures of such strongly correlated states, in view of the promising outlook created by the latest experimental improvements that are achieving unprecedented control over the range and strength of interparticle interactions. ###### Contents 1. 1 Introduction 1. 1.1 Preamble 2. 1.2 Spontaneous symmetry breaking: confined geometries versus extended systems 3. 1.3 Historical background from nuclear physics and chemistry 4. 1.4 Scope of the review 5. 1.5 Using a hierarchy of approximations versus probing of exact solutions 6. 1.6 Experimental signatures of quantum correlations 7. 1.7 Plan of the report 2. 2 Symmetry breaking and subsequent symmetry restoration for electrons in confined geometries: Theoretical framework 1. 2.1 Mean-field description and unrestricted Hartree-Fock 1. 2.1.1 The self-consistent Pople-Nesbet equations. 2. 2.1.2 The Wigner parameter and classes of spontaneous symmetry breaking solutions. 3. 2.1.3 Unrestricted Hartree-Fock solutions representing Wigner molecules. 4. 2.1.4 Unrestricted Hartree-Fock solutions representing electron puddles. 5. 2.1.5 Unrestricted Hartree-Fock solutions representing pure spin density waves within a single quantum dot. 2. 2.2 Projection techniques and post-Hartree-Fock restoration of broken symmetries 1. 2.2.1 The REM microscopic method in medium and high magnetic field. 2. 2.2.2 Group structure and sequences of magic angular momenta. 3. 2.3 The symmetry breaking dilemma and density functional theory 4. 2.4 More on symmetry restoration methods 1. 2.4.1 Symmetry restoration via random phase approximation. 2. 2.4.2 The generator coordinate method. 3. 3 Symmetry breaking and subsequent symmetry restoration for neutral and charged bosons in confined geometries: Theoretical framework 1. 3.1 Symmetry breaking for bosons, Gross-Pitaevskii wave functions, and permanents 2. 3.2 Repelling bosons in harmonic traps: Restoration of broken symmetry. 4. 4 Other many-body methods 1. 4.1 Exact diagonalization methods: Theoretical framework 1. 4.1.1 An example involving spin-resolved CPDs. 2. 4.2 Particle localization in Monte-Carlo approaches 5. 5 The strongly correlated regime in two-dimensional quantum dots: The two-electron problem and its significance 1. 5.1 Two-electron elliptic dot at low magnetic fields 1. 5.1.1 Generalized Heitler-London approach. 2. 5.1.2 Exact diagonalization. 3. 5.1.3 Results and comparison with measurements. 4. 5.1.4 Degree of entanglement. 2. 5.2 Two-electron circular dots at zero magnetic field 3. 5.3 Historical significance of the two-electron problem 6. 6 Rotating electron molecules in two-dimensional quantum dots under a strong magnetic field: The case of the lowest Landau level ($\omega_{c}/2\omega_{0}\rightarrow\infty$) 1. 6.1 REM analytic trial wave functions 2. 6.2 Yrast rotational band in the lowest Landau level 3. 6.3 Inconsistencies of the composite-fermion view for semiconductor quantum dots. 4. 6.4 REM versus Laughlin wave functions: Conditional probability distributions and multiplicity of zeroes 7. 7 Rotating electron molecules in two-dimensional quantum dots under a strong, but finite external magnetic field ($\omega_{c}/2\omega_{0}>1$) 1. 7.1 Ground-state energies in medium and high magnetic field 2. 7.2 The case of $N=11$ electrons. 3. 7.3 Approximate analytic expression for the yrast-band spectra 4. 7.4 Possible implications for the thermodynamic limit 8. 8 Bosonic molecules in rotating traps: Original results and applications 1. 8.1 Variational description of rotating boson molecules 2. 8.2 Exact diagonalization for bosons in the lowest Landau level 1. 8.2.1 The case of $N=6$ bosons in the lowest Landau level. 9. 9 Summary 10. A 1. A.1 Two-dimensional isotropic oscillator in a perpendicular magnetic field 2. A.2 Two-dimensional rotating harmonic oscillator [Symmetry breaking and quantum correlations] ## 1 Introduction ### 1.1 Preamble Fermionic or bosonic particles confined in manmade devices, i.e., electrons in two-dimensional (2D) quantum dots (QDs, referred to also as artificial atoms) or ultracold atoms in harmonic traps, can localize and form structures with molecular, or crystalline, characteristics. These molecular states of localized particles differ in an essential way from the electronic-shell- structure picture of delocalized electrons filling successive orbitals in a central-mean-field potential (the Aufbau principle), familiar from the many- body theory of natural atoms and the Mendeleev periodic table; they also present a different regime from that exhibited by a Bose-Einstein condensate (BEC, associated often with the mean-field Gross-Pitaevskii equation). The molecular states originate from strong correlations between the constituent repelling particles and they are called electron (and often Wigner) or boson molecules. Such molecular states forming within a single confining potential well constitute new phases of matter and allow for investigations of novel strongly-correlated phenomena arising in physical systems with a range of materials’ characteristics unavailable experimentally (and theoretically unexplored) until recently. One example is the range of values of the socalled Wigner parameter (denoted as $R_{W}$ for charged particles and $R_{\delta}$ for neutral ones, see section 2.1.2) which expresses the relative strength of the two-body repulsion and the one-particle kinetic energy, reflecting and providing a measure of the strength of correlations in the system under study. For the two-dimensional systems which we discuss here, these values are often larger than the corresponding ones for natural atoms and molecules. Other research opportunities offered by the quantum-dot systems are related to their relatively large (spatial) size (arising from a small electron effective mass and large dielectric constant), which allows the full range of orbital magnetic effects to be covered for magnetic fields that are readily attained in the laboratory (less than 40 T). In contrast, for natural atoms and molecules, magnetic fields of sufficient strength (i.e., larger than $10^{5}$ T) to produce novel phenomena related to orbital magnetism (beyond the perturbative regime) are known to occur only in astrophysical environments (e.g., on the surface of neutron stars) [1]. For ultracold gases, a similar extraordinary physical regime can be reached via the fast rotation of the harmonic trap. In addition to the fundamental issues unveiled through investigations of molecular states in quantum dots, these strongly-correlated states are of technological significance because of the potential use of manmade nanoscale systems for the implementation of qubits and quantum logic gates in quantum computers. The existence of electron and boson molecules is supported by large-scale exact diagonalization (EXD) calculations, which provide the ultimate theoretical test. The discovery of these “crystalline” states has raised important fundamental aspects, including the nature of quantum phase transitions and the conceptual issues relating to spontaneous symmetry breaking (SSB) in small finite-size systems. The present report addresses primarily the physics, theoretical description, and fundamental many-body aspects of molecular (crystalline) states in small systems. For a comprehensive description of the electronic-shell-structure regime (Aufbau-principle regime) in quantum dots and of Bose-Einstein condensates in harmonic traps, see the earlier reviews by Kouwenhoven et al[2] (QDs), Reimann and Manninen [3] (QDs), Dalfovo et al[4] (BECs), and Leggett [5] (BECs). Furthermore, in larger quantum dots, the symmetries of the external confinement that lead to shell structure are broken, and such dots exhibit mesoscopic fluctuations and interplay between single-particle quantum chaos [6] and many-body correlations. For a comprehensive description of this mesoscopic regime in quantum dots, see the reviews by Beenakker [7] and Alhassid [8]. ### 1.2 Spontaneous symmetry breaking: confined geometries versus extended systems Spontaneous symmetry breaking is a ubiquitous phenomenon in the macroscopic world. Indeed, there is an abundance of macroscopic systems and objects that are observed, or can be experimentally prepared, with effective many-body ground states whose symmetry is lower than the symmetry of the underlying many-body quantum-mechanical Hamiltonian; one says that in such cases the system lowers its energy through spontaneous symmetry breaking, resulting in a state of lower symmetry and higher order. It is important to stress that macroscopic SSB strongly suppresses quantum fluctuations and thus it can be described appropriately by a set of non-linear mean-field equations for the “order parameter.” The appearance of the order parameter is governed by bifurcations associated with the non-linearity of the mean-field equations and has led to the notion of “emergent phenomena,” a notion that helped promote condensed-matter physics as a branch of physics on a par with high-energy particle physics (in reference to the fundamental nature of the pursuit in these fields; see the seminal paper by Anderson in Ref. [9]). Our current understanding of the physics of SSB in the thermodynamic limit (when the number of particles $N\rightarrow\infty$) owes a great deal to the work of Anderson [10], who suggested that the broken-symmetry state can be safely taken as the effective ground state. In arriving at this conclusion Anderson invoked the concept of (generalized) rigidity. As a concrete example, one would expect a crystal to behave like a macroscopic body, whose Hamiltonian is that of a heavy rigid rotor with a low-energy excitation spectrum $L^{2}/2{\cal J}$ of angular-momentum $(L)$ eigenstates, with the moment of inertia ${\cal J}$ being of order $N$ (macroscopically large when $N\rightarrow\infty$). The low-energy excitation spectrum of this heavy rigid rotor above the ground-state ($L=0$) is essentially gapless (i.e., continuous). Thus although the formal ground state posseses continuous rotational symmetry (i.e., $L=0$), “there is a manifold of other states, degenerate in the $N\rightarrow\infty$ limit, which can be recombined to give a very stable wave packet with essentially the nature” of the broken-symmetry state (see p 44 in Ref. [10]). As a consequence of the “macroscopic heaviness” as $N\rightarrow\infty$, the relaxation of the system from the wave packet state (i.e., the broken-symmetry state) to the exact symmetrical ground state becomes exceedingly long. Consequently, in this limit, when symmetry breaking occurs, there is practically no need to follow up with a symmetry restoration step; that is the symmetry-broken state is admissible as an effective ground state. The present report addresses the much less explored question of symmetry breaking in finite condensed-matter systems with a small number of particles. For small systems, spontaneous symmetry breaking appears again at the level of mean-field description [e.g., the Hartree-Fock (HF) level]. A major difference from the $N\rightarrow\infty$ limit, however, arises from the fact that quantum fluctuations in small systems cannot be neglected. To account for the large fluctuations, one has to perform a subsequent post-Hartree-Fock step that restores the broken symmetries (and the linearity of the many-body Schrödinger equation). Subsequent to symmetry restoration, the ground state obeys all the original symmetries of the many-body Hamiltonian; however, effects of the mean-field symmetry breaking do survive in the properties of the ground state of small systems and lead to emergent phenomena associated with formation of novel states of matter and with characteristic behavior in the excitation spectra. In the following, we will present an overview of the current understanding of SSB in small systems focusing on the essential theoretical aspects, as well as on the contributions made by SSB-based approaches to the fast developing fields of two-dimensional semiconductor quantum dots and ultracold atomic gases in harmonic and toroidal traps. ### 1.3 Historical background from nuclear physics and chemistry The mean field approach, in the form of the Hartree-Fock theory and of the Gross-Pitaevskii (GP) equation, has been a useful tool in elucidating the physics of finite-size fermionic and bosonic systems, respectively. Its applications cover a wide range of systems, from natural atoms, natural molecules, and atomic nuclei, to metallic nanoclusters, and most recently two- dimensional quantum dots and ultracold gases confined in harmonic (parabolic) traps. Of particular interest for the present review (due to spatial-symmetry- breaking aspects) has been the mean-field description of deformed nuclei [11, 12, 13] and metal clusters [14, 15, 16] (exhibiting ellipsoidal shapes). At a first level of description, deformation effects in these latter systems can be investigated via semi-empirical mean-field models, like the particle-rotor model [11] of Bohr and Mottelson (nuclei), the anisotropic-harmonic-oscillator model of Nilsson (nuclei [12] and metal clusters [14]), and the shell- correction method of Strutinsky (nuclei [17] and metal clusters [15, 16]). At the microscopic level, the mean field for fermions is often described [18, 19] via the self-consistent single-determinantal Hartree-Fock theory. At this level, the description of deformation effects mentioned above requires [18] consideration of unrestricted Hartree-Fock (UHF) wave functions that break explicitly the rotational symmetries of the original many-body Hamiltonian, but yield HF Slater determinants with lower energy compared to the symmetry- adapted restricted Hartree-Fock (RHF) solutions.111See in particular Ch 5.5 and Ch 11 in Ref. [18]. However, our terminology (i.e., UHF vs. RHF) follows the practice in quantum chemistry (see Ref. [19]). In earlier publications [20, 21, 22, 23, 24, 25, 26], we have shown that, in the strongly correlated regime, UHF solutions that violate the rotational (circular) symmetry arise most naturally in the case of two-dimensional single quantum dots, for both the cases of zero and high magnetic field; for a UHF calculation in the lowest Landau level (LLL), see also Ref. [27]. Unlike the case of atomic nuclei, however, where (due to the attractive interaction) symmetry breaking is associated primarily with quadrupole shape deformations (a type of Jahn-Teller distortion), spontaneous symmetry breaking in 2D quantum dots induces electron localization (or “crystallization”) associated with formation of electron, or Wigner, molecules). The latter name is used in honor of Eugene Wigner who predicted the formation of a classical rigid Wigner crystal for the 3D electron gas at very low densities [28]. We stress, however, that because of the finite size, Wigner molecules are most often expected to show a physical behavior quite different from the classical Wigner crystal. Indeed, for finite $N$, Wigner molecules exhibit analogies closer to natural molecules, and the Wigner-crystal limit is expected to be reached only for special limiting conditions. For a small system the violation in the mean-field approximation of the symmetries of the original many-body Hamiltonian appears to be paradoxical at a first glance, and some times it has been described mistakenly as an “artifact” (in particular in the context of density-functional theory [29]). However, for the specific cases arising in Nuclear Physics and Quantum Chemistry, two theoretical developments had already resolved this paradox. They are: (1) the theory of restoration of broken symmetries via projection techniques222For the restoration of broken rotational symmetries in atomic nuclei, see Ref. [30] and Ch 11 in Ref. [18]. For the restoration of broken spin symmetries in natural 3D molecules, see Ref. [31]. [30, 31, 32], and (2) the group theoretical analysis of symmetry-broken HF orbitals and solutions in chemical reactions, initiated by Fukutome and coworkers [33] who used the symmetry groups associated with the natural 3D molecules. Despite the different fields, the general principles established in these earlier theoretical developments in nuclear physics and quantum chemistry have provided a wellspring of assistance in our investigations of symmetry breaking for electrons in quantum dots and bosons in harmonic traps. In particular, the restoration of broken symmetries in QDs and ultracold atomic traps via projection techniques constitutes a main theme of the present report. The theory of restoration of broken symmetries has been developed into a sophisticated computational approach in modern nuclear physics. Using the broken-symmetry solutions of the Hartree-Fock-Bogoliubov theory333See Ch 7 in Ref. [18]. (that accounts for nuclear pairing and superfluidity), this approach has been proven particularly efficient in describing the competition between shape deformation and pairing in nuclei. For some recent papers in nuclear physics, see, e.g., Refs. [34, 35, 36, 37, 38, 39]; for an application to superconducting metallic grains, see Ref. [40]. Pairing effects arise only in the case of attractive interactions and they are not considered in this report, since we deal only with repulsive two-body interactions. ### 1.4 Scope of the review Having discussed earlier the general context and historical background from other fields regarding symmetry breaking, we give here an outline of the related methodologies and of the newly discovered strongly correlated phenomena that are discussed in this report in the area of condensed-matter nanosystems. In particular, a two-step method [20, 21, 22, 23, 24, 25] of symmetry breaking at the unrestricted Hartree-Fock level and of subsequent post-Hartree-Fock restoration of the broken symmetries via projection techniques is reviewed for the case of two-dimensional (2D) semiconductor quantum dots and ultracold bosons in rotating traps with a small number ($N$) of particles. The general principles of the two-step method can be traced to nuclear theory (Peierls and Yoccoz, see the original Ref. [30], but also the recent Refs. [34, 35, 36, 37, 38, 39]) and quantum chemistry (Löwdin, see Ref. [31]); in the context of condensed-matter nanophysics and the physics of ultracold atomic gases, it constitutes a novel powerful many-body approach that has led to unexpected discoveries in the area of strongly correlated phenomena. The successes of the method have generated a promising theoretical outlook, bolstered by the unprecedented experimental and technological advances, pertaining particularly to control of system parameters (most importantly of the strength and variety of two-body interactions), that can be achieved in manmade nanostructures. In conjunction with exact diagonalization calculations [26, 41, 42, 43, 44] and recent experiments [41, 44, 45], it is shown that the two-step method can describe a wealth of novel strongly correlated phenomena in quantum dots and ultracold atomic traps. These include: (I) Chemical bonding, dissociation, and entanglement in quantum dot molecules [20, 22, 46] and in electron molecular dimers formed within a single elliptic QD [41, 42, 43, 44], with potential technological applications to solid-state quantum logic gates [47, 48, 49]. (II) Electron crystallization, with localization on the vertices of concentric polygonal rings, and formation of rotating electron molecules (REMs) in circular QDs. At zero magnetic field ($B$), the REMs can approach the limit of a rigid rotor [50, 51]; at high $B$, the REMs are highly floppy and “supersolid”-like, that is, they exhibit [51, 52, 53] a non-rigid rotational inertia [54], with the rings rotating independently of each other [52, 53]. (III) At high magnetic fields and under the restriction of the many-body Hilbert space to the lowest Landau level, the two-step method yields fully analytic many-body wave functions [24, 26], which are an alternative to the Jastrow/Laughlin (JL) [55] and composite-fermion (CF) [56, 57] approaches, offering a new point of view of the fractional quantum Hall regime (FQHE) [58, 59] in quantum dots (with possible implications for the thermodynamic limit). Large scale exact-diagonalization calculations [26, 52, 53] support the results of the two-step method outlined in items II and III above. (IV) The two-step method has been used [60] to discover crystalline phases of strongly repelling ultracold bosons (impenetrable bosons/ Tonks-Girardeau regime [61, 62]) in 2D harmonic traps. In the case of rotating traps, such repelling bosons form rotating boson molecules (RBMs) [63] that are energetically favorable compared to the Gross-Pitaevkii solutions, even for weak repulsion and, in particular, in the regime of GP vortex formation. We will not discuss in this report specific applications of the two-step method to atomic nuclei. Rather, as the title conveys, the report aims at exploring the universal characteristics of quantum correlations arising from symmetry breaking across various fields dealing with small finite systems, such as 2D quantum dots, trapped ultracold atoms, and nuclei – and even natural 3D molecules. Such universal characteristics and similarities in related methodologies persist across the aforementioned fields in spite of the differences in the size of the physical systems and in the range, nature, and strength of the two-body interactions. For specific applications to atomic nuclei, the interested reader is invited to consult the nucler physics literature cited in this report. Figure 1: (Color online) Synopsis of the method of hierarchical approximations (also referred to as the “two-step method,” emphasizing that symmetry breaking at the mean-field level must be accompanied by a subsequent post-Hartree-Fock step of symmetry restoration, with a subsequent further lowering of the energy). See text for a detailed description. ### 1.5 Using a hierarchy of approximations versus probing of exact solutions Figure 1 presents a synopsis of the hierarchy of approximations associated with the two-step method, and in particular for the case of 2D quantum dots. (A similar synopsis can also be written for the case of bosonic systems.) This method produces approximate wave functions with lower energy at each approximation level (as indicated by the downward vertical arrow on the left of the figure). At the lowest level of approximation (corresponding to higher energy with no correlations included), one places the restricted Hartree-Fock, whose main restriction is the double occupancy (up and down spins) of each space orbital. The many-body wave function is a single Slater determinant associated with a “central mean field.” The RHF preserves all spin and space symmetries. For 2D quantum dots, the single-particle density [also referred to as electron density ($e$-density)] is circularly symmetric. The next approximation involves the unrestricted Hartree-Fock, which employs different space orbitals for the two different spin directions. The UHF preserves the spin projection, but allows the total-spin and space symmetries (i.e., rotational symmetries or parity) to be broken. The broken symmetry solutions, however, are not devoid of any symmetry; they exhibit characteristic lower symmetries (point-group symmetries) that are explicit in the electron densities. The UHF many-body wave function is a single Slater determinant associated with a “non-central mean field.” Subsequent approximations aim at restoring the broken symmetries via projection techniques. The restoration-of-symmetry step goes beyond the mean field approximation and it provides a many-body wave function $|\Phi^{\rm PRJ}\rangle$ that is a linear superposition of Slater determinants (see detailed description in section 2.2 below). The projected (PRJ) many-body wave function $|\Phi^{\rm PRJ}\rangle$ preserves all the symmetries of the original many-body Hamiltonian; it has good total spin and angular momentum quantum numbers, and as a result the circular symmetry of the electron densities is restored. However, the lower (point-group) spatial symmetry found at the broken-symmetry UHF level (corresponding to the first step in this method) does not disappear. Instead, it becomes intrinsic or hidden, and it can be revealed via an inspection of conditional probability distributions (CPDs), defined as (within a proportionality constant) $P({\bf r},{\bf r}_{0})=\langle\Phi^{\rm PRJ}|\sum_{i\neq j}\delta({\bf r}_{i}-{\bf r})\delta({\bf r}_{j}-{\bf r}_{0})|\Phi^{\rm PRJ}\rangle,$ (1.1) where $\Phi^{\rm PRJ}({\bf r}_{1},{\bf r}_{2},\ldots,{\bf r}_{N})$ denotes the projected many-body wave function under consideration. If one needs to probe the intrinsic spin distribution of the localized electrons, one has to consider spin-resolved two-point correlation functions (spin-resolved CPDs), defined as $P_{\sigma\sigma_{0}}({\bf r},{\bf r}_{0})=\langle\Phi^{\rm{PRJ}}|\sum_{i\neq j}\delta({\bf r}-{\bf r}_{i})\delta({\bf r}_{0}-{\bf r}_{j})\delta_{\sigma\sigma_{i}}\delta_{\sigma_{0}\sigma_{j}}|\Phi^{\rm{PRJ}}\rangle.$ (1.2) The spin-resolved CPD gives the spatial probability distribution of finding a second electron with spin projection $\sigma$ under the condition that a first electron is located (fixed) at ${\bf r}_{0}$ with spin projection $\sigma_{0}$; $\sigma$ and $\sigma_{0}$ can be either up $(\uparrow$) or down ($\downarrow$). The meaning of the space-only CPD in (1.1) is analogous, but without consideration of spin. Further signatures of the intrinsic lower symmetry occur in the excitation spectra of circular quantum dots that exhibit ro-vibrational character related to the intrinsic molecular structure, or in the dissociation of quantum dot molecules. As the scheme in figure 1 indicates, the mean-field HF equations are non- linear and the symmetry breaking is associated with the appearance of bifurcations in the total HF energies. The occurrence of such bifurcations cannot be predicted a priori from a mere inspection of the many-body Hamiltonian itself; it is a genuine many-body effect that belongs to the class of so-called emergent phenomena [9, 64, 65] that may be revealed only through the solutions of the Hamiltonian themselves (if obtainable) or through experimental signatures. We note that the step of symmetry restoration recovers also the linear properties of the many-body Schrödinger equation. The relation between quantum correlations and the two-step method (also called the method of hierarchical approximations) is portrayed by the downward vertical arrow on the right of figure 1. Indeed, the correlation energy is defined [66] as the difference between the restricted Hartree-Fock and exact ground-state energies, i.e., $E_{\rm corr}=E_{\rm RHF}-E_{\rm EXD}.$ (1.3) As seen from figure 1, starting with the broken-symmetry UHF solution, each further approximation captures successively a larger fraction of the correlation energy (1.3); a specific example of this process is given in figure 5 below (in section 2.2). An alternative approach for studying the emergence of crystalline structures is the exact-diagonalizaion method that will be discussed in detail in section 4.1. Like the projected wave functions, the EXD many-body wave functions preserve of course all the symmetries of the original Hamiltonian. As a result, the intrinsic, or hidden, point-group symmetry associated with particle localization and molecule formation is not explicit, but it is revealed through inspection of CPDs [one simply uses the exact-diagonalization wave function $\Phi^{\rm EXD}({\bf r}_{1},{\bf r}_{2},\ldots,{\bf r}_{N})$ in Equation (1.1)] and Equation (1.2), or recognized via characteristic trends in the calculated excitation spectra. When feasible, the EXD results provide a definitive answer in terms of numerical accuracy, and as such they serve as a test to the results obtained through approximation methods (e.g., the above two-step method). However, the underlying physics of electron or boson molecule formation is less transparent when analyzed with the exact- diagonalization method compared to the two-step approach. Indeed, many exact- diagonalization studies of 2D quantum dots and trapped bosons in harmonic traps have focused simply on providing high accuracy energetics and they omitted calculation of CPDs. However, the importance of using CPDs as a tool for probing the many-body wave functions cannot be overstated. For example, while exact-diagonalization calculations for bosons in the lowest Landau level have been reported rather early [67, 68, 69, 70, 71], the analysis in these studies did not include calculations of the CPDs, and consequently formation of rotating boson molecules and particle “crystallization” was not recognized (for further discussion of these issues, see Romanovsky et al[60, 63] and Baksmaty et al[72]). From the above, it is apparent that both methods, i.e., the two-step method and the exact-diagonalization one, complement each other, and it is in this spirit that we use them in this report. ### 1.6 Experimental signatures of quantum correlations Historically, the isolation of a small number ($N<20$) of electrons down to a single electron was experimentally realized in the so-called “vertical” quantum dots [2]. The name vertical QDs derives from the fact that the leads and voltage gates are located in a vertical arrangement, on top and below the two-dimensional dot. At zero magnetic field, experimental measurements [2, 73] of addition energies, $\Delta^{2}E_{N}=\mu_{N+1}-\mu_{N},$ (1.4) where the chemical potential $\mu_{N}=E_{N}-E_{N-1}$, indicated that correlation effects at zero and low $B$ are rather weak in such dots, a property that later was attributed to the strong screening of the Coulomb interaction in these devices. The measured addition energies exhibited maxima at closed electronic shells ($N=2,6,12,\ldots$) and at mid-shells ($N=4,9,\ldots$) in agreement with a 2D-harmonic-oscillator central-mean-field model and the Hund’s rules, and in analogy with the Aufbau principle and the physics of natural 3D atoms. It was found that the measured ground-state energy spectra for low magnetic fields could be understood on the basis of a simple “constant-interaction” model where the effect of the two-body Coulomb interaction is reduced phenomenologically to an overall classical capacitance, $C$, characterizing the charging energy $Z^{2}e^{2}/(2C)$ of the quantum dot. As a result of screening, strong correlation effects and formation of Wigner molecules can be expected to occur in vertical dots particularly under the influence of high magnetic fields. Evidence about the formation of Wigner molecules in vertical quantum dots has been provided recently in Ref. [74], where measured ground-state spectra as a function of $B$ for $N=3e$ and $N=4e$ were reanalyzed with exact-diagonalization calculations that included screening. At the time of submission of this report, a second ground-state crossing at high $B$ due to strong correlations was also demonstrated experimentally in a two-electron vertical quantum dot with an external confinement that was smaller than the previously used ones [75]. Early theoretical work [20] at zero magnetic field using simply the symmetry broken UHF solutions suggested that an unscreened Coulomb repulsion may result in a violation of Hund’s rules. However, following the two-step method of Refs. [20, 21, 22, 23, 24, 25], it has been shown [76] most recently that the companion step of symmetry restoration recovers the Hund’s rules in the case of $N=4e$. In addition, the $B=0$ results of Ref. [20] suggested that both the maxima of the addition energies at closed shells and at mid-shells become gradually weaker (and they eventually disappear) as the strength of the Coulomb interaction (and consequently the strength of correlations) increases, leading to formation of “strong” Wigner molecules. The qualitative trend of formation of strong Wigner molecules obtained from a relatively simple UHF calculation at $B=0$ has been confirmed later by more accurate EXD [50, 77] and quantum Monte Carlo [78] calculations, as well as through symmetry restoration calculations [23, 76], although its experimental demonstration remains still a challenge. A more favorable experimental configuration for the development and observation of strong interelectron correlations is the so-called “lateral” dot, where the leads and gates are located on the sides of the dot and thus screening effects are reduced. Tunability of these dots down to a single electron has been achieved only in the last few years [79]. Most recently, continually improving experimental techniques have allowed precise measurements of excitation spectra of $2e$ lateral (and anisotropic) quantum dots at zero and low magnetic fields [41, 45, 80]. As discussed in detail in section 5, the behavior of these excitation spectra [41, 45] as a function of $B$ provides unambiguous signatures for the presence of strong correlations and the formation of Wigner molecules. Experimentally observed behavior of two electrons in lateral double QDs [81] provides further evidence for strong correlation phenomena. Indeed, instead of successively populating delocalized states over both QDs according to a molecular-orbital scheme, the two electrons localize on the individual dots according to a Heitler-London picture [82]. Theoretically, such strongly correlated phenomena in double quantum dots were described in Refs. [20, 22, 46]; see section 2.1.4 below. Correlations are expected to influence not only the spectral properties of quantum dots, but also to effect transport characteristics. Indeed correlation effects may underlie the behavior of the transmission amplitudes (magnitude and phase) of an electron tunneling through a quantum dot. Such transmission measurements have been performed using Aharonov-Bohm interferometry [83], and an interpretation involving strongly correlated states in the form of Wigner molecules has been proposed recently [84]. The quantity that links transport experiments with many-body theory of electrons in QDs is the overlap between many-body states with $N-1$ and $N$ electrons, i.e., $\langle\Phi(N-1)|c_{j}|\Phi(N)\rangle$, where $c_{j}$ annihilates the $j$th electron. The strength of correlations in quantum dots at zero $B$ can be quantified by the Wigner parameter $R_{W}$, which is the ratio between the strength of the Coulomb repulsion and the one-electron kinetic energy (see section 2.1.2). Naturally, for the case of neutral repelling bosons, the corresponding parameter is the ratio between the strength of the contact interaction and the one-particle kinetic energy in the harmonic trap, and it is denoted as $R_{\delta}$. Larger values of these parameters ($R_{W}$ or $R_{\delta}$) result in stronger correlation effects. Progress in the ability to experimentally control the above parameters has been particularly impressive in the case of ultracold trapped bosons. Indeed, realizations of continuous tunability of $R_{\delta}$ over two orders of magnitude (from 1 to 5 [85] and from 5 to 200 [86]) has been most recently reported in quasi-linear harmonic traps. Such high values of $R_{\delta}$ allowed experimental realization of novel strongly correlated states drastically different from a Bose-Einstein condensate. This range of high values of $R_{\delta}$ is known as the Tonks-Girardeau regime and the corresponding states are one-dimensional analogues of molecular structures made out of localized bosons. In two dimenional traps, it has been predicted that such large values of $R_{\delta}$ lead to the emergence of crystalline phases [60, 63]. The high experimental control of optical lattices has also been exploited for the creation [87] of novel phases of ultracold bosons analogous to Mott insulators; such phases are related to the formation of electron puddles discussed in section 2.1.4 and to the fragmentation of Bose-Einstein condensates [88]. ### 1.7 Plan of the report The plan of the report can be visualized through the table of contents. Special attention has been given to the Introduction, which offers a general presentation of the subject of symmetry breaking and quantum correlations in confined geometries – including a discussion of the differences with the case of extended systems, a historical background from other fields, and a diagrammatic synopsis of the two-step method of symmetry breaking/symmetry restoration. The theoretical framework and other technical methodological background are presented in Section 2 (symmetry breaking/symmetry restoration in quantum dots), Section 3 (symmetry breaking/symmetry restoration for trapped ultracold bosons), and Section 4 (exact-diagonalization approaches). Section 4 includes also a commentary on quantum Monte Carlo methods. For the case of semiconductor quantum dots, the main results and description of the strongly correlated regime are presented in Sections 5, 6, and 7, with Section 5 focusing on the case of two electrons and its historical significance. Section 8 is devoted to a description of the strongly-correlated regime of trapped repelling bosons. Finally, a summary is given in Section 9, and the Appendix offers an outline of the Darwin-Fock single-particle spectra for a two-dimensional isotropic oscillator under a perpendicular magnetic field or under rotation. We note that the sections on trapped bosons (Section 3 and Section 8) can be read independently from the sections on quantum dots. ## 2 Symmetry breaking and subsequent symmetry restoration for electrons in confined geometries: Theoretical framework The many-body Hamiltonian describing $N$ electrons confined in a two- dimensional QD and interacting via a Coulomb repulsion is written as ${\cal H}=\sum_{i=1}^{N}H(i)+\sum_{i=1}^{N}\sum_{j>i}^{N}\frac{e^{2}}{\kappa r_{ij}}.$ (2.1) In Equation (2.1), $\kappa$ is the dielectric constant of the semiconducting material and $r_{ij}=|{\bf r}_{i}-{\bf r}_{j}|$. The single-particle Hamiltonian in a perpendicular external magnetic field $B$ is given by $H=\frac{({\bf p}-e{\bf A}/c)^{2}}{2m^{*}}+V(x,y)+\frac{g^{*}\mu_{B}}{\hbar}{\bf B\cdot s},$ (2.2) where the external confinement is denoted by $V(x,y)$, the vector potential ${\bf A}$ is given in the symmetric gauge by ${\bf A}({\bf r})=\frac{1}{2}{\bf B}\times{\bf r}=\frac{1}{2}(-By,Bx,0),$ (2.3) and the last term in (2.2) is the Zeeman interaction with $g^{*}$ being the effective Landé factor, $\mu_{B}$ the Bohr magneton, ${\bf s}$ the spin of an individual electron and $m^{*}$ is the effective electron mass. The external potential confinement $V(x,y)$ can assume various parametrizations in order to model a single circular or elliptic quantum dot, or a quantum dot molecule. Of course, in the case of an elliptic QD, one has $V(x,y)=\frac{1}{2}m^{*}(\omega_{x}^{2}x^{2}+\omega_{y}^{2}y^{2}),$ (2.4) which reduces to the circular QD potential when $\omega_{x}=\omega_{y}=\omega_{0}$. The appropriate parametrization of $V(x,y)$ in the case of a double QD is more complicated. In our work, we use a parametrization based on a 2D two-center oscillator with a smooth necking. This latter parametrization is described in detail in Refs. [23, 46], where readers are directed for further details. In contrast with other parametrizations based on two displaced inverted Gaussians [89], the advantage of the two-center oscillator is that the height of the interdot barrier, the distance between the dots, the ellipticity of each dot, and the gate potentials of the two dots (i.e., the relative potential wells in the neighboring dots) can be varied independently of each other. A prefactor multiplying the Coulomb term in Equation (2.1) (being either an overall constant $\gamma$ as in section 5.1 below, or having an appropriate position-dependent functional form [42, 43]) is used to account for the reduction of the Coulomb interaction due to the finite thickness of the electron layer and to additional screening (beyond that produced by the dielectric constant of the material) arising from the formation of image charges in the gate electrodes [90]. ### 2.1 Mean-field description and unrestricted Hartree-Fock Vast literature is available concerning mean-field studies of electrons in quantum dots. Such publications are divided mainly into applications of density functional theory [3, 91, 92, 93, 94, 95, 96] and the use of Hartree- Fock methods [20, 25, 27, 93, 97, 98, 99, 100, 101, 102, 103, 104]. The latter include treatments according to the restricted Hartree-Fock [97], unrestricted Hartree-Fock with spin, but not space, symmetry breaking [98, 99, 100], unrestricted Hartree-Fock with spin and/or space symmetry breaking [20, 25, 27, 93, 101, 102, 103], and the so-called Brueckner Hartree-Fock [104, 105]. From the several Hartree-Fock variants mentioned above, only the UHF with consideration of both spin and space symmetry unrestrictions has been able to describe formation of Wigner molecules, and in the following we will exclusively use this unrestricted version of Hartree-Fock theory. The inadequacy of the density-functional theory in describing Wigner molecules will be discussed in section 2.3. #### 2.1.1 The self-consistent Pople-Nesbet equations. The unrestricted Hartree-Fock equations used by us are an adaptation of the Pople-Nesbet [106] equations described in detail in Ch 3.8 of Ref. [19]. For completeness, we present here a brief description of these equations, along with pertinent details of their computational implementation by us to the 2D case of semiconductor QDs. We start by requesting that the unrestricted Hartree-Fock many-body wave function for $N$ electrons is represented by a single Slater determinant $\Psi_{\rm UHF}({\bf x}_{1},\ldots,{\bf x}_{N})=\frac{1}{\sqrt{N!}}{\rm det}[\chi_{1}({\bf x}_{1}),\chi_{2}({\bf x}_{2}),\ldots,\chi_{N}({\bf x}_{N}],$ (2.5) where $[\chi_{i}({\bf x})]$ are a set of $N$ spin orbitals, with the index ${\bf x}$ denoting both the space and spin coordinates. Furthermore, we take $\chi_{i}({\bf x})=\psi_{i}({\bf r})\alpha$ for a spin-up electron and $\chi_{i}({\bf x})=\psi_{i}({\bf r})\beta$ for a spin-down electron. As a result, the UHF determinants in this report are eigenstates of the projection of the total spin with eigenvalue $S_{z}=(N^{\alpha}-N^{\beta})/2$, where $N^{\alpha(\beta)}$ denotes the number of spin up (down) electrons. However, these Slater determinants are not eigenstates of the square of the total spin, ${\bf S}^{2}$, except in the fully spin polarized case. According to the variational principle, the best spin orbitals must minimize the total energy $\langle\Psi_{\rm UHF}|{\cal H}|\Psi_{\rm UHF}\rangle$. By varying the spin orbitals $[\chi_{i}({\bf x})]$ under the constraint that they remain orthonormal, one can derive the UHF Pople-Nesbet equations described below. A key point is that electrons with $\alpha$ (up) spin will be described by one set of spatial orbitals $\\{\psi^{\alpha}_{j}|j=1,2,\dots,K\\}$, while electrons with $\beta$ (down) spin are described by a different set of spatial orbitals $\\{\psi^{\beta}_{j}|j=1,2,\ldots,K\\}$; of course in the restricted Hartree-Fock $\psi^{\alpha}_{j}=\psi^{\beta}_{j}=\psi_{j}$. Next, one introduces a set of basis functions $\\{\varphi_{\mu}|\mu=1,2,\ldots,K\\}$ (constructed to be orthonormal in our 2D case), and expands the UHF orbitals as $\psi^{\alpha}_{i}=\sum_{\mu=1}^{K}C_{\mu i}^{\alpha}\varphi_{\mu},~{}~{}~{}i=1,2,\ldots,K,$ (2.6) $\psi^{\beta}_{i}=\sum_{\mu=1}^{K}C_{\mu i}^{\beta}\varphi_{\mu},~{}~{}~{}i=1,2,\ldots,K.$ (2.7) The UHF equations are a system of two coupled matrix eigenvalue problems resolved according to up and down spins, ${\bf F}^{\alpha\beta}{\bf C}^{\alpha}={\bf C}^{\alpha}{\bf E}^{\alpha}$ (2.8) ${\bf F}^{\beta\alpha}{\bf C}^{\beta}={\bf C}^{\beta}{\bf E}^{\beta},$ (2.9) where ${\bf F}^{\alpha\beta(\beta\alpha)}$ are the Fock-operator matrices and ${\bf C}^{\alpha(\beta)}$ are the vectors formed with the coefficients in the expansions (2.6) and (2.7). The matrices ${\bf E}^{\alpha(\beta)}$ are diagonal, and as a result equations (2.8) and (2.9) are canonical (standard). Notice that noncanonical forms of HF equations are also possible (see Ch 3.2.2 of Ref. [19]). Since the self-consistent iterative solution of the HF equations can be computationally implemented only in their canonical form, canonical orbitals and solutions will always be implied, unless otherwise noted explicitly. We note that the coupling between the two UHF equations (2.8) and (2.9) is given explicitly in the expressions for the elements of the Fock matrices below [(2.12) and (2.13)]. Introducing the density matrices ${\bf P}^{\alpha(\beta)}$ for $\alpha(\beta)$ electrons, $P^{\alpha}_{\mu\nu}=\sum_{a}^{N^{\alpha}}C^{\alpha}_{\mu a}(C^{\alpha}_{\nu a})^{*}$ (2.10) $P^{\beta}_{\mu\nu}=\sum_{a}^{N^{\beta}}C^{\beta}_{\mu a}(C^{\beta}_{\nu a})^{*},$ (2.11) where $N^{\alpha}+N^{\beta}=N$, the elements of the Fock-operator matrices are given by $F^{\alpha\beta}_{\mu\nu}=H_{\mu\nu}+\sum_{\lambda}\sum_{\sigma}P^{\alpha}_{\lambda\sigma}[(\mu\sigma|\nu\lambda)-(\mu\sigma|\lambda\nu)]+\sum_{\lambda}\sum_{\sigma}P^{\beta}_{\lambda\sigma}(\mu\sigma|\nu\lambda)$ (2.12) $F^{\beta\alpha}_{\mu\nu}=H_{\mu\nu}+\sum_{\lambda}\sum_{\sigma}P^{\beta}_{\lambda\sigma}[(\mu\sigma|\nu\lambda)-(\mu\sigma|\lambda\nu)]+\sum_{\lambda}\sum_{\sigma}P^{\alpha}_{\lambda\sigma}(\mu\sigma|\nu\lambda),$ (2.13) where $H_{\mu\nu}$ are the elements of the single electron Hamiltonian (with an external magnetic field $B$ and an appropriate potential confinement), and the Coulomb repulsion is expressed via the two-electron integrals $(\mu\sigma|\nu\lambda)=\frac{e^{2}}{\kappa}\int\rmd{\bf r}_{1}\rmd{\bf r}_{2}\varphi^{*}_{\mu}({\bf r}_{1})\varphi^{*}_{\sigma}({\bf r}_{2})\frac{1}{|{\bf r}_{1}-{\bf r}_{2}|}\varphi_{\nu}({\bf r}_{1})\varphi_{\lambda}({\bf r}_{2}),$ (2.14) with $\kappa$ being the dielectric constant of the semiconductor material. Of course, the Greek indices $\mu$, $\nu$, $\lambda$, and $\sigma$ run from 1 to $K$. The system of the two coupled UHF matrix equations (2.8) and (2.9) is solved selfconsistently through iteration cycles. For obtaining the numerical solutions, we have used a set of $K$ basis states $\varphi_{i}$’s that are chosen to be the product wave functions formed from the eigenstates of one- center (single QD) and/or two-center [22, 46] (double QD) one-dimensional oscillators along the $x$ and $y$ axes. Note that for a circular QD a value $K=78$ corresponds to all the states of the associated 2D harmonic oscillator up to and including the 12th major shell. The UHF equations preserve at each iteration step the symmetries of the many- body Hamiltonian, if these symmetries happen to be present in the input (initial) electron density of the iteration (see section 5.5 of Ref. [18]). The input densities into the iteration cycle are controlled by the values of the $P^{\alpha}_{\lambda\sigma}$ and $P^{\beta}_{\lambda\sigma}$ matrix elements. Two cases arise in practice: (i) Symmetry adapted RHF solutions are extracted from (2.8) and (2.9) by using as input $P^{\alpha}_{\lambda\sigma}=P^{\beta}_{\lambda\sigma}$=0 for the case of closed shells (with or without an infinitesimally small $B$ value). For open shells, one needs to use an infinitesimally small value of $B$. With these choices, the output of the first iteration (for either closed or open shells) is the single-particle spectrum and corresponding electron densities at $B=0$ associated with the Hamiltonian in (2.2) (the small value of $B$ mentioned above guarantees that the single-particle total and orbital densities are circular). (ii) For obtaining broken-symmetry UHF solutions, the input densities must be different in an essential way from the ones mentioned above. We have found that the choice $P^{\alpha}_{\lambda\sigma}=1$ and $P^{\beta}_{\lambda\sigma}=0$ usually produces broken-symmetry solutions (in the regime where symmetry breaking occurs). Having obtained the selfconsistent solution, the total UHF energy is calculated as $E_{\rm UHF}=\frac{1}{2}\sum_{\mu}\sum_{\nu}[(P^{\alpha}_{\nu\mu}+P^{\beta}_{\nu\mu})H_{\mu\nu}+P^{\alpha}_{\nu\mu}F^{\alpha\beta}_{\mu\nu}+P^{\beta}_{\nu\mu}F^{\beta\alpha}_{\mu\nu}].$ (2.15) We note that the Pople-Nesbet UHF equations are primarily employed in Quantum Chemistry for studying the ground states of open-shell molecules and atoms. Unlike our studies of QDs, however, such chemical UHF studies consider mainly the breaking of the total spin symmetry, and not that of the space symmetries. As a result, for purposes of emphasis and clarity, we have often used (see, e.g., our previous papers) prefixes to indicate the specific unrestrictions (that is removal of symmetry restrictions) involved in our UHF solutions, i.e., the prefix s- for the total-spin and the prefix S- for the space unrestriction. The emergence of broken-symmetry solutions is associated with instabilities of the restricted HF solutions, i.e., the restricted HF energy is an extremum whose nature as a minimum or maximum depends on the positive or negative value of the second derivative of the HF energy. The importance of this instability problem was first highlighted in a paper by Overhauser [107]. Soon afterwards, the general conditions for the appearance of such instabilities (analyzed within linear response and the random-phase approximation) were discussed by Thouless in the context of nuclear physics [108]. Subsequently, the Hartree- Fock stability/instability conditions were re-examined [109, 110], using a language from (and applications to) the field of quantum chemistry. For comprehensive reviews of mean-field symmetry breaking and the Hartree-Fock methods and instabilities in the context of quantum chemistry, see the collection of papers in Ref. [111]. #### 2.1.2 The Wigner parameter and classes of spontaneous symmetry breaking solutions. Using the self-consistent (spin-and-space) unrestricted Hartree-Fock equations presented in the previous section, we found [20], for zero and low magnetic fields, three classes of spontaneous symmetry breakings in circular single QDs and in lateral quantum dot molecules (i.e., formation of ground states of lower symmetry than that of the confining potentials). These include the following: (I) Wigner molecules in both QDs and quantum dot molecules, i.e., (spatial) localization of individual electrons within a single QD or within each QD comprising the quantum dot molecule. (II) formation of electron puddles in quantum dot molecules, that is, localization of the electrons on each of the individual dots comprising the quantum dot molecule, but without localization within each dot, and (III) pure spin-density waves (SDWs) which are not accompanied by spatial localization of the electrons [91]. Figure 2: (Color online) UHF electron density in a parabolic QD for $N=19$ and $S_{z}=19/2$, exhibiting breaking of the circular symmetry at $R_{W}=5$ and $B=0$. The choice of the remaining parameters is: $\hbar\omega_{0}=5$ meV and $m^{*}=0.067m_{e}$. Distances are in nanometers and the electron density in $10^{-4}$ nm-2. It can be shown that a central-mean-field description (associated with the RHF) at zero and low magnetic fields may apply in the case of a circular QD only for low values of the Wigner parameter $R_{\rm W}\equiv Q/\hbar\omega_{0},$ (2.16) where $Q$ is the Coulomb interaction strength and $\hbar\omega_{0}$ is the energy quantum of the harmonic potential confinement (being proportional to the one-particle kinetic energy); $Q=e^{2}/(\kappa l_{0})$, with $\kappa$ being the dielectric constant, $l_{0}=(\hbar/(m^{*}\omega_{0}))^{1/2}$ the spatial extension of the lowest state’s wave function in the harmonic (parabolic) confinement, and $m^{*}$ the effective electron mass. Furthermore, we find that Wigner molecules (SSB class I) occur in both QDs and quantum dot molecules for $R_{\rm W}>1$. Depending on the value of $R_{W}$, one may distinguish between “weak” (for smaller $R_{W}$ values) and “strong” (for larger $R_{W}$ values) Wigner molecules, with the latter termed sometimes as “Wigner crystallites” or “electron crystallites.” The appearance of such crystalline structures may be regarded as a quantum phase transition of the electron liquid upon increase of the parameter $R_{W}$. Of course, due to the finite size of QDs, this phase transition is not abrupt, but it develops gradually as the parameter $R_{W}$ varies. For quantum dot molecules with $R_{\rm W}<1$, Wigner molecules do not develop and instead electron puddles may form (SSB class II). For single QDs with $R_{\rm W}<1$, we find in the majority of cases that the ground-states exhibit a central-mean-field behavior without symmetry breaking; however, at several instances (see an example below), a pure SDW (SSB class III) may develop. #### 2.1.3 Unrestricted Hartree-Fock solutions representing Wigner molecules. As a typical example of a Wigner-molecule solution that can be extracted from the UHF equations, we mention the case of $N=19$ electrons for $\hbar\omega_{0}=5$ meV, $R_{W}=5$ $(\kappa=3.8191)$, and $B=0$. Figure 2 displays the total electron density of the broken-symmetry UHF solution for these parameters, which exhibits breaking of the rotational symmetry. In accordance with electron densities for smaller dot sizes published by us earlier [20, 21] the electron density in figure 2 is highly suggestive of the formation of a Wigner molecule, with a (1,6,12) ring structure in the present case; the notation $(n_{1},n_{2},\ldots,n_{r})$ signifies the number of electrons in each ring: $n_{1}$ in the first, $n_{2}$ in the second, and so on. This polygonal ring structure agrees with the classical one (that is the most stable arrangement of 19 point charges in a 2D circular harmonic confinement [112, 113, 114]444These references presented extensive studies pertaining to the geometrical arrangements of classical point charges in a two-dimensional harmonic confinement.), and it is sufficiently complex to instill confidence that the Wigner-molecule interpretation is valid. The following question, however, arises naturally at this point: is such molecular interpretation limited to the intuition provided by the landscapes of the total electron densities, or are there deeper analogies with the electronic structure of natural 3D molecules? The answer to the second part of this question is in the affirmative. Indeed, it was found [25] that SSB results in the replacement of a higher symmetry by a lower one. As a result, the molecular UHF solutions exhibit point-group spatial symmetries that are amenable to a group-theoretical analysis in analogy with the case of 3D natural molecules. Figure 3: (Color online) UHF ground-state of a 6$e$ quantum dot molecule (double dot), with parameters resulting in formation of two non-crystallized electron puddles (akin to dissociation of the quantum dot molecule in two QDs with 3 electrons each). Left: total electronic density. Right: contour plots of the densities (orbital squares) of the three individual orbitals localized on the left dot, with spin polarization of the orbitals as indicated. The choice of parameters is: $\hbar\omega_{0}=5$ meV (harmonic confinement of each dot), $d=70$ nm (distance berween dots), $V_{b}=10$ meV (interdot barrier), $m^{*}=0.067m_{e}$ (electron effective mass), and $\kappa=20$ (dielectric constant). Lengths ($x$ and $y$ axes) in nm, density distribution (vertical axis) in 10-3 nm-2. #### 2.1.4 Unrestricted Hartree-Fock solutions representing electron puddles. An example of formation of electron puddles in quantum dot molecules, that is, localization of the electrons on each of the individual dots comprising the quantum dot molecule, but without localization within each dot, is presented in figure 3. We consider the case of $N=6$ electrons in a double dot under field-free conditions ($B=0$); with parameters $\hbar\omega_{0}=5$ meV (harmonic confinement of each dot), $d=70$ nm (distance berween dots), $V_{b}=10$ meV (interdot barrier) and $m^{*}=0.067m_{e}$ (electron effective mass). Reducing the $R_{W}$ value (with reference to each constituent QD) to 0.95 (i.e., for a dielectric constant $\kappa=20$) guarantees that the ground- state of the $6e$ quantum dot molecule consists of electron puddles [SSB of type II, figure 3]. In this case, each of the electron puddles (on the left and right dots) is spin-polarized with total spin projection $S_{z}=1/2$ on the left QD and $S_{z}=-1/2$ on the right QD. As a result, the singlet and triplet states of the whole quantum dot molecule are essentially degenerate. Note that the orbitals on the left and right dots [see, e.g., those on the left dot in figure 3 (right)] are those expected from a central-mean-field treatment of each individual QD, but with slight (elliptical) distortions due to the interdot interaction and the Jahn-Teller distortion associated with an open shell of three electrons (in a circular harmonic confinement). Note the sharp contrast between these central-mean-field orbitals and corresponding electron density (figure 3) with the electron density and the three orbitals associated with formation of a Wigner molecule inside a single QD [see, e.g., figure 6 in section 2.2.2 below]. The formation of electron puddles described above can be also seen as a form of dissociation of the quantum dot molecule. We found that only for much lower values of $R_{W}$ ($<0.20$, i.e., $\kappa>90.0$) the electron orbitals do extend over both the left and right QDs, as is usually the case with 3D natural molecules (molecular-orbital theory). Further examples and details of these two regimes (dissociation versus molecular-orbital description) can be found in Refs. [22, 46]. Figure 4: (Color online) UHF solution in a parabolic QD exhibiting a pure spin density wave for $N=14$, $S_{z}=0$, $R_{\rm W}=0.8$, and $B=0$. (a) The total electron density exhibiting circular symmetry; (b) The spin density exhibiting azimuthal modulation (note the 12 humps whose number is smaller than the number of electrons; on the contrary in the case of a Wigner molecule, the number of humps in the electron density is always equal to $N$). The choice of the remaining parameters is: $\hbar\omega_{0}=5$ meV and $m^{*}=0.067m_{e}$. Distances are in nanometers and the electron (ED) and spin (SD) densities in $10^{-4}$ nm-2. #### 2.1.5 Unrestricted Hartree-Fock solutions representing pure spin density waves within a single quantum dot. Another class of broken-symmetry solutions that can appear in single QDs are the spin density waves. The SDWs are unrelated to electron localization and thus are quite distinct from the Wigner molecules [20]; in single QDs, they were obtained [91] earlier within the framework of spin density functional theory. To emphasize the different nature of spin density waves and Wigner molecules, we present in figure 4 an example of a SDW obtained with the UHF approach [the corresponding parameters are: $N=14$, $S_{z}=0$, $R_{\rm W}=0.8$ ($\kappa=23.8693$), and $B=0$]. Unlike the case of Wigner molecules, the SDW exhibits a circular electron density [see figure 4(a)], and thus it does not break the rotational symmetry. Naturally, in keeping with its name, the SDW breaks the total spin symmetry and exhibits azimuthal modulations in the spin density [see figure 4(b); however, the number of humps is smaller than the number of electrons]. We mention here that the possibility of ground-state configurations with uniform electron density, but nonuniform spin density, was first discussed for 3D bulk metals using the HF method in Ref. [115]. The SDWs in single QDs appear for $R_{\rm W}\leq 1$ and are of lesser importance; thus in the following we will exclusively study the case of Wigner molecules. However, for $R_{\rm W}\leq 1$, formation of a special class of SDWs (often called electron puddles, see section 2.1.4) plays an important role in the coupling and dissociation of quantum dot molecules (see Ref. [22] and Ref. [46]). Figure 5: (Color online) Various approximation levels for the lowest singlet state of a field-free two-electron QD with $R_{W}=2.40$. The corresponding energies (in meV) are shown at the bottom of the figure. (a): Electron density of the RHF solution, exhibiting circular symmetry (due to the imposed symmetry restriction). The correlation energy $E_{\rm corr}=2.94$ meV, is defined as the difference between the energy of this state and the exact solution [shown in frame (e)]. (b1) and (b2): The two occupied orbitals (modulus square) of the symmetry-broken “singlet” UHF solution (b1), with the corresponding total electron density exhibiting non-circular shape (b2). The energy of the UHF solution shows a gain of 44.3% of the correlation energy. (c): Electron density of the spin-projected singlet (Spin-P), showing broken spatial symmetry, but with an additional gain of correlation energy. (d): the spin- and-angular-momentum projected state (S&AMP) exhibiting restored circular symmetry with a 73.1% gain of the correlation energy. The choice of parameters is: dielectric constant $\kappa=8$, parabolic confinement $\hbar\omega_{0}=5$ meV, and effective mass $m^{*}=0.067m_{e}$. Distances are in nanometers and the densities in $10^{-4}$ nm-2. ### 2.2 Projection techniques and post-Hartree-Fock restoration of broken symmetries As discussed in section 1.5, for finite systems the symmetry broken UHF solutions are only an intermediate approximation. A subsequent step of post- Hartree-Fock symmetry restoration is needed. Here we present the essentials of symmetry restoration while considering for simplicity the case of two electrons in a circular parabolic QD. Results obtained for various approximation levels for a two-electron QD with $B=0$ and $R_{W}=2.40$ (that is, in the Wigner-molecule regime) are displayed in figure 5. In these calculations [23], the spin projection was performed following reference [31], i.e., one constructs the wave function $\Psi_{\rm Spin-P}(s)={\cal P}_{\rm spin}(s)\Psi_{\rm UHF},$ (2.17) where $\Psi_{\rm UHF}$ is the original symmetry-broken UHF determinant (which is already by construction an eigenstate of the projection $S_{z}$ of the total spin). In (2.17), the spin projection operator (projecting into a state which is an eigenstate of the square of the total spin) is given by ${\cal P}_{\rm spin}(s)\equiv\prod_{s^{\prime}\neq s}\frac{\hat{\bf S}^{2}-s^{\prime}(s^{\prime}+1)}{s(s+1)-s^{\prime}(s^{\prime}+1)},$ (2.18) where the index $s^{\prime}$ runs over the quantum numbers associated with the eigenvalues $s^{\prime}(s^{\prime}+1)$ of $\hat{\bf S}^{2}$ (in units of $\hbar^{2}$), with $\hat{\bf S}$ being the total spin operator. For two electrons, the projection operator reduces to ${\cal P}_{\rm spin}^{s,t}=1\mp\varpi_{12}$, where the operator $\varpi_{12}$ interchanges the spins of the two electrons; the upper (minus) sign corresponds to the singlet ($s$ supersript), and the lower (plus) sign corresponds to the triplet ($t$ superscript) state. The angular momentum projector (projecting into a state with total angular momentum $L$) is given by $2\pi{\cal P}_{L}\equiv\int_{0}^{2\pi}\rmd\gamma\exp[-\rmi\gamma(\hat{L}-L)],$ (2.19) where $\hat{L}=\hat{l}_{1}+\hat{l}_{2}$ is the total angular momentum operator. As seen from (2.19), application of the projection operator ${\cal P}_{L}$ to the spin-restored state $\Psi_{\rm Spin-P}(s)$ corresponds to a continuous configuration interaction expansion of the wave function that uses, however, non-orthogonal orbitals (compare section 4.1). The application of the projection operator ${\cal P}_{L}$ to the state $\Psi_{\rm Spin-P}(s)$ generates a whole rotational band of states with good angular momenta (yrast band). The energy of the projected state with total angular momentum $L$ is given by $E_{\rm PRJ}(L)=\left.\int_{0}^{2\pi}h(\gamma)\rme^{\rmi\gamma L}\rmd\gamma\right/\int_{0}^{2\pi}n(\gamma)\rme^{\rmi\gamma L}\rmd\gamma,$ (2.20) with $h(\gamma)=\langle\Psi_{\rm Spin-P}(s;0)|{\cal H}|\Psi_{\rm Spin-P}(s;\gamma)\rangle$ and $n(\gamma)=\langle\Psi_{\rm Spin-P}(s;0)|\Psi_{\rm Spin-P}(s;\gamma)\rangle$, where $\Psi_{\rm Spin-P}(s;\gamma)$ is the spin-restored (i.e., spin-projected) wave function rotated by an azimuthal angle $\gamma$ and ${\cal H}$ is the many-body Hamiltonian. We note that the UHF energies are simply given by $E_{\rm UHF}=h(0)/n(0)$. In the following we focus on the ground state of the two-electron system, i.e., $L=0$. The electron densities corresponding to the initial RHF approximation [shown in figure 5(a)] and the final spin-and-angular-momentum projection (S&AMP) [shown in figure 5(d)], are circularly symmetric, while those corresponding to the two intermediate approximations, i.e., the UHF and spin-projected solutions [figure 5(b2) and figure 5(c), respectively] break the circular symmetry. This behavior illustrates graphically the meaning of the term “restoration of symmetry,” and the interpretation that the UHF broken-symmetry solution refers to the intrinsic (rotating) frame of reference of the electron molecule. In light of this discussion the final projected state is called a rotating electron or (Wigner) molecule. Expressions (2.19) and (2.20) apply directly to REMs having a single polygonal ring of $N$ localized electrons, with $\hat{L}=\sum_{i=1}^{N}{\hat{l}}_{i}$. For a generalization to electron molecules with multiple concentric polygonal rings, see section 2.2.1 below. For restoring the total spin, an alternative method to the projection formula (2.18) can be found in the literature [33]. We do not make use of this alternative formulation in this report, but we briefly describe it here for the sake of completeness. Based on the formal similarity between the 3D angular momentum and the total spin, one can apply the formula by Peierls and Yoccoz [30] and obtain the projection operator ${\cal P}^{s}_{{S_{z}}{q}}=\frac{2s+1}{8\pi^{2}}\int\rmd\Gamma D^{s*}_{{S_{z}}{q}}(\Gamma){\cal R}(\Gamma),$ (2.21) where $D^{s*}_{{S_{z}}{q}}(\Gamma)$ are the 3D Wigner $D$ functions [116], $\Gamma$ is a shorthand notation for the set of the three Euler angles $(\phi,\theta,\psi)$, and ${\cal R}(\Gamma)=\rme^{-\rmi\phi\hat{S}_{z}}\rme^{-\rmi\theta\hat{S}_{y}}\rme^{-\rmi\psi\hat{S}_{z}}$ (2.22) is the rotation operator in spin space. In (2.21), the indices of the Wigner $D$ functions are $s$, $S_{z}$, and $q$. The operator ${\cal P}^{s}_{{S_{z}}{q}}$ extracts from the symmetry broken wave function a state with a total spin $\hat{\bf S}$ and projection $S_{z}$ along the laboratory $z$ axis. However, $q$ is not a good quantum number of the many-body Hamiltonian, and the most general symmetry restored state is written as a superposition over the components of $q$, i.e., $\Psi_{\rm Spin-P}(s,S_{z};i)=\sum_{q}g_{q}^{i}{\cal P}^{s}_{{S_{z}}{q}}\Psi_{\rm UHF},$ (2.23) where the coefficients $g_{q}^{i}$ are determined through a diagonalization of the many-body Hamiltonian in the space spanned by the nonorthogonal ${\cal P}^{s}_{{S_{z}}{q}}\Psi_{\rm UHF}$ (see also Refs. [117, 118]). In (2.23), the index $i$ reflects the possible degeneracies of spin functions with a given good total-spin quantum number $s$ [119], which is not captured by (2.18). The Peierls-Yoccoz formulation for recovering spin-corrected wave functions applies also in the case when the UHF determinants violate in addition the conservation of spin projection [33], unlike the projector ${\cal P}_{\rm spin}(s)$ [see (2.18)] which acts on UHF determinants having a good $S_{z}=(N^{\alpha}-N^{\beta})/2$ according to the Pople-Nesbet theory presented in section 2.1.1. In the literature [18], there are two distinguishable implementations of symmetry restoration: variation before projection (VBP) and variation after projection (VAP). In the former, which is the one that we mostly use this report, mean-field solutions with broken symmetry are first constructed and then the symmetry is restored via projection techniques as described above. In the latter, the projected wave function is used as the trial wave function directly in the variational principle (in other words the trial function is assured to have the proper symmetry). The VAP is in general more accurate, but more difficult to implement numerically, and it has been used less often in the nuclear-physics literature. In quantum chemistry, the generalized valence bond method [120], or the spin-coupled valence bond method [121], describing covalent bonding between pairs of electrons, employ a variation after projection. For quantum dots, the variation after projection looks promising for reducing the error of the VBP techniques in the transition region from mean-field to Wigner-molecule behavior, where this error is the largest. In fact, it has been found that the discrepancy between variation-before-projection techniques and exact solutions is systematically reduced [23, 76, 122] for stronger symmetry breaking (increasing $R_{W}$ and/or increasing magnetic field). Moreover, in the case of an applied magnetic field (quantum dots) or a rotating trap (Bose gases), our VBP implementation corresponds to projecting cranked symmetry-unrestricted Slater determinants [123]. This is because of the “cranking” terms $-\hbar\omega_{c}L/2$ or $-\hbar\Omega L$ that contribute to the many-body Hamiltonian ${\cal H}$, respectively, with $\omega_{c}=eB/(m^{*}c)$ being the cyclotron frequency and $\Omega$ the rotational frequency of the trap; these terms arise in the single-particle component of ${\cal H}$ [see Equation (2.2) in section 2 and Equation (8.3) in section 8]. The cranking form of the many-body Hamiltonian is particularly advantageous to the variation before projection, since the cranking method provides a first-order approximation to the variation-after-projection restoration of the total angular-momentum $\hat{L}$ [124] (see also Ch 11.4.4 in Ref. [18]). #### 2.2.1 The REM microscopic method in medium and high magnetic field. In our method of hierarchical approximations, we begin with a static electron molecule, described by an unrestricted Hartree-Fock determinant that violates the circular symmetry [20, 23, 25]. Subsequently, the rotation of the electron molecule is described by a post-Hartree-Fock step of restoration of the broken circular symmetry via projection techniques [22, 23, 24, 25, 26, 51, 53]. Since we focus here on the case of strong $B$, we can approximate the UHF orbitals (first step of our procedure) by (parameter free) displaced Gaussian functions; that is, for an electron localized at ${\bf R}_{j}$ ($Z_{j}$), we use the orbital [53] $u(z,Z_{j})=\frac{1}{\sqrt{\pi}\lambda}\exp\left(-\frac{|z-Z_{j}|^{2}}{2\lambda^{2}}-\rmi\varphi(z,Z_{j};B)\right),$ (2.24) with $\lambda=\tilde{l}\equiv\sqrt{\hbar/m^{*}\tilde{\omega}}$; $\tilde{\omega}=\sqrt{\omega_{0}^{2}+\omega_{c}^{2}/4}$, where $\omega_{c}=eB/(m^{*}c)$ is the cyclotron frequency and $\omega_{0}$ specifies the external parabolic confinement. We have used complex numbers to represent the position variables, so that $z=x+\rmi y$, $Z_{j}=X_{j}+\rmi Y_{j}$. The phase guarantees gauge invariance in the presence of a perpendicular magnetic field and is given in the symmetric gauge by $\varphi(z,Z_{j};B)=(xY_{j}-yX_{j})/2l_{B}^{2}$, with $l_{B}=\sqrt{\hbar c/eB}$. For an extended 2D system, the $Z_{j}$’s form a triangular lattice [59, 125]. For finite $N$, however, the $Z_{j}$’s coincide [24, 26, 51, 52, 53] with the equilibrium positions [forming $r$ concentric regular polygons denoted as ($n_{1},n_{2},\ldots,n_{r}$)] of $N=\sum_{q=1}^{r}n_{q}$ classical point charges inside an external parabolic confinement [114]. In this notation, $n_{1}$ corresponds to the innermost ring with $n_{1}>0$. For the case of a single polygonal ring, the notation $(0,N)$ is often used; then it is to be understood that $n_{1}=N$. The wave function of the static electron molecule is a single Slater determinant $|\Psi^{\rm{UHF}}[z]\rangle$ made out of the single-electron wave functions $u(z_{i},Z_{i})$, $i=1,\ldots,N$. Correlated many-body states with good total angular momenta $L$ can be extracted [24, 26, 51, 53] (second step) from the UHF determinant using projection operators. The projected rotating electron molecule state is given by $\displaystyle|\Phi^{\rm{REM}}_{L}\rangle=\int_{0}^{2\pi}\ldots\int_{0}^{2\pi}\rmd\gamma_{1}\ldots\rmd\gamma_{r}|\Psi^{\rm{UHF}}(\gamma_{1},\ldots,\gamma_{r})\rangle\exp\left(\rmi\sum_{q=1}^{r}\gamma_{q}L_{q}\right).$ (2.25) Here $L=\sum_{q=1}^{r}L_{q}$ and $|\Psi^{\rm{UHF}}[\gamma]\rangle$ is the original Slater determinant with all the single-electron wave functions of the $q$th ring rotated (collectively, i.e., coherently) by the same azimuthal angle $\gamma_{q}$. Note that (2.25) can be written as a product of projection operators acting on the original Slater determinant [i.e., on $|\Psi^{\rm{UHF}}(\gamma_{1}=0,\ldots,\gamma_{r}=0)\rangle$]. Setting $\lambda=l_{B}\sqrt{2}$ restricts the single-electron wave function in (2.24) to be entirely in the lowest Landau level (see Appendix in Ref. [53]). The continuous-configuration-interaction form of the projected wave functions [i.e., the linear superposition of determimants in (2.25)] implies a highly entangled state. We require here that $B$ is sufficiently strong so that all the electrons are spin-polarized and that the ground-state angular momentum $L\geq L_{0}\equiv\sum_{i=0}^{N-1}i=N(N-1)/2$ (or equivalently that the fractional filling factor $\nu\equiv L_{0}/L\leq 1$). The state corresponding to $L_{0}$ is a single Slater determinant in the lowest Landau level and is called the “maximum density droplet” [126]. For high $B$, the calculations in this paper do not include the Zeeman contribution, which, however, can easily be added (for a fully polarized dot, the Zeeman contribution to the total energy is $Ng^{*}\mu_{B}B/2$, with $g^{*}$ being the effective Landé factor and $\mu_{B}$ the Bohr magneton). Due to the point-group symmetries of each polygonal ring of electrons in the UHF wave function, the total angular momenta $L$ of the rotating crystalline electron molecule are restricted to the so-called magic angular momenta, i.e., $L_{m}=L_{0}+\sum_{q=1}^{r}k_{q}n_{q},$ (2.26) where the $k_{q}$’s are non-negative integers (when $n_{1}=1$, $k_{1}=0$). Magic angular momenta associated with multiple rings have been discussed in Refs. [24, 26, 51, 52, 53]. For the simpler cases of $(0,N)$ or $(1,N-1)$ rings, see, e.g., Ref. [127] and Ref. [128]. The partial angular momenta associated with the $q$th ring, $L_{q}$ [see (2.25)], are given by $L_{q}=L_{0,q}+k_{q}n_{q},$ (2.27) where $L_{0,q}=\sum_{i=i_{q}+1}^{i_{q}+n_{q}}(i-1)$ with $i_{q}=\sum_{s=1}^{q-1}n_{s}$ $(i_{1}=0)$, and $L_{0}=\sum_{q=1}^{r}L_{0,q}$. The energy of the REM state (2.25) is given [24, 51, 52, 53] by $E^{\rm{REM}}_{L}=\left.{\int_{0}^{2\pi}h([\gamma])\rme^{\rmi[\gamma]\cdot[L]}\rmd[\gamma]}\right/{\int_{0}^{2\pi}n([\gamma])\rme^{\rmi[\gamma]\cdot[L]}\rmd[\gamma]},$ (2.28) with the Hamiltonian and overlap matrix elements $h([\gamma])=\langle\Psi^{\rm{UHF}}([0])|{\cal H}|\Psi^{\rm{UHF}}([\gamma])\rangle$ and $n([\gamma])=\langle\Psi^{\rm{UHF}}([0])|\Psi^{\rm{UHF}}([\gamma])\rangle$, respectively, and $[\gamma]\cdot[L]=\sum_{q=1}^{r}\gamma_{q}L_{q}$. The UHF energies are simply given by $E_{\rm{UHF}}=h([0])/n([0])$. The crystalline polygonal-ring arrangement $(n_{1},n_{2},\ldots,n_{r})$ of classical point charges is portrayed directly in the electron density of the broken-symmetry UHF, since the latter consists of humps centered at the localization sites $Z_{j}$’s (one hump for each electron). In contrast, the REM has good angular momentum and thus its electron density is circularly uniform. To probe the crystalline character of the REM, we use the conditional probability distribution (CPD) defined in (1.1). $P({\bf r},{\bf r}_{0})$ is proportional to the conditional probability of finding an electron at ${\bf r}$, given that another electron is assumed at ${\bf r}_{0}$. This procedure subtracts the collective rotation of the electron molecule in the laboratory frame of referenece, and, as a result, the CPDs reveal the structure of the many body state in the intrinsic (rotating) reference frame. #### 2.2.2 Group structure and sequences of magic angular momenta. It has been demonstrated [25] that the broken-symmetry UHF determinants and orbitals describe 2D electronic molecular stuctures (Wigner molecules) in close analogy with the case of natural 3D molecules. However, the study of Wigner molecules at the UHF level restricts their description to the intrinsic (nonrotating) frame of reference. Motivated by the case of natural atoms, one can take a subsequent step and address the properties of collectively rotating Wigner molecules in the laboratory frame of reference. As is well known, for natural atoms, this step is achieved by writing the total wave function of the molecule as the product of the electronic and ionic partial wave functions. In the case of the purely electronic Wigner molecules, however, such a product wave function requires the assumption of complete decoupling between intrinsic and collective degrees of freedom, an assumption that might be justifiable in limiting cases only. The simple product wave function was used in earlier treatments of Wigner molecules; see, e.g., Ref. [128]. The projected wave functions employed here are integrals over such product wave functions, and thus they account for quantal fluctuations in the rotational degrees of freedom. The reduction of the projected wave functions to the limiting case of a single product wave function is discussed in Ch 11.4.6.1 of Ref. [18]. As was discussed earlier, in the framework of the broken-symmetry UHF solutions, a further step is needed – and this companion step can be performed by using the post-Hartree-Fock method of restoration of broken symmetries via projection techniques (see section 2.2). In this section, we use this approach to illustrate through a couple of concrete examples how certain universal properties of the exact solutions, i.e., the appearance of magic angular momenta [127, 128, 129, 130, 131, 132, 133] in the exact rotational spectra, relate to the symmetry broken UHF solutions. Indeed, we demonstrate that the magic angular momenta are a direct consequence of the symmetry breaking at the UHF level and that they are determined fully by the molecular symmetries of the UHF determinant. Figure 6: (Color online) The UHF solution exhibiting breaking of the circular symmetry for $N=3$ and $S_{z}=1/2$ at $R_{W}=10$ and $B=0$. (a-b): real orbitals for the two spin-up electrons. (c): real orbital for the single spin- down electron. (d): total electron density (ED). (e): spin density (SD, difference of the spin-up minus the spin-down partial electron densities). The choice of the remaining parameters is: $\hbar\omega_{0}=5$ meV and $m^{*}=0.067m_{e}$. Distances are in nanometers. The real orbitals are in 10-3 nm-1 and the densities (electron density and spin density) in 10-4 nm-2. The arrows indicate the spin direction. As an illustrative example, we have chosen the relatively simple, but non trivial case, of $N=3$ electrons. For $B=0$, both the $S_{z}=1/2$ and $S_{z}=3/2$ polarizations can be considered. We start with the $S_{z}=1/2$ polarization, whose broken-symmetry UHF solution [25] is portayed in figure 6 and which exhibits a breaking of the total spin symmetry in addition to the rotational symmetry. Let us denote the corresponding UHF determinant [made out of the three spin orbitals in figure 6(a), figure 6(b), and figure 6(c)] as $|\downarrow\uparrow\uparrow\rangle$. We first proceed with the restoration of the total spin by noticing that $|\downarrow\uparrow\uparrow\rangle$ has a lower point-group symmetry (see Ref. [25]) than the $C_{3v}$ symmetry of an equilateral triangle. The $C_{3v}$ symmetry, however, can be readily restored by applying the projection operator (2.19) to $|\downarrow\uparrow\uparrow\rangle$ and by using the character table of the cyclic $C_{3}$ group (see Table I in Ref. [25]). Then for the intrinsic part of the many-body wave function, one finds two different three-determinantal combinations, namely $\Phi_{\rm{intr}}^{E^{\prime}}(\gamma_{0})=|\downarrow\uparrow\uparrow\rangle+\rme^{2\pi\rmi/3}|\uparrow\downarrow\uparrow\rangle+\rme^{-2\pi\rmi/3}|\uparrow\uparrow\downarrow\rangle,$ (2.29) and $\Phi_{\rm{intr}}^{E^{\prime\prime}}(\gamma_{0})=|\downarrow\uparrow\uparrow\rangle+\rme^{-2\pi\rmi/3}|\uparrow\downarrow\uparrow\rangle+\rme^{2\pi\rmi/3}|\uparrow\uparrow\downarrow\rangle,$ (2.30) where $\gamma_{0}=0$ denotes the azimuthal angle of the vertex of the equilateral triangle associated with the original spin-down orbital in $|\downarrow\uparrow\uparrow\rangle$. We note that, unlike the intrinsic UHF Slater determinant, the intrinsic wave functions $\Phi_{\rm{intr}}^{E^{\prime}}$ and $\Phi_{\rm{intr}}^{E^{\prime\prime}}$ here are eigenstates of the square of the total spin operator ${\hat{\bf S}}^{2}$ ($\hat{\bf S}=\sum_{i=1}^{3}\hat{\bf s}_{i}$) with quantum number $s=1/2$. This can be verified directly by applying ${\hat{\bf S}}^{2}$ to them.555For the appropriate expression of ${\bf S}^{2}$, see equation (6) in Ref. [46]. To restore the circular symmetry in the case of a (0,N) ring arrangement, one applies the projection operator (2.19). Note that the operator ${\cal P}_{L}$ is a direct generalization of the projection operators for finite point-groups discussed in Ref. [25] to the case of the continuous cyclic group $C_{\infty}$ [the phases $\exp(i\gamma L)$ are the characters of $C_{\infty}$]. The symmetry-restored projected wave function, $\Psi_{\rm{PRJ}}$, (having both good total spin and angular momentum quantum numbers) is of the form, $2\pi\Psi_{\rm{PRJ}}=\int^{2\pi}_{0}\rmd\gamma\Phi_{\rm{intr}}^{E}(\gamma)\rme^{\rmi\gamma L},$ (2.31) where now the intrinsic wave function [given by (2.29) or (2.30)] has an arbitrary azimuthal orientation $\gamma$. We note that, unlike the phenomenological Eckardt-frame model [128, 132] where only a single product term is involved, the PRJ wave function in (2.31) is an average over all azimuthal directions over an infinite set of product terms. These terms are formed by multiplying the intrinsic part $\Phi_{\rm{intr}}^{E}(\gamma)$ with the external rotational wave function $\exp(i\gamma L)$ (the latter is properly characterized as “external”, since it is an eigenfunction of the total angular momentum $\hat{L}$ and depends exclusively on the azimuthal coordinate $\gamma$).666Although the wave functions of the Eckardt-frame model are inaccurate compared to the PRJ ones [see (2.31)], they are able to yield the proper magic angular momenta for $(0,N)$ rings. This result is intuitively built in this model from the very beginning via the phenomenological assumption that the intrinsic wave function, which is never specified, exhibits $C_{Nv}$ point-group symmetries. The operator ${\hat{R}(2\pi/3)\equiv\exp(-\rmi 2\pi{\hat{L}}/3})$ can be applied onto $\Psi_{\rm{PRJ}}$ in two different ways, namely either on the intrinsic part $\Phi_{\rm{intr}}^{E}$ or the external part $\exp(\rmi\gamma L)$. Using (2.29) and the property $\hat{R}(2\pi/3)\Phi_{\rm{intr}}^{E^{\prime}}=\exp(-2\pi\rmi/3)\Phi_{\rm{intr}}^{E^{\prime}}$, one finds, $\hat{R}(2\pi/3)\Psi_{\rm{PRJ}}=\exp(-2\pi\rmi/3)\Psi_{\rm{PRJ}},$ (2.32) from the first alternative, and $\hat{R}(2\pi/3)\Psi_{\rm{PRJ}}=\exp(-2\pi L\rmi/3)\Psi_{\rm{PRJ}},$ (2.33) from the second alternative. Now if $\Psi_{\rm{PRJ}}\neq 0$, the only way that equations (2.32) and (2.33) can be simultaneously true is if the condition $\exp[2\pi(L-1)\rmi/3]=1$ is fulfilled. This leads to a first sequence of magic angular momenta associated with total spin $s=1/2$, i.e., $L=3k+1,\;k=0,\pm 1,\pm 2,\pm 3,\ldots$ (2.34) Using (2.30) for the intrinsic wave function, and following similar steps, one can derive a second sequence of magic angular momenta associated with good total spin $s=1/2$, i.e., $L=3k-1,\;k=0,\pm 1,\pm 2,\pm 3,\ldots$ (2.35) Figure 7: (Color online) The UHF case exhibiting breaking of the circular symmetry for $N=3$ and $S_{z}=3/2$ at $R_{W}=10$ and $B=0$. (a-c): real orbitals. (d): the corresponding electron density (ED). The choice of the remaining parameters is: $\hbar\omega_{0}=5$ meV and $m^{*}=0.067m_{e}$. Distances are in nanometers. The real orbitals are in 10-3 nm-1 and the total electron density in 10-4 nm-2. The arrows indicate the spin direction. In the fully spin-polarized case, the UHF determinant is portrayed in figure 7. This UHF determinant, which we denote as $|\uparrow\uparrow\uparrow\;\rangle$, is already an eigenstate of $\hat{\bf S}^{2}$ with quantum number $s=3/2$. Thus only the rotational symmetry needs to be restored, that is, the intrinsic wave function is simply $\Phi^{A}_{\rm{intr}}(\gamma_{0})=|\uparrow\uparrow\uparrow\;\rangle$. Since $\hat{R}(2\pi/3)\Phi^{A}_{\rm{intr}}=\Phi^{A}_{\rm{intr}}$, the condition for the allowed angular momenta is $\exp[-2\pi L\rmi/3]=1$, which yields the following magic angular momenta, $L=3k,\;k=0,\pm 1,\pm 2,\pm 3,\ldots$ (2.36) We note that in high magnetic fields only the fully polarized case is relevant and that only angular momenta with $k>0$ enter in (2.36) (see Ref. [24]). In this case, in the thermodynamic limit, the partial sequence with $k=2q+1$, $q=0,1,2,3,\ldots$ is directly related to the odd filling factors $\nu=1/(2q+1)$ of the fractional quantum Hall effect [via the relation $\nu=N(N-1)/(2L)$]. This suggests that the observed hierarchy of fractional filling factors in the quantum Hall effect may be viewed as a signature originating from the point group symmetries of the intrinsic wave function $\Phi_{\rm{intr}}$, and thus it is a manifestation of symmetry breaking at the UHF mean-field level. We further note that the discrete rotational (and more generally rovibrational) collective spectra associated with symmetry-breaking in a QD may be viewed as finite analogs to the Goldstone modes accompanying symmetry breaking transitions in extended media (see Ref. [10]). Recently there has been some interest in studying Goldstone-mode analogs in the framework of symmetry breaking in trapped BECs with attractive interactions [88]. ### 2.3 The symmetry breaking dilemma and density functional theory Density functional theory (and its extension for cases with a magnetic field known as current density functional theory) was initially considered [3] (and was extensively applied [3, 92, 94]) as a promising method for studying 2D semiconductor QDs. However, it soon became apparent [22, 23, 25, 46] that density functional approaches exhibited severe drawbacks when applied to the regime of strong correlations in QDs, where the underlying physics is associated with symmetry breaking leading to electron localization and formation of Wigner molecules. The inadequacies of density functional approaches in the field of QDs have by now gained broad recognition [41, 134, 135]. In particular, unlike the Hartree-Fock case for which a consistent theory for the restoration of broken symmetries has been developed (see, e.g., the earlier Refs. [18, 30, 31, 32, 33]; for developments in the area of quantum dots, see the more recent Refs. [22, 23, 24, 25, 46]), the breaking of space symmetry within the spin-dependent density functional theory poses [136] a serious dilemma. This dilemma has not been resolved [137] to date; several remedies are being proposed, but none of them appears to be completely devoid of inconsistencies. In particular, a theory for symmetry restoration of broken-symmetry solutions [134, 135] within the framework of density functional theory has not been developed as yet. This puts the density functional methods in a clear disadvantage with regard to the modern fields of quantum information and quantum computing; for example, the description of quantum entanglement (see section 5.1.4 below) requires the ability to calculate many-body wave functions exhibiting good quantum numbers, and thus it lies beyond the reach of density functional theory. Moreover, due to the unphysical self-interaction error, the density-functional theory becomes erroneously more resistant to space symmetry breaking [138] compared to the UHF (which is free from such an error), and thus it fails to describe a whole class of broken symmetries involving electron localization, e.g., the formation at $B=0$ of Wigner molecules in quantum dots [20, 46] and in thin quantum wires [139], the hole trapping at Al impurities in silica [140], or the interaction driven localization-delocalization transition in $d$\- and $f$\- electron systems, like Plutonium [141]. Recently, the shortcomings of the density functional theory to properly describe magnetic phenomena (such as exchange coupling constants associated with symmetry breaking of the total spin) has attracted significant attention in the quantum chemistry literature (see, e.g., Refs. [142, 143, 144]). ### 2.4 More on symmetry restoration methods In the framework of post-Hartree-Fock hierarchical approximations, projection techniques are one of methods used to treat correlations beyond the unrestricted Hartree-Fock. Two other methods are briefly discussed in this section, i.e., the method of symmetry restoration via random phase approximation (RPA) and the generator coordinate method (GCM). #### 2.4.1 Symmetry restoration via random phase approximation. This method introduces energy correlations by considering the effect of the zero-point motion of normal vibrations associated with the small amplitude motion of the time-dependent-Hartree-Fock mean field (which is equivalent to the RPA). In the case of space symmetry breaking, one of the RPA vibrational frequencies vanishes, and the corresponding motion is associated with the rotation of the system as a whole (rotational Goldstone mode), with a moment of inertia given by the so-called Thouless-Valatin expression [145]. The method has been used to calculate correlation energies of atomic nuclei [146, 147] and most recently to restore the broken symmetry in circular quantum dots [148] (mainly for the case of two electrons at zero magnetic field). As discussed in Ref. [148], restoration of the total spin cannot be treated within RPA. #### 2.4.2 The generator coordinate method. The projection techniques by themselves do not take into account quantum correlation effects arising from the vibrations and other large-amplitude intrinsic collective distortions of the Wigner molecule. For the inclusion of the effects of such collective motions, a natural extension beyond projection techniques is the generator coordinate method (see Ch 10 in Ref. [18]). Unlike the RPA, the GCM can treat large-amplitude collective motion in combination with the retoration of the total spin. Indeed, it has been shown that the RPA harmonic vibrations are a limiting small-amplitude case of the large-amplitude collective motion described via the generator coordinate method [18]. The GCM represents an additional step in the hierarchy of approximations described in section 1.5 and its use will result in a further reduction of the difference from the exact solutions. The GCM is complicated and computationally more expensive compared to projection techniques. Recent computational advances, however, have allowed rather extensive applications of the method in nuclear physics (see, e.g., Ref. [38]). As yet, applications of the GCM to quantum dots or trapped atomic gases have not been reported. The GCM employs a very general form for the trial many-body wave functions expressed as a continuous superposition of determinants $|\Psi[a]\rangle$ (or permanents for bosons), i.e., $|\Phi\rangle=\int\rmd[a]f[a]|\Psi[a]\rangle,$ (2.37) where $[a]=(a_{1},a_{2},\ldots,a_{k})$ is a set of collective parameters depending on the physics of the system under consideration. An example of such parameters are the azimuthal angles $(\gamma_{1},\gamma_{2},\ldots,\gamma_{r})$ in the REM trial wave function (2.25). Of course the crucial difference between the REM wave function (2.25) and the general GCM function (2.37) is the fact that the weight coefficients $f[a]$ in the former are known in advance (they coincide with the characters of the underlying symmetry group), while in the latter they are calculated numerically via the Hill-Wheeler-Griffin equations [149, 150] $\int\rmd[a^{\prime}]h(a,a^{\prime})f[a^{\prime}]=E\int\rmd[a^{\prime}]n(a,a^{\prime})f[a^{\prime}],$ (2.38) where $E$ are the eigenenergies, and $h(a,a^{\prime})=\langle\Psi[a]|{\cal H}|\Psi[a^{\prime}]\rangle,$ (2.39) $n(a,a^{\prime})=\langle\Psi[a]|\Psi[a^{\prime}]\rangle$ (2.40) are the Hamiltonian and overlap kernels. The Hill-Wheeler-Griffin equation (2.38) is usually solved numerically by discretization; then one can describe it as a diagonalization of the many-body hamiltonian in a nonorthogonal basis formed with the determinants $|\Psi[a]\rangle$. An example of a potential case for the application of the GCM is an anisotropic quamtum dot ($\zeta<1$, with $\zeta=\omega_{x}/\omega_{y}$) In this case, one cannot use projection techniques to restore the total angular momentum, since the external confinement does not possess circular symmetry. Application of the GCM, however, will produce numerical values for the expansion coefficients $f[\gamma]$, and these values will reduce to $\exp[\rmi\sum_{q=1}^{r}\gamma_{q}L_{q}]$ for the circular case $\zeta=1$ [while the GCM wave function will reduce to the REM wave function (2.25)]. It is apparent that the GCM many-body wave function changes continuously with varying anisotropy $\zeta$, although the symmetry properties of the confinement potential change in an abrupt way at the point $\zeta\rightarrow 1$. ## 3 Symmetry breaking and subsequent symmetry restoration for neutral and charged bosons in confined geometries: Theoretical framework ### 3.1 Symmetry breaking for bosons, Gross-Pitaevskii wave functions, and permanents Mean-field symmetry breaking for bosonic systems is transparent in the context of two-component condensates, where each species is necessarily associated with a different space orbital [151, 152]. For one species of bosons, symmetry breaking can be considered through a generalization of the UHF method of different orbitals for different spins known from the case of electrons in quantum chemistry and in quantum dots (section 2.1.1). Indeed, as shown in Refs. [60, 63, 153], one can allow each bosonic particle to occupy a different orbital $\phi_{i}({\bf r}_{i})$. Then the permanent $|\Phi_{N}\rangle={\rm Perm}[\phi_{1}({\bf r}_{1}),\ldots,\phi_{N}({\bf r}_{N})]$ serves as the many- body wave function of an unrestricted Bose-Hartree-Fock (UBHF) approximation. This wave function reduces to the Gross-Pitaevskii form with the restriction that all bosons occupy the same orbital $\phi_{0}({\bf r})$, i.e., $|\Phi^{\rm{GP}}_{N}\rangle=\prod_{i=1}^{N}\phi_{0}({\bf r}_{i})$, and $\phi_{0}({\bf r})$ is determined self-consistently at the restricted Bose- Hartree-Fock (RBHF) level via the equation [154] $[H_{0}({\bf r}_{1})+(N-1)\int d{\bf r}_{2}\phi^{*}_{0}({\bf r}_{2})v({\bf r}_{1},{\bf r}_{2})\phi_{0}({\bf r}_{2})]\phi_{0}({\bf r}_{1})=\varepsilon_{0}\phi_{0}({\bf r}_{1}).$ (3.1) Here $v({\bf r}_{1},{\bf r}_{2})$ is the two-body repulsive interaction, which is taken to be a contact potential, $v_{\delta}=g\delta({\bf r}_{1}-{\bf r}_{2})$, for neutral bosons, or the Coulomb repulsion $v_{C}=e^{2}Z^{2}/(\kappa|{\bf r}_{1}-{\bf r}_{2}|)$ for charged bosons. The single-particle Hamiltonian is given by $H_{0}({\bf r})=-\hbar^{2}\nabla^{2}/(2m)+m\omega_{0}^{2}{\bf r}^{2}/2,$ (3.2) where $\omega_{0}$ characterizes the circular harmonic confinement, and where we have considered a non-rotating trap. Compared to the fermionic case (see section 2.1.1), the self-consistent determination of the UBHF orbitals is rather complicated and numerically highly demanding (see section 2.5.3 in Ref. [155]). In fact, the self- consistent UHBF equations cannot be put into a standard (canonical) eigenvalue-problem form due to two reasons:(i) unitary transformations cannot be used to simplify the equations, since the permanent of the product of two matrices does not factorize into a product of two simpler terms (unlike the electronic case where the determinant of the product of two matrices is equal to the product of the corresponding determinants) and (ii) as a result of boson statistics, the bosonic orbitals cannot be assumed to be (and remain) orthogonal, which leads to additional coupling terms between the non- orthogonal orbitals. In the literature [156], an attempt has been made to derive unrestricted self- consistent equations for bosons by disregarding point (ii) mentioned above and invoking the assumption of orthonormal orbitals. Such equations of course are not of general validity, although they appear to be useful for describing fragmentation of Bose condensates in double wells. The difficulties of the self-consistent treatment can be bypassed and the UBHF problem can be simplified through consideration of explicit analytic expressions for the space orbitals $\phi_{i}({\bf r}_{i})$. In particular, for repulsive interactions, the bosons must avoid each other in order to minimize their mutual repulsion, and thus, in analogy with the case of electrons in QDs, one can take all the orbitals to be of the form of displaced Gaussians, namely, $\phi_{i}({\bf r}_{i})=\pi^{-1/2}\sigma^{-1}\exp[-({\bf r}_{i}-{\bf a}_{i})^{2}/(2\sigma^{2})].$ (3.3) The positions ${\bf a}_{i}$ describe the vertices of concentric regular polygons, with both the width $\sigma$ and the radius $a=|{\bf a}_{i}|$ of the regular polygons determined variationally through minimization of the total energy $E_{\rm{UBHF}}=\langle\Phi_{N}|{\cal H}|\Phi_{N}\rangle/\langle\Phi_{N}|\Phi_{N}\rangle,$ (3.4) where ${\cal H}=\sum_{i=1}^{N}H_{0}({\bf r}_{i})+\sum_{i<j}^{N}v({\bf r}_{i},{\bf r}_{j})$ (3.5) is the many-body Hamiltonian. With the above choice of localized orbitals the unrestricted permanent $|\Phi_{N}\rangle$ breaks the continuous rotational symmetry. However, the resulting energy gain becomes substantial for stronger repulsion. Controlling this energy gain (the strength of correlations) is the ratio $R_{\delta}$ (or $R_{W}$) between the strength of the repulsive potential and the zero-point kinetic energy. Specifically, for a 2D trap, one has $R_{\delta}=gm/(2\pi\hbar^{2})$ (3.6) for a contact potential (for $R_{W}$, see section 2.1.2). Note that in this section, we refer to the case of a non-rotating trap; the generalization to rotating traps is presented later in section 8. ### 3.2 Repelling bosons in harmonic traps: Restoration of broken symmetry. Although the optimized UBHF permanent $|\Phi_{N}\rangle$ performs very well regarding the total energies of the trapped bosons, in particular in comparison to the resctricted wave functions (e.g., the GP anzatz), it is still incomplete. Indeed, due to its localized orbitals, $|\Phi_{N}\rangle$ does not preserve the circular (rotational) symmetry of the 2D many-body Hamiltonian $H$. Instead, it exhibits a lower point-group symmetry, i.e., a $C_{2}$ symmetry for $N=2$ and a $C_{5}$ one for the $(1,5)$ structure of $N=6$ (see section 8 below). As a result, $|\Phi_{N}\rangle$ does not have a good total angular momentum. In analogy with the case of electrons in quantum dots, this paradox is resolved through a post-Hartree-Fock step of restoration of broken symmetries via projection techniques [23, 24, 25, 60], yielding a new wave function $|\Psi_{N,L}^{\rm{PRJ}}\rangle$ with a definite angular momentum $L$, that is $2\pi|\Psi_{N,L}^{\rm{PRJ}}\rangle=\int^{2\pi}_{0}\rmd\gamma|\Phi_{N}(\gamma)\rangle\rme^{\rmi\gamma L},$ (3.7) where $|\Phi_{N}(\gamma)\rangle$ is the original UBHF permanent having each localized orbital rotated by an azimuthal angle $\gamma$, with $L$ being the total angular momentum. The projection yields wave functions for a whole rotational band. Note that the projected wave function $|\Psi_{N,L}^{\rm{PRJ}}\rangle$ in (3.7) may be regarded as a superposition of the rotated permanents $|\Phi_{N}(\gamma)\rangle$, thus corresponding to a “continuous-configuration-interaction” solution. The energies of the projected states are given by $E_{L}^{\rm{PRJ}}=\langle\Psi_{N,L}^{\rm{PRJ}}|{\cal H}|\Psi_{N,L}^{\rm{PRJ}}\rangle/\langle\Psi_{N,L}^{\rm{PRJ}}|\Psi_{N,L}^{\rm{PRJ}}\rangle.$ (3.8) ## 4 Other many-body methods ### 4.1 Exact diagonalization methods: Theoretical framework We will discuss the essential elements of the exact-diagonalization method here by considering the special, but most important case of a many-body Hilbert space defined via the restriction that the single-particle states belong exclusively to the lowest Landau level. For 2D electrons, this LLL restriction is appropriate in the case of very high $B$. For rotating bosons in a harmonic trap, the LLL restriction is appropriate for $\Omega\sim\omega_{0}$ and a very weak repulsive contact potential. The particulars of the EXD method for quantum dots in the case of field-free (and/or low $B$) conditions will be discussed below in section 5.1.2. For sufficiently high magnetic field values (i.e., in the fractional quantum Hall effect, regime), the electrons are fully spin-polarized and the Zeeman term (not shown here) does not need to be considered. In the $B\rightarrow\infty$ limit, the external confinement $V(x,y)$ can be neglected, and the many-body $H$ can be restricted to operate in the lowest Landau level, reducing to the form [24, 26, 52, 53] ${\cal H}_{\rm LLL}=N\frac{\hbar\omega_{c}}{2}+\sum_{i=1}^{N}\sum_{j>i}^{N}\frac{e^{2}}{\kappa r_{ij}},$ (4.1) where $\omega_{c}=eB/(m^{*}c)$ is the cyclotron frequency. Namely, one needs to diagonalize the interaction Hamiltonian only. For the case of rotating bosons in the LLL, one needs to replace in (4.1) the Coulomb interaction by $g\delta({\bf r}_{i}-{\bf r}_{j})$ and the cyclotron frequency by $2\Omega$ (see the Appendix for the details of the equivalence between magnetic field $B$ and rotational frequency $\Omega$; in short $\omega_{c}\rightarrow 2\Omega$). For a given total angular momentum $L=\sum_{k=1}^{N}l_{k}$, the exact- diagonalization $N$-body wave function is a linear superposition of Slater determinants for fermions (or permanents for bosons) $\Psi(J)$ made out of lowest-Landau-level single-particle wave functions (see the Appendix), $\phi_{l}(z)=\frac{1}{\Lambda\sqrt{\pi l!}}\left(\frac{z}{\Lambda}\right)^{l}e^{-zz^{*}/(2\Lambda^{2})},$ (4.2) where $\Lambda=\sqrt{2\hbar c/(eB)}=l_{B}\sqrt{2}$ for the case of electrons in QDs ($l_{B}$ being the magnetic length), and $\Lambda=\sqrt{\hbar/(m\omega_{0})}$ for the case of bosons in rotating traps. In (4.2), we used complex coordinates $z=x+\rmi y$, instead of the usual vector positions ${\bf r}=(x,y)$; below we will use either notation interchangeably as needed. Thus, the many-body EXD wave function is written as $\Phi^{\rm EXD}(z_{1},z_{2},\ldots,z_{N})=\sum_{J=1}^{K}C_{J}\Psi(J)$ (4.3) with the index $J$ denoting any set of $N$ single-particle angular momenta $\\{l_{1},l_{2},\ldots,l_{N}\\}$ such that $l_{1}<l_{2}<\ldots<l_{N}$ (4.4) for fermions and $l_{1}\leq l_{2}\leq\ldots\leq l_{N}$ (4.5) for bosons, the absence or presence of the equal signs being determined by the different statistics between fermions and bosons, respectively. Using the expansion (4.3), one transforms the many-body Schrödinger equation ${\cal H}\Phi^{\rm EXD}(z_{1},z_{2},\ldots,z_{N})=E\Phi^{\rm EXD}(z_{1},z_{2},\ldots,z_{N}),$ (4.6) into a matrix eigenvalue problem. When the single-particle states are orthonormal [like the LLL ones in (4.2)], the matrix elements $\langle\Psi(I)|{\cal H}|\Psi(J)\rangle$ between two Slater determinants are calculated using the so-called Slater rules (see, e.g., Ch 2.3.3 of Ref. [19]). For the case of bosons, the correpsonding rules for the matrix elements $\langle\Psi(I)|{\cal H}|\Psi(J)\rangle$ between two permanents are given in the appendix of Ref. [72]. We remark here that the calculation of energies associated with projected wave functions [see, e.g., Equation (3.8)] requires calculation of similar matrix elements between two Slater determinants (or permanents) with non-orthogonal orbitals; the corresponding formulas for the case of fermions can be found in Ch 6.3. of Ref. [157], and for the case of bosons in Ref. [155]. Of course a necessary ingredient for the application of the above rules is the knowledge of the matrix elements of the two-body interaction $v({\bf r}_{1},{\bf r}_{2})$, i.e., $v_{\alpha\beta\gamma\Delta}\equiv\int\rmd{\bf r}_{1}\rmd{\bf r}_{2}\phi^{*}_{\alpha}(1)\phi^{*}_{\beta}(2)v(1,2)\phi_{\gamma}(1)\phi_{\Delta}(2).$ (4.7) In the general case, these two-body matrix elements need to be calculated numerically. For the simpler case specified by the LLL orbitals (4.2), the two-body matrix elements are given by analytic expressions. In particular, for the Coulomb repulsion, see Refs. [26, 158]; for a contact potential, see Ref. [159]. The dimension ${\cal D}$ of the Hilbert space increases very fast with the number of particles $N$ and the value of the total angular momentum $L$, and is is controlled by the maximum allowed single-particle angular momentum $l_{\rm{max}}$, such that $l_{k}\leq l_{\rm{max}}$, $1\leq k\leq N$. By varying $l_{\rm{max}}$, we can check that this choice produces well converged numerical results. For the solution of the large scale, but sparse, EXD matrix eigenvalue problem associated with the special Hamiltonian $H_{\rm{LLL}}$ [or the general one in (2.1)], we use the ARPACK computer code [160]. The availabilty of analytic expressions for the two-body interaction has greatly facilitated exact-diagonalization calculations in the lowest Landau level (appropriate for quantum dots at high $B$), and in this case (starting with Refs. [26, 161]) diagonalization of large matrices of dimensions of order 500,000 x 500,000 has become a commom occurrence. For circular quantum dots, similar analytic expressions for the matrix elements of the Coulomb interaction between general Darwin-Fock orbitals [162, 163] (i.e., the single- particle orbitals of a circular 2D harmonic oscillator under a perpendicular magnetic field $B$) are also available [129, 164], but they are not numerically as stable as Tsiper’s expressions [158] in the lowest Landau level. Exact-diagonalization calculations for field-free (and/or low $B$) conditions have been presented in several papers. Among them, we note the exact- diagonalization calculations of Refs. [77, 97, 165, 166, 167, 168, 169, 170]. EXD calculations employing Coulombic two-body matrix elements that are calculated numerically have also been reported for elliptic quantum dots (see section 5.1.2 below). Furthermore, some authors have used the method of hyperspherical harmonics [171] for circular quantum dots, while others have carried out exact-diagonalization calculations for quantum dots with a polygonal external confinement [172]. Concerning EXD calculations in the lowest Landau level, we mention Refs. [26, 95, 128, 129, 130, 131, 132, 173, 174, 175, 176] for the case of quantum dots (high $B$) and Refs. [67, 68, 69, 72, 177, 178] for the case of bosons in rapidly rotating traps. A version of EXD in the LLL uses a correlated basis constructed out of composite-fermion wave functions [176], while another exact-diagonalization version used non-orthogonal floating Gaussians in the place of the usual single-particle states (4.2) in the LLL. For two electrons in a single quantum dot, exact calculations have been carried out through separation into center-of-mass and relative coordinates [50, 179, 180]. In addition, EXD calculations have been reported for two electrons in a double quantum dot [89, 181]. It is of interest to note that the EXD approach is also used in other fields, but under different names. In particular, the term “shell model calculations” is used in nuclear theory, while the term “full configuration interaction” is employed in quantum chemistry. Figure 8: (Color online) Exact-diagonalization electron densities (EDs) and spin-resolved CPDs for $N=3$ electrons in an anisotropic quantum dot at zero magnetic field ($B=0$) and in the case of a strong Coulomb repulsion with a dielectric constant $\kappa=1$. (a-b) Results for the ground state (which has total spin $s=1/2$ and spin projection $S_{z}=1/2$) when $\hbar\omega_{x}=4.23$ meV and $\hbar\omega_{y}=5.84$ meV (i.e., for an intermediate anisotropy $\zeta=0.724$). (c-f) Results for the first excited state (which has also total spin $s=1/2$ and spin projection $S_{z}=1/2$) when $\hbar\omega_{x}=3.137$ meV and $\hbar\omega_{y}=6.274$ meV (i.e., for a strong anisotropy $\zeta=0.5$). The thick arrows and solid dots in the CPDs indicate the spin direction $\sigma_{0}$ and position ${\bf r}_{0}$ of the fixed electron [see (1.2)]. The thin arrows indicate the spin direction of the remaining two electrons. The effective mass is $m^{*}=0.070m_{e}$ for the intermediate anisotropy (a-b) and $m^{*}=0.067m_{e}$ for the strong anisotropy (c-f). Lengths are in nanometers. The vertical axes are in arbitrary units. #### 4.1.1 An example involving spin-resolved CPDs. Here we present an example of an EXD calculation exhibiting formation of a Wigner molecule in quantum dots. The case we chose is that of $N=3$ electrons under zero magnetic field in an anisotropic quantum dot with $\hbar\omega_{x}=4.23$ meV and $\hbar\omega_{y}=5.84$ meV (i.e., with an intermediate anisotropy $\zeta=\omega_{x}/\omega_{y}=0.724$) and dielectric constant $\kappa=1$ (strong interelectron repulsion). In particular, figure 8 displays results for the ground state of the three electrons with total spin $s=1/2$ and spin projection $S_{z}=1/2$. The electron density in figure 8(a) has the shape of a diamond and suggests formation of a Wigner molecule resonating between two isosceles triangular isomers (which are the mirror image of each other). The detailed interlocking of the two triangular configurations is further revealed in the spin-resolved CPD that is displayed in figure 8(b). It can be concluded that one triangle is formed by the points ${\bf R}_{1}=(0,-20)$ nm, ${\bf R}_{2}=(-43,10)$ nm, and ${\bf R}_{3}=(43,10)$ nm, while the second one (its mirror with respect to the $x$-axis) is formed by the points ${\bf R^{\prime}}_{1}=(0,20)$ nm, ${\bf R^{\prime}}_{2}=(-43,-10)$ nm, and ${\bf R^{\prime}}_{3}=(43,-10)$ nm. The two-triangle configuration discussed for three electrons above may be seen as the embryonic precursor of a quasilinear structure of two intertwined “zig- zag” crystalline chains. Such double zig-zag crystaline chains may also be related to the single zig-zag Wigner-crystal chains discussed recently in relation to spontaneous spin polarization in quantum wires [182, 183]. For strong anisotropies (e.g., $\zeta\leq 1/2$), the three electrons form a straightforward linear Wigner molecule [see the electron density in figure 8(c)], and the spin-resolved CPDs can be used to demonstrate [184] formation of prototypical entangled states, like the so-called $W$ states [185, 186]. From the CPDs [displayed in figure 8(d), figure 8(e) and figure 8(f)] of the first excited state (having $s=1/2$ and $S_{z}=1/2$), one can infer that its intrinsic spin structure is of the form $|\uparrow\uparrow\downarrow\;\rangle-|\downarrow\uparrow\uparrow\;\rangle$. The ground state (not shown) of this linear Wigner molecule has also a total spin $s=1/2$ and spin projection $S_{z}=1/2$, and its intrinsic spin structure corresponds to a form $2|\uparrow\downarrow\uparrow\;\rangle-|\uparrow\uparrow\downarrow\;\rangle-|\downarrow\uparrow\uparrow\;\rangle$ [184]. ### 4.2 Particle localization in Monte-Carlo approaches Quantum Monte-Carlo (MC) approaches [187] have been successfully used in many areas of condensed-matter physics; they are divided in two main branches, path-integral MC (PIMC) and variational/diffusion MC (V/DMC). Unlike the exact diagonalization, quantum MC approaches cannot calculate excited states, and they are restricted to the description of ground-state properties Applied to the case of circular quantum dots at zero magnetic field, PIMC calculationss [188, 189, 190, 191, 192] have been able to reproduce and describe electron localization with increasing $R_{W}$ and formation of Wigner molecules. In 2D quantum dots, a focus of the PIMC studies [188, 191] has been the determination of the critical value, $R_{W}^{\rm cr}$, for the Wigner parameter at which the phase transition from a Fermi liquid to a Wigner molecule occurs. Naturally, only an estimate of this critical value can be determined, since the phase transition is not sharp, but smooth, due to the finite size of the quantum dot. Obtaining a precise value of $R_{W}^{\rm cr}$ is also hampered by the variety of criteria employed by different researchers in the determination of this transition (e.g., height of localized density humps, appearance of a hump at the center of the dot, etc.). In the literature of PIMC studies [188, 191], one finds the critical value $R_{W}^{\rm cr}\sim 4$, which is in agreement with exact-diagonalization studies [77]. This is also in general agreement with the estimate $R_{W}^{\rm cr}\sim 1$ based on the abrupt onset of spatial symmetry breaking in unrestricted Hartree-Fock [20]. Of course the unrestricted-Hartree-Fock estimate has to be refined through the subsequent step of symmetry restoration. We believe that it is most appropriate to consider these two estimates mentioned above as the lower and upper limit of a transition region. The important conclusion is that the transition to Wigner crystallization in quantum dots takes place for much higher electron densities compared to the infinite two-dimensional electron gas (for which a value $R_{W}^{\rm cr}\sim 37$ [193] has been reported.)777Often the Wigner-Seitz radius $r_{s}$, in units of the effective Bohr radius $a^{*}_{B}=\hbar^{2}\kappa/(m^{*}e^{2})$ of the quantum dot, is used instead of the Wigner parameter $R_{W}$ (denoted some times by $\lambda$). In these units, one has $r_{s}\approx R_{W}$. A disadvantage of the PIMC method is that the case of an applied magnetic field cannot be easily incorporated in its formalism, and therefore related studies have not been reported. Other well known difficulties are the fermion sign problem and the nonconservation of total spin [77, 188]. Commenting on the other main branch of quantum Monte Carlo, i.e., the variational/diffusion MC, we wish to stress the crucial role played by the general form of the trial wave function used. Indeed, an early V/DMC study [194] using a single configurational state function (i.e., a primitive combination of products of Slater determinants for the two spin directions that is an eigenstate of the total angular momentum $\hat{L}$, the square of the total spin $\hat{\bf S}^{2}$, and the total-spin projection $S_{z}$) was unable to describe the formation of Wigner molecules in quatum dots at zero magnetic field. Another V/DMC study [195] managed to demonstrate electron localization, but at the cost of using a single product of two Slater determinants (multiplied by a Jastrow factor) which violated the conservation of both the total angular momentum and total spin (without the possibility of further corrections related to symmetry restoration). Most recently, more sophisticated trial wave functions involving a large number of configurational state functions with good total angular momentum and total spin have been employed, which enabled eventually confirmation of the formation of Wigner molecules via V/DMC methods, both at zero [78] and high magnetic field [196]. There are, however, disagreements between the V/DMC results [197] and those from PIMC and EXD calculations concerning the details of Wigner-molecule formation in circular quantum dots in the absence of an applied magnetic field. In particular, these disagreements focus on the density scale for the cross-over and the strength of azimuthal and radial electron correlations as a function of $R_{W}$. Such disagreements remain an open question for two reasons: (i) The criterion of lowest energy (evoked by the V/DMC approaches) is not sufficient to guarantee the quality of the variational many-body wave function. A counterexample to this lowest-energy criterion was presented by us for the case of the Laughlin wave functions in Refs. [24, 52] (see also section 6.4). Most recently, this point was also illustrated within the framework of variational Monte Carlo calculations [198]. (ii) The V/DMC studies for larger $N$ [78, 197] have presented only calculations for CPDs. However, due to the presence of dummy integrations in (1.1) (which result in an averaging over the remaining $N-2$ particles), the ability of the CPDs to portray the intrinsic crystalline structure of the Wigner molecule diminishes with increasing $N$. As a result, higher-order correlation functions, like $N$-point correlations, may be required. The fact that higher-order correlation functions reflect the crystalline correlations more accurately than the CPDs was illustrated for the case of rotating boson molecules in Ref. [72] (see also section 8.2). A detailed comparison between ground-state energies calculated with quantum MC and exact-diagonalization methods can be found in Ref. [77]. For a comparison between variation-before-projection (see section 2.2) and V/DMC total energies, see Ref. [122]. ## 5 The strongly correlated regime in two-dimensional quantum dots: The two- electron problem and its significance In section 2 and section 3, we focused on the general principles and the essential theoretical framework of the method of symmetry breaking and of subsequent symmetry restoration for finite condensed-matter systems. In addition, in section 4, we presented the basic elements of the exact- diagonalization approach. In the following four sections, we will focus on specific applications and predictions from these methods in the field of semiconductor quantum dots and of ultracold bosons in harmonic traps, in particular regarding the emergence and properties of Wigner molecules under various circumstances. At the same time we will continue to elaborate and further expand on more technical aspects of these methods. In this section, we start by concentrating on the description of two-electron molecules in QDs. A discussion on the importance of the two-electron problem is given in section 5.3. ### 5.1 Two-electron elliptic dot at low magnetic fields Here, we present an exact diagonalization and an approximate (generalized Heitler-London, GHL) microscopic treatment for two electrons in a single elliptic QD specified by parameters that correspond to a recently fabricated experimental device [41]. The two-dimensional Hamiltonian for the two interacting electrons is given by ${\cal H}=H({\bf r}_{1})+H({\bf r}_{2})+\gamma e^{2}/(\kappa r_{12}),$ (5.1) where the last term is the Coulomb repulsion, $\kappa$ (12.5 for GaAs) is the dielectric constant, and $r_{12}=|{\bf r}_{1}-{\bf r}_{2}|$. The prefactor $\gamma$ accounts for the reduction of the Coulomb strength due to the finite thickness of the electron layer in the $z$ direction and for any additional screening effects due to the gate electrons. $H({\bf r})$ is the single- particle Hamiltonian for an electron in an external perpendicular magnetic field ${\bf B}$ and an appropriate confinement potential (2.2). For an elliptic QD, the external potential is written as $V(x,y)=\frac{1}{2}m^{*}(\omega^{2}_{x}x^{2}+\omega^{2}_{y}y^{2}).$ (5.2) Here the effective mass is taken to be $m^{*}=0.07m_{0}$. In the Hamiltonian (2.2), we neglect the Zeeman contribution due to the negligible value ($g^{*}\approx 0$) of the effective Landé factor in our sample [199]. #### 5.1.1 Generalized Heitler-London approach. The GHL method for solving the Hamiltoninian (5.1) consists of two steps. In the first step, we solve selfconsistently the ensuing unrestricted Hartree- Fock equations allowing for lifting of the double-occupancy requirement (imposing this requirement gives the restricted HF method, RHF). For the $S_{z}=0$ solution, this step produces two single-electron orbitals $u_{L,R}({\bf r})$ that are localized left $(L)$ and right $(R)$ of the center of the QD [unlike the RHF method that gives a single doubly-occupied elliptic (and symmetric about the origin) orbital]. At this step, the many-body wave function is a single Slater determinant $\Psi_{\rm{UHF}}(1\uparrow,2\downarrow)\equiv|u_{L}(1\uparrow)u_{R}(2\downarrow)\rangle$ made out of the two occupied UHF spin-orbitals $u_{L}(1\uparrow)\equiv u_{L}({\bf r}_{1})\alpha(1)$ and $u_{R}(2\downarrow)\equiv u_{R}({\bf r}_{2})\beta(2)$, where $\alpha(\beta)$ denotes the up (down) [$\uparrow(\downarrow)$] spin. This UHF determinant is an eigenfunction of the projection $S_{z}$ of the total spin $\hat{S}=\hat{s}_{1}+\hat{s}_{2}$, but not of $\hat{S}^{2}$ (or the parity space-reflection operator). In the second step, we restore the broken parity and total-spin symmetries by applying to the UHF determinant the projection operator (2.18). For two electrons, this operator reduces to ${\cal P}_{\rm spin}^{s,t}=1\mp\varpi_{12}$, where the operator $\varpi_{12}$ interchanges the spins of the two electrons; the upper (minus) sign corresponds to the singlet. The final result is a generalized Heitler-London two-electron wave function $\Psi^{s,t}_{\rm{GHL}}({\bf r}_{1},{\bf r}_{2})$ for the ground-state singlet (index $s$) and first-excited triplet (index $t$), which uses the UHF localized orbitals, $\Psi^{s,t}_{\rm{GHL}}({\bf r}_{1},{\bf r}_{2})\propto{\bf(}u_{L}({\bf r}_{1})u_{R}({\bf r}_{2})\pm u_{L}({\bf r}_{2})u_{R}({\bf r}_{1}){\bf)}\chi^{s,t},$ (5.3) where $\chi^{s,t}=(\alpha(1)\beta(2)\mp\alpha(2)\beta(1))$ is the spin function for the 2$e$ singlet and triplet states. The general formalism of the 2D UHF equations and of the subsequent restoration of broken spin symmetries was presented in section 2.2. The use of optimized UHF orbitals in the generalized Heitler-London method is suitable for treating single elongated QDs [46], including the special case of elliptically deformed ones discussed in this section. The GHL is equally applicable to double QDs with arbitrary interdot-tunneling coupling [46]. In contrast, the Heitler-London (HL) treatment [82] (known also as the simple Valence bond), where non-optimized “atomic” orbitals of two isolated QDs are used, is appropriate only for the case of a double dot with small interdot- tunneling coupling [48]. The orbitals $u_{L,R}({\bf r})$ are expanded in a real Cartesian harmonic- oscillator basis, i.e., $u_{L,R}({\bf r})=\sum_{j=1}^{K}C_{j}^{L,R}\varphi_{j}({\bf r}),$ (5.4) where the index $j\equiv(m,n)$ and $\varphi_{j}({\bf r})=X_{m}(x)Y_{n}(y)$, with $X_{m}(Y_{n})$ being the eigenfunctions of the one-dimensional oscillator in the $x$($y$) direction with frequency $\omega_{x}$($\omega_{y}$). The parity operator ${\cal P}$ yields ${\cal P}X_{m}(x)=(-1)^{m}X_{m}(x)$, and similarly for $Y_{n}(y)$. The expansion coefficients $C_{j}^{L,R}$ are real for $B=0$ and complex for finite $B$. In the calculations we use $K=54$ and/or $K=79$, yielding convergent results. #### 5.1.2 Exact diagonalization. In the EXD method, the many-body wave function is written as a linear superposition over the basis of non-interacting two-electron determinants, i.e., $\Psi^{s,t}_{\rm{EXD}}({\bf r}_{1},{\bf r}_{2})=\sum_{i<j}^{2K}{\cal A}_{ij}^{s,t}|\psi(1;i)\psi(2;j)\rangle,$ (5.5) where $\psi(1;i)=\varphi_{i}(1\uparrow)$ if $1\leq i\leq K$ and $\psi(1;i)=\varphi_{i-K}(1\downarrow)$ if $K+1\leq i\leq 2K$ [and similarly for $\psi(2,j)$]. The total energies $E^{s,t}_{\rm{EXD}}$ and the coefficients ${\cal A}_{ij}^{s,t}$ are obtained through a “brute force” diagonalization of the matrix eigenvalue equation corresponding to the Hamiltonian in (5.1). The exact-diagonalization wave function does not immediately reveal any particular form, although, our calculations below show that it can be approximated by a GHL wave function in the case of the elliptic dot under consideration. #### 5.1.3 Results and comparison with measurements. To model the experimental quantum dot device, we take, following Ref. [41], $\hbar\omega_{x}=4.23$ meV, $\hbar\omega_{y}=5.84$ meV, $\kappa=12.5$, and $\gamma=0.862$. The corresponding anisotropy is $\omega_{y}/\omega_{x}=1.38$, indicating that the quantum dot considered here is closer to being circular than in other experimental systems [45, 80]. Figure 9: (Color online) Differentiated current $\rmd I/\rmd V_{\mathrm{pg}}$ at $V_{\mathrm{bias}}=2.5$ mV (the subscript ${\rm pg}$ denotes the plunger gate). Gray striped regions (red online) marked by symbols correspond to positive (peaks) $dI/dV_{\mathrm{pg}}$. The dark black region (also black online) corresponds to negative $dI/dV_{\mathrm{pg}}$. Electron numbers $N$ are indicated. Transitions between the one-electron ground state and the 2$e$ spin-singlet ground (excited for $B>3.8$ T) state ($S_{0}$), spin-triplet excited (ground for $B>3.8$ T) state ($T_{+}$), spin-singlet excited state ($S_{2}$), and spin-triplet plus center-of-mass excited state ($T_{+,{\rm CM}}$) are labeled. Figure 10: (Color online) Calculated exact- diagonalization energy spectrum in a magnetic field, referenced to $2\hbar\sqrt{\omega_{0}^{2}+\omega_{c}^{2}/4}$, of a 2$e$ dot with anisotropic harmonic confinement (for the dot parameters, see text). We have adopted the notation $(N_{x},N_{y},n,m)$, where $(N_{x},N_{y})$ refer to the CM motion along the $x$ and $y$ axes and $(n,m)$ refer to the number of radial nodes and angular momentum of the relative motion in the corresponding circular dot. Inset: The EXD spectrum of the corresponding circular dot. Only the $(n,m)$ indices are shown, since $N_{x}=N_{y}=0$ for all the plotted curves. Solid lines denote singlets. Dashed lines denote triplets. As shown in Ref. [41], the experimental findings can be quantitatively interpreted by comparing to the results of the EXD calculations for two electrons in an anisotropic harmonic confinement potential with the parameters listed above. All the states observed in the measured spectra (as a function of the magnetic field) can be unambiguously identified [41] with calculated ground-state and excited states of the two-electron Hamiltonian (compare figure 9 and figure 10). Moreover, the calculated magnetic-field-dependent energy splitting, $J_{\rm{EXD}}(B)=E^{t}_{\rm{EXD}}(B)-E^{s}_{\rm{EXD}}(B)$, between the two lowest singlet ($S_{0}$) and triplet ($T_{+}$) states is found to be in remarkable agreement with the experiment [see figure 11]. Figure 11: (Color online) Comparison between the lowest-triplet/lowest-singlet energy splitting [$J(B)$] calculated with different methods and the experimental results (open squares). Solid line (online magenta): EXD. Dotted line (online green): GHL. Dashed line (online red): RHF. For the parameters used in the calculation to model the anisotropic QD, see text. A deeper understanding of the structure of the many-body wave function can be acquired by comparing the measured $J(B)$ with that calculated within the GHL and RHF approximations. To facilitate the comparisons, the calculated $J_{\rm{GHL}}(B)$ and $J_{\rm{RHF}}(B)$ curves are plotted also in figure 11, along with the exact-diagonalization result and the measurements. Both the RHF and GHL schemes are appealing intuitively, because they minimize the total energy using single-particle orbitals. It is evident, however, from figure 11 that the RHF method, which assumes that both electrons occupy a common single- particle orbital, is not able to reproduce the experimental findings. On the contrary, the generalized Heitler-London approach, which allows the two electrons to occupy two spatially separated orbitals, appears to be a good approximation. Plotting the two GHL orbitals [see figure 12] for the singlet state clearly demonstrates that the two electrons do not occupy the same orbital, but rather fill states that are significantly spatially separated. Figure 12: (Color online) Single-particle UHF orbitals (modulus square) that are used in the construction of the GHL wave function in (5.3). Arrows indicate up and down spins. For the parameters used in the calculation to model the anisotropic QD, see text. Lengths in nm and orbital densities in 10-3 nm-2. The UHF orbitals from which the GHL singlet state is constructed [see (5.3)] are displayed on figure 12 for both the $B=0$ and $B=3.8$ T cases. The spatial shrinking of these orbitals at the higher $B$-value illustrates the “dissociation” of the electron dimer with increasing magnetic field. The asymptotic convergence (beyond the ST point) of the energies of the singlet and triplet states, [i.e., $J(B)\rightarrow 0$ as $B\rightarrow\infty$] is a reflection of the dissociation of the 2$e$ molecule, since the ground-state energy of two fully spatially separated electrons (zero overlap) does not depend on the total spin. We stress again that the RHF, which corresponds to the more familiar physical picture of a QD-Helium atom, fails to describe this dissociation, because $J_{\rm{RHF}}(B)$ diverges as the value of the magnetic field increases. In contrast to the RHF, the GHL wave function is able to capture the importance of correlation effects. Further insight into the inportance of correlations in our QD device can be gained through inspection [41] of the conditional probability distributions [see (1.1)] associated with the EXD solutions; see an illustration in figure 13. Indeed, already at zero magnetic field, the calculated CPDs provide further support of the physical picture of two localized electrons forming a state resembling an H2-type [23, 46, 41] Wigner molecule [20, 188]. Figure 13: (Color online) CPDs extracted from the exact-diagonalization wave function for the singlet state for $B=0$ and $B=3.8$ T. The CPD expresses the conditional probability for finding the second electron at position ${\bf r}$ given that the first electron is located at ${\bf r}_{0}$ (denoted by a heavy solid dot). For the parameters used in the calculation to model the anisotropic QD, see text. Lengths in nm and CPDs in arbitrary units. #### 5.1.4 Degree of entanglement. Further connections between the strong correlations found in our microscopic treatment and the theory of quantum computing [48] can be made through specification of the degree of entanglement between the two localized electrons in the molecular dimer. For two electrons, we can quantify the degree of entanglement by calculating a well-known measure of entanglement such as the von Neumann entropy [42, 200] for indistinguishable particles. To this effect, one needs to bring the EXD wave function into a diagonal form (the socalled “canonical form” [200, 201]), i.e., $\Psi^{s,t}_{\rm{EXD}}({\bf r}_{1},{\bf r}_{2})=\sum_{k=1}^{M}z^{s,t}_{k}|\Phi(1;2k-1)\Phi(2;2k)\rangle,$ (5.6) with the $\Phi(i)$’s being appropriate spin orbitals resulting from a unitary transformation of the basis spin orbitals $\psi(j)$’s [see (5.5)]; only terms with $z_{k}\neq 0$ contribute. The upper bound $M$ can be smaller (but not larger) than $K$ (the dimension of the single-particle basis); $M$ is referred to as the Slater rank. One obtains the coefficients of the canonical expansion from the fact that the $|z_{k}|^{2}$ are eigenvalues of the hermitian matrix ${\cal A}^{\dagger}{\cal A}$ [${\cal A}$, see (5.5), is antisymmetric]. The von Neumann entropy is given by ${\cal S}=-\sum_{k=1}^{M}|z_{k}|^{2}\log_{2}(|z_{k}|^{2})$ with the normalization $\sum_{k=1}^{M}|z_{k}|^{2}=1$. Figure 14: Von Neumann entropy for the lowest singlet EXD state of the elliptic dot as a function of the magnetic field $B$. On the top, we show histograms for the $|z_{k}|^{2}$ coefficients [see (5.6)] of the singlet state at $B=3.8$ T (left) and $B=8.0$ T (right) illustrating the dominance of two determinantal configurations (in agreement with the generalized Heitler-London picture). Note the small third coefficient $|z_{3}|^{2}=0.081$ for $B=8.0$ T. For the parameters used to model the experimental device, see text. The EXD singlet has obviously a Slater rank $M>2$. The von Neumann entropy for the EXD singlet (${\cal S}^{s}_{\rm{EXD}}$) is displayed in figure 14. It is remarkable that ${\cal S}^{s}_{\rm{EXD}}$ increases with increasing $B$, but remains close to unity for large $B$, although the maximum allowed mathematical value is $\log_{2}(K)$ [for example, for $K=79$, $\log_{2}(79)=6.3$]. The saturation of the entropy for large $B$ to a value close to unity reflects the dominant (and roughly equal at large $B$) weight of two configurations in the canonical expansion [see (5.6)] of the exact- diagonalization wave function, which are related [42] to the two terms in the canonical expansion of the GHL singlet. This is illustrated by the histograms of the $|z_{k}^{s}|^{2}$ coefficients for $B=3.8$ T and $B=8.0$ T in figure 14 (top). Notice that the ratio $|z_{2}|^{2}/|z_{1}|^{2}$ reflects the extent of the overlap between the two GHL orbitals [42], with the ratio increasing for smaller overlaps (corresponding to a more complete dissociation of the Wigner molecule). The above discussion illustrates that microscopic calculations that are shown to reproduce experimental spectra [41] can be used to extract valuable information that allows assessment of the suitability of a given device for quantum computations. ### 5.2 Two-electron circular dots at zero magnetic field In section 2.2, we illustrated the formation of “rotating electron molecules” in the case of a two-electron circular QD, where one needs to consider restoration of the rotational symmetry as well, in addition to the restoration of the total spin. There, we focused on properties of the ground state ($L=0$). Figure 15: (Color online) The calculated spectrum of a two-electron circular parabolic quantum dot, with $R_{W}=200$. The quantum numbers are $(N,M,n,m)$ with $N$ corresponding to the number of radial nodes in the center of mass (CM) wavefunction, and $M$ is the CM azimuthal quantum number. The integers $n$ and $m$ are the corresponding quantum numbers for the electrons’ relative motion (RM) and the total energy is given by $E_{NM,nm}=E^{\rm CM}_{NM}+E^{\rm RM}(n,|m|)$. The spectrum may be summarized by the “spectral rule” given in the figure, with an effective rigid moment of inertia $C=0.037$ (corresponding to an angular momentum ${\cal L}=\hbar m$), the phonon for the stretching vibration $\hbar\omega_{s}=3.50$, and the phonon for the bending vibration coincides with that of the CM motion, i.e., $\hbar\omega_{b}=\hbar\omega_{0}=2$. The quantum numbers $(N_{0},M_{0},n_{0},m)$ specifying each rotational band are given at the bottom, with $m=0,1,2,\ldots$ (the levels $m=0$ and $m=1$ in each band may not be resolved on the scale of the figure). We note that the energy separation between levels in a given rotational band increases as $(2m+1)$ with increasing $m$, which is a behavior characteristic of a rigid rotor. All energies are in units of $\hbar\omega_{0}/2$, where $\omega_{0}$ is the parabolic confinement frequency. In this section, we will further examine the excitation spectra of a two- electron QD by using the rather simple exact solution of this problem provided through separation of the center-of-mass and inter-electron relative-distance degrees of freedom [50]. The spectrum obtained for $R_{W}=200$ (figure 15), exhibits features that are characteristic of a collective rovibrational dynamics, akin to that of a natural “near-rigid” triatomic linear molecule with an infinitely heavy middle particle representing the center of mass of the dot. This spectrum transforms to that of a “floppy” molecule for smaller value of $R_{W}$ (i.e., for stronger confinements characterized by a larger value of $\omega_{0}$, and/or for weaker inter-electron repulsion), ultimately converging to the independent-particle picture associated with the circular central mean-field of the QD. Further evidence for the formation of the electron molecule and the emergence of a rovibrational spectrum was found through examination [50] of the conditional probability distributions for various states $(N,M,n,m)$ (see the caption of figure 15 for the precise meaning of these quantum numbers labeling the spectra). As an example, we display in figure 16 the CPD for the bottom state ($m=0$) of the rotational band $(1,0,1,m)$ (not shown in figure 15); it reveals that this state corresponds to a vibrational motion of the electron molecule both along the interelectron axis (one excited stretching-mode phonon; see figure 15) and perpendicularly to this axis (two excited bending- mode phonons; see figure 15). It is instructive to note here certain similarities between the formation of a “two-electron molecule” in man-made quantum dots, and the collective (rovibrational) features observed in the electronic spectrum of doubly-excited helium atoms [202, 203, 204]. Figure 16: CPD of the excited multi-vibrational state (1,0,1,0) of a $2e$ circular parabolic QD with $R_{W}=200$ (see text for details). The solid dot portrays the position of the reference point ${\bf r}_{0}=(d_{0},0)$, where $d_{0}=2.6$ is half the interelectron distance at the ground state $(0,0,0,0)$. Distances are in units of $l_{0}\sqrt{2}$; the scale of the vertical axis is arbitrary. ### 5.3 Historical significance of the two-electron problem In spite of being the simplest many-body system, the significance of the problem of two interacting electrons confined in an external potential cannot be overstated. Historically it played a central role in the development of the quantum theory of matter through the failure of the Bohr-type semiclassical models to account for the natural He atom. Most recently it has influenced the development of several fields like nonlinear physics, atomic physics, semiconductror quantum dots, and quantum computing. It is instructive to make here a historical detour. Indeed, the failure of Bohr-type semiclassical models, based on the orbiting of spatially correlated (antipodal) electrons in conjunction with the Bohr-Sommerfeld quantization rule, to yield a reasonable estimate of the ground state of the He atom signaled a looming crisis in physics in the 1920’s, which Bohr himself, as well as others, had been keenly aware of, as summarized succintly by Sommerfeld: “All attempts made hitherto to solve the problem of the neutral helium atom have proved to be unsuccessful” [205]; see also the 10th chapter entitled “It was the Spring of hope, it was the Winter of despair” in the book by Pais [206], the review by Van Vleck [207] and the book by Born [208]. While, since, numerical solutions of the two-electron Schrödinger equation provided a quantitative resolution to the problem, the first successful semiclassical treatment of the three-body Coulomb system awaited till 1980 [209, 210]. Furthermore, based on rather general group-theoretical arguments arising from the observation of hierarchies with lower symmetry in the excited spectra, and motivated by ideas originating in nuclear-physics spectroscopy, it has been discovered in the late 1970’s and early 1980’s that electron correlations in doubly excited He lead to quantization of the spectrum much like in a linear triatomic molecule, $e$-He2+-$e$. This molecular picture, with near rigidity and separability, results in “infinite sequences of vibrational levels, on each of which is built an infinite sequence of rotational levels” [202, 211, 212]. The two previous section 5.1 and section 5.2 describing the formation and properties of a 2$e$ Wigner molecule in a single QD may be viewed as the culmination of this historical background. Interestingly, as in the aforementioned semiclassical treatments, the collinear configuration plays a special role in the molecule-like model, serving perhaps as “partial vindication” of the geometry considered originally by Niels Bohr. ## 6 Rotating electron molecules in two-dimensional quantum dots under a strong magnetic field: The case of the lowest Landau level ($\omega_{c}/2\omega_{0}\rightarrow\infty$) ### 6.1 REM analytic trial wave functions In the last ten years, and in particular since 1999 (when it was explicitly demonstrated [20] that Wigner crystallization for small systems is related to symmetry breaking at the unrestricted Hartree-Fock mean-field level), the number of publications addressing the formation and properties of Wigner (or electron) molecules in 2D QDs and quantum dot molecules has grown steadily [20, 21, 23, 24, 27, 50, 101, 102, 128, 132, 167, 171, 175, 188, 189, 195, 213, 214, 215, 216, 217, 218, 219, 220, 221]. A consensus has been reached that rotating electron molecules are formed both in zero [21, 23, 24, 46, 50, 101, 167, 175, 188, 189, 195, 213, 214, 215, 216, 217, 218, 219] and high [26, 27, 46, 102, 128, 131, 132, 171, 220, 221] magnetic fields. At $B=0$, in spite of considerable differences explored in this report (see next paragraph), formation of REMs in quantum dots is driven by the same physical factors as Wigner crystallization in infinite 2D media, i.e., when the strength of the interelectron repulsion relative to the zero-point kinetic energy ($R_{W}$) exceeds a certain critical value, electrons spontaneously crystallize around sites forming geometric molecular structures. At high magnetic fields, the formation of Wigner molecules may be thought of as involving a two-step crystallization process: (I) the localization of electrons results from the shrinkage of the orbitals due to the increasing strength of the magnetic field; (II) then, even a weak interelectron Coulomb repulsion is able to arrange the localized electrons according to geometric molecular structures (thus this process is independent of the value of $R_{W}$). It has been found [24, 26, 27, 128] that the molecular structures at high $B$ coincide with the equilibrium configurations at $B=0$ of $N$ classical point charges [112, 113, 114, 222]. Due to the small finite number, $N$, of electrons, however, there are two crucial differences between the REM and the bulk Wigner crystal. Namely, (I) the crystalline structure is that of the equilibrium 2D configuration of $N$ classical point charges, and thus consists of nested polygonal rings, in contrast to the well known hexagonal bulk crystal; (II) in analogy with the case of 3D natural molecules, the Wigner molecules rotate as a whole (collective rotations); they behave, however, as highly floppy (non-rigid) rotors. A most striking observation concerning the REMs is that their formation and properties have been established with the help of traditional ab initio many- body methods, i.e., exact diagonalization, [50, 128, 131, 167, 171, 213, 221] quantum Monte Carlo [78, 188, 195, 196, 216], and the systematic controlled hierarchy [20, 22, 23, 24, 25, 27, 102, 217] of approximations involving the UHF and subsequent post-Hartree-Fock methods. This contrasts with the case of the Jastrow/Laughlin [55] and composite-fermion [56, 57] wave functions, which were constructed through “intuition-based guesswork.” In spite of its appearance in the middle nineties and its firm foundation in many-body theory, however, the REM picture had not, until recently, successfully competed with the CF/JL picture; indeed many research papers [181, 223, 224, 225, 226, 227, 228] and books [133] describe the physics of quantum dots in high magnetic fields following exclusively notions based on CF/JL functions, as expounded in 1983 (see Ref. [55]) and developed in detail in 1995 in Ref. [229] and Ref. [174]. One of the main obstacles for more frequent use of the REM picture had been the lack of analytic correlated wave functions associated with this picture. This situation, however, changed with the recent explicit derivation of such REM wave functions [24]. The approach used in Ref. [24] for constructing the analytic REM functions in high $B$ consists of two-steps: First the breaking of the rotational symmetry at the level of the single-determinantal unrestricted Hartree-Fock approximation yields states representing electron molecules. Subsequently the rotation of the electron molecule is described through restoration of the circular symmetry via post Hartree-Fock methods, and in particular Projection Techniques [18]. The restoration of symmetry goes beyond the single determinantal mean-field description and yields multi-determinantal wave functions. In the zero and low-field cases, the broken symmetry UHF orbitals need to be determined numerically, and, in addition, the restoration of the total-spin symmetry needs to be considered for unpolarized and partially polarized cases. The formalism and mathematical details of this procedure at $B=0$ have been elaborated in previous sections. In the case of high magnetic fields, the spins of the electrons are fully polarized. Furthermore, one can specifically consider the limit when the confining potential can be neglected compared to the confinement induced by the magnetic field, so that the Hilbert space is restricted to the lowest Landau level. Then, assuming a symmetric gauge, the UHF orbitals can be represented [24, 230] by displaced Gaussian analytic functions, centered at different positions $Z_{j}\equiv X_{j}+\rmi Y_{j}$ according to the equilibrium configuration of $N$ classical point charges [112, 113, 114, 222] arranged at the vertices of nested regular polygons (each Gaussian representing a localized electron). Such displaced Gaussians in the lowest Landau level are written as $u(z,Z_{j})=(1/\sqrt{\pi})\exp[-|z-Z_{j}|^{2}/2]\exp[-\rmi(xY_{j}-yX_{j})],$ (6.1) where the phase factor is due to the gauge invariance. $z\equiv x+\rmi y$, and all lengths are in dimensionless units of ${l_{B}}\sqrt{2}$ with the magnetic length being $l_{B}=\sqrt{\hbar c/eB}$. Note that expression (6.1) is a special case of the more general expression (2.24) for a displaced Gaussian which corresponds to situations with smaller magnetic fields when the restriction to the lowest Landau level breaks down. The notation $z\equiv x+\rmi y$ is associated with positive angular momenta for the single-particle states in the lowest Landau level. Ref. [24] used $z\equiv x-\rmi y$ and negative single-particle angular momenta in the lowest Landau level. The final expressions for the trial wave functions do not depend on these choices. Table 1: Comparison of yrast-band energies obtained from REM and EXD calculations for $N=6$ electrons in the lowest Landau level, that is in the limit $B\rightarrow\infty$. In this limit the external confinement can be neglected and only the interaction energy contributes to the yrast-band energies. Energies in units of $e^{2}/(\kappa l_{B})$. For the REM results, the (1,5) polygonal-ring arrangement was considered. The values of the fractional filling may be obtained for each $L$ as $\nu=N(N-1)/(2L)$. * $L$ | REM | EXD | Error (%) | $L$ | REM | EXD | Error (%) ---|---|---|---|---|---|---|--- 70 | 2.3019 | 2.2824 | 0.85 | 140 | 1.6059 | 1.6006 | 0.33 75 | 2.2207 | 2.2018 | 0.85 | 145 | 1.5773 | 1.5724 | 0.31 80 | 2.1455 | 2.1304 | 0.71 | 150 | 1.5502 | 1.5455 | 0.30 85 | 2.0785 | 2.0651 | 0.65 | 155 | 1.5244 | 1.5200 | 0.29 90 | 2.0174 | 2.0054 | 0.60 | 160 | 1.4999 | 1.4957 | 0.28 95 | 1.9614 | 1.9506 | 0.55 | 165 | 1.4765 | 1.4726 | 0.27 100 | 1.9098 | 1.9001 | 0.51 | 170 | 1.4542 | 1.4505 | 0.26 105 | 1.8622 | 1.8533 | 0.48 | 175 | 1.4329 | 1.4293 | 0.25 110 | 1.8179 | 1.8098 | 0.45 | 180 | 1.4125 | 1.4091 | 0.24 115 | 1.7767 | 1.7692 | 0.42 | 185 | 1.3929 | 1.3897 | 0.23 120 | 1.7382 | 1.7312 | 0.40 | 190 | 1.3741 | 1.3710 | 0.23 125 | 1.7020 | 1.6956 | 0.38 | 195 | 1.3561 | 1.3531 | 0.22 130 | 1.6681 | 1.6621 | 0.36 | 200 | 1.3388 | 1.3359 | 0.21 135 | 1.6361 | 1.6305 | 0.34 | | | | Ref. [24] used these analytic orbitals to first construct the broken symmetry UHF determinant, $\Psi^{\rm{UHF}}_{N}$, and then proceeded to derive analytic expressions for the many-body REM wave functions by applying onto $\Psi^{\rm{UHF}}_{N}$ an appropriate projection operator ${\cal P}_{L}$ (see section 2.2.1) that restores the circular symmetry and generates correlated wave functions with good total angular momentum $L$. These REM wave functions can be easily written down [24] in second-quantized form for any classical polygonal ring arrangement $(n_{1},n_{2},\ldots,n_{r})$ by following certain simple rules for determining the coefficients of the determinants $D(l_{1},l_{2},\ldots,l_{N})\equiv{\rm{det}}[z_{1}^{l_{1}},z_{2}^{l_{2}},\cdot\cdot\cdot,z_{N}^{l_{N}}]$, where the $l_{j}$’s denote the angular momenta of the individual electrons. The REM functions associated with the $(0,N)$ and $(1,N-1)$ ring arrangements, respectively [here $(0,N)$ denotes a regular polygon with $N$ vertices, such as an equilateral triangle or a regular hexagon, and $(1,N-1)$ is a regular polygon with $N-1$ vertices and one occupied site in its center], are given by $\displaystyle\Phi_{L}^{(0,N)}(z_{1},z_{2},\ldots,z_{N})$ $\displaystyle=\sum^{l_{1}+\cdot\cdot\cdot+l_{N}=L}_{0\leq l_{1}<l_{2}<\cdot\cdot\cdot<l_{N}}\left(\prod_{i=1}^{N}l_{i}!\right)^{-1}\left(\prod_{1\leq i<j\leq N}\sin\left[\frac{\pi}{N}(l_{i}-l_{j})\right]\right)$ (6.2) $\displaystyle\times\;D(l_{1},l_{2},\ldots,l_{N})\exp(-\sum_{i=1}^{N}z_{i}z_{i}^{*}/2),$ with $L=L_{0}+Nm,\;\;m=0,1,2,3,\ldots,$ (6.3) and $\displaystyle\Phi_{L}^{(1,N-1)}(z_{1},z_{2},\ldots,z_{N})$ $\displaystyle=\sum^{l_{2}+\cdot\cdot\cdot+l_{N}=L}_{1\leq l_{2}<l_{3}<\cdot\cdot\cdot<l_{N}}\left(\prod_{i=2}^{N}l_{i}!\right)^{-1}\left(\prod_{2\leq i<j\leq N}\sin\left[\frac{\pi}{N-1}(l_{i}-l_{j})\right]\right)$ (6.4) $\displaystyle\times\;D(0,l_{2},\ldots,l_{N})\exp(-\sum_{i=1}^{N}z_{i}z_{i}^{*}/2),$ with $L=L_{0}+(N-1)m,\;\;m=0,1,2,3,\ldots,$ (6.5) where $L_{0}=N(N-1)/2$ is the minimum allowed total angular momentum for $N$ (fully spin polarized) electrons in high magnetic fields. Notice that the REM wave functions [Equation (6.2) and Equation (6.4)] vanish identically for values of the total angular momenta outside the specific values given by the sequences (6.3) and (6.5), respectively; these sequences are termed as magic angular momentum sequences. Figure 17: Total interaction energy from exact-diagonalization calculations as a function of the total angular momentum $(19\leq L\leq 140)$ for $N=6$ electrons in the lowest Landau level. The upwards pointing arrows indicate the magic angular momenta corresponding to the classically most stable (1,5) polygonal ring arrangement of the Wigner molecule. The short downwards pointing arrows indicate successful predictions of the composite-fermion model. The medium-size downwards pointing arrows indicate predictions of the composite-fermion model that fail to materialize as magic angular momenta. The long downward arrows indicate EXD magic angular momenta not predicted by the composite-fermion model. Energies in units of $e^{2}/\kappa l_{B}$, where $\kappa$ is the dielectric constant. We remark that, while the original REM analytic wave function was derived in the context of a high magnetic field (that is in the fractional quantum Hall effect regime), it is valid for any circumstance where the spectrum consists of a degenerate manifold of LLL-like states (even with no magnetic field present). Indeed a wave function having the form of the REM wave function discussed by us above has been employed recently for graphene quantum dots with a zig-zag boundary condition and in the absence of a magnetic field [231]. In the remaining of this section, we continue discussing the properties of analytic REM wave functions associated with fully spin polarized electrons. However, we mention here that, following the methodology of Ref. [24] for fully spin polarized REMs. Dai et al[232] and Shi et al[233] have most recently presented analytic trial wave functions for rotating electron molecules with partial spin polarizations. ### 6.2 Yrast rotational band in the lowest Landau level As an accuracy test, we compare in table 1 REM and exact-diagonalization results for the interaction energies of the yrast band associated with the magic angular momenta $L_{m}$ [see (2.26)] of $N=6$ electrons in the lowest Landau level. An yrast888The word yrast is the superlative of the Swedish yr, which means dizzy [11]. The term yrast is widely used in nuclear spectroscopy. state is defined as the lowest-energy state for a given angular momentum $L$. As a result, the yrast band represents excitations with purely rotational motion; no other excitations, like center-of-mass or vibrational modes, are present. As seen from table 1, the REM wave functions offer an excellent approximation to the EXD ones, since the relative error of the REM energies is smaller than 0.3%, and it decreases steadily for larger $L$ values. Of course, a small difference in the energies between approximate and exact-diagonalization results is only one of several tests for deciding whether a given trial wave function is a good approximation. As will be discussed below, comparison of conditional probability distrubutions is an equally (if not more) important test. ### 6.3 Inconsistencies of the composite-fermion view for semiconductor quantum dots. Before the development of the REM approach, electrons in the lowest Landau level in two-dimensional quantum dots were thought as being well approximated by composite fermion trial wave functions. However, results obtained with the REM and exact-diagonalization calculations led researchers to examine inconsistencies and discrepancies of the CF approach in the context of quantum dots. This section focuses on these issues. For $N=6$, figure 17 displays (in four frames) the total interaction energy from exact-diagonalization as a function of the total angular momentum $L$ in the range $19\leq L\leq 140$. [The total kinetic energy in the Hamiltonian (4.1), being a constant, can be disregarded.] One can immediately observe the appearance of downward cusps, implying states of enhanced stability, at certain magic angular momenta. For the CF theory, the magic angular momenta can be determined by $L=L^{*}+mN(N-1)=L^{*}+2mL_{0}.$ (6.6) Namely, for $N=6$, if one knows the non-interacting $L^{*}$’s, the CF magic $L$’s in any filling-factor interval $1/(2m-1)\geq\nu\geq 1/(2m+1)$ [corresponding to the angular-momentum interval $15(2m-1)\leq L\leq 15(2m+1)]$, $m=1,2,3,4,\ldots$, can be found by adding $2mL_{0}=30m$ units of angular momentum to each of the $L^{*}$’s. To obtain the non-interacting $L^{*}$’s, one needs first to construct [26, 131, 229] the compact Slater determinants. Let $N_{n}$ denote the number of electrons in the $n$th Landau level with $\sum_{n=0}^{t}N_{n}=N$; $t$ is the index of the highest occupied Landau level and all the lower Landau levels with $n\leq t$ are assumed to be occupied. The compact determinants are defined as those in which the $N_{n}$ electrons occupy contiguously the single-particle orbitals (of each $n$th Landau level) having the lowest angular momenta $l=-n,-n+1,\ldots,-n+N_{n}-1$. The compact Slater determinants are usually denoted as $[N_{0},N_{1},\ldots,N_{t}]$; see Refs. [25, 229] for details. The compact determinants and the corresponding non-interacting $L^{*}$’s for $n=6$ are listed in Table 2. Table 2: Compact Slater determinants and associated angular momenta $L^{*}$ for $N=6$ electrons according to the CF presciption. Both $L^{*}=-3$ and $L^{*}=3$ are associated with two compact states each, the one with lowest energy being the preferred one. * Compact state | $L^{*}$ ---|--- $[$1,1,1,1,1,1$]$ | $-$15 $[$2,1,1,1,1$]$ | $-$9 $[$2,2,1,1$]$ | $-$5 $[$3,1,1,1$]$ | $-$3 $[$2,2,2$]$ | $-$3 $[$3,2,1$]$ | 0 $[$4,1,1$]$ | 3 $[$3,3$]$ | 3 $[$4,2$]$ | 5 $[$5,1$]$ | 9 $[$6$]$ | 15 There are nine different values of $L^{*}$’s, and thus the CF theory for $N=6$ predicts that there are always nine magic numbers in any interval $15(2m-1)\leq L\leq 15(2m+1)$ between two consecutive angular momenta of Jastrow/Laughlin states, $15(2m-1)$ and $15(2m+1)$, $m=1,2,3,\ldots$ (henceforth we will denote this interval as ${\cal I}_{m}$). For example, using Table 2 and (6.6), the CF magic numbers in the interval $15\leq L\leq 45$ ($m=1$) are found to be the following nine $15,\;21,\;25,\;27,\;30,\;33,\;35,\;39,\;45.$ (6.7) On the other hand, in the interval $105\leq L\leq 135$ ($m=4$), the CF theory predicts the following set of nine magic numbers, $105,\;111,\;115,\;117,\;120,\;123,\;125,\;129,\;135.$ (6.8) An inspection of the total-energy-vs-$L$ plots in figure 17 reveals that the CF prediction misses the actual magic angular momenta specified by the exact- diagonalization calculations as those associated with the downward cusps. It is apparent that the number of downward cusps in any interval ${\cal I}_{m}$ is always different from 9. Indeed, there are 10 cusps in ${\cal I}_{1}$ [including that at $L=15$, not shown in figure 17(a)], 10 in ${\cal I}_{2}$ [see figure 17(b)], 7 in ${\cal I}_{3}$ [see figure 17(c)], and 7 in ${\cal I}_{4}$ [see figure 17(d)]. In detail, the CF theory fails in the following two aspects: (I) There are exact magic numbers that are consistently missing from the CF prediction in every interval; with the exception of the lowest $L=20$, these exact magic numbers (marked by a long downward arrow in the figures) are given by $L=10(3m-1)$ and $L=10(3m+1)$, $m=1,2,3,4,\ldots$; (II) There are CF magic numbers that do not correspond to downward cusps in the EXD calculations (marked by medium-size downward arrows in the figures). This happens because cusps associated with $L$’s whose difference from $L_{0}$ is divisible by 6 (but not simultaneously by 5) progressively weaken and completely disappear in the intervals ${\cal I}_{m}$ with $m\geq 3$; only cusps with the difference $L-L_{0}$ divisible by 5 survive. On the other hand, the CF model predicts the appearance of four magic numbers with $L-L_{0}$ divisible solely by 6 in every interval ${\cal I}_{m}$, at $L=30m\mp 9$ and $30m\mp 3$, $m=1,2,3,\ldots$ The overall extent of the inadequacy of the CF model can be appreciated better by the fact that there are six false predictions (long and medium-size downward arrows) in every interval ${\cal I}_{m}$ with $m\geq 3$, compared to only five correct ones [small downward arrows, see figure 17(c) and figure 17(d)]. In contrast to the CF model, the magic angular momenta in the REM theory are associated with the polygonal ring configurations of $N$ classical point charges. This is due to the fact that the enhanced stability of the downward cusps results from the coherent collective rotation of the regular-polygon REM structures. Due to symmetry requirements, such collective rotation can take place only at magic-angular-momenta values. The in-between angular momenta require the excitation of additional degrees of freedom (like the center of mass and/or vibrational modes), which raises the total energy with respect to the values associated with the magic angular momenta. For $N=6$, the ring configuration of lowest energy is the (1,5), while there exists a (0,6) isomer [114, 222] with higher energy. As a result, our exact- diagonalization calculations [26] (as well as earlier ones [131, 132, 171] for lower angular momenta $L\leq 70$) have found that there exist two sequences of magic angular momenta, a primary one $(S_{p})$ with $L=15+5m$ [see Equation (6.5)], associated with the most stable (1,5) classical molecular configuration, and a secondary one $(S_{s})$ with $L=15+6m$ [see Equation (6.3)], associated with the metastable $(0,6)$ ring arrangement. Furthermore, our calculations (see also Refs. [132, 171]) show that the secondary sequence $S_{s}$ contributes only in a narrow range of the lowest angular momenta; in the region of higher angular momenta, the primary sequence $S_{p}$ is the only one that survives and the magic numbers exhibit a period of five units of angular momentum. It is interesting to note that the initial competition between the primary and secondary sequences, and the subsequent prevalence of the primary one, has been seen in other sizes as well [171] i.e., $N=5,7,8$. Furthermore, this competition is reflected in the field-induced molecular phase transitions associated with broken symmetry UHF solutions in a parabolic QD. Indeed, Ref. [53] demonstrated recently that, as a function of increasing $B$, the UHF solutions for $N=6$ first depict the transformation of the maximum density droplet [126] (see definition in section 2.2.1) into the (0,6) molecular configuration; then (at higher $B$) the (1,5) configuration replaces the (0,6) structure as the one having the lower HF energy. The extensive comparisons in this section lead to the conclusion that the composite-fermion model does not explain the systematic trends exhibited by the magic angular momenta in 2D quantum dots in high magnetic fields. These trends, however, were shown to be a natural consequence of the formation of REMs and their metastable isomers. These results motivated a reexamination of the original composite-fermion approach (the mean-field CF) and led to a reassessment of the significance of the residual interaction, neglected in the mean-field CF theory. Initially, it has been reported that some CF functions away from the main fractions [e.g., for $N=19$ and $L=1845$ and $N=19$ and $L=3555$] may reproduce the aforementioned crystalline patterns [234]. Subsequently, Jain and coworkers have found that inclusion of the residual interaction is absolutely necessary to account for the full range of inconsistencies of the mean-field CF theory [176]. However, this latter development was achieved with the trade off of abandoning the fundamental nature of the composite fermion as an elementary, independent and weakly interacting quasi-particle. Indeed, the revised [176] CF picture amounts to an exact diagonalization method which uses a correlated basis set (made out of CF wave functions). Another attempt to update the CF theory in order to account for crystallization consists of combining the REM analytic wave function $\Phi^{\rm REM}_{L}(z_{1},z_{2},\ldots,z_{N})$ (see section 6.1) with Jastrow prefactors [235], namely one uses a variational wave function of the form $\Psi^{{2p,}{\rm CFC}}_{L}=\prod_{i<j}(z_{i}-z_{j})^{2p}\Phi^{\rm REM}_{L^{*}},$ (6.9) with $L=L^{*}+pN(N-1)$ and $p$ serves as an additional variational parameter. Obviously, the crystalline patterns in such an approach originate from the REM wave function and the Jastrow prefactors simply increase the variational freedom, leading to a numerical improvement. Although this approach is a straightforward variational improvement of the analytic REM method [24], it is being referred to [233, 235] as a composite-fermion crystal (CFC). More direct variational improvements of the analytic REM wave functions can be devised in the spirit of the variation-after-projection method. For example, one can use angular-momentum conserving variational parameters in front of the sine coefficients in the REM expansion [231]. Figure 18: (Color online) Conditional probability distributions at high $B$ for $N=6$ electrons and $L=75$ ($\nu=1/5$, left column) and for $N=7$ electrons and $L=105$ (again $\nu=1/5$, right column). Top row: REM case. Middle row: The case of exact diagonalization. Bottom row: The Jastrow/Laughlin case. The exact diagonalization and REM wave functions have a pronouned crystalline character, corresponding to the (1,5) polygonal configuration of the REM for $N=6$, and to the (1,6) polygonal configuration for $N=7$. In contrast, the Jastrow/Laughlin wave functions exhibit a characteristic liquid profile that depends smoothly on the number $N$ of electrons. The observation point (identified by a solid dot) is located at $r_{0}=5.431l_{B}$ for $N=6$ and $L=75$ and $r_{0}=5.883l_{B}$ for $N=7$ and $L=105$. The EXD Coulomb interaction energies (lowest Landau level) are 2.2018 and 2.9144 $e^{2}/\kappa l_{B}$ for $N=6,L=75$ and $N=7,L=105$, respectively. The errors relative to the corresponding exact-diagonalization energies and the overlaps of the trial functions with the EXD ones are: (I) For $N=6,L=75$, REM: 0.85%, 0.817; JL: 0.32%, 0.837. (II) For $N=7,L=105$, REM: 0.59%, 0.842; JL: 0.55%, 0.754. Figure 19: (Color online) CPDs at high $B$ for $N=7$ and $L=63$ ($\nu=1/3$). Top: REM case; Middle: EXD case; Bottom: JL case. Unlike the JL CPD (which is liquid), the CPDs for the exact-diagonalization and REM wave functions exhibit a well developed crystalline character [corresponding to the (1,6) polygonal configuration of the REM for $N=7$ electrons]. The observation point (identified by a solid dot) is located at $r_{0}=4.568l_{B}$. ### 6.4 REM versus Laughlin wave functions: Conditional probability distributions and multiplicity of zeroes Recent extensive numerical calculations [24, 52] have revealed major disagreements between the intrinsic structure of the Jastrow/Laughlin trial wave functions [55] for the main fractions $\nu=1/(2m+1)$ and that of the exact-diagonalization and REM wave functions. Indeed, it was found that both EXD and REM wave functions exhibit crystalline correlations, while the Jastrow/Laughlin ones are liquid-like as originally described in Ref. [55]. To illustrate the differences between the intrinsic structure of the REM and EXD states in the lowest Landau level versus the familiar Jastrow/Laughlin ones, we display in figure 18 the CPDs for cusp states corresponding to a low filling factor $\nu=1/5$ and for two different sizes, i.e., for $N=6$ electrons ($L=75$, left column) and $N=7$ electrons ($L=105$, right column). In figure 18, the top row depicts the REM case; the EXD case is given by the middle row, while the CF case [which reduces to the JL wave functions for fractions $1/(2p+1)$] are given by the bottom row. There are three principal conclusions that can be drawn from an inspection of figure 18 (and the many other cases studied in Ref. [26]). (I) The character of the exact-diagonalization states is unmistakably crystalline with the EXD CPDs exhibiting a well developed molecular polygonal configuration [(1,5) for $N=6$ and (1,6) for $N=7$, with one electron at the center], in agreement with the explicitly crystalline REM case. (II) For all the examined instances covering the low fractional fillings 1/9, 1/7, and 1/5, the Jastrow/Laughlin wave functions fail to capture the intrinsic crystallinity of the exact-diagonalization states. In contrast, they represent “liquid” states in agreement with an analysis that goes back to the original papers [55, 236] by Laughlin. In particular, Ref. [236] investigated the character of the JL states through the use of a pair correlation function [usually denoted by $g(R)$] that determines the probability of finding another electron at the absolute relative distance $R=|{\bf r}-{\bf r}_{0}|$ from the observation point ${\bf r}_{0}$. Our anisotropic CPD of Equation (1.1) is of course more general (and more difficult to calculate) than the $g(R)$ function of Ref. [236]. However, both our $P({\bf r},{\bf r}_{0})$ (for $N=6$ and $N=7$ electrons) and the $g(R)$ (for $N=1000$ electrons, and for $\nu=1/3$ and $\nu=1/5$) in Ref. [236] reveal a similar characteristic liquid-like and short-range-order behavior for the JL states, eloquently described in Ref. [236] (see p. 249 and p. 251). Indeed, we remark that only the first-neighbor electrons on the outer rings can be distinguished as separate localized electrons in our CPD plots of the JL functions [see figure 18]. (III) For a finite number of electrons, pronounced crystallinity of the exact- diagonalization states occurs already at the $\nu=1/5$ value (see figure 18). This finding is particularly interesting in light of expectations [234, 237] (based on comparisons [55, 236, 238] between the JL states and the static bulk Wigner crystal) that a liquid-to-crystal phase transition may take place only at lower fillings with $\nu\leq 1/7$. Of interest also is the case of $\nu=1/3$. Indeed, for this fractional filling, the liquid JL function is expected to provide the best approximation, due to very high overlaps (better than 0.99) with the exact wave function [58, 59, 161]. In figure 19, we display the CPDs for $N=7$ and $L=63$ ($\nu=1/3$), and for the three cases of REM, EXD, and JL wave functions. Again, even in this most favorable case, the CPD of the JL function disagrees with the EXD one, which exhibits clearly a (1,6) crystalline configuration in agreement with the REM CPD. Similar crystalline correlations at higher fractions were also found for quantum dots of larger sizes, e.g., $N=8$, and $N=9$ electrons. As illustrative examples for these additional sizes [see also the EXD CPD for $N=12$ electrons in figure 24 below (in section 7.3)], we displayed in figure 5 of Ref. [26] the CPDs for $N=8$ and $L=91$ ($1/5<\nu=4/13<1/3$) and for $N=9$ and $L=101$ ( $1/3<\nu=36/101<1$). Again, the CPDs (both for the REM and the EXD wave functions) exhibit a well developed crystalline character in accordance with the (1,7) and (2,7) polygonal configurations of the REM, appropriate for $N=8$ and $N=9$ electrons, respectively. Another area of disagreement between REM and Laughlin wave functions concerns the properties of the zero points. In this respect, we recall that the Jastrow/Laughlin trial functions for $N$ electrons have the form $\Phi^{\rm JL}(z_{1},z_{2},\ldots,z_{N})=\left(\prod_{1\leq i<j\leq N}(z_{i}-z_{j})^{2m+1}\right)\exp(-\sum_{i=1}^{N}z_{i}z_{i}^{*}/2).$ (6.10) Due to the Jastrow factors $(z_{i}-z_{j})^{2m+1}$, it is apparent that the Laughlin expressions (6.10) (as a function of a given $z_{i}$) have $N-1$ zero points, each of order $2m+1$, which are bound to the positions of the remaining $N-1$ electrons. In contrast, as discussed in Ref. [24], the analytic REM wave functions do not have zeroes with order higher than unity. In particular, only $N-1$ of the REM zeroes are bound to the positions of the remaining electrons, while the rest of them are free. Recently, it has been shown through extensive numerical studies [239] that the properties of REM zeroes are in agreement with the behavior of the zeroes in exact- diagonalization wave functions; this is another indication of the superiority of the REM picture compared to the Laughlin theory. Before exiting this discussion, we remark about discrepancies of the Laughlin quasihole theory in the context of quantum dots. In particular, we recall that the Laughlin quasihole, with $N$ additional units of angular momentum, has been conjectured to be the first excited state. However, LLL exact- diagonalization calculations for $N$ electrons in a quantum dot have revealed that this is not the case. Instead, the first excited state corresponds to an increment in the total angular momentum which varies as the number of electrons localized on one of the rings of the rotating electron molecule, usually the outermost one; see figure 26 in section 7.4 below. ## 7 Rotating electron molecules in two-dimensional quantum dots under a strong, but finite external magnetic field ($\omega_{c}/2\omega_{0}>1$) Figure 20: (Color online) Two-step-method versus exact-diagonalization calculations: Ground-state energies for $N=4$ electrons (referenced to $4\hbar\tilde{\omega}$) as a function of the magnetic field $B$. Thick dashed line (red): broken-symmetry UHF (static electron molecule). Solid line (green): EXD (from Ref. [171]). Thick dashed-dotted line (blue): REM. Thin dashed line (violet, marked LLL): the commonly used approximate energies $\tilde{E}^{\rm{EXD}}_{\rm{LLL}}(B)$ (see text for details). Thin dotted line (black): $\tilde{E}^{\rm{REM}}_{\rm{LLL}}(B)$ (see text). For $B<8$ T, the $\tilde{E}^{\rm{EXD}}_{\rm{LLL}}(B)$ and $\tilde{E}^{\rm{REM}}_{\rm{LLL}}(B)$ curves coincide; we have checked that these curves approach each other also at larger values of $B$, outside the plotted range. Numbers near the bottom curves denote the value of magic angular momenta [$L_{m}$, see (2.26)] of the ground state. Corresponding fractional filling factors are specified by $\nu=N(N-1)/(2L_{m})$. Parameters used: confinement $\hbar\omega_{0}=3.60$ meV, dielectric constant $\kappa=13.1$, effective mass $m^{*}=0.067m_{e}$. ### 7.1 Ground-state energies in medium and high magnetic field The general form (2.24) for the displaced Gaussian orbitals [in conjunction with the projected REM wave function (2.25)] enables us to calculate REM ground-state energies for moderately-high $B$, when corrections arising from higher Landau levels must be taken into consideration. Unlike the lowest- Landau-level case, where the azimuthal integration can be carried out analytically, the energies (2.28) (and corresponding CPDs) associated with the general REM wave function (2.25) require numerical integration over the azimuthal angles $\gamma_{q}$. Before proceeding with the presentation of results for $N>10$, we demonstrate the accuracy of the two-step method embodied in Equation (2.25) through comparisons with existing exact-diagonalization results for smaller sizes. In figure 20, our REM calculations for the ground-state energies as a function of $B$ are compared to EXD calculations [171] for $N=4$ electrons in an external parabolic confinement. The thick dotted line (red) represents the broken- symmetry UHF approximation (first step of our method), which naturally is a smooth curve lying above the EXD one [solid line (green)]. The results obtained after restoration of symmetry [dashed-dotted line (blue); marked as REM] agree very well with the EXD one in the whole range 2 T$<B<$ 15 T. We recall here that, for the parameters of the quantum dot, the electrons form in the intrinsic frame of reference a square about the origin of the dot, i.e., a (0,4) configuration, with the zero indicating that no electron is located at the center. According to (2.26), $L_{0}=6$, and the magic angular momenta are given by $L_{m}=6+4k$, $k=0$, 1, 2, $\ldots$ Note that the REM energies are slightly lower than the EXD ones in several subranges. According to the Rayleigh-Ritz variational theorem, this indicates that the hyperspherical- harmonics calculation (equivalent to an exact-diagonalization approach) of [171] did not converge fully in these subranges. To further evaluate the accuracy of the two-step method, we also display in figure 20 [thin dashed line (violet)] ground-state energies $\tilde{E}^{\rm{EXD}}_{\rm LLL}(B)$ calculated with the commonly used approximate LLL Hamiltonian [128, 229, 237, 240] $\tilde{\cal H}_{\rm LLL}=N\hbar\tilde{\omega}+\hbar(\tilde{\omega}-\frac{\omega_{c}}{2})L+\sum_{j>i=1}^{N}\frac{e^{2}}{\kappa r_{ij}},$ (7.1) where $\tilde{\omega}=\sqrt{\omega_{0}^{2}+\omega_{c}^{2}/4}$. The LLL Hamiltonian $\tilde{\cal H}_{\rm LLL}$ reduces to the previously introduced Hamiltonian ${\cal H}_{\rm LLL}$ [see Equation (4.1)] in the limit $B\rightarrow\infty$. Both Hamiltonians restrict the many-body wave functions within the lowest Landau level, and they both accept the same set of eigenstates as solutions. Indeed the term $\hbar(\tilde{\omega}-\omega_{c}/2)L$ is proportional to the total angular momentum, and thus its presence influences only the eigenvalues, but not the composition of the eigenstates. $\tilde{\cal H}_{\rm LLL}$ corresponds to a situation where the external harmonic confinement is added to ${\cal H}_{\rm LLL}$ as a perturbation (see section II.B in Ref. [53]). As a result, (i) the degeneracy of the single-particle levels in the lowest Landau level is lifted and (ii) there is an eigenstate with minimum energy (the ground state) at each value of $B$ (expressed through the cyclotron frequency $\omega_{c}$). Naturally, the LLL levels used in the exact diagonalization of $\tilde{\cal H}_{\rm LLL}$ are given by expression (4.2), but with $\Lambda=\tilde{l}=\sqrt{\hbar/(m^{*}\tilde{\omega})}$. We find that the energies $\tilde{E}^{\rm EXD}_{\rm LLL}(B)$ tend to substantially overestimate the REM (and EXD) energies for lower values of $B$ (e.g., by as much as 5.5% at $B\sim 4$ T). On the other hand, for higher values of $B$ ($>$ 12 T), the energies $\tilde{E}^{\rm{EXD}}_{\rm{LLL}}(B)$ tend to agree rather well with the REM ones. We stress that the results labelled simply as EXD correspond to exact diagonalizations without any restrictions on the Hilbert space, i.e., the full Darwin-Fock single-particle spectrum is considered at a given $B$. A behavior similar to $\tilde{E}^{\rm EXD}_{\rm LLL}(B)$ is exhibited also by the $\tilde{E}^{\rm{REM}}_{\rm{LLL}}(B)$ ground-state energies [which are calculated using the Hamiltonian (7.1) and the LLL analytic REM wave functions in section 6.1 with lengths in units of $\sqrt{\hbar/(m^{*}\tilde{\omega})}$ instead of $l_{B}\sqrt{2}$; dotted line (black)]. A similar agreement between REM and EXD results, and a similar inaccurate behavior of the LLL approximate Hamiltonian (7.1) was found by us also for $N=3$ electrons in the range 2 T $<B<$ 16 T shown in figure 2 of Ref. [53] (the exact-diagonalization calculation in this figure was taken from Ref. [166]). In all cases, the total energy of the REM is lower than that of the UHF Slater determinant (see, e.g., figure 20). Indeed, a theorem discussed in section 3 of Ref. [241], pertaining to the energies of projected wave functions, guarantees that this lowering of energy applies for all values of $N$ and $B$. ### 7.2 The case of $N=11$ electrons. Figure 21: (Color online) Ground-state energies for $N=11$ electrons (per particle, referenced to $\hbar\tilde{\omega}$) as a function of the magnetic field $B$. Dashed line (red): UHF (static electron molecule). Solid line (blue): REM. Dotted line (black): Approximate energies $\tilde{E}^{\rm{REM}}_{\rm{LLL}}(B)$ (see text). Parameters used: confinement $\hbar\omega_{0}=3.60$ meV, dielectric constant $\kappa=13.1$, effective mass $m^{*}=0.067m_{e}$. The inset shows a magnification of the REM curve in the range 5 T $<B<$ 12 T. Table 3: Ground-state magic angular momenta and their decomposition $\\{k_{1},k_{2}\\}$ for $N=11$ in the nagnetic-field range 5 T $\leq B\leq$ 25 T. The results correspond to the REM (see lower curve in figure 21). The parameters used are as in figure 21. * $L_{m}$ | $k_{1}$ | $k_{2}$ | $L_{m}$ | $k_{1}$ | $k_{2}$ ---|---|---|---|---|--- 55 | 0 | 0 | 165 | 2 | 13 63 | 0 | 1 | 173 | 2 | 14 71 | 0 | 2 | 181 | 2 | 15 79 | 0 | 3 | 189 | 2 | 16 90 | 1 | 4 | 197 | 2 | 17 98 | 1 | 5 | 205 | 2 | 18 106 | 1 | 6 | 213 | 2 | 19 114 | 1 | 7 | 224 | 3 | 20 122 | 1 | 8 | 232 | 3 | 21 130 | 1 | 9 | 240 | 3 | 22 138 | 1 | 10 | 248 | 3 | 23 146 | 1 | 11 | 256 | 3 | 24 154 | 1 | 12 | | | Figure 21 presents the case for the ground-state energies of a quantum dot with $N=11$ electrons, which have a nontrivial double-ring configuration $(n_{1},n_{2})$. The most stable [114] classical configuration is $(3,8)$, for which we have carried UHF (static electron molecule) and REM (projected) calculations in the magnetic field range 5 T $<B<$ 25 T. Figure 21 also displays the LLL ground-state energies $\tilde{E}^{\rm{REM}}_{\rm{LLL}}(B)$ [dotted curve (black)], which, as in previous cases, overestimate the ground- state energies for smaller $B$. The approximation $\tilde{E}^{\rm{REM}}_{\rm{LLL}}(B)$, however, can be used to calculate ground-state energies for higher values of $B$. In keeping with the findings for smaller sizes [51] [with $(0,N)$ or $(1,N-1)$ configurations], we found that both the UHF and the REM ground-state energies approach, as $B\rightarrow\infty$, the classical equilibrium energy of the (3,8) polygonal configuration [i.e., 19.94 meV; 4.865$E_{0}$ in the units of Ref. [114], $E_{0}\equiv(m^{*}\omega_{0}^{2}e^{4}/2\kappa^{2})^{1/3}$]. In analogy with smaller sizes (see, e.g., figure 20 and Ref. [53]), the REM ground-state energies in figure 21 exhibit oscillations as a function of $B$ (see in particular the inset). These oscillations are associated with magic angular momenta, specified by the number of electrons on each ring. For $N=11$ they are given by (2.26), i.e., $L_{m}=55+3k_{1}+8k_{2}$, with the $k_{q}$’s being nonnegative integers. As was the case with $N=9$ electrons [53], an analysis of the actual values taken by the set of indices $\\{k_{1},k_{2}\\}$ reveals several additional trends that further limit the allowed values of ground-state $L_{m}$’s. In particular, starting with the values $\\{0,0\\}$ at $B=5$ T ($L_{0}=55$), the indices $\\{k_{1},k_{2}\\}$ reach the values $\\{3,24\\}$ at $B=25$ T ($L_{m}=256$). As seen from table 3, the outer index $k_{2}$ changes faster than the inner index $k_{1}$. This behavior minimizes the total kinetic energy of the independently rotating rings; indeed, the kinetic energy of the inner ring (as a function of $k_{1}$) rises faster than that of the outer ring (as a function of $k_{2}$) due to smaller moment of inertia (smaller radius) of the inner ring [see (7.2)]. Figure 22: (Color online) Conditional probability distributions for the REM ground state of $N=11$ electrons at $B=10$ T ($L=106$). The electrons are arranged in a (3,8) structure. The observation point (solid dot) is placed on (left) the outer ring at $r_{0}=1.480R_{0}$, and (right) on the inner ring at $r_{0}=0.557R_{0}$. Parameters used: confinement $\hbar\omega_{0}=3.60$ meV, dielectric constant $\kappa=13.1$, effective mass $m^{*}=0.067m_{e}$. Lengths in units of $R_{0}=(2e^{2}/m^{*}\kappa\omega_{0}^{2})^{1/3}$. CPDs (vertical axes) in arbitrary units. In addition to the overestimation of the ground-state energy values for smaller magnetic fields (see figure 21 and our discussion above), there are additional shortcomings of the lowest-Landau-level approximation pertaining to the ground-state ring configurations. In particular, for $N=11$, we find that according to the LLL approximation the ground-state angular momentum immediately after the maximum density droplet ($L_{0}=55$) is $L_{m}=66$, i.e., the one associated with the $(0,N)$ vortex-in-the-center configuration. This result, erroneously stated in Refs. [242, 243] as the ground state, disagrees with the correct result that includes the full effect of the confinement and is listed in table 3, where the ground-state angular momentum immediately following the maximum density droplet is $L_{m}=63$. This angular- momentum value corresponds to the classicaly most stable (3,8) ring configuration, that is, a configuration with no vortex at all (see also the case of $N=9$ electrons in Ref. [53]). Figure 22 displays the REM conditional probability distributions for the ground state of $N=11$ electrons at $B=10$ T ($L_{m}=106$). The (3,8) ring configuration is clearly visible. We note that when the observation point is placed on the outer ring (left panel), the CPD reveals the crystalline structure of this ring only; the inner ring appears to have a uniform density. To reveal the crystalline structure of the inner ring, the observation point must be placed on this ring; then the outer ring appears to be uniform in density. This behavior suggests that the two rings rotate independently of each other, a property that is explored in the next section to derive an approximate quasiclassical expression for the yrast rotational spectra associated with an arbitrary number of electrons. ### 7.3 Approximate analytic expression for the yrast-band spectra In figure 23, we display the CPD for the REM wave function of $N=17$ electrons. This case has a nontrivial three-ring structure (1,6,10) [114] which is sufficiently complex to allow generalizations for larger numbers of particles. The remarkable floppy character (leading to a non-classical, non- rigid rotational inertia, see section VI of Ref. [53]) of the REM is illustrated in the CPDs of figure 23. Indeed, as the two CPDs [reflecting the choice of taking the observation point [${\bf r}_{0}$ in (1.1)] on the outer (left frame) or the inner ring (right frame)] reveal, the polygonal electron rings rotate independently of each other. Thus, e.g., to an observer located on the inner ring, the outer ring will appear as having a uniform density, and vice versa. The wave functions obtained from exact diagonalization exhibit also the property of independently rotating rings [see, e.g., the $N=12$ and $L=132$ ($\nu=1/2$) case in figure 24], which is a testimony to the ability of the REM wave function to capture the essential physics of a finite number of electrons in high $B$. In particular, the conditional probability distribution displayed in figure 24 for exact-diagonalization wave functions exhibits the characteristics expected from the CPD evaluated using REM wave functions for the (3,9) configuration and with an angular-momentum decomposition into shell contributions [see Equation (2.25) and Equation (2.27)] $L_{1}=3+3k_{1}$ and $L_{2}=63+9k_{2}$ ($L_{1}+L_{2}=L_{m}$; for $L_{m}=132$ the angular-momentum decomposition is $L_{1}=6$ and $L_{2}=126$). Figure 23: (Color online) Ground-state conditional probability distributions obtained from REM wave functions for the ground state of $N=17$ electrons at $B=10$ T ($L=228$). The electrons are arranged in a (1,6,10) structure. The observation point (solid dot) is placed on the outer ring at $r_{0}=1.858R_{0}$ (left frame), and on the inner ring at $r_{0}=0.969R_{0}$ (right frame). The rest of the parameters are: confinement $\hbar\omega_{0}=3.6$ meV, dielectric constant $\kappa=13.1$, effective mass $m^{*}=0.067m_{e}$. Lengths in units of $R_{0}=(2e^{2}/(\kappa m^{*}\omega_{0}^{2}))^{1/3}$. CPDs (vertical axes) in arbitrary units. Figure 24: (Color online) CPDs for $N=12$ electrons and with angular momentum $L=132$ ($\nu=1/2$) calculated using exact diagonalization in the lowest Landau level. The electrons are arranged in a (3,9) structure. The observation point (solid dot) is placed on the outer ring at $r_{0}=5.22l_{B}$ (left frame), and on the inner ring at $r_{0}=1.87l_{B}$ (right frame). Lengths in units of $l_{B}$. CPDs (vertical axes) in arbitrary units. In addition to the conditional probabilities, the floppy-rotor character of the REM is revealed in its excited rotational spectrum for a given $B$. From our microscopic calculations based on the wave function in (2.25), we have derived (see below) an approximate (denoted as “app”), but analytic and parameter-free, expression [see (7.7) below] which reflects directly the nonrigid character of the REM for arbitrary size. This expression allows calculation of the energies of REMs for arbitrary $N$, given the corresponding equilibrium configuration of confined classical point charges. We focus on the description of the yrast band at a given $B$. Motivated by the aforementioned nonrigid character of the rotating electron molecule, we consider the following kinetic-energy term corresponding to a $(n_{1},\ldots,n_{q},\ldots,n_{r})$ configuration (with $\sum_{q=1}^{r}n_{q}=N$): $E_{\rm{app}}^{\rm{kin}}(N)=\sum_{q=1}^{r}\hbar^{2}L_{q}^{2}/(2{\cal J}_{q}(a_{q}))-\hbar\omega_{c}L/2,$ (7.2) where $L_{q}$ is the partial angular momentum associated with the $q$th ring about the center of the dot and the total angular momentum is $L=\sum_{q=1}^{r}L_{q}$. ${\cal J}_{q}(a_{q}))\equiv n_{q}m^{*}a_{q}^{2}$ is the rotational moment of inertia of each individual ring, i.e., the moment of inertia of $n_{q}$ classical point charges on the $q$th polygonal ring of radius $a_{q}$. To obtain the total energy, $E_{L}^{\rm{REM}}$, we include also the term $E_{\rm{app}}^{\rm{hc}}(N)=\sum_{q=1}^{r}{\cal J}_{q}(a_{q})\tilde{\omega}^{2}/2$ due to the effective harmonic confinement $\tilde{\omega}$ (see Appendix A.1), as well as the interaction energy $E_{\rm{app}}^{C}$, $E_{\rm{app}}^{C}(N)=\sum_{q=1}^{r}\frac{n_{q}S_{q}}{4}\frac{e^{2}}{\kappa a_{q}}+\sum_{q=1}^{r-1}\sum_{s>q}^{r}V_{C}(a_{q},a_{s}).$ (7.3) The first term is the intra-ring Coulomb-repulsion energy of $n_{q}$ point- like electrons on a given ring, with a structure factor $S_{q}=\sum_{j=2}^{n_{q}}(\sin[(j-1)\pi/n_{q}])^{-1}.$ (7.4) The second term is the inter-ring Coulomb-repulsion energy between rings of uniform charge distribution corresponding to the specified numbers of electrons on the polygonal rings. The expression fo $V_{C}$ is $V_{C}(a_{q},a_{s})=n_{q}{n_{s}}e^{2}[\kappa(a_{q}^{2}+a_{s}^{2})^{1/2}]^{-1}\;{{}_{2}F_{1}}[3/4,1/4;1;4a_{q}^{2}a_{s}^{2}(a_{q}^{2}+a_{s}^{2})^{-2}],$ (7.5) where ${{}_{2}F_{1}}$ is the hypergeometric function. For large $L$ (and/or $B$), the radii of the rings of the rotating molecule can be found by neglecting the interaction term in the total approximate energy, thus minimizing only $E_{\rm{app}}^{\rm{kin}}(N)+E_{\rm{app}}^{\rm{hc}}(N)$. One finds $a_{q}=\lambda\sqrt{L_{q}/n_{q}},$ (7.6) with $\lambda=\tilde{l}=\sqrt{\hbar/m^{*}\tilde{\omega}}$; i.e., the ring radii depend on the partial angular momentum $L_{q}$, reflecting the lack of radial rigidity. Substitution into the above expressions for $E_{\rm{app}}^{\rm{kin}}$, $E_{\rm{app}}^{\rm{hc}}$, and $E_{\rm{app}}^{C}$ yields for the total approximate energy the final expression: $E_{\rm{app},L}^{\rm{REM}}(N)=\hbar(\tilde{\omega}-\omega_{c}/2)L+\sum_{q=1}^{r}\frac{C_{V,q}}{L_{q}^{1/2}}+\sum_{q=1}^{r-1}\sum_{s>q}^{r}V_{C}(\lambda\sqrt{\frac{L_{q}}{n_{q}}},\lambda\sqrt{\frac{L_{s}}{n_{s}}}),$ (7.7) where the constants $C_{V,q}=0.25n_{q}^{3/2}S_{q}e^{2}/(\kappa\lambda).$ (7.8) For simpler $(0,N)$ and $(1,N-1)$ ring configurations, Equation (7.7) reduces to the expressions reported earlier [51, 128]. The floppy-rotor character of the REM under strong magnetic field is reflected in the absence in (7.7) of a kinetic-energy term proportional to $L^{2}$. This contrasts with the rigid-rotor behavior of an electron molecule at zero magnetic field (see section 5.2 and Ref. [51]). ### 7.4 Possible implications for the thermodynamic limit While our focus in this section is on the behavior of trial and exact wave functions in (finite) quantum dots in high magnetic fields, it is natural to inquire about possible implications of our findings to fractional-quantum- Hall-effect systems in the thermodynamic limit. We recall that appropriate trial wave functions for clean FQHE systems possess a good angular momentum $L\geq L_{0}$, a property shared by both the CF/JL and REM functions [24, 55, 57, 236]. We also recall the previous finding [55, 236] that for large fractional fillings $\nu>1/7$, the liquid-like (and circularly uniform) Jastrow/Laughlin function is in the thermodynamic limit energetically favored compared to the broken-symmetry static Wigner crystal (which has no good angular momentum); for $\nu<1/7$, the static Wigner crystal becomes lower in energy. This finding was enabled by the simple form of the JL functions, which facilitated computations of total energies as a function of size for sufficiently large $N$ (e.g., $N=1000$). A main finding of the recent literature on quantum dots is that the exact- numerical-diagonalization wave functions of small systems ($N\leq 12$) are crystalline in character for both low and high fractional fillings. This finding contradicts earlier suggestions [55, 229, 236] that, for high $\nu$’s, small systems are accurately described by the liquid-like JL wave functions and their descendants, e.g., the composite-fermion ones. Of course, for the same high $\nu$’s, our small-size results cannot exclude the possibility that the CPDs of the exact solution may exhibit with increasing $N$ a transition from crystalline to liquid character, in agreement with the JL function. However, as of now the existence of such a transition remains an open theoretical subject. For the low fractions, the rotating-electron-molecule theory raises still another line of inquiry. Due to the specific form of the REM wave functions, computational limitations (in the so-called disk geometry that is natural to quantum dots) prevent us at present from making extrapolations of total energies at a given $\nu$ as $N\rightarrow\infty$. Nevertheless, from the general theory of projection operators, one can conclude that the REM energies exhibit a different trend compared to the JL ones, whose energies were found [55, 236] to be higher than the static Wigner crystal. Indeed the rotating- electron-molecule wave functions remain lower in energy than the corresponding static crystalline state for all values of $N$ and $\nu$, even in the thermodynamic limit. This is due to an “energy gain” theorem (see Section 3 in Ref. [241]) stating that at least one of the projected states (i.e., the ground state) has an energy lower than that of the original broken-symmetry trial function (e.g., the UHF determinant), and this theorem applies for any number of electrons $N$ and for all values of the magnetic field $B$. Naturally, the REM wave functions will be physically relevant compared to those of the broken-symmetry crystal at the thermodynamic limit if the energy gain does not vanish when $N\rightarrow\infty$; otherwise, one needs to consider the posssibility that the static crystal is the relevant physical picture. Figure 25: Stabilization energies $\Delta E^{\rm{gain}}_{\rm{gs}}=E^{\rm{gs}}_{\rm{REM}}-E_{\rm{UHF}}$ for $N=6$ (dashed curve) and $N=7$ (solid curve) fully polarized electrons in a parabolic QD as a function of $B$. The troughs associated with the major fractional fillings (1/3, 1/5, and 1/7) and the corresponding ground-state angular momenta [$L=N(N-1)/(2\nu)$] are indicated with arrows. We have extended the calculations up to $B=120$ T (not shown), and verified that $\Delta E^{\rm{gain}}_{\rm{gs}}$ remains negative while its absolute value vanishes as $B\rightarrow\infty$. The choice of parameters is: $\hbar\omega_{0}=3$ meV (parabolic confinment), $m^{*}=0.067m_{e}$ (electron effective mass), and $\kappa=12.9$ (dielectric constant). The discussion in the above paragraph may be recapitulated by the following question: which state is the relevant one in the thermodynamic limit $(N\rightarrow\infty)$ – the broken-symmetry one (i.e., the static crystal) or the symmetry restored (i.e., rotating crystal) state? This question, in the context of bulk broken-symmetry systems, has been addressed in the early work of Anderson [10] who concluded that the broken-symmetry state (here the UHF static crystalline solution) can be safely taken as the effective ground state. In arriving at this conclusion Anderson invoked the concept of (generalized) rigidity. As a concrete example, one would expect a crystal to behave like a macroscopic body, whose Hamiltonian is that of a heavy rigid rotor with a low-energy excitation spectrum $L^{2}/2{\cal J}$, the moment of inertia ${\cal J}$ being of order $N$ (macroscopically large when $N\rightarrow\infty$). The low-energy excitation spectrum of this heavy rigid rotor above the ground-state ($L=0$) is essentially gapless (i.e., continuous). Thus although the formal ground state posseses continuous rotational symmetry (i.e., $L=0$), “there is a manifold of other states, degenerate in the $N\rightarrow\infty$ limit, which can be recombined to give a very stable wave packet with essentially the nature” [10] of the broken- symmetry state (i.e., the static Wigner crystal in our case). As a consequence of the “macroscopic heaviness” as $N\rightarrow\infty$, one has: (I) The energy gain due to symmetry restoration (i.e., the stabilization energy $\Delta E^{\rm{gain}}_{\rm{gs}}=E^{\rm{gs}}_{\rm{REM}}-E_{\rm{UHF}}$, see figure 25) vanishes as $N\rightarrow\infty$, and (II) The relaxation of the system from the wave packet state (i.e., the static Wigner crystal) to the symmetrized one (i.e., the rotating crystal) becomes exceedingly long. This picture underlies Anderson’s aforementioned conclusion that in the thermodynamic limit the broken-symmetry state may be used as the effective ground state. Consequently, in the rest of this section we will focus on issues pertaining to the “rigidity” of the rotating electron molecule in high magnetic fields. In particular, using our projection method and exact diagonalization, we have demonstrated explicitly [50, 51] that the rigid-rotor picture applies to an $N$-electron QD only when $B=0$. In contrast, in the presence of a high magnetic field, we found [51, 52, 53] that the electrons in the quantum dot do not exhibit global rigidity and therefore cannot be modeled as a macroscopic rotating crystal. Instead, a more appropriate model is that of a highly non- rigid rotor whose moment of inertia depends strongly on the value of the angular momentum $L$. This behavior originates from the dominance of the magnetic field over the Coulomb repulsion. The non-rigid rotor at high $B$ has several unique properties: (I) The ground state has angular momentum $L_{\rm{gs}}>0$; (II) While the rotating electron molecule does not exhibit global rigidity, it possesses azimuthal rigidity (i.e., all electrons on a given ring rotate coherently), with the rings, however, rotating independently of each other. Furthermore, the radii of the rings vary for different values of $L$, unlike the case of a rigid rotor; (III) The excitation spectra do not vary as $L^{2}$; instead they consist of terms that vary as $aL+\sum_{q=1}^{r}b_{q}/\sqrt{L_{q}}$ (with $\sum_{q=1}^{r}L_{q}=L$; for the precise values of the constants $a$ and $b$ see section 7.3 and Refs. [51, 53]); (IV) The angular momentum values are given by the magic values [see section 6.1] $L=L_{0}+\sum_{q=1}^{r}k_{q}n_{q}$, where $(n_{1},n_{2},\ldots,n_{r})$ is the polygonal ring arrangement of the static Wigner molecule (with $n_{q}$ the number of electrons on the $q$th ring) and $k_{1}<k_{2}<\ldots<k_{q}$ are nonnegative integers. These magic $L$’s are associated with the cusp states which exhibit a relative energy gain with respect to neighboring excitations. Thus the low-energy excitation spectrum of the non-rigid rotor is not dense and exhibits gaps due to the occurrence of the magic (cusp) states (see figure 26). Furthermore, these gaps are reflected in the oscillatory behavior of $\Delta E^{\rm{gain}}_{\rm{gs}}$ (see, e.g., figure 25) as a function of $B$ (or $\nu$). Figure 26: Low-energy part of the spectrum of the parabolic QD whose parameters are the same as those in figure 25, calculated as a function of the angular momentum $L$ through exact diagonalization for $N=7$ electrons at a magnetic field $B=18.8$ T. We show here the spectrum in the interval $95\leq L\leq 115$ (in the neighborhood of $\nu=1/5$). The magic angular momentum values corresponding to cusp states are marked (99, 105, and 111), and they are seen to be separated from the rest of the spectrum. For the given value of $B$, the global energy minimum (ground state) occurs for $L_{\rm{gs}}=105$, and the gap $\Delta$ to the first excited state ($L=99$) is indicated. The lowest energies for the different $L$’s (the yrast band) in the plotted range are connected by a dashed line, as a guide to the eye. The zero of energy corresponds to $7\hbar\tilde{\omega}$, where $\tilde{\omega}=(\omega_{0}^{2}+\omega_{c}^{2}/4)^{1/2}$ and $\omega_{c}=eB/(m^{*}c)$. The horizontal arrow denotes the energy of the Laughlin quasihole at L=112. It is seen that the Laughlin quasihole is not the lowest excited state, as presumed in Ref. [55]. As $N$ increases, more polygonal rings are successively added, and since the polygonal rings rotate independently of each other (see, e.g., the case of $N=12$ in figure 24), we expect that the non-rigid-rotor picture remains valid even as $N\rightarrow\infty$. As a result, it is plausible to conjecture the following properties at high $B$ in the thermodynamic limit: (I) the oscillatory character of $\Delta E^{\rm{gain}}_{\rm{gs}}$ will maintain, yielding nonvanishing stabilization energies at the fractional fillings $\nu$, and (II) the low-energy excitation spectra of the system will still exhibit gaps in the neighborhood of the magic angular momenta (see figure 26). Of course, these conjectures need to be further supported through numerical calculations for large $N$. Nevertheless, the above discussion indicates that the question of which state is physically relevant for low fractions in the thermodynamic limit at high $B$ – i.e., the broken-symmetry static crystal or the symmetrized rotating crystal – remains open, and cannot be answered solely following the path of Anderson as described in Ref. [10]. The rotating Wigner crystal has properties characteristic of FQHE states, i.e., it is incompressible (connected to the presence of an excitation gap) and carries a current (while the broken-symmetry static crystal is insulating). Thus, we may conjecture that a transition at lower fractional fillings from a conducting state with good circular symmetry to an insulating Wigner crystal cannot occur spontaneously for clean systems. Therefore, it should be possible to observe FQHE-type behavior at low fractional fillings in a clean system – a prediction that could explain the observations of Ref. [244], where FQHE behavior has been observed for low fractional fillings typically associated with the formation of a static Wigner crystal. In practice, however, impurities and defects may influence the properties of the rotating crystal (and its excitations), depending on the magnitude of the excitation gap (see, e.g., figure 26). Thus one of the main challenges for observation of the fractional quantum Hall effect at such low fillings relates to fabrication of high mobility (nearly impurity-free) samples [245]. We remark, however, that the stabilization energy and the gap $\Delta$ (see, e.g., figure 26) diminish as the magnetic field increases, and as a result the impurities become more efficient in influencing the rotating Wigner crystal for the lower fractional fillings (i.e., higher angular momenta). ## 8 Bosonic molecules in rotating traps: Original results and applications ### 8.1 Variational description of rotating boson molecules Recent experimental advances in the field of trapped ultracold neutral bosonic gases have enabled control of the strength of interatomic interactions over wide ranges [85, 86, 87, 246], from the very weak to the very strong. This control is essential for experimental realizations of novel states of matter beyond the well known Bose-Einstein condensate [85, 86, 87]. In this context, the linear 1D Tonks-Girardeau regime of impenetrable trapped bosons has generated intensive theoretical activity [247, 248] and several experimental realizations of it have been reported most recently [85, 86]. In this section, we address the properties of strongly-repelling impenetrable bosons in rotating ring-shaped or 2D harmonic traps. It has been found that impenetrable bosons are “localized” relative to each other [60, 63, 85] and exhibit nontrivial intrinsic crystalline correlations [60, 63]. For a small number of bosons, $N$, these crystalline arrangements are reminiscent of the structures exibited by the well-studied rotating electron molecules in quantum dots under high magnetic fields [26, 52, 53]. Consequently, we use in the following the term rotating boson molecules. A central result of our study is that the point-group symmetries of the intrinsic crystalline structures give rise to characteristic regular patterns (see below) in the ground-state spectra and associated angular momenta of the RBMs as a function of the rotational frequency for neutral bosons (or the magnetic field for charged bosons). An unexpected result of our studies is that the rotation of repelling bosons (even those interacting weakly) does not necessarily lead to formation of vortices, as is familiar from the case of rotating Bose-Einstein condenstates. In particular, for small $N$, we will show that the Gross-Pitaevskii energies (including those corresponding to formation of vortices) remain always higher compared to the ground-state energies of the RBMs. Of course, we expect that the rotating BEC will become the preferred ground state for sufficiently large $N$ in the case of weakly repelling neutral bosons. We anticipate, however, that it will be feasible to test our unexpected results for small $N$ by using rotating optical lattices, where it is established that a small finite number of atoms can be trapped per given site [87]. In a non-rotating trap, it is natural to describe a localized boson (at a position ${\bf R}_{j}$) by a simple displaced Gaussian [60]. When the rotation of the trap is considered, the Gaussian needs to be modified by a phase factor, determined through the analogy between the one-boson Hamiltonian in the rotating frame of reference and the planar motion of a charged particle under the influence of a perpendicular magnetic field $B$ (described in the symmetric gauge). That is, the single-particle wave function of a localized boson is $\varphi_{j}({\bf r})\equiv\varphi({\bf r},{\bf R}_{j})=\frac{1}{\sqrt{\pi}\lambda}\exp\left[\frac{({\bf r}-{\bf R}_{j})^{2}}{2\lambda^{2}}-\rmi{\bf r}{\bf\cdot}({\bf Q}\times{\bf R}_{j})\right],$ (8.1) with ${\bf Q}\equiv{\bf\hat{z}}/\Lambda^{2}$ and the width of the Gaussian $\lambda$ is a variational parameter; $\Lambda\equiv l_{B}\sqrt{2}=\sqrt{2\hbar c/(eB)}=\sqrt{2\hbar/(m\omega_{c})}$ for the case of a perpendicular magnetic field ${\bf B}$, and $\Lambda\equiv l_{\Omega}\sqrt{2}=\sqrt{\hbar/(m\Omega)}$ in the case of a rotating trap with rotational frequency ${\bf\Omega}$ (we recall that $\omega_{c}\rightarrow 2\Omega$, see the Appendix). Note that we consider a 2D trap, so that ${\bf r}\equiv(x,y)$ and ${\bf R}\equiv(X,Y)$. Figure 27: (Color online) Properties of $N=8$ neutral repelling bosons in a rotating toroidal trap as a function of the reduced rotational frequency $\Omega/\omega_{0}$. The confining potential is given by (8.4) with $n=2$ and radius $r_{0}=3l_{0}$, and the interaction-strength parameter was chosen as $R_{\delta}=50$. (a) RBM ground-state energies, $E^{\rm{PRJ}}$. The inset shows the range $0\leq\Omega/\omega_{0}\leq 0.3$. The numbers denote ground- state magic angular momenta. (b) Energy difference $E^{\rm{PRJ}}-E^{\rm{UBHF}}$. (c) Total angular momenta associated with (i) the RBM ground states [thick solid line (showing steps and marked as PRJ); online black] and (ii) the UBHF solutions (thin solid line; online red). In the figures, we may use the symbol $L_{z}$, instead of simply $L$, to denote the 2D total angular momentum. The Hamiltonian corresponding to the single-particle kinetic energy is given by $H_{K}({\bf r})=({\bf p}-\hbar{\bf Q}\times{\bf r})^{2}/(2m),$ (8.2) for the case of a magnetic field, and by $H_{K}({\bf r})=({\bf p}-\hbar{\bf Q}\times{\bf r})^{2}/(2m)-m\Omega^{2}{\bf r}^{2}/2,$ (8.3) for the case of a rotating frame of reference.999The single-particle wave function in (8.1) and the many-body projected wave function in (8.5) contain contributions from higher Landau levels. These wave functions belong exclusively to the lowest Landau level only in the limit when $\lambda=\sqrt{2}l_{B}$ in the case of a magnetic field, or $\lambda=\sqrt{2}l_{\Omega}$ and $\Omega/\omega_{0}=1$ in the case of a rotating trap. A toroidal trap with radius $r_{0}$ can be specified by the confining potential $V({\bf r})=\frac{\hbar\omega_{0}}{2}(r-r_{0})^{n}/l_{0}^{n},$ (8.4) with $l_{0}=\sqrt{\hbar/(m\omega_{0})}$ being the characteristic length of the 2D trap. For $n\gg 2$ and $l_{0}/r_{0}\to 0$ this potential approaches the limit of a toroidal trap with zero width, which has been considered often in previous theoretical studies (see, e.g., Ref. [249]). In the following, we consider the case with $n=2$, which is more realistic from the experimental point of view. In this case, in the limit $r_{0}=0$, one recovers a harmonic trapping potential. To construct an RBM variational many-body wave function describing $N$ impenetrable bosons in the toroidal trap, we use $N$ displaced orbitals $\varphi({\bf r},{\bf R}_{i})$, $i=1,2,\ldots,N$ [see (8.1)] centered at the vertices of a regular polygon. Then, we first construct an unrestricted Bose Hartree-Fock permanent [60, 63] $|\Phi^{\rm{UBHF}}_{N}\rangle\propto\sum_{P(i_{m})}\varphi_{1}({\bf r}_{i_{1}})\varphi_{2}({\bf r}_{i_{2}})\ldots\varphi_{N}({\bf r}_{i_{N}})$. The UBHF permanent breaks the circular symmetry of the many-body Hamiltonian. As was discussed in section 3.2, the “symmetry dilemma” is resolved through a subsequent “symmetry-restoration” step accomplished via projection techniques [23, 24, 30, 31, 52, 53], i.e., we construct a many-body wave function with good total angular momentum by applying the projection operator $\hat{\cal P}_{L}=(1/2\pi)\int_{0}^{2\pi}\rmd\theta\exp[\rmi\theta(L-\hat{L})]$, so that the final RBM wave function is given by $|\Psi^{\rm{PRJ}}_{N,L}\rangle=\frac{1}{2\pi}\int_{0}^{2\pi}\rmd\theta|\Phi^{\rm{UBHF}}_{N}(\theta)\rangle\rme^{\rmi\theta L}.$ (8.5) $|\Phi^{\rm{UBHF}}_{N}(\theta)\rangle$ is the original UBHF permanent rotated by an azimuthal angle $\theta$. We note that, in addition to having good angular momenta, the projected wave function $|\Psi^{\rm{PRJ}}_{N,L}\rangle$ has also a lower energy than that of $|\Phi^{\rm{UBHF}}_{N}\rangle$ [see, e.g. $E_{L}^{\rm{PRJ}}-E^{\rm{UBHF}}$ in figure 27(b). The projected ground-state energy is given by $E_{L}^{\rm{PRJ}}=\int_{0}^{2\pi}h(\theta)\rme^{\rmi\theta L}\rmd\theta\left/\int_{0}^{2\pi}n(\theta)\rme^{\rmi\theta L}\right.\rmd\theta,$ (8.6) where $h(\theta)=\langle\Phi_{N}^{\rm{UBHF}}(\theta=0)|{\cal H}|\Phi_{N}^{\rm{UBHF}}(\theta)\rangle$ and $n(\theta)=\langle\Phi_{N}^{\rm{UBHF}}(\theta=0)|\Phi_{N}^{\rm{UBHF}}(\theta)\rangle$; the latter term ensures proper normalization. Figure 28: (Color online) Single-particle densities and CPDs for $N=8$ bosons in a rotating toroidal trap with $\Omega/\omega_{0}=0.2$ and $R_{\delta}=50$. The remaining trap parameters are as in figure 27. (a) Gross-Pitaevskii single-particle density. (b) UBHF single-particle density exhibiting breaking of the circular symmetry. (c) RBM single-particle density exhibiting circular symmetry. (d) CPD for the RBM wave function [PRJ wave function, see Equation (8.5)] revealing the hidden point-group symmetry in the intrinsic frame of reference. The observation point is denoted by a white dot. The RBM ground- state angular momentum is $L_{z}=16$. Lengths in units of $l_{0}$. The vertical scale is the same for (b), (c), and (d), but different for (a). The many-body Hamiltonian in the rotating trap is given by ${\cal H}=\sum_{i=1}^{N}[H_{K}({\bf r}_{i})+V({\bf r}_{i})]+\sum_{i<j}^{N}v({\bf r}_{i},{\bf r}_{j}),$ (8.7) with the interparticle interaction being given by a contact potential $v_{\delta}({\bf r}_{i},{\bf r}_{j})=g\delta({\bf r}_{i}-{\bf r}_{j})$ for neutral bosons and a Coulomb potential $v_{C}({\bf r}_{i},{\bf r}_{j})=Z^{2}e^{2}/|{\bf r}_{i}-{\bf r}_{j}|$ for charged bosons. The parameter that controls the strength of the interparticle repulsion relative to the zero-point kinetic energy is given by $R_{\delta}=gm/(2\pi\hbar^{2})$ [60, 63] for a contact potential and $R_{W}=Z^{2}e^{2}/(\hbar\omega_{0}l_{0})$ [20, 60] for a Coulomb repulsion. For a given value of the dimensionless rotational frequency, $\Omega/\omega_{0}$, the projection yields wave functions and energies for a whole rotational band comprising many angular momenta. In the following, we focus on the ground-state wave function (and corresponding angular momentum and energy) associated with the lowest energy in the band. Figure 29: (Color online) Properties of $N=6$ neutral bosons in a rotating harmonic trap as a function of the reduced rotational frequency $\Omega/\omega_{0}$. The confining potential is given by (8.4) with $n=2$ and $r_{0}=0$, and the interaction-strength parameter was chosen as $R_{\delta}=50$. The intrinsic molecular structure is $(1,5)$. (a) RBM ground- state energies, $E^{\rm{PRJ}}$. The inset shows a smaller range. The numbers denote ground-state angular momenta. (b) Total angular momenta associated with (i) the RBM ground states (thick solid line showing steps; online black) and (ii) the UBHF solutions (thin solid line; online red). Figure 27(a) displays the ground-state energy $E_{\rm{PRJ}}$ of $N=8$ bosons in a toroidal trap as a function of the dimensionless rotational frequency $\Omega/\omega_{0}$, with $\omega_{0}$ being the trap frequency. The prominent features in figure 27(a) are: (i) the energy diminishes as $\Omega/\omega_{0}$ increases; this is an effect of the centrifugal force, and (ii) the $E_{\rm{PRJ}}$ curve consists of linear segments, each one associated with a given angular momentum $L$. Most remarkable is the regular variation of the values of $L$ with a constant step of $N$ units (here $N=8$) [see inset in figure 27(a) and figure 27(c)]. These preferred angular momenta $L=kN$ with integer $k$, are reminiscent of the so called “magic angular momenta” familiar from studies of electrons under high-magnetic fields in 2D semiconductor quantum dots [26, 52, 53]. Figure 30: (Color online) Properties of GP solutions (thin solid line; online red) versus those of RBM wave functions (thick solid line; online green) for $N=6$ neutral bosons as a function of the reduced rotational frequency $\Omega/\omega_{0}$. A harmonic trap is considered, and the interaction strength equals $R_{\delta}=50$. (a) Ground-state energies. (b) Associated ground-state angular momenta. (c) GP (BEC) single-particle density at $\Omega/\omega_{0}=0.65$ having 7 vortices with a 6-fold symmetry (thus exhibiting breaking of the circular symmetry). (d) RBM single-particle density at $\Omega/\omega_{0}=0.65$ which does not break the circular symmetry. (e) CPD of the RBM at $\Omega/\omega_{0}=0.65$ revealing the intrinsic (1,5) crystalline pattern. The white dot denotes the observation point ${\bf r}_{0}$. Note the dramatic difference in spatial extent between the GP and RBM wave functions [compare (c) with (d) and (e). Lengths in units of $l_{0}$. The vertical scale is the same for (d) and (e), but different for (c). The preferred angular momenta reflect the intrinsic molecular structure of the localized impenetrable bosons. We note, that the (0,8) polygonal-ring arrangement is obvious in the single-particle density associated with the UBHF permanent [see figure 28(b)]; (0,8) denotes no particles in the inner ring and 8 particles in the outer one. After restoration of symmetry, however, the single-particle density is circularly symmetric [see the PRJ single-particle density in figure 28(c)] and the intrinsic crystallinity becomes “hidden”; it can, however, be revealed via the conditional probability distribution [20, 52, 53, 60] [CPD, see figure 28(d)]. We note the Gross-Pitaevskii single- particle density in figure 28(a), which is clearly different from the PRJ density in figure 28(c). The internal structure for charged bosons in a toroidal trap (not shown) is similar to that of neutral bosons (figure 28), i.e., a (0,8) ring arrangement, portrayed also in the stepwise variation (in steps of 8 units) of the total angular momenta. The internal structure is also reflected in the variation of the ground-state total energy as a function of the magnetic field. In contrast to the case of neutral bosons, however, the ground-state energy curve for charged bosons is not composed of linear segments, but of intersecting inverted-parabola-type pieces; this is due to the positive contribution of the Lorentz force compared to the negative contribution of the centrifugal force in a rotating trap. For RBMs in rotating harmonic traps, the polygonal-ring pattern of localized bosons becomes more complex than the simple $(0,N)$ arrangement that appears naturally in a toroidal trap. Indeed, in harmonic traps, one anticipates the emergence of concentric ring structures. For $N=6$ neutral bosons in a harmonic trap, we observe that, as in the case of a toroidal trap, the ground- state energy as a function of the reduced rotational frequency, $\Omega/\omega_{0}$, [figure 29(a)] is composed of linear segments, but now the corresponding magic angular momenta [figure 29(b)] vary in steps of $N-1=5$ units. This indicates a rotating boson molecule consisting of two polygonal rings; denoted as a $(1,5)$ structure, with the inner ring having a single boson and the outer ring five. Figure 31: (Color online) Properties of GP solutions (thin solid line; online red) versus those of RBM wave functions (thick solid line; online green) for $N=6$ bosons as a function of the interaction strength $R_{\delta}$. A harmonic trap is considered, and the reduced rotational frequency equals $\Omega/\omega_{0}=0.85$. (a) Ground-state energies (b) Associated ground- state angular momenta. In figure 30(a), we display the rotating-boson-molecule and mean-field Gross- Pitaevskii ground-state energies of $N=6$ strongly repelling (i.e., $R_{\delta}=50$) neutral bosons in a harmonic trap as a function of the reduced angular frequency of the trap. The GP curve (thin solid line; online red) remains well above the RBM curve (thick solid line; online green) in the whole range $0\leq\Omega/\omega_{0}\leq 1$. The RBM ground-state angular momenta exhibit again the periodicity in steps of five units [figure 30(b)]. As expected, the GP total angular momenta are quantized [$L_{z}=0$ (no-vortex) or $L_{z}=6$ (one central vortex)] only for an initial range $0\leq\Omega/\omega_{0}\leq 0.42$. For $\Omega/\omega_{0}\geq 0.42$, the GP total angular momentum takes non-integer values and ceases to be a good quantum number, reflecting the broken-symmetry character of the associated mean field, with each kink signaling the appearance of a different vortex pattern of $p$-fold symmetry ($p=1,2,3,4,\ldots$) [250]; see an example in figure 30(c). The energetic superiority of the RBM wave function over the GP solution demonstrated in figure 30(a) was to be expected, since we considered the case of strongly repelling bosons. Unexpectedly, however, for a small number of neutral bosons the energetic advantage of the rotating boson molecule persists even for weakly repelling bosons, as illustrated in figure 31(a). Indeed, figure 31(a) displays the RBM (thick solid line; online green) and GP (thin solid line; online red) ground-state energies for $N=6$ neutral bosons in a trap rotating with $\Omega/\omega_{0}=0.85$ as a function of the interaction parameter $R_{\delta}$. The surprising result in figure 31(a) is that the GP energy remains above the RBM curve even for $R_{\delta}\rightarrow 0$. Of course the RBM wave function is very close to that of a BEC without vortices when $R_{\delta}\rightarrow 0$ (BECs without vortices are approximately feasible for small $N$). However, for small $N$, our results show that BECs with vortices (i.e., for $L_{z}\geq N$) are not the preferred many-body ground states; instead, formation of RBMs is favored. Note that the energy difference $E^{\rm{GP}}-E^{\rm{PRJ}}$ increases rapidly with increasing $R_{\delta}$, reflecting the fact that the RBM energies saturate (as is to be expected from general arguments), while the GP energies (even with vortices fully accounted for) exhibit an unphysical divergence as $R_{\delta}\rightarrow\infty$ [figure 31(a)]; we have checked this trend up to values of $R_{\delta}=100$ (not shown). Of interest again is the different behavior of the RBM and GP ground state angular momenta [figure 31(b)] (see also discussion of figure 30(b)). To summarize this section: We have studied the ground-state properties of a variational many-body wave function for repelling bosons in rotating traps that incorporates correlations beyond the Gross-Pitaevskii mean-field approximations. This variational wave function describes rotating boson molecules, i.e., localized bosons arranged in polygonal-ring-type patterns in their intrinsic frame of reference. For small numbers of neutral bosons, and in particular in the case of GP vortex formation, the RBM ground-state energies are lower than those associated with the corresponding Gross- Pitaevskii BEC solutions. Given the large differences between the properties of the RBM and BEC wave functions (which become more pronounced for larger interaction parameter $R_{\delta}$), and the recently demonstrated ability to experimentally control $R_{\delta}$ [85, 86, 87, 246], we anticipate that our results could be tested in experiments involving rotating optical lattices. Detection of rotating boson molecules could be based on a variety of approaches, such as the measurement of the spatial extent [contrast the RBM and BEC spatial extents in figure 30(c) - figure 30(e)], or the use of Hanbury Brown-Twiss-type experiments [251] to directly detect the intrinsic crystalline structure of the RBM. ### 8.2 Exact diagonalization for bosons in the lowest Landau level Rotating ultracold trapped Bose condensed systems are most commonly discussed in the context of formation of vortex lattices, which are solutions to the Gross-Pitaevskii mean-field equation [4, 5, 252, 253, 254, 255, 256, 257]. Such vortex lattices have indeed been found experimentally for systems containing a large number of bosons [258, 259, 260]. Nevertheless, several theoretical investigations [67, 68, 69, 70, 71] of rapidly rotating trapped bosonic systems suggested formation of strongly correlated exotic states which differ drastically from the aforementioned vortex-lattice states. While experimental realizations of such strongly correlated states have not been reported yet, there is already a significant effort associated with two- dimenional exact-diagonalization studies of a small number of particles ($N$) in the lowest Landau level; the LLL restriction corresponds to the regime of rapid rotation, where the rotational frequency of the trap $\Omega$ equals the frequency of the confining potential. The large majority [67, 69, 70, 71] of such exact-diagonalization studies have attempted to establish a close connection between rapidly rotating bosonic gases and the physics of electrons under fractional-quantum-Hall-effect conditions employing the bosonic version of “quantum-liquid” analytic wave functions, such as the Laughlin wave functions, composite-fermion, Moore-Read, and Pfaffian functions. As described in section 6, the “quantum-liquid” picture for a small number of trapped electrons in the FQHE regime has been challenged in a series of extensive studies [24, 26, 42, 51, 52, 53] of electrons in 2D quantum dots under high magnetic fields. Such studies (both exact-diagonalization and variational) revealed that, at least for finite systems, the underlying physical picture governing the behavior of strongly-correlated electrons is not that of a “quantum liquid.” Instead, the appropriate description is in terms of a “quantum crystal,” with the localized electrons arranged in polygonal concentric rings [24, 26, 51, 52, 53, 127, 128, 131]. These “crystalline” states lack [52, 53] the familiar rigidity of a classical extended crystal, and are better described [24, 26, 42, 51, 52, 53] as rotating electron (or Wigner) molecules. Motivated by the discovery in the case of electrons of REMs at high $B$ (and from the fact that Wigner molecules form also at zero magnetic field [20, 25, 41, 50, 167, 188]) some theoretical studies have most recently shown that analogous molecular patterns of localized bosons do form in the case of a small number of particles inside a static or rotating harmonic trap [43, 60, 63, 177, 178]. In analogy with the electron case, the bosonic molecular structures can be referred to [63] as rotating boson molecules; a description of RBMs via a variational wave function built from symmetry-breaking displaced Gaussian orbitals with subsequent restoration of the rotational symmetry was presented in Refs. [43, 60, 63] and reviewed in section 8.1. In a recent paper, Baksmaty et al[72] used exact diagonalization in the lowest Landau level to investigate the formation and properties of RBMs focusing on a larger number of particles than previously studied, in particular for sizes where multiple-ring formation can be expected based on our knowledge of the case of 2D electrons in high $B$. A finite number of particles ($N\leq 11$) at both low ($\nu<1/2$) and high ($\nu\geq 1/2$) filling fractions $\nu\equiv N(N-1)/2L$ (where $L\equiv{\cal L}/\hbar$ is the quantum number associated with the total angular momentum ${\cal L}$) was studied and both the cases of a long-range (Coulomb) and a short-range ($\delta$-function) repulsive interaction were investigated. In this section, we report some main results from Ref. [72]. As in the case of electrons in 2D quantum dots, we probe the crystalline nature of the bosonic ground states by calculating the full anisotropic two- point correlation function $P({\bf r},{\bf r}_{0})$ [see Equation (1.1)] associated with the exact wavefunction $\Psi({\bf r}_{1},{\bf r}_{2},\ldots,{\bf r}_{N})$. The quantity $P({\bf r},{\bf r}_{0})$ is proportional to the probability of finding a boson at ${\bf r}$ given that there is another boson at the observation point ${\bf r}_{0}$, and it is often referred to as the conditional probability distribution (section 1.5). A main finding of our studies is that consideration solely of the CPDs is not sufficient for the boson case at high fractional fillings $\nu\geq 1/2$; in this case, one needs to calculate even higher-order correlation functions, e.g., the full $N$-point correlation function defined as the modulus square of the full many-body EXD wave function, i.e., $P({\bf r};{\bf r}_{1},{\bf r}_{2},\ldots,{\bf r}_{N-1})=|\Psi({\bf r};{\bf r}_{1},{\bf r}_{2},\ldots,{\bf r}_{N-1})|^{2},$ (8.8) where one fixes the positions of $N-1$ particles and inquiries about the (conditional) probability of finding the $N$th particle at any position ${\bf r}$. The investigations in this section are also motivated by recent experimental developments, e.g., the realization of trapped ultracold gas assemblies featuring bosons interacting via a long-range dipole-dipole interaction [261, 262]. We expect the results presented in this section to be directly relevant to systems with a two-body repulsion intermediate between the Coulomb and the delta potentials. Additionally, we note the appearance of promising experimental techniques for measuring higher-order correlations in ultra-cold gases employing an atomic Hanbury Brown-Twiss scheme [251] or shot-noise interferometry [263, 264]. Experimental realization of few-boson rotating systems can be anticipated in the near future as a result of increasing sophistication of experiments involving periodic optical lattices co-rotating with the gas, which are capable of holding a few atoms in each site. A natural first step in the study of such systems is the analysis of the physical properties of a few particles confined in a rotating trap with open boundary conditions (i.e., conservation of the total angular momentum $L$). Figure 32: Ground-state angular momenta, $L_{\rm{gs}}$, for $N=6$ bosons in a rapidly rotating trap [described by the LLL Hamiltonian in (8.9)], as a function of the rotational frequency $\Omega$ expressed in units of $\omega_{0}$. The bosons interact via a Coulombic repulsion (left) and a delta repulsion (right), and the many-body Hilbert space is restricted to the lowest Landau level. The angular momentum associated with the first bosonic Laughlin state occurs at $L=30$, i.e., at $N(N-1)$. The value of $c=0.2\hbar\omega_{0}\Lambda$ for the Coulomb case (left) and the value of $g=2\pi\hbar\omega_{0}\Lambda^{2}/N$ for the case of a delta repulsion (right); the many-body wave functions do not depend on these choices. In the delta-interaction case, the values of the angular momenta terminate with the value $L=30$ (the Laughlin value) at $\Omega/\omega_{0}=1$. In contrast, in the Coulomb-interaction case (left), the values of the ground-state angular momenta do not terminate, but diverge as $\Omega/\omega_{0}\rightarrow 1$. Note the stepwise variation of the values of the ground-state angular momenta in both cases, indicating the presence of an intrinsic point-group symmetry associated with the (0,6) and (1,5) polygonal-ring structure of a rotating Boson molecule. The main results of Ref. [72] can be summarized as follows: Similar to the well-established (see section 6 and section 7) emergence of rotating electron molecules in quantum dots, rotating boson molecules form in rotating harmonic traps as well. The RBMs are also organized in concentric polygonal rings that rotate independently of each other, and the polygonal rings correspond to classical equilibrium configurations and/or their low-energy isomers. Furthermore, the degree of crystallinity increases gradually with larger angular momenta $L$’s (smaller filling fractions $\nu$’s), as was the trend [26, 52, 53] for the REMs and as was observed also for $\nu<1/2$ in another study [178] for rotating bosons in the lowest Landau level with smaller $N$ and single-ring structures. We finally note that the crystalline character of the RBMs appears to depend only weakly on the range of the repelling interaction, for both the low (see also Ref. [178]) and high (unlike Ref. [177]) fractional fillings. In studies of 2D quantum dots, CPDs were used some time ago in Refs. [50, 128, 131]. For probing the intrinsic molecular structure in the case of ultracold bosons in 2D traps, however, they were introduced only recently by Romanovsky et al[60]. The importance of using CPDs as a probe can hardly be underestimated. Indeed, while exact-diagonalization calculations for bosons in the lowest Landau level have been reported earlier [67, 68, 69, 70, 71], the analysis in these studies did not include calculations of the CPDs, and consequently formation of rotating boson molecules was not recognized. Figure 33: (a) Single-particle densities [$n({\bf r})$; left column], (b) CPDs $\left[P({\bf r},{\bf r}_{0})\right]$ in 3D plots (middle column), and (c) CPDs in contour plots (right column), portraying the strengthening of the crystalline RBM structure for $N=6$ bosons interacting via a repulsive Coulomb interaction as the filling fraction $\nu$ is reduced. The white dots in the CPD plots indicate the reference point ${\bf r}_{0}$. We note in particular the gradual enhancement of the peak at the center of the plots, and the growth of the radius of the outer ring; the latter reflects the nonrigid-rotor nature of the RBMs (in analogy with the findings of Ref. [52] regarding the properties of rotating electron molecules). The cases of $\nu=1/4$ and $\nu=1/8$ exhibit a clear $(1,5)$ crystalline arrangement, while the case of $\nu=1/2$ (first Laughlin state) is intermediate between a $(1,5)$ and a $(0,6)$ pattern (see text for details). Lengths in units of $\Lambda$. The vertical scales are in arbitrary units, which however do not change for the panels within the same column (a), (b), or (c). #### 8.2.1 The case of $N=6$ bosons in the lowest Landau level. As a specific example of the points discussed above in section 8.2, we present here results for $N=6$ bosons in the lowest Landau level. For additional cases (e.g., $N=9$ and $N=11$), see Ref. [72]. In analogy with the magnetic-field Hamiltonian of Equation (7.1), the many- body Hamiltonian for $N$ bosons in a rotating trap is reduced in the lowest Landau level to the expression $\tilde{\cal H}^{\Omega}_{\rm{LLL}}=N\hbar\omega_{0}+\hbar(\omega_{0}-\Omega)L+\sum_{i<j}^{N}v({\bf r}_{i},{\bf r}_{j}),$ (8.9) where $\omega_{0}$ specifies the 2D harmonic trap and $\Omega$ denotes the rotational frequency. The interparticle interaction is given by a contact potential $v_{\delta}({\bf r}_{i},{\bf r}_{j})=g\delta({\bf r}_{i}-{\bf r}_{j})$ for neutral bosons and a Coulomb potential $v_{C}({\bf r}_{i},{\bf r}_{j})=c/|{\bf r}_{i}-{\bf r}_{j}|$ for charged bosons. Since $\tilde{\cal H}^{\Omega}_{\rm{LLL}}$ is rotationally invariant, i.e., $[\tilde{\cal H}^{\Omega}_{\rm{LLL}},L]=0$, its eigenstates $\Psi_{L}$ must also be eigenstates of the total angular momentum with eigenvalue $\hbar L$. For a given rotational frequency $\Omega$, the eigenstate with lowest energy is the ground state; we denote the corresponding angular momentum as $L_{\rm{gs}}$. We proceed to describe the EXD results for $N=6$ particles interacting via a Coulomb repulsion by referring to figure 32, where we plot $L_{\rm{gs}}$ against the angular frequency $\Omega$ of the rotating trap. A main result from all our calculations is that $L_{\rm{gs}}$ increases in characteristic (larger than unity) steps that take only a few integer values, i.e., for $N=6$ the variations of $L_{\rm{gs}}$ are in steps of 5 or 6. In keeping with previous work on electrons [24, 26, 42, 51, 52, 53] at high $B$, and very recently on bosons in rotating traps [43, 60, 63, 177, 178], we explain these magic-angular-momenta patterns (i.e., for $N=6$, $L_{\rm{gs}}=L_{0}+5k$ or $L_{\rm{gs}}=L_{0}+6k$, with $L_{0}=0$) as manifestation of an intrinsic point-group symmetry associated with the many-body wave function. This point- group symmetry emerges from the formation of RBMs, i.e., from the localization of the bosons at the vertices of concentric regular polygonal rings; it dictates that the angular momentum of a purely rotational state can only take values $L_{\rm{gs}}=L_{0}+\sum_{i}k_{i}n_{i}$, where $n_{i}$ is the number of localized particles on the $i$th polygonal ring. [We remind the reader that for spin-polarized electrons in the lowest Landau level, the corresponding value is $L_{0}=N(N-1)/2$.] Thus for $N=6$ bosons, the series $L_{\rm{gs}}=5k$ is associated with an $(1,5)$ polygonal ring structure, while the series $L_{\rm{gs}}=6k$ relates to an $(0,6)$ arrangement of particles. It is interesting to note that in classical calculations [114] for $N=6$ particles in a harmonic 2D trap, the $(1,5)$ arrangement is found to be the global energy minimum, while the $(0,6)$ structure is the lowest metastable isomer. This fact is apparently reflected in the smaller weight of the $L_{\rm{gs}}=6k$ series compared to the $L_{\rm{gs}}=5k$ series, and the gradual disappearance of the former with increasing $L$. Magic values dominate also the ground state angular momenta of neutral bosons (delta repulsion) in rotating traps, as shown for $N=6$ bosons in the right panel of figure 32. Although the corresponding $\Omega$-ranges along the horizontal axis may be different compared to the Coulomb case, the appearance of only the two series $5k$ and $6k$ is remarkable — pointing to the formation of RBMs with similar $(1,5)$ and $(0,6)$ structures in the case of a delta interaction as well (see also Refs. [177, 178]). An important difference, however, is that for the delta interaction both series end at $\Omega/\omega_{0}=1$ with the value $L=N(N-1)=30$ (for $N=6$ the bosonic Laughlin value at $\nu=1/2$), while for the Coulomb interaction this $L$ value is reached for $\Omega/\omega_{0}<1$ — allowing for an infinite set of magic angular momenta [larger than $N(N-1)$] to develop as $\Omega/\omega_{0}\rightarrow 1$. Beyond the analysis of the ground-state spectra as a function of $\Omega$, the intrinsic crystalline point-group structure can be revealed by an inspection of the CPDs [and to a much lesser extent by an inspection of single-particle densities]. Because the EXD many-body wave function is an eigenstate of the total angular momentum, the single-particle densities are circularly symmetric and can only reveal the presence of concentric rings through oscillations in the radial direction. The localization of bosons within the same ring can only be revealed via the azimuthal variations of the anisotropic CPD [Equation (1.1)]. One of our findings is that for a given $N$ the crystalline features in the CPDs develop slowly as $L$ increases (or $\nu$ decreases). For $\nu<1/2$, we find that the crystalline features are well developed for all sizes studied by us. In figure 33, we present some concrete examples of CPDs from exact-diagonalization calculations associated with the ground-states of $N=6$ bosons in a rotating trap interacting via a repulsive Coulomb potential. In particular, we present the CPDs for $L_{\rm{gs}}=30$ (bosonic Laughlin for $\nu=1/2$), 60, and 120; these angular momenta are associated with ground states at specific $\Omega$-ranges [see figure 32]. All three of these angular momenta are divisible by both 5 and 6. However, only the $L_{\rm{gs}}=30$ CPD (figure 33 top row) has a structure that is intermediate between the $(1,5)$ and the $(0,6)$ polygonal-ring arrangements. The two other CPDs, associated with the higher $L_{\rm{gs}}=60$ and $L_{\rm{gs}}=120$ exhibit clearly only the $(1,5)$ structure, illustrating our statement above that the quantum-mechanical CPDs conform to the structure of the most stable arrangement [i.e., the $(1,5)$ for $N=6$] of classical point-like charges as the fractional filling decreases. Figure 34: Contour plots of the CPD (a) and $N$-point correlation function (b) and (c) for $N=6$ bosons with $L_{\rm{gs}}=15$ interacting via a Coulomb repulsion. The white squares indicate the positions of the fixed particles. The black square in (b) and (c) indicates the position of the 6th particle according to the classical $(1,5)$ molecular configuration. Note the different arrangements of the five fixed particles, i.e., (b) one fixed particle at the center and (c) no fixed particle at the center. Note also that the CPD in (a) fails to reveal the $(1,5)$ pattern, which, however, is clearly seen in the $N$-point correlation functions in both (b) and (c). Lengths in units of $\Lambda$. The vertical scales are arbitrary, but the same in (b) and (c). Figure 35: Contour plots of the CPD (a) and $N$-point correlation function (b) and (c) for $N=6$ bosons with $L_{\rm{gs}}=15$ interacting via a $\delta$-repulsion. The white squares indicate the positions of the fixed particles. The black square in (b) and (c) indicates the position of the 6th particle according to the classical $(1,5)$ molecular configuration. Note the different arrangements of the five fixed particles, i.e., (b) one fixed particle at the center and (c) no fixed particle at the center. Note also that the CPD in (a) fails to reveal the $(1,5)$ pattern, which, however, is clearly seen in the $N$-point correlation functions in both (b) and (c). Lengths in units of $\Lambda$. The vertical scales are arbitrary, but the same in (b) and (c). However, for $\nu>1/2$, the azimuthal variations may not be visible in the CPDs, in spite of the characteristic step-like ground-state spectra [see figure 32 for $N=6$ bosons]. This paradox is resolved when one considers higher-order correlations, and in particular $N$-point correlations [see Equation (8.8)]. In figure 34 and figure 35, we plot the $N$-point correlation functions for $N=6$ bosons and $L_{\rm{gs}}=15$ for both the Coulomb interaction and $\delta$-repulsion, respectively, and we compare them against the corresponding CPDs. The value of 15 is divisible by 5, and one expects this state to be associated with a $(1,5)$ molecular configuration. It is apparent that the CPDs fail to portray such fivefold azimuthal pattern. The $(1,5)$ pattern, however, is clear in the $N$-point correlations (middle and right panels). One has two choices for choosing the positions of the first five particles (white dots), i.e., one choice places one white dot at the center and the other choice places all five white dots on the vertices of a regular pentagon. For both choices, as shown by the contour lines in the figures, the position of maximum probability for the sixth boson coincides with the point that completes the $(1,5)$ configuration [see the black dots in the middle and right panels]. Note that the differences in the CPDs and $N$-point correlation functions between the Coulomb and the $\delta$-repulsion are rather minimal. ## 9 Summary This report reviewed the physics of strong correlations in two-dimensional small finite-size condensed-matter systems, such as electrons in quantum dots and repelling bosons in harmonic traps. It was shown that strong correlations in such systems relate to the appearance of symmetry breaking at the mean- field level of description. Particular attention was given to the similarities of symmetry breaking in these systems despite the different interparticle interactions (Coulombic repulsion in quantum dots versus a contact potential for neutral bososns in harmonic traps). The universal aspects of symmetry breaking in small systems (including nuclei and molecules in quantum chemistry) have been exploited to develop a two-step method of symmetry breaking at the unrestricted Hartree-Fock level and subsequent symmetry restoration via post Hartree-Fock projection techniques. In conjunction with exact-diagonalization calculations, the two-step method was used to describe a vast range of strongly-correlated phenomena associated with particle localization and formation of crystalline (molecular) structures of electrons in quantum dots and bosons in harmonic traps. Due to their finite size, these crystalline structures are different from the familiar rigid crystals of extended systems; they rather resemble and exhibit similarities with the natural 3D molecules (e.g., ro-vibrational spectra). It was shown that strongly-correlated phenomena emerging from symmetry breaking include: (I) Chemical bonding, dissociation, and entanglement in quantum dot molecules and in electron molecular dimers formed within a single anisotropic quantum dot, with potential technological applications to solid-state quantum- computing devices. (II) Electron crystallization, with localization on the vertices of concentric polygonal rings, and formation of rotating electron molecules in circular quantum dots. At zero magnetic field, the REMs can approach the limit of a rigid rotor; at high magnetic field, the REMs are highly floppy, with the rings rotating independently of each other. (III) In the lowest Landau level, the rotating electron molecules are described by parameter-free analytic many-body wave functions, which are an alternative to the composite-fermion and Jastrow/Laughlin approaches, offering a new point of view of the fractional quantum Hall regime in quantum dots (with possible implications for the thermodynamic limit). (IV) Crystalline phases of strongly repelling bosons. In the case of rotating traps and in analogy with the REMs, such repelling bosons form rotating boson molecules, which are energetically favored compared to the Gross-Pitaevkii solutions, in particular in the regime of vortex formation. This work is supported by the US D.O.E. (Grant No. FG05-86ER45234). Calculations were peformed at the Georgia Institute of Technology Center for Computational Materials Science. ## Appendix A In this Appendix, we briefly review the single-particle wave functions and associated energy spectra of a two-dimensional circular harmonic oscillator under the influence of a perpendicular magnetic field $B$ (relevant to the case of quantum dots) or under rotation with angular frequency $\Omega$ (relevant to the case of trapped atomic gases in rotating harmonic traps). These single-particle wave functions and associated spectra are known as the Darwin-Fock states and energy levels, after the names of the authors of two original papers [162, 163] on this subject. Figure 36: (Color online) The Darwin-Fock single-particle energy levels of a 2D harmonic oscillator under the effect of a perpendicular magnetic field $B$ as a function of $\eta=\omega_{c}/\omega_{0}$, where $\omega_{c}$ is the cyclotron frequency and $\omega_{0}$ is the frequency specifying the 2D harmonic confinement. A specific color (online) indicates orbitals with the same number of radial nodes. ### A.1 Two-dimensional isotropic oscillator in a perpendicular magnetic field In this case, the Hamiltonian (for an electron of mass $m^{*}$) is given by: $H=\frac{1}{2m^{*}}({\bf p}-\frac{e}{c}{\bf A})^{2}+\frac{1}{2}m^{*}\omega_{0}^{2}{\bf r}^{2},$ (1.1) where ${\bf r}=(x,y)$ and $\omega_{0}$ is the frequency of the oscillator. In the symmetric gauge, the vectror potential is given by ${\bf A}=({\bf B}\times{\bf r})/2$, and the Hamiltonian (1.1) can be rewritten in the form $H=\frac{{\bf p}^{2}}{2m^{*}}-\frac{1}{2}\omega_{c}\hat{l}+\frac{1}{2}m^{*}\tilde{\omega}^{2}{\bf r}^{2},$ (1.2) where $\hat{l}=-\rmi\hbar(x\partial/\partial y-y\partial/\partial x)$ is the angular momentum operator of the electron (in the $z$ direction), $\omega_{c}=eB/(m^{*}c)$ is the cyclotron frequency, and $\tilde{\omega}=\sqrt{\omega_{0}^{2}+\omega_{c}^{2}/4}$ is the effective- confinement frequency. The eigenfunctions of the Hamiltonian (1.2) have the same functional form as those of a 2D harmonic oscillator at B=0, but with an effective frequency $\tilde{\omega}$, i.e., in polar coordinates $\phi_{n,l}(\rho,\theta)={\cal N}_{n,l}\rho^{|l|}\rme^{-\rho^{2}/2}\rme^{\rmi l\theta}L_{n}^{|l|}(\rho^{2}),$ (1.3) with $\rho=r/\tilde{l}$; the characteristic length $\tilde{l}$ is given by $\tilde{l}=\sqrt{\hbar/(m^{*}\tilde{\omega})}$. In (1.3), the index $n$ denotes the number of nodes in the radial direction, and $l$ (without any subscript or tilde) denotes the angular-momentum quantum numbers; the $L_{n}^{|l|}$’s are associated Laguerre polynomials. The single-particle energy spectrum corresponding to the Hamiltonian (1.2) is plotted in figure 36; the associated eigenenergies are given by $\frac{E_{n,l}}{\hbar\omega_{0}}=(2n+|l|+1)\sqrt{1+\frac{\eta^{2}}{4}}-\frac{l}{2}\eta,$ (1.4) with $\eta=\omega_{c}/\omega_{0}$. In the limit of $\omega_{c}/(2\omega_{0})\rightarrow\infty$, one can neglect the external confinement, and the energy spectrum in Equation (1.4) reduces to that of the celebrated Landau levels, i.e., $E_{\cal M}=\hbar\omega_{c}({\cal M}+\frac{1}{2}),$ (1.5) where ${\cal M}=n+(|l|-l)/2$ is the Landau-level index. We remark that the Landau levels are infinitely degenerate. The lowest Landau level ${\cal M}=0$ contains all nodeless levels ($n=0$) with arbitrary positive angular momentum $l\geq 0$. Figure 37: (Color online) The Darwin-Fock single-particle energy levels of a 2D harmonic oscillator rotating with angular frequency $\Omega$ as a function of $\eta=\Omega/\omega_{0}$, where $\omega_{0}$ is the frequency specifying the 2D harmonic confinement. A specific color (online) indicates orbitals with the same number of radial nodes. ### A.2 Two-dimensional rotating harmonic oscillator In the case of a rotating isotropic oscillator, instead of the expression (1.2), one has the following single-particle Hamiltonian: $H=\frac{{\bf p}^{2}}{2m}-\Omega\hat{l}+\frac{1}{2}m\omega_{0}^{2}{\bf r}^{2},$ (1.6) where the mass of the particle (e.g., a bosonic or fermionic atom) is denoted by $m$; $\Omega$ denotes the rotational frequency. From a comparison of the second terms in (1.2) and (1.6), one derives the correspondence $\Omega\rightarrow\omega_{c}/2$. We note that, unlike the application of a perpendicular magnetic field, the rotation does not generate an effective confinement different from the original external one [compare the third terms between (1.2) and (1.6)]. As a result, the eigenfunctions of the Hamiltonian (1.6) are given by the expressions (1.3), but with $\rho=r/l_{0}$ where the characteristic length $l_{0}=\sqrt{\hbar/(m\omega_{0})}$. The single-particle energy spectrum corresponding to the Hamiltonian (1.6) is plotted in figure 37 and the associated eigenenergies are given by $\frac{E_{n,l}}{\hbar\omega_{0}}=(2n+|l|+1)-l\eta,$ (1.7) with $\eta=\Omega/\omega_{0}$. For $\Omega/\omega_{0}=1$, the energy spectrum in Equation (1.7) reduces to that of the corresponding Landau levels, i.e., $E_{\cal M}=2\hbar\omega_{0}({\cal M}+\frac{1}{2}),$ (1.8) where ${\cal M}=n+(|l|-l)/2$ is the Landau-level index. As was the case with the perpendicular magnetic field, the Landau levels are infinitely degenerate, and the lowest Landau level ${\cal M}=0$ contains all nodeless levels ($n=0$) with arbitrary positive angular momentum $l\geq 0$. However, unlike the magnetic-field case where $\hbar\omega_{c}$ depends on $B$, the energy gap between the Landau levels in the case of rotation is independent of $\Omega$ and equals $2\hbar\omega_{0}$. ## References * [1] Ruder H, Wunner G, Herold H and Geyer F 1994 Atoms in Strong Magnetic Fields (Berlin: Springer Verlag) * [2] Kouwenhoven L P, Austing D G and Tarucha S 2001 Rep. Prog. Phys. 64 701 * [3] Reimann S M and Manninen M 2002 Rev. Mod. Phys. 74 1283 * [4] Dalfovo F, Giorgini S, Pitaevskii L P and Stringari S 1999 Rev. Mod. Phys. 71 463 * [5] Leggett A J 2001 Rev. Mod. Phys. 73 307 * [6] Bohigas O 1990 In Chaos and Quantum Physics edited by Giannoni M-J, Voros A and Zinn-Justin J (Amsterdam: North-Holland) p 87 * [7] Beenakker C W J 1997 Rev. Mod. Phys. 69 731 * [8] Alhassid Y 2000 Rev. Mod. Phys. 72 895 * [9] Anderson P W 1972 More Is Different: Broken symmetry and the hierarchical structure of science Science 177 393 * [10] Anderson P W 1984 Basic Notions of Condensed Matter Physics (Reading, MA: Addison-Wesley) pp 44 - 47 * [11] Bohr Å and Mottelson B R 1953 K. Danske Vidensk. Selsk. Mat.-Fys. Medd. 27 No 16 * [12] Nilsson S G 1955 K. Danske Vidensk. Selsk. Mat.-Fys. Medd. 29 No 16 * [13] Nazarewicz W 1994 Nucl. Phys. A 574 c27 * [14] Clemenger K L 1985 Phys. Rev. B 32 1359 * [15] Yannouleas C and Landman U 1995 Phys. Rev. B 51 1902 * [16] Yannouleas C, Landman U and Barnett R N 1999 In Metal Clusters edited by Ekardt W (Chichester: Wiley) p 145 * [17] Strutinsky V M 1967 Nucl. Phys. A95 420; Strutinsky V M 1968 Nucl. Phys. A122 1 * [18] Ring P and Schuck P 1980 The Nuclear Many-body Problem (New York: Springer-Verlag) * [19] Szabo A and Ostlund N S 1989 Modern Quantum Chemistry (New York: McGraw-Hill) * [20] Yannouleas C and Landman U 1999 Phys. Rev. Lett. 82 5325; 2000 Phys. Rev. Lett. 85 2220(E) * [21] Yannouleas C and Landman U 2000 Phys. Rev. B 61 15895 * [22] Yannouleas C and Landman U 2001 Eur. Phys. J. D 16 373 * [23] Yannouleas C and Landman U 2002 J. Phys.: Condens. Matter 14 L591 * [24] Yannouleas C and Landman U 2002 Phys. Rev. B 66 115315 * [25] Yannouleas C and Landman U 2003 Phys. Rev. B 68 035325 * [26] Yannouleas C and Landman U 2003 Phys. Rev. B 68 035326 * [27] Müller H-M and Koonin S E 1996 Phys. Rev. B 54 14532 * [28] Wigner E P 1934 Phys. Rev. 46 1002 * [29] Hirose K and Wingreen N S 1999 Phys. Rev. B 59 4604 * [30] Peierls R E and Yoccoz J 1957 Proc. Phys. Soc. (London) A70 381 * [31] Löwdin P-O 1955 Phys. Rev. 97 1509 * [32] Löwdin P-O 1964 Rev. Mod. Phys. 36 966 * [33] Fukutome H 1981 Int. J. Quantum Chem. 20 955, and references therein * [34] Rodriguez T R, Egido J L and Robledo L M 2005 Phys. Rev. C 72 064303 * [35] Schmid K W 2004 Prog. Part. Nucl. Phys. 52 565 * [36] Bender M, Flocard H and Heenen P H 2003 Phys. Rev. C 68 044321 * [37] Stoitsov M V, Dobaczewski J, Nazarewicz W, Pittel S and Dean D J 2003 Phys. Rev. C 68 054312 * [38] Bender M, Heenen P H and Reinhard P-G (2003) Rev. Mod. Phys. 75 121 * [39] Reinhard P-G, Nazarewicz W, Bender M and Maruhn J A 1996 Phys. Rev. C 53 2776 * [40] Fernandez M A and Egido J L 2003 Phys. Rev. B 68 184505 * [41] Ellenberger C, T. Ihn T, Yannouleas C, Landman U, Ensslin K, Driscoll D and Gossard A C 2006 Phys. Rev. Lett. 96 126806 * [42] Yannouleas C and Landman U 2006 phys. stat. sol. (a) 203 1160 * [43] Yannouleas C and Landman U 2006 Proc. Natl. Acad. Sci. USA 103 10600 * [44] Ihn T, Ellenberger C, Ensslin K, Yannouleas C, Landman U, Driscoll D C and Gossard A C 2007 Int. J. Mod. Phys. B 21 1316 * [45] Zumbühl D M, Marcus C M, Hanson M P and Gossard A C 2004 Phys. Rev. Lett. 93 256801 * [46] Yannouleas C and Landman U 2002 Int. J. Quantum Chem. 90 699 * [47] Loss D and D.P. DiVincenzo D P 1998 Phys. Rev. A 57 120 * [48] Burkard G, Loss D and DiVincenzo D P 1999 Phys. Rev. B 59 2070 * [49] Taylor J M, Engel H A, Dür W, Yacoby A, Marcus C M, Zoller P and Lukin M D 2005 Nature Phys. 1 177 * [50] Yannouleas C and Landman U 2000 Phys. Rev. Lett. 85 1726 * [51] Yannouleas C and Landman U 2004 Phys. Rev. B 69 113306 * [52] Yannouleas C and Landman U 2004 Phys. Rev. B 70 235319 * [53] Li Y S, Yannouleas C and Landman U 2006 Phys. Rev. B 73 075301 * [54] Leggett T 2004 Science 305 1921 * [55] Laughlin R B 1983 Phys. Rev. Lett. 50 1395 * [56] Jain J K 1989 Phys. Rev. Lett. 63 199 * [57] Jain J K 1990 Phys. Rev. B 41 7653 * [58] Yoshioka D 2002 The Quantum Hall Effect (Berlin: Springer) * [59] Jain J K 2007 Composite Fermions (New York: Cambridge University Press) * [60] Romanovsky I, Yannouleas C and Landman U 2004 Phys. Rev. Lett. 93 230405 * [61] Tonks L 1936 Phys. Rev. 50 955 * [62] Girardeau M 1960 J. Math. Phys. (N.Y.) 1 516 * [63] Romanovsky I, Yannouleas C, Baksmaty L O and Landman U 2006 Phys. Rev. Lett. 97 090401 * [64] Laughlin R B 1999 Rev. Mod. Phys. 71 863 * [65] Landman U 2005 Proc. Natl. Acad. Sci. USA 102 6671 * [66] Löwdin P-O 1959 Adv. Chem. Phys. vol II p 207 * [67] Wilkin N K and Gunn J M F 2000 Phys. Rev. Lett. 84 6 * [68] Viefers S, Hansson T H and Reimann S M 2000 Phys. Rev. A 62 053604 * [69] Cooper N R, Wilkin N K and Gunn J M F 2001 Phys. Rev. Lett. 87 120405 * [70] Regnault N and Jolicoeur Th 2003 Phys. Rev. Lett. 91 030402 * [71] Chang C-C, Regnault N, Jolicoeur Th and Jain J K 2005 Phys. Rev. A 72 013611 * [72] Baksmaty L O, Yannouleas C and Landman U 2007 Phys. Rev. A 75 023620 * [73] Tarucha S, Austing D G, Honda T, van der Hage R J, Kouwenhoven L P 1996 Phys. Rev. Lett. 77 3613 * [74] Nishi Y, Maksym P A, Austing D G, Hatano T, Kouwenhoven L P, Aoki H, Tarucha S 2006 Phys. Rev. B 74 033306 * [75] Nishi Y, Tokura Y, Gupta J, Austing G and Tarucha S 2007 Phys. Rev. B 75 121301 * [76] De Giovannini U, Cavaliere F, Cenni R, Sassetti M and Kramer B 2007 New J. Phys. 9 93 * [77] Rontani M, Cavazzoni C, Bellucci D and Goldoni G 2006 J. Chem. Phys. 124 124102 * [78] Ghosal A, Güçlü A D, Umrigar C J, Ullmo D and Baranger H U 2006 Nature Phys. 2 336 * [79] Ciorga M, Sachrajda A S, Hawrylak P, Gould C, Zawadzki P, Jullian S, Feng Y and Wasilewski Z 2000 Phys. Rev. B 61 R16315 * [80] Kyriakidis J, Pioro-Ladriere M, Ciorga M, Sachrajda A S and Hawrylak P 2002 Phys. Rev. B 66 035320 * [81] Hatano T, Stopa M and Tarucha S 2005 Science 309 268 * [82] Heitler H and London F 1927 Z. Phys. 44 455 * [83] Avinun-Kalish M, Heiblum M, Zarchin O, Mahalu D and Umansky V 2005 Nature 436 529 * [84] Gurvitz S A 2007 arXiv:0704.1260v1 * [85] Kinoshita G T, Wenger T and and Weiss D S 2004 Science 305 1125 * [86] Paredes B, Widera A, Murg V, Mandel O, Folling S, Cirac I, Shlyapnikov G V, Hansch T W and Bloch I 2004 Nature (London) 429 277 * [87] Greiner M, Mandel O, Esslinger T, Hansch T W and Bloch I 2002 Nature (London) 415 39 * [88] Mueller E J, Ho T L, Ueda M and Baym G 2006 Phys. Rev. A 74 033612 * [89] Zhang L-X, Melnikov D V and Leburton J-P 2006 Phys. Rev. B 74 205306 * [90] Hallam L D, Weis J and Maksym P A 1996 Phys. Rev. B 53 1452 * [91] Koskinen M, Manninen M and Reimann S M 1997 Phys. Rev. Lett. 79 1389 * [92] Lee I H, Rao V, Martin R M and Leburton J-P 1998 Phys. Rev. B 57 9035 * [93] Barnett R N, Cleveland C L, Hakkinen H, Luedtke W D, Yannouleas C and Landman U 1999 Eur. Phys. J. D 9 95 * [94] Matagne P, Leburton J-P, Austing D G and Tarucha S 2002 Phys. Rev. B 65 085325 * [95] Saarikoski H, Harju A, Puska M J and Nieminen R M 2004 Phys. Rev. Lett. bf 93 116802 * [96] Borgh M, Toreblad M, Koskinen M, Manninen M, Åberg S and Reimann S M 2005 Int. J. Quantum Chem. 105 817 * [97] Pfannkuche D, Gudmundsson V and Maksym P A 1993 Phys. Rev. B 47 2244 * [98] Fujito M, Natori A and Yasunaga H 1996 Phys. Rev. B 53 9952 * [99] Bednarek S, Szafran B and Adamowski J 1999 Phys. Rev. B 59 013036 * [100] Rontani M, Rossi F, Manghi F and Molinari E 1999 Phys. Rev. B 59 010165 * [101] Reusch B, Häusler W and Grabert H 2001 Phys. Rev. B 63 113313 * [102] Szafran B, Bednarek S and Adamowski J 2003 Phys. Rev. B 67 045311 * [103] Szafran B, Bednarek S, Adamowski J, Tavernier M B, Anisimovas E and Peeters F M 2004 Eur. Phys. J. D 28 373 * [104] Emperador A, Lipparini E and Serra L 2006 Phys. Rev. B 73 235341 * [105] Lipparini E 2003 Modern Many-Particle Physics: Atomic Gases, Quantum Dots and Quantum Fluids (London: World Scientific) * [106] Pople J A and Nesbet R K 1954 J. Chem. Phys. 22 571 * [107] Overhauser A W 1960 Phys. Rev. Lett. 4 415 * [108] Thouless D J 1961 The Quantum Mechanics of Many-Body Systems (New York: Academic Press) * [109] Adams W H 1962 Phys. Rev. 127 1650 * [110] Čížek J and Paldus J 1967 J. Chem. Phys. 47 3976 * [111] Carbó R and Klobukowski M 1990 Self-Consistent Field: Theory and Applications (Amsterdam: Elsevier Science Publishers) * [112] Lozovik Yu E 1987 Usp. Fiz. Nauk 153 356 [1987 Sov. Phys. Usp. 30 912] * [113] Bedanov V M and Peeters F M 1994 Phys. Rev. B 49 2667 * [114] Kong M, Partoens B and Peeters F M 2002 Phys. Rev. E 65 046602 * [115] Overhauser A W 1962 Phys. Rev. 128 1437 * [116] Wigner E P 1959 Group Theory and its Application to the Quantum Mechanics of Atomic Spectra (New York: Academic Press) * [117] Hashimoto K 1982 Int. J. Quantum Chem. 22 397 * [118] Igawa A 1995 Int. J. Quantum Chem. 54 235 * [119] Pauncz R 2000 The Construction of Spin Eigenfunctions: An Exercise Book (New York: Kluwer/Plenum) * [120] Goddard W A III, Dunning T H Jr, Hunt W J and Hay P J 1973 Accounts Chem. Res. 6 368 * [121] Gerratt J and Lipscomb W N 1968 Proc. Natl. Acad. Sci. USA 59 332 * [122] De Giovannini U, Cavaliere F, Cenni R, Sassetti M and Kramer B 2007 arXiv:0707.3705v1 * [123] Zduńczuk H, Dobaczewski J and Satuła W 2007 Int. J. Mod. Phys. E 16 377 * [124] Kamlah A 1968 Z. Phys. 216 52 * [125] Yoshioka D J and Lee P A 1983 Phys. Rev. B 27 4986 * [126] MacDonald A H, Yang S R E and Johnson M D 1993 Aust. J. Phys. 46 345 * [127] Ruan W Y, Liu Y Y, Bao C G and Zhang Z Q 1995 Phys. Rev. B 51 7942 * [128] Maksym P A 1996 Phys. Rev. B 53 10871 * [129] Girvin S M and Jach T 1983 Phys. Rev. B 28 4506 * [130] Maksym P A and Chakraborty T 1990 Phys. Rev. Lett. 65 108 * [131] Seki T, Kuramoto Y and Nishino T 1996 J. Phys. Soc. Jpn. 65 3945 * [132] Maksym P A, Imamura H, Mallon G P and Aoki H 2000 J. Phys.: Conden. Matter 12 R299 * [133] Jacak L, Hawrylak P and Wojs A 1998 Quantum Dots (Berlin: Springer), and references therein. * [134] Melnikov D V, Matagne P, Leburton J-P, Austing D G, Yu G, Tarucha S, Fettig J and Sobh N 2005 Phys. Rev. B 72 085331 * [135] Melnikov D V and Leburton J-P 2006 Phys. Rev. B 73 155301 * [136] Perdew J P, Savin A and Burke K 1995 Phys. Rev. A 51 4531 * [137] Savin A 1996 In Recent Developments and Applications of Modern Density Functional Theory edited by Seminario J M (Amsterdam: Elsevier) p 327 * [138] Bauernschmitt R and Ahlrichs R J 1996 Chem. Phys. 104 9047 * [139] Abedinpour S H, Polini M, Xianlong G and Tosi M P 2007 Eur. Phys. J. B 56 127 * [140] Laegsgaard J and Stokbro K 2001 Phys. Rev. Lett. 86 2834 * [141] Savrasov S Y, Kotliar G and Abrahams E 2001 Nature (London) 410 793 * [142] Adamo C, Barone V, Bencini A, Broer R, Filatov M, Harrison N M, Illas F, Malrieu J P and Moreira I D R 2006 J. Chem. Phys. 124 107101 * [143] Rong C, Lian S, Yin D, Shen B, Zhong A, Bartolotti L and Liu S 2006 J. Chem. Phys. 125 174102 * [144] Illas F, Moreira I D R, Bofill J M and Filatov M 2006 Theor. Chem. Acc. 116 587 * [145] Thouless D J and Valatin J G 1962 Nucl. Phys. 31 211 * [146] Bertsch G F and Hagino K 2001 Yad. Fiz. 64 646 [2001 Phys. Part. Nucl. 64 588] * [147] Nazmitdinov R G, Almehed D and Dönau F 2002 Phys. Rev. C 65 041307 * [148] Serra Ll, Nazmitdinov R G and Puente A 2003 Phys. Rev. B 68 035341 * [149] Hill D L and Wheeler J A 1953 Phys. Rev. 89 1102 * [150] Griffin J J and Wheeler J A 1957 Phys. Rev. 108 311 * [151] Öhberg P and Stenholm S 1998 Phys. Rev. A 57 1272 * [152] Esry B D and Greene C H 1999 Phys. Rev. A 59 1457 * [153] Alon O E, Streltsov A I and Cederbaum L S 2005 Phys. Rev. Lett. 95 030405 * [154] Esry B D 1997 Phys. Rev. A 55 1147 * [155] Romanovsky I A 2006 Ph. D. Thesis, Georgia Institute of Technology, Atlanta, http://etd.gatech.edu/theses/available/etd-07102006-041659. * [156] Cederbaum L S and Streltsov A I 2003 Phys. Lett. A 318 564 * [157] Hurley A C 1976 Introduction to the Electron Theory of Small Molecules (London: Academic Press) * [158] Tsiper E V 2002 J. Math. Phys. 43 1664 * [159] Bertsch G F and Papenbrock T 1999 Phys. Rev. Lett. 83 5412 * [160] Lehoucq R B, Sorensen D C and Yang C 1998 ARPACK Users’ Guide: Solution of Large-Scale Eigenvalue Problems with Implicitly Restarted Arnoldi Methods (Philadelphia: SIAM) * [161] Tsiper E V and Goldman V J 2001 Phys. Rev. B 64 165311 * [162] Darwin C G 1930 Proc. Cambridge Philos. Soc. 27 86 * [163] Fock V 1928 Z. Phys. 47 446 * [164] Anisimovas E and Matulis A 1998 J. Phys.: Condens. Matter 10 601 * [165] Eto M 1997 Jpn. J. Appl. Phys. 36 3924 * [166] Hawrylak P and Pfannkuche D 1993 Phys. Rev. Lett. 70 485 * [167] Mikhailov S A 2002 Phys. Rev. B 65 115312 * [168] Mikhailov S A and Savostianova N A 2002 Phys. Rev. B 66 033307 * [169] Tavernier M B, Anisimovas E, Peeters F M, Szafran B, Adamowski J and Bednarek S 2003 Phys. Rev. B 68 205305 * [170] Wensauer A, Korkusinski M and Hawrylak P 2004 Solid State Commun. 130 115 * [171] Ruan W Y and Cheung H-F 1999 J. Phys.: Condens. Matter 11 435 * [172] Creffield C E, Hausler W, Jefferson J H and Sarkar S 1999 Phys. Rev. B 59 10719 * [173] Laughlin R B 1983 Phys. Rev. B 27 3383 * [174] Kawamura T and Jain J K 1996 J. Phys.: Condens. Matter 8 2095 * [175] Reimann S M, Koskinen M and Manninen M 2000 Phys. Rev. B 62 8108 * [176] Jeon G S, Chang C-C and Jain J K 2007 Eur. Phys. J. B 55 271 * [177] Barberan N, Lewenstein M, Osterloh K and Dagnino D 2006 Phys. Rev. A 73 063623 * [178] Reimann S M, Koskinen M, Yu Y and Manninen M 2006 New J. Phys. 8 59 * [179] Wagner M, Merkt U and Chaplik A V 1992 Phys. Rev. B 45 R1951 * [180] Drouvelis P S, Schmelcher P and Diakonos F K 2004 Phys. Rev. B 69 035333 * [181] Harju A, Siljamäki S and Nieminen R M 2002 Phys. Rev. Lett. 88 226804 * [182] Klironomos A D, Meyer J S and Matveev K A 2006 Europhys. Lett. 74 679 * [183] Piacente G, Schweigert I V, Betouras J J and Peeters F M 2004 Phys. Rev. B 69 045324 * [184] Li Y, Yannouleas C and Landman U 2007 to be published * [185] Dür W, Vidal G and Cirac J I 2000 Phys. Rev. A 62 062314 * [186] Coffman V, Kundu J and Wootters W K 2000 Phys. Rev. A 61 052306 * [187] Ohno K, Esfarjani K and Kawazoe Y 1999 Computational Materials Science: From Ab Initio to Monte Carlo Methods (New York: Springer) * [188] Egger R, Häusler W, Mak C H and Grabert H 1999 Phys. Rev. Lett. 82 3320; 1999 Phys. Rev. Lett. 83 462(E) * [189] Filinov A V, Lozovik Yu E and Bonitz M 2000 Phys. Status Solidi B 221 231 * [190] Harting J, Mülken and Borrmann P 2000 Phys. Rev. B 62 10207 * [191] Reusch B and Egger R 2003 Europhys. Lett. 64 84 * [192] Weiss S and Egger R 2005 Phys. Rev. B 72 245301 * [193] Tanatar B and Ceperley D M 1989 Phys. Rev. B 39 5005 * [194] Pederiva F, Umrigar C J and Lipparini E 2000 Phys. Rev. B 62 8120 * [195] Harju A, Siljamäki S and Nieminen R M 2002 Phys. Rev. B 65 075309 * [196] Güçlü A D and Umrigar C J 2005 Phys. Rev. B 72 045309 * [197] Ghosal A, Güçlü A D, Umrigar C J, Ullmo D and Baranger H U 2007 arXiv:cond-mat/0702666v1 * [198] Mora Ch and Waintal X 2007 Phys. Rev. Lett. 99 030403 * [199] Salis G, Kato Y, Ensslin K, Driscoll D C, Gossard A C and Awschalom D D 2001 Nature 414 619 * [200] Paskauskas R and You L 2001 Phys. Rev. A 64 042310 * [201] Schliemann J, Cirac J, Kus M, Lewenstein M and Loss D 2001 Phys. Rev. A 64 022303 * [202] Kellman M E and Herrick D R 1980 Phys. Rev. A 22 1536 * [203] Kellman M E 1997 Int. J. Quantum Chem. 65 399 * [204] Berry R S 1989 Contemp. Phys. 30 1 * [205] Sommerfeld A 1923 Rev. Sci. Inst. 7 509 * [206] Pais A 1991 Niels-Bohr’s Times (Oxford: Clarendon) * [207] Van Vleck J H 1922 Phil. Mag. 44 842 * [208] Born M 1927 The Mechanics of the Atom (New York: Ungar) * [209] Leopold J G and Percival I C 1980 J. Phys. B 13 1037 * [210] Wintgen D, Richter K and Tanner G 1992 Chaos 2 19 * [211] Ezra G S and Berry R S 1983 Phys. Rev. A 28 1974 * [212] Watanabe S and Lin C D 1986 Phys. Rev. A 34 823 * [213] Creffield C E, Häusler W, Jefferson J H and Sarkar S 1999 Phys. Rev. B 59 10719 * [214] Häusler W, Reusch B, Egger R and Grabert H 2000 Physica B 284 1772 * [215] Matulis A and Peeters F M 2001 Solid State Commun. 117 655 * [216] Filinov A V, Bonitz M and Lozovik Yu E 2001 Phys. Rev. Lett. 86 3851 * [217] Sundqvist P A, Volkov S Y, Lozovik Yu E and Willander M 2002 Phys. Rev. B 66 075335 * [218] Mikhailov S A and Ziegler K 2002 Eur. Phys. J. B 28 117 * [219] Mikhailov S A 2002 Physica E 12 884 * [220] Creffield C E, Jefferson J H, Sarkar S and Tipton D L J 2000 Phys. Rev. B 62 7249 * [221] Rontani M, Goldoni G, Manghi F and Molinari E 2002 Europhys. Lett. 58 555 * [222] Bolton F and Rössler U Superlatt. Microstruct. 13 139 * [223] Jain J K and Kamilla R K 1997 Int. J. Mod. Phys. B 11 2621 * [224] Goldmann E and Renn S R 1997 Phys. Rev. B 56 13296 * [225] Eric Yang S-R and Han J H 1998 Phys. Rev. B 57 R12681 * [226] Taut M 2000 J. Phys.: Condens. Matter 12 3689 * [227] Manninen M, Viefers S, Koskinen M and Reimann S M 2001 Phys. Rev. B 64 245322 * [228] Wan X, Yang K and Rezayi E H 2002 Phys. Rev. Lett. 88 056802 * [229] Jain J K and Kawamura T 1995 Europhys. Lett. 29 321 * [230] Maki K and Zotos X 1983 Phys. Rev. B 28 4349 * [231] Wunsch B, Stauber T and Guinea F 2007 arXiv:0707.2948v2 * [232] Dai Z, Zhu J-L, Yang N and Wang Y 2007 arXiv:cond-mat/0702256v2 * [233] Shi C, Jeon G S and Jain J K 2007 Phys. Rev. B 75 165302 * [234] Jeon G S, Chang C-C and Jain J K 2004 J. Phys.: Condens. Matter 16 L271 * [235] Chang C-C, Jeon G S and Jain J K 2005 Phys. Rev. Lett. 94 016809 * [236] Laughlin R B 1987 In The Quantum Hall Effect Edited by Prange R E and Girvin S M (New York: Springer) p 233 * [237] Jeon G S, Chang C-C and Jain J K 2004 Phys. Rev. B 69 R241304 * [238] Lam P K and Girvin S M 1984 Phys. Rev. B 30 473 * [239] Tavernier M B, Anisimovas E and Peeters F M 2004 Phys. Rev. B 70 155321 * [240] Yang S R E, MacDonald A H and Johnson M D 1993 Phys. Rev. Lett. 71 3194 * [241] Löwdin P-O 1962 Rev. Mod. Phys. 34 520 * [242] Toreblad M, Borgh M, Koskinen M, Manninen M and Reimann S M 2004 Phys. Rev. Lett. 93 090407 * [243] Toreblad M, Yu Y, Reimann S M, Koskinen M and Manninen M 2006 J. Phys. B 39 2721 * [244] Pan W, Stormer H L, Tsui D C, Pfeiffer L N, Baldwin K W and West K W 2002 Phys. Rev. Lett. 88 176802 * [245] Pfeiffer L and West K W 2003 Physica E 20 57 * [246] Cornish S L, Claussen N R, Roberts J L, Cornell E A and Wieman C E 2000 Phys. Rev. Lett. 85 1795 * [247] Girardeau M D and Wright E M 2002 Laser Physics 12 8 * [248] Dunjko V, Lorent V and Olshanii M 2001 Phys. Rev. Lett. 86 5413 * [249] Kartsev P F 2003 Phys. Rev. A 68 063613 * [250] Butts R A and Rokhsar D S 1999 Nature (London) 397 327 * [251] Schellekens M, Hoppeler R, Perrin A, Gomes J V, Boiron D, Aspect A and Westbrook C I 2005 Science 310 648 * [252] Ho T L 2001 Phys. Rev. Lett. 87 060403 * [253] Fetter A L 2001 Phys. Rev. A 64 063608 * [254] Pethick C J and Smith H 2002 Bose-Einstein Condensation in Dilute Gases (Cambridge: Cambridge University Press) * [255] Pitaevskii L P and Stringari S 2003 Bose-Einstein Condensation (Oxford: Oxford University Press) * [256] Watanabe G, Baym G and Pethick C J 2004 Phys. Rev. Lett. 93 190401 * [257] Baym G 2005 J. Low Temp. Phys. 138 601 * [258] Madison K W, Chevy F, Wohlleben W and Dalibard J 2000 Phys. Rev. Lett. 84 806 * [259] Raman C, Abo-Shaeer J R, Vogels J M, Xu K and Ketterle W 2001 Phys. Rev. Lett. 87 210402 * [260] Engels P, Coddington I, Haljan P C and Cornell E A 2002 Phys. Rev. Lett. 89 100403 * [261] Griesmaier A, Werner J, Hensler S, Stuhler J and Pfau T 2005 Phys. Rev. Lett. 94 160401 * [262] Baranov M, Dobrek Ł, Góral K, Santos L and Lewenstein M 2002 Physica Scripta 102 74 * [263] Altman E, Demler E and Lukin M D 2004 Phys. Rev. A 70 013603 * [264] Fölling S, Gerber F, Widera A, Mandel O, Gericke T and Bloch I 2005 Nature (London) 434 481

arxiv-papers

2024-09-04T02:48:53.310711

0711.0642

††thanks: This work is supported by the NWO VICI grant 639.043.201 “Optical Interferometry: A new Method for Studies of Extrasolar Planets” to A. Quirrenbach. # Geodetic Line at Constant Altitude above the Ellipsoid Richard J. Mathar mathar@strw.leidenuniv.nl http://www.strw.leidenuniv.nl/~mathar Leiden Observatory, Leiden University, P.O. Box 9513, 2300 RA Leiden, The Netherlands ###### Abstract The two-dimensional surface of a bi-axial ellipsoid is characterized by the lengths of its major and minor axes. Longitude and latitude span an angular coordinate system across. We consider the egg-shaped surface of constant altitude above (or below) the ellipsoid surface, and compute the geodetic lines—lines of minimum Euclidean length—within this surface which connect two points of fixed coordinates. This addresses the common “inverse” problem of geodesics generalized to non-zero elevations. The system of differential equations which couples the two angular coordinates along the trajectory is reduced to a single integral, which is handled by Taylor expansion up to fourth power in the eccentricity. geodesy; ellipsoid; eccentricity; geodetic line; inverse problem ###### pacs: 91.10.By, 91.10.Ws, 02.40.Hw ## I Contents The common three parameters employed to relate Cartesian coordinates to an ellipsoidal surface are the angles of latitude and longitude in a grid on the surface, plus an altitude which is a shortest (perpendicular) distance to the surface. The well-known functional relations (coordinate transformations) are summarized in Section II. The inverse problem of geodesy is to find the line embedded in the ellipsoid surface which connects two fixed points subject to minimization of its length. We pose the equivalent problem for lines at constant altitude, as if one would ask for the shortest track of the center of a sphere of given radius which rolls on the ellipsoid surface and meets two points of the same, known altitude. In Section III, we formulate this in terms of the generic differential equations of geodesy parametrized by the Christoffel symbols. In the main Section IV, the coupled system of differential equations of latitude and longitude as a function of path length is reduced to one degree of freedom, here chosen to be the direction along the path at one of the fixed terminal points, measured in the topocentric horizontal system, and dubbed the launching angle. Closed-form expressions of this parameter in terms of the coordinates of the terminal points have not been found; instead, the results are presented as series expansions up to fourth power in the eccentricity. The standard treatment of this analysis is the projection on an auxiliary sphere; this technique is (almost) completely ignored but for the pragmatic aspect that the case of zero eccentricity is a suitable zeroth-order reference of series expansions around small eccentricities. ## II Spheroidal Coordinates ### II.1 Surface The cross section of an ellipsis of equatorial radius $\rho_{e}>\rho_{p}$ with eccentricity $e$ in a Cartesian ($x,z$) system is: $\rho_{p}^{2}=\rho_{e}^{2}(1-e^{2});\quad\frac{x^{2}}{\rho_{e}^{2}}+\frac{z^{2}}{\rho_{p}^{2}}=1.$ (1) The ellipsis defines a geocentric latitude $\phi^{\prime}$ and a geodetic latitude $\phi$, the latter measured by intersection of the normal to the tangential plane with the equatorial plane (Figure 1). Figure 1: Major semi-axis $\rho_{e}$, minor semi-axis $\rho_{p}$, straight distance to the center of coordinates $\rho$, geocentric latitude $\phi^{\prime}$, geodetic latitude $\phi$, and their pointing difference $v$. On the surface of the ellipsoid (Smart, 1949, §IX)Dufour (1958)(Bouasse, 1919, §140): $\tan\phi=\frac{z}{x}\frac{\rho_{e}^{2}}{\rho_{p}^{2}};\quad\frac{z}{x}=\tan\phi^{\prime}=\frac{\rho_{p}^{2}}{\rho_{e}^{2}}\tan\phi.$ (2) Supposed $\rho$ denotes the distance from the center of coordinates to the point on the geoid surface, the transformation from $(x,z)$ to $\phi$ is $x=\rho\cos\phi^{\prime}=\frac{\rho_{e}\cos\phi}{\sqrt{1-e^{2}\sin^{2}\phi}},\quad z=\rho\sin\phi^{\prime}=\frac{\rho_{e}(1-e^{2})\sin\phi}{\sqrt{1-e^{2}\sin^{2}\phi}}.$ (3) $v=\phi-\phi^{\prime}$ defines the difference between the geodetic (astronomical) and the geocentric latitudes, $\tan v=\frac{e^{2}\sin(2\phi)}{2(1-e^{2}\sin^{2}\phi)}=\frac{m\sin(2\phi)}{1+m\cos(2\phi)}=m\sin(2\phi)-m^{2}\sin(2\phi)\cos(2\phi)+m^{3}\sin(2\phi)\cos^{2}(2\phi)+\cdots;$ (4) where the expansion of the denominator has been given in terms of the geometric series of the parameter $m\equiv\frac{e^{2}}{2-e^{2}}.$ (5) The Taylor series in powers of $\sin(2\phi)$ and in powers of $m$ are $\displaystyle v$ $\displaystyle=$ $\displaystyle\frac{m}{1+m}\sin(2\phi)+\frac{m^{2}(2+m)}{6(1+m)^{3}}\sin^{3}(2\phi)+\frac{m^{2}(5+25m+15m^{2}+3m^{3})}{40(1+m)^{5}}\sin^{5}(2\phi)+\cdots$ (6) $\displaystyle=$ $\displaystyle\sin(2\phi)m-\frac{1}{2}\sin(4\phi)m^{2}+\frac{1}{3}\sin(6\phi)m^{3}-\frac{1}{4}\sin(8\phi)m^{4}+\frac{1}{5}\sin(10\phi)m^{5}+\cdots.$ (7) A flattening factor $f$ is also commonly defined Kaplan (2006), $f=\frac{\rho_{e}-\rho_{p}}{\rho_{e}}=1-\sqrt{1-e^{2}};\quad e^{2}=f(2-f).$ (8) The reference values of the Earth ellipsoid adopted in the WGS84 National Imagery and Mapping Agency (2000) are $f=1/298.257223563,\quad\rho_{e}=6378137.0\text{ m}.$ (9) The numerical evaluation of (6) in terms of a Fourier series with this parametrization is, in units of radians and arcseconds: $\displaystyle v$ $\displaystyle=$ $\displaystyle 0.0033584338\sin(2\phi)-0.56395388\times 10^{-5}\sin(4\phi)+0.12626678\times 10^{-7}\sin(6\phi)+\cdots$ (10) $\displaystyle=$ $\displaystyle 692.^{\prime\prime}72669\sin(2\phi)-1.^{\prime\prime}16324\sin(4\phi)+0.^{\prime\prime}00260\sin(6\phi)+\cdots.$ (11) Slightly different values would emerge according to IERS conventions of 2003 McCarthy and Petit (2003). ### II.2 General Altitude If one moves along the direction of $\phi$ a distance $h$ away from the surface of the geoid, the new coordinates relative to (3) are Fukushima (2006); Vermeille (2004); Jones (2004); Pollard (2002); Zhang _et al._ (2005); Hradilek (1976) $x=\rho\cos\phi^{\prime}+h\cos\phi;\quad z=\rho\sin\phi^{\prime}+h\sin\phi,$ (12) which can be written in terms of a distance $N(\phi)$, $N(\phi)\equiv\frac{\rho_{e}}{\sqrt{1-e^{2}\sin^{2}\phi}}$ (13) as $x=(N+h)\cos\phi;\quad z=[N(1-e^{2})+h]\sin\phi.$ (14) Figure 2: A point with Cartesian coordinates $(x,y,z)$ has a distance $z$ to the equatorial plane, a distance $\sqrt{x^{2}+y^{2}}$ to the polar axis, a distance $h$ to the surface of the ellipsoid, and a distance $N+h$ to the polar axis, measured along the local normal to the surface Long (1975). The vector ${\mathbf{e}}_{\phi}$ points North at this point. Rotation of (14) around the polar axis with geographic longitude $\lambda$ defines the full 3D transformation between $(x,y,z)$ and $(\lambda,\phi,h)$, $\left(\begin{array}[]{c}x\\\ y\\\ z\\\ \end{array}\right)=\left(\begin{array}[]{c}\left[N(\phi)+h\right]\cos\phi\cos\lambda\\\ \left[N(\phi)+h\right]\cos\phi\sin\lambda\\\ \left[N(\phi)(1-e^{2})+h\right]\sin\phi\\\ \end{array}\right).$ (15) The only theme of this paper is to generalize the geodetic lines of the literature Dufour (1958); Sodano (1958); Saito (1970); Thomas (1965); Ölander (1952); Bowring (1983) to the case of finite altitude $h\neq 0$. The physics of gravimetric or potential theory is not involved, only the mathematics of the geometry. It should be noted that points with constant, non-zero $h$ do _not_ define a surface of an ellipsoid with effective semi-axes $\rho_{e,p}+h$—otherwise the geodetic line could be deduced by mapping the problem onto an equivalent ellipsoidal surface Dufour (1958). ## III Inverse Problem of Geodesy ### III.1 Topocentric Coordinate System A line of shortest distance at constant height $h=$const between two points 1 and 2 is defined by minimizing the Euclidean distance, the line integral $S=\int_{1}^{2}\sqrt{dx^{2}+dy^{2}+dz^{2}}=\int_{1}^{2}\sqrt{ds^{2}}\equiv\int_{1}^{2}{\cal L}ds$ (16) for some parametrization $\lambda(\phi)$. The integrand is equivalent to a Lagrange function ${\cal L}(\phi,\lambda,d\lambda/d\phi)$ or ${\cal L}(\phi,\lambda,d\phi/d\lambda)$. The gradient of of (15) with respect to $\lambda$ and $\phi$ defines the vectors ${\mathbf{e}}_{\lambda,\phi}$ that span the topocentric tangent plane ${\bf e}_{\lambda}=\left(\begin{array}[]{c}-\left[N(\phi)+h\right]\cos\phi\sin\lambda\\\ \left[N(\phi)+h\right]\cos\phi\cos\lambda\\\ 0\\\ \end{array}\right);\quad{\bf e}_{\phi}=\left(\begin{array}[]{c}-\sin\phi\cos\lambda\left[M(\phi)+h\right]\\\ -\sin\phi\sin\lambda\left[M(\phi)+h\right]\\\ \cos\phi\left[M(\phi)+h\right]\\\ \end{array}\right),$ (17) where a meridional radius of curvature Tienstra (1951) $M(\phi)\equiv\frac{\rho_{e}(1-e^{2})}{(1-e^{2}\sin^{2}\phi)^{3/2}}=N(\phi)\frac{1-e^{2}}{1-e^{2}\sin^{2}\phi}$ (18) is defined to simplify the notation. Building squares and dot products computes the three Gauss Fundamental parameters of the surface Reichardt (1957) $\displaystyle{\bf e}_{\lambda}^{2}=E$ $\displaystyle=$ $\displaystyle(N+h)^{2}\cos^{2}\phi;$ (19) $\displaystyle{\bf e}_{\lambda}\cdot{\bf e}_{\phi}=F$ $\displaystyle=$ $\displaystyle 0;$ (20) $\displaystyle{\bf e}_{\phi}^{2}=G$ $\displaystyle=$ $\displaystyle\left[N\frac{1-e^{2}}{1-e^{2}\sin^{2}\phi}+h\right]^{2}=(M+h)^{2}.$ (21) Specializing to $h=0$ we get the You formulae You (1999). $E$ and $G$ provide the principal curvatures along the meridian and azimuth Dorrer (1999), and the coefficients of the metric tensor in the quadratic form of $ds^{2}$, $S=\int\sqrt{Ed\lambda^{2}+2Fd\lambda d\phi+Gd\phi^{2}}.$ (22) ### III.2 Christoffel Symbols Christoffel symbols are the connection coefficients between differentials $d{\bf e}_{\epsilon}$ of the topocentric axis and of the positions $d{\bf x}^{\beta}$, in a generic definition $d{\bf e}_{\epsilon}=\sum_{\alpha,\beta}{\bf e}_{\alpha}\Gamma_{\beta\epsilon}^{\alpha}d{\bf x}^{\beta}.$ (23) This format is matched by first computing the derivative of ${\bf e}_{\lambda}$ with respect to $\lambda$ and $\phi$ (at $h=const$): $d{\bf e}_{\lambda}=\left(\begin{array}[]{c}-(N+h)\cos\phi\cos\lambda\\\ -(N+h)\cos\phi\sin\lambda\\\ 0\\\ \end{array}\right)d\lambda+\left(\begin{array}[]{c}(M+h)\sin\phi\sin\lambda\\\ -(M+h)\sin\phi\cos\lambda\\\ 0\\\ \end{array}\right)d\phi$ (24) and of ${\bf e}_{\phi}$ with respect to $\lambda$ and $\phi$: $d{\bf e}_{\phi}=\left(\begin{array}[]{c}(M+h)\sin\phi\sin\lambda\\\ -(M+h)\sin\phi\cos\lambda\\\ 0\\\ \end{array}\right)d\lambda+\left(\begin{array}[]{c}\left[Ne^{2}\frac{\cos^{2}\phi}{1-e^{2}\sin^{2}\phi}-3Ne^{2}(1-e^{2})\frac{\sin^{2}\phi}{(1-e^{2}\sin^{2}\phi)^{2}}-(N+h)\right]\cos\phi\cos\lambda\\\ \left[Ne^{2}\frac{\cos^{2}\phi}{1-e^{2}\sin^{2}\phi}-3Ne^{2}(1-e^{2})\frac{\sin^{2}\phi}{(1-e^{2}\sin^{2}\phi)^{2}}-(N+h)\right]\cos\phi\sin\lambda\\\ \left[Ne^{2}(1-e^{2})\left(3\frac{\cos^{2}\phi}{(1-e^{2}\sin^{2}\phi)^{2}}-\frac{\sin^{2}\phi}{1-e^{2}\sin^{2}\phi}\right)-[N(1-e^{2})+h]\right]\sin\phi\\\ \end{array}\right)d\phi.$ (25) The next step splits these two equations to the expanded version of (23), $\displaystyle d{\bf e}_{\lambda}$ $\displaystyle=$ $\displaystyle({\bf e}_{\lambda}\Gamma_{\lambda\lambda}^{\lambda}+{\bf e}_{\phi}\Gamma_{\lambda\lambda}^{\phi})d\lambda+({\bf e}_{\lambda}\Gamma_{\phi\lambda}^{\lambda}+{\bf e}_{\phi}\Gamma_{\phi\lambda}^{\phi})d\phi;$ (26) $\displaystyle d{\bf e}_{\phi}$ $\displaystyle=$ $\displaystyle({\bf e}_{\lambda}\Gamma_{\lambda\phi}^{\lambda}+{\bf e}_{\phi}\Gamma_{\lambda\phi}^{\phi})d\lambda+({\bf e}_{\lambda}\Gamma_{\phi\phi}^{\lambda}+{\bf e}_{\phi}\Gamma_{\phi\phi}^{\phi})d\phi.$ (27) The eight $\Gamma$ are extracted by evaluating dot products of the four vector coefficients in (24)–(25) by ${\bf e}_{\lambda}$ and ${\bf e}_{\phi}$, $\Gamma_{\lambda\lambda}^{\lambda}=\Gamma_{\phi\lambda}^{\phi}=\Gamma_{\lambda\phi}^{\phi}=\Gamma_{\phi\phi}^{\lambda}=0;$ (28) $\Gamma_{\lambda\lambda}^{\phi}=(N+h)\cos\phi\sin\phi\frac{1-e^{2}\sin^{2}\phi}{h(1-e^{2}\sin^{2}\phi)+N(1-e^{2})};$ (29) $\Gamma_{\phi\lambda}^{\lambda}=\Gamma_{\lambda\phi}^{\lambda}=-\sin\phi\frac{h(1-e^{2}\sin^{2}\phi)+N(1-e^{2})}{(N+h)(1-e^{2}\sin^{2}\phi)\cos\phi};$ (30) $\Gamma_{\phi\phi}^{\phi}=3Ne^{2}(1-e^{2})\sin\phi\cos\phi\frac{1}{[h(1-e^{2}\sin^{2}\phi)+N(1-e^{2})](1-e^{2}\sin^{2}\phi)}.$ (31) The Euler-Lagrange Differential Equations $\delta\int_{1}^{2}\sqrt{Ed\lambda^{2}+Gd\phi^{2}}=0$ for a stationary Lagrange density $\cal L$ (at $F=0$) become the differential equations of the geodesic Reichardt (1957), in the generic format $\frac{d^{2}x^{\epsilon}}{ds^{2}}+\sum_{\mu\nu}\Gamma_{\mu\nu}^{\epsilon}\frac{dx^{\mu}}{ds}\frac{dx^{\nu}}{ds}=0.$ (32) The explicit write-up $\frac{d^{2}\lambda}{ds^{2}}+\Gamma_{\lambda\lambda}^{\lambda}\frac{d\lambda}{ds}\frac{d\lambda}{ds}+\Gamma_{\lambda\phi}^{\lambda}\frac{d\lambda}{ds}\frac{d\phi}{ds}+\Gamma_{\phi\lambda}^{\lambda}\frac{d\phi}{ds}\frac{d\lambda}{ds}+\Gamma_{\phi\phi}^{\lambda}\frac{d\phi}{ds}\frac{d\phi}{ds}=0;$ (33) $\frac{d^{2}\phi}{ds^{2}}+\Gamma_{\lambda\lambda}^{\phi}\frac{d\lambda}{ds}\frac{d\lambda}{ds}+\Gamma_{\lambda\phi}^{\phi}\frac{d\lambda}{ds}\frac{d\phi}{ds}+\Gamma_{\phi\lambda}^{\phi}\frac{d\phi}{ds}\frac{d\lambda}{ds}+\Gamma_{\phi\phi}^{\phi}\frac{d\phi}{ds}\frac{d\phi}{ds}=0,$ (34) simplifies with (28) to $\displaystyle\frac{d^{2}\lambda}{ds^{2}}+2\Gamma_{\lambda\phi}^{\lambda}\frac{d\lambda}{ds}\frac{d\phi}{ds}$ $\displaystyle=$ $\displaystyle 0;$ (35) $\displaystyle\frac{d^{2}\phi}{ds^{2}}+\Gamma_{\lambda\lambda}^{\phi}\left(\frac{d\lambda}{ds}\right)^{2}+\Gamma_{\phi\phi}^{\phi}\left(\frac{d\phi}{ds}\right)^{2}$ $\displaystyle=$ $\displaystyle 0.$ (36) ## IV Reduction of the Differential Equations ### IV.1 Separation of Angular Variables Decoupling of the two differential equations (35)–(36) starts with the separation of variables in (35), $\frac{\frac{d^{2}\lambda}{ds^{2}}}{\frac{d\lambda}{ds}}=-2\Gamma_{\phi\lambda}^{\lambda}\frac{d\phi}{ds}.$ (37) Change of the integration variable on the right hand side from $s$ to $\phi$ allows to use the underivative of (30) $\int\Gamma_{\lambda\phi}^{\phi}(\phi)d\phi=\log\left\\{[N(\phi)+h]\cos\phi\right\\}+const$ (38) to generate a first integral $\log\frac{d\lambda}{ds}=-2\log[(N+h)\cos\phi]+const.$ (39) Exponentiation yields $\frac{d\lambda}{ds}=c_{3}\frac{1}{(N+h)^{2}\cos^{2}\phi},$ (40) where $c_{3}\equiv\frac{d\lambda}{ds}_{\mid 1}(N_{1}+h)^{2}\cos^{2}\phi_{1}$ (41) is a constant for each geodesic. Is has been defined with the azimuth $\phi_{1}$ and $N_{1}\equiv N(\phi_{1})$ at the start of the line, but could as well be associated with any other point or the end point $2$. $c_{3}$ is positive for trajectories starting into eastwards direction, negative for the westwards heading, zero for routes to the poles. Moving the square of (40) into (36) yields $\frac{d^{2}\phi}{ds^{2}}+\frac{\sin\phi}{(N+h)^{3}\cos^{3}\phi}\,\frac{1-e^{2}\sin^{2}\phi}{h(1-e^{2}\sin^{2}\phi)+N(1-e^{2})}c_{3}^{2}+\frac{3Ne^{2}(1-e^{2})\sin\phi\cos\phi}{[h(1-e^{2}\sin^{2}\phi)+N(1-e^{2})](1-e^{2}\sin^{2}\phi)}\left(\frac{d\phi}{ds}\right)^{2}=0.$ (42) To solve this differential equation, we substitute the variable $\phi$ by its projection $\tau$ onto the polar axis, $\tau\equiv\sin\phi,$ (43) which implies transformations in the derivatives: $\frac{d\tau}{ds}=\cos\phi\frac{d\phi}{ds};$ (44) $\frac{d^{2}\tau}{ds^{2}}=-\sin\phi\frac{d\phi}{ds}\frac{d\phi}{ds}+\cos\phi\frac{d^{2}\phi}{ds^{2}};$ (45) $\cos\phi\frac{d^{2}\phi}{ds^{2}}=\frac{d^{2}\tau}{ds^{2}}+\tau\left(\frac{d\phi}{ds}\right)^{2}=\frac{d^{2}\tau}{ds^{2}}+\frac{\tau}{\cos^{2}\phi}\left(\frac{d\tau}{ds}\right)^{2}=\frac{d^{2}\tau}{ds^{2}}+\frac{\tau}{1-\tau^{2}}\left(\frac{d\tau}{ds}\right)^{2}.$ (46) We multiply (42) by $\cos\phi$, then replace $d\phi/ds$ and $d^{2}\phi/ds^{2}$ as noted above, $\frac{d^{2}\tau}{ds^{2}}+\frac{\tau}{1-\tau^{2}}\left(\frac{d\tau}{ds}\right)^{2}+\frac{\tau}{(N+h)^{3}(1-\tau^{2})}\,\frac{1-e^{2}\tau^{2}}{h(1-e^{2}\tau^{2})+N(1-e^{2})}c_{3}^{2}+\frac{3Ne^{2}(1-e^{2})\tau}{[h(1-e^{2}\tau^{2})+N(1-e^{2})](1-e^{2}\tau^{2})}\left(\frac{d\tau}{ds}\right)^{2}=0;$ $\frac{d^{2}\tau}{ds^{2}}+\frac{\tau}{1-\tau^{2}}\,\frac{h(1-e^{2}\tau^{2})^{2}+N(1-4e^{2}\tau^{2}+2e^{2}+4e^{4}\tau^{2}-3e^{4})}{[h(1-e^{2}\tau^{2})+N(1-e^{2})][1-e^{2}\tau^{2}]}\left(\frac{d\tau}{ds}\right)^{2}\\\ +\frac{\tau}{(N+h)^{3}(1-\tau^{2})}\,\frac{1-e^{2}\tau^{2}}{h(1-e^{2}\tau^{2})+N(1-e^{2})}c_{3}^{2}=0.$ (47) This is a differential equation with no explicit appearance of the independent variable $s$, $(1-\tau^{2})\frac{d^{2}\tau}{ds^{2}}+\frac{h(1-e^{2}\tau^{2})^{2}+N(1-e^{2})(1+3e^{2}-4e^{2}\tau^{2})}{[h(1-e^{2}\tau^{2})+N(1-e^{2})][1-e^{2}\tau^{2}]}\tau\left(\frac{d\tau}{ds}\right)^{2}+\frac{\tau c_{3}^{2}}{(N+h)^{3}}\,\frac{1-e^{2}\tau^{2}}{h(1-e^{2}\tau^{2})+N(1-e^{2})}=0,$ and the standard way of progressing is the substitution $\frac{d\tau}{ds}\equiv p;\quad\frac{d^{2}\tau}{ds^{2}}=p\frac{dp}{d\tau};$ (48) $(1-\tau^{2})p\frac{dp}{d\tau}+\frac{h(1-e^{2}\tau^{2})^{2}+N(1-e^{2})(1+3e^{2}-4e^{2}\tau^{2})}{[h(1-e^{2}\tau^{2})+N(1-e^{2})][1-e^{2}\tau^{2}]}\tau p^{2}+\frac{\tau c_{3}^{2}}{(N+h)^{3}}\,\frac{1-e^{2}\tau^{2}}{h(1-e^{2}\tau^{2})+N(1-e^{2})}=0.$ (49) This is transformed to a linear differential equation by the further substitution $P\equiv p^{2}$, $dP/d\tau=2p\,dp/d\tau$, $\frac{1}{2}(1-\tau^{2})\frac{dP}{d\tau}+\frac{h(1-e^{2}\tau^{2})^{2}+N(1-e^{2})(1+3e^{2}-4e^{2}\tau^{2})}{[h(1-e^{2}\tau^{2})+N(1-e^{2})][1-e^{2}\tau^{2}]}\tau P+\frac{\tau c_{3}^{2}}{(N+h)^{3}}\,\frac{1-e^{2}\tau^{2}}{h(1-e^{2}\tau^{2})+N(1-e^{2})}=0.$ (50) The standard approach is to solve the homogeneous differential equation first, $\frac{dP}{d\tau}=-2\tau\frac{h(1-e^{2}\tau^{2})^{2}+N(1-e^{2})(1+3e^{2}-4e^{2}\tau^{2})}{[1-\tau^{2}][h(1-e^{2}\tau^{2})+N(1-e^{2})][1-e^{2}\tau^{2}]}P.$ (51) After division through $P$, the left hand side is easily integrated, and the right hand side (incompletely) decomposed into partial fractions, $\displaystyle\log P$ $\displaystyle=$ $\displaystyle-2\int\tau\frac{h(1-e^{2}\tau^{2})^{2}+N(1-e^{2})(1+3e^{2}-4e^{2}\tau^{2})}{[1-\tau^{2}][h(1-e^{2}\tau^{2})+N(1-e^{2})][1-e^{2}\tau^{2}]}d\tau$ (52) $\displaystyle=$ $\displaystyle\int\frac{-2\tau}{1-\tau^{2}}d\tau+3\int\frac{-2e^{2}\tau}{1-e^{2}\tau^{2}}d\tau+6he^{2}\int\frac{\tau}{h(1-e^{2}\tau^{2})+N(1-e^{2})}d\tau$ $\displaystyle=$ $\displaystyle\log(1-\tau^{2})+3\log(1-e^{2}\tau^{2})-2\log\left[h(1-e^{2}\tau^{2})^{3/2}+\rho_{e}(1-e^{2})\right]+const.$ $P=const\cdot\frac{(1-\tau^{2})(1-e^{2}\tau^{2})^{3}}{\left[h(1-e^{2}\tau^{2})^{3/2}+\rho_{e}(1-e^{2})\right]^{2}}=const\cdot\frac{(1-\tau^{2})(1-e^{2}\tau^{2})^{2}}{\left[h(1-e^{2}\tau^{2})+N(1-e^{2})\right]^{2}}=\frac{const\cdot(1-\tau^{2})}{G(\tau)}.$ Solution of the _inhomogeneous_ differential equation (50) proceeds with the variation of the constant, the ansatz $P=c(\tau)\cdot\frac{(1-\tau^{2})(1-e^{2}\tau^{2})^{2}}{\left[h(1-e^{2}\tau^{2})+N(1-e^{2})\right]^{2}}.$ (53) Back insertion into (50) leads to a first order differential equation for $c(\tau)$, $\frac{dc(\tau)}{d\tau}\frac{1-\tau^{2}}{G(\tau)}=-\frac{2\tau c_{3}^{2}}{(1-\tau^{2})(N+h)^{3}}\,\frac{1-e^{2}\tau^{2}}{h(1-e^{2}\tau^{2})+N(1-e^{2})},$ which is decomposed into partial fractions $\frac{dc(\tau)}{d\tau}=c_{3}^{2}\left[eN\frac{2e\tau}{1-e^{2}\tau^{2}}\frac{1}{(N+h)^{3}}+\frac{-2\tau}{1-\tau^{2}}\frac{1}{(N+h)^{2}}\right]\frac{1}{1-\tau^{2}}.$ The ensuing integral over $d\tau$ is solved by aid of the substitution $\tau^{2}=u$, $c(\tau)=-c_{3}^{2}\frac{1}{(N+h)^{2}(1-\tau^{2})}+const.$ Back into (53)—using $const$ to indicate placement of any member of an anonymous bag of constants of integration, $\displaystyle P$ $\displaystyle=$ $\displaystyle\left[const-\frac{1}{(N+h)^{2}(1-\tau^{2})}\right]\frac{c_{3}^{2}(1-\tau^{2})(1-e^{2}\tau^{2})^{2}}{\left[h(1-e^{2}\tau^{2})+N(1-e^{2})\right]^{2}}$ (54) $\displaystyle=$ $\displaystyle c_{5}\frac{(1-\tau^{2})(1-e^{2}\tau^{2})^{2}}{\left[h(1-e^{2}\tau^{2})+N(1-e^{2})\right]^{2}}-\frac{c_{3}^{2}(1-e^{2}\tau^{2})^{2}}{(N+h)^{2}\left[h(1-e^{2}\tau^{2})+N(1-e^{2})\right]^{2}}$ (55) $\displaystyle=$ $\displaystyle c_{5}\frac{1}{(h+M)^{2}}\left(1-\tau^{2}-\frac{c_{3}^{2}}{(N+h)^{2}}\right)=c_{5}\frac{1-\tau^{2}}{G(\tau)}\left(1-\frac{c_{3}^{2}}{E(\tau)}\right)=p^{2}.$ (56) The subscript $1$ denotes values at the starting point of the curve, $\displaystyle P_{1}$ $\displaystyle=$ $\displaystyle c_{5}\frac{\cos^{2}\phi_{1}(1-e^{2}\sin^{2}\phi_{1})^{2}}{\left[h(1-e^{2}\sin^{2}\phi_{1})+N_{1}(1-e^{2})\right]^{2}}-\frac{c_{3}^{2}(1-e^{2}\sin^{2}\phi_{1})^{2}}{(N_{1}+h)^{2}\left[h(1-e^{2}\sin^{2}\phi_{1})+N_{1}(1-e^{2})\right]^{2}}=p_{1}^{2}$ (57) $\displaystyle=$ $\displaystyle c_{5}\frac{\cos^{2}\phi_{1}(1-e^{2}\sin^{2}\phi_{1})^{2}}{\left[h(1-e^{2}\sin^{2}\phi_{1})+N_{1}(1-e^{2})\right]^{2}}-\frac{(d\lambda/ds)_{1}^{2}(N_{1}+h)^{2}\cos^{4}\phi_{1}(1-e^{2}\sin^{2}\phi_{1})^{2}}{\left[h(1-e^{2}\sin^{2}\phi_{1})+N_{1}(1-e^{2})\right]^{2}}.$ (58) Solving for $c_{5}$ yields $\displaystyle c_{5}$ $\displaystyle=$ $\displaystyle\left(\frac{d\lambda}{ds}_{\mid 1}\right)^{2}\cos^{2}\phi_{1}(N_{1}+h)^{2}+\frac{p_{1}^{2}\left[h(1-e^{2}\sin^{2}\phi_{1})+N_{1}(1-e^{2})\right]^{2}}{\cos^{2}\phi_{1}(1-e^{2}\sin^{2}\phi_{1})^{2}}=\left(\frac{d\lambda}{ds}_{\mid 1}\right)^{2}\cos^{2}\phi_{1}(N_{1}+h)^{2}+\frac{p_{1}^{2}}{\cos^{2}\phi_{1}}G_{1}$ (59) $\displaystyle=$ $\displaystyle\left(\frac{d\lambda}{ds}_{\mid 1}\right)^{2}E_{1}+\left(\frac{d\phi}{ds}_{\mid 1}\right)^{2}G_{1}=\frac{c_{3}^{2}}{(N_{1}+h)^{2}\cos^{2}\phi_{1}}+\left(\frac{d\phi}{ds}_{\mid 1}\right)^{2}(M_{1}+h)^{2}.$ Compared with the differential version of (22), $ds^{2}=Ed\lambda^{2}+Gd\phi^{2};\quad 1=E\left(\frac{d\lambda}{ds}\right)^{2}+G\left(\frac{d\phi}{ds}\right)^{2},$ (60) we must have $c_{5}=1$. Note that (56) is essentially a write-up for $p^{2}\sim(d\phi/ds)^{2}$ and could be derived quickly by inserting (40) directly into (60). The square root of (56) is $p=\pm\frac{1-e^{2}\tau^{2}}{h(1-e^{2}\tau^{2})+N(1-e^{2})}\sqrt{1-\tau^{2}-\frac{c_{3}^{2}}{(N+h)^{2}}}=\pm\frac{\sqrt{1-\tau^{2}-\frac{c_{3}^{2}}{(N+h)^{2}}}}{h+M(\tau)}=\frac{d\tau}{ds}.$ (61) The sign in front of the square root is to be chosen positive for pieces of the trajectory with $d\tau/ds>0$ (northern direction), negative where $d\tau/ds<0$ (southern direction). The value in the square root may run through zero within one curve, $d\tau/ds=0$ at one point $\tau_{m}$, such that the square root switches sign there (Figure 3). This happens whenever $(1-\tau_{m}^{2})[N(\tau_{m})+h]^{2}=c_{3}^{2}$ (62) yields a vanishing discriminant of the square root for some $(\lambda,\phi)$, that is, whenever the difference $\lambda_{2}-\lambda_{1}$ is sufficiently large to create a point of minimum polar distance along the trajectory that is not one of the terminal points. Figure 3: Two examples of trajectories of $d\tau/ds$ [left, equation (61)] or $ds/d\tau$ [right] as a function of $\tau$. They may or may not pass through a zero $\tau_{m}$ of (61) while connecting the starting abscissa $\tau_{1}$ with the final abscissa $\tau_{2}$. There are four basic topologies, depending on whether $\tau_{m}$ is positive or negative, and depending on whether the sign change in $d\tau/ds$ is from $+$ to $-$ or from $-$ to $+$ at this point. ### IV.2 Launching Direction So far we have written the bundle of all geodesics through $(\lambda_{1},\phi_{1})$ in the format (61), which specifies the change of latitude as a function of distance traveled. The direction at point 1 in the topocentric system of coordinates is represented by $c_{3}$. To address the inverse problem of geodesy, that is to pick the particular geodesics which also runs through the terminal point $(\lambda_{2},\phi_{2})$ in fulfillment of the Dirichlet boundary conditions, the associated change in longitude, some form of (40), $\frac{d\lambda}{ds}=c_{3}\frac{1}{(N+h)^{2}(1-\tau^{2})}=\frac{c_{3}}{E},$ (63) must obviously get involved. The strategy is to write down $\tau(\lambda)$ or $\lambda(\tau)$ with $c_{3}$ as a parameter, then to adjust $c_{3}$ to ensure with what would be called a shooting method that starting at point 1 at that angle eventually passes through point 2. Coupling of $\lambda$ to $\tau$ is done by division of (61) and (63), $\frac{d\tau}{d\lambda}=\frac{d\tau}{ds}\frac{ds}{d\lambda}=\frac{d\tau}{ds}/\frac{d\lambda}{ds}=\pm\frac{(N+h)^{2}(1-\tau^{2})\sqrt{1-\tau^{2}-\frac{c_{3}^{2}}{(N+h)^{2}}}}{c_{3}(h+M)}.$ (64) Separation of both variables yields $\pm\int_{1}d\tau\frac{c_{3}(h+M)}{(N+h)^{2}(1-\tau^{2})\sqrt{1-\tau^{2}-\frac{c_{3}^{2}}{(N+h)^{2}}}}=\lambda_{\mid 1}.$ (65) For short paths, small $|\lambda_{2}-\lambda_{1}|$, the integral is simply $\int_{\tau_{1}}^{\tau_{2}}$. If one of the cases occurs, where the difference in $\lambda$ is too large for a solution, the additional path through the singularity $\tau_{m}$ is to be used [with some local extremum in the graph $\tau(s)$], and the integral is to be interpreted as $\int_{\tau_{1}}^{\tau_{m}}-\int_{\tau_{m}}^{\tau_{2}}$. Taking the sign change and the symmetry of the integrand into account, this is $\int_{\tau_{1}}^{\tau_{2}}+2\int_{\tau_{2}}^{\tau_{m}}=\int_{\tau_{1}}^{\tau_{m}}+\int_{\tau_{2}}^{\tau_{m}}$, twice the underivative at $\tau_{m}$ minus the sum of the underivative at $\tau_{1}$ and $\tau_{2}$. In the right plot of Figure 3, the difference is in including or not including the loudspeaker shaped area between $\tau_{2}$ and $\tau_{m}$. $\tau_{m}$ is the solution of (62), the positive value of the solution taken if $d\tau/ds>0$ at $\phi_{1}$, the negative value if $d\tau/ds<0$ at $\phi_{1}$. The squared zero of $d\tau/ds$, $\tau_{m}^{2}$, is a root of the fourth-order polynomial which emerges by rewriting (62), $e^{4}h^{4}(\tau_{m}^{2})^{4}+2e^{2}h^{2}\left[\rho_{e}^{2}-h^{2}+e^{2}\left(c_{3}^{2}-h^{2}\right)\right](\tau_{m}^{2})^{3}\\\ +\left[\left(\rho_{e}^{2}-h^{2}\right)^{2}+2e^{2}\left(2h^{4}-2c_{3}^{2}h^{2}-2\rho_{e}^{2}h^{2}-\rho_{e}^{2}c_{3}^{2}\right)+e^{4}\left(c_{3}^{2}-h^{2}\right)^{2}\right](\tau_{m}^{2})^{2}\\\ +2\left[-\left(\rho_{e}^{2}-h^{2}\right)^{2}+c_{3}^{2}\left(\rho_{e}^{2}+h^{2}\right)+e^{2}\left(-\left(c_{3}^{2}-h^{2}\right)^{2}+\rho_{e}^{2}(c_{3}^{2}+h^{2})\right)\right]\tau_{m}^{2}\\\ +\left[(\rho_{e}+h)^{2}-c_{3}^{2}\right]\left[(\rho_{e}-h)^{2}-c_{3}^{2}\right]=0.$ (66) Alternatively, $\tau_{m}$ admits a Taylor expansion by expanding the zero of (62) in orders of $e$: $\pm\tau_{m}=\sqrt{1-\frac{c_{3}^{2}}{H^{2}}}+\frac{c_{3}^{2}\rho_{e}}{2H^{3}}\sqrt{1-\frac{c_{3}^{2}}{H^{2}}}e^{2}-\frac{3c_{3}^{2}\rho_{e}}{8H^{3}}\left[\left(\frac{\rho_{e}}{H}-\frac{c_{3}^{2}}{H^{2}}\right)^{2}-1\right]\sqrt{1-\frac{c_{3}^{2}}{H^{2}}}e^{4}+O(e^{6}),$ (67) where some maximum distance $H\equiv\rho_{e}+h$ (68) to the polar axis has been defined to condense the notation. The integrand in (65) has a Taylor expansion in $e$, $\pm\int d\tau\left[\frac{c_{3}/(\rho_{e}+h)}{(1-\tau^{2})\sqrt{1-\tau^{2}-\frac{c_{3}^{2}}{(\rho_{e}+h)^{2}}}}-\frac{c_{3}\rho_{e}[(\rho_{e}+h)^{2}(2-\tau^{2})-2c_{3}^{2}]}{2(\rho_{e}+h)^{4}\left(1-\tau^{2}-\frac{c_{3}^{2}}{(\rho_{e}+h)^{2}}\right)^{3/2}}e^{2}+O(e^{4})\right]=\lambda_{\mid 1}.$ (69) We integrate the left hand side of (69) separately for each power of $e$ up to $O(e^{4})$, $\Big{[}\arctan\frac{\tau\frac{c_{3}}{H}}{\sqrt{T}}-\frac{c_{3}\rho_{e}}{2H^{2}}\left(\frac{\tau}{\sqrt{T}}+\arctan\frac{\tau}{\sqrt{T}}\right)e^{2}\\\ -\frac{c_{3}\rho_{e}}{16H^{7}}\Big{(}\frac{\tau}{T^{3/2}}[-2H^{2}c_{3}^{2}\rho_{e}+4H^{2}c_{3}^{2}\rho_{e}T+2c_{3}^{4}\rho_{e}-6H^{3}c_{3}^{2}T+6H^{5}T-9H^{5}T^{2}-4H^{4}\rho_{e}T+6H^{4}\rho_{e}T^{2}]\\\ +H^{2}[3H^{3}-2\rho_{e}H^{2}-3c_{3}^{2}H+4c_{3}^{2}\rho_{e}]\arctan\frac{\tau}{\sqrt{T}}\Big{)}e^{4}+O(e^{6})\Big{]}_{\tau_{1}}^{\tau_{2}}=\pm(\lambda_{2}-\lambda_{1}),$ (70) where $T\equiv 1-\tau^{2}-\left(\frac{c_{3}}{H}\right)^{2}$ (71) is some convenient definition of the trajectory’s distance to its solstice $\tau_{m}$. The first term is not the principal value of the arc tangent but its steadily defined extension through the entire interval of $\tau$-values, $|\tau|\leq\tau_{m}$. It is an odd function of $\tau$, and phase jumps are corrected as follows: the inclination at $\tau=0$ has (by inspection of the derivative above at $\tau=0$) the sign of $c_{3}$. So whenever the triple product $\operatorname{sgn}\tau\operatorname{sgn}c_{3}\operatorname{sgn}(\arctan.)$ is negative, one must shift the branch of the arctan by adding multiples of $\pi\operatorname{sgn}\tau\operatorname{sgn}c_{3}$ to the principal value. Whether this is to be taken between the limits $\tau_{2}$ and $\tau_{1}$ or as the sum of two components (see above) can be tested by integrating up to $\tau_{m}$, $\int_{\tau_{1}}^{\tau_{m}}$, where the value of the underivative at $\tau_{m}$ is given by $(1/2)\arctan 0$, effectively $\frac{\pi}{2}\operatorname{sgn}\tau_{m}\operatorname{sgn}c_{3}$ after selecting the branch of the inverse trigonometric function as described above. Equation (70) is solved numerically, where $c_{3}/H$ is the unknown, where $\tau_{1,2}$ and $\lambda_{1,2}$ are known from the coordinates of the two points that define the boundary value problem, and where $e$ and $H$ are constant parameters. The complexity of the equation suggests use of a Newton algorithm to search for the zero, starting from (120), $c_{3}/H\approx\cos\phi_{1}\cos\phi_{2}\sin(\lambda_{2}-\lambda_{1})/\sin Z$, as the initial estimate. An alternative is to insert the power series $c_{3}=c_{3}^{(0)}+c_{3}^{(2)}e^{2}+c_{3}^{(4)}e^{4}+\cdots$ (72) right into (65), integrate the orders of $e$ term-by-term, and to obtain the coefficients $c_{3}^{(2)}$, $c_{3}^{(4)}$ etc. by comparison with the equivalent powers of the right hand side. $c_{3}^{(0)}$ is given by (120); the coefficients of the higher powers are recursively calculated from linear equations. $c_{3}^{(2)}$, for example, is determined via $\left[\frac{\rho_{e}c_{3}^{(0)}}{H^{2}}\sqrt{1-\tau^{2}-\left(\frac{c_{3}^{(0)}}{H}\right)^{2}}\left[1-\frac{{c_{3}^{(0)}}^{2}}{H^{2}}\right]\arctan\frac{H\tau}{\sqrt{1-\tau^{2}-\left(\frac{c_{3}^{(0)}}{H}\right)^{2}}}+\frac{\rho_{e}c_{3}^{(0)}}{H^{2}}\left[\frac{1-{c_{3}^{(0)}}^{2}}{H^{2}}\right]\tau-2\frac{c_{3}^{(2)}}{H}\tau\right]_{\tau_{1}}^{\tau_{2}}=0.$ (73) The corresponding equation for $c_{3}^{(4)}$ is already too lengthy to be reproduced here, so no real advantage remains in comparison with solving the non-linear equation (69). Once the parameter $c_{3}$ is known for a particular set of terminal coordinates, $\lambda(\phi)$ is given by replacing $\tau_{2}$ and $\lambda_{2}$ in (70) by any other generic pair of values. Two other variables of interest along the curve, the direction and integrated distance from the starting point, are then accessible with methods summarized in the next, final two sub-chapters. ### IV.3 Nautical course The nautical course at any point of the trajectory $\phi(\lambda)$ is the angle $\kappa$ in the topocentric coordinate system spanned by ${\bf e}_{\phi}$ (direction North) and ${\bf e}_{\lambda}$ (direction East), measured North over East, $\frac{d\bf r}{ds}=\cos\kappa\frac{{\bf e}_{\phi}}{|{\bf e}_{\phi}|}+\sin\kappa\frac{{\bf e}_{\lambda}}{|{\bf e}_{\lambda}|}=\sin\kappa\frac{{\bf e}_{\lambda}}{\sqrt{E}}+\cos\kappa\frac{{\bf e}_{\phi}}{\sqrt{G}}.$ (74) $d{\bf r}/ds$ is the differential of (15) with respect to $s$, where $d/ds=(d\lambda/ds)(d/d\lambda)+(d\phi/ds)(d/d\phi)$, $\frac{d\bf r}{ds}\propto\frac{d\lambda}{ds}{\bf e}_{\lambda}+\frac{d\phi}{ds}{\bf e}_{\phi}.$ (75) The sign $\propto$ indicates that the left hand side of this equation is a vector normalized to unity, but not the right hand side. $\frac{\sin\kappa/\sqrt{E}}{\cos\kappa/\sqrt{G}}=\frac{d\lambda/ds}{d\phi/ds}=\frac{d\lambda}{d\phi}.$ (76) Insertion of (19), (21) and (44) yield $\frac{M+h}{(N+h)\cos\phi}\tan\kappa=\frac{1}{\frac{d\phi}{d\lambda}}=\frac{1}{\frac{d\phi}{d\tau}\frac{d\tau}{d\lambda}}=\frac{d\tau/d\phi}{\frac{d\tau}{d\lambda}}=\frac{\cos\phi}{\frac{d\tau}{d\lambda}}.$ (77) Mixing (64) into this yields the course, supposed $\phi$ and $c_{3}$ are known, $\tan\kappa=\pm\,\frac{c_{3}}{(N+h)\sqrt{\cos^{2}\phi-(\frac{c_{3}}{N+h})^{2}}}.$ (78) ### IV.4 Distance To Terminal Points An implicit write-up for the distance from the origin ${\mathbf{s}}_{1}$ measured along the curve is given by separating variables in (61): $s=\pm\int_{\tau_{1}}d\tau\frac{h(1-e^{2}\tau^{2})+N(1-e^{2})}{(1-e^{2}\tau^{2})\sqrt{1-\tau^{2}-\frac{c_{3}^{2}}{(N+h)^{2}}}}.$ (79) Power series expansion of the integrand in powers of $e$, then integration term-by-term, generate a Taylor series of the form $s=\pm(s^{(0)}+s^{(2)}e^{2}+s^{(4)}e^{4}+\ldots)\big{|}_{\tau_{1}}^{\tau}.$ In the notation (71), the components of the underivative read: $s^{(0)}=H\arctan\frac{\tau}{\sqrt{1-\tau^{2}-c_{3}^{2}/H^{2}}}=H\arctan\frac{\tau}{\sqrt{T}};$ (80) $s^{(2)}=-\frac{\rho_{e}}{4}\Big{[}(1+c_{3}^{2}/H^{2})\arctan\frac{\tau}{\sqrt{T}}+\tau\frac{3(1-\tau^{2})-c_{3}^{2}/H^{2}}{\sqrt{T}}\Big{]};$ (81) $s^{(4)}=\frac{\rho_{e}}{64}\Big{\\{}\left[9(c_{3}/H)^{4}-6(c_{3}/H)^{2}-12(c_{3}/H)^{4}(\rho_{e}/H)+4(c_{3}/H)^{2}(\rho_{e}/H)-3\right]\arctan\frac{\tau}{\sqrt{T}}\\\ +\frac{\tau}{T^{3/2}}\Big{[}24T(c_{3}/H)^{4}-24T(c_{3}/H)^{2}+63T^{2}(c_{3}/H)^{2}-8(\rho_{e}/H)(c_{3}/H)^{6}+8(\rho_{e}/H)(c_{3}/H)^{4}\\\ -8(\rho_{e}/H)T(c_{3}/H)^{4}+8(\rho_{e}/H)T(c_{3}/H)^{2}-12(\rho_{e}/H)T^{2}(c_{3}/H)^{2}-27T^{2}+30T^{3}\Big{]}\Big{\\}}.$ (82) ## V Summary Computation of the geodetic line within the iso-surface of constant altitude above the ellipsoid is of the same complexity as on its surface at zero altitude. Although the surface is no longer an ellipsoid, mathematics reaches an equivalent level of simplification at which one integral is commonly expanded in powers of the eccentricity or flattening factor. We have done this first for the parameter which provides a solution to the inverse problem, then for two of the basic functions, distance from the starting point and compass course. ## Appendix A Reference: Spherical case The limit of vanishing eccentricity, $e=0$, simplifies the curved trajectories to arcs of great circles, and presents an easily accessible first estimate of the series expansions in orders of $e$. The Cartesian coordinates of the two points to be connected then are ${\mathbf{s}}_{1}=(\rho_{e}+h)\left(\begin{array}[]{c}\cos\phi_{1}\cos\lambda_{1}\\\ \cos\phi_{1}\sin\lambda_{1}\\\ \sin\phi_{1}\\\ \end{array}\right);\quad{\mathbf{s}}_{2}=(\rho_{e}+h)\left(\begin{array}[]{c}\cos\phi_{2}\cos\lambda_{2}\\\ \cos\phi_{2}\sin\lambda_{2}\\\ \sin\phi_{2}\\\ \end{array}\right).$ (83) The angular separation $Z$ is derived from the dot product ${\bf s}_{1}\cdot{\bf s}_{2}=|{\bf s}_{1}||{\bf s}_{2}|\cos Z$, $\cos Z=\sin\phi_{1}\sin\phi_{2}+\cos\phi_{1}\cos\phi_{2}\cos(\lambda_{1}-\lambda_{2});\quad 0\leq Z\leq\pi.$ (84) Each point ${\bf s}$ on the great circle in between lies in the plane defined by the sphere center and the terminal points, and can therefore be written as a linear combination ${\bf s}=\alpha{\mathbf{s}}_{1}+\beta{\mathbf{s}}_{2},\quad 0\leq\alpha,\beta.$ (85) The square must remain normalized to the squared radius $s^{2}=\alpha^{2}{\mathbf{s}}_{1}^{2}+\beta^{2}{\mathbf{s}}_{2}^{2}+2\alpha\beta{\mathbf{s}}_{1}\cdot{\mathbf{s}}_{2}=\alpha^{2}(\rho_{e}+h)^{2}+\beta^{2}(\rho_{e}+h)^{2}+2\alpha\beta(\rho_{e}+h)^{2}\cos Z,$ (86) which couples the two expansion coefficients via $\alpha^{2}+\beta^{2}+2\alpha\beta\cos Z=1.$ (87) One reduction to a single parameter $\xi$ to enforce this condition is $\alpha=\frac{\cos\xi}{\sqrt{1+\sin(2\xi)\cos Z}};\quad\beta=\frac{\sin\xi}{\sqrt{1+\sin(2\xi)\cos Z}};\quad 0\leq\xi\leq\pi/2.$ (88) In summary, the point (85) on the great circle of radius $\rho_{e}+h$ between ${\mathbf{s}}_{1}$ and ${\mathbf{s}}_{2}$ has the Cartesian coordinates $\displaystyle{\mathbf{s}}$ $\displaystyle=$ $\displaystyle(\rho_{e}+h)\frac{\cos\xi}{\sqrt{1+\sin(2\xi)\cos Z}}\left(\begin{array}[]{c}\cos\phi_{1}\cos\lambda_{1}\\\ \cos\phi_{1}\sin\lambda_{1}\\\ \sin\phi_{1}\\\ \end{array}\right)+(\rho_{e}+h)\frac{\sin\xi}{\sqrt{1+\sin(2\xi)\cos Z}}\left(\begin{array}[]{c}\cos\phi_{2}\cos\lambda_{2}\\\ \cos\phi_{2}\sin\lambda_{2}\\\ \sin\phi_{2}\\\ \end{array}\right)$ (95) $\displaystyle\equiv$ $\displaystyle(\rho_{e}+h)\left(\begin{array}[]{c}\cos\phi(\xi)\cos\lambda(\xi)\\\ \cos\phi(\xi)\sin\lambda(\xi)\\\ \sin\phi(\xi)\\\ \end{array}\right).$ (99) The $z$-component of this, $\sin\phi(\xi)=\frac{\cos\xi\sin\phi_{1}+\sin\xi\sin\phi_{2}}{\sqrt{1+\sin(2\xi)\cos Z}},$ (100) allows to convert $\xi$ into $\varphi$. No ambiguity with respect to the branch of the $\arcsin$ arises since $-\pi/2\leq\phi\leq\pi/2$. The ratio of the $y$ and $x$-components demonstrates the dependence of $\lambda$ on $\xi$, $\tan\lambda(\xi)=\frac{\cos\xi\cos\phi_{1}\sin\lambda_{1}+\sin\xi\cos\phi_{2}\sin\lambda_{2}}{\cos\xi\cos\phi_{1}\cos\lambda_{1}+\sin\xi\cos\phi_{2}\cos\lambda_{2}},$ (101) where the $\arctan$ branch is defined by considering separately the numerator and denominator under the restrictions that no signs are canceled. The azimuth angle $\sigma$ between the points at $\xi=0$ and at $\xi$ follows from ${\mathbf{s}}\cdot{\mathbf{s}}_{1}=(\rho_{e}+h)^{2}\cos\sigma$, $\displaystyle\cos\sigma$ $\displaystyle=$ $\displaystyle\left(\frac{\cos\xi}{\sqrt{1+\sin(2\xi)\cos Z}}\left(\begin{array}[]{c}\cos\phi_{1}\cos\lambda_{1}\\\ \cos\phi_{1}\sin\lambda_{1}\\\ \sin\phi_{1}\\\ \end{array}\right)+\frac{\sin\xi}{\sqrt{1+\sin(2\xi)\cos Z}}\left(\begin{array}[]{c}\cos\phi_{2}\cos\lambda_{2}\\\ \cos\phi_{2}\sin\lambda_{2}\\\ \sin\phi_{2}\\\ \end{array}\right)\right)\cdot\left(\begin{array}[]{c}\cos\phi_{1}\cos\lambda_{1}\\\ \cos\phi_{1}\sin\lambda_{1}\\\ \sin\phi_{1}\\\ \end{array}\right)$ (111) $\displaystyle=$ $\displaystyle\frac{\sin\xi\cos Z+\cos\xi}{\sqrt{1+\sin(2\xi)\cos Z}};\quad 0\leq\sigma\leq Z.$ (112) The path length along the great circle perimeter is simply the radial distance $\rho_{e}+h$ to the center of coordinates times the azimuth angle $\sigma$ measured in radians, $s=\int_{1}\sqrt{ds^{2}}=(\rho_{e}+h)\sigma=(\rho_{e}+h)\arccos\frac{\sin\xi\cos Z+\cos\xi}{\sqrt{1+\sin(2\xi)\cos Z}};\quad 0\leq s\leq(\rho_{e}+h)Z.$ (113) To calculate the direction in the ${\mathbf{e}}_{\lambda}$\- ${\mathbf{e}}_{\phi}$-plane at the starting point ${\mathbf{s}}_{1}$, we employ $\xi$ as the parameter that mediates between $\lambda$ and $\phi$: $\frac{d\lambda}{ds}=\frac{1}{\rho_{e}+h}\frac{d\lambda}{d\sigma}=\frac{1}{\rho_{e}+h}\frac{d\lambda}{d\xi}\,\frac{d\xi}{d\sigma}=\frac{1}{\rho_{e}+h}\frac{d\lambda}{d\xi}/\frac{d\sigma}{d\xi}.$ (114) To calculate $d\lambda/d\xi$, use the derivative of (101) with respect to $\xi$, $\displaystyle\frac{1}{\cos^{2}\lambda}\frac{d\lambda}{d\xi}$ $\displaystyle=$ $\displaystyle\frac{-\sin\xi\cos\phi_{1}\sin\lambda_{1}+\cos\xi\cos\phi_{2}\sin\lambda_{2}}{\cos\xi\cos\phi_{1}\cos\lambda_{1}+\sin\xi\cos\phi_{2}\cos\lambda_{2}}$ (115) $\displaystyle-(\cos\xi\cos\phi_{1}\sin\lambda_{1}+\sin\xi\cos\phi_{2}\sin\lambda_{2})\frac{-\sin\xi\cos\phi_{1}\cos\lambda_{1}+\cos\xi\cos\phi_{2}\cos\lambda_{2}}{(\cos\xi\cos\phi_{1}\cos\lambda_{1}+\sin\xi\cos\phi_{2}\cos\lambda_{2})^{2}}.$ In particular at the starting point, where $\xi=0$, $\lambda=\lambda_{1}$ and $\phi=\phi_{1}$, $\frac{1}{\cos^{2}\lambda_{1}}\frac{d\lambda}{d\xi}_{\mid 1}=\frac{\cos\phi_{2}\sin\lambda_{2}}{\cos\phi_{1}\cos\lambda_{1}}-\cos\phi_{1}\sin\lambda_{1}\frac{\cos\phi_{2}\cos\lambda_{2}}{(\cos\phi_{1}\cos\lambda_{1})^{2}}.$ By multiplication with $\cos\phi_{1}\cos^{2}\lambda_{1}$ $\cos\phi_{1}\frac{d\lambda}{d\xi}_{\mid 1}=\cos\phi_{2}\sin\lambda_{2}\cos\lambda_{1}-\sin\lambda_{1}\cos\phi_{2}\cos\lambda_{2}=\cos\phi_{2}\sin(\lambda_{2}-\lambda_{1}).$ (116) To calculate $d\sigma/d\xi$, we convert the cosine in (112) to the sine, $\displaystyle\sin\sigma$ $\displaystyle=$ $\displaystyle\sqrt{1-\cos^{2}\sigma}=\frac{\sin\xi\sin Z}{\sqrt{1+\sin(2\xi)\cos Z}}.$ (117) The derivative of this with respect to $\xi$ is $\cos\sigma\frac{d\sigma}{d\xi}=\cos\xi\sin Z\frac{1}{\sqrt{1+\sin(2\xi)\cos Z}}-\frac{1}{2}\sin\xi\sin Z\frac{2\cos(2\xi)\cos Z}{(1+\sin(2\xi)\cos Z)^{3/2}},$ in particular at the starting point, $\sigma=\xi=0$, $\frac{d\sigma}{d\xi}_{\mid 1}=\sin Z.$ (118) Insert this derivative and (116) back into (114) $\frac{d\lambda}{ds}_{\mid 1}=\frac{1}{\rho_{e}+h}\frac{\cos\phi_{2}\sin(\lambda_{2}-\lambda_{1})}{\cos\phi_{1}}\frac{1}{\sin Z},$ (119) to obtain the master parameter $c_{3}$ for the spherical case with (41), $c_{3}^{(0)}=(\rho_{e}+h)\frac{\cos\phi_{1}\cos\phi_{2}\sin(\lambda_{2}-\lambda_{1})}{\sin Z};\quad(e=0).$ (120) ## Appendix B Zero Altitude In particular _on_ the ellipsoid, at $h=0$, (formal) solutions to some integrals exist. An underivative of (79) in terms of Elliptic Integrals of the Second Kind is (Gradstein and Ryshik, 1981, 3.158.11)(Byrd and Friedman, 1971, 219.07), $s=\pm\rho_{e}\left[\sqrt{1-\frac{c_{3}^{2}e^{2}}{\rho_{e}^{2}}}E\left(\arcsin(\frac{e\tau}{\sin\zeta})\setminus\zeta\right)-e^{2}\tau\sqrt{\frac{1-e^{2}}{1-e^{2}\tau^{2}}-\frac{c_{3}^{2}}{\rho_{e}^{2}}}\right]+const;\quad h=0,$ (121) where an auxiliary modular angle $\zeta$ is defined from $\sin\zeta\equiv e\sqrt{\frac{1-c_{3}^{2}/\rho_{e}^{2}}{1-c_{3}^{2}e^{2}/\rho_{e}^{2}}}.$ The left hand side of (65) can be written as an Elliptic Integral of the Third Kind (Gradstein and Ryshik, 1981, 3.157.7), $\pm\frac{c_{3}}{\rho_{e}}\,\frac{1-e^{2}}{\sqrt{1-c_{3}^{2}e^{2}/\rho_{e}^{2}}}\Pi\left(\frac{\sin^{2}\zeta}{e^{2}};\arcsin(\frac{e\tau}{\sin\zeta})\setminus\zeta\right)+const=\lambda;\quad h=0.$ (122) ## Appendix C Notations $c_{5},c_{3}$ | constants of integration ---|--- $E,E_{1}$ | Gauss parameter eq. (19) and its value at curve origin $(\lambda_{1},\phi_{1},h)$ $E(.\setminus.)$ | Incomplete Elliptic Integral of the Second Kind in the Abramowitz-Stegun notation (Abramowitz and Stegun, 1972, §17) $e$ | eccentricity of the ellipsoid, eq. (1) ${\mathbf{e}}_{\lambda,\phi}$ | horizontal topocentric coordinate vectors at ($\lambda,\phi,h$) $f$ | flattening, eq. (8) $F$ | Gauss parameter, eq. (20) $G,G_{1}$ | Gauss parameter, eq. (21) and its value at curve origin $\Gamma_{..}^{.}$ | Christoffel symbols in $(\lambda,\phi,h)$ coordinates $h$ | vertical distance to surface of ellipsoid $H$ | maximum distance to polar axis (in the equatorial plane), eq. (68) $\kappa$ | nautical angle, North over East in the topocentric tangential plane $\lambda$ | longitude $M$ | a radius of curvature on the ellipsoid surface, eq. (18) $N$ | a radius of curvature on the ellipsoid surface, eq. (13) $\rho$ | distance ellipsoid center to foot point on the surface $\rho_{e,p}$ | equatorial, polar radius of ellipsoid, eq. (1) $s$ | distance along geodetic line measured from curve origin ${\mathbf{s}}$ | vector from ellipsoid center to point on geodetic line $S$ | length of curved geodetic trajectory; equals $s_{2}$ at $(\lambda_{2},\phi_{2},h)$ $\operatorname{sgn}$ | sign function, $\pm 1$ or $0$ $\sigma$ | azimuth along great circle, eq. (113) $T$ | normalized distance from closest polar approach, eq (71) $\tau,\tau_{1}$ | $\sin\phi$ and its value at start of curve, eq. (43) $v$ | geodetic minus geocentric latitude $\phi$ | geodetic latitude $\phi^{\prime}$ | geocentric latitude $\Pi(.;.\setminus.)$ | Incomplete Elliptic Integral of the Third Kind (Abramowitz and Stegun, 1972, §17) $x,y,z$ | Cartesian coordinates from ellipsoid center, eq. (15) $\xi$ | parametrization of great circle (spherical case), eq. (95) $Z$ | cone angle of circular section (spherical case), eq. (84) ## Appendix D FEM implementation The simplest numerical solution of the inverse problem without restriction on the eccentricity could be a finite-element (FEM) integration of (65) and iterative adjustment of the single free parameter $c_{3}$. The following Java program implements this approach. Member functions in overview: the constructors Geod define a surface from the parameters $\rho_{e}$, $e$ and $h$ in which the geodetic line is embedded. getCartesian computes the vector (15). curvN computes (13). dNdtau computes $\frac{dN}{d\tau}=N(\tau)\frac{e^{2}\tau}{1-e^{2}\tau^{2}}.$ (123) flatt computes (8). curvM computes (18). dMdtau computes $\frac{dM}{d\tau}=3M(\tau)\frac{e^{2}\tau}{1-e^{2}\tau^{2}}.$ (124) d2Mdtau2 computes $\frac{d^{2}M}{d\tau^{2}}=3M(\tau)\frac{e^{2}(1+4e^{2}\tau^{2})}{(1-e^{2}\tau^{2})^{2}}.$ (125) GaussE computes (19). dEdtau computes $\frac{dE}{d\tau}=-2\tau(N+h)(M+h).$ (126) d2Edtau2 computes the next higher derivative, $d^{2}E/d\tau^{2}$. GaussG computes (21). tauTropic solves (66) for $\tau_{m}$ with a first estimate taken from (67); the function is not actually called in this implementation. dtaudlambda computes $d\tau/d\lambda$ via (64), referencing one factor to (61). dsdlambda is $ds/d\lambda$, the inverse of (63). discrT computes $1-\tau^{2}-c_{3}^{2}/(N+h)^{2}$, which generalizes (71) to nonzero $e$. Its derivatives $\displaystyle\frac{d}{d\tau}\left[1-\tau^{2}-\frac{c_{3}^{2}}{[N(\tau)+h]^{2}}\right]$ $\displaystyle=$ $\displaystyle 2\tau\left[\frac{Ne^{2}c_{3}^{2}}{(N+h)^{3}(1-e^{2}\tau^{2})}-1\right];$ (127) $\displaystyle\frac{d^{2}}{d\tau^{2}}\left[1-\tau^{2}-\frac{c_{3}^{2}}{[N(\tau)+h]^{2}}\right]$ $\displaystyle=$ $\displaystyle-2\left[1-\frac{Nc_{3}^{2}e^{2}}{(N+h)^{3}(1-e^{2}\tau^{2})}\left(1+\frac{3he^{2}\tau^{2}}{(N+h)(1-e^{2}\tau^{2})}\right)\right],$ (128) are implemented in dTdtau and d2Tdtau2. dtauds calculates (61). d2taudlambda2 calculates the derivative of (64), $\frac{d^{2}\tau}{d\lambda^{2}}=\frac{d}{d\lambda}\frac{E(\tau)\sqrt{1-\tau^{2}-c_{3}^{2}/(N+h)^{2}}}{c_{3}(h+M)}=\frac{d\tau}{d\lambda}\frac{d}{d\tau}\frac{E(\tau)\sqrt{1-\tau^{2}-c_{3}^{2}/(N+h)^{2}}}{c_{3}(h+M)}.$ (129) d3taudlambda3 is the next higher order application of Bruno di Faà’s formula to relegate derivatives $d/d\lambda$ to derivatives $d/d\tau$, (Gradstein and Ryshik, 1981, 0.430), $\frac{d^{3}\tau}{d\lambda^{3}}=\frac{d^{2}\tau}{d\lambda^{2}}\frac{d}{d\tau}\frac{E(\tau)\sqrt{1-\tau^{2}-c_{3}^{2}/(N+h)^{2}}}{c_{3}(h+M)}+\left(\frac{d\tau}{d\lambda}\right)^{2}\frac{d^{2}}{d\tau^{2}}\frac{E(\tau)\sqrt{1-\tau^{2}-c_{3}^{2}/(N+h)^{2}}}{c_{3}(h+M)}.$ (130) ds2dlambda2 calculates $\frac{d^{2}s}{d\lambda^{2}}=\frac{d}{d\lambda}\frac{E}{c_{3}}=\frac{1}{c_{3}}\,\frac{d\tau}{d\lambda}\,\frac{d}{d\tau}E.$ (131) c3Sphere returns the estimate (120). dtaudsSignum returns the sign of $d\tau/ds$ at $\tau_{1}$, obtained by considering the sign of the derivative of (100) with respect to $\xi$. adjLambdaEnd modifies $\phi_{2}$ modulo $2\pi$ to select the smallest value of $|\phi_{2}-\phi_{1}|$. nautAngle computes $\kappa$ from (78). tauShoot walks along a geodetic line on a discrete mesh of width $\Delta\lambda$ by extrapolating $\tau_{\lambda+\Delta\lambda}\approx\tau_{\lambda}+\frac{d\tau}{d\lambda}\Delta\lambda+\frac{1}{2}\frac{d^{2}\tau}{d\lambda^{2}}(\Delta\lambda)^{2}+\frac{1}{6}\frac{d^{3}\tau}{d\lambda^{3}}(\Delta\lambda)^{3},$ (132) initialized at $\lambda_{1},\phi_{1}$, given $c_{3}$. The equivalent formula is used to build up $s_{\lambda+\Delta\lambda}$, currently only implemented up to the order $(\Delta\lambda)^{2}$, since d3sdlambda3 returns 0. c3shoot calls tauShoot four times to adjust $c_{3}$ such that the error by which $\phi_{2}$ was missed—returned by tauShoot—is minimized. The first call assumes (120), the second takes an arbitrary small offset, and the third and fourth estimates are from linear and quadratic interpolations in the earlier calls to zoom into a root of this error as a function of $c_{3}$. The last of these runs tabulates the Cartesian coordinates (15), $\lambda$, $\phi$, $\kappa$ and $s$ on a subgrid of the $\lambda$-mesh. main collects some adjustable parameters plus the pairs $(\lambda_{1},\phi_{1})$ and $(\lambda_{2},\phi_{2})$, and calls c3shoot to solve the inverse problem of geodetics. package org.nevec.rjm ; /** Solution of the inverse problem of geodesy for biaxial ellipsoid. * The inverse problem of geodesy is solved given the parameters of * a biaxial ellipsoid (equatorial radius and eccentricity), a common * altitude of a surface above the ellipsoid, and pairs of geodetic * coordinates (latitude, longitude) for a starting and a final point * of a trajectory. * <p> * * The first order differential equation of the smooth geodetic latitude * as a function of longitude is solved by moving from one point to the * next on a finite grid of equidistant sampling points on the longitude, * using a predictor-only Newton approximation. * The parameter which defines the initial direction in the local * tangential plane is gathered by a shooting method with a starting * guess obtained from a spherical approximation and a fixed number of * iteration through the parameter space. * * @since 2008-04-12 * @see preprint <a href="http://arxiv.org/abs/0711.0642">Geodetic line at constant altitude above the Ellipsoid</a>, * arXiv:0711.0642 [math.MG] * @author Richard J. Mathar */ class Geod { /** The inverse flattening factor of the IERS 2003 convention. */ static public final double IERS_TN32_FLAT= 298.25642 ; /** The inverse flattening factor of the IERS 1996 convention. */ static public final double IERS_TN21_FLAT= 298.25642 ; /** The inverse flattening factor of the GRS80 model. */ static public final double GRS80_FLAT= 298.257222101 ; /** The inverse flattening factor of the WGS84 model. */ static final double WGS84_FLAT= 298.257223563 ; /** Equatorial radius of the IERS 2003 convention, meters. */ static public final double IERS_TN32_RHO_E=6378136.6 ; /** Equatorial radius of the IERS 1996 convention, meters. */ static public final double IERS_TN21_RHO_E=6378136.49 ; /** Equatorial radius of the GRS 1980 system, meters. */ static public final double GRS80_RHO_E=6378137.0 ; /** Equatorial radius of the WGS84, meters. */ static public final double WGS84_RHO_E=6378137.0 ; /** Three enumeration values to convert angular units on user input and output */ interface AngleUnit { int RAD =0, DEG = 1, GON = 2; } ; /** Equatorial radius in meters. */ double rho_equat ; /** Eccentricity. * For the Earth of the order of 0.08. */ double eccen ; /** Altitude of the inverse problem above the ellipsoid. * Value equals 0 on the ellipsoid surface. May have both signs. * Measured in the same units as Geod::rho_equat. */ double altit ; /** Default Constructor. * The variables are set to the defaults of the surface of the WGS84 geoid. */ public Geod() { this(WGS84_RHO_E,Math.sqrt((2.0-1./WGS84_FLAT)/WGS84_FLAT),0.) ; } /** Constructor. * @param rho equatorial radius [m] * @param e eccentricity * @param h altitude above the reference ellipsoid [m] * @todo catch error conditions of e>=1, rho<0 or h< polar radius. */ public Geod(double rho, double e, double h) { rho_equat = rho ; eccen = e ; altit = h ; } /** Convert geodetic to Cartesian coordinates. * @param tau the sine of the geodetic latitude * @param lambd the geodetic longitude [rad] * @return the three components of the Cartesian coordinates * in the same units as rho_equat and altit. * @see eq (15) of the preprint */ public double[] getCartesian(final double tau, final double lambd) { double[] cart = new double[3] ; final double N = curvN(tau) ; final double sinphi = Math.sqrt(1.-tau*tau) ; cart[0] = (N+altit)*sinphi*Math.cos(lambd) ; cart[1] = (N+altit)*sinphi*Math.sin(lambd) ; cart[2] = (N*(1.+eccen)*(1.-eccen)+altit)*tau ; return cart ; } /** Curvature parameter N [m] * @param tau sine of the latitude * @return N according to eq (13) of the preprint. */ public double curvN(final double tau) { return rho_equat/Math.sqrt( (1.+eccen*tau)*(1.-eccen*tau) ) ; } /** Derivative d N/d tau [m]. * @param tau sine of the latitude * @return the derivative of N(tau) */ double dNdtau(double tau) { return curvN(tau)*eccen*eccen*tau/((1.+eccen*tau)*(1.-eccen*tau)) ; } /** Convert eccentricity to flattening factor. * @param e the (first) eccentricity * @return f = 1-sqrt(1-eccentricity^2). */ public static double flatt(final double e) { final double e2 = e*e ; if ( e2 < 3.8e-3) /* a Taylor approximation up to O(e^14), relative accuracy to E-16. */ return e2*(0.5+e2*(1./8.+e2*(1./16.*e2*(5./128.+e2*(7./256.+e2*(21./1024.+e2*33./2048.)))))) ; else return 1.-Math.sqrt(1.-e2) ; } /** Flattening factor. * @return f = 1-sqrt(1-eccentricity^2). */ public double flatt() { return flatt(eccen) ; } /** Equatorial radius of curvature M [m] * @param tau sine of the latitude * @return M according to eq (18) of the preprint. */ public double curvM(final double tau) { return curvN(tau)*(1.-eccen*eccen)/( (1.+eccen*tau)*(1.-eccen*tau) ) ; } /** Derivative d M /d tau [m] * @param tau sine of the latitude * @return the derivative of curvM() with respect to the parameter tau. */ public double dMdtau(final double tau) { return 3.*curvM(tau)*eccen*eccen*tau/( (1.+eccen*tau)*(1.-eccen*tau) ) ; } /** second Derivative d^2 M /d tau^2 [m] * @param tau sine of the latitude * @return the second derivative of curvM() with respect to the parameter tau. */ public double d2Mdtau2(final double tau) { final double etau2 = eccen*eccen*tau*tau ; return 3.*curvM(tau)*Math.pow(eccen/(1.-etau2),2.)*(1.+4.*etau2) ; } /** Gauss parameter E [m^2]. * @param tau sine of the latitude * @return E according to eq (19) of the preprint. */ double GaussE(final double tau) { return Math.pow(curvN(tau)+altit,2.)*(1.-tau*tau) ; } /** Derivative d E/ d tau [m^2] * @param tau sine of the latitude * @return the derivative of GaussE() with respect to the parameter tau */ public double dEdtau(final double tau) { return -2.*tau*(curvN(tau)+altit)*(curvM(tau)+altit) ; } /** Derivative d^2 E/ d tau^2 [m^2] * @param tau sine of the latitude * @return the 2nd derivative of GaussE() with respect to the parameter tau */ public double d2Edtau2(final double tau) { /* Use the Leibniz rule applied to the first derivative, which was -2*tau*(N+h)(M+h) */ final double Nh = curvN(tau)+altit ; final double Mh = curvM(tau)+altit ; return -2.*(Nh*Mh+tau*dNdtau(tau)*Mh+tau*Nh*dMdtau(tau)) ; } /** Gauss parameter G [m^2]. * @param tau sine of the latitude * @return G according to eq (21) of the preprint. * @since 2008-04-22 */ double GaussG(final double tau) { return Math.pow(curvM(tau)+altit,2.) ; } /** Minimum polar distance. * @param c3 the constant c3 along one individual trajectory * @return the value of tau_m according to eq (66) of the preprint that could be reached if the trajectory were * closed over the surface. That is the maximum of |sin(phi)|, or the value * of |sin(phi)| where the derivative of sin(phi) with respect to the longitude * becomes zero. * @since 2008-04-22 */ double tauTropic(final double c3) { final double c3sqr = c3*c3 ; final double H = rho_equat+altit ; final double rhoebar = rho_equat/H ; final double esqr = eccen*eccen ; /* Initial estimate from Taylor expansion in orders of squared eccentricity * See eq (67) of the preprint */ double tau2 = (1.-c3sqr/(H*H))*(1+c3sqr*rhoebar*esqr/(H*H)* (1.+0.25*(3.*(1.-c3sqr/(H*H))+rhoebar*(7.*c3sqr/(H*H)-3.))*esqr) ) ; final double hsqr = altit*altit ; final double rhosqr = rho_equat*rho_equat ; /* coefficients in the 4th order polynomial of tau^2 which is to be solved */ double[] coeff = new double[5] ; coeff[0] = (Math.pow(rho_equat+altit,2.)-c3sqr)*(Math.pow(rho_equat-altit,2.)-c3sqr) ; coeff[1] = 2.*(-Math.pow(rhosqr-hsqr,2.) +c3sqr*(rhosqr+hsqr) +esqr*(-Math.pow(c3sqr-hsqr,2.)+rhosqr*(c3sqr+hsqr)) ) ; coeff[2] = Math.pow(rhosqr-hsqr,2.) +2.*esqr*(2.*hsqr*hsqr-2.*c3sqr*hsqr-2.*rhosqr*hsqr-rhosqr*c3sqr) +esqr*esqr*Math.pow(c3sqr-hsqr,2.) ; coeff[3] = 2.* Math.pow(altit*eccen,2.)*(rhosqr-hsqr+esqr*(c3sqr-hsqr)) ; coeff[4] = Math.pow(altit*eccen,4.) ; /* Small number of loops for self-consistency, using some Horner scheme for * the function to be zeroed and its first derivative */ for(int loop=0; loop < 4 ;loop++) { System.out.println("# taum2 "+tau2) ; tau2 -= ( coeff[0]+tau2*(coeff[1]+tau2*(coeff[2]+tau2*(coeff[3]+tau2*coeff[4]))) )/( coeff[1]+tau2*(2.*coeff[2]+tau2*(3.*coeff[3]+tau2*4.*coeff[4])) ) ; } return Math.sqrt(tau2) ; } /** Derivative d tau/d lambda. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return absolute value of the derivative according to eq (64) of the preprint. */ double dtaudlambda(double tau, double c3) { /* treat the factors (N+h)^2*(1-tau^2) and the sqrt(..)/(h+M) separately */ return GaussE(tau)*dtauds(tau,c3)/c3 ; } /** Derivative d s/d lambda. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return absolute value of the derivative of the path length according to eq (63) of the preprint. */ double dsdlambda(double tau, double c3) { return GaussE(tau)/c3 ; } /** Discriminant under the square root of d tau / ds. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return 1-tau^2-(c3/(N+h))^2. */ double discrT(final double tau, final double c3) { /* N+altitude */ final double Nh = curvN(tau)+altit ; return 1.-tau*tau-Math.pow(c3/Nh,2.) ; } /** Derivative of T with respect to tau. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return derivative of discrT() with respect to tau */ double dTdtau(final double tau, final double c3) { final double N = curvN(tau) ; final double Nh = N+altit ; return 2*tau*( N*Math.pow(c3*eccen/Nh,2.)/(Nh*(1.+eccen*tau)*(1.-eccen*tau)) -1. ) ; } /** Second derivative of T with respect to tau. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return the second derivative of discrT() with respect to tau */ double d2Tdtau2(final double tau, final double c3) { final double N = curvN(tau) ; final double Nh = N+altit ; final double e2 = eccen*eccen ; final double Oneetau = (1.+eccen*tau)*(1.-eccen*tau) ; return -2.*( 1.-N*c3*c3*e2*(1.+3.*altit*e2*tau*tau/Nh/Oneetau)/Math.pow(Nh,3.)/Oneetau ) ; } /** Derivative d tau/d s [1/m]. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return absolute value of the derivative according to eq (61) of the preprint. */ double dtauds(double tau, double c3) { return Math.sqrt( discrT(tau,c3) )/(altit+curvM(tau)) ; } /** Second derivative d^2 tau/d lambda^2. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return value of the derivative of eq (64) of the preprint. */ double d2taudlambda2(double tau, double c3) { /* The array contains values of E, sqrt(T) and 1/(h+M) in the components facto[][0], * and their first derivatives with respect to tau in facto[][1]. */ double[][] facto = new double[3][2] ; facto[0][0] = GaussE(tau) ; facto[1][0] = Math.sqrt(discrT(tau,c3) ) ; facto[2][0] = 1./( curvM(tau)+altit) ; facto[0][1] = dEdtau(tau) ; /* derivative of square root is one half of the inverse multiplied by interior derivative */ facto[1][1] = dTdtau(tau,c3)/(2.*facto[1][0]) ; /* derivative of 1/(h+M) is negative of its square multiplied by interior derivative */ facto[2][1] = -dMdtau(tau)*facto[2][0]*facto[2][0] ; /* The Leibniz rule of derivatives is applied to collect the derivative of * the product of facto[i][0], i=0..2. */ final double resul = facto[0][1]*facto[1][0]*facto[2][0] +facto[0][0]*facto[1][1]*facto[2][0] +facto[0][0]*facto[1][0]*facto[2][1] ; /* resul contains so far the derivative (d/dtau) (d tau/d lambda), without * the c3 in the denominator. Postmultiply with d tau/ d lambda to generate * the d^2 tau/ d lambda^2. */ return dtaudlambda(tau,c3)*resul/c3 ; } /** Third derivative d^3 tau/d lambda^3. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return value of the 2nd derivative of eq (64) of the preprint. */ double d3taudlambda3(double tau, double c3) { final double T = discrT(tau,c3) ; /* Faa di Bruno formula: * d^3 tau/d lambda^3 = (d^2tau/d lambda^2)* (d/dtau) (dtau/dlambda)+ (dtau/dlambda)^2* (d^2/dtau^2) (dtau/dlambda */ /* The array contains values of E, sqrt(T) and 1/(h+M) in the components [][0], * and their first derivatives with respect to tau in [][1], 2nd derivatives in [][2]. */ double[][] facto = new double[3][3] ; facto[0][0] = GaussE(tau) ; facto[1][0] = Math.sqrt(T) ; facto[2][0] = 1./( curvM(tau)+altit) ; facto[0][1] = dEdtau(tau) ; /* derivative of square root is one half of the inverse multiplied by interior derivative */ final double Tprime = dTdtau(tau,c3) ; facto[1][1] = Tprime/(2.*facto[1][0]) ; /* derivative of 1/(h+M) is negative of its square multiplied by interior derivative */ final double Mprime = dMdtau(tau) ; facto[2][1] = -Mprime*facto[2][0]*facto[2][0] ; facto[0][2] = d2Edtau2(tau) ; /* First derivative of sqrt(T) was dTdtau/(2*sqrt(T)). The 2nd derivative therefore is * [ (d^2/dtau^2) T -(d T/d tau)^2/(2T)]/[2sqrt(T)], which we store in facto[1][2]. */ facto[1][2] = ( d2Tdtau2(tau,c3) -Tprime*Tprime/( 2.*T ) )/(2.*facto[1][0]) ; /* Second derivative of 1/(h+M). First one was -(dM/dtau)/(h+M)^2. Second one is /* -(d^2M/dtau^2)/(h+M)^2+2(dM/dtau)^2/(h+M)^3. */ facto[2][2] = ( 2.*Mprime*Mprime*facto[2][0]-d2Mdtau2(tau) )*facto[2][0]*facto[2][0] ; /* The value of (d/d tau) (d tau/d lambda). The implementation is not * tuned for quick evaluation, since the same value is calculated again in #d2taudlambda2. */ final double dtaudlambdadtau = ( facto[0][1]*facto[1][0]*facto[2][0] +facto[0][0]*facto[1][1]*facto[2][0] +facto[0][0]*facto[1][0]*facto[2][1] )/c3 ; /* the value of (d^2/dtau^2) (dtau/dlambda) with Leibniz rule, multinomial case */ final double d3taudlambdadtau2 = ( facto[0][2]*facto[1][0]*facto[2][0] +2.*facto[0][1]*facto[1][1]*facto[2][0] +2.*facto[0][1]*facto[1][0]*facto[2][1] +facto[0][0]*facto[1][2]*facto[2][0] +2.*facto[0][0]*facto[1][1]*facto[2][1] +facto[0][0]*facto[1][0]*facto[2][2] )/c3 ; /* Faa di Bruno rule of derivatives is applied to collect the derivative * (d^2/dlambda^2) dtau/dlambda * = (d^2 tau/dlambda^2)*(d/dtau) (d tau/d lambda)+(d tau/d lambda)^2* (d^2/d tau^2) (d tau/d lambda) */ return d2taudlambda2(tau,c3)*dtaudlambdadtau + Math.pow(dtaudlambda(tau,c3),2.)* d3taudlambdadtau2; } /** Second derivative d^2 s/d lambda^2 = (d/d lambda) (E/c3). * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return value of the 2nd derivative of eq (63) of the preprint. */ double d2sdlambda2(double tau, double c3) { /* First the derivative (d/dtau) (d s/d lambda), without * the c3 in the denominator. Postmultiply with d tau/ d lambda to generate * the d^2 s/ d lambda^2. */ return dtaudlambda(tau,c3)*dEdtau(tau)/c3 ; } /** Third derivative d^3 s /d lambda ^3. * This is currently implemented as a dummy to return 0. Effectively * the order of the approximation remains second order. * @param tau sine of the latitude * @param c3 the constant c3 along one individual trajectory * @return value of the third derivative of eq (67) of the preprint */ double d3sdlambda3(double tau, double c3) { return 0. ; } /** Spherical approximation to parameter c3 * @param phi the latitudes at start and end [rad] * @param lambd the longitudes at start and end [rad] * @return the parameter c3 obtained form the spherical approximation. * @see eq (A19) in the preprint. */ double c3Sphere(final double[] phi, final double lambd[]) { /* cosine of the angle Z */ final double cosZ = Math.sin(phi[0])*Math.sin(phi[1]) +Math.cos(phi[0])*Math.cos(phi[1])*Math.cos(lambd[1]-lambd[0]) ; return (altit+rho_equat)*Math.cos(phi[0])*Math.cos(phi[1])*Math.sin(lambd[1]-lambd[0])/Math.sqrt(1.-cosZ*cosZ) ; } /** Approximate sign d tau / ds * @param phi the latitudes at start and end [rad] * @param lambd the longitudes at start and end [rad] * @return the sign of the square root in eq (64) of the preprint, as obtained from * the spherical approximation. */ double dtaudsSignum(final double[] phi, final double lambd[]) { /* cosine of the angle Z */ final double cosZ = Math.sin(phi[0])*Math.sin(phi[1]) +Math.cos(phi[0])*Math.cos(phi[1])*Math.cos(lambd[1]-lambd[0]) ; return Math.signum(Math.sin(phi[1])-Math.sin(phi[0])*cosZ) ; } /** Adjust longitude of the final point. * @param lambd the longitudes at start and end [rad] * The value of lambd[1] is adjusted to the range lambd[0]+-pi to ensure that the * trajectory chooses a path along the correct hemisphere. */ static void adjLambdaEnd(double lambd[]) { final double dl = lambd[1]-lambd[0] ; if ( Math.abs( dl+2.*Math.PI) < Math.abs(dl) ) lambd[1] += 2.*Math.PI ; if ( Math.abs( dl-2.*Math.PI) < Math.abs(dl) ) lambd[1] -= 2.*Math.PI ; } /** Nautical course in the local oblique horizontal [rad] * @param tau sine of the latitude * @param c3 the directional parameter of the solution to the inverse problem [m] * @param north sign of the square root, +1 or -1 * @return the direction angle (North over East) in the local oblique tangential plane [rad] * @see eq (78) of the preprint */ double nautAngle(final double tau, final double c3, final double north) { /* sinkappa and coskappa are the sine and cosine of the angle multiplied by * a common positive factor, which retains the correct quadrant of the solution. */ final double sinkappa = c3/(curvN(tau)+altit) ; final double coskappa = north* Math.sqrt( discrT(tau,c3) ) ; return Math.atan2(sinkappa,coskappa) ; } /** Compute one trajectory over the lambda interval, assuming c3 given. * @param phi start and (target) value of the geodetic latitude [rad] * @param lambd start and end value of the longitude [rad] * @param c3 fixed parameter of c3 * @param Nsampl number of steps into which the interval lambd[0]-lambd[1] is divided * @param usampl if positive, each usampl’th point of the result is printed * @param useRad one value of the AngleUnit to indicate which units are preferred for angles * @param taylOrd the order of the Taylor approximation in the FEM, must be 2 or 3. * @return the overshooting of the value of tau versus the sine of phi[1] [rad] */ double tauShoot(final double[] phi, final double lambd[], final double c3, final int Nsampl, final int usampl, final int useRad, final int taylOrd) { /* step width in the longitudes, including a sign */ final double dlam = (lambd[1]-lambd[0])/Nsampl ; /* current value of tau = sin(phi), initialized with the value at the start */ double tau = Math.sin(phi[0]) ; /* current value of s, the length along the line, initialized with the value at the start */ double s = 0. ; /* the sign which chooses a N or a S route. +1 or -1 */ double north = dtaudsSignum(phi,lambd) ; if ( usampl > 0 ) System.out.println("\n\n") ; double kappa = nautAngle(tau,c3,north) ; System.out.println("# start course "+ kappa +" rad, " + Math.toDegrees(kappa)+ " deg") ; /* loop over the finite elements */ for(int i=0; i < Nsampl; i++) { /* in each of the elements, a 2nd order Taylor approximation * for dtau/ds is applied with (64) of the preprint */ double deriv1 = north*dtaudlambda(tau,c3) ; /* below, the ’north’ factor appears twice and cancels */ double deriv2 = d2taudlambda2(tau,c3) ; double deriv3 = (taylOrd >= 3) ? north*d3taudlambda3(tau,c3) : 0.; /* in each of the elements, a 2nd order Taylor approximation * for ds/dlambda is applied with (64) of the preprint */ double deriv1s = dsdlambda(tau,c3) ; double deriv2s = d2sdlambda2(tau,c3) ; double deriv3s = (taylOrd >= 3) ? d3sdlambda3(tau,c3) : 0.; /* Taylor extrapolation. Current value +(df/dx)*delta+(1/2)(d^2/dx^2)*delta^2+(1/6)*(d^3/dx^3)*delta^3 * Update to the value at the right end of the interval. */ tau += dlam*(deriv1+dlam*(0.5*deriv2+dlam*deriv3/6.) ) ; /* We may have passed by a maximum of the equatorial distance * and sense this by comparing the location of the extremum on the current parabolic * estimator for tau(lambda) with the current interval. Setting the derivative of the * Taylor estimator to zero, (df/dx)+(d^2/dx^2)*(lambda-current lambda)=0, the * location of the extremum is -(df/dx)/( d^2f/dx^2) relative to the current lambda. * @todo implement the 3rd order Taylor Approximation as well. */ final double lamFlat = -deriv1/deriv2 ; /* Update of the path length along the trajectory in the Taylor approximation */ s += dlam*(deriv1s+dlam*(0.5*deriv2s+dlam*deriv3s/6.) ) ; if ( usampl > 0 ) if ( i % usampl == 0 || i == Nsampl-1 ) { /* longitude at right end of interval [rad] */ final double longi = lambd[0]+(i+1)*dlam ; double[] cart = getCartesian(tau, longi ) ; /* Print the x,y,z Cartesian coordinates as the first 3 columns per line */ System.out.print(""+cart[0]+" "+cart[1]+" "+cart[2]) ; kappa = nautAngle(tau,c3,north) ; /* Print longitude, geodetic latitude, nautical angle as columns 4 to 6 */ if ( useRad == AngleUnit.RAD ) System.out.print(" "+ longi +" "+ Math.asin(tau)+ " " +kappa ) ; else if ( useRad == AngleUnit.DEG ) System.out.print(" "+ Math.toDegrees(longi) + " " + Math.toDegrees(Math.asin(tau)) +" "+Math.toDegrees(kappa) ) ; else System.out.print(" "+ 10.*Math.toDegrees(longi)/9. + " " + 10.*Math.toDegrees(Math.asin(tau))/9. +" "+10.*Math.toDegrees(kappa)/9. ) ; /* Print length of trajectory in column 7 */ System.out.print(" "+s) ; System.out.println("") ; } if ( lamFlat/dlam >= 0. && lamFlat/dlam < 1.) { north *= -1. ; } } /* tau now is the estimate of the final position; return the error */ return tau-Math.sin(phi[1]) ; } /** Perform 4 iterations on adjusting the parameter c3. * @param phi the latitudes at start and end [rad] * @param lambd the longitudes at start and end [rad] * @param Nsampl the number of divisions of the longitude axis * @param usampl undersampling factor for the printout in the converged trajectory * @param useRad the unit of the angles which are reported * @param taylOrd the order of the Taylor expansion of the core integral, 2 or 3. * @return a convergent to the parameter c3 */ public double c3shoot(final double[] phi, final double lambd[],final int Nsampl, final int usampl, final int useRad, final int taylOrd) { /* For a parabolic (2nd order) approximation of the shooting * error (units of tau) use 3 abscissa and target missing parameters */ double[] err = new double[4] ; double[] c3spread = new double [4] ; /* start with the estimate from the spherical approximation, zero eccentricity */ c3spread[0] = c3Sphere(phi,lambd) ; err[0] = tauShoot(phi,lambd,c3spread[0],Nsampl,-1,useRad,taylOrd) ; System.out.println("# c3 (spherical) "+c3spread[0]+" err "+err[0]) ; /* A stepping parameter for the variation */ final double c3Delt = -c3spread[0]*0.005 ; c3spread[1] = c3spread[0]+c3Delt ; err[1] = tauShoot(phi,lambd,c3spread[1],Nsampl,-1,useRad,taylOrd) ; /* estimate of the 3rd c3 by a linear interpolation * between the previous 2 to get zero error. The ansatz is err=err[0]+(err[1]-err[0])/c3Delt*(c-c[0]) = 0, * where (err[1]-err[0])/c3Delt is the slope of the error curve, which is solved for c. */ c3spread[2] = c3spread[0]-c3Delt*err[0]/(err[1]-err[0]) ; err[2] = tauShoot(phi,lambd,c3spread[2],Nsampl,-1,useRad,taylOrd) ; System.out.println("# c3 (linear extrapol) "+c3spread[2]+" err "+err[2]) ; /* Lagrange quadratic interpolation between (c3spred[0],err[0]), (c3spread[1],err[1]) * and (c3srepad[2],err[2]): err = eofCinter[0]+eofCinter[1]*c3spread+eofCinter[2]*c3spread^2 * up to a common factor which is omitted. */ double[] eofCinter = new double [3] ; eofCinter[0] = err[2]*c3spread[0]*c3spread[1]*(c3spread[0]-c3spread[1]) +err[1]*c3spread[0]*c3spread[2]*(c3spread[2]-c3spread[0]) +err[0]*c3spread[1]*c3spread[2]*(c3spread[1]-c3spread[2]) ; eofCinter[1] = err[2]*(Math.pow(c3spread[1],2.)-Math.pow(c3spread[0],2.)) + err[1]*(Math.pow(c3spread[0],2.)-Math.pow(c3spread[2],2.)) + err[0]*(Math.pow(c3spread[2],2.)-Math.pow(c3spread[1],2.)) ; eofCinter[2] = err[2]*(c3spread[0]-c3spread[1]) +err[1]*(c3spread[2]-c3spread[0]) +err[0]*(c3spread[1]-c3spread[2]) ; /* Set this quadratic to zero; normalize to monic polynomial */ eofCinter[0] /= eofCinter[2] ; eofCinter[1] /= eofCinter[2] ; eofCinter[2] = Math.sqrt(0.25*Math.pow(eofCinter[1],2)-eofCinter[0]) ; /* From the two solutions take the one closer to the previous estimate */ c3spread[3] = (c3spread[2]+0.5*eofCinter[1] > 0.) ? -0.5*eofCinter[1]+eofCinter[2] : -0.5*eofCinter[1]-eofCinter[2] ; err[3] = tauShoot(phi,lambd,c3spread[3],Nsampl,usampl,useRad,taylOrd) ; System.out.println("# c3 (quadratic extrapol) "+c3spread[3]+" err "+err[3]) ; return c3spread[3] ; } /** * usage: * <tt> * java Geod [-h altitude/m] [-s #longitude steps] [-u sfac] [-R equat radius/m] [-e eccent] [-r] * [-g] [-2] phi1 lam1 phi2 lam2 * </tt> * <br> * Command line parameters * <ul> * <li> * -h followed by a floating point number specifies the altitude of the * trajectory above the ellipsoid. This is given in the same units as those of the * parameter of the equatorial radius. The default is zero. * </li> * <li> * -s followed by an integer denotes the number of steps between the starting and * final point of the trajectory in the finite element method. The default is 400. * </li> * <li> * -u followed by an integer specifies an undersampling factor for listing the * parameters of the trajectory on standard output. The elements actually listed * are given by the sampling steps divided by the undersampling factor. * The default is 10. * </li> * <li> * -R followed by a floating point number defines the equatorial radius of * the surface of the ellipsoid in the same units as the option for the altitude. * The default is the WGS84 value. * </li> * <li> * -e followed by a floating point number defines the eccentricity of the * surface of the ellipsoid. The default is the WGS84 value. * </li> * <li> * -r without any additional number specifies that the four mandatory command * line parameters are provided in units of radians. The default is degrees. * </li> * <li> * -g without any additional number specifies that the four mandatory command * line parameters are provided in units of gons. The default is degrees. * </li> * </ul> * There are four mandatory parameters, the goedetic latitude and longitude * of the start of the trajectory, and the geodetic latitude and longitude of * the final point of the trajectory. * <br> * The standard output contains comment lines (starting with hashes) and * one line per point on the trajectory with the following columns (separated by blanks): * <ul> * <li> * The first three colums are the x, y and z Cartesian coordinate of the point in the * same units as with the equatorial radius and altitude parameters. * </li> * <li> * Columns 4 and 5 are the longitude and geodetic latitude of the point * in the same units (deg or rad) as the command line parameters. * </li> * <li> * Column 6 is the nautical angle in the local topocentric tangential plane * in the same units (deg or rad or gon) as the command line parameters. * </li> * <li> * Column 7 is the distance along the geodetic * in the same units as the command line parameters. * </li> * </ul> * <br> * Examples * <ul> * <li> * <tt> * java Geod -s 1000 -e 0.08227185417541244347812117 -R 6378206.4 8.973611111 -79.573333333333 21.435000000 -158.02583333 * </tt> * computes the trajectory for the case from Panama to Hawaii by Paul D. Thomas, * J. Geophys. Res. vol 70 (1965) p 3331 in his Figure 3. * <li> * <tt> * java Geod -R 637838.799919243 -g -e 0.0815815883368028 54.262654 0 11.977469 115.959876 * </tt> * computes the trajectory for the case from Paris to Saigon by H. M. Dufour * in Bull. Geo. vol 32 (1953) p 26, with f=1/300, angles in gons. * </ul> * @since 2008-04-12 */ public static void main(String[] args) throws Exception { /* default equatorial radius set to WGS84 */ double rho = WGS84_RHO_E ; /* default eccentricity set to WGS84 * that is 0.08195646813308029492885488 ; */ double e = (2.-1./WGS84_FLAT)/WGS84_FLAT ; /* default altitude above ellipsoid is zero */ double h = 0. ; /* default number of elements */ int Nsampl = 400 ; /* Default undersampling factor for the printed output. */ int usampl = 10 ; /* Default Taylor order in the FEM is 3. */ int taylOrd = 3 ; /* no defaults for the coordinates */ double[] phi = new double[2] ; double[] lambd = new double [2] ; /* units of final angular arguments. */ int useRad = AngleUnit.DEG ; /* parse options and parameters */ for(int argc =0 ; argc < args.length; argc++) { if ( args[argc].compareTo("-h") == 0) h = Double.parseDouble(args[++argc]) ; else if ( args[argc].compareTo("-s") == 0 ) Nsampl = Integer.parseInt(args[++argc]) ; else if ( args[argc].compareTo("-u") == 0 ) usampl = Integer.parseInt(args[++argc]) ; else if ( args[argc].compareTo("-R") ==0 ) rho = Double.parseDouble(args[++argc]) ; else if ( args[argc].compareTo("-e") ==0 ) e = Double.parseDouble(args[++argc]) ; else if ( args[argc].compareTo("-r") ==0 ) useRad = AngleUnit.RAD ; else if ( args[argc].compareTo("-g") ==0 ) useRad = AngleUnit.GON ; else if ( args[argc].compareTo("-2") == 0 ) taylOrd = 2 ; else { phi[0] = Double.parseDouble(args[argc++]) ; lambd[0] = Double.parseDouble(args[argc++]) ; phi[1] = Double.parseDouble(args[argc++]) ; lambd[1] = Double.parseDouble(args[argc++]) ; } } /* If user provided degrees or gons on the command line, convert to radians for further processing * 400 gons are 360 degrees. */ switch( useRad) { case AngleUnit.DEG : for(int i=0 ; i < 2 ; i++) { phi[i] = Math.toRadians(phi[i]) ; lambd[i] = Math.toRadians(lambd[i]) ; } break ; case AngleUnit.GON : for(int i=0 ; i < 2 ; i++) { phi[i] = Math.toRadians(360.*phi[i]/400.) ; lambd[i] = Math.toRadians(360.*lambd[i]/400.) ; } break ; } /* normalize input such that the final longitude is on the same hemisphere */ adjLambdaEnd(lambd) ; /* log the input parameters */ System.out.println("# equat Radius "+ rho+", eccentricity "+e+", altitude "+h) ; System.out.println("# inverse flattening "+ 1./(1.-Math.sqrt(1.-e*e)) ) ; System.out.println("# start "+phi[0] + " rad, "+lambd[0]+" rad; end "+phi[1] + " rad, "+lambd[1]+" rad") ; switch( useRad) { case AngleUnit.DEG : System.out.println("# start "+Math.toDegrees(phi[0]) + " deg, "+ Math.toDegrees(lambd[0]) +" deg; end "+Math.toDegrees(phi[1]) + " deg, "+Math.toDegrees(lambd[1])+" deg") ; break ; case AngleUnit.GON : System.out.println("# start "+10.*Math.toDegrees(phi[0])/9. + " gon, "+ 10.*Math.toDegrees(lambd[0])/9. +" gon; end "+10.*Math.toDegrees(phi[1])/9. + " gon, "+10.*Math.toDegrees(lambd[1])/9.+" gon") ; } System.out.println("# "+Nsampl + " elements, Taylor order " + taylOrd) ; /* define the shape of the ellipsoid */ Geod g = new Geod(rho,e,h) ; /* Determine the main parameter of the bundle of geodesics through initial point to * solve the inverse problem, and print parameters along the trajectory * in the 4th iteration. */ g.c3shoot(phi,lambd,Nsampl,usampl,useRad,taylOrd) ; } } /* Geoid */ ## References * Abramowitz and Stegun (1972) Abramowitz, M., and I. A. Stegun (eds.), 1972, _Handbook of Mathematical Functions_ (Dover Publications, New York), 9th edition, ISBN 0-486-61272-4. * Bouasse (1919) Bouasse, H., 1919, _Geographie Mathématique_ (Librairie Delagrave, Paris). * Bowring (1983) Bowring, B. R., 1983, Bull. Geod. 57(1–4), 109. * Byrd and Friedman (1971) Byrd, P. F., and M. D. Friedman, 1971, _Handbook of elliptical Integrals for Engineers and Physicists_ , volume LXVII of _Die Grundlehren der mathematischen Wissenschaften in Einzeldarstellungen_ (Springer, Berlin, Göttingen), 2nd edition. * Dorrer (1999) Dorrer, E., 1999, in _Quo vadis geodesia…?_ , edited by F. Krumm and V. S. Schwarze (Universität Stuttgart), Technical Reports of Geodesy and GeoInformatics, ISSN 0933-2839, URL http://www.uni-stuttgart.de/gi/research/schriftenreihe/quo_vadis/contents.html. * Dufour (1958) Dufour, H. M., 1958, Bull. Geod. 32(2), 26. * Fukushima (2006) Fukushima, T., 2006, J. Geod. 79(12), 689. * Gradstein and Ryshik (1981) Gradstein, I., and I. Ryshik, 1981, _Summen-, Produkt- und Integraltafeln_ (Harri Deutsch, Thun), 1st edition, ISBN 3-87144-350-6. * Hradilek (1976) Hradilek, L., 1976, Bull. Geod. 50(4), 301. * Jones (2004) Jones, G. C., 2004, J. Geod. 76(8), 437. * Kaplan (2006) Kaplan, G. H., 2006, arXiv:astro-ph/0602086 . * Long (1975) Long, S. A. T., 1975, Cel. Mech. 12(2), 225. * McCarthy and Petit (2003) McCarthy, D. D., and G. Petit, 2003, _IERS Technical Note No 32_ , Technical Report, IERS Convention Centre, URL http://www.iers.org/iers/publications/tn/tn32/. * National Imagery and Mapping Agency (2000) National Imagery and Mapping Agency, 2000, _Department Of Defense World Geodetic System 1984_ , Technical Report TR8350.2, NIMA, URL http://earth-info.nga.mil/GandG/publications/tr8350.2/tr8350_2.html. * Ölander (1952) Ölander, V. R., 1952, Bull. Geod. 26(3), 337. * Pollard (2002) Pollard, J., 2002, J. Geod. 76(1), 36. * Reichardt (1957) Reichardt, H., 1957, _Vorlesungen über Vektor- und Tensorrechnung_ , volume 34 of _Hochschulbücher für Mathematik_ (Deutscher Verlag der Wissenschaften, Berlin). * Saito (1970) Saito, T., 1970, Bull. Geod. 44(4), 341. * Smart (1949) Smart, W. M., 1949, _Text-book on Spherical Astronomy_ (Cambridge University Press, Cambridge), 4th edition. * Sodano (1958) Sodano, E. M., 1958, Bull. Geod. 32(2), 13. * Thomas (1965) Thomas, P. D., 1965, J. Geophys. Res. 70(14), 3331. * Tienstra (1951) Tienstra, J. M., 1951, Bull. Geod. 25(1), 7. * Vermeille (2004) Vermeille, H., 2004, J. Geod. 78(1–2), 94. * You (1999) You, R.-J., 1999, in _Quo vadis geodesia…?_ , edited by F. Krumm and V. S. Schwarze (Universität Stuttgart), Technical Reports of Geodesy and GeoInformatics, ISSN 0933-2839, URL http://www.uni-stuttgart.de/gi/research/schriftenreihe/quo_vadis/contents.html. * Zhang _et al._ (2005) Zhang, C.-D., H. T. Hsu, X. P. Wu, S. S. Li, Q. B. Wang, H. Z. Chai, and L. Du, 2005, J. Geod. 79(8), 413.

arxiv-papers

2024-09-04T02:48:53.329243

0711.1380

# Preliminaries in Many-Particle Quantum Gravity. Einstein–Friedmann Spacetime L.A. Glinka laglinka@gmail.com, glinka@theor.jinr.ru Bogoliubov Laboratory of Theoretical Physics, Joint Institute for Nuclear Research, 141980 Dubna, Russia ###### Abstract Preliminaries for the Many-Particle approach to quantization of the Einstein–Hilbert theory of gravitation are presented in this paper. The Einstein–Friedmann Spacetime is detailed discussed from this point of view, and the Von Neumann–Araki–Woods second quantization in the Fock space of the canonical Hamilton field equations is presented. In result the Spacetime as system with the Bose–Einstein statistics is described and thermodynamics of the system is computed. Quantum Cosmology, Quantum Gravity, Boson systems, Einstein–Hilbert theory ###### pacs: 04.60.-m, 98.80.Qc, 05.30.Jp ## I Introduction In this paper introductory results of the Fock space formulation of Quantum Gravity are discussed by example of Einstein–Friedmann Spacetime. Firstly the classical Dirac–Arnowitt–Deser–Misner Hamiltonian approach to General Relativity for the Spacetime is briefly presented, and the Wheeler–DeWitt evolution as primary quantization of constraints is formulated. Next evolution is separated on the canonical Hamilton field equations system, and consistent the von Neumann–Araki–Woods quantization in the Fock space is realized. For diagonalization the Bogoliubov–Heisenberg basis is used, and internal logarithmic conformal field theory structure in this basis is proved. By treatment of the Spacetime as the Bose–Einstein system thermodynamics of the system is computed. ## II Classical approach ### II.1 The Dirac constraints The flat, homogenous and isotropic Spacetime investigated by Einstein e and Friedmann f is described by an interval $ds^{2}=a^{2}(\eta)\left[(d\eta)^{2}-(dx^{i})^{2}\right],~{}~{}d\eta=N_{d}(x^{0})dx^{0},$ (1) here $a(\eta)$ is the Friedmann conformal scale factor, $\eta$ is the Dirac–Arnowitt–Deser–Misner conformal time d ; adm , and $N_{d}$ is lapse function. The Hilbert action h for (1) can be written as $\mathcal{A}[a]=\int{dx^{0}}\left\\{p_{a}\dfrac{da}{dx^{0}}-N_{d}\left[-\dfrac{p_{a}^{2}}{4V_{0}^{2}}+\rho(a)\right]V_{0}\right\\},$ (2) where $p_{a}$ is nontrivial canonical conjugate momentum, $\rho(a)$ is energy density $\displaystyle p_{a}=-\dfrac{2V_{0}}{N_{d}}\dfrac{da}{dx^{0}},~{}~{}\rho(a)=\dfrac{a^{4}}{V_{0}}\int{d^{3}x}~{}\mathcal{H}_{M},$ (3) and $V_{0}=\int{d^{3}}x<\infty$ is finite space volume. Variational principle with respect to $N_{d}$ applied to action (2) leads to constraints $\dfrac{\delta\mathcal{A}[a]}{\delta N_{d}}=0\Rightarrow-\dfrac{p_{a}^{2}}{4V_{0}^{2}}+\rho(a)=0,$ (4) with solution given by the Hubble law $\int_{a_{I}}^{a}\dfrac{da}{\sqrt{\rho(a)}}=\pm|\eta-\eta_{I}|,$ (5) where index $I$ means initial data. ### II.2 The Hamilton field equations One can see that the primary constraints (4) simplify to a form $p_{a}^{2}-{\omega}^{2}(a)=0,~{}~{}\omega(a)\equiv\pm 2V_{0}\sqrt{\rho(a)},$ (6) and that direct canonical quantization $i\left[\hat{\mathrm{p}}_{a},a\right]=1,~{}~{}\hat{\mathrm{p}}_{a}\equiv-i\dfrac{\partial}{\partial{a}},$ (7) leads to the Wheeler–DeWitt evolution equation w1 ; w2 $\left(\dfrac{\partial^{2}}{\partial{a^{2}}}+{\omega}^{2}(a)\right)\Psi(a)=0.$ (8) This equation can be represented as the canonical Hamilton field equations system $\left\\{\begin{array}[]{ccc}\dfrac{\partial}{\partial\Pi_{\Psi}}\mathrm{H}(\Pi_{\Psi},\Psi)\\!\\!\\!\\!\\!&=&\\!\\!\\!\\!\\!\dfrac{\partial}{\partial{a}}\Psi,\vspace*{5pt}\\\ -\dfrac{\partial}{\partial\Psi}\mathrm{H}(\Pi_{\Psi},\Psi)\\!\\!\\!\\!&=&\\!\dfrac{\partial}{\partial{a}}\Pi_{\Psi},\end{array}\right.$ (9) where considered Hamiltonian $\mathrm{H}(\Pi_{\Psi},\Psi)$ has a form $\mathrm{H}(\Pi_{\Psi},\Psi)=\dfrac{1}{2}\left(\Pi^{2}_{\Psi}+{\omega}^{2}(a)\Psi^{2}\right).$ (10) In further text we will use compact form of (9) $\dfrac{\partial}{\partial{a}}\left[\begin{array}[]{c}\Psi\\\ \Pi_{\Psi}\end{array}\right]=\left[\begin{array}[]{cc}0&1\\\ -{\omega}^{2}(a)&0\end{array}\right]\left[\begin{array}[]{c}\Psi\\\ \Pi_{\Psi}\end{array}\right].$ (11) ## III Quantization We focus attention on the Hamilton field equations (11). ### III.1 Von Neumann–Araki–Woods quantization Essence of quantization constitutes general transition between classical fields and field operators $\left[\begin{array}[]{c}\Psi(a)\\\ \Pi_{\Psi}(a)\end{array}\right]\longrightarrow\left[\begin{array}[]{c}\mathbf{\Psi}[a(\eta)]\\\ \mathbf{\Pi_{\Psi}}[a(\eta)]\end{array}\right],$ (12) with standard Canonical Commutation Relations (CCRs) $\left[\mathbf{\Pi_{\Psi}}[a(\eta)],\mathbf{\Psi}[a(\eta^{\prime})]\right]=-i\delta\left(a(\eta)-a(\eta^{\prime})\right).$ (13) Problem of CCRs representations is not new, and was investigated in ccr1 and ccr2 . In g was proposed the Von Neumann–Araki–Woods quantization in the Fock space $\left(\mathcal{G},\mathcal{G}^{\dagger}\right)$ of annihilation and creation operators $\left[\begin{array}[]{c}\mathbf{\Psi}\\\ \mathbf{\Pi_{\Psi}}\end{array}\right]=\left[\begin{array}[]{cc}\dfrac{1}{\sqrt{2{\omega}(a)}}&\dfrac{1}{\sqrt{2{\omega}(a)}}\\\ -i\sqrt{\dfrac{{\omega}(a)}{2}}&i\sqrt{\dfrac{{\omega}(a)}{2}}\end{array}\right]\left[\begin{array}[]{c}\mathcal{G}[a(\eta)]\\\ \mathcal{G}^{\dagger}[a(\eta)]\end{array}\right],$ (14) which is consistent with (13) by CCRs $\displaystyle\left[\mathcal{G}[a(\eta)],\mathcal{G}^{\dagger}[a(\eta^{\prime})]\right]$ $\displaystyle=$ $\displaystyle\delta(a(\eta)-a(\eta^{\prime})),$ (15) $\displaystyle\left[\mathcal{G}[a(\eta)],\mathcal{G}[a(\eta^{\prime})]\right]$ $\displaystyle=$ $\displaystyle 0.$ (16) Quantization (14) applied to equations (11) gives $\dfrac{\partial}{\partial{a}}\left[\begin{array}[]{c}\mathbf{\Psi}\\\ \mathbf{\Pi_{\Psi}}\end{array}\right]\\!=\\!\left[\begin{array}[]{cc}0&1\\\ -{\omega}^{2}(a)&0\end{array}\right]\left[\begin{array}[]{c}\mathbf{\Psi}\\\ \mathbf{\Pi_{\Psi}}\end{array}\right],$ (17) or in terms of the Fock space $\dfrac{\partial}{\partial{a}}\left[\begin{array}[]{c}\mathcal{G}[a(\eta)]\\\ \mathcal{G}^{\dagger}[a(\eta)]\end{array}\right]\\!=\\!\left[\begin{array}[]{cc}-i{\omega}(a)&\Delta\\\ \Delta&i{\omega}(a)\end{array}\right]\left[\begin{array}[]{c}\mathcal{G}[a(\eta)]\\\ \mathcal{G}^{\dagger}[a(\eta)]\end{array}\right],$ (18) where $\Delta=\partial_{a}\ln\left|\sqrt{\dfrac{\omega(a)}{\omega(a_{I})}}\right|$, and $\omega(a_{I})=\pm 4V_{0}\sqrt{\rho(a_{I})}$ is initial value of $\omega(a)$. ### III.2 The Bogoliubov–Heisenberg basis Evolution (18) should be diagonalize in order to build correct quantum theory. We apply the Bogoliubov automorphism for Bose–Einstein systems, which lies in accordance with considered field equations (8) $\displaystyle\left[\begin{array}[]{cc}\mathcal{W}[a(\eta)]\\\ \mathcal{W}^{\dagger}[a(\eta)]\end{array}\right]$ $\displaystyle=$ $\displaystyle\left[\begin{array}[]{cc}u(a)&v(a)\\\ v^{\ast}(a)&u^{\ast}(a)\end{array}\right]\left[\begin{array}[]{cc}\mathcal{G}[a(\eta)]\\\ \mathcal{G}^{\dagger}[a(\eta)]\end{array}\right],$ (26) $\displaystyle|u(a)|^{2}-|v(a)|^{2}=1.$ This transformation preserves CCRs (15) and (16) in new basis $\displaystyle\left[\mathcal{W}[a(\eta)],\mathcal{W}^{\dagger}[a(\eta^{\prime})]\right]$ $\displaystyle=$ $\displaystyle\delta(a(\eta)-a(\eta^{\prime})),$ (27) $\displaystyle\left[\mathcal{W}[a(\eta)],\mathcal{W}[a(\eta^{\prime})]\right]$ $\displaystyle=$ $\displaystyle 0.$ (28) Next we reduce evolution (18) to Heisenberg type one $\dfrac{\partial}{\partial{a}}\left[\begin{array}[]{c}\mathcal{W}[a(\eta)]\\\ \mathcal{W}^{\dagger}[a(\eta)]\end{array}\right]=\left[\begin{array}[]{cc}-i\lambda&0\\\ 0&i\lambda\end{array}\right]\left[\begin{array}[]{c}\mathcal{W}[a(\eta)]\\\ \mathcal{W}^{\dagger}[a(\eta)]\end{array}\right].$ (29) In result we obtain $\lambda=0$, $\mathcal{W}\equiv\mathrm{w}=\mathit{constant}$, and vacuum state $|0\rangle$ defined by relation $\mathrm{w}|0\rangle=0$. Evolution (18) is equivalent to equations on functions $u$ and $v$ $\dfrac{\partial}{\partial{a}}\left[\begin{array}[]{c}v(a)\\\ u(a)\end{array}\right]=\left[\begin{array}[]{cc}-i{\omega}(a)&-\Delta\\\ -\Delta&i{\omega}(a)\end{array}\right]\left[\begin{array}[]{c}v(a)\\\ u(a)\end{array}\right],$ (30) that can be solved in hyperbolic parametrization $\displaystyle v(a)=e^{i\theta(a)}\sinh\phi(a),~{}~{}u(a)=e^{i\theta(a)}\cosh\phi(a).$ (31) Here functions–parameters $\displaystyle\theta(a)=\int_{a_{I}}^{a}p_{a}da,~{}\phi(a)=\ln{\left|\sqrt{\dfrac{\omega(a_{I})}{\omega(a)}}\right|}.$ (32) are classical phase integral and logarithmic conformal field. In this manner we have proved that theory builded by the Von Neumann–Araki–Woods quantization in form (14) has internal logarithmic conformal field theory structure. ## IV Physical implications Now some physical results will present. ### IV.1 Density functional We take density functional operator in standard form of occupation number consistent with quantization (14) $\varrho=\mathcal{G}^{\dagger}\mathcal{G},$ (33) which in the Bogoliubov–Heisenberg basis has a following matrix representation $\varrho=\left[\begin{array}[]{cc}|u|^{2}&-uv\\\ -u^{\ast}v^{\ast}&|v|^{2}\end{array}\right].$ (34) One can see that the Von Neumann entropy for considered system has the Boltzmann one form $\mathrm{S}\equiv-\dfrac{\mathrm{tr}(\varrho\ln\varrho)}{\mathrm{tr}(\varrho)}=\ln\left\\{\dfrac{1}{2|u|^{2}-1}\right\\}.$ (35) It creates opportunity to formulate of thermodynamics for quantized the Einstein–Friedmann Spacetime represented by entropy (35). ### IV.2 Thermodynamics of Spacetime In opposite to results of paper g , in this text we propose use of (10) as the Hamiltonian of theory. Theory constructed by this Hamiltonian is conceptually simpler than theory presented in g , because in (10) superfluidity term is absent. However, as it will turn out physical consequences presented below are principally the same as in g . The Hamiltonian (10) has a following matrix representation in the Bogoliubov–Heisenberg basis $\displaystyle\mathrm{H}=\left[\begin{array}[]{cc}\dfrac{|u|^{2}+|v|^{2}}{2}\omega&-uv\omega\\\ -u^{\ast}v^{\ast}\omega&\dfrac{|u|^{2}+|v|^{2}}{2}\omega\end{array}\right].$ (38) Statistical mechanics understand internal energy as average $\mathrm{U}=\dfrac{\mathrm{tr}(\varrho\mathrm{H})}{\mathrm{tr}\varrho}$, which for considered case is $\mathrm{U}=\left(\dfrac{1}{2}+\dfrac{4\mathrm{n}+3}{2\mathrm{n}+1}\mathrm{n}\right)\omega(a).$ (39) Here we introduced quantity $\mathrm{n}=\langle 0|\mathcal{G}^{\dagger}\mathcal{G}|0\rangle=\dfrac{1}{4}\Bigg{|}\sqrt{\left|\dfrac{\omega(a_{I})}{\omega(a)}\right|}-\sqrt{\left|\dfrac{\omega(a)}{\omega(a_{I})}\right|}\Bigg{|}^{2},$ (40) which is number of particles produced from vacuum. Averaged number of particles is $\langle\mathrm{n}\rangle=2\mathrm{n}+1.$ (41) Chemical potential is defined as $\mu=\dfrac{\partial\mathrm{U}}{\partial\mathrm{n}}$, in this case $\mu=\left(1+\dfrac{1}{(2\mathrm{n}+1)^{2}}-\dfrac{1}{2}\dfrac{4\mathrm{n}+1}{4\mathrm{n}^{2}+2\mathrm{n}}\sqrt{\dfrac{\mathrm{n}}{\mathrm{n}+1}}\right)\omega(a).$ (42) Description of the system according to quantum theory principles demands using of the thermal Gibbs ensemble. The system is characterized by CCRs (27, 28). It leads to identification of function under logarithm in (35) with the Bose–Einstein partition function $\dfrac{1}{2|u|^{2}-1}\equiv\left\\{\exp\left[\dfrac{\mathrm{\mathrm{U}-\mu\mathrm{n}}}{\mathrm{T}}\right]-1\right\\}^{-1}.$ (43) Using of relations (26), (39), (40), (42), and (43) give the formula for temperature of considered system $\displaystyle\mathrm{T}=\dfrac{1+\left(\dfrac{2\mathrm{n}}{2\mathrm{n}+1}\right)^{2}+\dfrac{8\mathrm{n}^{2}+8\mathrm{n}+1}{4\mathrm{n}+2}\sqrt{\dfrac{\mathrm{n}}{\mathrm{n}+1}}}{2\ln(2\mathrm{n}+2)}\omega(a).$ (44) This formula describes temperature of the system as a function of Friedmann conformal scale factor $a$. ## V Summary Above formalism points at direct way to the Quantum Cosmology understood as many-particle open quantum system. On base of this type reasoning formulation of physics is obvious and clear - in presented case description of the Einstein–Friedmann Spacetime as the quantum Bose–Einstein system leads to well-defined thermodynamics. Furthermore, for considered Spacetime open quantum system point of view gives opportunity to formulate Quantum Gravity as theory with internal structure of logarithmic conformal field theory. In opinion of the author presented way creates substantially wider opportunities for construction of similar formalism for general gravitational fields and to research connections between geometry and physics in terms of logarithmic conformal field theory. ## Acknowledgements The author is especially thankful for valuable discussions to B.M. Barbashov, I. Białynicki-Birula, A. Borowiec, S.J. Brodsky, G. `t Hooft, J. Lukierski, V.N. Pervushin, V.B. Priezzhev, and C. Rovelli. ## References * (1) Einstein A., The meaning of relativity (Pricenton University Press, Princeton, 1922). * (2) Friedmann A.A., Z. Phys. 10, 377, (1922). * (3) Dirac P.A.M., Phys. Rev. 114, 924, (1959). * (4) Arnowitt R., Deser S., Misner C.W., Phys. Rev. 122, 997, (1961). * (5) Hilbert D., Gott. Nachr. 27, 395, (1915). * (6) Wheeler J.A., in Battelle Rencontres: 1967 Lectures in Mathematics and Physics, edited by C.M. DeWitt and J.A. Wheeler (New York, 1968), p. 242. * (7) DeWitt B.S., Phys. Rev. 160, 1113, (1967). * (8) Neumann von J., Math. Ann 104, 570, (1931). * (9) Araki H., Woods E.J., J. Math. Phys 4, 637, (1963). * (10) Glinka L.A., SIGMA 3, 087, (2007).

arxiv-papers

2024-09-04T02:48:53.344544

0711.2010

# A Polynomial Time Algorithm for Graph Isomorphism Reiner Czerwinski Institut für Softwaretechnik und Theoretische Informatik TU Berlin ###### Abstract Algorithms testing two graphs for isomorphism known as yet in computer science have exponential worst case complexity. In this paper we propose an algorithm that has polynomial complexity and constructively supplies the evidence that the graph isomorphism lies in P. ## 1 Introduction Graph isomorphism is a crucial problem in computer science and has been investigated intensely in the past. Applications can be found both inside and outside the computer science area. Symmetry breaking is an important example, which is a current topic in constraint solving and may be used for instance in register allocation, which is a critical performance issue. In chemistry, graph isomorphism algorithms are employed for molecule analysis. Known algorithms like the ones presented by Nauty [1] and VF [2] do indeed indicate an instant isomorphism efficiently in the majority of cases. Yet, for each of these techniques, graphs leading to an exponential growth in calculation time can be constructed. For a long time, the existence of a method with polynomial runtime was in doubt, albeit Köbler [3],[4] had expressed that there were reasons to believe graph isomorphism not to be NP hard, unlike the subgraph isomorphism problem or the travelling salesman problem for which it is assumed that fast algorithms do not exist. He pointed out that the problem of finding an isomorphism is equally hard to the one of finding all isomorphisms which is unusual in NP-hard problems. It has been well known [5] that a necessary (but not sufficient) criterion for graph isomorphism are identical eigenvalues of the adjacency matrix. This has been exploited [6] to develop an algorithm for the class of graphs with bounded eigenvalues. The matrices will still have same eigenvalues, if you perturbs the diagonal elements of the matrices, such that a vertex in the first and in the second graph have same values for the diagonal elements, if they are candidates for an isomorphism. By iterative perturbation of the diagonal elements the two graphs can be checked for isomorphism in polynomial time. A program for graph isomorphism with polynomial run time was already implemented by Trofimov and Smolenskii two years ago [7] by using this trick. ## 2 Terms and linear algebra A graph $G=(V,E)$ consists of a set of vertices $V$ and a set of edges $E$. Two vertices $v1$ and $v2$ are connected or adjacent iff $\\{v1,v2\\}\in E$. Two graphs $G=(V,E)$ and $G^{\prime}=(V^{\prime},E^{\prime})$ are isomorph, iff a bijective mapping $\pi:V\rightarrow V^{\prime}$ exists with $\\{v1,v2\\}\in E\Leftrightarrow\\{\pi(v1),\pi(v2)\\}\in E^{\prime}$ . In the adjacency matrix an entry is 1 in the case when the two respective vertices are connected through an edge in $E$, otherwise 0. As we will only consider undirectional graphs, the adjacency matrix is symmetric. Köbler [4] has proven this to be sufficient as the graph isomorphism problem for directed graphs can be reduced to the one for undirected ones. A symmetric matrix can be decomposed by means of eigenvalue decomposition into $A=V*D*V^{T}$, $D$ being a diagonal matrix, whose main diagonal entries are filled by the eigenvalues of the matrix A. $V$ is a unitary matrix, that is $V^{T}*V=I$. The columns of $V$ are the eigenvectors of $A$. If $V=(v_{1},...,v_{n})$ and $D=diag(\mu_{1},...,\mu_{n})$, then $Au_{i}=\mu_{i}u_{i}$. $v_{1},...,v_{n}$ are orthonormal[8], p. 291, thus $<v_{i},v_{j}>=v_{i}^{T}v_{j}=\begin{cases}1&\text{if $i=j$ }\\\ 0&\text{otherwise}\end{cases}$ . Let $\\{\mu_{1},...,\mu_{m}\\}$ be the set of distinct eigenvalues. Associated with the eigenvalues $\mu_{i}$ is the eigenspace $S_{i}$ containing the eigenvectors associated with $\mu_{i}:S_{i}=\\{x\in R^{n}\mid Ax=\mu_{i}x\\}$ . By virtue of the symmetry of A, we have: 1. 1. If $\mu_{i}$ is an eigenvalue with multiplicity $m_{i}$ then $S_{i}$ has dimension $m_{i}$. 2. 2. The direct sum $S_{1}\oplus S_{2}\oplus...\oplus S_{m}$ equals $R^{n}$ 3. 3. If $i\not=j$ then $S_{i}$ and $S_{j}$ are mutually orthogonal. ## 3 Derivation If two graphs G and G’ are isomorphic, so for the respective adjacency matrices $A^{\prime}=P*A*P^{T}$ must hold, where $P$ denotes a permutation matrix. This entails $A^{\prime k}=P*A^{k}*P^{T}$ for arbitrary $k$. $A^{\prime}$ and $A$ have the same eigenvalues, because $P$ is an unitary matrix. ###### Proposition 1. If $A$ is an adjacency matrix of a graph and $P$ is a permutation matrix, then $P$ describes an automorphism of the graph, iff $PA=AP$. The proof is written in [5]. ###### Lemma 1. Let $A$ and $A^{\prime}=P*A*P^{T}$ be adjacency matrices of isomorphic graphs and $D$ a diagonal matrix, so their characteristic polynomials are equal and $\chi(A+D)=\chi(A^{\prime}+P*D*P^{T})$ holds, i. e. the two matrices have the same eigenvalues. ###### Proof. $A^{\prime}+P*D*P^{T}=P*(A+D)*P^{T}$, applies, which results in the eigenvalues and consequently the characteristic polynomials being equal. ∎ ###### Lemma 2. Let $A$ and $B$ be two positive definite matrices, and it furthermore, let $\pi$ be a permutation $\pi$ with $\sum_{j=1}^{n}(A^{k}_{ji})^{2}=\sum_{j=1}^{n}(B^{k}_{j\pi_{i}})^{2}$ for $k=1,...,n$, then they possess identical eigenvalues and consequently identical characteristic polynomials. ###### Proof. Let $e_{1},...,e_{n}$ be the unit vectors, let $\mu_{1},...,\mu_{m}$ be the possibly multiple eigenvalues and $v_{i1},..,v_{im_{i}}$ the eigenvectors belonging to $\mu_{i}$. $e_{i}$ can be written as a linear combination of the eigenvectors. Let now be $e_{i}=\sum_{l=1}^{m}\sum_{p=1}^{m_{l}}x_{il_{p}}v_{lp}$, so applies: $A^{k}*e_{i}=\sum_{l=1}^{m}\left(\mu_{l}^{k}*\sum_{p=1}^{m_{l}}x_{il_{p}}v_{lp}\right)$ is the $i$th column of $A^{k}$, called $a^{(k)}_{i}$. We observe that $\displaystyle{a^{(k)}_{i}}^{T}a^{(k)}_{i}$ $\displaystyle=\sum_{l=1}^{m}\left(\mu_{l}^{k}*\sum_{p=1}^{m_{l}}x_{il_{p}}v_{l_{p}}\right)^{T}\sum_{l=1}^{m}\left(\mu_{l}^{k}*\sum_{p=1}^{m_{l}}x_{il_{p}}v_{l_{p}}\right)$ $\displaystyle=\left\langle\sum_{l=1}^{m}\mu_{l}^{k}*\sum_{p=1}^{m_{l}}x_{il_{p}}v_{l_{p}}\quad,\quad\sum_{l=1}^{m}\mu_{l}^{k}*\sum_{p=1}^{m_{l}}x_{il_{p}}v_{l_{p}}\right\rangle$ $\displaystyle=\sum_{l=1}^{m}\sum_{l^{\prime}=1}^{m}\mu_{l}^{k}\mu_{l^{\prime}}^{k}*\sum_{p=1}^{m_{l}}\sum_{p^{\prime}=1}^{m_{l^{\prime}}}x_{il_{p}}x_{il^{\prime}_{p^{\prime}}}\langle v_{l_{p}},v_{{l^{\prime}}_{p^{\prime}}}\rangle$ $\displaystyle=\sum_{l=1}^{m}\mu_{l}^{k*2}*\sum_{p=1}^{m_{l}}x_{il_{p}}^{2}=\sum_{l=1}^{m}\mu_{l}^{2k}E_{il}$ . $E_{il}^{0.5}$ is thereby the component of the vector $e_{i}$ that is within the vector space that $\mu_{l}$ spans. Once $E_{il}$ are known, we can unambiguously determine the $\mu_{l}^{2}$ from the linear system of equations $\sum_{l=1}^{m}(\mu_{l}^{2})^{k}E_{il}={a^{(k)}_{i}}^{T}a^{(k)}_{i}$ with $i\in\\{1,...,n\\},k\in\\{1,...,m\\}$ according to Vandermonde and, as all eigenvalues are positive, the $\mu_{l}$ as well. Yet, the $E_{il}$ still have to be identified. The fact that the eigenvectors form an orthonormal basis, yields the condition $\sum_{l=1}^{m}E_{il}=1$ for $i=1,...,m$. This property results in a linear system of equations which is unambiguous given that all values are positive. As this argumentation equally holds for $B$, both matrices have the same eigenvalues. ∎ ###### Gerschgorin’s Disc Theorem 1. The eigenvalues of a matrix $A=(a_{ij})$ are inside the Gerschgorin discs $G_{i}=\\{z\in C:|z-a_{ii}\leq\sum_{j=1,j\not=i}^{n}|a_{ij}|\\}\quad i=1,..,n\quad.$ Every disc holds an eigenvalue. More information about the theorem can be read, for example, in [9]. If $A$ is symmetric, the eigenvalues are real, thus every eigenvalue is in one interval $[a_{ii}-\sum_{j=1,j\not=i}^{n}|a_{ij}|\;,\;a_{ii}+\sum_{j=1,j\not=i}^{n}|a_{ij}|]$. ###### Lemma 3. Let $B=D+A$ with $D=diag(d_{1},...,d_{n})$ is a diagonal matrix and $A$ is an adjacency matrix of an undirected graph. Let $|d_{i}-d_{j}|\geq 4n$ and $d_{i}>n-1$ when $d_{i}\not=d_{j}$, otherwise $||B^{k}*e_{i}||^{2}=||B^{k}*e_{j}||^{2}$ for all $k$. If $d_{i}=d_{j}$ then $j$ is in the orbit of $i$, i.e. there is an automorphism $\psi$ with $j=\psi(i)$ ###### Proof. Let $N(w):=\\{v|v$ is adjacent to vertex $w\\}$, then $Be_{i}=d_{i}*e_{i}+\sum_{i^{\prime}\in N(i)}e_{i^{\prime}}$ . If $d_{i}=d_{j}$, the vertices $i$ and $j$ have the same degrees, because $||Be_{i}||^{2}=d_{i}^{2}+\sum_{i^{\prime}\in N(i)}1^{2}=d_{j}^{2}+\sum_{j^{\prime}\in N(j)}1^{2}=||Be_{j}||^{2}$. $e_{i}$ can be written $e_{i}=\sum_{j=1}^{n}x_{ij}v_{j}$, where $v_{1},...,v_{n}$ are eigenvectors, which are an orthonormal basis. Thus $1=\langle e_{i},e_{i}\rangle=\langle\sum_{j=1}^{n}x_{ij}v_{j},\sum_{j=1}^{n}x_{ij}v_{j}\rangle=\sum_{j=1}^{n}\sum_{l=1}^{n}x_{ij}x_{il}\langle v_{j},v_{l}\rangle\\\ =\sum_{j=1}^{n}x_{ij}^{2}\langle v_{j},v_{j}\rangle+\sum_{j\not=l}x_{ij}x_{il}\langle v_{j},v_{l}\rangle=\sum_{j=1}^{n}x_{ij}^{2}*1+\sum_{j\not=l}x_{ij}*0=\sum_{j=1}^{n}x_{ij}^{2}$ and $\displaystyle e_{i}^{T}Be_{i}$ $\displaystyle=\langle e_{i},Be_{i}\rangle$ $\displaystyle=\left\langle\sum_{j=1}^{n}x_{ij}v_{j}\quad,\quad B\sum_{j=1}^{n}x_{ij}v_{j}\right\rangle$ $\displaystyle=\left\langle\sum_{j=1}^{n}x_{ij}v_{j}\quad,\quad\sum_{j=1}^{n}Bx_{ij}v_{j}\right\rangle$ $\displaystyle=\left\langle\sum_{j=1}^{n}x_{ij}v_{j}\quad,\quad\sum_{j=1}^{n}\mu_{j}x_{ij}v_{j}\right\rangle$ $\displaystyle=\sum_{j=1}^{n}\sum_{l=1}^{n}\mu_{l}x_{ij}x_{il}\langle v_{j},v_{l}\rangle$ $\displaystyle=\sum_{j=1}^{n}\mu_{j}x_{ij}^{2}=d_{i}$ and it is also $\displaystyle(Be_{i})^{T}(Be_{i})$ $\displaystyle=\langle Be_{i},Be_{i}\rangle$ $\displaystyle=\left\langle B\sum_{j=1}^{n}x_{ij}v_{j}\quad,\quad B\sum_{j=1}^{n}x_{ij}v_{j}\right\rangle$ $\displaystyle=\left\langle\sum_{j=1}^{n}Bx_{ij}v_{j}\quad,\quad\sum_{j=1}^{n}Bx_{ij}v_{j}\right\rangle$ $\displaystyle=\left\langle\sum_{j=1}^{n}x_{ij}\mu_{j}v_{j}\quad,\quad\sum_{j=1}^{n}\mu_{j}x_{ij}v_{j}\right\rangle$ $\displaystyle=\sum_{j=1}^{n}\sum_{l=1}^{n}\mu_{j}\mu_{l}x_{ij}x_{il}\langle v_{j},v_{l}\rangle$ $\displaystyle=\sum_{j=1}^{n}\mu_{j}^{2}x_{ij}^{2}=d_{i}^{2}+\delta_{i}\quad\text{,}$ where $\delta_{i}$ is the degree of vertex $i$. Because of Gerschgorin’s disc theorem $B$ is positive definite, thus $e_{i}$ and $e_{j}$ are equally distributed to the eigenspaces as shown in Lemma 2, if $d_{i}=d_{j}$. Let $E_{ir}=\sum_{j\in\\{j|v_{j}\in S_{r}\\}}x_{ij}^{2}$ be the energy of $e_{i}$ in the eigenspace $S_{r}=\\{v|Bv=\mu_{r}v\\}$, then $\sum_{r=1}^{m}E_{ir}=1$. We can look at $E_{i1},...,E_{im}$ as if it is a probability distribution. In this case $E[\mu_{r}]=\sum_{r=1}^{m}\mu_{r}E_{ir}=<e_{i},Be_{i}>=d_{i}$ and $Var[\mu_{r}]=E[\mu_{r}^{2}]-E[\mu_{r}]^{2}=\sum_{r=1}^{m}\mu_{r}^{2}E_{ir}-[\sum_{r=1}^{m}\mu_{r}E_{ir}]^{2}=<Be_{i},Be_{i}>-<e_{i},Be_{i}>=d_{i}^{2}+\delta_{i}-d_{i}^{2}=\delta_{i}$. By Chebyshev’s inequality there will be $Pr(|\mu_{r}-d_{i}|\geq 2n)\leq\frac{\delta_{i}}{4n^{2}}<\frac{n}{4n^{2}}=\frac{1}{4n}$. If an eigenvalue $\mu_{r}$ is outside the Gerschgorin disc, then $E_{ir}<\frac{1}{4n}$. Let $e_{i}$ and $e_{j}$ two vectors with $\forall rE_{ir}=E_{jr}$, so the energy of $Be_{i}$ and $Be_{j}$ is also equaly distributed to the eigenspaces and so $\forall r\sum_{i^{\prime}\in N(i)}E_{i^{\prime}r}=\sum_{j^{\prime}\in N(j)}E_{j^{\prime}r}$. For any $d$ let $k_{i}$ the number of vertices $i^{\prime}\in N(i)$ with $d_{i^{\prime}}=d$ and $k_{j}$ the number of vertices $j^{\prime}\in N(j)$ with $d_{j^{\prime}}=d$ . $k_{i}=k_{j}$, because : Assumption: $k_{j}<k_{i}$. Let $G(d)$ be the Gerschgorin disk $[d-2n,d+2n]$. Let $E_{G(d)i}=\sum_{r\in\\{r\mid\mu_{r}\in G(D)\\}}E_{ri}$ the energy in the eigenspaces with eigenvalues inside the Gerschgorin disc, then $\sum_{i^{\prime}\in N(i)}E_{G(d)i^{\prime}}\geq k_{i}(1-\frac{1}{4n})>k_{i}-\frac{1}{4}$ and $\sum_{j^{\prime}\in N(j)}E_{G(d)j^{\prime}}\leq k_{j}+(N(j)-k_{j})\frac{1}{4n}<k_{j}+\frac{1}{4}\quad.$ From that follows $k_{i}=k_{j}$. So there is a bijection between adjacent vertices of $i$ and $j$ with $l$ and $\psi(l)$ with $d_{l}=d_{\psi(l)}$. This way we find a bijection on every vertex of the graph, which is described by a permutation matrix $P$. If $i$ and $j$ are vertices with $e_{j}=Pe_{i}$, then $B_{ii}=d_{i}=d_{j}=B_{jj}$ and so $E_{ir}=E_{jr}$ for every $r\in\\{1,...,m\\}$. Let $e_{i}=\sum_{r=1}^{m}u_{ir}$ and $e_{j}=\sum_{r=1}^{m}u_{jr}$, where $u_{ir}$ and $u_{jr}$ are eigenvectors in the eigenspace $S_{r}$, then $||u_{ir}||^{2}=E_{ir}=E_{jr}=||u_{ir}||^{2}$. $\forall r\;u_{jr}=Pu_{ir}$, because $A^{k}P\sum_{r}\mu_{r}^{k}u_{jr}=A^{k}Pe_{i}=A^{k}e_{j}=\sum_{r}\mu_{r}^{k}u_{jr}$ for every $k$ and $\mu_{r}>0$ are distinct eigenvalues, thus a linear system of equations with exactly one solution is build. This causes $PBe_{i}=P*\sum_{r=1}^{m}\mu_{r}u_{ir}=\sum_{r=1}^{m}\mu_{r}P*u_{ir}=\sum_{r=1}^{m}\mu_{r}u_{jr}=B*e_{j}=B*Pe_{j}$, thus $PB=BP$. For the adjacency matrix $A=B-diag(d_{1},...,d_{n})$ we get $PA=AP$ and so $P$ is an automorphism, as proven in [5]. ∎ ## 4 Algorithms ### 4.1 Algorithm 1 input : | adjacency matrices of graphs $A$ und $A^{\prime}$ ---|--- output: | graphs isomorphic ? $C1_{1}$ := $\\{1,..,n\\}$ --- $C2_{1}$ := $\\{1,..,n\\}$ $A1$ := $A$ $A2$ := $A^{\prime}$ repeat | $c$ := number of $C1_{x}$ | for $i$:=$1$ to $n$ | | A1[i,i] := $4n*x$, where $i\in C1_{x}$ | | A2[i,i] := $4n*x$, where $i\in C2_{x}$ | for $k$:=$1$ to $n$ | | for $i$:=$1$ to $n$ | | | $a1_{i}^{(k)}$:= $i$th column vector of $A1^{k}$ | | | $a2_{i}^{(k)}$:= $i$th column vector of $A2^{k}$ | | | $\hat{A1}[i][k]$ := ${a1_{i}^{(k)}}^{T}a1_{i}^{(k)}$ | | | $\hat{A2}[i][k]$ := ${a2_{i}^{(k)}}^{T}a2_{i}^{(k)}$ | sort lists $\hat{A1}$ and $\hat{A2}$ lexicographically | compare $\hat{A1}$ and $\hat{A2}$ | if sorted lists are not equal | | then return false | put vertices into equivalence classes $C1_{x}$, such that $i\equiv j$ iff $\hat{A1}[i]=\hat{A1}[j]$ | put vertices into equivalence classes $C2_{x}$, such that $i\equiv j$ iff $\hat{A2}[i]=\hat{A2}[j]$ | | and $\hat{A1}[l]=\hat{A2}[i]$ iff $l\in C1_{x}$ and $i\in C2_{x}$ until $c=$ number of $C_{x}$ return true If the graphs are isomorph, the algorithm will return true, because not only the adjacent matrices have the same spectra, also the matrices A1 ad A2 have the same spectra in every iteration step because of Lemma 1. If Algorithm 1 returns true, we can look at the graph as two not connected parts of a big graph. Because of Lemma 3 we find for every vertex in the first graph a vertex in the second graph, with is in the same orbit, belonging to the big graph. That way, the algorithm construct an isomorphism. ### 4.2 Algorithm 2 Algorithm to construct an isomorphism input : | adjacency matrices of two isomorphic graphs $A$ und $A^{\prime}$ ---|--- output: | graph isomorphism, if exists | make $A1$,$A2$,$C1$,$C2$ and $c$ as described in Algorithm 1 ---|--- | while $c<n$ | | get $x$ with $|C1_{x}|=|C2_{x}|>1$ | | get an $i\in C1_{x}$ and $j\in C2_{x}$ | | $A1[i,i]:=A2[j,j]:=4*(c+1)*n$ | | for $k$:=$1$ to $n$ | | | for $i$:=$1$ to $n$ | | | | $a1_{i}^{(k)}$:= $i$th column vector of $A1^{k}$ | | | | $a2_{i}^{(k)}$:= $i$th column vector of $A2^{k}$ | | | | $\hat{A1}[i][k]$ := ${a1_{i}^{(k)}}^{T}a1_{i}^{(k)}$ | | | | $\hat{A2}[i][k]$ := ${a2_{i}^{(k)}}^{T}a2_{i}^{(k)}$ | | put vertices into equivalence classes $C1_{x}$, such that $i\equiv j$ iff $\hat{A1}[i]=\hat{A1}[j]$ | | put vertices into equivalence classes $C2_{x}$, such that $i\equiv j$ iff $\hat{A2}[i]=\hat{A2}[j]$ | | | and $\hat{A1}[l]=\hat{A2}[i]$ iff $l\in C1_{x}$ and $i\in C2_{x}$ | | $c$ := number of $C1_{x}$ | sort lists $\hat{A1}$ and $\hat{A2}$ lexicographically | return permutation induced by the sorted lists When $c=n$, we have only single eigenvalues for $A1$ and $A2$. In this case, there is only one solution for $A1=P*A2*P^{T}$ with $P$ unitary, the permutation the algorithm has produced. Now $A=P*A^{\prime}*P^{T}$ because of Lemma 1 and so the permutation is an isomorphism of the graphs. ## 5 Complexity The main effort of Algorithm 1 consists in $n$ matrix multiplications in every iteration of the outer loop. The outer loop is a fixed point iteration, because it terminates, when no more equivalence classes can be constructed. In case of constructing new classes, the number of classes will grow. Algorithm 1 must not make more then $n$ iterations in the outer loop. In the outer loop of Algorithm 2 vertex $i$ of the first graph and vertex $j$ of the second graph will put into a single class separated to the other vertices. Algorithm 2 also have not more then $n$ iterations in outer loop. Both algorithms consists in $n^{2}$ matrix multiplications in $n$ iterations of the outer loop. Thus we have to perform $O(n^{8})$ integer additions or multiplications. The maximum eigenvalue of $A1$ is according to Gerschgorin’s circle theorem $\leq 4\frac{n(n+1)}{2}=2n^{2}+2n$. Consequently the maximum eigenvalue of $A1^{n}\leq(2n^{2}+2n)^{n}$. Since all matrix entries are positive, none of them can be bigger than the maximum eigenvalue and therefore $\leq(2n^{2}+2n)^{n}$. The maximum value occurs in calculating the sum of the quadratic entries and is $\leq n*(2n^{2}+2n)^{2n}$. The coding length for an entry is therefore $(6n)*(\log(n)+C)=O(n*\log(n))$. Algorithm 2 needs also at most $O(n^{2})$ matrix multiplications with maximum coding length of an entry in $O(n*\log(n))$. ## 6 Literature 1. 1. McKay, Brendan D. (1981), Practical graph isomorphism, Congressus Numerantium 30: 10th. Manitoba Conference on Numerical Mathematics and Computing (Winnipeg, 1980) 2. 2. amalfi.dis.unina.it/graph/db/papers/benchmark.pdf 3. 3. Köbler, Johannes; Schöning, Uwe & Torán, Jacobo (1992), Graph isomorphism is low for PP, Computational Complexity 2 (4) 4. 4. Köbler, Johannes; Schöning, Uwe & Torán, Jacobo (1993), The Graph Isomorphism Problem: Its Structural Complexity, Birkhäuser, ISBN 978-0817636807. 5. 5. Biggs, Norman (1974), Algebraic Graph Theory, Cambridge University Press 6. 6. Babai, László; Grigoryev, D. Yu. & Mount, David M. (1982), Isomorphism of graphs with bounded eigenvalue multiplicity, pp. 310-324, Proceedings of the 14th Annual ACM Symposium on Theory of Computing 7. 7. Trofimov, M. I.; Smolenskii, E. A. (2005), Application of the electronegativity indices of organic molecules to tasks of chemical informatics, Russian Chemical Bulletin, International Edition, Vol. 54, No. 9, pp. 2235-2246 8. 8. Bronstein, I. N.; Semendjajew, K. A.; Musiol, G.; Mühlig, H. (2001), Taschenbuch der Mathematik, 5. Auflage, Verlag Harri Deutsch, Thun und Frankfurth am Main 9. 9. Plato, Robert (2003), Concise Numerical Mathematics, Graduate Studies in Mathematics, Volume 57, American Mathematical Society

arxiv-papers

2024-09-04T02:48:53.355284

0711.2605

# Generalized D-Forms Have No Spurious Creases Erik D. Demaine111Partially supported by NSF CAREER award CCF-0347776, DOE grant DE-FG02-04ER25647, and AFOSR grant FA9550-07-1-0538. and Gregory N. Price222Partially supported by an NSF Graduate Fellowship. MIT Computer Science and Artificial Intelligence Laboratory 32 Vassar St., Cambridge, MA 02139, USA –edemaine,price˝@mit.edu ###### Abstract A convex surface that is flat everywhere but on finitely many smooth curves (or _seams_) and points is a _seam form_. We show that the only creases through the flat components of a seam form are either between vertices or tangent to the seams. As corollaries we resolve open problems about certain special seam forms: the flat components of a D-form have no creases at all, and the flat component of a pita-form has at most one crease, between the seam’s endpoints. ## 1 Introduction Given any metric space $S$ with the topology and local geometry required of the surface of a convex three-dimensional body, there is exactly one convex body up to isometry whose surface has the intrinsic geometry of $S$. This is the endpoint of a line of research pursued in the middle of the last century by Alexandrov and Pogorelov [Ale50], and it implies a strong correspondence between the geometry of a convex body and the intrinsic geometry of its surface. On the other hand, the exact nature of this correspondence is not yet well understood—which properties in the surface geometry imply what properties in the body geometry, and vice versa. For example, if $S$ is a _D-form_ , obtained by sewing together two smooth convex shapes of the same perimeter, then even for this special case the most basic questions are open. These forms were invented by an artist [Wil] and introduced into the literature in [PW01]. The latter study poses three problems: (1) when is the D-form the convex hull of a space curve, (2) when are the two pieces free of creases, and (3) how can one compute the D-form numerically from the two shapes. A later treatment [DO07] suggests an informal argument for Problem 1 (arguing that the D-form is always the convex hull of its seam) and leaves Problems 2 and 3 open. The same book [DO07] introduces also a related special case where $S$ is obtained by sewing up a single smooth convex shape along its boundary in one seam, calling these _pita-forms_ and suggesting, based on paper experiments, that pita-forms might never have creases. We resolve Problems 1 and 2: both D-forms and pita-forms are always the convex hull of their seams, and (excluding the seam) D-forms are always free of creases but a pita-form may have one crease. Our results apply to a natural generalization of both D-forms and pita-forms, the _seam form_ , which roughly consists of intrinsically flat pieces joined along finitely many seams. Because the original sources of the problem are stated informally, we first introduce precise definitions that we believe capture the intuitive picture. Then we show the following theorems: ###### Theorem 1. Every three-dimensional convex body is the convex hull of the nonflat points on its surface. ###### Corollary 2. Every seam form is the convex hull of its seams and vertices. ###### Theorem 3. In a flat component of a seam form, every crease lies on a line segment composed of creases, and each endpoint of such a segment is either a strict vertex or a point of tangency to a seam. ###### Corollary 4. The flat components of a D-form are without creases; in the flat component of a pita-form, the only crease(s) make up the line segment between the endpoints of the seam. Intuitively the line segment between a pita-form’s endpoints in Corollary 4 should be thought of as one “crease”; it is a consequence of our definitions, below, that this segment may be arbitrarily subdivided into several segments we call creases. Problem 3, to efficiently compute the three-dimensional shape of a D-form or seam form from its two-dimensional intrinsic geometry, has now been largely resolved. To make this problem well posed, one needs a finite representation of the input geometry, which is most naturally done by a piecewise-linear or polyhedral approximation. With considerable effort, the problem of reconstructing a three-dimensional convex polyhedron from its intrinsic geometry can be reduced to the solution of a high-dimensional ordinary differential equation [BI06]. The numerical solution of this equation appears to be achievable efficiently in practice, and is provably achievable within pseudopolynomial time [KPD09]. We introduce terminology in Section 2, prove Theorem 1 and its Corollary 2 in Section 3, and prove Theorem 3 and its Corollary 4 in Section 4. In Section 5 we describe counterexamples that show the necessity of some of the hypotheses in our results. ## 2 Background and Notation For us a surface is a metric 2-manifold embedded in $\mathbb{R}^{3}$. The surface is $C^{k}$ if the manifold and its embedding are $C^{k}$. The surface is piecewise-$C^{k}$ if it can be decomposed as a complex of vertices, $C^{k}$ open edges, and $C^{k}$ open regions. A good surface is a piecewise-$C^{2}$ surface. A good surface $S$ therefore decomposes into a union of $C^{2}$ surfaces $S_{i}$, called pieces, $C^{2}$ edges $\gamma_{j}$, which we call semicreases, and vertices. If $S$ is itself $C^{1}$ everywhere on a semicrease, we call it a proper semicrease; otherwise it is a crease. (This conservative definition of crease, where some parts may be $C^{1}$ but not $C^{2}$, only broadens our characterization of creases in seam forms.) A point on a surface is flat if it has a neighborhood isometric to a region in the plane. A surface or part of a surface is called flat if all of its points are flat. A surface $S$ is convex if $S\subseteq\partial X$ for some bounded convex body $X$ in $\mathbb{R}^{3}$. A normal to a convex body $X$ at a point $x$ is a unit vector $n$ with $n\cdot x=\sup_{x^{\prime}\in X}n\cdot x^{\prime}$. The relation between points on the boundary of $X$ and their normals is traditionally called the Gauss map, though it need not be a map—one point may have many normals. We write $G(x)$ for the normals at $x$, and $G(U)$ for all the normals to any point in $U\subset X$. Observe that $G(x)$ is always a convex subset of the sphere. A consequence of Gauss’ celebrated Theorema Egregium [Gau02] is that a convex surface $U$ is flat just if $G(U)$ has zero area. If $G(x)$ has positive spherical area, then we call $x$ a strict vertex. The $C^{2}$ condition prevents a strict vertex $x$ from being on a semicrease or a piece, so for good surfaces, strict vertices are indeed vertices. Figure 1: A D-form, constructed by sewing together two ellipses (at left). The solid line is the seam, and the dotted lines are the false edges from projection into the page. A seam form $S$ is a good convex surface in which each piece $S_{i}$ is flat. It is simple to verify that such a surface decomposes uniquely into maximal connected flat open subcomplexes, which we call flat components, and some leftover semicreases and vertices, the connected components of which we call seams. A simple seam form is a seam form in which each flat component is isometric to a convex plane region, and a simple smooth seam form is a simple seam form in which these regions have smooth ($C^{\infty}$) boundary. A simple smooth seam form with one flat component is called a pita form, and with two flat components is called a D-form. See Figures 1 and 3. Given a convex body $X$ and a point $x\in X$, we say that a line $\ell$ traverses $x$ if $x\in\ell$ and some open neighborhood of $x$ on $\ell$ is contained in $X$. An extreme point of $X$ is a point $x\in X$ not traversed by any line. ## 3 Convex Hull In this section we prove Theorem 1, that every three-dimensional convex body is the convex hull of the nonflat points on its surface. From convex geometry we have the following characterization of the minimal set from which a convex body can be recovered as the convex hull: we need only the extreme points. ###### Theorem 5 (Minkowski’s Theorem). Every convex body in $\mathbb{R}^{n}$ is the convex hull of its extreme points. The proof is a straightforward induction on the dimension of the body and can be found as Corollary 1.4.5 in Schneider’s textbook [Sch93]. It remains to describe the extreme points of a seam form. To do so we begin with the following proposition: ###### Proposition 6. If $p$ is an extreme point of the convex body $X$, then for every open neighborhood $U$ of $p$ in $X$, some hyperplane has $p$ strictly on one side and all of $X\setminus U$ strictly on the other. ###### Proof. This is Lemma 1.4.6 in [Sch93]. For completeness we give the proof. Because $p$ is an extreme point, it cannot by definition be the convex combination of any two other points in $X$. Therefore it is not the convex combination $a_{1}x_{1}+\dotsb+a_{k}x_{k}$, with all $a_{i}>0$, of any $k$ other points in $X$, because otherwise we would have $p=(1-a_{k})(\frac{a_{1}}{1-a_{k}}x_{1}+\dotsb+\frac{a_{k-1}}{1-a_{k}}x_{k-1})+a_{k}x_{k}$, and certainly the convex combination $\frac{a_{1}}{1-a_{k}}x_{1}+\dotsb+\frac{a_{k-1}}{1-a_{k}}x_{k-1}$ is a point in $X$. In other words, $p$ lies outside the convex hull of $X\setminus\\{p\\}$, and consequently outside the convex hull $Y$ of $X\setminus U$. Now $Y$ is itself a convex body, and $p$ a point outside it. By the Separating Hyperplane Theorem, some hyperplane strictly separates them, and because $X\setminus U\subset Y$, it strictly separates $p$ and $X\setminus U$ as required. ∎ ###### Proposition 7. On the surface of a convex body, there are no flat extreme points. ###### Proof. Suppose some extreme point $p$ of a convex body $X$ was flat, with a neighborhood $S\subset\partial X$ isometric to a plane region. Let $U$ be an open neighborhood of $p$ in $X$ with $U\cap\partial X\subset S$. Let the hyperplane $H$ guaranteed by Proposition 6 separate $X$ into convex bodies $C$ and $Y$ with $p\in C$, and let $D=C\cap\partial X$. Because $C\subset U$, we have $D\subset S$ so that $D$ is flat. Now consider the normals to $X$ along the portion $D$ of its surface. Let $d$ be the distance from $p$ to $H$, and let $r$ be the maximum distance from the projection of $p$ onto $H$ to any point in $H\cap X$. Then any plane through $p$ and making an angle at most $\theta=\tan^{-1}(d/r)$ to $H$ fails to intersect $H\cap X$ and therefore fails to intersect $Y$. Therefore the normals to these planes, covering a spherical area of $2\pi(1-\cos\theta)>0$, all are normals to $X$ somewhere on $D$. This gives $G(D)$ a positive area, contradicting that $D\subset S$ is flat. ∎ Theorem 1 is now immediate from Theorem 5 and Proposition 7, and Corollary 2 follows. ## 4 Creases In this section we prove Theorem 3, characterizing the possible creases of a seam form. ###### Proposition 8. Let $\gamma$ be a crease in a flat component of a seam form $S$. Then $\gamma$ lies on a line segment $[p,q]$ between endpoints $p$ and $q$ that lie on seams, and the whole segment is composed of creases and vertices. ###### Proof. Let $S_{1}$ and $S_{2}$ be the open pieces bordered by $\gamma$ in the decomposition of the good surface $S$, and let $x\in\gamma$ be a point at which $S$ is not $C^{1}$. Then $S_{1}$ and $S_{2}$ are $C^{2}$ surfaces, so they have normals $n_{1}$ and $n_{2}$ at $x$, and because $S$ is not $C^{1}$ at $x$, these normals are distinct. Therefore $G(x)$ contains at least two distinct vectors. By Proposition 7, $x$ must be traversed by some line $\ell$, so that $\ell\cap S=[p,q]$ for some $p$ and $q$. Necessarily $\ell$ is perpendicular to all of $G(x)$, so for each $y\in[p,q]$ and each normal $n\in G(x)$, $n\cdot y=n\cdot x=\sup_{x^{\prime}\in X}n\cdot x^{\prime}$ and $n$ is a normal of $y$. Therefore each $G(y)$ contains $G(x)$, and so like $G(x)$ has at least two distinct vectors. The multiple normals in $G(x)$ and hence in each $G(y)$ determine a unique perpendicular line, so that no other line may traverse any point of $[p,q]$. In particular no line traverses $p$ or $q$, so by Proposition 7, these points are not flat and must lie on seams or vertices. At the same time, because a $C^{1}$ surface has only one normal at each point, no point of $[p,q]$ can be on a $C^{2}$ piece or a semicrease. The whole segment is therefore made up of creases and (nonstrict) vertices. Because a crease is defined from a cell-complex decomposition, only one crease runs through a given point, so because $\gamma$ runs through $x$ it must be one of the creases making up $[p,q]$. ∎ In order to analyze the Gauss map at seam and vertex points, we introduce some additional notation. Let $x\in S$ be incident to a 1- or 2-cell $C$, a (semi)crease or piece. Then we define $G_{C}(x)=\varliminf_{U\ni x}\overline{G(C\cap U)}$ as the Gauss map at $x$ on $C$. For comparison, observe that $G(x)=\varliminf_{U\ni x}\overline{G(U)}$ because the relation $G$ is closed, and in particular $G_{C}(x)\subseteq G(x)$. ###### Proof of Theorem 3. Let $\gamma$ be a crease in a flat component of a seam form $S$. By Proposition 8, $\gamma$ lies on a segment $[p,q]$ composed of creases and vertices and whose endpoints lie on seams. It remains to prove that if an endpoint, say $p$, lies on a seam and is not a strict vertex, then the seam is tangent to $[p,q]$. Let $G_{\gamma}(p)$ be the great circular arc $mn$, and let the pieces of $S$ bordering $\gamma$ be $S_{1}$ and $S_{2}$. By continuity, $G_{S_{1}}(p)\ni m$ and $G_{S_{2}}(p)\ni n$ (possibly after exchanging the names $m,n$), and because $m\neq n$ the Gauss map at $p$ on at least one of the cells $C$ surrounding $p$ from $S_{1}$ to $S_{2}$ apart from $\gamma$ must be a positive-length spherical curve in order to complete the path from $m$ to $n$. If $p$ is not a strict vertex, then $G(p)$ is a convex spherical shape of zero area, so it is a great circular arc, and $G_{C}(p)\subseteq G(p)$ is also a great circular arc. If $C$ is a piece, then $G_{C}(p)$ is either a singleton or a curve not lying on a great circle, because a great-circle Gauss map makes parallel rule lines that cannot converge at $p$. Therefore $C$ is a semicrease. Because $G(C)$ must be more than a single point, $C$ is a crease, and to make the Gauss map lie within the arc $G(p)$, $C$ must be tangent to $[p,q]$ as required. Finally, because $p$ is the endpoint of the intersection of the line $pq$ with $S$, the crease $C$ must not be a line segment, so by Proposition 8, it is actually part of the seam and the proof is complete. ∎ Of course, in a convex plane region, no line segment in the interior is tangent to the boundary, from which follows a corollary about simple seam forms. ###### Corollary 9. In a simple seam form, every crease in a flat component is on a line segment between two strict vertices. Finally, in a simple smooth seam form such as a pita-form or a D-form, the requirement of smoothness sharply limits the possible configurations. By (local) convexity, no vertex can be incident to three or more semicreases as part of its seam, and a vertex through which a seam passes cannot be a strict vertex. Consequently a pita-form must have a single path for its seam and just two strict vertices located at the seam’s endpoints, and a D-form must have a single cycle for its seam and no strict vertices. Corollary 4 follows. ## 5 Counterexamples We have required the flat components of a D-form to be convex. We could relax this requirement, requiring instead only that the metric space resulting from joining the two components be locally convex, and the Alexandrov-Pogorelov theorem would still guarantee a unique convex embedding in three-dimensional space. Of course Corollary 2 would still guarantee that the resulting body would be the convex hull of its seam, but it turns out that Corollary 4, whose conditions would no longer be satisfied, really would fail in its conclusion: one can construct a “D-form” under this relaxed definition which contains creases in its flat components. Indeed it is not hard to construct such an example, if one keeps in mind Theorem 3 that the offending crease must be tangent to a seam; see Figure 2. Figure 2: A “D-form” with a relaxed convexity condition. The solid dark line is the seam, the dotted lines are false edges from projection onto the page, and the solid light lines are the creases through a flat component. Actual D-forms have no such creases. For pita-forms, we have concluded in Corollary 4 that a pita-form may have at most one crease. Indeed this is tight, and it is easy to construct an example pita-form with a crease; see Figure 3. This possibility of creases therefore represents a real difference from D-forms. It represents also a contrast from the appearance of the natural paper experiments, which led the authors first introducting pita-forms ([DO07]) to suggest that pita-forms might never have creases; in fact, once one is familiar with Corollary 2, it is clear that the same experiments really would have to be creased if only the paper were behaving ideally. Figure 3: A typical pita-form. A crease runs between the endpoints of the seam. For some theorems of the same flavor as our results here, one might hope to obtain proofs by showing that the desired properties hold of convex polyhedra, which are relatively concrete and amenable to reasoning, and then that they carry over to general convex bodies as limits of polyhedra. Indeed, this is the approach taken in [DO07] to argue for Corollary 2 for the case of D-forms. Unfortunately this approach does not hold as widely as one might like. In particular, Corollary 9 cannot be proven by a limiting argument of the obvious form, even when restricted to D-forms. When each flat component of the D-form is approximated by a sequence of polygons, it is possible for the dihedral angles inside the components to approach positive limits, even when the angles of the polygonal approximations are required to converge to zero. For example, in the D-form obtained from two circular disks (which is just the double cover of a disk), the components may be approximated by regular $n$-gons for increasing $n$, and the resulting approximations to the D-form may be antiprisms of two smaller $n$-gons and $2n$ triangles. In this approximation sequence, the dihedral angles between each $n$-gon and its $n$ neighboring triangles approach $\pi/4$, not zero, even though they lie inside the flat components. For this and other reasons we have chosen direct proofs that attack the general case of convex bodies. ## 6 Acknowledgments We thank Jonathan Kelner, Joseph O’Rourke, and Johannes Wallner for helpful discussions, and the anonymous referees for helpful comments. ## References * [Ale50] A. D. Alexandrov, Convex Polyhedra, Springer-Verlag, Berlin, 2005. See especially note 21, page 189. * [BI06] Alexander I. Bobenko and Ivan Izmestiev, Alexandrov’s theorem, weighted Delaunay triangulations, and mixed volumes, Annales de l’Institut Fourier, in press. arXiv:math.DG/0609447. * [DO07] Erik D. Demaine and Joseph O’Rourke, Geometric Folding Algorithms, Cambridge University Press, Cambridge, 2007. Pages 352–354. * [Gau02] Gauss, Carl Friedrich, General Investigations of Curved Surfaces, 1827, Morehead and Hiltebeitel, tr., Princeton, 1902. * [KPD09] Daniel Kane, Gregory N. Price, and Erik D. Demaine, A pseudopolynomial algorithm for Alexandrov’s theorem, Algorithms and Data Structures Symposium (WADS) 2009. arXiv:0812.5030. * [PW01] Helmut Pottmann and Johannes Wallner, Computational Line Geometry, Springer-Verlag, Berlin, 2001. Page 418\. * [Sch93] Rolf Schneider, Convex Bodies: The Brunn-Minkowski Theory, Cambridge University Press, Cambridge, 1993\. * [Wil] Tony Wills, DForms: 3D forms from two 2D sheets, Bridges: Mathematical Connections in Art, Music, and Science, Reza Sarhangi and John Sharp, eds, London, pp. 503–510.

arxiv-papers

2024-09-04T02:48:53.364689

0711.3992

# Radiatively induced finite and (un)determined Chern-Simons-like terms Eduardo Passos Departamento de Física, Universidade Federal da Paraíba, Caixa Postal 5008, 58051-970 João Pessoa, Paraíba, Brazil E-mail passos@fisica.ufpb.br ###### Abstract: The problem of Chern-Simons-like term induction via quantum corrections in four-dimensions is investigated in two different cases. In the first case, we consider two distinct approaches to deal with the exact fermion propagator of the extended QED theory up to the first order in the $b$-coefficient. We find different results for distinct approaches in the same regularization scheme. In the second case, we show that when we use a modified derivative expansion method and another regularization scheme, we obtain a result that exactly coincides with one of the results obtained in the former case. This seems to imply an ambiguity absence as one treats the fermion propagator and the self- energy tensor properly. Chern-Simons Theories, Space-Time Symmetries, Quantum Corrections ## 1 Introduction The induction of the Chern-Simons-like Lorentz and CPT violating term, given by ${\cal L}_{CS}=\frac{1}{2}k_{\mu}\epsilon^{\mu\nu\lambda\theta}F_{\nu\lambda}A_{\theta}$, being $k_{\mu}$ a constant vector characterizing the preferred direction of the space-time, is one of the most important results in the study of Lorentz symmetry violation [2, 3]. This term which is known to have some important implications, such as birefringence of light in the vacuum [4], naturally emerges as a quantum correction in the theory suggested in [3] as a possible extension of QED: $\displaystyle{\cal L}_{QED}=\bar{\psi}(i\partial\\!\\!\\!/-m)\psi-\bar{\psi}b\\!\\!\\!/\gamma_{5}\psi-e\bar{\psi}A\\!\\!\\!/\psi,$ (1) where $b_{\mu}$ is a parameter introducing CPT symmetry breaking. Carrying out the integration over fermions, the relation between the coefficients $k_{\mu}$ and $b_{\mu}$ could be obtained in terms of some loop integrals with some of them being divergent. Therefore one has to implement some regularization scheme to calculate these integrals. Thus, the constant relating the coefficient $k_{\mu}$ and $b_{\mu}$ turns out to be dependent on the regularization scheme used [5, 6]. Such dependence on the regularization scheme has been intensively discussed in [7, 8, 9]. However, there is an alternative study with absence of ambiguities put forward recently in [10]. Based on the theory (1), the purpose of our study is to investigate different possibilities of finding ambiguities inherent to generation of Chern-Simons- like term via quantum corrections in four dimensions. We do this by using derivative expansion method of the fermion determinant [11] and the imaginary time formalism. The work structure is organized as follows: In section 2 we will investigate the induction of the Chern-Simons-like term via quantum corrections by using distinct approaches to deal with the exact fermion propagator. We find distinct relations between the coefficients $k_{\mu}$ and $b_{\mu}$ for different approaches, but with a same regularization scheme. Therefore, we conclude that different approaches to deal with the exact fermion propagator of the theory, leads to a new ambiguity for the problem of radiatively induced Chern-Simons-like term. In section 3 we develop another method to investigate the present issue. By modifying the derivative expansion method, we obtain a different self-energy tensor. We then use a specific regularization scheme to find a finite result identical to that obtained in section 2 by using another regularization scheme. This effect seems to imply ambiguity absence in our calculations. Finally, in section 4 we present our conclusions. ## 2 Inducing Chern-Simons-like term: Two different approaches In this section, we focus on the induction of the Chern-Simons-like term coefficient by expanding the self-energy (7) and using two distinct approaches to deal with the exact fermion propagator (8) up to the leading order in $b$. We find a new ambiguity because two different results appear. The one-loop effective action $S_{eff}[b,A]$ of the gauge field $A_{\mu}$ related to theory (1), can be expressed in the form of the following functional trace: $\displaystyle S_{eff}[b,A]=-i\,{\rm Tr}\,\ln(p\\!\\!\\!/-m-b\\!\\!\\!/\gamma_{5}-eA\\!\\!\\!/).$ (2) This functional trace can be represented as $S_{eff}[b,A]=S_{eff}[b]+S_{eff}^{\,\prime}[b,A]$, where the first term $S_{eff}[b]=-i\,{\rm Tr}\ln(p\\!\\!\\!/-m-b\\!\\!\\!/\gamma_{5})$ does not depend on the gauge field. The only nontrivial dynamics is concentrated in the second term $S_{eff}^{\,\prime}[b,A]$, which is given by the following power series: $\displaystyle S_{eff}^{\,\prime}[b,A]=i\,{\rm Tr}\sum_{n=1}^{\infty}\frac{1}{n}\Biggl{[}\frac{1}{p\\!\\!\\!/-m-b\\!\\!\\!/\gamma_{5}}\,eA\\!\\!\\!/\Biggr{]}^{n}.$ (3) To obtain the Chern-Simons-like term we should expand this expression up to the second order in the gauge field: $\displaystyle S_{eff}^{\,\prime}[b,A]=S_{eff}^{(2)}[b,A]+\ldots$ (4) The dots in (4) stand for higher order terms in the gauge field. Here $\displaystyle S_{eff}^{(2)}[b,A]=\frac{ie^{2}}{2}{\rm Tr}\Biggl{[}\frac{1}{p\\!\\!\\!/-m-b\\!\\!\\!/\gamma_{5}}\;A\\!\\!\\!/\;\frac{1}{p\\!\\!\\!/-m-b\\!\\!\\!/\gamma_{5}}\,\;A\\!\\!\\!/\Biggl{]}.$ (5) Using the derivative expansion method [11] one can find that the one-loop contribution to $S_{\rm eff}^{(2)}[b,A]$ reads $S_{\rm eff}^{(2)}[b,A(x)]=\frac{1}{2}\int d^{4}x\;\Pi^{\alpha\mu\nu}F_{\alpha\mu}A_{\nu},$ (6) where the one-loop self-energy $\Pi^{\alpha\mu\nu}$ is given by $\Pi^{\alpha\mu\nu}=-\frac{ie^{2}}{2}\int\,\frac{d^{4}p}{(2\pi)^{4}}\,\text{tr}\bigl{[}S_{b}(p)\,\gamma^{\mu}\,S_{b}(p)\,\gamma^{\alpha}\,S_{b}(p)\,\gamma^{\nu}\bigl{]},$ (7) where $\displaystyle S_{b}(p)=\frac{i}{p\\!\\!\\!/-m-b\\!\\!\\!/\gamma_{5}}$ (8) is a $b^{\mu}$ dependent exact fermion propagator of the theory. ### 2.1 Approach I: Fermion propagator rationalized Firstly, we use the approximation developed in [12], where the exact propagator (8) is rationalized in the form $S_{b}(p)=i\Big{[}\frac{p\\!\\!\\!/+m-\gamma_{5}b\\!\\!\\!/}{(p^{2}-m^{2})}-\frac{2\gamma_{5}(mb\\!\\!\\!/-(b\cdot p))(p\\!\\!\\!/+m)}{(p^{2}-m^{2})^{2}}\Big{]}+\cdots.$ (9) Substituting (9) into (7), we can calculate the trace of gamma matrices, resulting in the following expression for the self-energy tensor [5]: $\displaystyle\Pi^{\mu\alpha\nu}_{\bf r}$ $\displaystyle=$ $\displaystyle-2ie^{2}\int\,\frac{d^{4}p}{(2\pi)^{4}}\frac{1}{(p^{2}+m^{2})^{3}}\\{3\varepsilon^{\alpha\mu\nu\theta}\bigl{[}b_{\theta}(p^{2}-m^{2})-2p_{\theta}(b\cdot p)\bigl{]}$ (10) $\displaystyle-2b_{\theta}\bigl{[}\varepsilon^{\beta\mu\nu\theta}p_{\beta}p^{\alpha}+\varepsilon^{\alpha\beta\nu\theta}p_{\beta}p^{\mu}+\varepsilon^{\alpha\mu\beta\theta}p_{\beta}p^{\nu}\bigl{]}\\},$ where $\Pi^{\mu\alpha\nu}_{\bf r}$ means self-energy tensor is rationalized up to the first order in $b$-coefficient. In Eq.(10), we turn the Minkowski space to an Euclidean space by performing the Wick rotation $x_{0}\to-ix_{0}$, $p_{0}\to ip_{0}$, $b_{0}\to ib_{0}$, $d^{4}x\to-id^{4}x$ and $d^{4}p\to id^{4}p$. Note that by power counting, the momentum integral in (10) involves a finite term and terms with logarithmic divergences. In order to regularize such divergence, we use a scheme which we implement translation only on space coordinates of the momentum $p_{\rho}$ [13]. Hence, we have $\displaystyle p_{\rho}\to\vec{p}_{\rho}+p_{0}\delta_{0\rho}.$ (11) We use the covariance under spatial rotations which allows us to carry out the following replacement $\displaystyle\vec{p}_{\rho}\vec{p}^{\sigma}\to\frac{\vec{p}^{2}}{D}(\delta^{\sigma}_{\rho}-\delta_{\rho 0}\delta^{\sigma}_{0}).$ (12) Thus, $\displaystyle 2p_{\rho}(b\cdot p)\to 2\left(b_{\rho}\frac{\vec{p}^{2}}{D}-b_{0}\delta_{\rho 0}(\frac{\vec{p}^{2}}{D}-p^{2}_{0})\right),$ $\displaystyle 2p_{\beta}p^{\alpha}\to 2\left(\delta_{\beta}^{\alpha}\,\frac{\vec{p}^{\,2}}{D}-\delta_{\beta 0}\delta_{0}^{\alpha}(\frac{\vec{p}^{2}}{D}-p^{2}_{0})\right).$ (13) We have that only the terms above can contribute to the Chern-Simons structure. Therefore, we can split the expression (5) into a sum of two parts, “covariant” and “noncovariant”, i.e., $\displaystyle S^{\rm cov}_{\rm eff}=\frac{(-i)}{2}\int\,d^{4}x\,I_{1}\varepsilon^{\alpha\mu\nu\beta}\,b_{\beta}F_{\alpha\mu}A_{\nu},$ (14) with $\displaystyle I_{1}$ $\displaystyle=$ $\displaystyle\frac{-3ie^{2}}{2\pi}\,\int\,\frac{d^{D}\vec{p}}{(2\pi)^{D}}\,\int_{-\infty}^{+\infty}\,dp_{0}\,\Big{[}\frac{(1-\frac{4}{D})\vec{p}^{2}+p_{0}^{2}-m^{2}}{(\vec{p}^{2}+p_{0}^{2}+m^{2})^{3}}\Big{]},$ (15) and $\displaystyle S^{\rm ncv}_{\rm eff}$ $\displaystyle=$ $\displaystyle\frac{i}{2}\int\,d^{4}xI_{2}\,\bigl{[}3\varepsilon^{\alpha\mu\nu 0}\,b_{0}F_{\alpha\mu}A_{\nu}+b_{\theta}(\varepsilon^{0\mu\nu\theta}F_{0\mu}A_{\nu}$ (16) $\displaystyle+\varepsilon^{\alpha 0\nu\theta}F_{\alpha 0}A_{\nu}+\varepsilon^{\alpha\mu 0\theta}F_{\alpha\mu}A_{0})\bigl{]},$ with $\displaystyle I_{2}=\frac{ie^{2}}{\pi}\,\int\,\frac{d^{D}\vec{p}}{(2\pi)^{D}}\int_{-\infty}^{+\infty}\,dp_{0}\,\Big{[}\frac{\frac{\vec{p}^{2}}{D}-p_{0}^{2}}{(\vec{p}^{2}+p_{0}^{2}+m^{2})^{3}}\Big{]}.$ (17) The integrals $I_{1}$ and $I_{2}$ over $p_{0}$ are finite and can be calculated by residues theorem. Hence, we have $\displaystyle I_{1}=\frac{3ie^{2}}{8D}\,\int\,\frac{d^{D}\vec{p}}{(2\pi)^{D}}\,\Big{[}\frac{2(3-D)\vec{p}^{\,2}+Dm^{2}}{(\vec{p}^{2}+m^{2})^{5/2}}\Big{]},$ (18) and $\displaystyle I_{2}=-\frac{ie^{2}}{8D}\,\int\,\frac{d^{D}\vec{p}}{(2\pi)^{D}}\,\Big{[}\frac{(3-D)\vec{p}^{\,2}-Dm^{2}}{(\vec{p}^{2}+m^{2})^{5/2}}\Big{]}.$ (19) Now, we can integrate over the spatial momentum in $D$-dimensions [16], and we find, $\displaystyle I_{1}=\frac{3ie^{2}}{8(4\pi)^{D/2}}\frac{\Gamma{(1+\frac{\epsilon}{2})}}{\Gamma(\frac{5}{2})(m^{2})^{\epsilon/2}}$ (20) where $\epsilon=3-D$. The integral $I_{2}$ presents a value identically equal to zero. Therefore, the effective action Eq.(14) in the limit $D=3$ can be written in the form $\displaystyle S^{\rm cov}_{\rm eff}=\frac{1}{2}\int\,d^{4}x\,\varepsilon^{\alpha\mu\nu\beta}\,k_{\beta}F_{\alpha\mu}A_{\nu},$ (21) where $\displaystyle k_{\beta}=\frac{e^{2}}{16\pi^{2}}\,b_{\beta},$ (22) that coincides with the resulted of work [17] in the Schwinger constant field approximation. ### 2.2 Approach II: Fermion propagator expansion Now we use the approximation developed in [14] to expand the exact propagator (8) up to the first order in $b$-coefficient $\displaystyle S_{b}(p)=\frac{i}{p\\!\\!\\!/-m}+\frac{i}{p\\!\\!\\!/-m}\,(-ib\\!\\!\\!/\gamma_{5})\,\frac{i}{p\\!\\!\\!/-m}+\cdot\cdot\cdot.$ (23) In this case, we have our self-energy tensor in the form [6, 18]: $\displaystyle\Pi^{\mu\alpha\nu}_{\bf e}$ $\displaystyle=$ $\displaystyle-2ie^{2}\epsilon^{\mu\alpha\nu\rho}\int\,\frac{d^{4}p}{(2\pi)^{4}}\,\frac{1}{(p^{2}+m^{2})^{3}}$ (24) $\displaystyle\times[b_{\rho}(p^{2}-3m^{2})-4p_{\rho}(b\cdot p)],$ where $\Pi^{\mu\alpha\nu}_{\bf e}$ means the expanded self-energy tensor. In Eq.(24), we also change the Minkowski space to Euclidean space. Note that by power counting, the integral in the momentum space Eq.(24) also present a finite term and another that diverges logarithmically. Thus, we use the same regularization scheme adopted in the rationalized fermion propagator approach. In this way, we obtain the following effective action in the limit of $D=3$: $\displaystyle\tilde{S}^{\rm cov}_{\rm eff}=\frac{1}{2}\int\,d^{4}x\,\varepsilon^{\alpha\mu\nu\beta}\,\tilde{k}_{\beta}F_{\alpha\mu}A_{\nu},$ (25) where $\displaystyle\tilde{k}_{\beta}=\frac{e^{2}}{4\pi^{2}}\,b_{\beta}.$ (26) Here, we also have the absence of “noncovariant” part. The result (26) is equivalent to the obtained in [19] where one uses a physical cutoff for fermions. According to the results (22) and (26), we found that the use of different approaches to deal with the exact fermion propagator leads to distinct relations between the coefficient $k_{\mu}$ and $b_{\mu}$ for the Chern-Simons-like term. Our results establish the following relation: $\displaystyle\tilde{k}_{\beta}=k_{\beta}+\frac{3e^{2}}{16\pi^{2}}\,b_{\beta},$ (27) that means $\displaystyle\Delta k_{\beta}=\frac{3e^{2}}{16\pi^{2}}\,b_{\beta}.$ (28) The result (28) corresponds to variation between the results (22) and (26), that is identical to the result of Ref. [18] originated from the self-energy (24), in another regularization scheme. The same result has been found in the literature [14, 15]. ## 3 Other aspects on induced Chern-Simons-like term In this section, we present an alternative method to compute induced Chern- Simons-like term, that is independent of the approach used to deal with the exact fermion propagator. Let us rewrite the Eq.(3) in the form, $\displaystyle S_{eff}^{\,\prime}[b,A]=i\,{\rm Tr}\sum_{n=1}^{\infty}\frac{1}{n}\Biggl{[}\frac{1}{p\\!\\!\\!/-m-eA\\!\\!\\!/}\,b\\!\\!\\!/\gamma_{5}\Biggr{]}^{n}.$ (29) To obtain the Chern-Simons-like term we should expand this expression up to the leading order in $b$. Thus, for $n=1$, we have $\displaystyle S_{eff}^{(1)}[b,A]$ $\displaystyle=$ $\displaystyle i{\rm Tr}\Biggl{[}\frac{1}{p\\!\\!\\!/-m-eA\\!\\!\\!/}\;b\\!\\!\\!/\gamma_{5}\Biggl{]}.$ (30) Using the relation $\frac{1}{A-B}=\frac{1}{A}+\frac{1}{A}B\frac{1}{A}+\frac{1}{A}B\frac{1}{A}B\frac{1}{A}+\cdot\cdot\cdot$ (31) for $A=p\\!\\!\\!/-m$ and $B=eA\\!\\!\\!/$, we find $\displaystyle S_{eff}^{(1)}[b,A]=ie^{2}{\rm Tr}\Biggl{[}\frac{1}{p\\!\\!\\!/-m}\,b\\!\\!\\!/\gamma_{5}\,\frac{1}{p\\!\\!\\!/-m}\,A\\!\\!\\!/\,\frac{1}{p\\!\\!\\!/-m}\,A\\!\\!\\!/\Biggl{]},$ (32) where one was considered the cyclic property of the trace in the product of the $\gamma$-matrices. By using derivative expansion method, we find the following effective action: $S_{\rm eff}^{(1)}[b,A(x)]=\frac{1}{2}\int d^{4}x\;\Pi^{\alpha\mu\nu}F_{\alpha\mu}A_{\nu},$ (33) where the one-loop self-energy $\Pi^{\alpha\mu\nu}$ is given by $\displaystyle\Pi^{\mu\alpha\nu}$ $\displaystyle=$ $\displaystyle-2ie^{2}\int\,\frac{d^{4}p}{(2\pi)^{4}}\frac{1}{(p^{2}-m^{2})^{3}}\\{\varepsilon^{\alpha\mu\nu\theta}b_{\theta}(p^{2}-m^{2})$ (34) $\displaystyle-2b_{\theta}\bigl{[}\varepsilon^{\alpha\nu\theta\beta}p_{\beta}p^{\mu}-\varepsilon^{\alpha\mu\theta\beta}p_{\beta}p^{\nu}\bigl{]}\\}.$ Note that in the self-energy tensor (34) there exists a convergent contribution and the remaining term diverges logarithmically. However, differently of the previous situations, in this case the calculation of the divergent integrals is very delicate. Calculating this self-energy tensor by using the same regularization scheme of the section 2, we find a null result, and in turn, the absence of the Chern-Simons term. On the other hand, there also exists the possibility of using in Eq.(34) the relation $\int\frac{d^{4}p}{(2\pi)^{4}}p_{\mu}p_{\nu}\,f(p^{2})=\frac{g_{\mu\nu}}{4}\int\frac{d^{4}p}{(2\pi)^{4}}p^{2}\,f(p^{2}),$ (35) that naturally removes the logarithmic divergence. As a result, we have only the finite contribution $\Pi^{\mu\alpha\nu}=\frac{e^{2}}{16\pi^{2}}\,\varepsilon^{\mu\alpha\nu\theta}\,b_{\theta}.$ (36) In this case, we find $S_{\rm eff}^{(1)}[b,A(x)]=\frac{1}{2}\int d^{4}x\;k_{\beta}\varepsilon^{\mu\alpha\nu\beta}\partial_{\alpha}A_{\mu}A_{\nu},$ (37) where $k_{\beta}=\frac{e^{2}}{16\pi^{2}}b_{\beta}.$ (38) Note that the result (38) is the same as the result found in (22) where the exact fermion propagator was rationalized up to the first order in the $b$-coefficient. Thus, we have here another surprising effect: The result $k_{\beta}=\frac{e^{2}}{16\pi^{2}}b_{\beta}$ given in (22) as result of dimensional regularization and also given in (38) as result of the Lorentz preserving regularization (35), appears to be $independent$ of the regularization scheme used by properly carrying out the self-energy tensor. The result (38) was also obtained in the work [15] by using massless exact fermion propagator. We observe that the factor $(e^{2}/16\pi^{2})$, is exactly the same as found in the well-known Adler-Bell-Jackiw anomaly [20, 21, 22]. ## 4 Conclusions We have investigated the induction of Chern-Simons-like term via quantum corrections in two different situations. Firstly, we use a same regularization to different approaches to deal with the exact fermion propagators up to the leading order in the $b$-coefficient. In this case, our results are finite and agree with other results in the literature, but they do not agree with each other, because they are different depending on the approach used. Moreover, it generates values whose difference is exactly: $\Delta k_{\beta}=(3e^{2}/{16\pi^{2}})\,b_{\beta}$, which agrees with the result of Ref. [18] found in another context. We conclude that this is due to the different approximation of the exact fermion propagator of theory. The problem was also investigated in a context independent of the approaches used to deal with the exact fermion propagator. In this case, we modify the derivative expansion method and obtain a new self-energy tensor for the effective action. The momentum integrals were calculated by using another regularization scheme. As a result, we obtained a relation between the coefficients $k_{\beta}$ and $b_{\beta}$ identical to the one obtained in the case where it was used the fermion propagator expansion. We also observed that the parameter of proportionality between these two coefficients is exactly the same as the one found in the well-known Adler-Bell-Jackiw anomaly. In our calculations, we also observed that the “noncovariant” contributions for the Chern-Simons-like term are absent, as was anticipated in [5] in the finite temperature context. Therefore, we insist that a complete comprehension of these question will require further investigations. Acknowledgments. The author would like to thank A. Kostelecký for useful comments and F.A. Brito for interesting discussions. This work was partially supported by Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq). ## References * [1] * [2] S. Carroll, G. Field, R. Jackiw, Phys. Rev. D 41, 1231 (1990). * [3] D. Kolladay, V. A. Kostelecký, Phys. Rev. D 55, 6760 (1997). * [4] R. Jackiw, Nucl. Phys. Proc. Suppl. 108, 30 (2002), hep-th/0110057. * [5] F. A. Brito, J. R. Nascimento, E. Passos, A. Yu. Petrov, JHEP 06, 016 (2007), arXiv: 0705.1338 [hep-th]. * [6] M. Gomes, J. R. Nascimento, E. Passos, A. Yu. Petrov, A. J. da Silva, Phys. Rev. D 76, 047701 (2007). * [7] R. Jackiw, Int. J. Mod. Phys. B 14, 2011 (2000), hep-th/9903044; J. M. Chung, Phys. Rev. D60, 127901 (1999), hep-th/9904037. * [8] M. Perez-Victoria, JHEP 04, 032 (2001), hep-th/0102021. * [9] G. Bonneau, Nucl. Phys. B 593 (2001) 398. * [10] F. A. Brito, J. R. Nascimento, E. Passos, A. Yu. Petrov, arXiv: 0709.3090v2 [hep-th]. * [11] I. J. R. Aitchison and C. M. Fraser, Phys. Lett. B 146, 63 (1984); Phys. Rev. D 31, 2605 (1985); C. M. Fraser, Z. Phys. C 28, 101 (1985); A. I. Vainshtein, V. I. Zakharov, V. A. Novikov, and M. A. Shifman, Yad. Fiz. (Sov. J. Nucl. Phys.) 39, 77 (1984); J. A. Zuk, Phys. Rev. D 32, 2653 (1985); L.-H. Chan, Phys. Rev. Lett. 54, 1222 (1985); ibid. 55, 21 (1985); M. K. Gaillard, Nucl. Phys. B268, 669 (1986); A. Das and A. Karev, Phys. Rev. D 36, 623 (1987); K. S. Babu, A. Das, and P. Panigrahi, Phys. Rev. D 36, 3725 (1987). * [12] D. Ebert, V. Ch. Zhukovsky and A. S. Razumovsky, Phys. Rev. D 70, 025003 (2004). * [13] L. Cervi, L. Griguolo, D. Seminara, Phys. Rev. D 64, 105003 (2001). * [14] R. Jackiw and V. A. Kostelecký, Phys. Rev. Lett. 82, 3572 (1999), hep-ph/9901358. * [15] M. Perez-Victoria, Phys. Rev. Lett. 83 (1999), 2518. * [16] G. ‘t Hooft and M. Veltman, Nucl. Phys. B44, 189 (1972). * [17] M. Chaichian, W. F. Chen and R. González Filipe, Phys.Lett.B 503, 215-222 (2001). * [18] T. Mariz, J. R. Nascimento, E. Passos and R. F. Ribeiro, JHEP 10, 019 (2005). * [19] A. A. Andrianov, P. Giacconi and R. Soldati, JHEP 02, 030 (2002). * [20] S.L. Adler, Phys. Rev. 177, 2426 (1969); S. L. Adler and W. A Bardeen, Phys. Rev. 182, 1517 (1969); J. S. Bell and R. Jackiw, Nuovo Cimento 60A 47 (1969). * [21] Michael E. Peskin and Daniel. V. Schroeder, An Introduction to Quantum Field Theory, Persus Books (1995). * [22] H. Banerjee, arXiv: 9907:162 v2 [hep-th].

arxiv-papers

2024-09-04T02:48:53.383690

0712.1056

# Can the tidal quality factors of terrestrial planets and moons scale as positive powers of the tidal frequency? Michael Efroimsky US Naval Observatory, Washington DC 20392 USA e-mail: me @ usno.navy.mil ###### Abstract In geophysics and seismology, it is a common knowledge that the quality factors $Q$ of the mantle and crust materials scale as the tidal frequency to a positive fractional power (Karato 2007, Efroimsky & Lainey 2007). In astronomy, there exists an equally common belief that such rheological models introduce discontinuities into the equations and thus are unrealistic at low frequencies. We demonstrate that, while such models indeed make the conventional expressions for the tidal torque diverge for vanishing frequencies, the emerging infinities reveal not the impossible nature of one or another rheology, but a subtle flaw in the underlying mathematical model of friction. Flawed is the common misassumption that damping merely provides phase lags to the terms of the Fourier series for the tidal potential. A careful hydrodynamical treatment by Sir George Darwin (1879), with viscosity explicitly included, had demonstrated that the magnitudes of the terms, too, get changed – a fine detail later neglected as “irrelevant”. Reinstating of this detail tames the fake infinities and rehabilitates the “impossible” scaling law (which happens to be the actual law the terrestrial planets obey at low frequencies). This preprint is a pilot paper. A more comprehensive treatise on tidal torques is to be published. (Efroimsky & Williams 2009) ## 1 Introduction. We are considering bodily tides in a primary perturbed by a secondary. Each elementary volume of the primary is subject to a tide-raising potential, which in general is not periodic but can be expanded into a sum of periodic terms. We shall assume that the primary is homogeneous and incompressible. Although simplistic, the model provides a good qualitative understanding of tidal evolution of both the primary’s spin and the secondary’s orbit ## 2 Linearity of the tide ### 2.1 Two aspects of linearity We assert deformations to be linear. Each tidal harmonic $\,W_{\it{l}}\,$ of the potential disturbance produced by the secondary generates a linear deformation of the primary’s shape, while each such deformation amends the potential of the primary with an addition proportional to the Love number $\,k_{\it{l}}\,$. Linearity also implies that the energy attenuation rate $\langle\stackrel{{\scriptstyle\centerdot}}{{E}}(\chi)\rangle$ at each frequency $\,\chi\,$ depends solely on the value of $\,\chi\,$ and on the amplitude $\,E_{peak}(\chi)\,$, and is not influenced by the other harmonics. Thus, $\displaystyle\langle\,\dot{E}(\chi)\,\rangle\;=\;-\;\frac{\textstyle\chi E_{peak}(\chi)}{\textstyle Q(\chi)}\;$ (1) or, equivalently: $\displaystyle\Delta E_{cycle}(\chi)\;=\;-\;\frac{2\;\pi\;E_{peak}(\chi)}{Q(\chi)}\;\;\;,$ (2) $\Delta E_{cycle}(\chi)\,$ being the one-cycle energy loss. The so-defined quality factor $\,Q(\chi)\,$ corresponding to some frequency $\,\chi\,$ is interconnected with the phase lag $\,\epsilon(\chi)\,$ corresponding to the same frequency:111 The interconnection becomes more complicated than (3) when the values of $\,Q\,$ are about ten or less – see the Appendix below. $\displaystyle{Q}\,=\,\cot|\epsilon|\;\;\;.$ (3) This makes the essence of the standard, linear theory of bodily tides. The model permits for the freedom of choice of the functional dependency of the quality factor upon the tidal frequency. Whatever the form of this dependence, the basic idea of the theory is the following: the tide-raising potential is expanded over periodic terms, whereafter (a) the material’s response is assumed to be linear, and (b) the overall attenuation rate is asserted to be a sum of rates corresponding to the involved frequencies. Leaving the first of these assertions intact, we shall make in the subsequent subsection an aside on the applicability limits of the second assertion. In the Appendix below, we shall also probe some of the applicability limits of (3). ### 2.2 Goldreich’s admonition: a general difficulty stemming from nonlinearity Introduced empirically as a means to figleaf our lack of knowledge of the attenuation process in its full complexity, the notion of $\,Q\,$ has proven to be practical due to its smooth and universal dependence upon the frequency and temperature. At the same time, this empirical treatment has its predicaments and limitations. Its major inborn defect was brought to light by Peter Goldreich who pointed out that the attenuation rate at a particular frequency depends not only upon the appropriate Fourier component of the stress, but also upon the _overall_ stress. This happens because for real minerals each quality factor $\,Q(\chi_{\textstyle{{}_{i}}})\,$ bears dependence not only on the frequency $\,\chi_{\textstyle{{}_{i}}}\,$ itself, but also on the magnitude of the $\,\chi_{\textstyle{{}_{i}}}-$component of the stress and, most importantly, also on the _overall_ stress. This, often- neglected, manifestation of nonlinearity may be tolerated only when the amplitudes of different harmonics of stress are comparable. However, when the amplitude of the principal mode is orders of magnitude higher than that of the harmonics (tides being the case), then the principal mode will, through this nonlinearity, make questionable our entire ability to decompose the overall attenuation into a sum over frequencies. Stated differently, the quality factors corresponding to the weak harmonics will no longer be well defined physical parameters. Here follows a quotation from Goldreich (1963): _“… Darwin and Jeffreys both wrote the tide-raising potential as the sum of periodic potentials. They then proceeded to consider the response of the planet to each of the potentials separately. At first glance this might seem proper since the tidal strains are very small and should add linearly. The stumbling block in this procedure, however, is the amplitude dependence of the specific dissipation function. In the case of the Earth, it has been shown by direct measurement that $\,Q\,$ varies by an order of magnitude if we compare the tide of frequency $\,2\omega-2n\,$ with the tides of frequencies $\,2\omega-n\,$, $\,2\omega-3n\,$, and $\,\frac{\textstyle 3}{\textstyle 2}n\,$. This is because these latter tides have amplitudes which are smaller than the principle tide (of frequency $\,2\omega-2n\,$ ) by a factor of eccentricity or about 0.05. It may still appear that we can allow for this amplitude dependence of Q merely by adopting an amplitude dependence for the phase lags of the different tides. Unfortunately, this is really not sufficient since a tide of small amplitude will have a phase lag which increases when its peak is reinforcing the peak of the tide of the major amplitude. This non-linear behaviour cannot be treated in detail since very little is known about the response of the planets to tidal forces, except for the Earth.”_ On these grounds, Goldreich concluded the paragraph with an important warning that we _“use the language of linear tidal theory, but we must keep in mind that our numbers are really only parametric fits to a non-linear problem.”_ In order to mark the line beyond which this caveat cannot be ignored, let us first of all recall that the linear approximation remains applicable insofar as the strains do not approach the nonlinearity threshold, which for most minerals is of order $\,10^{-6}\,$. On approach to that threshold, the quality factors may become dependent upon the strain magnitude. In other words, in an attempt to extend the expansion (1 \- 2) to the nonlinear case, we shall have to introduce, instead of $\,Q(\chi_{\textstyle{{}_{i}}})\,$, some new functions $\,Q(\,\chi_{\textstyle{{}_{i}}}\,,\;E_{{}_{peak}}(\chi_{\textstyle{{}_{i}}})\,,\;E_{{}_{overall}}\,)\;$. (Another complication is that in the nonlinear regime new frequencies will be generated, but we shall not go there.) Now consider a superposition of two forcing stresses – one at the frequency $\,\chi_{\textstyle{{}_{1}}}\,$ and another at $\,\chi_{\textstyle{{}_{2}}}\,$. Let the amplitude $\;E_{{}_{peak}}(\chi_{\textstyle{{}_{1}}})\;$ be close or above the nonlinearity threshold, and $\;E_{{}_{peak}}(\chi_{\textstyle{{}_{2}}})\;$ be by an order or two of magnitude smaller than $\;E_{{}_{peak}}(\chi_{\textstyle{{}_{1}}})\;$. To adapt the linear machinery (1 \- 2) to the nonlinear situation, we have to write it as $\displaystyle\langle\,\dot{E}~{}\rangle~{}=~{}\langle\,\dot{E}_{1}~{}\rangle~{}+~{}\langle\,\dot{E}_{2}~{}\rangle~{}=~{}-~{}\chi_{\textstyle{{}_{1}}}~{}\frac{~{}E_{{}_{peak}}(\chi_{\textstyle{{}_{1}}})~{}}{Q(\,\chi_{\textstyle{{}_{1}}}\,,~{}E_{{}_{peak}}(\chi_{\textstyle{{}_{1}}})\,)}~{}-~{}\chi_{\textstyle{{}_{2}}}~{}\frac{\,E_{{}_{peak}}(\chi_{\textstyle{{}_{2}}})\,}{Q(\chi_{\textstyle{{}_{2}}}\,,\;E_{{}_{peak}}(\chi_{\textstyle{{}_{1}}})\,,\;E_{{}_{peak}}(\chi_{\textstyle{{}_{2}}})\,)}\;\;\;,\;\;\;\;\;$ (4) the second quality factor bearing a dependence not only upon the frequency $\,\chi{\textstyle{{}_{2}}}\,$ and the appropriate magnitude $\;E_{{}_{peak}}(\chi_{\textstyle{{}_{2}}})\;$, but also upon the magnitude of the _first_ mode, $\;E_{{}_{peak}}(\chi_{\textstyle{{}_{1}}})\;$, – this happens because it is the first mode which makes a leading contribution into the overall stress. Even if (4) can be validated as an extension of (1 \- 2) to nonlinear regimes, we should remember that the second term in (4) is much smaller than the first one (because we agreed that $\;E_{{}_{peak}}(\chi_{\textstyle{{}_{2}}})\,\ll\,E_{{}_{peak}}(\chi_{\textstyle{{}_{1}}})\;$). This results in two quandaries. The first one (not mentioned by Goldreich) is that a nonlinearity-caused non-smooth behaviour of $\;Q(\,\chi_{\textstyle{{}_{1}}}\,,\;E_{{}_{peak}}(\chi_{\textstyle{{}_{1}}})\,)\;$ will cause variations of the first term in (4), which may exceed or be comparable to the entire second term. The second one (mentioned in the afore quoted passage from Goldreich) is the phenomenon of nonlinear superposition, i.e., the fact that the smaller-amplitude tidal harmonic has a higher dissipation rate (and, therefore, a larger phase lag) whenever the peak of this harmonic is reinforcing the peak of the principal mode. Under all these circumstances, fitting experimental data to (4) will become a risky business. Specifically, it will become impossible to reliably measure the frequency dependence of the second quality factor; therefore the entire notion of the quality factor will, in regard to the second frequency, become badly defined. The admonition by Goldreich had been ignored until an alarm sounded. This happened when the Jet Propulsion Laboratory’s Lunar-ranging team applied the linear approach to determining the frequency-dependence of the Lunar quality factor, $\,Q(\chi)\,$. They obtained (or, as Peter Goldreich rightly said, _fitted_ their data to) the dependency $\,Q(\chi)\,\sim\,\chi^{\alpha}\,$ (Williams et al. 2001). The value of the exponential for the Moon turned out to be _negative_ : $\,\alpha\,=\,-\,0.07\;$, a result firmly tabooed by the condensed-matter physics for this range of frequencies (Karato 2007). One possible approach to explaining this result may be the following. Let us begin with a very crude estimate for the tidal strains. For the Moon, the tidal displacements are of order $\,0.1m\,$ (zero to peak). The most rough estimate for the strain can be obtained through dividing the displacement by the radius of the body. While this ratio is almost twenty times less than the $\,10^{-6}\,$ nonlinearity threshold, we should keep in mind that in reality the distribution of the tidal strain is a steep function of the radius, with the strain getting its maximum near the centre of the body, as can be seen from equations (48.17) in Sokolnikoff (1956). The values of strain can vary in magnitude, over the radius, by about a factor of five.222 Accordingly, the tidal-energy radial distribution, too, is strongly inhomogeneous, with a maximum in the planet’s centre – see Fig. 3 in Peale & Cassen (1978). Another order of magnitude, at least, will come from the fact that the Moon is inhomogeneous and that its warmer layers are far more elastic than its rigid surface. As a result, deep in the Lunar interior the tidal strain will exceed the afore mentioned nonlinearity threshold. Hence the uncertainties in determination of $\,Q(\chi)\,$ by the Lunar-ranging team. We shall not dwell on this topic in quantitative detail, leaving it for a future work. Our only goal here has been to draw the readers’ attention to the existing difficulty stemming from the shortcomings of the extension of (1 \- 2) to nonlinear regimes. ## 3 Darwin (1879) and Kaula (1964) The potential produced at point $\mbox{{\boldmath$\vec{R}$}}=(R\,,\,\lambda\,,\,\phi)\,$ by a mass $\,M^{*}$ located at $\,\mbox{{\boldmath$\vec{r}$}}^{\;*}=(r^{*},\,\lambda^{*}\,,\,\phi^{*})\,$ is $\displaystyle W(\mbox{{\boldmath$\vec{R}$}}\,,\,\mbox{{\boldmath$\vec{r}$}}^{\;*})\,=\,-\,\frac{G\;M^{*}}{r^{\,*}}\sum_{{\it{l}}=2}^{\infty}\left(\frac{R}{r^{\;*}}\right)^{\textstyle{{}^{\it{l}}}}\sum_{m=0}^{\it l}\frac{({\it l}-m)!}{({\it l}+m)!}(2-\delta_{0m})P_{{\it{l}}m}(\sin\phi)P_{{\it{l}}m}(\sin\phi^{*})\;\cos m(\lambda-\lambda^{*})~{}~{}.~{}~{}~{}~{}~{}~{}$ (5) When a tide-raising secondary located at $\,\mbox{{\boldmath$\vec{r}$}}^{\;*}\,$ distorts the shape of the primary, the potential generated by the primary at some exterior point $\vec{r}$ gets changed. In the linear approximation, its variation is: $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;\sum_{{\it l}=2}^{\infty}\;k_{\it l}\;\left(\,\frac{R}{r}\,\right)^{{\it l}+1}\;W_{\it{l}}(\mbox{{\boldmath$\vec{R}$}}\,,\;\mbox{{\boldmath$\vec{r}$}}^{\;*})~{}~{}~{},~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (6) $\,k_{\it{l}}\,$ being the _l_ th Love number, $R\,$ now being the mean equatorial radius of the primary, $\,\mbox{{\boldmath$\vec{R}$}}\,=\,(R\,,\,\phi\,,\,\lambda)\,$ being a surface point, $\,\mbox{{\boldmath$\vec{r}$}}^{\;*}\,=\,(r^{*}\,,\,\phi^{*}\,,\,\lambda^{*})\,$ being the coordinates of the tide-raising secondary, $\,\mbox{{\boldmath$\vec{r}$}}\,=\,(r\,,\,\phi\,,\,\lambda)\,$ being an exterior point located above the surface point $\vec{R}$ at a radius $\,r\,\geq\,R\,$, and the longitudes being reckoned from a fixed meridian on the primary. Substitution of (5) into (6) entails $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;\,-\,{G\;M^{*}}\sum_{{\it{l}}=2}^{\infty}k_{\it l}\;\frac{R^{\textstyle{{}^{2\it{l}+1}}}}{r^{\textstyle{{}^{\it{l}+1}}}{r^{\;*}}^{\textstyle{{}^{\it{l}+1}}}}\sum_{m=0}^{\it l}\frac{({\it l}-m)!}{({\it l}+m)!}(2-\delta_{0m})P_{{\it{l}}m}(\sin\phi)P_{{\it{l}}m}(\sin\phi^{*})\;\cos m(\lambda-\lambda^{*})~{}~{}~{}.~{}~{}~{}~{}~{}$ (7) A different expression for the tidal potential was offered by Kaula (1961, 1964), who developed a powerful technique that enabled him to switch from the spherical coordinates to the Kepler elements $\,(\,a^{*},\,e^{*},\,{\it i}^{*},\,\Omega^{*},\,\omega^{*},\,{\cal M}^{*}\,)\,$ and $\,(\,a,\,e,\,{\it i},\,\Omega,\,\omega,\,{\cal M}\,)\,$ of the secondaries located at $\,\mbox{{\boldmath$\vec{r}$}}^{\;*}\,$ and $\vec{r}$ . Application of this technique to (7) results in $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;-\;\sum_{{\it l}=2}^{\infty}\;k_{\it l}\;\left(\,\frac{R}{a}\,\right)^{\textstyle{{}^{{\it l}+1}}}\frac{G\,M^{*}}{a^{*}}\;\left(\,\frac{R}{a^{*}}\,\right)^{\textstyle{{}^{\it l}}}\sum_{m=0}^{\it l}\;\frac{({\it l}-m)!}{({\it l}+m)!}\;\left(\,2\;\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (8) $\displaystyle~{}~{}~{}~{}~{}\left.-\,\delta_{0m}\,\right)\,\sum_{p=0}^{\it l}F_{{\it l}mp}({\it i}^{*})\sum_{q=-\infty}^{\infty}G_{{\it l}pq}(e^{*})\sum_{h=0}^{\it l}F_{{\it l}mh}({\it i})\sum_{j=-\infty}^{\infty}G_{{\it l}hj}(e)\;\cos\left[\left(v_{{\it l}mpq}^{*}-m\theta^{*}\right)-\left(v_{{\it l}mhj}-m\theta\right)\right]~{}~{}_{\textstyle{{}_{\textstyle,}}}$ where $\displaystyle v_{{\it l}mpq}^{*}\;\equiv\;({\it l}-2p)\omega^{*}\,+\,({\it l}-2p+q){\cal M}^{*}\,+\,m\,\Omega^{*}~{}~{}~{},$ (9) $\displaystyle v_{{\it l}mhj}\;\equiv\;({\it l}-2h)\omega\,+\,({\it l}-2h+j){\cal M}\,+\,m\,\Omega~{}~{}~{},$ (10) and $\theta\,=\,\theta^{*}\,$ is the sidereal angle. While (8) and (7) are equivalent for an idealised elastic planet with an instant response of the shape, the situation becomes more involved when dissipation-caused delays come into play. Kaula’s expression (8), as well as its truncated, Darwin’s version,333 While the treatment by Kaula (1964) entails the infinite Fourier series (8), the developments by Darwin (1879) furnish its partial sum with $\,|{\it{l}}|,\,|q|,\,|j|\,\leq\,2\,.$ For a simple introduction into Darwin’s method see Ferraz-Mello, Rodríguez & Hussmann (2008). is capable of accommodating separate phase lags for each harmonic involved: $\displaystyle U(\mbox{{\boldmath$\vec{r}$}})\;=\;-\;\sum_{{\it l}=2}^{\infty}\;k_{\it l}\;\left(\,\frac{R}{a}\,\right)^{\textstyle{{}^{{\it l}+1}}}\frac{G\,M^{*}}{a^{*}}\;\left(\,\frac{R}{a^{*}}\,\right)^{\textstyle{{}^{\it l}}}\sum_{m=0}^{\it l}\;\frac{({\it l}-m)!}{({\it l}+m)!}\;\left(\,2\;-\right.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (11) $\displaystyle\left.\delta_{0m}\,\right)\,\sum_{p=0}^{\it l}F_{{\it l}mp}({\it i}^{*})\sum_{q=-\infty}^{\infty}G_{{\it l}pq}(e^{*})\sum_{h=0}^{\it l}F_{{\it l}mh}({\it i})\sum_{j=-\infty}^{\infty}G_{{\it l}hj}(e)\;\cos\left[\left(v_{{\it l}mpq}^{*}-m\theta^{*}\right)-\left(v_{{\it l}mhj}-m\theta\right)-\epsilon_{{\it l}mpq}\right]~{}~{}_{\textstyle{{}_{\textstyle.}}}$ where $\displaystyle\epsilon_{{\it l}mpq}=\left[\,({\it l}-2p)\,\dot{\omega}^{*}\,+\,({\it l}-2p+q)\,\dot{\cal{M}}^{*}\,+\,m\,(\dot{\Omega}^{*}\,-\,\dot{\theta}^{*})\,\right]\,\Delta t_{\it{l}mpq}=\,\omega^{*}_{\it{l}mpq}\,\Delta t_{\it{l}mpq}=\,\pm\,\chi^{*}_{\it{l}mpq}\,\Delta t_{\it{l}mpq}~{}~{},~{}~{}~{}~{}$ (12) is the phase lag interconnected with the quality factor via $\;Q_{{\it{l}}mpq}\,=\,\cot\,|\epsilon_{{\it{l}}mpq}|\;$. The tidal harmonic $\,\omega^{*}_{\it{l}mpq}\,$ introduced in (12) is $\displaystyle\omega^{*}_{{\it l}mpq}\;\equiv\;({\it l}-2p)\;\dot{\omega}^{*}\,+\,({\it l}-2p+q)\;\dot{\cal{M}}^{*}\,+\,m\;(\dot{\Omega}^{*}\,-\,\dot{\theta}^{*})\;~{}~{},~{}~{}~{}$ (13) while the positively-defined quantity $\displaystyle\chi^{*}_{{\it l}mpq}\,\equiv\,|\,\omega^{*}_{{\it l}mpq}\,|\,=\,|\,({\it l}-2p)\,\dot{\omega}^{*}\,+\,({\it l}-2p+q)\,\dot{\cal{M}}^{*}\,+\,m\,(\dot{\Omega}^{*}\,-\,\dot{\theta}^{*})\;|~{}~{}~{}~{}~{}$ (14) is the actual physical $\,{{\it l}mpq}\,$ frequency excited by the tide in the primary. The corresponding positively-defined time delay $\,\Delta t_{\it{l}mpq}\,$ depends on this physical frequency, the functional forms of this dependence being different for different materials. Formulae (8) and (11) constitute the principal result of Kaula’s theory of tides. Most importantly, Kaula’s formalism imposes no _a priori_ constraint on the form of frequency-dependence of the lags. ## 4 The Darwin-Kaula-Goldreich expansion for the tidal torque Now we are prepared to calculate the planet-perturbing tidal torque. Since in what follows we shall dwell on the low-inclination case, it will be sufficient to derive the torque’s component orthogonal to the planetary equator: $\displaystyle{\tau}\;=\;-\;{M}\;\frac{\partial U(\mbox{{\boldmath$\vec{r}$}})}{\partial\theta}\;\;\;,$ (15) $M\,$ being the mass of the tide-disturbed satellite, and the “minus” sign emerging due to our choice not of the astronomical but of the physical sign convention. Adoption of the latter convention implies the emergence of a “minus” sign in the expression for the potential of a point mass: $\;-\,Gm/r\,$. This “minus” sign then shows up on the right-hand sides of (5), (7), (8), and (11). It is then compensated by the “minus” sign standing in (15). The right way of calculating $\,{\partial U(\mbox{{\boldmath$\vec{r}$}})}/{\partial\theta}\,$ is to take the derivative of (11) with respect to $\,\theta\,$, and then444 Be mindful that our intention here is to differentiate not $\;\cos\left[\left(v_{{\it l}mpq}^{*}-m\theta^{*}\right)-\left(v_{{\it l}mhj}-m\theta\right)\right]\;$ but $\;\cos\left[(v_{{\it l}mpq}^{*{(delayed)}}-m{\theta^{*}}^{(delayed)})-\left(v_{{\it l}mhj}-m\theta\right)\right]\;$. Hence the said sequence of operations. to get rid of the sidereal angle completely, by imposing the constraint $\,\theta^{*}\,=\,\theta\,$. This will yield: $\displaystyle{\tau}=-\,\sum_{{\it l}=2}^{\infty}k_{\it l}\left(\frac{R}{a}\right)^{\textstyle{{}^{{\it l}+1}}}\frac{G\,M^{*}\,M}{a^{*}}\left(\frac{R}{a^{*}}\right)^{\textstyle{{}^{\it l}}}\sum_{m=0}^{\it l}\frac{({\it l}-m)!}{({\it l}+m)!}2m\;\sum_{p=0}^{\it l}F_{{\it l}mp}({\it i}^{*})\sum_{q=-\infty}^{\infty}G_{{\it l}pq}(e^{*})~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ $\displaystyle\sum_{h=0}^{\it l}F_{{\it l}mh}({\it i})\sum_{j=-\infty}^{\infty}G_{{\it l}hj}(e)\;\sin\left[\,v^{*}_{{\it l}mpq}\,-\;v_{{\it l}mhj}\,-\;\epsilon_{{\it l}mpq}\,\right]~{}~{}_{\textstyle{{}_{\textstyle,}}}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (16) In the case of the tide-raising satellite coinciding with the tide-perturbed one, all the elements become identical to their counterparts with an asterisk. For a primary body not in a tidal lock with its satellite,555 With $\alpha$ denoting the librating angle, the locking condition reads: $\;\,\theta\,=\,\Omega\,+\,\omega\,+\,{\cal M}\,+\,180^{o}\,+\,\alpha\,+\,O(i^{2})\;\,$. Insertion thereof into (13) results in: $\;\omega^{*}_{{\it l}mpq}\;\equiv\;({\it l}-2p-m)\;\dot{\omega}^{*}\,+\,({\it l}-2p+q-m)\;\dot{\cal{M}}^{*}\,$, where we have neglected $\,\;-m\dot{\alpha}\,$ on account of $\,{\alpha}\,$ being extremely small. Clearly, the indices can assume more than one set of values corresponding to one tidal frequency. This way, the case of libration is more involved than that of tidal despinning, and deserves a separate consideration. it is sufficient to limit our consideration to the constant part of the torque,666 The tide-raising and tidally-perturbed satellites being the same body does _not_ yet mean that the indices $\,(p\,,\,q)\,$ coincide with with $\,(h\,,\,j)\,$. These are two independent sets of indices, wherewith the terms of two Fourier series are numbered, expression (16) being a product of those two series. This product contains a constant part, as well as short- period terms proportional to $\;\dot{\cal{M}}\;$ and long-period terms proportional to $\,\dot{\omega}\,$. The short-period terms get averaged out over a period of the tidal flexure, while the long-period terms get averaged out over longer times, _provided the periapse is precessing and not librating_. Expression (17) furnishes the constant part of the torque. Fortunately, this is sufficient for our further calculations. a part for which the indices $\,(p\,,\,q)\,$ coincide with $\,(h\,,\,j)\,$, and therefore $\,v_{{\it l}mhj}\,$ cancels with $\,v_{{\it l}mpq}^{*}\,$. This will give us: $\displaystyle{\tau}~{}=~{}\sum_{{\it{l}}=2}^{\infty}2~{}k_{\it l}~{}G~{}M^{\textstyle{{}^{2}}}~{}R^{\textstyle{{}^{2{\it{l}}\,+\,1}}}a^{\textstyle{{}^{-\,2\,{\it{l}}\,-\,2}}}\sum_{m=0}^{\it l}\frac{({\it{l}}\,-\,m)!}{({\it{l}}\,+\,m)!}\;m\;\sum^{\it l}_{p=0}\;F^{\textstyle{{}^{2}}}_{{\it{l}}mp}({\it i})\sum^{\it\infty}_{q=-\infty}G^{\textstyle{{}^{2}}}_{{\it{l}}pq}(e)\;\sin\epsilon_{{\it{l}}mpq}\;\;\;.~{}~{}~{}$ (17) The expression gets considerably simplified if we restrict ourselves to the case of $\,{\it l}\,=\,2\,$. Since $\,0\,\leq\,m\,\leq\,{\it l}\,$, and since $\,m\,$ enters the expansion as a multiplier, we see that only $\,m\,=\,1\,,\,2\,$ actually matter. As $\,0\,\leq\,p\,\leq\,{\it l}\,$, we are left with only six relevant $\,F$’s, those corresponding to $\;(\it{l}mp)\,=\,$ (210), (211), (212), (220), (221), and (222). By a direct inspection of the table of $\,F_{\it{l}mp}\,$ we find that five of these six functions happen to be $\,O({\it i})\,$ or $\,O(\,{\it i}^{2}\,)\,$, the sixth one being $\,F_{220}\,=\,\frac{\textstyle 3}{\textstyle 4}\,\left(\,1\,+\,\cos{\it i}\,\right)^{2}\,=\,3\,+\,O({\it i}^{2})\,$. Thus we obtain, in the leading order of $\,{\it i}\;$: $\displaystyle{\tau}_{\textstyle{{}_{{}_{\textstyle{{}_{l=2}}}}}}~{}=~{}\frac{3}{2}~{}\sum_{q=-\infty}^{\infty}~{}G~{}M^{\textstyle{{}^{2}}}~{}~{}R^{\textstyle{{}^{5}}}\;a^{-6}\;G^{\textstyle{{}^{2}}}_{\textstyle{{}_{\textstyle{{}_{20\mbox{\it{q}}}}}}}(e)\;k_{{2}}\;\sin\epsilon_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}\;+\;O({\it i}^{2}/Q)\;\;\;.$ (18) The leading term of the expansion is $\displaystyle{\tau}_{\textstyle{{}_{\textstyle{{}_{\textstyle{{}_{2200}}}}}}}~{}=~{}\frac{3}{2}~{}G\,M^{2}~{}k_{2}~{}R^{\textstyle{{}^{5}}}\;a^{{{-\,6}}}\sin\epsilon_{\textstyle{{}_{2200}}}\;\;\;.$ (19) Switching from the lags to quality factors via formula777 The phase lag $\,\epsilon_{\it{l}mpq}\,$ is introduced in (12), while the tidal harmonic $\,\omega_{\it{l}mpq}\,$ is given by (13). The quality factor $\,Q_{\it{l}mpq}\,=\,|\,\cot\epsilon_{\it{l}mpq}\,|$ is, for physical reasons, positively defined. Hence the multiplier $\,\mbox{sgn}\,\omega_{\it{l}mpq}\,$ in (21). (As ever, the function $\,\mbox{sgn}(x)\,$ is defined to assume the values $\,+1\,$, $\,-1\,$, or $\,0\,$ for positive, negative, or vanishing $\,x\,$, correspondingly.) Mind that no factor of two appears in (20 \- 21), because $\epsilon$ is a phase lag, not a geometric angle. $\displaystyle Q_{\it{l}mpq}\,=\,|\,\cot\epsilon_{\it{l}mpq}\,|\;\;\;,$ (20) we obtain: $\displaystyle\sin\epsilon_{\textstyle{{}_{\textstyle{{}_{{{{\it{l}}mpq}}}}}}}=\,\sin|\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}|\;\,\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}=\,\frac{\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}\cot^{2}\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}=\;\frac{\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\;}{\sqrt{{\textstyle 1~{}+~{}Q^{\textstyle{{}^{2}}}_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}}}=~{}\frac{~{}\mbox{sgn}\,\omega_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}~{}}{Q_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}}+O(Q^{-3})~{}~{},~{}~{}~{}$ (21) whence $\displaystyle{\tau}_{\textstyle{{}_{{}_{\textstyle{{}_{l=2}}}}}}~{}=~{}\frac{3}{2}~{}\sum_{q=-\infty}^{\infty}~{}G~{}M^{2}~{}\;R^{\textstyle{{}^{5}}}\;a^{-6}\;G^{\textstyle{{}^{2}}}_{\textstyle{{}_{20\mbox{\it{q}}}}}(e)\;k_{\textstyle{{}_{2}}}\;\frac{~{}\mbox{sgn}\,\omega_{\textstyle{{}_{220\mbox{\it{q}}}}}\,}{Q_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}}\;\,\,+\,O({\it i}^{2}/Q)\,+\,O(Q^{-3})\;\;\;.$ Now, let us simplify the sign multiplier. If in expression (13) for $\omega_{\textstyle{{}_{{\it{l}}mpq}}}$ we get rid of the redundant asterisks, replace888 While in the undisturbed two-body setting $\,{\cal{M}}\,=\,{\cal{M}}_{0}+n\,(t-t_{0})\,$ and $\,\dot{\cal{M}}=n\,$, under perturbation these relations get altered. One possibility is to introduce (following Tisserand 1893) an _osculating mean motion_ $\,n(t)\,\equiv\,\sqrt{\mu/a(t)^{3}}\,$, and to stick to this definition under perturbation. Then the mean anomaly will evolve as $\,{\cal{M}}\,=\,{\cal{M}}_{0}(t)+\int_{t_{o}}n(t)\,dt\,$, whence $\,\dot{\cal{M}}=\dot{\cal{M}}_{0}(t)+n(t)\,$. Other possibilities include introducing an _apparent_ mean motion, i.e., defining $\,n\,$ either as the mean-anomaly rate $\,d{\cal M}/dt\,$, or as the mean-longitude rate $\,dL/dt\,=\,d\Omega/dt\,+\,d\omega/dt\,+\,d{\cal{M}}/dt\,$ (as was done by Williams et al. 2001). While the first-order perturbations in $a(t)$ and in the osculating mean motion $\sqrt{\mu/a(t)^{3}}$ do not have secular rates, the epoch terms typically do have secular rates. Hence the difference between the apparent mean motion defined as $dL/dt$ (or as $d{\cal M}/dt$) and the osculating mean motion $\sqrt{\mu/a(t)^{3}}$. I am thankful to James G. Williams for drawing my attention to this circumstance (J.G. Williams, private communication). $\;\dot{\cal{M}}$ with $\dot{\cal{M}}_{0}+n\approx n$, and set ${\it{l}}=m=2$ and $p=0$, the outcome will be: $\displaystyle\mbox{sgn}\,\omega_{\textstyle{{}_{220\mbox{\it{q}}}}}\;=\;\mbox{sgn}\,\left[\,2\;\dot{\omega}\,+\,(2+q)\;n\,+\,2\,\dot{\Omega}-\,2\,\dot{\theta}\,\right]\;=\;\mbox{sgn}\,\left[\,\dot{\omega}\,+\,\left(1\,+\,\frac{\textstyle q}{\textstyle 2}\,\right)\;n\,+\,\dot{\Omega}-\,\dot{\theta}\,\right]~{}~{}~{}.$ As the node and periapse precessions are slow, the above expression may be simplified to $\displaystyle\mbox{sgn}\,\left[\,\left(1\,+\,\frac{\textstyle q}{\textstyle 2}\,\right)\;n\,-\,\dot{\theta}\,\right]~{}~{}~{}.$ All in all, the approximation for the torque assumes the form: $\displaystyle{\tau}_{\textstyle{{}_{{}_{\textstyle{{}_{l=2}}}}}}\,=\,\frac{3}{2}~{}\sum_{q=-\infty}^{\infty}~{}G~{}M^{2}~{}\;R^{\textstyle{{}^{5}}}\;a^{-6}\,G^{\textstyle{{}^{2}}}_{\textstyle{{}_{20\mbox{\it{q}}}}}(e)\;k_{\textstyle{{}_{2}}}\;{Q^{\textstyle{{}^{-1}}}_{\textstyle{{}_{\textstyle{{}_{220\mbox{\it{q}}}}}}}}\;~{}\mbox{sgn}\,\left[\,\left(1\,+\,\frac{\textstyle q}{\textstyle 2}\,\right)\;n\,-\,\dot{\theta}\,\right]+O({\it i}^{2}/Q)+O(Q^{-3})\;\;.~{}~{}~{}$ (22) That the sign of the right-hand side in the above formula is correct can be checked through the following obvious observation: for a sufficiently high spin rate $\,\dot{\theta}\,$ of the planet, the multiplier $\,\mbox{sgn}\,\left[\,\left(1\,+\,\frac{\textstyle q}{\textstyle 2}\,\right)\;n\,-\,\dot{\theta}\,\right]\,$ becomes negative. Thereby the overall expression for $\,{\tau}_{\textstyle{{}_{{}_{\textstyle{{}_{l=2}}}}}}\,$ acquires a “minus” sign, so that the torque points out in the direction of rotation opposite to the direction of increase of the sidereal angle $\,\theta\,$. This is exactly how it should be, because for a fixed $\,q\,$ and a sufficiently fast spin the $\,q$’s component of the tidal torque must be decelerating and driving the planet to synchronous rotation. Expansion (22) was written down for the first time, without proof, by Goldreich & Peale (1966). A schematic proof was later offered by Dobrovolskis (2007). ## 5 Can the quality factor scale as a positive power of the tidal frequency? As of now, the functional form of the dependence $\,Q(\chi)\,$ for Jovian planets remains unknown. For terrestrial planets, the model $\,Q\,\sim\,1/\chi\,$ is definitely incompatible with the geophysical data. A convincing volume of measurements firmly witnesses that $\,Q\,$ of the mantle scales as the tidal frequency to a _positive_ fractional power: $\displaystyle Q\;=\;{\cal E}^{\alpha}\;\chi^{\alpha}\;\;\;,~{}~{}~{}~{}~{}\mbox{where}\;\;\;\alpha\,=\,0.3\,\pm\,0.1\;\;\;,$ (23) ${\cal E}\,$ being an integral rheological parameter with dimensions of time. This rheology is incompatible with the postulate of frequency-independent time-delay. Therefore an honest calculation should be based on averaging the Darwin-Kaula-Goldreich formula (22), with the actual scaling law (23) inserted therein, and with the appropriate dependence $\,\Delta t_{\textstyle{{}_{lmpq}}}(\,\chi_{\textstyle{{}_{lmpq}}}\,)\,$ taken into account.999 For the dependence of $\,\Delta t_{\textstyle{{}_{lmpq}}}\,$ upon $\,\chi_{\textstyle{{}_{lmpq}}}\,$ see Efroimsky & Lainey 2007. ### 5.1 The “paradox” Although among geophysicists the scaling law (23) has long become common knowledge, in the astronomical community it is often met with prejudice. The prejudice stems from the fact that, in the expression for the torque, $\,Q\,$ stands in the denominator: $\displaystyle\tau\;\sim\;\frac{1}{Q}\;\;\;\,.\,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (24) At the instant of crossing the synchronous orbit, the principal tidal frequency $\,\chi_{\textstyle{{}_{\textstyle{{}_{2200}}}}}\,$ becomes nil, for which reason insertion of $\displaystyle Q\;\sim\;\chi^{\alpha}\;\;\;,~{}~{}~{}~{}~{}\alpha\,>\,0\;\;\;$ (25) into (24) seems to entail an infinitely large torque at the instant of crossing: $\displaystyle\tau\;\sim\;\frac{1}{Q}\;\sim\;\frac{1}{\chi^{\alpha}}\;\rightarrow\;\infty~{}~{},~{}~{}~{}~{}~{}{\mbox{for}}~{}~{}~{}\chi\;\rightarrow\;0~{}~{}~{},$ (26) a clearly unphysical result. Another, very similar objection to (23) originates from the fact that the quality factor is inversely proportional to the phase shift: $\,Q\,\sim\,1/\epsilon\,$. As the shift (12) vanishes on crossing the synchronous orbit, one may think that the value of the quality factor must, effectively, approach infinity. On the other hand, the principal tidal frequency vanishes on crossing the synchronous orbit, for which reason (23) makes the quality factor vanish. Thus we come to a contradiction. For these reasons, the long-entrenched opinion is that these models introduce discontinuities into the expression for the torque, and can thus be considered as unrealistic. It is indeed true that, while law (23) works over scales shorter than the Maxwell time (about $\,10^{2}$ yr for most minerals), it remains subject to discussion in regard to longer timescales. Nonetheless, it should be clearly emphasised that the infinities emerging at the synchronous-orbit crossing can in no way disprove any kind of rheological model. They can only disprove the flawed mathematics whence they provene. ### 5.2 A case for reasonable doubt To evaluate the physical merit of the alleged infinite-torque “paradox”, recall the definition of the quality factor. As part and parcel of the linearity approximation, the overall damping inside a body is expanded in a sum of attenuation rates corresponding to each periodic disturbance: $\displaystyle\langle\,\dot{E}\;\rangle\;=\;\sum_{i}\;\langle\,\dot{E}(\chi_{\textstyle{{}_{i}}})\;\rangle$ (27) where, at each frequency $\,\chi_{i}\,$, $\displaystyle\langle\,\dot{E}(\chi_{\textstyle{{}_{i}}})~{}\rangle~{}=~{}-~{}2~{}\chi_{\textstyle{{}_{i}}}~{}\frac{\,\langle\,E(\chi_{\textstyle{{}_{i}}})~{}\rangle\,}{Q(\chi_{\textstyle{{}_{i}}})}~{}=\,\;-\;\chi_{\textstyle{{}_{i}}}\;\frac{\,E_{{}_{peak}}(\chi_{\textstyle{{}_{i}}})\,}{Q(\chi_{\textstyle{{}_{i}}})}\;\;\;,$ (28) $\langle\,.\,.\,.\,\rangle~{}$ designating an average over a flexure cycle, $\,E(\chi_{\textstyle{{}_{i}}})\,$ denoting the energy of deformation at the frequency $\,\chi_{\textstyle{{}_{i}}}\,$, and $Q(\chi_{\textstyle{{}_{i}}})\,$ being the quality factor of the medium at this frequency. This definition by itself leaves enough room for doubt in the above “paradox”. As can be seen from (28), the dissipation rate is proportional not to $\;1/Q(\chi)\;$ but to $\;\chi/Q(\chi)\;$. This way, for the dependence $\,Q\,\sim\,\chi^{\alpha}\,$, the dissipation rate $\,\langle\dot{E}\rangle\,$ will behave as $\,\chi^{1-\alpha}\,\;$. In the limit of $\,\chi\,\rightarrow\,0\,$, this scaling law portends no visible difficulties, at least for the values of $\,\alpha\,$ up to unity. While raising $\,\alpha\,$ above unity may indeed be problematic, there seem to be no fundamental obstacle to having materials with positive $\,\alpha\,$ taking values up to unity. So far, such values of $\,\alpha\,$ have caused no paradoxes, and there seems to be no reason for any infinities to show up. ### 5.3 The phase shift and the quality factor As another preparatory step, we recall that, rigorously speaking, the torque is proportional not to the phase shift $\,\epsilon\,$ itself but to $\,\sin\epsilon\,$. From (21) and (23) we obtain: $\displaystyle|\,\sin\epsilon\,|\;=\;\frac{1}{\textstyle\sqrt{1\,+\,Q^{\textstyle{{}^{2}}}}}\;=\;\frac{1}{\sqrt{1\,+\;{\cal E}^{\textstyle{{}^{2\alpha}}}\;{\chi}^{\textstyle{{}^{2\alpha}}}\;}}\;\;\;.$ (29) We see that only for large values of $\,Q\,$ one can approximate $\;\,|\,\sin\epsilon\,|\;\,$ with $\,1/Q\,$ (crossing of the synchronous orbit _not_ being the case). Generally, in any expression for the torque, the factor $\,1/Q\,$ must always be replaced with $\,1/\sqrt{1\,+\,Q^{2}}\,\;$. Thus instead of (24) we must write: $\displaystyle\tau\;\sim\;|\,\sin\epsilon\,|\;=\;\frac{1}{\sqrt{1\,+\,Q^{\textstyle{{}^{2}}}\;}}\;=\;\frac{1}{\sqrt{1\,+\;{\cal E}^{\textstyle{{}^{2\alpha}}}\;{\chi}^{\textstyle{{}^{2\alpha}}}\;}}\;\;\;\;\,,\,~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (30) ${\cal E}\,$ being a dimensional constant from (23). Though this immediately spares us from the fake infinities at $\,\chi\,\rightarrow\,0\,$, we still are facing this strange situation: it follows from (29) that, for a positive $\,\alpha\,$ and vanishing $\,\chi\,$, the phase lag $\,\epsilon\,$ must be approaching $\,\pi/2\,$, thereby inflating the torque to its maximal value (while on physical grounds the torque should vanish for zero $\,\chi\,$). Evidently, some important details are still missing from the picture. ### 5.4 The stone rejected by the builders To find the missing link, recall that Kaula (1964) described tidal damping by employing the method suggested by Darwin (1880): he accounted for attenuation by merely adding a phase shift to every harmonic involved – an empirical approach intended to make up for the lack of a consistent hydrodynamical treatment with viscosity included. It should be said, however, that prior to the work of 1880 Darwin had published a less known article (Darwin 1879), in which he attempted to construct a self-consistent theory, one based on the viscosity factor of the mantle, and not on empirical phase shifts inserted by hand. Darwin’s conclusions of 1879 were summarised and explained in a more general mathematical setting by Alexander (1973). The pivotal result of the self-consistent hydrodynamical study is the following. When a variation of the potential of a tidally disturbed planet, $\,U(\mbox{{\boldmath$\vec{r}$}})\,$, is expanded over the Legendre functions $\,P_{{\it{l}}m}(\sin\phi)\,$, each term of this expansion will acquire not only a phase lag but also a factor describing a change in amplitude. This forgotten factor, derived by Darwin (1879), is nothing else but $\;\,\cos\epsilon\;$. Its emergence should in no way be surprising if we recall that the damped, forced harmonic oscillator $\displaystyle\ddot{x}\;+\;2\;\gamma\;\dot{x}\;+\;\omega^{2}_{o}\,x\;=\;F\;e^{{\it i}\,\lambda\,t}$ (31) evolves as $\displaystyle x(t)~{}=~{}C_{1}~{}\,e^{\textstyle{{}^{(\,-\,\gamma\,+\,{\it i}\,\sqrt{\omega_{o}^{2}-\lambda^{2}\,}\,)\;t}}}\,+\;C_{2}~{}\,e^{\textstyle{{}^{(\,-\,\gamma\,+\,{\it i}\,\sqrt{\omega_{o}^{2}-\lambda^{2}\,}\,)~{}t}}}\,+~{}\frac{F~{}\cos\epsilon}{\omega_{o}^{2}\,-\,\lambda^{2}}~{}\,e^{\textstyle{{}^{{\it i}\,(\lambda\,t\,-\,\epsilon)}}}\;\;\;,$ (32) where the phase lag is $\displaystyle\tan\epsilon\;=\;2\;\gamma\;\lambda\;\left(\,\omega_{o}^{2}\,-\;\lambda^{2}\,\right)\;\;\;,$ (33) and the first two terms in (32) are damped away in time.101010 As demonstrated by Alexander (1973), this example indeed has relevance to the hydrodynamical theory of Darwin, and is not a mere illustration. Alexander (1973) also explained that the emergence of the $\,\cos\epsilon\,$ factor is generic. (Darwin (1879) had obtained it in the simple case of $\,{\it l}\,=\,2\,$ and for a special value of the Love number: $\,k{\it{{}_{l}}}=\,1.5\,$.) A further investigation of this issue was undertaken in a comprehensive work by Churkin (1998), which unfortunately has never been published in English because of a tragic death of its Author. In this preprint, Churkin explored the frequency- dependence of both the Love number $\,k_{2}\,$ and the quality factor within a broad variety of rheological models, including those of Maxwell and Voight. It follows from Churkin’s formulae that within the Voight model the dynamical $\,k_{2}\,$ relates to the static one as $\,\cos\epsilon\,$. In the Maxwell and other models, the ratio approaches $\,\cos\epsilon\,$ in the low-frequency limit. In the works by Darwin’s successors, the allegedly irrelevant factor of $\,\cos\epsilon\,$ fell through the cracks, because the lag was always asserted to be small. In reality, though, each term in the Fourier expansions (11), (16 \- 19), and (22) should be amended with $\,\cos\epsilon_{\textstyle{{}_{\textstyle{{}_{{\it{l}}mpq}}}}}\,$. For the same reason, instead of (30), we should write down: $\displaystyle\tau\;\sim\;|\,\cos\epsilon\;\,\sin\epsilon\,|\;=\;\frac{Q}{\sqrt{1\,+\,Q^{\textstyle{{}^{2}}}\;}}\;\frac{1}{\sqrt{1\,+\,Q^{\textstyle{{}^{2}}}\;}}\;=\;\frac{{\cal E}^{\textstyle{{}^{\alpha}}}\;{\chi}^{\textstyle{{}^{\alpha}}}}{1\,+\;{\cal E}^{\textstyle{{}^{2\alpha}}}\;{\chi}^{\textstyle{{}^{2\alpha}}}}\;\;\;\;\,,\,~{}$ (34) At this point, it would be tempting to conclude that, since (71) vanishes in the limit of $\chi\rightarrow 0\;$, _for any sign_ _of_ $\alpha\;$, then no paradoxes happens on the satellite’s crossing the synchronous orbit. Sadly, this straightforward logic would be too simplistic. In fact, prior to saying that $\,\cos\epsilon\,\sin\epsilon\rightarrow 0$, we must take into consideration one more subtlety missed so far. As demonstrated in the Appendix, taking the limit of $Q\rightarrow 0$ is a nontrivial procedure, because at small values of $\,Q\,$ the interconnection between the lag and the Q factor becomes very different from the conventional $Q=\cot|\epsilon|$. A laborious calculation shows that, for $\;Q<1-\pi/4\,$, the relation becomes: $\displaystyle\sin\epsilon\,\cos\epsilon\,=\,\pm\;(3Q)^{1/3}\,\left[1-\frac{4}{5}(3Q)^{2/3}+O(Q^{4/3})\right]\;\;\;,$ which indeed vanishes for $Q\rightarrow 0$. Both $\,\epsilon_{\textstyle{{}_{2200}}}\,$ and the appropriate component of the torque change their sign on the satellite crossing the synchronous orbit. So the main conclusion remains in force: nothing wrong happens on crossing the synchronous orbit, Q.E.D. ## 6 Conclusions In the article thus far we have punctiliously spelled out some assumptions that often remain implicit, and brought to light those steps in calculations, which are often omitted as “self-evident”. This has helped us to explain that no “paradoxes” ensue from the frequency-dependence $\;Q\,\sim\,\chi^{\alpha}\;\;,\;\;\alpha\,=\,0.3\,\pm\,0.1\;$, which is in fact the actual dependence found for the mantle and crust. This preprint is a pilot paper. A more comprehensive treatise on tidal torques is to be published. (Efroimsky & Williams 2009) Acknowledgments I deeply thank Bruce Bills, Alessandra Celletti, Tony Dobrovolskis, Peter Goldreich, Shun-ichiro Karato, Valery Lainey, William Newman, Stan Peale, S. Fred Singer, and Gabriel Tobie – the colleagues with whom I on many occasions had stimulating conversations on the theory of tides, and from whom I have learned much. My special gratitude goes to Sylvio Ferraz Mello and James G. Williams, who kindly offered numerous valuable comments on my text. Appendix. The lag and the quality factor: is the formula $\boldmath{Q=\cot|\mbox{{\boldmath$\epsilon$}}|}$ universal? The interrelation between the quality factor $\,Q\,$ and the phase lag $\,\epsilon\,$ is long-known to be $\displaystyle Q\,=\,\cot|\epsilon|\;\;\;,$ (35) and its derivation can be found in many books. In Appendix A2 of Efroimsky & Lainey(2007), that derivation is reproduced, with several details that are normally omitted in the literature. Among other things, we pointed out that the interrelation has exactly the form (35) only in the limit of small lags. For large phase lags, the form of this relation will change considerably. Since in section 9 of the current paper we address the case of large lags, it would be worth reconsidering the derivation presented in Efroimsky & Lainey (2007), and correcting a subtle omission made there. Before writing formulae, let us recall that, at each frequency $\,\chi\,$ in the spectrum of the deformation, the quality factor (divided by $\,2\,\pi\,$) is defined as the peak energy stored in the system divided by the energy damped over a cycle of flexure: $\displaystyle{Q}(\chi)\;\equiv\;-\;\frac{2\;\pi\;E_{peak}(\chi)}{\Delta E_{cycle}(\chi)}\;\;\;,$ (36) where $\,\Delta E_{cycle}(\chi)\,<\,0\,$ as we are talking about energy losses.111111 We are considering flexure in the linear approximation. Thus at each frequency $\,\chi\,$ the appropriate energy loss over a cycle, $\,\Delta E_{cycle}(\chi)\,$, depends solely on the maximal energy stored at that same frequency, $\,E_{peak}(\chi)\,$. An attempt to consider large lags (all the way up to $\,|\epsilon|\,=\,\pi/2\,$) sets the values of $\,Q/2\pi\,$ below unity. As the dissipated energy cannot exceed the energy stored in a free oscillator, the question becomes whether the values of $\,Q/2\pi\,$ can be that small. To understand that they can, recall that in this situation we are considering an oscillator, which is not free but is driven (and is overdamped). The quality factor being much less than unity simply implies that the eigenfrequencies get damped away during less than one oscillation. Nonetheless, motion goes on due to the driving force. Now let us switch to the specific context of tides. To begin with, let us recall that the dissipation rate in a tidally distorted primary is well approximated by the work that the secondary carries out to deform the primary: $\displaystyle\dot{E}\;=\;-\;\int\,\rho\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\;\cdot\;\nabla W\;d^{3}x$ (37) $\rho\,,\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\,$, and $\,W\,$ denoting the density, velocity, and tidal potential inside the primary. The expression on the right-hand side can be transformed by means of the formula $\displaystyle\rho\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot\nabla W\,=\,\nabla\cdot(\rho\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\,W)\,-\,W\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot\nabla\rho\,-\,W\,\nabla\cdot(\rho\,\mbox{\boldmath$\vec{\boldmath{\,V}}$})\,=\,\nabla\cdot(\rho\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\,W)\,-\,W\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot\nabla\rho\,+\,W\,\frac{\partial\rho}{\partial t}\;\;,\;\;\;\;$ (38) where the $\,W\,\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot\nabla\rho\,$ and $\,\partial\rho/\partial t\,$ terms may be omitted under the assumption that the primary is homogeneous and incompressible. In this approximation, the attenuation rate becomes simply $\displaystyle\dot{E}\;=\;-\;\int\,\nabla\,\cdot\,(\rho\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\;W)\,d^{3}x\;=\;-\;\int\,\rho\;W\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\,\cdot\,{\vec{\bf{n}}}\;\,dA\;\;\;,$ (39) ${\vec{\bf{n}}}\,$ being the outward normal to the surface of the primary, and $\,dA\,$ being an element of the surface area. It is now clear that, under the said assertions, it is sufficient to take into account only the radial elevation rate, not the horizontal distortion. This way, formula (37), in application to a unit mass, will get simplified to $\displaystyle\dot{E}\;=\;\left(-\,\frac{\partial W}{\partial r}\right)\;\mbox{\boldmath$\vec{\boldmath{\,V}}$}\cdot{\vec{\bf{n}}}\;=\;\left(-\,\frac{\partial W}{\partial r}\right)\frac{d\zeta}{dt}\;\;\;,$ (40) $\zeta\,$ standing for the vertical displacement (which is, of course, delayed in time, compared to $\,W\,$). The amount of energy dissipated over a time interval $\,(t_{o}\,,\;t)\,$ is then $\displaystyle\Delta{E}\;=\;\int^{t}_{t_{o}}\;\left(-\,\frac{\partial W}{\partial r}\right)\;d\zeta\;\;\;.$ (41) We shall consider the simple case of an equatorial moon on a circular orbit. At each point of the planet, the variable part of the tidal potential produced by this moon will read $\displaystyle W\;=\;W_{o}\;\cos\chi t\;\;\;,$ (42) the tidal frequency being given by $\displaystyle\chi\,=\,2~{}|n\;-\;\omega_{p}|~{}~{}~{}.~{}~{}~{}$ (43) Let g denote the surface free-fall acceleration. An element of the planet’s surface lying beneath the satellite’s trajectory will then experience a vertical elevation of $\displaystyle\zeta\;=\;h_{2}\;\frac{W_{o}}{\mbox{g}}\;\cos(\chi t\;-\;|\epsilon|)\;\;\;,$ (44) $\,h_{2}\,$ being the corresponding Love number, and $\,|\epsilon|\,$ being the _positive_ 121212 Were we not considering the simple case of a circular orbit, then, rigorously speaking, the expression for $\,W\,$ would read not as $\,W_{o}\,\cos\chi t\,$ but as $\,W_{o}\,\cos\omega_{\textstyle{{}_{tidal}}}t\,$, the tidal frequency $\,\omega_{\textstyle{{}_{tidal}}}\,$ taking both positive and negative values, and the physical frequency of flexure being $\,\chi\,\equiv\,|\omega_{\textstyle{{}_{tidal}}}|\,$. Accordingly, the expression for $\,\zeta\,$ would contain not $\,\cos(\chi t\,-\,|\epsilon|)\,$ but $\,\cos(\omega_{\textstyle{{}_{tidal}}}t\,-\,\epsilon)\,$. As we saw in equation (24), the sign of $\,\epsilon\,$ is always the same as that of $\,\omega_{\textstyle{{}_{tidal}}}\,$. For this reason, one may simply deal with the physical frequency $\,\chi\,\equiv\,|\omega_{\textstyle{{}_{tidal}}}|\,$ and with the absolute value of the phase lag, $\;|\epsilon|\;$. phase lag, which for the principal tidal frequency is simply the double geometric angle $\,\delta\,$ subtended at the primary’s centre between the directions to the secondary and to the main bulge: $\displaystyle|\epsilon|\;=\;2\;\delta\;\;\;.$ (45) Accordingly, the vertical velocity of this element of the planet’s surface will amount to $\displaystyle u\;=\;\dot{\zeta}\;=\;-\;h_{2}\;\chi\;\frac{W_{o}}{\mbox{g}}\;\sin(\chi t\;-\;|\epsilon|)\;=\;-\;h_{2}\;\chi\;\frac{W_{o}}{\mbox{g}}\;\left(\sin\chi t\;\cos|\epsilon|\;-\;\cos\chi t\;\sin|\epsilon|\right)\;\;.\;\;$ (46) The expression for the velocity has such a simple form because in this case the instantaneous frequency $\,\chi\,$ is constant. The satellite generates two bulges – on the facing and opposite sides of the planet – so each point of the surface is uplifted twice through a cycle. This entails the factor of two in the expression (43) for the frequency. The phase in (45), too, is doubled, though the necessity of this is less evident, – see footnote 4 in Appendix A1 to Efroimsky & Lainey (2007). The energy dissipated over a time cycle $\,T\,=\,2\pi/\chi\,$, per unit mass, will, in neglect of horizontal displacements, be $\displaystyle\Delta E_{{}_{cycle}}$ $\displaystyle=$ $\displaystyle\int^{T}_{0}u\left(-\,\frac{\partial W}{\partial r}\right)dt=\,-\left(-\,h_{2}\;\chi\frac{W_{o}}{\mbox{g}}\right)\,\frac{\partial W_{o}}{\partial r}\int^{t=T}_{t=0}\cos\chi t\,\left(\sin\chi t\,\cos|\epsilon|\,-\,\cos\chi t\,\sin|\epsilon|\right)dt$ (47) $\displaystyle=$ $\displaystyle\,-\;h_{2}\;\chi\;\frac{W_{o}}{\mbox{g}}\;\frac{\partial W_{o}}{\partial r}\;\sin|\epsilon|\,\;\frac{1}{\chi}\;\int^{\chi t\,=\,2\pi}_{\chi t\,=\,0}\;\cos^{2}\chi t\;\;d(\chi t)\;=\;-\;h_{2}\;\frac{W_{o}}{\mbox{g}}\;\frac{\partial W_{o}}{\partial r}\;\pi\;\sin|\epsilon|\;\;,\;\;\;~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ while the peak energy stored in the system during the cycle will read: $\displaystyle E_{{}_{peak}}$ $\displaystyle=$ $\displaystyle\int^{T/4}_{|\epsilon|/\chi}u\left(-\,\frac{\partial W}{\partial r}\right)dt=\,-\left(-\,h_{2}\;\chi\,\frac{W_{o}}{\mbox{g}}\right)\frac{\partial W_{o}}{\partial r}\int^{t=T/4}_{t=|\epsilon|/\chi}\cos\chi t\,\left(\sin\chi t\,\cos|\epsilon|\,-\,\cos\chi t\,\sin|\epsilon|\right)dt$ (48) $\displaystyle=$ $\displaystyle\;\chi\;h_{2}\;\frac{W_{o}}{\mbox{g}}\;\frac{\partial W_{o}}{\partial r}\;\left[\;\frac{\cos|\epsilon|}{\chi}\;\int^{\chi t\,=\,\pi/2}_{\chi t\,=\,|\epsilon|}\;\cos\chi t\;\sin\chi t\;\;d(\chi t)\;-\;\frac{\sin|\epsilon|}{\chi}\;\int^{\chi t\,=\,\pi/2}_{\chi t\,=\,|\epsilon|}\;\cos^{2}\chi t\;\;d(\chi t)\;\right]\;\;.~{}~{}~{}~{}~{}~{}~{}~{}\,$ In the appropriate expression in Appendix A1 to Efroimsky & Lainey (2007), the lower limit of integration was erroneously set to be zero. To understand that in reality integration over $\,\chi\,t\,$ should begin from $\,|\epsilon|\,$, one should superimpose the plots of the two functions involved, $\,\cos\chi t\,$ and $\,\sin(\chi t\,-\,|\epsilon|)\,$. The maximal energy gets stored in the system after integration through the entire interval over which both functions have the same sign. Hence $\,\chi t=|\epsilon|\,$ as the lower limit. Evaluation of the integrals entails: $\displaystyle E_{peak}\;=\;h_{2}\;\frac{W_{o}}{\mbox{g}}\;\frac{\partial W_{o}}{\partial r}\;\left[\;\frac{1}{2}\;\cos|\epsilon|\;-\;\frac{1}{2}\;\left(\;\frac{\pi}{2}\;-\;|\epsilon|\;\right)\;\sin|\epsilon|\;\right]~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (49) whence $\displaystyle Q^{-1}\;=\;\frac{-\;\Delta E_{{}_{cycle}}}{2\,\pi\,E_{{}_{peak}}}\;=\;\frac{1}{2\,\pi}\;\,\frac{\pi\;\sin|\epsilon|}{~{}\frac{\textstyle 1}{\textstyle 2}\;\cos|\epsilon|\;-\;\frac{\textstyle 1}{\textstyle 2}\;\left(\;\frac{\textstyle\pi}{\textstyle 2}\;-\;|\epsilon|\;\right)\;\sin|\epsilon|}\;=\;\frac{\tan|\epsilon|}{1\;-\;\left(\;\frac{\textstyle\pi}{\textstyle 2}\;-\;|\epsilon|\;\right)\;\tan|\epsilon|}\;\;\;.~{}~{}~{}~{}~{}$ (50) As can be seen from (50), both the product $\,\sin\epsilon\,\cos\epsilon\,$ and the appropriate component of the torque attain their maxima when $\,Q\,=\,1\,-\,\pi/4\,$. Usually, $\,|\epsilon|\,$ is small, and we arrive at the customary expression $\displaystyle Q^{-1}\,=\,\tan|\epsilon|\;+\;O(\epsilon^{2})\;\;\;.$ (51) In the opposite situation, when $\,Q\rightarrow 0$ and $\,|\epsilon|\rightarrow\pi/2\,$, it is convenient to consider the small difference $\displaystyle\xi\;\equiv\;\frac{\pi}{2}\;-\;|\epsilon|\;\;\;,$ (52) in terms whereof the inverse quality factor will read: $\displaystyle Q^{-1}\,=\,\frac{\cot\xi}{1\;-\;\xi\;\cot\xi}\;=\;\frac{1}{\tan\xi\;-\;\xi}\;=\;\frac{1}{z\;-\;\arctan z}\;=\;\frac{1}{\frac{\textstyle 1}{\textstyle 3}\;z^{3}\;\left[\,1\;-\;\frac{\textstyle 3}{\textstyle 5}\;z^{2}\,+\;O(z^{4})\,\right]}\;\;\;,\;\;\;$ (53) where $\;z\,\equiv\,\tan\xi\;$ and, accordingly, $\;\xi\;=\;\arctan z\;=\;z\,-\frac{\textstyle 1}{\textstyle 3}\,z^{3}\,+\,\frac{\textstyle 1}{\textstyle 5}\,z^{5}\,+\,O(z^{7})\;\,.\;$ Formula (53) may, of course, be rewritten as $\displaystyle z^{3}\;\left[\,1\;-\;\frac{3}{5}\;z^{2}\;+\;O(z^{4})\;\right]\;=\;3\;Q\;\;\;$ (54) or, the same, as $\displaystyle z\;=\;(3\,Q)^{1/3}\;\left[\,1\;+\;\frac{1}{5}\;z^{2}\;+\;O(z^{4})\,\right]\;\;\;.$ (55) While the zeroth approximation is simply $\;z\,=\,(3Q)^{1/3}\,+\,O(Q)\;$, the first iteration gives: $\displaystyle\tan\xi\;\equiv\;z\;=\;(3Q)^{1/3}\,\left[\,1\;+\;\frac{1}{5}\;(3Q)^{2/3}\;+\;O(Q^{4/3})\,\right]\;=\;q\;\left[\,1\;+\;\frac{1}{5}\;q^{2}\;+\;O(q^{4})\,\right]\;\;\;,~{}~{}~{}$ (56) with $\,q\,=\,(3Q)^{1/3}\,$ playing the role of a small parameter. We now see that the customary relation (51) should be substituted, for large lags, i.e., for small131313 The afore-employed expansion of $\,\arctan z\,$ is valid for $\,|z|\,<\,1\,$. This inequality, along with (53), entails: $\,Q\,=\,z\,-\,\arctan z\,<\,1\,-\,\pi/4\,$. values of $\,Q\,$, with: $\displaystyle\tan|\epsilon|\;=\;(3Q)^{-1/3}\,\left[\,1\;-\;\frac{1}{5}\;(3Q)^{2/3}\;+\;O(Q^{4/3})\,\right]$ (57) The formula for the tidal torque contains a multiplier $\,\sin\epsilon\,\cos\epsilon\,$, whose absolute value can, for our purposes, be written down as $\displaystyle\sin|\epsilon|\,\cos|\epsilon|=\cos\xi\,\sin\xi=\frac{\tan\xi}{1+\tan^{2}\xi}\,=\,\frac{q\,\left[1+\frac{\textstyle 1}{\textstyle 5}\,q^{2}+O(q^{4})\right]}{1+q^{2}\left[1+O(q^{2})\right]}=(3Q)^{1/3}\left[1-\frac{4}{5}(3Q)^{2/3}+O(Q^{4/3})\right]\;,~{}~{}~{}$ (58) whence $\displaystyle\sin\epsilon\;\cos\epsilon\;=\;\pm\;(3Q)^{1/3}\left[1-\frac{4}{5}(3Q)^{2/3}+O(Q^{4/3})\right]\;,~{}~{}~{}$ (59) an expression vanishing for $\,Q\,\rightarrow\,0\;$. Be mindful that both $\,\epsilon_{\textstyle{{}_{2200}}}\,$ and the appropriate component of the torque change their sign on the satellite crossing the synchronous orbit. ## References * [1] Alexander, M. E. 1973. “The weak-friction approximation and tidal evolution in close binary systems.” _Astrophysics and Space Sciences_ , Vol. 23, pp. 459 - 510 * [2] Churkin, V. A. 1998. “The Love numbers for the models of inelastic Earth.” Preprint No 121. Institute of Applied Astronomy. St.Petersburg, Russia. /in Russian/ * [3] Correia, A. C. M., & Laskar, J. 2004. “Mercury’s capture into the $\,3/2\,$ spin-orbit resonance as a result of its chaotic dynamics.” _Nature_ , Vol. 429, pp. 848 - 850 * [4] Darwin, G. H. 1879. “On the precession of a viscous spheroid and on the remote history of the Earth” Philosophical Transactions of the Royal Society of London, Vol.170, pp 447 -530 http://www.jstor.org/view/02610523/ap000081/00a00010/ * [5] Darwin, G. H. 1880. “On the secular change in the elements of the orbit of a satellite revolving about a tidally distorted planet.” Philosophical Transactions of the Royal Society of London, Vol. 171, pp. 713 - 891 http://www.jstor.org/view/02610523/ap000082/00a00200 * [6] Darwin, G. H. 1908. “Tidal friction and cosmogony.” In: Darwin, G. H., _Scientific Papers,_ Vol.2. Cambridge University Press, NY 1908. * [7] Dobrovolskis, A. 2007. “Spin states and climates of eccentric exoplanets.” _Icarus_ , Vol. 192, pp. 1 - 23 * [8] Efroimsky, M., & Williams, J. G. 2009. “Tidal torques. A critical review of some techniques.” _Celestial Mechanics and Dynamical Astronomy_ , Vol. 104, pp. 257 - 289 arXiv:0803.3299 * [9] Efroimsky, M., & V. Lainey. 2007. “The Physics of Bodily Tides in Terrestrial Planets, and the Appropriate Scales of Dynamical Evolution.” _Journal of Geophysical Research – Planets_ , Vol. 112, p. E12003 doi:10.1029/2007JE002908 * [10] Ferraz-Mello, S., Rodríguez, A., & Hussmann, H. 2008. “Tidal friction in close-in satellites and exoplanets: The Darwin theory re-visited.” _Celestial mechanics and Dynamical Astronomy,_ Vol. 101, pp. 171 - 201 * [11] Goldreich, P. 1963. “On the eccentricity of satellite orbits in the Solar system.” _The Monthly Notes of the Royal Astronomical Society of London_ , Vol. 126, pp. 259 - 268\. * [12] Goldreich, P., & Peale, S. 1966. “Spin-orbit coupling in the Solar System.” _The Astronomical Journal_. Vol. 71, pp. 425 - 438 * [13] Karato, S.-i. 2007. _Deformation of Earth Materials. An Introduction to the Rheology of Solid Earth._ Cambridge University Press, UK. * [14] Kaula, W. M. 1961. “Analysis of gravitational and geometric aspects of geodetic utilisation of satellites.” The Geophysical Journal, Vol. 5, pp. 104 - 133 * [15] Kaula, W. M. 1964. “Tidal Dissipation by Solid Friction and the Resulting Orbital Evolution.” Reviews of Geophysics, Vol. 2, pp. 661 - 684 * [16] Mignard, F. 1979. “The Evolution of the Lunar Orbit Revisited. I.” The Moon and the Planets. Vol. 20, pp. 301 - 315. * [17] Mignard, F. 1980. “The Evolution of the Lunar Orbit Revisited. II.” The Moon and the Planets. Vol. 23, pp. 185 - 201 * [18] Peale, S. 2005. “The free precession and libration of Mercury.” Icarus, Vol 178, pp 4 - 18 * [19] Peale, S. J., and Cassen, P. 1978. “Contribution of Tidal Dissipation to Lunar Thermal History.” Icarus, Vol. 36, pp. 245 - 269 * [20] Singer, S. F. 1968. “The Origin of the Moon and Geophysical Consequences.” The Geophysical Journal of the Royal Astronomical Society, Vol. 15, pp. 205 - 226 * [21] Sokolnikoff, I. S. 1956. _Mathematical Theory of Elasticity._ 2nd Ed., McGraw-Hill NY 1956. * [22] Tisserand, F.-F. 1896. _Traité de Mécanique Céleste_. Tome I. _Perturbations des planètes d’après la méthode de la variation des constantes arbitraires._ Gauthier Villars, Paris 1896. Chapitre X. * [23] Williams, J. G., Boggs, D. H., Yoder, C. F., Ratcliff, J. T. & Dickey, J. O. 2001. “Lunar rotational dissipation in solid body and molten core.” _The Journal of Geophysical Research - Planets_ , Vol. 106, No E11, pp. 27933 - 27968.

arxiv-papers

2024-09-04T02:48:53.412420

0801.3099

# Gradient flow approach to geometric convergence analysis of preconditioned eigensolvers ††thanks: Received by the editors June 16, 2008; revised March 5, 2009; accepted March 16, 2009; published electronically ???????, 2009. Preliminary version http://arxiv.org/abs/0801.3099 Andrew V. Knyazev Department of Mathematical and Statistical Sciences, University Colorado Denver, P.O. Box 173364, Campus Box 170, Denver, CO 80217-3364 (andrew.knyazev at ucdenver.edu, http://math.ucdenver.edu/$\sim$aknyazev/). Supported by the NSF-DMS 0612751. Klaus Neymeyr Universität Rostock, Institut für Mathematik, Universitätsplatz 1, 18055 Rostock, Germany (klaus.neymeyr at mathematik.uni-rostock.de, http://cat.math.uni-rostock.de/$\sim$neymeyr/). ###### Abstract Preconditioned eigenvalue solvers (eigensolvers) are gaining popularity, but their convergence theory remains sparse and complex. We consider the simplest preconditioned eigensolver—the gradient iterative method with a fixed step size—for symmetric generalized eigenvalue problems, where we use the gradient of the Rayleigh quotient as an optimization direction. A sharp convergence rate bound for this method has been obtained in 2001–2003. It still remains the only known such bound for any of the methods in this class. While the bound is short and simple, its proof is not. We extend the bound to Hermitian matrices in the complex space and present a new self-contained and significantly shorter proof using novel geometric ideas. ###### keywords: iterative method; continuation method; preconditioning; preconditioner; eigenvalue; eigenvector; Rayleigh quotient; gradient iteration; convergence theory; spectral equivalence ###### AMS: 49M37 65F15 65K10 65N25 (Place for Digital Object Identifier, to get an idea of the final spacing.) ## 1 Introduction We consider a generalized eigenvalue problem (eigenproblem) for a linear pencil $B-\mu A$ with symmetric (Hermitian in the complex case) matrices $A$ and $B$ with positive definite $A$. The eigenvalues $\mu_{i}$ are enumerated in decreasing order $\mu_{1}\geq\dotsc\geq\mu_{\min}$ and the $x_{i}$ denote the corresponding eigenvectors. The largest value of the Rayleigh quotient $\mu(x)={(x,Bx)}/{(x,Ax)},$ where $(\cdot,\cdot)$ denotes the standard scalar product, is the largest eigenvalue $\mu_{1}$. It can be approximated iteratively by maximizing the Rayleigh quotient in the direction of its gradient, which is proportional to $(B-\mu(x)A)x$. Preconditioning is used to accelerate the convergence; see, e.g., [2, 4, 5, 6, 8] and the references therein. Here we consider the simplest preconditioned eigenvalue solver (eigensolver)—the gradient iterative method with an explicit formula for the step size, cf. [2], one step of which is described by (1) $x^{\prime}=x+\frac{1}{\mu(x)-\mu_{\min}}T(Bx-\mu(x)Ax),\quad\mu(x)=\frac{(x,Bx)}{(x,Ax)}.$ The symmetric (Hermitian in the complex case) positive definite matrix $T$ in (1) is called the _preconditioner_. Since $A$ and $T$ are both positive definite, we assume that (2) $(1-\gamma)(z,T^{-1}z)\leq(z,Az)\leq(1+\gamma)(z,T^{-1}z),\,\forall z,\text{ for a given }\gamma\in[0,1).$ The following result is proved in [8, 9, 10] for symmetric matrices in the real space. ###### Theorem 1. If $\mu_{i+1}<\mu(x)\leq\mu_{i}$ then $\mu(x^{\prime})\geq\mu(x)$ and (3) $\frac{\mu_{i}-\mu(x^{\prime})}{\mu(x^{\prime})-\mu_{i+1}}\leq\sigma^{2}\frac{\mu_{i}-\mu(x)}{\mu(x)-\mu_{i+1}},\quad\sigma=1-(1-\gamma)\frac{\mu_{i}-\mu_{i+1}}{\mu_{i}-\mu_{\min}}.$ The convergence factor $\sigma$ cannot be improved with the chosen terms and assumptions. Compared to other known non-asymptotic convergence rate bounds for similar preconditioned eigensolvers, e.g., [1, 2, 4, 5], the advantages of (3) are in its sharpness and elegance. Method (1) is the easiest preconditioned eigensolver, but (3) still remains the only known sharp bound in these terms for any of preconditioned eigensolvers. While bound (3) is short and simple, its proof in [8] is quite the opposite. It covers only the real case and is not self-contained—in addition it requires most of the material from [9, 10]. Here we extend the bound to Hermitian matrices and give a new much shorter and self-contained proof of Theorem 1, which is a great qualitative improvement compared to that of [8, 9, 10]. The new proof is not yet as elementary as we would like it to be; however, it is easy enough to hope that a similar approach might be applicable in future work on preconditioned eigensolvers. Our new proof is based on novel techniques combined with some old ideas of [3, 9, 10]. We demonstrate that, for a given initial eigenvector approximation $x$, the next iterative approximation $x^{\prime}$ described by (1) belongs to a cone if we apply any preconditioner satisfying (2). We analyze a corresponding continuation gradient method involving the gradient flow of the Rayleigh quotient and show that the smallest gradient norm (evidently leading to the slowest convergence) of the continuation method is reached when the initial vector belongs to a subspace spanned by two specific eigenvectors, namely $x_{i}$ and $x_{i+1}$. This is done by showing that Temple’s inequality, which provides a lower bound for the norm of the gradient $\nabla\mu(x)$, is sharp only in $\mathrm{span}\\{x_{i},x_{i+1}\\}$. Next, we extend by integration the result for the continuation gradient method to our actual fixed step gradient method to conclude that the point on the cone, which corresponds to the poorest convergence and thus gives the guaranteed convergence rate bound, belongs to the same two-dimensional invariant subspace $\mathrm{span}\\{x_{i},x_{i+1}\\}$. This reduces the convergence analysis to a two-dimensional case for shifted inverse iterations, where the sharp convergence rate bound is established. ## 2 The proof of Theorem 1 We start with several simplifications: ###### Theorem 2. We can assume that $\gamma>0$, $A=I$, $B>0$ is diagonal, eigenvalues are simple, $\mu(x)<\mu_{i}$, and $\mu(x^{\prime})<\mu_{i}$ in Theorem 1 without loss of generality. ###### Proof. First, we observe that method (1) and bound (3) are evidently both invariant with respect to a real shift $s$ if we replace the matrix $B$ with $B+sA$, so without loss of generality we need only consider the case $\mu_{\min}=0$ which makes $B\geq 0.$ Second, by changing the basis from coordinate vectors to the eigenvectors of $A^{-1}B$ we can make $B$ diagonal and $A=I$. Third, having $\mu(x^{\prime})\geq\mu(x)$ if $\mu(x)=\mu_{i}$ or $\mu(x^{\prime})\geq\mu_{i}$, or both, bound (3) becomes trivial. The assumption $\gamma>0$ is a bit more delicate. The vector $x^{\prime}$ depends continuously on the preconditioner $T$, so we can assume that $\gamma>0$ and extend the final bound to the case $\gamma=0$ by continuity. Finally, we again use continuity to explain why we can assume that all eigenvalues (in fact, we only need $\mu_{i}$ and $\mu_{i+1}$) are simple and make $\mu_{\min}>0$ and thus $B>0$ without changing anything. Let us list all $B$-dependent terms, in addition to all participating eigenvalues, in method (4): $\mu(x)$ and $x^{\prime}$; and in bound (3): $\mu(x)$ and $\mu(x^{\prime})$. All these terms depend on $B$ continuously if $B$ is slightly perturbed into $B_{\epsilon}$ with some $\epsilon\to 0$, so we increase arbitrarily small the diagonal entries of the matrix $B$ to make all eigenvalues of $B_{\epsilon}$ simple and $\mu_{\min}>0$. If we prove bound (3) for the matrix $B_{\epsilon}$ with simple positive eigenvalues, and show that the bound is sharp as $0<\mu_{\min}\to 0$ with $\epsilon\to 0$, we take the limit $\epsilon\to 0$ and by continuity extend the result to the limit matrix $B\geq 0$ with $\mu_{\min}=0$ and possibly multiple eigenvalues. ∎ It is convenient to rewrite (1)–(3) equivalently by Theorem 2 as follows111Here and below $\|\cdot\|$ denotes the Euclidean vector norm, i.e., $\|x\|^{2}=(x,x)=x^{H}x$ for a real or complex column-vector $x$, as well as the corresponding induced matrix norm. (4) $\displaystyle\mu(x)x^{\prime}=Bx-(I-T)(Bx-\mu(x)x),\quad\mu(x)=\frac{(x,Bx)}{(x,x)},$ (5) $\displaystyle\|I-T\|\leq\gamma,\quad 0<\gamma<1;$ and if $\mu_{i+1}<\mu(x)<\mu_{i}$ and $\mu(x^{\prime})<\mu_{i}$ then $\mu(x^{\prime})\geq\mu(x)$ and (6) $\frac{\mu_{i}-\mu(x^{\prime})}{\mu(x^{\prime})-\mu_{i+1}}\leq\sigma^{2}\frac{\mu_{i}-\mu(x)}{\mu(x)-\mu_{i+1}},\quad\sigma=1-(1-\gamma)\frac{\mu_{i}-\mu_{i+1}}{\mu_{i}}=\gamma+(1-\gamma)\frac{\mu_{i+1}}{\mu_{i}}.$ Now we establish the validity and sharpness of bound (6) assuming (4) and (5). ###### Theorem 3. Let us define222We define angles in $[0,\pi/2]$ between vectors by $\cos\angle\\{x,y\\}=|(x,y)|/(\|x\|\|y\|)$. $\phi_{\gamma}(x)=\arcsin\left(\gamma{\|Bx-\mu(x)x\|}/{\|Bx\|}\right),$ then $\phi_{\gamma}(x)<\pi/2$ and $\angle\\{x^{\prime},Bx\\}\leq\phi_{\gamma}(x)$. Let $w\neq 0$ be defined as the vector constrained by $\angle\\{w,Bx\\}\leq\phi_{\gamma}(x)$ and with the smallest value $\mu(w)$. Then $\mu(x^{\prime})\geq\mu(w)>\mu(x)$. ###### Proof. Orthogonality $(x,Bx-\mu(x)x)=0$ by the Pythagorean theorem implies $\|Bx\|^{2}=\|\mu(x)x\|^{2}+\|Bx-\mu(x)x\|^{2}$, so $\|Bx-\mu(x)x\|<\|Bx\|$, since $\mu(x)>0$ as $B>0$, and $\sin\angle\\{x,Bx\\}=\sin\phi_{1}(x)={\|Bx-\mu(x)x\|}/{\|Bx\|}<1,$ where $Bx\neq 0$ as $B>0$. A ball with the radius $\gamma\|Bx-\mu(x)x\|\geq\|I-T\|\|Bx-\mu(x)x\|$ by (5) centered at $Bx$ contains $\mu(x)x^{\prime}$ by (4), so $\sin\angle\\{x^{\prime},Bx\\}\leq\gamma\|Bx-\mu(x)x\|/\|Bx\|<\gamma<1$. The statement $\mu(x^{\prime})\geq\mu(w)$ follows directly from the definition of $w$. Now, $0<\frac{(x,Bx)}{\|x\|\|Bx\|}=\cos\phi_{1}(x)<\cos\angle\\{w,Bx\\}=\frac{|(w,Bx)|}{\|w\|\|Bx\|}\leq\frac{(w,Bw)^{1/2}(x,Bx)^{1/2}}{\|w\|\|Bx\|}$ as $B>0$, so $\sqrt{\mu(x)}<\sqrt{\mu(w)}$ and $\mu(x)<\mu(w).$ ∎ We denote by $C_{\phi_{\gamma}(x)}(Bx):=\\{y:\angle\\{y,Bx\\}\leq\phi_{\gamma}(x)\\}$ the circular cone around $Bx$ with the opening angle $\phi_{\gamma}(x)$. Theorem 3 replaces $x^{\prime}$ with the minimizer $w$ of the Rayleigh quotient on the cone $C_{\phi_{\gamma}(x)}(Bx)$ in the rest of the paper, except at the end of the proof of Theorem 8, where we show that bounding below the value $\mu(w)$ instead of $\mu(x^{\prime})$ still gives the sharp estimate. Later on, in the proof of Theorem 5, we use an argument that holds easily only in the real space, so we need the following last simplification. ###### Theorem 4. Without loss of generality we can consider only the real case. ###### Proof. The key observation is that for our positive diagonal matrix $B$ the Rayleigh quotient depends evidently only on the absolute values of the vector components, i.e., $\mu(x)=\mu(|x|)$, where the absolute value operation is applied component-wise. Moreover, $\|Bx-\mu(x)x\|=\|B|x|-\mu(|x|)|x|\|$ and $\|Bx\|=\|B|x|\|$, so $\phi_{\gamma}(x)=\phi_{\gamma}(|x|)$. The cone $C_{\phi_{\gamma}(x)}(Bx)$ lives in the complex space, but we also need its substitute in the real space. Let us introduce the notation $C^{R}_{\phi_{\gamma}(|x|)}(B|x|)$ for the real circular cone with the opening angle $\phi_{\gamma}(|x|)$ centered at the real vector $B|x|.$ Next we show that in the real space we have the inclusion $\left|C_{\phi_{\gamma}(x)}(Bx)\right|\subseteq C^{R}_{\phi_{\gamma}(|x|)}(B|x|)$. For any complex nonzero vectors $x$ and $y$, we have $|(y,Bx)|\leq(|y|,B|x|)$ by the triangle inequality, thus $\angle\\{|y|,B|x|\\}\leq\angle\\{y,Bx\\}$. If $y\in C_{\phi_{\gamma}(x)}(Bx)$ then $\angle\\{|y|,B|x|\\}\leq\angle\\{y,Bx\\}\leq\phi_{\gamma}(x)=\phi_{\gamma}(|x|)$, i.e., indeed, $|y|\in C^{R}_{\phi_{\gamma}(|x|)}(B|x|)$, which means that $\left|C_{\phi_{\gamma}(x)}(Bx)\right|\subseteq C^{R}_{\phi_{\gamma}(|x|)}(B|x|)$ as required. Therefore, changing the given vector $x$ to take its absolute value $|x|$ and replacing the complex cone $C_{\phi_{\gamma}(x)}(Bx)$ with the real cone $C^{R}_{\phi_{\gamma}(|x|)}(B|x|)$ lead to the relations $\min_{y\in C_{\phi_{\gamma}(x)}(Bx)}\mu(y)=\min_{|y|\in\left|C_{\phi_{\gamma}(x)}(Bx)\right|}\mu(|y|)\geq\min_{|y|\in C^{R}_{\phi_{\gamma}(|x|)}(B|x|)}\mu(|y|)$, but does not affect the starting Rayleigh quotient $\mu(x)=\mu(|x|).$ This proves the theorem with the exception of the issue of whether the sharpness in the real case implies the sharpness in the complex case; see the end of the proof of Theorem 8. ∎ ###### Theorem 5. We have $w\in\partial C_{\phi_{\gamma}(x)}(Bx)$ and $\exists\,\alpha=\alpha_{\gamma}(x)>-\mu_{i}$ such that $(B+\alpha I)w=Bx$. The inclusion $x\in\mathrm{span}\\{x_{i},x_{i+1}\\}$ implies $w\in\mathrm{span}\\{x_{i},x_{i+1}\\}.$ ###### Proof. Fig. 1: The cone $C_{\phi_{\gamma}(x)}(Bx)$. Assuming that $w$ is strictly inside the cone $C_{\phi_{\gamma}(x)}(Bx)$ implies that $w$ is a point of a local minimum of the Rayleigh quotient. The Rayleigh quotient has only one local (and global) minimum, $\mu_{\min}$, but the possibility $\mu(w)=\mu_{\min}$ is eliminated by Theorem 3, so we obtain a contradiction, thus $w\in\partial C_{\phi_{\gamma}(x)}(Bx)$. The necessary condition for a local minimum of a smooth real-valued function on a smooth surface in a real vector space is that the gradient of the function is orthogonal to the surface at the point of the minimum and directed inwards. In our case, $C_{\phi_{\gamma}(x)}(Bx)$ is a circular cone with the axis $Bx$ and the gradient $\nabla\mu(w)$ is positively proportional to $Bw-\mu(w)w$; see Figure 1. We first scale the vector $w$ such that $(Bx-w,w)=0$ so that the vector $Bx-w$ is an inward normal vector for $\partial C_{\phi_{\gamma}(x)}(Bx)$ at the point $w$. This inward normal vector must be positively proportional to the gradient, $\beta(Bx-w)=Bw-\mu(w)w$ with $\beta>0$, which gives $(B+\alpha I)w=\beta Bx$, where $\alpha=\beta-\mu(w)>-\mu(w)>-\mu_{i}$. Here $\beta\neq 0$ as otherwise $w$ would be an eigenvector, but $\mu(x)<\mu(w)<\mu(x^{\prime})$ by Theorem 3, where by assumptions $\mu_{i+1}<\mu(x)$, while $\mu(x^{\prime})<\mu_{i}$ by Theorem 2, which gives a contradiction. As the scaling of the minimizer is irrelevant, we denote $w/\beta$ here by $w$ with a slight local notation abuse. Finally, since $(B+\alpha I)w=Bx$, inclusion $x\in\mathrm{span}\\{x_{i},x_{i+1}\\}$ gives either the required inclusion $w\in\mathrm{span}\\{x_{i},x_{i+1}\\}$ or $w\in\mathrm{span}\\{x_{i},x_{i+1},x_{j}\\}$ with $\alpha=-\mu_{j}$ for some $j\neq i$ and $j\neq i+1.$ We now show that the latter leads to a contradiction. We have just proved that $\alpha>-\mu_{i}$, thus $j>i+1$. Let $x=c_{i}x_{i}+c_{i+1}x_{i+1}$, where we notice that $c_{i}\neq 0$ and $c_{i+1}\neq 0$ since $x$ is not an eigenvector. Then we obtain $w=a_{i}c_{i}x_{i}+a_{i+1}c_{i+1}x_{i+1}+c_{j}x_{j}$ where $(B-\mu_{j})w=Bx$, therefore $a_{k}=\mu_{k}/(\mu_{k}-\mu_{j}),\,k=i,i+1$. Since all eigenvalues are simple, $\mu_{i+1}\neq\mu_{j}.$ We observe that $0<a_{i}<a_{i+1}$, i.e., in the mapping of $x$ to $w$ the coefficient in front of $x_{i}$ changes by a smaller absolute value compared to the change in the coefficient in front of $x_{i+1}$. Thus, $\mu(x)>\mu(a_{i}c_{i}x_{i}+a_{i+1}c_{i+1}x_{i+1})\geq\mu(w)$ using the monotonicity of the Rayleigh quotient in the absolute values of the coefficients of the eigenvector expansion of its argument, which contradicts $\mu(w)>\mu(x)$ proved in Theorem 3. ∎ Theorem 5 characterizes the minimizer $w$ of the Rayleigh quotient on the cone $C_{\phi_{\gamma}(x)}(Bx)$ for a fixed $x$. The next goal is to vary $x$, preserving its Rayleigh quotient $\mu(x)$, and to determine conditions on $x$ leading to the smallest $\mu(w)$ in such a setting. Intuition suggests (and we give the exact formulation and the proof later in Theorem 7) that the poorest convergence of a gradient method corresponds to the smallest norm of the gradient, so in the next theorem we analyze the behavior of the gradient $\|\nabla\mu(x)\|$ of the Rayleigh quotient and the cone opening angle $\phi_{\gamma}(x).$ ###### Theorem 6. Let $\kappa\in(\mu_{i+1},\mu_{i})$ be fixed and the level set of the Rayleigh quotient be denoted by ${\mathcal{L}}(\kappa):=\\{x\neq 0:\mu(x)=\kappa\\}$ . Both $\|\nabla\mu(x)\|\|x\|$ and $\phi_{1}(x)-\phi_{\gamma}(x)$ with $0<\gamma<1$ attain their minima on $x\in{\mathcal{L}}(\kappa)$ in $\mathrm{span}\\{x_{i},x_{i+1}\\}.$ ###### Proof. By definition of the gradient, $\|\nabla\mu(x)\|\|x\|=2\|Bx-\kappa x\|/\|x\|$ for $x\in{\mathcal{L}}(\kappa)$. The Temple inequality $\|Bx-\kappa x\|^{2}/\|x\|^{2}\geq(\mu_{i}-\kappa)(\kappa-\mu_{i+1})$ is equivalent to the operator inequality $(B-\mu_{i}I)(B-\mu_{i+1}I)\geq 0$, which evidently holds. The equality here is attained only for $x\in\mathrm{span}\\{x_{i},x_{i+1}\\}.$ Finally, we turn our attention to the angles. For $x\in{\mathcal{L}}(\kappa),$ the Pythagorean theorem $\|Bx\|^{2}=\|\kappa x\|^{2}+\|Bx-\kappa x\|^{2}$ shows that $a^{2}:=\frac{\|Bx-\kappa x\|^{2}}{\|Bx\|^{2}}=\frac{\|Bx-\kappa x\|^{2}/\|x\|^{2}}{\kappa^{2}+\|Bx-\kappa x\|^{2}/\|x\|^{2}}\in(0,1)$ is minimized together with $\|Bx-\kappa x\|/\|x\|$. But for a fixed $\gamma\in(0,1)$ the function $\arcsin(a)-\arcsin(\gamma a)$ is strictly increasing in $a\in(0,1)$ which proves the proposition for $\phi_{1}(x)-\phi_{\gamma}(x)=\arcsin(a)-\arcsin(\gamma a).$ ∎ Now we are ready to show that the same subspace $\mathrm{span}\\{x_{i},x_{i+1}\\}$ gives the smallest change in the Rayleigh quotient $\mu(w)-\kappa$. The proof is based on analyzing the negative normalized gradient flow of the Rayleigh quotient. ###### Theorem 7. Under the assumptions of Theorems 5 and 6 we denote $I_{\gamma}(\kappa):=\\{w:w\in\arg\min\mu(C_{\phi_{\gamma}(x)}(Bx));\;x\in{\mathcal{L}}(\kappa)\\}$—the set of minimizers of the Rayleigh quotient. Then $\arg\min\mu(I_{\gamma}(\kappa))\in\mathrm{span}\\{x_{i},x_{i+1}\\}.$ (See Figure 2). Fig. 2: The Rayleigh quotient gradient flow integration on the unit ball. ###### Proof. The initial value problem for a gradient flow of the Rayleigh quotient, (7) $y^{\prime}(t)=-\frac{\nabla\mu(y(t))}{\|\nabla\mu(y(t))\|},\,t\geq 0,\,y(0)=w\in I_{\gamma}(\kappa),$ has the vector-valued solution $y(t),$ which preserves the norm of the initial vector $w$ since $d\|y(t)\|^{2}/dt=2(y(t),y^{\prime}(t))=0$ as $(y,\nabla\mu(y))=0$. Without loss of generality we assume $\|w\|=1=\|y(t)\|.$ The Rayleigh quotient function $\mu(y(t))$ is decreasing since $\frac{d}{dt}\mu(y(t))=\left(\nabla\mu(y(t)),y^{\prime}(t)\right)=\left(\nabla\mu(y(t)),-\frac{\nabla\mu(y(t))}{\|\nabla\mu(y(t))\|}\right)=-\|\nabla\mu(y(t))\|\leq 0.$ As $\mu(y(0))=\mu(w)<\mu_{i}$, the function $\mu(y(t))$ is strictly decreasing at least until it reaches $\kappa>\mu_{i+1}$ as there are no eigenvalues in the interval $[\kappa,\mu(y(0))]\subset(\mu_{i+1},\mu_{i}),$ but only eigenvectors can be special points of ODE (7). The condition $\mu(y(\bar{t}))=\kappa$ thus uniquely determines $\bar{t}$ for a given initial value $w$. The absolute value of the decrease of the Rayleigh quotient along the path $L:=\\{y(t),\,0\leq t\leq\bar{t}\\}$ is $\mu(w)-\kappa=\mu(y(0))-\mu(y(\bar{t}))=\int_{0}^{\bar{t}}\|\nabla\mu(y(t))\|dt>0.$ Our continuation method (7) using the _normalized_ gradient flow is nonstandard, but its advantage is that it gives the following simple expression for the length of $L$, $\mathrm{Length}(L)=\int_{0}^{\bar{t}}\|y^{\prime}(t)\|dt=\int_{0}^{\bar{t}}1dt=\bar{t}.$ Since the initial value $w$ is determined by $x$, we compare a generic $x$ with the special choice $x=x^{*}\in\mathrm{span}\\{x_{i},x_{i+1}\\}$, using the superscript $*$ to denote all quantities corresponding to the choice $x=x^{*}$. By Theorem 5 $x^{*}\in\mathrm{span}\\{x_{i},x_{i+1}\\}$ implies $w^{*}\in\mathrm{span}\\{x_{i},x_{i+1}\\}$, so we have $y^{*}(t)\in\mathrm{span}\\{x_{i},x_{i+1}\\},$ $0\leq t\leq\bar{t}^{*}$ as $\mathrm{span}\\{x_{i},x_{i+1}\\}$ is an invariant subspace for the gradient of the Rayleigh quotient. At the end points, $\mu(y(\bar{t}))=\kappa=\mu(x)=\mu(y^{*}(\bar{t}^{*})),$ by their definition. Our goal is to bound the initial value $\mu(w^{*})=\mu(y^{*}(0))$ by $\mu(w)=\mu(y(0))$, so we compare the lengths of the corresponding paths $L^{*}$ and $L$ and the norms of the gradients along these paths. We start with the lengths. We obtain $\phi_{1}(x^{*})-\phi_{\gamma}(x^{*})\leq\phi_{1}(x)-\phi_{\gamma}(x)$ by Theorem 6. Here the angle $\phi_{1}(x)-\phi_{\gamma}(x)$ is the smallest angle between any two vectors on the cones boundaries $\partial C_{\phi_{\gamma}(x)}(Bx)$ and $\partial C_{\phi_{1}(x)}(Bx)$. Thus, $\phi_{1}(x)-\phi_{\gamma}(x)\leq\angle\\{y(0),y(\bar{t})\\}$ as our one vector $y(0)=w\in\partial C_{\phi_{\gamma}(x)}(Bx)$ by Theorem 5, while the other vector $y(\bar{t})$ cannot be inside the cone $C_{\phi_{\gamma}(x)}(Bx)$ since $\mu(w)>\kappa=\mu(y(\bar{t}))$ by Theorem 3. As $y(t)$ is a unit vector, $\angle\\{y(0),y(\bar{t})\\}\leq\mathrm{Length}(L)=\bar{t}$ as the angle is the length of the arc—the shortest curve from $y(0)$ to $y(\bar{t})$ on the unit ball. For our special $*$-choice, inequalities from the previous paragraph turn into equalities, as $y^{*}(t)$ is in the intersection of the unit ball and the subspace $\mathrm{span}\\{x_{i},x_{i+1}\\}$, so the path $L^{*}$ is the arc between $y^{*}(0)$ to $y^{*}(\bar{t}^{*})$ itself. Combining everything together, $\displaystyle\bar{t}^{*}=\mathrm{Length}(L^{*})=\angle\\{y^{*}(0),y^{*}(\bar{t}^{*})\\}=\angle\\{w^{*},x^{*}\\}=\varphi_{1}(x^{*})-\varphi_{\gamma}(x^{*})$ $\displaystyle\leq\varphi_{1}(x)-\varphi_{\gamma}(x)\leq\angle\\{y(0),y(\bar{t})\\}\leq\mathrm{Length}(L)=\bar{t}.$ By Theorem 6 on the norms of the gradient, $-\|\nabla\mu(y^{*}(t^{*}))\|\geq-\|\nabla\mu(y(t))\|$ for each pair of independent variables $t^{*}$ and $t$ such that $\mu(y^{*}(t^{*}))=\mu(y(t)).$ Using Theorem 9, we conclude that $\mu(w^{*})=\mu(y^{*}(0))\leq\mu(y(\bar{t}-\bar{t}^{*}))\leq\mu(y(0))=\mu(w)$ as $\bar{t}-\bar{t}^{*}\geq 0$, i.e., the subspace $\mathrm{span}\\{x_{i},x_{i+1}\\}$ gives the smallest value $\mu(w).$ ∎ By Theorem 7 the poorest convergence is attained with $x\in\mathrm{span}\\{x_{i},x_{i+1}\\}$ and with the corresponding minimizer $w\in\mathrm{span}\\{x_{i},x_{i+1}\\}$ described in Theorem 5, so finally our analysis is now reduced to the two-dimensional space $\mathrm{span}\\{x_{i},x_{i+1}\\}$. ###### Theorem 8. Bound (6) holds and is sharp for $x\in\mathrm{span}\\{x_{i},x_{i+1}\\}$. ###### Proof. Assuming $\|x\|=1$ and $\|x_{i}\|=\|x_{i+1}\|=1$, we derive (8) $|(x,x_{i})|^{2}=\frac{\mu(x)-\mu_{i+1}}{\mu_{i}-\mu_{i+1}}>0\mbox{ and }|(x,x_{i+1})|^{2}=\frac{\mu_{i}-\mu(x)}{\mu_{i}-\mu_{i+1}},$ and similarly for $w\in\mathrm{span}\\{x_{i},x_{i+1}\\}$ where $(B+\alpha I)w=Bx$. Since $B>0$, we have $x=(I+\alpha B^{-1})w$. Assuming $\alpha=-\mu_{i+1}$, this identity implies $x=x_{i},$ which contradicts our assumption that $x$ is not an eigenvector. For $\alpha\neq-\mu_{i+1}$ and $\alpha>-\mu_{i}$ by Theorem 5, the inverse $(B+\alpha I)^{-1}$ exists. Next we prove that $\alpha>0$ and that it is a strictly decreasing function of $\kappa:=\mu(x)\in(\mu_{i+1},\mu_{i}).$ Indeed, using $Bx=(B+\alpha I)w$ and our cosine-based definition of the angles, we have $0<(w,(B+\alpha I)w)^{2}=(w,Bx)^{2}=\|w\|^{2}\|Bx\|^{2}\cos^{2}\phi_{\gamma}(x),$ where $\|Bx\|^{2}\cos^{2}\phi_{\gamma}(x)=\|Bx\|^{2}-\gamma^{2}\|Bx-\kappa x\|^{2}.$ We substitute $w=(B+\alpha I)^{-1}Bx$, which gives $((B+\alpha I)^{-1}Bx,Bx)^{2}=\|(B+\alpha I)^{-1}Bx\|^{2}\left(\|Bx\|^{2}-\gamma^{2}\|Bx-\kappa x\|^{2}\right).$ Using (8), multiplication by $(\mu_{i}+\alpha)^{2}(\mu_{i+1}+\alpha)^{2}$ leads to a simple quadratic equation, $a\alpha^{2}+b\alpha+c=0,\,a=\gamma^{2}(\kappa(\mu_{i}+\mu_{i+1})-\mu_{i}\mu_{i+1}),\,b=2\gamma^{2}\kappa\mu_{i}\mu_{i+1},\,c=-(1-\gamma^{2})\mu_{i}^{2}\mu_{i+1}^{2}$ for $\alpha.$ As $a>0$, $b>0$, and $c<0$, the discriminant is positive and the two solutions for $\alpha$, corresponding to the minimum and maximum of the Rayleigh quotient on $C_{\phi_{\gamma}(x)}(Bx)$, have different signs. The proof of Theorem 5 analyzes the direction of the gradient of the Rayleigh quotient to conclude that $\beta>0$ and $\alpha>-\mu(w)$ correspond to the minimum. Repeating the same arguments with $\beta<0$ shows that $\alpha<-\mu(w)$ corresponds to the maximum. But $\mu(w)>0$ since $B>0$, hence the negative $\alpha$ corresponds to the maximum and thus the positive $\alpha$ corresponds to the minimum. We observe that the coefficients $a>0$ and $b>0$ are evidently increasing functions of $\kappa\in(\mu_{i+1},\mu_{i})$, while $c<0$ does not depend on $\kappa$. Thus $\alpha>0$ is strictly decreasing in $\kappa$, and taking $\kappa\to\mu_{i}$ gives the smallest $\alpha=\mu_{i+1}(1-\gamma)/\gamma>0.$ Since $(B+\alpha I)w=Bx$ where now $\alpha>0$, condition $(x,x_{i})\neq 0$ implies $(w,x_{i})\neq 0$ and $(x,x_{i+1})=0$ implies $(w,x_{i+1})=0,$ so we introduce the convergence factor as $\sigma^{2}(\alpha):=\frac{\mu_{i}-\mu(w)}{\mu(w)-\mu_{i+1}}\frac{\mu(x)-\mu_{i+1}}{\mu_{i}-\mu(x)}=\left|\frac{(w,x_{i+1})}{(w,x_{i})}\right|^{2}\left|\frac{(x,x_{i})}{(x,x_{i+1})}\right|^{2}=\left(\frac{\mu_{i+1}}{\mu_{i}}\frac{\mu_{i}+\alpha}{\mu_{i+1}+\alpha}\right)^{2},$ where we use (8) and again $(B+\alpha I)w=Bx$. We notice that $\sigma(\alpha)$ is a strictly decreasing function of $\alpha>0$ and thus takes its largest value for $\alpha=\mu_{i+1}(1-\gamma)/\gamma$ giving $\sigma=\gamma+(1-\gamma){\mu_{i+1}}/{\mu_{i}},$ i.e., bound (6) that we are seeking. The convergence factor $\sigma^{2}(\alpha)$ cannot be improved without introducing extra terms or assumptions. But $\sigma^{2}(\alpha)$ deals with $w\in C_{\phi_{\gamma}(x)}(Bx)$, not with the actual iterate $x^{\prime}$. We now show that for $\kappa\in(\mu_{i+1},\mu_{i})$ there exist a vector $x\in\mathrm{span}\\{x_{i},x_{i+1}\\}$ and a preconditioner $T$ satisfying (5) such that $\kappa=\mu(x)$ and $x^{\prime}\in\mathrm{span}\\{w\\}$ in both real and complex cases. In the complex case, let us choose $x$ such that $\mu(x)=\kappa$ and $x=|x|$ according to (8), then the real vector $w=|w|\in C_{\phi_{\gamma}(x)}(Bx)$ is a minimizer of the Rayleigh quotient on $C_{\phi_{\gamma}(x)}(Bx)$, since $\mu(w)=\mu(|w|)$ and $|(w,B|x|)|\leq(|w|,B|x|)$. Finally, for a real $x$ with $\mu(x)=\kappa$ and a real properly scaled $y\in C_{\phi_{\gamma}(x)}(Bx)$ there is a real matrix $T$ satisfying (5) such that $y=Bx-(I-T)(Bx-\kappa x)$, which leads to (4) with $\mu(x)x^{\prime}=y.$ Indeed, for the chosen $x$ we scale $y\in C_{\phi_{\gamma}(x)}(Bx)$ such that $(y,Bx-y)=0$ so $\|Bx-y\|=\sin\phi_{\gamma}(x)\|Bx\|=\gamma\|Bx-\kappa x\|$. As vectors $Bx-y$ and $\gamma(Bx-\kappa x)$ are real and have the same length there exists a _real_ Householder reflection $H$ such that $Bx-y=H\gamma(Bx-\kappa x)$. Setting $T=I-\gamma H$ we obtain the required identity. Any Householder reflection is symmetric and has only two distinct eigenvalues $\pm 1$, so we conclude that $T$ is real symmetric (and thus Hermitian in the complex case) and satisfies (5). ∎ ## 3 Appendix The integration of inverse functions theorem follows. ###### Theorem 9. Let $f,\>g:[0,b]\to{\bf R}$ for $b>0$ be strictly monotone increasing smooth functions and suppose that for $a\in[0,b]$ we have $f(a)=g(b)$. If for all $\alpha,\>\beta\in[0,b]$ with $f(\alpha)=g(\beta)$ the derivatives satisfy $f^{\prime}(\alpha)\leq g^{\prime}(\beta),$ then for any $\xi\in[0,a]$ we have $f(a-\xi)\geq g(b-\xi).$ ###### Proof. For any $\xi\in[0,a]$ we have (using $f(a)=g(b)$) $\xi=\int_{g(b-\xi)}^{g(b)}\left(g^{-1}\right)^{\prime}\\!(y)\>dy=\int_{f(a-\xi)}^{g(b)}\left(f^{-1}\right)^{\prime}\\!(y)\>dy.$ If $y=f(\alpha)=g(\beta)$, then for the derivatives of the inverse functions it holds that $\left(g^{-1}\right)^{\prime}\\!(y)\leq\left(f^{-1}\right)^{\prime}\\!(y).$ Since $f$ and $g$ are strictly monotone increasing functions the integrands are positive functions and $g(b-\xi)<g(b)$ as well as $f(a-\xi)<f(a)=g(b)$. Comparing the lower limits of the integrals gives the statement of the theorem. ∎ ## Conclusions We present a new geometric approach to the convergence analysis of a preconditioned fixed-step gradient eigensolver which reduces the derivation of the convergence rate bound to a two-dimensional case. The main novelty is in the use of a continuation method for the gradient flow of the Rayleigh quotient to locate the two-dimensional subspace corresponding to the smallest change in the Rayleigh quotient and thus to the slowest convergence of the gradient eigensolver. An elegant and important result such as Theorem 1 should ideally have a textbook-level proof. We have been trying, unsuccessfully, to find such a proof for several years, so its existence remains an open problem. ## Acknowledgments We thank M. Zhou of University of Rostock, Germany for proofreading. M. Argentati of University of Colorado Denver, E. Ovtchinnikov of University of Westminster, and anonymous referees have made numerous great suggestions to improve the paper and for future work. ## REFERENCES * [1] J. H. Bramble, J. E. Pasciak, and A. V. Knyazev, A subspace preconditioning algorithm for eigenvector/eigenvalue computation, Adv. Comput. Math., 6 (1996), pp. 159–189. * [2] E. G. D’yakonov, _Optimization in solving elliptic problems_ , CRC Press, 1996. * [3] A. V. Knyazev, Computation of eigenvalues and eigenvectors for mesh problems: algorithms and error estimates, (In Russian), Dept. Num. Math., USSR Ac. Sci., Moscow, 1986. * [4] A. V. Knyazev, Convergence rate estimates for iterative methods for a mesh symmetric eigenvalue problem, Russian J. Numer. Anal. Math. Modelling, 2 (1987), pp. 371–396. * [5] A. V. Knyazev, Preconditioned eigensolvers—an oxymoron?, Electron. Trans. Numer. Anal., 7 (1998), pp. 104–123. * [6] A. V. Knyazev, Preconditioned eigensolvers: practical algorithms, In Z. Bai, J. Demmel, J. Dongarra, A. Ruhe, and H. van der Vorst, editors, Templates for the Solution of Algebraic Eigenvalue Problems: A Practical Guide, pp. 352–368. SIAM, Philadelphia, 2000. * [7] A. V. Knyazev, Toward the optimal preconditioned eigensolver: Locally optimal block preconditioned conjugate gradient method, SIAM J. Sci. Comput., 23 (2001), pp. 517–541. * [8] A. V. Knyazev and K. Neymeyr, A geometric theory for preconditioned inverse iteration. III: A short and sharp convergence estimate for generalized eigenvalue problems, Linear Algebra Appl., 358 (2003), pp. 95–114. * [9] K. Neymeyr, A geometric theory for preconditioned inverse iteration. I: Extrema of the Rayleigh quotient, Linear Algebra Appl., 322 (2001), pp. 61–85. * [10] K. Neymeyr, A geometric theory for preconditioned inverse iteration. II: Convergence estimates, Linear Algebra Appl., 322 (2001), pp. 87–104.

arxiv-papers

2024-09-04T02:48:53.483794

0801.4603

# The $Y(2175)$ State in the QCD Sum Rule Hua-Xing Chen1,2 hxchen@rcnp.osaka-u.ac.jp Xiang Liu1 xiangliu@pku.edu.cn Atsushi Hosaka2 hosaka@rcnp.osaka-u.ac.jp Shi-Lin Zhu1 zhusl@phy.pku.edu.cn 1Department of Physics, Peking University, Beijing 100871, China 2Research Center for Nuclear Physics, Osaka University, Ibaraki 567–0047, Japan ###### Abstract We study the mass of the state $Y(2175)$ of $J^{PC}=1^{--}$ in the QCD sum rule. We construct both the diquark-antidiquark currents $(ss)(\bar{s}\bar{s})$ and the meson-meson currents $(\bar{s}s)(\bar{s}s)$. We find that there are two independent currents for both cases, and derive the relations between them. The OPE convergence of these two currents is sufficiently fast, which enables us to perform good sum rule analysis. Both the SVZ sum rule and the finite energy sum rule lead to a mass around $2.3\pm 0.4$ GeV, which is consistent with the observed mass within the uncertainties of the present QCD sum rule. The coupling of the four-quark currents to lower lying states such as $\phi(1020)$ turns out to be rather small. We also discuss possible decay properties of $Y(2175)$ if it is a tetraquark state. tetraquark, QCD sum rule, vector meson ###### pacs: 12.39.Mk, 12.38.Lg, 12.40.Yx ## I Introduction The theory of the strong interactions, Quantum Chromodynamics (QCD), originated from the systematics of hadron spectroscopy. The spectroscopy contains meson and baryon states, many of which are well classified by the quark model with the quark content $q\bar{q}$ and $qqq$. Besides the quark model, QCD allows much richer hadron spectrum such as multiquark states, hadron molecules, hybrid states, glueballs etc. However the spectrum of QCD seems to saturate at $q\bar{q}$ and $qqq$. Since 2003, there have been important developments in hadron spectroscopy, which is triggered by the observation of the pentaquark $\Theta^{+}$. After three years of intense study, the status of $\Theta^{+}$ is still controversial. However, we have the charm-strange mesons $D_{sJ}(2317)$, $D_{sJ}(2460)$ 2317 ; 2460 ; the charmonium state $X(3872)$ 3872 , $Y(4260)$ 4260 , and many $X$s and $Y$s, whose properties seem difficult to be explained by the conventional picture of $q\bar{q}$. Recently Babar Collaboration observed a resonance $Y(2175)$ near the threshold in the process $e^{+}e^{-}\rightarrow\phi f_{0}(980)$ via initial-state radiation Aubert:2006bu ; Aubert:2007ur ; Aubert:2007ym . It has the quantum numbers $J^{PC}=1^{--}$. The Breit-Wigner mass is $M=2.175\pm 0.010\pm 0.015$ GeV, and width is $\Gamma=0.058\pm 0.016\pm 0.020$ GeV. It has been also confirmed by BES collaboration in the process $J/\psi\rightarrow\eta\phi f_{0}(980)$. A fit with a Breit-Wigner function gives the peak mass and width of $M=2.186\pm 0.010\pm 0.006$ GeV and $\Gamma=0.065\pm 0.023\pm 0.017$ GeV :2007yt . There are many suggestions to interpret this resonance. Ding and Yan interpreted it as a strangeonium hybrid and studied its decay properties in the flux-tube model and the constituent gluon model. Furthermore, for testing $s\bar{s}g$ scenario, they suggested searching decay modes such as $Y(2175)\to K_{1}(1400)K\to\pi K^{*}(892)K$, $Y(2175)\to K_{1}(1270)K\to\rho KK$ and $Y(2175)\to K_{1}(1270)K\to\pi K_{0}^{*}(1430)$ Ding:2006ya . In Ref. Ding-Yan , the authors explored $Y(2175)$ as a $2^{3}D_{1}$ $s\bar{s}$ meson, and calculated its decay modes by using both the ${}^{3}P_{0}$ model and the flux- tube model. They suggested experimental search of the decay modes $KK$, $K^{*}K^{*}$, $K(1460)K$ and $h_{1}(1380)\eta$. The characteristic decay modes of $Y(2175)$ as either a hybrid state or an $s\bar{s}$ state are quite different, which may be used to distinguish the hybrid and $s\bar{s}$ schemes. Wang studied $Y(2175)$ as a tetraquark state $ss\bar{s}\bar{s}$ by using QCD sum rule and suggested that there may be some tetraquark components in the state $Y(2175)$ Wang:2006ri . In a recent article Zhu:2007wz , Zhu reviewed $Y(2175)$ and indicated that the possibility of $Y(2175)$ arising from S-wave threshold effects can not be excluded. Napsuciale, Oset, Sasaki and Vaquera- Araujo studied the reaction $e^{+}e^{-}\to\phi\pi\pi$ for pions in an isoscalar S-wave channel which is dominated by the loop mechanism. By selecting the $\phi f_{0}(980)$ contribution as a function of the $e^{+}e^{-}$ energy, they also reproduced the experimental data except for the narrow peak Napsuciale:2007wp . Bystritskiy, Volkov, Kuraev, Bartos and Secansky calculated the total probability and the differential cross section of the process $e^{+}e^{-}\to\phi f_{0}(980)$ by using the local NJL model Bystritskiy:2007wq . Anikin, Pire and Teryaev studied the reaction $\gamma^{*}\gamma\to\rho\rho$, and calculated the mass of the isotensor exotic meson Anikin:2005ur . In Ref. Guo:2007uz , the authors performed a QCD sum rule study for $1^{--}$ hybrid meson, and the mass is predicted to be $2.3-2.4$, $2.3-2.5$, and $2.5-2.6$ GeV for $q\bar{q}g$, $q\bar{s}g$, and $s\bar{s}g$, respectively. In this work, we revisit the possibility of $Y(2175)$ as an tetraquark state $ss\bar{s}\bar{s}$. With the approach developed in our previous work Chen:2007zzg , we construct the general tetraquark interpolating currents with the quantum numbers $J^{PC}=1^{--}$. We find that there are two independent currents. They can have a structure of diquark-antidiquark $(ss)(\bar{s}\bar{s})$, or have a structure of meson-meson $(\bar{s}s)(\bar{s}s)$. We show that they are equivalent, and derive the relations between them. Then by using these two independent currents, we also perform a QCD sum rule analysis. We calculate the OPE up to the dimension 12, which contains the $\langle\bar{q}q\rangle^{4}$ condensates. In these two respects, our study differs from the previous one of Ref. Wang:2006ri . This paper is organized as follows. In Sec. II, we construct the tetraquark currents using both diquark ($qq$) and antidiquark ($\bar{q}\bar{q}$) fields, as well as quark-antiquark ($\bar{q}q$) pairs. In Sec. III, we perform a QCD sum rule analysis by using these currents. In Sec. IV, the numerical result is obtained for the mass of $Y(2175)$. In Sec. V, we use finite energy sum rule to calculate its mass again. Sec. VI is a summary. ## II Interpolating Currents In this section, we construct currents for the state $Y(2175)$ of $J^{PC}=1^{--}$. From the decay pattern $Y(2175)\to\phi(1020)f_{0}(980)$, we expect that there is a large $ss\bar{s}\bar{s}$ component in $Y(2175)$. We may add further quark and antiquark pairs, but the simplest choice would be $ss\bar{s}\bar{s}$. We will discuss later how this simplest quark content will be compatible with the above decay pattern when considering the possible structure of $\phi(1020)$ and $f_{0}(980)$. Let us now briefly see the flavor structure of the current. In the diquark- antidiquark construction $(ss)(\bar{s}\bar{s})$ where $ss$ and $\bar{s}\bar{s}$ pairs have a symmetric flavor structure, the flavor decomposition goes as $\mathbf{6_{f}}\otimes\mathbf{\bar{6}_{f}}=\mathbf{1_{f}}\oplus\mathbf{8_{f}}\oplus\mathbf{27_{f}}\,.$ (1) Therefore, the $(ss)(\bar{s}\bar{s})$ state is a mixing of $1_{f},8_{f}$ and $27_{f}$ multiplets in the ideal mixing scheme. Now we find that there are two non-vanishing currents for each state with the quantum number $J^{PC}=1^{--}$. For the state $ss\bar{s}\bar{s}$: $\displaystyle\eta_{1\mu}$ $\displaystyle=$ $\displaystyle(s_{a}^{T}C\gamma_{5}s_{b})(\bar{s}_{a}\gamma_{\mu}\gamma_{5}C\bar{s}_{b}^{T})-(s_{a}^{T}C\gamma_{\mu}\gamma_{5}s_{b})(\bar{s}_{a}\gamma_{5}C\bar{s}_{b}^{T})\,,$ (2) $\displaystyle\eta_{2\mu}$ $\displaystyle=$ $\displaystyle(s_{a}^{T}C\gamma^{\nu}s_{b})(\bar{s}_{a}\sigma_{\mu\nu}C\bar{s}_{b}^{T})-(s_{a}^{T}C\sigma_{\mu\nu}s_{b})(\bar{s}_{a}\gamma^{\nu}C\bar{s}_{b}^{T})\,,$ (3) where the sum over repeated indices ($\mu$ for Dirac spinor indices, and $a,b$ for color indices) is taken. $C=i\gamma_{2}\gamma_{0}$ is the Dirac field charge conjugation operator, and the superscript $T$ represents the transpose of the Dirac indices only. Besides the diquark-antidiquark currents, we can also construct the tetraquark currents by using quark-antiquark ($\bar{s}s$) pairs. We find that there are four non-vanishing currents: $\displaystyle\eta_{3\mu}$ $\displaystyle=$ $\displaystyle(\bar{s}_{a}s_{a})(\bar{s}_{b}\gamma_{\mu}s_{b})\,,$ $\displaystyle\eta_{4\mu}$ $\displaystyle=$ $\displaystyle(\bar{s}_{a}\gamma^{\nu}\gamma_{5}s_{a})(\bar{s}_{b}\sigma_{\mu\nu}\gamma_{5}s_{b})\,,$ $\displaystyle\eta_{5\mu}$ $\displaystyle=$ $\displaystyle{\lambda_{ab}}{\lambda_{cd}}(\bar{s}_{a}s_{b})(\bar{s}_{c}\gamma_{\mu}s_{d})\,,$ $\displaystyle\eta_{6\mu}$ $\displaystyle=$ $\displaystyle{\lambda_{ab}}{\lambda_{cd}}(\bar{s}_{a}\gamma^{\nu}\gamma_{5}s_{b})(\bar{s}_{c}\sigma_{\mu\nu}\gamma_{5}s_{d})\,.$ In Ref. Wang:2006ri , the author used $\eta_{5\mu}$ to perform QCD sum rule analysis, which is a mixing of $\eta_{1\mu}$ and $\eta_{2\mu}$. We can verify the following relations by using the Fierz transformation: $\displaystyle\eta_{5\mu}=-\frac{5}{3}\eta_{3\mu}-i\eta_{4\mu}\,,\,\,\eta_{6\mu}=3i\eta_{3\mu}+\frac{1}{3}\eta_{4\mu}\,.$ (4) Therefore, among the four $(\bar{q}q)(\bar{q}q)$ currents, two are independent. We can also verify the relations between $(ss)(\bar{s}\bar{s})$ currents and $(\bar{s}s)(\bar{s}s)$ currents, by using the Fierz transformation: $\displaystyle\eta_{1\mu}=-\eta_{3\mu}+i\eta_{4\mu}\,,\,\,\eta_{2\mu}=3i\eta_{3\mu}-\eta_{4\mu}\,.$ (5) Therefore, these two constructions are equivalent, and we will use $\eta_{1\mu}$ and $\eta_{2\mu}$ for QCD sum rule analysis. ## III QCD sum rule Analysis For the past decades QCD sum rule has proven to be a very powerful and successful non-perturbative method Shifman:1978bx ; Reinders:1984sr , and it has been applied to study tetraquark states in many references Chen:2007zzg ; Zhang:2006xp ; Matheus:2006xi ; Lee:2007mva ; Sugiyama:2007sg ; Matheus:2007ta ; Lee:2007gs . In sum rule analyses, we consider two-point correlation functions: $\Pi_{\mu\nu}(q^{2})\,\equiv\,i\int d^{4}xe^{iqx}\langle 0|T\eta_{\mu}(x){\eta_{\nu}^{\dagger}}(0)|0\rangle\,,$ (6) where $\eta_{\mu}$ is an interpolating current for the tetraquark. The Lorentz structure can be simplified to be: $\Pi_{\mu\nu}(q^{2})=({q_{\mu}q_{\nu}\over q^{2}}-g_{\mu\nu})\Pi^{(1)}(q^{2})+{q_{\mu}q_{\nu}\over q^{2}}\Pi^{(0)}(q^{2})\,.$ (7) We compute $\Pi(q^{2})$ in the operator product expansion (OPE) of QCD up to certain order in the expansion, which is then matched with a hadronic parametrization to extract information of hadron properties. At the hadron level, we express the correlation function in the form of the dispersion relation with a spectral function: $\Pi^{(1)}(q^{2})=\int^{\infty}_{16m_{s}^{2}}\frac{\rho(s)}{s-q^{2}-i\varepsilon}ds\,,$ (8) where the subscript is $(4m_{s})^{2}=16m_{s}^{2}$, and $\displaystyle\rho(s)$ $\displaystyle\equiv$ $\displaystyle\sum_{n}\delta(s-M^{2}_{n})\langle 0|\eta|n\rangle\langle n|{\eta^{\dagger}}|0\rangle\ $ (9) $\displaystyle=$ $\displaystyle f^{2}_{Y}\delta(s-M^{2}_{Y})+\rm{higher\,\,states}\,.$ For the second equation, as usual, we adopt a parametrization of one pole dominance for the ground state $Y$ and a continuum contribution. The sum rule analysis is then performed after the Borel transformation of the two expressions of the correlation function, (6) and (8) $\Pi^{(all)}(M_{B}^{2})\equiv\mathcal{B}_{M_{B}^{2}}\Pi^{(1)}(p^{2})=\int^{\infty}_{16m_{s}^{2}}e^{-s/M_{B}^{2}}\rho(s)ds\,.$ (10) Assuming the contribution from the continuum states can be approximated well by the spectral density of OPE above a threshold value $s_{0}$ (duality), we arrive at the sum rule equation $\Pi(M_{B}^{2})\equiv f^{2}_{Y}e^{-M_{Y}^{2}/M_{B}^{2}}=\int^{s_{0}}_{16m_{s}^{2}}e^{-s/M_{B}^{2}}\rho(s)ds\,.$ (11) Differentiating Eq. (11) with respect to $1/M_{B}^{2}$ and dividing it by Eq. (11), finally we obtain $M^{2}_{Y}=\frac{\frac{\partial}{\partial(-1/M_{B}^{2})}\Pi(M_{B}^{2})}{\Pi(M_{B}^{2})}=\frac{\int^{s_{0}}_{16m_{s}^{2}}e^{-s/M_{B}^{2}}s\rho(s)ds}{\int^{s_{0}}_{16m_{s}^{2}}e^{-s/M_{B}^{2}}\rho(s)ds}\,.$ (12) In the following, we study both Eqs. (11) and (12) as functions of the parameters such as the Borel mass $M_{B}$ and the threshold value $s_{0}$ for various combinations of the tetraquark currents. For the currents $\eta_{1\mu}$ and $\eta_{2\mu}$, we have calculated the OPE up to dimension twelve, which contains the $\langle\bar{q}q\rangle^{4}$ condensate: $\displaystyle\Pi_{1}(M_{B}^{2})$ $\displaystyle=$ $\displaystyle\int^{s_{0}}_{16m_{s}^{2}}\Bigg{[}{s^{4}\over 18432\pi^{6}}-{m_{s}^{2}s^{3}\over 256\pi^{6}}+\Big{(}-{\langle g^{2}GG\rangle\over 18432\pi^{6}}+{m_{s}\langle\bar{s}s\rangle\over 48\pi^{4}}\Big{)}s^{2}$ $\displaystyle+\Big{(}{\langle\bar{s}s\rangle^{2}\over 18\pi^{2}}-{m_{s}\langle g\bar{s}\sigma Gs\rangle\over 48\pi^{4}}+{17m_{s}^{2}\langle g^{2}GG\rangle\over 9216\pi^{6}}\Big{)}s+\Big{(}{\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 12\pi^{2}}-{m_{s}\langle g^{2}GG\rangle\langle\bar{s}s\rangle\over 128\pi^{4}}-{29m_{s}^{2}\langle\bar{s}s\rangle^{2}\over 12\pi^{2}}\Big{)}\Bigg{]}e^{-s/M_{B}^{2}}ds$ $\displaystyle+\Big{(}{5\langle g^{2}GG\rangle\langle\bar{s}s\rangle^{2}\over 864\pi^{2}}+{\langle g\bar{s}\sigma Gs\rangle^{2}\over 48\pi^{2}}+{20m_{s}\langle\bar{s}s\rangle^{3}\over 9}-{5m_{s}\langle g^{2}GG\rangle\langle g\bar{s}\sigma Gs\rangle\over 2304\pi^{4}}-{3m_{s}^{2}\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 2\pi^{2}}\Big{)}$ $\displaystyle+{1\over M_{B}^{2}}\Big{(}-{32g^{2}\langle\bar{s}s\rangle^{4}\over 81}-{\langle g^{2}GG\rangle\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 576\pi^{2}}-{10m_{s}\langle\bar{s}s\rangle^{2}\langle g\bar{s}\sigma Gs\rangle\over 9}+{m_{s}^{2}\langle g^{2}GG\rangle\langle\bar{s}s\rangle^{2}\over 576\pi^{2}}+{m_{s}^{2}\langle g\bar{s}\sigma Gs\rangle^{2}\over 12\pi^{2}}\Big{)}\,,$ $\displaystyle\Pi_{2}(M_{B}^{2})$ $\displaystyle=$ $\displaystyle\int^{s_{0}}_{16m_{s}^{2}}\Bigg{[}{s^{4}\over 12288\pi^{6}}-{3m_{s}^{2}s^{3}\over 512\pi^{6}}+\Big{(}{\langle g^{2}GG\rangle\over 18432\pi^{6}}+{m_{s}\langle\bar{s}s\rangle\over 32\pi^{4}}\Big{)}s^{2}$ $\displaystyle+\Big{(}{\langle\bar{s}s\rangle^{2}\over 12\pi^{2}}-{m_{s}\langle g\bar{s}\sigma Gs\rangle\over 32\pi^{4}}+{35m_{s}^{2}\langle g^{2}GG\rangle\over 9216\pi^{6}}\Big{)}s+\Big{(}{\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 8\pi^{2}}-{3m_{s}\langle g^{2}GG\rangle\langle\bar{s}s\rangle\over 128\pi^{4}}-{29m_{s}^{2}\langle\bar{s}s\rangle^{2}\over 8\pi^{2}}\Big{)}\Bigg{]}e^{-s/M_{B}^{2}}ds$ $\displaystyle+\Big{(}{5\langle g^{2}GG\rangle\langle\bar{s}s\rangle^{2}\over 288\pi^{2}}+{\langle g\bar{s}\sigma Gs\rangle^{2}\over 32\pi^{2}}+{10m_{s}\langle\bar{s}s\rangle^{3}\over 3}-{5m_{s}\langle g^{2}GG\rangle\langle g\bar{s}\sigma Gs\rangle\over 768\pi^{4}}-{9m_{s}^{2}\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 4\pi^{2}}\Big{)}$ $\displaystyle+{1\over M_{B}^{2}}\Big{(}-{16g^{2}\langle\bar{s}s\rangle^{4}\over 27}-{\langle g^{2}GG\rangle\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 192\pi^{2}}-{5m_{s}\langle\bar{s}s\rangle^{2}\langle g\bar{s}\sigma Gs\rangle\over 3}-{m_{s}^{2}\langle g^{2}GG\rangle\langle\bar{s}s\rangle^{2}\over 576\pi^{2}}+{m_{s}^{2}\langle g\bar{s}\sigma Gs\rangle^{2}\over 8\pi^{2}}\Big{)}\,.$ In the above equations, $\langle\bar{s}s\rangle$ is the dimension $D=3$ strange quark condensate; $\langle g^{2}GG\rangle$ is a $D=4$ gluon condensate; $\langle g\bar{s}\sigma Gs\rangle$ is $D=5$ mixed condensate. There are many terms which give minor contributions, such as $\langle g^{3}G^{3}\rangle$, and we omit them. As usual, we assume the vacuum saturation for higher dimensional condensates such as $\langle 0|\bar{q}q\bar{q}q|0\rangle\sim\langle 0|\bar{q}q|0\rangle\langle 0|\bar{q}q|0\rangle$. To obtain these results, we keep the terms of order $O(m_{q}^{2})$ in the propagators of a massive quark in the presence of quark and gluon condensates: $\displaystyle iS^{ab}$ $\displaystyle\equiv$ $\displaystyle\langle 0|T[q^{a}(x)q^{b}(0)]|0\rangle$ (15) $\displaystyle=$ $\displaystyle{i\delta^{ab}\over 2\pi^{2}x^{4}}\hat{x}+{i\over 32\pi^{2}}{\lambda^{n}_{ab}\over 2}g_{c}G^{n}_{\mu\nu}{1\over x^{2}}(\sigma^{\mu\nu}\hat{x}+\hat{x}\sigma^{\mu\nu})-{\delta^{ab}\over 12}\langle\bar{q}q\rangle$ $\displaystyle+{\delta^{ab}x^{2}\over 192}\langle g_{c}\bar{q}\sigma Gq\rangle-{m_{q}\delta^{ab}\over 4\pi^{2}x^{2}}+{i\delta^{ab}m_{q}\langle\bar{q}q\rangle\over 48}\hat{x}+{i\delta^{ab}m_{q}^{2}\over 8\pi^{2}x^{2}}\hat{x}$ We find that there is an approximate relation between the correlation functions of $\eta_{1\mu}$ and $\eta_{2\mu}$: $3\Pi_{1}(M_{B}^{2})\sim 2\Pi_{2}(M_{B}^{2})\,,$ (16) which is valid for the continuum, $\langle\bar{s}s\rangle$, and $\langle g_{c}\bar{q}\sigma Gq\rangle$ terms, etc. So the numerical results by using them are also very similar. ## IV Numerical Analysis In our numerical analysis, we use the following values for various condensates and $m_{s}$ at 1 GeV and $\alpha_{s}$ at 1.7 GeV Yang:1993bp ; Narison:2002pw ; Gimenez:2005nt ; Jamin:2002ev ; Ioffe:2002be ; Ovchinnikov:1988gk ; Ellis:1996xc ; Yao:2006px : $\displaystyle\langle\bar{q}q\rangle=-(0.240\mbox{ GeV})^{3}\,,$ $\displaystyle\langle\bar{s}s\rangle=-(0.8\pm 0.1)\times(0.240\mbox{ GeV})^{3}\,,$ $\displaystyle\langle g_{s}^{2}GG\rangle=(0.48\pm 0.14)\mbox{ GeV}^{4}\,,$ $\displaystyle\langle g_{s}\bar{q}\sigma Gq\rangle=-M_{0}^{2}\times\langle\bar{q}q\rangle\,,$ (17) $\displaystyle M_{0}^{2}=(0.8\pm 0.2)\mbox{ GeV}^{2}\,,$ $\displaystyle m_{s}(1\mbox{ GeV})=125\pm 20\mbox{ MeV}\,,$ $\displaystyle\alpha_{s}(1.7\mbox{GeV})=0.328\pm 0.03\pm 0.025\,.$ There is a minus sign in the definition of the mixed condensate $\langle g_{s}\bar{q}\sigma Gq\rangle$, which is different from that used in some other QCD sum rule studies. This difference just comes from the definition of coupling constant $g_{s}$ Yang:1993bp ; Hwang:1994vp . First we want to study the convergence of the operator product expansion, which is the cornerstone of the reliable QCD sum rule analysis. By taking $s_{0}$ to be $\infty$ and the integral subscript $16m_{s}^{2}$ to be zero, we obtain the numerical series of the OPE as a function of $M_{B}$: $\displaystyle\Pi_{1}(M_{B}^{2})$ $\displaystyle=$ $\displaystyle 1.4\times 10^{-6}M_{B}^{10}-3.8\times 10^{-7}M_{B}^{8}-6.2\times 10^{-7}M_{B}^{6}+4.2\times 10^{-7}M_{B}^{4}$ (18) $\displaystyle-1.2\times 10^{-6}M_{B}^{2}+4.7\times 10^{-8}-1.5\times 10^{-7}M_{B}^{-2}\,,$ $\displaystyle\Pi_{2}(M_{B}^{2})$ $\displaystyle=$ $\displaystyle 2.0\times 10^{-6}M_{B}^{10}-5.7\times 10^{-7}M_{B}^{8}-8.0\times 10^{-7}M_{B}^{6}+6.4\times 10^{-7}M_{B}^{4}$ (19) $\displaystyle-1.7\times 10^{-6}M_{B}^{2}+1.0\times 10^{-7}-2.2\times 10^{-7}M_{B}^{-2}\,.$ After careful testing of the free parameter Borel mass $M_{B}$, we find for $M_{B}^{2}>2$ GeV2, which is the region suitable for the study of $Y(2175)$, the Borel mass dependence is weak. Moreover, the convergence of the OPE is satisfied in this region. The correlation function of the current $\eta_{1\mu}$ is shown in Fig. 1, when we take $s_{0}=5.7$ GeV2 (the integral subscript is still $16m_{s}^{2}$). We find that in the region of 2 GeV${}^{2}<M_{B}^{2}<5$ GeV2, the perturbative term (the solid line in Fig. 1) gives the most important contribution, and the convergence is quite good. Figure 1: Various contribution to the correlation function for the current $\eta_{1\mu}$ as functions of the Borel mass $M_{B}$ in units of GeV10 at $s_{0}=5.7$ GeV2. The labels indicate the dimension up to which the OPE terms are included. It is important to note that the $Y(2175)$ state is not the lowest state in the $1^{--}$ channel containing $s\bar{s}$ and that the interpolating currents see only the quantum number of the states. It is possible that the low-lying states $\phi(1020)$ and $\phi(1680)$ also couple to the tetraquark currents $\eta_{1\mu}$ and $\eta_{2\mu}$. If so, their contribution to the spectral density and the resulting correlation function should be positive definite. However, we find that (1) the spectral densities $\rho(s)$ of Eq. (9) for both currents $\eta_{1\mu}$ and $\eta_{2\mu}$ are negative when $s<2$ GeV2; (2) the Borel transformed correlation function $\Pi(M_{B}^{2})$ in Eq. (11) is also negative in the region $s_{0}<4.3$ GeV2 and $1$ GeV${}^{2}<M_{B}^{2}<4$ GeV2. As an illustration, we show the correlation function as a function of $s_{0}$ in Fig. 2. This fact indicates that the $ss\bar{s}\bar{s}$ tetraquark currents couple weakly to the lower states $\phi(1020)$ and $\phi(1680)$ in the present QCD sum rule analysis. Figure 2: The correlation function for the current $\eta_{1\mu}$ as a function of $s_{0}$ in units of GeV10. The curves are obtained by setting $M_{B}^{2}=1$ GeV2 (long-dashed line), 2 GeV2 (short-dashed line) and 4 GeV2 (solid line). The pole contribution is not large enough for both currents due to $D=10$ perturbative term $\int_{0}^{s_{0}}e^{-s/M_{B}^{2}}s^{4}ds$, which is a common feature for any multiquark interpolating currents with high dimensions. The mixing of the currents $\eta_{1\mu}$ and $\eta_{2\mu}$ does not improve the rate of the pole contribution. The small pole contribution suggests that the continuum contribution to the spectral density is dominant, which demands a very careful choice of the parameters of the QCD sum rule. In our numerical analysis, we require the extracted mass have a dual minimum dependence on both the Borel parameter $M_{B}$ and the threshold parameter $s_{0}$. In this way, we can find a good working region of $M_{B}$ and $s_{0}$ (Borel window), where the mass of $Y(2175)$ can be determined reliably. Now the mass is shown as functions of the Borel mass $M_{B}$ and the threshold value $s_{0}$ in Fig. 3 and Fig. 4. The threshold value is taken to be around $5\sim 7$ GeV2, where its square root is around $2.2\sim 2.7$ GeV. We find that there is a mass minimum around 2.4 GeV for the current $\eta_{1\mu}$, when we take $M_{B}^{2}\sim 4$ GeV2 and $s_{0}\sim 5.7$ GeV2. While this minimum is around 2.3 GeV for the current $\eta_{2\mu}$, when we take $M_{B}^{2}\sim 4$ GeV2 and $s_{0}\sim 5.4$ GeV2. Figure 3: The mass of $Y(2175)$ as a function of $M_{B}$ (Left) and $s_{0}$ (Right) in units of GeV for the current $\eta_{1\mu}$. Figure 4: The mass of $Y(2175)$ as a function of $M_{B}$ (Left) and $s_{0}$ (Right) in units of GeV for $\eta_{2\mu}$. In short summary, we have performed the QCD sum rule analysis for both $\eta_{1\mu}$ and $\eta_{2\mu}$. The obtained results are quite similar. This is due to the similarity of the two correlation functions as shown in Eq. (16). We have also considered their mixing, which also give the similar result. The mass is predicted to be around $2.3\sim 2.4$ GeV in the QCD sum rule. ## V Finite Energy Sum Rule In this section, we use the method of finite energy sum rule (FESR). In order to calculate the mass in the FESR, we first define the $n$th moment by using the spectral function $\rho(s)$ in Eq. (9) $W(n,s_{0})=\int^{s_{0}}_{0}\rho(s)s^{n}ds\,.$ (20) This integral is used for the phenomenological side, while the integral along the circular contour of radius $s_{0}$ on the $q^{2}$ complex plain should be performed for the theoretical side. The lower integral bound $s=0$ is taken in order to include the $delta$-function contribution in the OPE (Eqs. (V) and (V)). With the assumption of quark-hadron duality, we obtain $W(n,s_{0})\Big{|}_{Hadron}=W(n,s_{0})\Big{|}_{OPE}\,.$ (21) The mass of the ground state can be obtained as $M^{2}_{Y}(n,s_{0})={W(n+1,s_{0})\over W(n,s_{0})}\,.$ (22) For the currents $\eta_{1\mu}$ and $\eta_{2\mu}$, the spectral functions $\rho_{1}(s)$ and $\rho_{2}(s)$ can be drawn from Eqs. (III) and (III). The d = 12 terms which are proportional to $1/(q^{2})^{2}$ do not contribute to the function $W(n,s_{0})$ of Eq. (20) for $n=0$, or they have a very small contribution for $n=1$, when the theoretical side is computed by the integral over the circle of radius $s_{0}$ on the complex $q^{2}$ plain. Therefore, the spectral densities for $\eta_{1\mu}$ and $\eta_{2\mu}$ take the following form up to dimension 10, $\displaystyle\rho_{1}(s)$ $\displaystyle=$ $\displaystyle{s^{4}\over 18432\pi^{6}}-{m_{s}^{2}s^{3}\over 256\pi^{6}}+\Big{(}-{\langle g^{2}GG\rangle\over 18432\pi^{6}}+{m_{s}\langle\bar{s}s\rangle\over 48\pi^{4}}\Big{)}s^{2}$ $\displaystyle+\Big{(}{\langle\bar{s}s\rangle^{2}\over 18\pi^{2}}-{m_{s}\langle g\bar{s}\sigma Gs\rangle\over 48\pi^{4}}+{17m_{s}^{2}\langle g^{2}GG\rangle\over 9216\pi^{6}}\Big{)}s+\Big{(}{\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 12\pi^{2}}-{m_{s}\langle g^{2}GG\rangle\langle\bar{s}s\rangle\over 128\pi^{4}}-{29m_{s}^{2}\langle\bar{s}s\rangle^{2}\over 12\pi^{2}}\Big{)}$ $\displaystyle+\Big{(}{5\langle g^{2}GG\rangle\langle\bar{s}s\rangle^{2}\over 864\pi^{2}}+{\langle g\bar{s}\sigma Gs\rangle^{2}\over 48\pi^{2}}+{20m_{s}\langle\bar{s}s\rangle^{3}\over 9}-{5m_{s}\langle g^{2}GG\rangle\langle g\bar{s}\sigma Gs\rangle\over 2304\pi^{4}}-{3m_{s}^{2}\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 2\pi^{2}}\Big{)}\delta(s)\,,$ $\displaystyle\rho_{2}(s)$ $\displaystyle=$ $\displaystyle{s^{4}\over 12288\pi^{6}}-{3m_{s}^{2}s^{3}\over 512\pi^{6}}+\Big{(}{\langle g^{2}GG\rangle\over 18432\pi^{6}}+{m_{s}\langle\bar{s}s\rangle\over 32\pi^{4}}\Big{)}s^{2}$ $\displaystyle+\Big{(}{\langle\bar{s}s\rangle^{2}\over 12\pi^{2}}-{m_{s}\langle g\bar{s}\sigma Gs\rangle\over 32\pi^{4}}+{35m_{s}^{2}\langle g^{2}GG\rangle\over 9216\pi^{6}}\Big{)}s+\Big{(}{\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 8\pi^{2}}-{3m_{s}\langle g^{2}GG\rangle\langle\bar{s}s\rangle\over 128\pi^{4}}-{29m_{s}^{2}\langle\bar{s}s\rangle^{2}\over 8\pi^{2}}\Big{)}$ $\displaystyle+\Big{(}{5\langle g^{2}GG\rangle\langle\bar{s}s\rangle^{2}\over 288\pi^{2}}+{\langle g\bar{s}\sigma Gs\rangle^{2}\over 32\pi^{2}}+{10m_{s}\langle\bar{s}s\rangle^{3}\over 3}-{5m_{s}\langle g^{2}GG\rangle\langle g\bar{s}\sigma Gs\rangle\over 768\pi^{4}}-{9m_{s}^{2}\langle\bar{s}s\rangle\langle g\bar{s}\sigma Gs\rangle\over 4\pi^{2}}\Big{)}\delta(s)\,.$ Figure 5: The mass of $Y(2175)$ by using the current $\eta_{1\mu}$ (solid line) and $\eta_{2\mu}$ (dashed line) as a function of $s_{0}$ in units of GeV. The mass is shown as a function of the threshold value $s_{0}$ in Fig. 5, where $n$ is chosen to be 1. We find that there is a mass minimum. It is around 2.3 GeV for the current $\eta_{1\mu}$ when we take $s_{0}\sim 5.2$ GeV2, while it is around 2.2 GeV for the current $\eta_{2\mu}$ when we take $s_{0}\sim 4.8$ GeV2. For the current $\eta_{1\mu}$, the minimum point occurs at $\sqrt{s_{0}}=2.28$ GeV where the mass takes 2.3 GeV, and the threshold value is slightly smaller than the mass, unlike the ordinary expectation that $\sqrt{s_{0}}$ is larger than the obtained mass. However, the minimum point is on the very shallow minimum curve and the resulting mass is rather insensitive to the change in the $\sqrt{s_{0}}$ value. Therefore, we can increase $\sqrt{s_{0}}$ slightly more, for example 2.45 GeV, but the mass still remains at around $2.35$ GeV, which is smaller than $\sqrt{s_{0}}$ now. This fact is due to the uncertainty of our sum rule analysis as well as the negative part of the spectral densities. For example, if we take the lower limit of integrations in Eq. (22) to be 1 GeV2 instead of $16m_{s}^{2}$ (the positive part starts at around 3 GeV2), the mass minimum will be around 2.1 GeV, when $s_{0}$ is around 4.5 GeV2; if we take the lower limit to be 2 GeV2, the mass minimum would be around 2.0 GeV, when $s_{0}$ is around 4 GeV2. We show the second case in Fig. 6. The region $5<s_{0}<6$ GeV2 is suitable for the QCD sum rule analysis, and the mass obtained is around 2.2 GeV. Therefore, considering the uncertainty of the QCD sum rule, we obtain the same result as the previous one. Figure 6: The mass of $Y(2175)$ by using the current $\eta_{1\mu}$ (solid line) and $\eta_{2\mu}$ (dashed line) as a function of $s_{0}$ in units of GeV, when the lower limit of integrations in Eq. (22) is 2 GeV2 instead of $16m_{s}^{2}$. ## VI Summary and Discussions In this work we have studied the mass of the state $Y(2175)$ with the quantum numbers $J^{PC}=1^{--}$ in the QCD sum rule. We have constructed both the diquark-antidiquark currents $(ss)(\bar{s}\bar{s})$ and the meson-meson currents $(\bar{s}s)(\bar{s}s)$. We find that there are two independent currents for both cases and verify the relations between them. Then using the two $(ss)(\bar{s}\bar{s})$ currents, we calculate the OPE up to dimension twelve, which contains the $\langle\bar{s}s\rangle^{4}$ condensates. The convergence of the OPE turns our to be very good. We find that the OPE’s of the two currents are similar, and therefore, the obtained results are also similar. By using both the SVZ sum rule and the finite energy sum rule, we find that there is a mass minimum. For SVZ sum rule, the minimum is in the region $5<s_{0}<7$ GeV2 and $2<M_{B}^{2}<4$ GeV2. For finite energy sum rule, the minimum is in the region $4.5<s_{0}<5.5$ GeV2. Considering the uncertainty, the mass obtained is around $2.3\pm 0.4$ GeV. The state $Y(2175)$ can be accommodated in the QCD sum rule formalism although the central value of the mass is about 100 MeV higher than the experimental value. We calculate the OPE up to dimension twelve and include many terms, but still the accuracy is around 20%. This is the usual accuracy of the QCD sum rule. In our analysis it is partly due to the many omitted condensates such as $\langle GGG\rangle$ etc. We have investigated the coupling of the currents to the lower lying states including $\phi(1020)$ and found that the relevant spectral density becomes negative, implying that the present four-quark currents can not describe those states properly. This fact indicates that the four-quark interpolating currents couple rather weakly to $\phi(1020)$, which is a pure $s\bar{s}$ state. We can test the tetraquark structure of $Y(2175)$ by considering its decay properties. Naively, the $ss\bar{s}\bar{s}$ tetraquark would fall apart via $S$-wave into the $\phi(1020)f_{0}(980)$ pair, and would have a very large width. The experimental width of $Y(2175)$ is only about $60$ MeV, which seems too narrow to be a pure tetraquark state. We can discuss the decay of the $Y(2175)$ by borrowing an argument based on a valence quark picture. The $(\bar{s}s)(\bar{s}s)$ configuration for $Y(2175)$ can be a combination of ${}^{3}S_{1}$ and ${}^{3}P_{0}$, which may fall apart into two mesons of $1^{-}$ and $0^{+}$ in the $s$-wave. In the QCD sum rule the $1^{-}$ $\bar{s}s$ meson is well identified with $\phi(1020)$, while the $0^{+}$ $\bar{s}s$ meson has a mass around $1.5$ GeV and is hard to be identified with the observed $f_{0}(980)$. Therefore, such a fall-apart decay would simply be suppressed due to the kinematical reason. The physical $f_{0}(980)$ state may be a tetraquark state as discussed in the previous QCD sum rule study Chen:2007zzg . Then the transition $Y(2175)\to\phi(1020)+f_{0}({\rm tetraquark})$ should be accompanied by a $\bar{q}q$ creation violating the OZI rule, as well as by an annihilation of one quanta of orbital angular momentum. These facts may once again suppress the decay of $Y(2175)\to\phi(1020)+f_{0}(980)$. This fact was studied in the recent paper by Torres, Khemchandani, Geng, Napsuciale and Oset Torres:2008gy . They studied the $\phi K\bar{K}$ system with the Faddeev equations where the contained $K\bar{K}$ form the $f_{0}(980)$ resonance. The decay width they calculated is around $18$ MeV, not far from the experimental value. The all above evidences would imply that the $Y(2175)$ is a possible candidate of a tetraquark state. $Y(2175)$ could be a threshold effect, a hybrid state $s\bar{s}G$, a tetraquark, an excited $s\bar{s}$ state or a mixture of all the above possibilities. Because of its non-exotic quantum number, it is not easy to establish its underlying structure. Clearly more experimental and theoretical investigations are required. One byproduct of the present work is the interesting observation that some type of four-quark interpolating currents may couple weakly to the conventional $q\bar{q}$ ground states. If future work confirms this point, we may have a novel framework to study the excited $q\bar{q}$ mesons using the four-quark interpolating currents, which is not feasible for the traditional $q\bar{q}$ interpolating currents. ## Acknowledgments The authors thank G. Erkol for useful discussions. H.X.C. is grateful for Monkasho support for his stay at the Research Center for Nuclear Physics where this work is done. X.L was supported by the National Natural Science Foundation of China under Grants 10705001 and the China Postdoctoral Science foundation (20060400376). A.H. is supported in part by the Grant for Scientific Research ((C) No.19540297) from the Ministry of Education, Culture, Science and Technology, Japan. S.L.Z. was supported by the National Natural Science Foundation of China under Grants 10625521 and 10721063 and Ministry of Education of China. ## References * (1) Babar Collaboration, B. Aubert et al., Phys. Rev. Lett. 90, 242001 (2003); F. Porter, Eur. Phys. J. C33, 219 (2004); Belle Collaboration, P. Krokovny et al., Phys. Rev. Lett. 91, 262002 (2003); CLEO Collaboration, D. Besson et al., Phys. Rev. D 68, 032002 (2003). * (2) Belle Collaboration, Y.Mikami et al., Phys. Rev. Lett. 92, 012002 (2004); Belle Collaboration, P. Krokovny et al., arXiv: hep-ex/0310053; FOCUS Collabortation, E. W. Vaandering, arXiv: hep-ex/0406044; Babar Collaboration, B. Aubert et al., Phys. Rev. Lett. 93, 181801 (2004); Babar Collaboration, B. Aubert et al., Phys. Rev. D 69, 031101 (2004); Babar Collaboration, G. Calderini et al., arXiv: hep-ex/0405081; Babar Collaboration, B. Aubert et al., arXiv: hep-ex/0408067. * (3) Belle Collaboration, S.K. Coi et al., Phys. Rev. Lett. 91, 262001 (2003); CDF Collaboration, D. Acosta et al., Phys. Rev. Lett. 93, 072001 (2004); D0 Collaboration, V.M. Abazov et al., Phys. Rev. Lett. 93, 162002 (2003); Babar Collaboration, B. Aubert et al., Phys. Rev. D 71, 071103 (2005); Belle Collaboration, K. Abe et al., arXiv: hep-ex/0505037; Babar Collaboration, B. Aubert, Phys. Rev. D 74, 071101 (2006); Belle Collaboration, G. Gokhroo et al., Phys. Rev. Lett. 97, 162002 (2006); Babar Collaboration, talk given by P. Grenier in Moriond QCD 2007, 17-24 March, http://moriond.in2p3.fr/QCD/2007/SundayAfternoon/Grenier.pdf; CDF Collaboration, A. Abulencia et al., Phys. Rev. Lett. 96, 102002 (2006); Belle Collaboration, K. Abe et al., arXiv: hep-ex/0505038; CDF Collaboration, A. Abulencia, Phys. Rev. Lett 98, 132002 (2007). * (4) Babar Collaboration, B. Aubert et al., Phys. Rev. Lett. 95, 142001 (2005); CLEO Collaboration, T. E. Coan et al.,Phys. Rev. Lett. 96, 162003 (2006); CLEO Collaboration, Q. He et al., Phys. Rev. D 74, 091104 (2006); Belle Collaboration, K. Abe et al., arXiv: hep-ex/0612006; Belle Collaboration, C.Z. Yuan et al., arXiv: 07072541v1 [hep-ex]. * (5) BABAR Collaboration, B. Aubert et al., Phys. Rev. D 74, 091103 (2006). * (6) BABAR Collaboration, B. Aubert et al., Phys. Rev. D 76, 012008 (2007). * (7) BaBar Collaboration, B. Aubert et al., arXiv:0710.4451 [hep-ex]. * (8) BES Collaboration, M. Ablikim et al., arXiv:0712.1143 [hep-ex]. * (9) G. J. Ding and M. L. Yan, Phys. Lett. B 650, 390 (2007). * (10) G.J. Ding and M.L. Yan, Phys. Lett. B 657, 49 (2007). * (11) Z. G. Wang, Nucl. Phys. A 791, 106 (2007). * (12) S. L. Zhu, arXiv:hep-ph/0703225. * (13) M. Napsuciale, E. Oset, K. Sasaki and C. A. Vaquera-Araujo, Phys. Rev. D 76, 074012 (2007). * (14) I. V. Anikin, B. Pire and O. V. Teryaev, Phys. Lett. B 626, 86 (2005). * (15) F. K. Guo, P. N. Shen, Z. G. Wang, W. H. Liang and L. S. Kisslinger, arXiv:hep-ph/0703062. * (16) Yu. M. Bystritskiy, M. K. Volkov, E. A. Kuraev, E. Bartos and M. Secansky, arXiv:0712.0304 [hep-ph]. * (17) H. X. Chen, A. Hosaka and S. L. Zhu, Phys. Lett. B 650, 369 (2007). * (18) M. A. Shifman, A. I. Vainshtein and V. I. Zakharov, Nucl. Phys. B 147, 385 (1979). * (19) L. J. Reinders, H. Rubinstein and S. Yazaki, Phys. Rept. 127, 1 (1985). * (20) A. Zhang, T. Huang and T. G. Steele, Phys. Rev. D 76, 036004 (2007). * (21) R. D. Matheus, S. Narison, M. Nielsen and J. M. Richard, Phys. Rev. D 75, 014005 (2007). * (22) H. J. Lee and N. I. Kochelev, arXiv:hep-ph/0702225. * (23) J. Sugiyama, T. Nakamura, N. Ishii, T. Nishikawa and M. Oka, Phys. Rev. D 76, 114010 (2007) * (24) R. D. Matheus, F. S. Navarra, M. Nielsen and R. Rodrigues da Silva, Phys. Rev. D 76, 056005 (2007). * (25) S. H. Lee, A. Mihara, F. S. Navarra and M. Nielsen, arXiv:0710.1029 [hep-ph]. * (26) W. M. Yao et al. [Particle Data Group], J. Phys. G 33, 1 (2006). * (27) K. C. Yang, W. Y. P. Hwang, E. M. Henley and L. S. Kisslinger, Phys. Rev. D 47, 3001 (1993). * (28) S. Narison, Camb. Monogr. Part. Phys. Nucl. Phys. Cosmol. 17, 1 (2002). * (29) V. Gimenez, V. Lubicz, F. Mescia, V. Porretti and J. Reyes, Eur. Phys. J. C 41, 535 (2005) * (30) M. Jamin, Phys. Lett. B 538, 71 (2002). * (31) B. L. Ioffe and K. N. Zyablyuk, Eur. Phys. J. C 27, 229 (2003). * (32) A. A. Ovchinnikov and A. A. Pivovarov, Sov. J. Nucl. Phys. 48, 721 (1988) [Yad. Fiz. 48, 1135 (1988)]. * (33) J. R. Ellis, E. Gardi, M. Karliner and M. A. Samuel, Phys. Rev. D 54, 6986 (1996). * (34) W. Y. P. Hwang and K. C. Yang, Phys. Rev. D 49, 460 (1994). * (35) A. M. Torres, K. P. Khemchandani, L. S. Geng, M. Napsuciale and E. Oset, arXiv:0801.3635 [Unknown].

arxiv-papers

2024-09-04T02:48:53.504219

0802.0368

Dynamical symmetry breaking of lambda and vee-type three-level systems on quantization of the field modes Mihir Ranjan Nath111mrnath_95@rediffmail.com and Surajit Sen222ssen55@yahoo.com Department of Physics Guru Charan College Silchar 788004, India Asoke Kumar Sen333asokesen@sancharnet.in Department of Physics Assam University Silchar 788011, India Gautam Gangopadhyay444gautam@bose.res.in S N Bose National Centre for Basic Sciences JD Block, Sector III Salt Lake City, Kolkata 700098, India ###### Abstract We develop a scheme to construct the Hamiltonians of the lambda, vee and cascade type of three-level configurations using the generators of $SU(3)$ group. It turns out that this approach provides a well defined selection rule to give different Hamitonians for each configurations. The lambda and vee type configurations are exactly solved with different initial conditions while taking the two-mode classical and quantized fields . For the classical field, it is shown that the Rabi oscillation of the lambda model is similar to that of the vee model and the dynamics of the vee model can be recovered from lambda model and vice versa simply by inversion. We then proceed to solve the quantized version of both models introducing a novel Euler matrix formalism. It is shown that this dynamical symmetry exhibited in the Rabi oscillation of two configurations for the semiclassical models is completely destroyed on quantization of the field modes. The symmetry can be restored within the quantized models when the field modes are both in the coherent states with large average photon number which is depicted through the collapse and revival of the Rabi oscillations. I.Introduction Quantum Optics gives birth to many novel proposals which are within reach of present-day ingenious experiments performed with intense narrow-band tunable laser and high-Q superconducting cavity [1]. Major thrust in the atomic, molecular and optical experiments primarily involves the coherent manipulation of the quantum states which may be useful to verify several interesting results of quantum information theory and also the experimental realization of the quantum computer [2,3]. The actual number of the quantum mechanical states of atoms involved in the interaction with light is of much importance in these days since many coherent effects are due to the level structure of the atom. It is well-known that the two-level system and its quantized version, namely, the Jaynes-Cummings model (JCM), have been proved to be an useful theoretical laboratory to understand many subtle issues of the cavity electrodynamics [4,5]. The two-level system is modeled using the Pauli’s spin matrices - the spin-half representation of $SU(2)$ group, where apart from the level number, the spectrum is designated by the photon number as the quantum number. A natural but non-trivial extension of the JCM is the three-level system and it exhibits plethora of optical phenomena such as, two-photon coherence [6], resonance Raman scattering [7], double resonance process [8],population trapping [9], three-level super radiance [10], three-level echoes [11], STIRAP [12], quantum jump [13], quantum zeno effect [14], Electromagnetically Induced Transparency [15,16] etc. There are three distinct schemes of the three-level configurations which are classified as the lambda, vee and cascade systems respectively. The Hamiltonians of these configurations are generally modeled by two two-level systems coupled by the two modes of cavity fields of different frequencies [17,18]. Although these Hamiltonians succeed in revealing several phenomena [19,20], however, their $ad$ $hoc$ construction subsides the underlying symmetry and its role in the population dynamics of these systems. The connection between the $SU(N)$ symmetry and the $N$-level system in general, was investigated extensively in recent past [21-27]. These studies not only mimic the possible connection between quantum optics with the octet symmetry, well-known paradigm of particle physics, but for $N=3$, it also reveals several interesting results such as the realization of the eight dimensional Bloch equation, existence of non-linear constants [18,22], population transfer via continuum [28], dynamical aspects of three-level system in absence of dissipation [29] etc. However, inspite of these progress, a general formalism as well as the $ab$ $initio$ solutions of all three configurations are yet to be developed for the reasons mentioned below. The generic model Hamiltonian of a three-level configuration with three well- defined energy levels can be represented by the hermitian matrix $H=\left[{\begin{array}[]{*{20}c}\Delta_{3}&h_{32}&h_{31}\\\ h_{32}&\Delta_{2}&h_{21}\\\ h_{31}&h_{21}&\Delta_{1}\\\ \end{array}}\right],\hfill(1)$ where $h_{ij}(i,j=1,2,3)$ be the matrix element of specific transition and $\Delta_{i}$ be the detuning which vanishes at resonance. We note that from Eq.(1), the lambda system, which corresponds to the transition $1\leftrightarrow 3\leftrightarrow 2$ shown in Fig.1a, can be described by the Hamiltonian with elements $h_{21}=0$, $h_{32}\neq 0$ and $h_{31}\neq 0$. Similarly the vee model, characterized by the transition $3\leftrightarrow 1\leftrightarrow 2$ shown in Fig.1b, corresponds to the elements $h_{21}\neq 0$, $h_{32}=0$ and $h_{31}\neq 0$ and for the cascade model we have transition $1\leftrightarrow 2\leftrightarrow 3$, we have $h_{21}\neq 0$, $h_{32}\neq 0$ and $h_{31}=0$ respectively. Thus we have distinct Hamiltonian for three different configurations which can be read off from Eq.(1) shown in Fig.1. This definition, however, differs from the proposal advocated by Hioe and Eberly, who argued the order of the energy levels to be $E_{1}<E_{3}<E_{2}$ for the lambda system, $E_{2}<E_{3}<E_{1}$ for the vee system and $E_{1}<E_{2}<E_{3}$ for the cascade system respectively [18,21,22]. In their scheme, the level-2 is always be the intermediary level which becomes the upper, lower and middle level to generate the lambda, vee and cascade configurations respectively. It is worth noting that, if we follow their scheme, these energy conditions map all three three-level configurations to a unique cascade Hamiltonian described by the matrix with elements $h_{12}\neq 0$, $h_{23}\neq 0$ and $h_{13}=0$ in Eq.(1). Thus because of the similar structure of the model Hamiltonian, if we start formulating the solutions of the lambda, vee and cascade configurations, then it would led to same spectral feature. Furthermore, due to the same reason, the eight dimensional Bloch equation always remains same for all three models [18,22]. Both of these consequences go against the usual notion because wide range of coherent phenomena mentioned above arises essentially due to different class of the three-level configurations. Thus it is worth pursing to formulate a comprehensive approach, where we have distinct Hamiltonian for three configurations without altering the second level for each model. The problem of preparing multilevel atoms using one or more laser pulses is of considerable importance from experimental point of view. Thus the completeness of the study of the three-level systems requires the exact solution of these models to find the probability amplitudes of all levels, the effect of the field quantization on the population oscillation and, most importantly, the observation of the collapse and revival effect. In recent past, the three- level systems and its several ramifications were extensively covered in a general framework of the $SU(N)$ group having $N$-levels [21-27,30,31]. Also, the semiclassical model [24,32,33] and its fully quantized version [23,34,35] are studied, but to our knowledge, the pursuit of the exact solutions of different three-level systems in the spirit of the theory of Electron Spin Resonance (ESR) model and JCM, are still to be facilitated analytically. In a recent paper, we have studied the exact solutions of the equidistant cascade system interacting with the single mode classical and quantized field with different initial conditions [36]. It is shown that for the semiclassical model the Rabi oscillation exhibits a symmetric pattern of evolution, which is destroyed on quantization of the cavity field. We also show that this symmetry is restored by taking the cavity mode to be the coherent state indicating the proximity of the coherent state to the classical field. We have further studied the equidistant cascade four-level system and obtain similar conclusions [37]. To extend above studies for the lambda and vee models we note that the vee configuration can be obtained from the lambda configuration simply by inversion. However, it is worth noting that, the lambda configuration is associated with processes such as STIRAP [12], EIT [15,16] etc, while the vee configuration corresponds to the phenomena such as quantum jump [13], quantum zeno effect [14], quantum beat [3] etc indicating that both the processes are fundamentally different. It is therefore natural to examine the inversion symmetry between the models by comparing their Rabi oscillations and study the effect of the field quantization on that symmetry. The comparison shows that the inversion symmetry exhibited by the semiclassical models is completely spoiled on quantization of the cavity modes indicating the non-trivial role of the vacuum fluctuation in the symmetry breaking. The remaining part of the paper is organized as follows. In Section-II, we discuss the basic tenets of the $SU(3)$ group necessary to develop the Hamiltonian of all possible three-level configurations. Section-III deals with the solution of the lambda model taking the two field modes as the classical fields and then in Section IV we proceed to solve the corresponding quantized version of the model using a novel Euler matrix formalism. Section-V and VI we present similar calculation for the vee model taking the mode fields to be first classical and then quantized respectively. In Section-VII we compare the population dynamics in both models and discuss its implications. Finally in Section-VIII we conclude our results. II.The Models The most general Hamiltonian of a typical three-level configuration is given by Eq.(1) which contains several non-zero matrix elements showing all possible allowed transitions. To show how the $SU(3)$ symmetry group provides a definite scheme of selection rule which forbids any one of the three transitions to give the Hamiltonian of a specific model, let us briefly recall the tenets of $SU(3)$ group described by the Gell-Mann matrices, namely, $\lambda_{1}=\left[{\begin{array}[]{*{20}c}0&1&0\\\ 1&0&0\\\ 0&0&0\\\ \end{array}}\right],\hskip 25.29494pt\lambda_{2}=\left[{\begin{array}[]{*{20}c}0&-i&0\\\ i&0&0\\\ 0&0&0\\\ \end{array}}\right],\hskip 25.29494pt\lambda_{3}=\left[{\begin{array}[]{*{20}c}1&0&0\\\ 0&-1&0\\\ 0&0&0\\\ \end{array}}\right],$ $\lambda_{4}=\left[{\begin{array}[]{*{20}c}0&0&1\\\ 0&0&0\\\ 1&0&0\\\ \end{array}}\right],\hskip 25.29494pt\lambda_{5}=\left[{\begin{array}[]{*{20}c}0&0&-i\\\ 0&0&0\\\ i&0&0\\\ \end{array}}\right],\hskip 25.29494pt\lambda_{6}=\left[{\begin{array}[]{*{20}c}0&0&0\\\ 0&0&1\\\ 0&1&0\\\ \end{array}}\right],$ $\lambda_{7}=\left[{\begin{array}[]{*{20}c}0&0&0\\\ 0&0&-i\\\ 0&i&0\\\ \end{array}}\right],\hskip 25.29494pt\lambda_{8}=\frac{1}{\sqrt{3}}\left[{\begin{array}[]{*{20}c}1&0&0\\\ 0&1&0\\\ 0&0&-2\\\ \end{array}}\right].\hfill(2)$ These matrices follow the following commutation and anti-commutation relations $[\lambda_{i},\lambda_{j}]=2if_{ijk}\lambda_{k}$, $\\{\lambda_{i},\lambda_{j}\\}=\frac{4}{3}\delta_{ij}+2d_{ijk}\lambda_{k}$, (3) respectively, where $d_{ijk}$ and $f_{ijk}$ ($i,j=1,2,..8$) represent completely symmetric and completely antisymmetric structure constants which characterizes $SU(3)$ group [39]. It is customary to define the shift operators $T$, $U$ and $V$ spin as $T_{\pm}=\frac{1}{2}(\lambda_{1}\pm i\lambda_{2})$, $U_{\pm}=\frac{1}{2}(\lambda_{6}\pm i\lambda_{7})$, $V_{\pm}=\frac{1}{2}(\lambda_{4}\pm i\lambda_{5})$. (4) They satisfy the closed algebra $[U_{+},U_{-}]=U_{3}$, $[V_{+},V_{-}]=V_{3}$, $[T_{+},T_{-}]=T_{3}$, (5) $[T_{3},T_{\pm}]=\pm 2T_{\pm}$, $[T_{3},U_{\pm}]=\mp U_{\pm}$, $[T_{3},V_{\pm}]=\pm V_{\pm}$, $[V_{3},T_{\pm}]=\pm T_{\pm}$, $[V_{3},U_{\pm}]=\pm U_{\pm}$, $[V_{3},V_{\pm}]=\pm 2V_{\pm}$, $[U_{3},T_{\pm}]=\mp T_{\pm}$, $[U_{3},U_{\pm}]=\pm 2U_{\pm}$, $[U_{3},V_{\pm}]=\pm V_{\pm}$, $[T_{+},V_{-}]=-U_{-}$, $[T_{+},U_{+}]=V_{+}$, $[U_{+},V_{-}]=T_{-}$, $[T_{-},V_{+}]=U_{+}$, $[T_{-},U_{-}]=-V_{-}$, $[U_{-},V_{+}]=-T_{+}$, where the diagonal terms are $T_{3}=\lambda_{3}$, $U_{3}=(\sqrt{3}\lambda_{8}-\lambda_{3})/2$ and $V_{3}=(\sqrt{3}\lambda_{8}+\lambda_{3})/2$, respectively. The Hamiltonian of the semiclassical lambda model is given by ${\rm H^{\Lambda}}={\rm H_{I}^{\Lambda}}+{\rm H_{II}^{\Lambda}}$, (6a) where the unperturbed and interaction parts including the detuning terms are given by ${\rm H_{I}^{\Lambda}}=\hbar(\Omega_{1}-\omega_{1}-\omega_{2})V_{3}+\hbar(\Omega_{2}-\omega_{1}-\omega_{2}){\rm T}_{3}$, (6b) and ${\rm H_{II}^{\Lambda}}=\hbar(\Delta^{\Lambda}_{1}V_{3}+\Delta^{\Lambda}_{2}T_{3})+$ $\hbar\kappa_{1}(V_{+}\exp(-i\Omega_{1}t)+V_{-}\exp(i\Omega_{1}t))+\hbar\kappa_{2}(T_{+}\exp(-i\Omega_{2}t)+T_{-}\exp(i\Omega_{2}t))$, (6c) respectively. In Eq.(6), $\Omega_{i}$ ($i=1,2$) are the external frequencies of the bi-chromatic field, $\kappa_{i}$ are the coupling parameters and $\hbar\omega_{1}(=-{\rm E}_{1}),\hbar\omega_{2}(=-{\rm E}_{2}),\hbar(\omega_{2}+\omega_{1})({=\rm E}_{3})$ be the respective energies of the three levels. $\Delta^{\Lambda}_{1}=(2\omega_{1}+\omega_{2}-\Omega_{1})$ and $\Delta^{\Lambda}_{2}=(\omega_{1}+2\omega_{2}-\Omega_{2})$ represent the respective detuning from the bi-chromatic external frequencies as shown in Fig.1. Proceeding in the same way, the semiclassical vee type three-level system can be written as ${\rm H^{V}}={\rm H_{I}^{V}}+{\rm H_{II}^{V}},$ (7a) where ${\rm H_{I}^{V}}=\hbar(\Omega_{1}-\omega_{1}-\omega_{2})V_{3}+\hbar(\Omega_{2}-\omega_{1}-\omega_{2}){\rm U}_{3},$ (7b) and ${\rm H_{II}^{V}}=\hbar(\Delta^{V}_{1}V_{3}+\Delta^{V}_{2}U_{3})+$ $\hbar\kappa_{1}(V_{+}\exp(-i\Omega_{1}t)+V_{-}\exp(i\Omega_{1}t))+\hbar\kappa_{2}(U_{+}\exp(-i\Omega_{2}t)+U_{-}\exp(i\Omega_{2}t))$ (7c) where $\Delta^{V}_{1}=(2\omega_{1}+\omega_{2}-\Omega_{1})$ and $\Delta^{V}_{2}=(2\omega_{2}+\omega_{1}-\Omega_{2})$ be the detuning shown in Fig.2. Similarly the semiclassical cascade three-level model is given by ${\rm H^{\Xi}}={\rm H_{I}^{\Xi}}+{\rm H_{II}^{\Xi}}$, (8a) where ${\rm H_{I}^{\Xi}}=\hbar(\Omega_{1}+\omega_{2}-\omega_{1}){\rm U}_{3}+\hbar(\Omega_{2}+\omega_{1}-\omega_{2}){\rm T}_{3}$, (8b) and ${\rm H_{II}^{\Xi}}=\hbar(\Delta_{1}^{\Xi}U_{3}+\Delta_{2}^{\Xi}T_{3})+$ $\hbar\kappa_{1}(U_{+}\exp(-i\Omega_{1}t)+U_{-}\exp(i\Omega_{1}t))+\hbar\kappa_{2}(T_{+}\exp(-i\Omega_{2}t)+T_{-}\exp(i\Omega_{2}t))$ (8c) respectively with respective detuning $\Delta^{\Xi}_{1}=(2\omega_{1}-\omega_{2}-\Omega_{1})$ and $\Delta^{\Xi}_{2}=(2\omega_{2}-\omega_{1}-\Omega_{2})$. Taking the fields to be the quantized cavity fields, in the rotating wave approximation, the Hamiltonian of the quantized lambda configuration is given by ${\it H^{\Lambda}}=H_{I}^{\Lambda}+H_{II}^{\Lambda}$, (9a) where, ${\it H_{I}^{\Lambda}}=\hbar(\Omega_{2}-\omega_{1}-\omega_{2}){\rm T}_{3}+\hbar(\Omega_{1}-\omega_{1}-\omega_{2})V_{3}+\sum\limits_{j=1}^{2}\Omega_{j}{a_{j}^{{\dagger}}a_{j}}$, (9b) ${\it H_{II}^{\Lambda}}=\hbar\Delta_{1}^{\Lambda}V_{3}+\hbar\Delta_{2}^{\Lambda}{\rm T}_{3}+\hbar g_{1}(V_{+}a_{1}+V_{-}a_{1}^{\dagger})+\hbar g_{2}(T_{+}a_{2}+T_{-}a_{2}^{\dagger}),$ (9c) where $a^{\dagger}_{i}$ and $a_{i}$ ($i=1,2$) be the creation and annihilation operators of the cavity modes, $g_{i}$ be the coupling constants and $\Omega_{i}$ be the mode frequencies. Proceeding in the similar pattern, the Hamiltonian of the quantized vee system is given by ${\it H^{V}}=H_{I}^{V}+H_{II}^{V}$, (10a) where, ${\it H_{I}^{V}}=\hbar(\Omega_{2}-\omega_{1}-\omega_{2}){\rm U}_{3}+\hbar(\Omega_{1}-\omega_{1}-\omega_{2})V_{3}+\sum\limits_{j=1}^{2}\Omega_{j}{a_{j}^{{\dagger}}a_{j}}$ (10b) ${\it H_{II}^{V}}=\hbar\Delta_{1}^{V}V_{3}+\hbar\Delta_{2}^{V}{\rm U}_{3}+\hbar g_{1}(V_{+}a_{1}+V_{-}a_{1}^{\dagger})+\hbar g_{2}(U_{+}a_{2}+U_{-}a_{2}^{\dagger}),$ (10c) respectively. Similarly the Hamiltonian of the quantized cascade system reads ${\it H^{\Xi}}=H_{I}^{\Xi}+H_{II}^{\Xi}$, (11a) where ${\it H_{I}^{\Xi}}=\hbar(\Omega_{2}-\omega_{1}-\omega_{2}){\rm T}_{3}+\hbar(\Omega_{1}-\omega_{1}-\omega_{2})U_{3}+\sum\limits_{j=1}^{2}\Omega_{j}{a_{j}^{{\dagger}}a_{j}}$, (11b) ${\it H_{II}^{\Xi}}=\hbar\Delta_{1}^{\Xi}U_{3}+\hbar\Delta_{2}^{\Xi}{\rm T}_{3}+\hbar g_{1}(U_{+}a_{1}+U_{-}a_{1}^{\dagger})+\hbar g_{2}(T_{+}a_{2}+T_{-}a_{2}^{\dagger}).$ (11c) Using the algebra given in Eq.(5) and that of field operators, it is easy to check that $[{\it H_{I}^{i}},{\it H_{II}^{i}}]=0$ for $\Delta^{i}_{1}=-\Delta^{i}_{2}$ ($i=\Lambda$ and $V$) for the lambda and vee model and $\Delta^{\Xi}_{1}=\Delta^{\Xi}_{2}$ for the cascade model which are identified as the two photon resonance condition and equal detuning conditions, respectively [18,21,22,24,26]. This ensures that each piece of the Hamiltonian has the simultaneous eigen functions. Thus we note that, unlike Ref.[18,21,22], precise formulation of the aforementioned three-level configurations require the use of a subset of Gell-Mann $\lambda_{i}$ matrices rather than the use of all matrices. We now proceed to solve the lambda and vee configurations for the classical and the quantized field separately. III.The semiclassical lambda system At zero detuning the Hamiltonian of the lambda type three-level system is given by $H^{\Lambda}=\left[{\begin{array}[]{*{20}c}\hbar(\omega_{1}+\omega_{2})&\hbar\kappa_{2}\exp[-i\Omega_{2}t]&\hbar\kappa_{1}\exp[-i\Omega_{1}t]\\\ \hbar\kappa_{2}\exp[i\Omega_{2}t]&-\hbar\omega_{2}&0\\\ \hbar\kappa_{1}\exp[i\Omega_{1}t]&0&-\hbar\omega_{1}\\\ \end{array}}\right].\hfill(12)$ The solution of the Schrodinger equation corresponding to Hamiltonian (12) is given by $\Psi(t)=C_{1}(t)\left|1\right\rangle+C_{2}(t)\left|2\right\rangle+C_{3}(t)\left|3\right\rangle$ (13) where $C_{1}(t)$, $C_{2}(t)$ and $C_{3}(t)$ be the time-dependent normalized amplitudes of the lower, middle and upper levels with the respective basis states, $\left|1\right\rangle=\left[{\begin{array}[]{*{20}c}0\\\ 0\\\ 1\\\ \end{array}}\right],$ $\left|2\right\rangle=\left[{\begin{array}[]{*{20}c}0\\\ 1\\\ 0\\\ \end{array}}\right],$ $\left|3\right\rangle=\left[{\begin{array}[]{*{20}c}1\\\ 0\\\ 0\\\ \end{array}}\right],$ (14) respectively. We now proceed to calculate the probability amplitudes of the three states. Substituting Eq.(13) in Schrödinger equation and equating the coefficients of $\left|2\right\rangle$ , $\left|3\right\rangle$ and $\left|1\right\rangle$ from both sides we obtain $i\frac{{\partial C_{3}}}{{\partial t}}=(\omega_{2}+\omega_{1})C_{3}+\kappa_{1}\exp(-i\Omega_{1}t)C_{1}+\kappa_{2}\exp(-i\Omega_{2}t)C_{2},$ (15a) $i\frac{{\partial C_{2}}}{{\partial t}}=-\omega_{2}C_{2}+\kappa_{2}\exp(i\Omega_{2}t)C_{3},$ (15b) $i\frac{{\partial C_{1}}}{{\partial t}}=-\omega_{1}C_{1}+\kappa_{1}\exp(i\Omega_{1}t)C_{3}.$ (15c) Let the solutions of Eqs.(15a-c) are of the following form, $C_{1}={\rm A}_{1}\exp(iS_{1}t),$ (16a) $C_{2}={\rm A}_{2}\exp(iS_{2}t),$ (16b) $C_{3}={\rm A}_{3}\exp(iS_{3}t),$ (16c) where $A_{i}$s’ are the time independent constants to be determined. Putting Eqs.(16a-c) in Eqs.(15a-c) we obtain $(S_{3}+\omega_{2}+\omega_{1}){\rm A}_{3}+\kappa_{2}{\rm A}_{2}+\kappa_{1}{\rm A}_{1}=0$, (17a) $(S_{3}+\Omega_{2}-\omega_{2}){\rm A}_{2}+\kappa_{2}{\rm A}_{3}=0$, (17b) $(S_{3}+\Omega_{1}-\omega_{1}){\rm A}_{1}+\kappa_{1}{\rm A}_{3}=0$. (17c) In deriving Eqs.(17), the time independence of the amplitudes ${\rm A}_{3}$, ${\rm A}_{2}$ and ${\rm A}_{1}$ are ensured by invoking the conditions $S_{2}=S_{3}+\Omega_{2}$ and $S_{1}=S_{3}+\Omega_{1}$. At resonance, we have $\Delta^{\Lambda}_{1}=0=-\Delta^{\Lambda}_{2}$ i.e, $(2\omega_{2}+\omega_{1})-\Omega_{2}=0=(\omega_{2}+2\omega_{1})-\Omega_{1}$ and the solution of Eq.(17) yields $S_{3}=-(\omega_{2}+\omega_{1})\pm\Delta$, (18a) $S_{3}=-(\omega_{2}+\omega_{1})$ (18b) where $\Delta=\sqrt{\kappa_{1}^{2}+\kappa_{2}^{2}}$ and we have three values of $S_{2}$ and $S_{1}$ namely $S_{2}^{1}=\omega_{2},S_{2}^{2,3}=\omega_{2}\pm\Delta$, (19a) $S_{1}^{1}=\omega_{1},S_{1}^{2,3}=\omega_{1}\pm\Delta$. (19b) Using Eqs.(18) and (19), Eq.(16) can be written as $C_{3}(t)={\rm A}_{3}^{1}\exp(-i(\omega_{2}+\omega_{1})t)$ $+{\rm A}_{3}^{2}\exp(i(-(\omega_{2}+\omega_{1})+\Delta)t)+{\rm A}_{3}^{3}\exp(i(-(\omega_{2}+\omega_{1})-\Delta)t)$, (20a) $C_{2}(t)={\rm A}_{2}^{1}\exp(i\omega_{2}t)+{\rm A}_{2}^{2}\exp(i(\omega_{2}+\Delta)t)+{\rm A}_{2}^{3}(i(\omega_{2}-\Delta)t)$, (20b) $C_{1}(t)={\rm A}_{1}^{1}\exp(i\omega_{1}t)+{\rm A}_{1}^{2}\exp(i(\omega_{1}+\Delta)t)+{\rm A}_{1}^{3}(i(\omega_{1}-\Delta)t)$, (20c) where $A_{i}$-s are the constants which can be calculated from the following initial conditions: Case-I: At $t=0$ let the atom is in level-1, i.e. $C_{1}(0)=1$, $C_{2}(0)=0$, $C_{3}(0)=0$. Using Eqns (15) and (20), the corresponding time-dependent probabilities of the three levels are $\left|{C_{3}(t)}\right|^{2}=\frac{{\kappa_{1}^{2}}}{{\Delta^{2}}}\sin^{2}\Delta t$, (21a) $\left|{C_{2}(t)}\right|^{2}=4\frac{{\kappa_{1}^{2}\kappa_{2}^{2}}}{{\Delta^{4}}}\sin^{4}\Delta t/2$, (21b) $\left|{C_{1}(t)}\right|^{2}=\frac{1}{{\Delta^{4}}}(\kappa_{2}^{2}+\kappa_{1}^{2}\cos\Delta t)^{2}$. (21c) Case-II: If the atom is initially in level-2, i.e. $C_{1}(0)=0$, $C_{2}(0)=1$ and $C_{3}(0)=0$, the probabilities of the three states are $\left|{C_{3}(t)}\right|^{2}=\frac{{\kappa_{2}^{2}}}{{\Delta^{2}}}\sin^{2}\Delta t$, (22a) $\left|{C_{2}(t)}\right|^{2}=\frac{1}{{\Delta^{4}}}(\kappa_{1}^{2}+\kappa_{2}^{2}\cos\Delta t)^{2}$, (22b) $\left|{C_{1}(t)}\right|^{2}=4\frac{{\kappa_{1}^{2}\kappa_{2}^{2}}}{{\Delta^{4}}}\sin^{4}\Delta t/2$. (22c) Case-III: When the atom is initially in level-3, i.e. $C_{1}(0)=0$, $C_{2}(0)=0$ and $C_{3}(0)=1$, the time evolution of the probabilities of the three states are $\left|{C_{3}(t)}\right|^{2}=\cos^{2}\Delta t$, (23a) $\left|{C_{2}(t)}\right|^{2}=\frac{{\kappa_{2}^{2}}}{{\Delta^{2}}}\sin^{2}\Delta t$, (23b) $\left|{C_{1}(t)}\right|^{2}=\frac{{\kappa_{1}^{2}}}{{\Delta^{2}}}\sin^{2}\Delta t$. (23c) We now proceed to solve the quantized version of the above model. IV. The quantized lambda system We now consider the three-level lambda system interacting with a bi-chromatic quantized fields described by the Hamiltonian Eq.(9). At zero detuning the solution of the Hamiltonian is given by $\left|{\Psi_{\Lambda}(t)}\right\rangle=\sum\limits_{n,m=0}^{\infty}{[C_{1}^{n-1,m+1}(t)\left|{n-1,m+1,1}\right\rangle+C_{2}^{n,m}(t)\left|{n,m,2}\right\rangle}+C_{3}^{n-1,m}(t)\left|{n-1,m,3}\right\rangle],$ (24) where $n$ and $m$ represent the photon number corresponding to two modes of the bi-chromatic fields. This interaction Hamiltonian that couples the atom- field states ${\left|{n-1,m,3}\right\rangle}$, ${\left|{n,m,2}\right\rangle}$ and ${\left|{n-1,m+1,1}\right\rangle}$ and forms the lambda configuration shown in Fig.1 is given by ${\it H_{II}^{\Lambda}}=\hbar\left[{\begin{array}[]{*{20}c}0&{g_{2}\sqrt{n}}&{g_{1}\sqrt{m+1}}\\\ {g_{2}\sqrt{n}}&0&0\\\ {g_{1}\sqrt{m+1}}&0&0\\\ \end{array}}\right].$ (25) The eigenvalues of the Hamiltonian are given by $\lambda_{\pm}=\pm\hbar\sqrt{ng_{2}^{2}+(m+1)g_{1}^{2}}\quad(=\pm\hbar\Omega_{nm})$ and $\lambda_{0}=0(=\Omega_{0})$, respectively with the corresponding dressed eigenstates $\left[{\begin{array}[]{*{20}c}{\left|{nm,3}\right\rangle}\\\ {\left|{nm,2}\right\rangle}\\\ {\left|{nm,1}\right\rangle}\\\ \end{array}}\right]=T_{n,m}(g_{1},g_{2})\left[{\begin{array}[]{*{20}c}{\left|{n-1,m,3}\right\rangle}\\\ {\left|{n,m,2}\right\rangle}\\\ {\left|{n-1,m+1,1}\right\rangle}\\\ \end{array}}\right].$ (26) In Eq.(26), the dressed states are constructed by rotating the bare states with the Euler matrix given by $T_{n,m}(g_{1},g_{2})=\left[{\begin{array}[]{*{20}c}{c_{3}c_{2}-c_{1}s_{2}s_{3}}&{c_{3}s_{2}-c_{1}c_{2}s_{3}}&{s_{3}s_{1}}\\\ {-s_{3}c_{2}-c_{1}s_{2}c_{3}}&{-s_{3}s_{2}+c_{1}c_{2}c_{3}}&{c_{3}s_{1}}\\\ {s_{1}s_{2}}&{-s_{1}c_{2}}&{c_{1}}\\\ \end{array}}\right]$ (27) where $s_{i}=\sin\theta_{i}$ and $c_{i}=\cos\theta_{i}$ ($i=1,2,3$). The elements of the matrix are found to $T_{n,m}(g_{1},g_{2})=\left[{\begin{array}[]{*{20}c}{\frac{1}{{\sqrt{2}}}}&{g_{2}\sqrt{\frac{{n}}{{2(ng_{2}^{2}+(m+1)g_{1}^{2})}}}}&{g_{1}\sqrt{\frac{{m+1}}{{2(ng_{2}^{2}+(m+1)g_{1}^{2})}}}}\\\ {0}&{g_{1}\sqrt{\frac{{m+1}}{{ng_{2}^{2}+(m+1)g_{1}^{2}}}}}&{-g_{2}\sqrt{\frac{{n}}{{ng_{2}^{2}+(m+1)g_{1}^{2}}}}}\\\ {-\frac{1}{{\sqrt{2}}}}&{g_{2}\sqrt{\frac{{n}}{{2(ng_{2}^{2}+(m+1)g_{1}^{2})}}}}&{g_{1}\sqrt{\frac{{m+1}}{{2(ng_{2}^{2}+(m+1)g_{1}^{2})}}}}\\\ \end{array}}\right],$ (28) with corresponding Euler angles, $\theta_{1}=\arccos[\frac{\sqrt{1+m}g_{1}}{\sqrt{2(1+m)g_{1}^{2}+2ng_{2}^{2}}}]$, $\theta_{2}=-\arccos[-\frac{\sqrt{n}g_{2}}{\sqrt{(1+m)g_{1}^{2}+2ng_{2}^{2}}}]$, $\theta_{3}=\arccos[-\frac{\sqrt{2n}g_{2}}{\sqrt{(1+m)g_{1}^{2}+2ng_{2}^{2}}}]$. (29) The time-dependent probability amplitudes of the three levels are given by $\left[{\begin{array}[]{*{20}c}{C_{3}^{n-1,m}(t)}\\\ {C_{2}^{n,m}(t)}\\\ {C_{1}^{n-1,m+1}(t)}\\\ \end{array}}\right]=T_{n,m}^{-1}(g_{1},g_{2})\left[{\begin{array}[]{*{20}c}{e^{-i\Omega_{nm}t}}&0&0\\\ 0&{e^{-i\Omega_{0}t}}&0\\\ 0&0&{e^{i\Omega_{nm}t}}\\\ \end{array}}\right]T_{n,m}(g_{1},g_{2})\left[{\begin{array}[]{*{20}c}{C_{3}^{n-1,m}(0)}\\\ {C_{2}^{n,m}(0)}\\\ {C_{1}^{n-1,m+1}(0)}\\\ \end{array}}\right].$ (30) Now similar to the semiclassical model the probabilities corresponding to different initial conditions are: Case-IV: When the atom is initially in level-1, i.e, $C_{1}^{n-1,m+1}=1$, $C_{2}^{n,m}=0$ and $C_{3}^{n-1,m}=0$, the time-dependent atomic populations of the three states are given by $\left|{C_{3}^{n-1,m}(t)}\right|^{2}=\frac{{(m+1)g_{1}^{2}}}{{\Omega_{nm}^{2}}}\sin^{2}\Omega_{nm}t$, (31a) $\left|{C_{2}^{n,m}(t)}\right|^{2}=4\frac{{g_{1}^{2}g_{2}^{2}n(m+1)}}{{\Omega_{nm}^{4}}}\sin^{4}\Omega_{nm}t/2$, (31b) $\left|{C_{1}^{n-1,m+1}(t)}\right|^{2}=\frac{1}{{\Omega_{nm}^{4}}}[ng_{2}^{2}+(m+1)g_{1}^{2}\cos\Omega_{nm}t]^{2}$. (31c) Case-V: When the atom is initially in level-2, i.e, $C_{1}^{n-1,m+1}=0$, $C_{2}^{n,m}=1$ and $C_{3}^{n-1,m}=0$, the probabilities of three states are $\left|{C_{3}^{n-1,m}(t)}\right|^{2}=\frac{{ng_{2}^{2}}}{{\Omega_{nm}^{2}}}\sin^{2}\Omega_{nm}t$, (32a) $\left|{C_{2}^{n,m}(t)}\right|^{2}=\frac{1}{{\Omega_{nm}^{4}}}[(m+1)g_{1}^{2}+ng_{2}^{2}\cos\Omega_{nm}t]^{2}$, (32b) $\left|{C_{1}^{n-1,m+1}(t)}\right|^{2}=4\frac{{g_{1}^{2}g_{2}^{2}n(m+1)}}{{\Omega_{nm}^{4}}}\sin^{4}\Omega_{mm}t/2$. (32c) Case-VI: If the atom is initially in level-3, then we have $C_{1}^{n-1,m+1}=0$, $C_{2}^{n,m}=0$ and $C_{3}^{n-1,m+1}=1$ and the corresponding probabilities are $\left|{C_{3}^{n-1,m}(t)}\right|^{2}=\cos^{2}\Omega_{nm}t$, (33a) $\left|{C_{2}^{n,m}(t)}\right|^{2}=\frac{{ng_{2}^{2}}}{{\Omega_{nm}^{2}}}\sin^{2}\Omega_{nm}t$, (33b) $\left|{C_{1}^{n-1,m+1}(t)}\right|^{2}=\frac{{(m+1)g_{1}^{2}}}{{\Omega_{nm}^{2}}}\sin^{2}\Omega_{nm}t$. (33c) We now proceed to evaluate the population oscillations of different levels of the vee system with similar initial conditions. V.The semiclassical vee system At zero detuning, the Hamiltonian of the semiclassical three-level vee system interacting with two-mode classical fields is given by $H^{V}=\left[{\begin{array}[]{*{20}c}\hbar\omega_{1}&0&\hbar\kappa_{1}\exp[-i\Omega_{1}t]\\\ 0&\hbar\omega_{2}&\hbar\kappa_{2}\exp[-i\Omega_{2}t]\\\ \hbar\kappa_{1}\exp[i\Omega_{1}t]&\hbar\kappa_{2}\exp[i\Omega_{2}t]&-\hbar(\omega_{1}+\omega_{2})\\\ \end{array}}\right].\hfill(34)$ Let the solution of the Schrodinger equation corresponding to Eq.(34) is given by $\Psi(t)=C_{1}(t)\left|1\right\rangle+C_{2}(t)\left|2\right\rangle+C_{3}(t)\left|3\right\rangle,$ (35) where $C_{1}(t)$, $C_{2}(t)$ and $C_{3}(t)$ are the time-dependent normalized amplitudes with the basis vectors defined in Eqs.(13). To calculate the probability amplitudes of three states, substituting Eq.(35) into the Schrödinger equation we obtain $i\frac{{\partial C_{3}}}{{\partial t}}=\omega_{1}C_{3}+\kappa_{1}\exp(-i\Omega_{1}t)C_{1}$, (36a) $i\frac{{\partial C_{2}}}{{\partial t}}=\omega_{2}C_{2}+\kappa_{2}\exp(-i\Omega_{2}t)C_{1}$, (36b) $i\frac{{\partial C_{1}}}{{\partial t}}=-(\omega_{1}+\omega_{2})C_{1}+\kappa_{2}\exp(i\Omega_{2}t)C_{2}+\kappa_{1}\exp(i\Omega_{1}t)C_{3}$. (36c) Let the solutions of Eqs.(36) are of the following form: $C_{3}(t)={\rm A}_{3}\exp(iS_{3}t),$ (37a) $C_{2}(t)={\rm A}_{2}\exp(iS_{2}t),$ (37b) $C_{1}(t)={\rm A}_{1}\exp(iS_{1}t),$ (37c) where $A_{i}$-s are the time independent constants to be determined from the boundary conditions. From Eq.(36) and Eq.(37) we obtain $(S_{1}-\Omega_{1}+\omega_{1}){\rm A}_{3}+\kappa_{1}{\rm A}_{1}=0$, (38a) $(S_{1}-\Omega_{2}+\omega_{2}){\rm A}_{2}+\kappa_{2}{\rm A}_{1}=0$, (38b) $(S_{1}-\omega_{2}-\omega_{1}){\rm A}_{1}+\kappa_{2}{\rm A}_{2}+\kappa_{1}{\rm A}_{3}=0$. (38c) In deriving Eqs.(38), the time independence of the amplitudes ${\rm A}_{3}$, ${\rm A}_{2}$ and ${\rm A}_{1}$ are ensured by invoking the conditions $S_{2}=S_{1}-\Omega_{2}$ and $S_{3}=S_{1}-\Omega_{1}$. At resonance, $\Delta^{V}_{1}=0=-\Delta^{V}_{2}$ i.e. $(2\omega_{2}+\omega_{1})-\Omega_{2}=0=(\omega_{2}+2\omega_{1})-\Omega_{1}$ and the solutions of Eq.(38) are given by $S_{1}=(\omega_{1}+\omega_{2})$ (39a) $S_{1}=(\omega_{1}+\omega_{2})\pm\Delta$ (39b) and we have three values of $S_{2}$ and $S_{3}$ $S_{2}^{1}=-\omega_{2},S_{2}^{2,3}=-\omega_{2}\pm\Delta$ (40a) $S_{3}^{1}=-\omega_{1},S_{3}^{2,3}=-\omega_{1}\pm\Delta$. (40b) Using Eqs.(39) and (40), Eqs. (37) can be written as $C_{3}(t)={\rm A}_{3}^{1}\exp(-i\omega_{1}t)+{\rm A}_{3}^{2}\exp(-i(\omega_{1}+\Delta)t)+{\rm A}_{3}^{3}(-i(\omega_{1}-\Delta)t)$, (41a) $C_{2}(t)={\rm A}_{2}^{1}\exp(-i\omega_{2}t)+{\rm A}_{2}^{2}\exp(-i(\omega_{2}+\Delta)t)+{\rm A}_{2}^{3}(-i(\omega_{2}-\Delta)t)$, (41b) $C_{1}(t)={\rm A}_{1}^{1}\exp(i(\omega_{2}+\omega_{1})t)$ $+{\rm A}_{1}^{2}\exp(i((\omega_{2}+\omega_{1})+\Delta)t)+{\rm A}_{1}^{3}\exp(i((\omega_{2}+\omega_{1})-\Delta)t)$, (41c), where $A_{i}$-s are the constants which are calculated below from the various initial conditions. Case-I: Let us consider initially at $t=0$, the atom is in level-1, i.e, $C_{1}(0)=1$, $C_{2}(0)=0$ and $C_{3}(0)=0$. Using Eqs. (36) and (41), the time dependent probabilities of the three levels are given by $\left|{C_{3}(t)}\right|^{2}=\frac{{\kappa_{1}^{2}}}{{\Delta^{2}}}\sin^{2}\Delta t$, (42a) $\left|{C_{2}(t)}\right|^{2}=\frac{{\kappa_{2}^{2}}}{{\Delta^{2}}}\sin^{2}\Delta t$, (42b) $\left|{C_{1}(t)}\right|^{2}=\cos^{2}\Delta t$. (42c) Case-II: If the atom is initially in level-2, i.e, $C_{1}(0)=0$, $C_{2}(0)=1$ and $C_{3}(0)=0$, the corresponding probabilities of the states are given by $\left|{C_{3}(t)}\right|^{2}=4\frac{{\kappa_{1}^{2}\kappa_{2}^{2}}}{{\Delta^{4}}}\sin^{4}\Delta t/2$, (43a) $\left|{C_{2}(t)}\right|^{2}=\frac{1}{{\Delta^{4}}}(\kappa_{1}^{2}+\kappa_{2}^{2}\cos\Delta t)^{2}$, (43b) $\left|{C_{1}(t)}\right|^{2}=\frac{{\kappa_{2}^{2}}}{{\Delta^{2}}}\sin^{2}\Delta t$. (43c) Case-III: When the atom is initially in level-3, i.e, $C_{1}(0)=0$, $C_{2}(0)=0$ and $C_{3}(0)=1$, we obtain the the occupation probabilities of the three states as follows: $\left|{C_{3}(t)}\right|^{2}=\frac{1}{{\Delta^{4}}}(\kappa_{2}^{2}+\kappa_{1}^{2}\cos\Delta t)^{2}$, (44a) $\left|{C_{2}(t)}\right|^{2}=4\frac{{\kappa_{1}^{2}\kappa_{2}^{2}}}{{\Delta^{4}}}\sin^{4}\Delta t/2$, (44b) $\left|{C_{1}(t)}\right|^{2}=\frac{{\kappa_{1}^{2}}}{{\Delta^{2}}}\sin^{2}\Delta t$. (44c) VI.The quantized vee system The eigenfunction of the quantized vee system described by the Hamiltonian in Eq.(10) is given by $\left|{\Psi_{V}(t)}\right\rangle=\sum\limits_{n,m=0}^{\infty}{[C_{1}^{n+1,m}(t)\left|{n+1,m,1}\right\rangle}+C_{2}^{n,m}(t)\left|{n,m,2}\right\rangle+C_{3}^{n+1,m-1}(t)\left|{n+1,m-1,3}\right\rangle].$ (45) Once again we note that the Hamiltonian couples the atom-field states ${\left|{n+1,m,1}\right\rangle}$, ${\left|{n,m,2}\right\rangle}$ and ${\left|{n+1,m-1,3}\right\rangle}$ forming vee configuration depicted in Fig.2. The interaction part of the Hamiltonian (45) can also be expressed in the matrix form ${\it H_{II}^{V}}=\hbar\left[{\begin{array}[]{*{20}c}0&0&{g_{1}\sqrt{m}}\\\ 0&0&{g_{2}\sqrt{n+1}}\\\ {g_{1}\sqrt{m}}&{g_{2}\sqrt{n+1}}&0\\\ \end{array}}\right],$ (46) and the corresponding eigenvalues are $\lambda_{\pm}=\pm\hbar\sqrt{mg_{1}^{2}+(n+1)g_{2}^{2}}\quad(=\pm\hbar\Omega_{nm})$ and $\lambda_{0}=0$ respectively. The dressed eigenstate is given by $\left[{\begin{array}[]{*{20}c}{\left|{nm,3}\right\rangle}\\\ {\left|{nm,2}\right\rangle}\\\ {\left|{nm,1}\right\rangle}\\\ \end{array}}\right]=T_{n,m}\left[{\begin{array}[]{*{20}c}{\left|{n+1,m-1,3}\right\rangle}\\\ {\left|{n,m,2}\right\rangle}\\\ {\left|{n+1,m,1}\right\rangle}\\\ \end{array}}\right],$ (47) the rotation matrix is found to be $T_{n,m}=\left[{\begin{array}[]{*{20}c}{g_{1}\sqrt{\frac{{m}}{{2((n+1)g_{2}^{2}+mg_{1}^{2})}}}}&{g_{2}\sqrt{\frac{{n+1}}{{2((n+1)g_{2}^{2}+mg_{1}^{2})}}}}&{\frac{1}{{\sqrt{2}}}}\\\ {-g_{2}\sqrt{\frac{{n+1}}{{(n+1)g_{2}^{2}+mg_{1}^{2}}}}}&{g_{1}\sqrt{\frac{{m}}{{(n+1)g_{2}^{2}+mg_{1}^{2}}}}}&{0}\\\ {-g_{1}\sqrt{\frac{{m}}{{2((n+1)g_{2}^{2}+mg_{1}^{2})}}}}&{-g_{2}\sqrt{\frac{{n+1}}{{2((n+1)g_{2}^{2}+mg_{1}^{2})}}}}&{\frac{1}{{\sqrt{2}}}}\\\ \end{array}}\right].$ (48) The straightforward evaluation yields the various Euler angles are $\theta_{1}=-\frac{\pi}{4}$, $\theta_{2}=\arccos[-\frac{\sqrt{n+1}g_{2}}{\sqrt{mg_{1}^{2}+(1+n)g_{2}^{2}}}]$, $\theta_{3}=-\frac{\pi}{2}$. (49) The time-dependent probability amplitudes of the three levels are given by $\left[{\begin{array}[]{*{20}c}{C_{3}^{n+1,m-1}(t)}\\\ {C_{2}^{n,m}(t)}\\\ {C_{1}^{n+1,m}(t)}\\\ \end{array}}\right]={T^{-1}_{n,m}}\left[{\begin{array}[]{*{20}c}{e^{-i\Omega_{nm}t}}&0&0\\\ 0&{e^{-i\Omega_{0}t}}&0\\\ 0&0&{e^{i\Omega_{nm}t}}\\\ \end{array}}\right]T_{n,m}\left[{\begin{array}[]{*{20}c}{C_{3}^{n+1,m-1}(0)}\\\ {C_{2}^{n,m}(0)}\\\ {C_{1}^{n+1,m}(0)}\\\ \end{array}}\right]$. (50) Once again we proceed to calculate the probabilities for different initial conditions. Case-IV: Here we consider initially the atom is in level-1 i.e, $C_{1}^{n+1,m}=1$, $C_{2}^{n,m}=0$ and $C_{3}^{n+1,m-1}=0$. Using Eqs.(49) and (50) the time-dependent probabilities of the three levels are given by $\left|{C_{3}^{n+1,m-1}(t)}\right|^{2}=\frac{{mg_{1}^{2}}}{{\Omega_{nm}^{2}}}\sin^{2}\Omega_{nm}t$, (51a) $\left|{C_{2}^{n,m}(t)}\right|^{2}=\frac{{(n+1)g_{2}^{2}}}{{\Omega_{nm}^{2}}}\sin^{2}\Omega_{nm}t$, (51b) $\left|{C_{1}^{n+1,m}(t)}\right|^{2}=\cos^{2}\Omega_{nm}t$. (51c) Case-V: If the atom is initially in level-2 i.e, $C_{3}^{n+1,m-1}=0$, $C_{2}^{n,m}=1$ and $C_{1}^{n+1,m}=0$, then corresponding probabilities are $\left|{C_{3}^{n+1,m-1}(t)}\right|^{2}=4\frac{{g_{2}^{2}g_{1}^{2}(n+1)(m)}}{{\Omega_{mn}^{4}}}\sin^{4}\Omega_{mn}t/2$, (52a) $\left|{C_{2}^{n,m}(t)}\right|^{2}=\frac{1}{{\Omega_{nm}^{4}}}[mg_{1}^{2}+(n+1)g_{2}^{2}\cos\Omega_{nm}t]^{2}$, (52b) $\left|{C_{1}^{n+1,m}(t)}\right|^{2}=\frac{{g_{2}^{2}(n+1)}}{{\Omega_{nm}^{2}}}\sin^{2}\Omega_{nm}t$. (52c) Case-VI: Finally if the atom is initially in level-3 i.e, $C_{1}^{n+1,m}=0$, $C_{2}^{n,m}=0$ and $C_{3}^{n+1,m-1}=1$, then $\left|{C_{3}^{n+1,m-1}(t)}\right|^{2}=\frac{1}{{\Omega_{nm}^{4}}}[mg_{1}^{2}\cos\Omega_{nm}t+(n+1)g_{2}^{2}]^{2}$, (53a) $\left|{C_{2}^{n,m}(t)}\right|^{2}=4\frac{{g_{2}^{2}g_{1}^{2}(n+1)(m)}}{{\Omega_{nm}^{4}}}\sin^{4}\Omega_{nm}t/2$, (53b) $\left|{C_{1}^{n+1,m}(t)}\right|^{2}=\frac{{mg_{1}^{2}}}{{\Omega_{nm}^{2}}}\sin^{2}\Omega_{nm}t$. (53c) Finally we note that for large values of n and m, Case-IV, V and VI become identical to Case-I, II and III, respectively. This precisely shows the validity of the Bohr’s correspondence principle indicating the consistency of our approach. VII.Numerical results and discussion Before going to show the numerical plots of the semiclassical and quantized lambda and vee systems, we first consider their analytical results. If we compare Case-I, II, III of both cases, we find that the probabilities in Case-I (Case-III)) of lambda system is the same as in Case-III (Case-I) of vee system except the populations of 1st and 3rd levels are interchanged. See Eqs.(21 $\&$ 44) and Eqs.(23 $\&$ 42) for detailed comparison. Also Case-II respective models are similar which is evident by comparing Eqs.(22 $\&$ 43). In contrast, for the quantized model, Case-IV (Case-VI) of the lambda system is no longer same as in Case-VI (Case-IV) of the vee system. This breaking of symmetry is evident by comparing the analytical results, Eqs.(31 $\&$ 53), Eqs.(32 $\&$ 52) and Eqs.(32 $\&$ 51) respectively. Unlike previous case, also Case-V both the models are different which is evident from Eqs.(22 $\&$ 43). In what follows, we compare the probabilities of the semiclassical and quantized lambda and vee systems respectively. Fig.3 and 4 show the plots of the probabilities $|C_{1}^{i}(t)|^{2}$ (blue line), $|C_{2}^{i}(t)|^{2}$ (green line) and $|C_{3}^{i}(t)|^{2}$ (red line) for the semiclassical lambda and vee models when the atom is initially at level-1 (Case-I), level-2 (Case- II) and level-3 (Case-III), respectively. The comparison of the plots shows that the pattern of the probability oscillation of the lambda system for Case-I shown in Fig.3a (Case-III in Fig.3c) is similar to that of Case-III shown in Fig.4c (Case-I in Fig.4a) of the vee system. More particularly we note that in all cases the oscillation of level-2 remains unchanged, while the oscillation of level-3 (level-1) of the lambda system for Case-I is identical to that of level-1 (level-3) of the vee system for Case-III. Furthermore, comparison of Fig.3b and Fig.4b for Case-II shows that the time evolution of the probabilities of level-2 of both systems also remains similar while those of level-3 and level-1 are interchanged. From the behaviour of the probability curve we can conclude that the lambda and vee configurations are essentially identical to each other as we can obtain one configuration from another simply by the inversion followed by the interchange of probabilities. For the quantized field, we first consider the time evolution of the probabilities taking the field is in a number state representation. In the number state representation, the vacuum Rabi oscillation corresponding to Case-IV, V and VI of the lambda and vee systems are shown in Fig.5 and Fig.6 respectively. We note that, unlike previous case, the Rabi oscillation for Case-IV shown in Fig.5a (Case-VI shown in Fig.5c) for the lambda model is no longer similar to Case-VI shown in Fig.6c (Case-IV shown in Fig.6a) for the vee model. Furthermore, we note that for Case-V, the oscillation patterns of Fig.5b is completely different from that of Fig.6b. In a word, for the quantized field, in contrast to the semiclassical case, the symmetry in the pattern of the vacuum Rabi oscillation in all cases is completely spoiled irrespective of the fact whether the system stays initially in any one of the three levels. The quantum origin of the breaking of the symmetric pattern of the Rabi oscillation is the following. We note that due to the appearance of the terms like $(n+1)$ or $(m+1)$, several elements in the probabilities given by Eqs.(31,32,33) for the lambda system and Eqs.(51,52,53) for the vee are non zero even at $m=0$ and $n=0$. We argue that the vacuum Rabi oscillation interferes with the probability oscillations of various levels and spoils their symmetric structure. Thus as a consequence of the vacuum fluctuation, the symmetry of probability amplitudes of the dressed states of both models formed by the coherent superposition of the bare states is also lost. In the other word, the invertibility between the lambda and vee models exhibited for the classical field disappears as the direct consequence of the quantization of the cavity modes. Finally we consider the lambda and vee models interacting with the bi- chromatic quantized fields which are in the coherent state. The coherently averaged probabilities of level-3, level-2 and level-1 are given by $\left\langle{P_{3}(t)}\right\rangle_{\Lambda}=\sum\limits_{n,m}{W_{n}}W_{m}\left|{C_{3}^{n-1,m}(t)}\right|^{2}$, (54a) $\left\langle{P_{2}(t)}\right\rangle_{\Lambda}=\sum\limits_{n,m}{W_{n}}W_{m}\left|{C_{2}^{n,m}(t)}\right|^{2}$, (54b) $\left\langle{P_{1}(t)}\right\rangle_{\Lambda}=\sum\limits_{n,m}{W_{n}}W_{m}\left|{C_{1}^{n-1,m+1}(t)}\right|^{2}$, (54c) for the lambda system and $\left\langle{P_{3}(t)}\right\rangle_{V}=\sum\limits_{n,m}{W_{n}}W_{m}\left|{C_{3}^{n+1,m-1}(t)}\right|^{2}$, (55a) $\left\langle{P_{2}(t)}\right\rangle_{V}=\sum\limits_{n,m}{W_{n}}W_{m}\left|{C_{2}^{n,m}(t)}\right|^{2}$, (55b) $\left\langle{P_{1}(t)}\right\rangle_{V}=\sum\limits_{n,m}{W_{n}}W_{m}\left|{C_{1}^{n+1,m}(t)}\right|^{2}$, (55c) for the vee system, where $W_{n}=\frac{1}{n!}\exp[-\bar{n}]{\bar{n}}^{n}$ and $W_{m}=\frac{1}{m!}\exp[-\bar{m}]{\bar{m}}^{m}$ with $\bar{n}$ and $\bar{m}$ be the mean photon numbers of the two quantized modes, respectively. Fig.7-9 display the numerical plots of Eq.(54) and Eq.(55) for Case-IV, V and VI respectively where the collapse and revival of the Rabi oscillation is clearly evident for large average photon numbers in both the fields. We note that in all cases, the collapse and revival of level-2 of both the systems are identical to each other. Furthermore, we note that the collapse and revival for lambda system initially in level-1 shown in Fig.7a, Fig.7b and Fig.7c (level-3 shown in Fig.9a, Fig.9b and Fig.9c) is the same as that of the vee system if it is initially in level-3 shown in Fig.7f, Fig.7e and Fig.7d (level-1 shown in Fig.9f, Fig.9e and Fig.9d) respectively. On the other hand, if the system is initially in level-2, the collapse and revival of the lambda systems shown in Fig.8a, Fig.8b and Fig.8c are identical to Fig.8f, Fig.8e and Fig.8d respectively for the vee system. This is precisely the situation what we obtained in case of the semiclassical model. Thus the symmetry broken in the case of the quantized model is restored back again indicating that the coherent state with large average photon number is very close to the classical state where the effect of field population in the vacuum level is almost zero. It is needless to say that the coherent state with very low average photon number in the field modes can not show the symmetric dynamics in lambda and vee systems. VIII.Conclusion This paper presents the explicit construction of the Hamiltonians of the lambda, vee and cascade type of three-level configurations from the Gell-Mann matrices of $SU(3)$ group and compares the exact solutions of the first two models with different initial conditions. It is shown that the Hamiltonians of different configurations of the three-level systems are different. We emphasize that there is a conceptual difference between our treatment and the existing approach by Hioe and Eberly [18,21,22]. These authors advocate the existence of different energy conditions which effectively leads to same cascade Hamiltonian ($h_{21}\neq 0$, $h_{32}\neq 0$ and $h_{31}=0$ in Eq.(1)) having similar spectral feature irrespective of the configuration. We justify our approach by noting the fact that the two-photon condition and the equal detuning condition is a natural outcome of our analysis. For the lambda and vee models, the transition probabilities of the three levels for different initial conditions are calculated while taking the atom interacting with the bi-chromatic classical and quantized field respectively. It is shown that due to the vacuum fluctuation, the inversion symmetry exhibited by the semiclassical models is completely destroyed. In other words, the dynamics for the semiclassical lambda system can be completely obtained from the knowledge of the vee system and vice versa while such recovery is not possible if the field modes are quantized. The symmetry is restored again when the field modes are in the coherent state with large average photon number. Such breaking of the symmetric pattern of the quantum Rabi oscillation is not observed in case of the two-level Jaynes-Cummings model and therefore it is essentially a nontrivial feature of the multi-level systems which is manifested if the number of levels exceeds two. This investigation is a part of our sequel studies of the symmetry breaking effect for the equidistant cascade three- level and equidistant cascade four-level systems respectively [36,37]. Following the scheme of constructing of the model Hamiltonians, it is easy to show that we have different eight dimensional Bloch equations and non-linear constants for different configurations of the three-level systems and these issues will be considered elsewhere [40]. The breaking of the inversion symmetry of the lambda and vee models as a direct effect of the field quantization is an intricate issue especially in context with future cavity experiments with the multilevel systems. Acknowledgement MRN thanks University Grants Commission and SS thanks Department of Science and Technology, New Delhi for partial financial support. We thank Dr T K Dey for discussions. SS is also thankful to S N Bose National Centre for Basic Sciences, Kolkata, for supporting his visit to the centre through the Visiting Associateship program. ## References * [1] A Joshi and S V Lawande, Current Science 82 816 (2002); ibid 82 958 (2002) * [2] L Hollberg, L Lugiato, A Oraevski, A sergienko and V Zadkov (Eds), J. Opt. B: Quan. Semiclass. Opt. 5 457 (2003) * [3] A Nielsen, I L Chuang, Quantum Computation (Cambridge University Press, Cambridge 2002) * [4] E T Jaynes and F W Cummings, Proc. IEEE, 51 89 (1963) * [5] G Rempe, H Walther and N Klien, Phys Rev Lett 58 353 (1987) * [6] R G Brewer and E L Hahn, Phys Rev A11 1641 (1975); P W Milloni and J H Eberly, J Chem Phys 68 1602 (1978), E M Belanov and I A Poluktov JETP 29 758 (1969); D Grischkowsky, M M T Loy and P F Liao, Phys Rev A12 2514 (1975) and references therein * [7] B Sobolewska, Opt Commun 19 185 (1976), C Cohen-Tannoudji and S Raynaud, J Phys B10 365 (1977) * [8] R M Whitley and C R Stroud Jr, Phys Rev A14 1498 (1976) * [9] E Arimondo, Coherent Population Trapping in Laser Spectroscopy, Prog in Optics XXXV Edited by E Wolf (Elsevier Science, Amsterdam, 1996) p257. * [10] C M Bowden and C C Sung, Phys Rev A18 1588 (1978) * [11] T Mossberg, A Flusberg, R Kachru and S R Hartman, Phys Rev Lett 39 1523 (1984); T W Mossberg and S R Hartman, Phys Rev A 39 1271 (1981) * [12] K Bergman, H Theuer and B W Shore, Rev Mod Phys 70 1003 (1998) * [13] R J Cook and H J Kimble, Phys Rev Lett 54 1023 (1985); R J Cook, Phys Scr, T21 49 (1988) * [14] B Misra and E C G Sudarshan, J Math Phys 18 756 (1977); C B Chiu, E C G Sudarshan and B Misra, Phys Rev D16 520 (1977); R J Cook, Phys Scr, T21 49 (1988) * [15] S E Harris, Phys Today 50 36 (1997) * [16] L V Lau, S E Harris, Z Dutton and C H Behroozi, Nature (London) 397 594 (1999) * [17] C C Gerry and J H Eberly, Phys Rev A 42 6805 (1990) * [18] F T Hioe and J H Eberly, Phys Rev A 25 2168 (1982) * [19] H I Yoo and J H Eberly, Phys Rep 118 239 (1985) * [20] B W Shore, P L Knight, J Mod Phy 40 1195 (1993) * [21] F T Hioe and J H Eberly, Phys Rev Lett 47 838 (1981) * [22] F T Hioe, Phys Rev A 28 879 (1983) * [23] X Li and N Bei, Phys. Lett A 101 169 (1984) * [24] J N Elgin, Phys Lett A80 140 (1980) * [25] R J Cook, B W Shore, Phys Rev A20, 539 (1979) * [26] F T Hioe and J H Eberly, Phys Rev 29 1164 (1984) * [27] B Buck, C V Sukumar, J Phys A17 877 (1984) * [28] T Nikajima, M Elk, Lambropoulos, Phys Rev A50 R913 (1994) * [29] Tak-San Ho and Shih-I Chu, Phys Rev A31 659 (1985) * [30] F Li, X Li, D L Lin and T F George, Phys Rev A40 5129 (1989) and references therein * [31] F T Hioe, Phys Lett A99 150 (1983) * [32] N V Kancheva, D Pushkarov and S Rashev, J Phys B: Mol Opt Phys 14 539 (1981) * [33] N V Vitanov, J Phys B: Mol Opt Phys 31 709 (1998) * [34] S Y Chu and D C Su, Phys Rev A25 3169 (2003) * [35] N N Bogolubov Jr, F L Kien and A S Shumovsky, Phys Lett 101A (1984) 201; ibid, 107A (1985) 173 * [36] M R Nath, S Sen and G Gangopadhyay, Pramana-J Phys 61 1089 (2003) [quant-ph:0711.3884] * [37] M R Nath, T K Dey, S Sen and G Gangopadhyay, Pramana-J Phys 70 141 (2008) [quant-ph:0712.2649v2] * [38] M S Scully and M O Zubairy, Quantum optics, (Cambridge University Press, Cambridge, 1997) p16 * [39] D B Litchenberg, Unitary Symmetry and Elementary Particles (Academic Press, New York 1970) * [40] M R Nath, S Sen and G Gangopadhyay, (In preparation) (100,90)(200,90)(210,89)[]$\Delta_{2}$(260,90)[]$|3;m,n-1>$(155,85)(170,40) (175,70)[]$\Omega_{2}$(100,40)(200,40) (250,40)[]$|2;m,n>$(100,80)(175,80)5 (150,85)(200,85)5 (110,100)(110,90) (110,70)(110,80) (195,100)(195,90) (195,75)(195,85) (120,10)(150,80) (140,20)[]$\Omega_{1}$(100,10)(200,10) (265,10)[]$|1;m+1,n-1>$(90,85)[]$\Delta_{1}$ Fig.1 : Lambda type transition (100,90)(200,90) (250,90)[]$|3;m-1,n+1>$(120,85)(140,10) (170,25)[]$\Omega_{2}$(100,40)(200,40) (250,40)[]$|2;m,n>$(100,85)(175,85)5 (140,35)(200,35)5 (145,10)(160,35) (120,60)[]$\Omega_{1}$(100,10)(200,10) (260,10)[]$|1;m,n+1>$(110,100)(110,90) (110,78)(110,85) (195,47)(195,40) (195,28)(195,35) (90,87)[]$\Delta_{1}$(210,39)[]$\Delta_{2}$ Fig.2 : Vee type transition [Fig.3]: The time evolution of the probabilities $|C_{1}(t)|^{2}$ (blue line), $|C_{2}(t)|^{2}$ (green line) and $|C_{3}(t)|^{2}$ (red line) of the semiclassical lambda system for Case-I, II and III respectively with values ${\bf\kappa_{1}=.2}$, ${\bf\kappa_{2}=.1}$. [Fig.4]: The time variation of the probabilities $|C_{1}(t)|^{2}$ (blue line), $|C_{2}(t)|^{2}$ (green line) and $|C_{3}(t)|^{2}$ (red line) of the semiclassical vee system for Case-I, II and III respectively with above values of ${\bf\kappa_{1}}$, ${\bf\kappa_{2}}$. [Fig.5]: The Rabi oscillation of the quantized lambda system when the fields are in the number state for Case-I, II and III, respectively with ${\bf g_{1}=.2}$, ${\bf g_{2}=.1}$, ${\bf n=1}$, ${\bf m=1}$. [Fig.6]: The Rabi oscillation of the quantized vee system when the fields are in the number states for Case-I, II and III, respectively for same values of $\bf g_{1}$, $\bf g_{2}$, $\bf n$, $\bf m$. [Fig.7]: Figs.7a-c display the time-dependent collapse and revival phenomenon of level-3, level-2 and level-1 of the lambda system for Case-IV, while Figs.7d-f show that of the level-3, level-2 and level-1 respectively for of Case-VI of the vee system taking the field modes are in coherent states with $\bar{n}=30$ and $\bar{m}=20$. [Fig.8]: Figs.8a-c display the time-dependent of collapse and revival of level-3, level-2 and level-1 of the lambda system for Case-V while Figs.8d-f show that of level-3, level-2 and level-1 of the vee system for Case-V with the same values of $\bar{n}$ and $\bar{m}$. as in Fig.7 [Fig.9]: Figs.9a-c display the time-dependent of collapse and revival of level-3, level-2 and level-1 of the lambda system for Case-VI while Figs.9d-f show that for level-3, level-2 and level-1 respectively for the vee system for Case-IV with the same values of $\bar{n}$ and $\bar{m}$ as in Fig.7.

arxiv-papers

2024-09-04T02:48:53.520986

0802.0451

# Qregularity and an Extension of Evans-Griffiths Criterion to Vector Bundles on Quadrics Edoardo Ballico and Francesco Malaspina Università di Trento 38050 Povo (TN), Italy e-mail: ballico@science.unitn.it Politecnico di Torino Corso Duca degli Abruzzi 24, 10129 Torino, Italy e-mail: alagalois@yahoo.it ###### Abstract Here we define the concept of Qregularity for coherent sheaves on quadrics. In this setting we prove analogs of some classical properties. We compare the Qregularity of coherent sheaves on $\hbox{${\cal Q}$}_{n}\subset\mathbb{P}^{n+1}$ with the Castelnuovo-Mumford regularity of their extension by zero in $\mathbb{P}^{n+1}$. We also classify the coherent sheaves with Qregularity $-\infty$. We use our notion of Qregularity in order to prove an extension of Evans-Griffiths criterion to vector bundles on Quadrics. In particular we get a new and simple proof of the Knörrer’s characterization of ACM bundles. ††Mathematics Subject Classification 2000: 14F05, 14J60. keywords: Spinor bundles; coherent sheaves on quadric hypersurfaces; Castelnuovo-Mumford regularity. ## 1 Introduction In chapter $14$ of [12] Mumford introduced the concept of regularity for a coherent sheaf on a projective space $\mathbb{P}^{n}$. Since then, Castelnuovo-Mumford regularity has become a fundamental invariant and was investigate by several people. Chipalkatti in [2] has generalized this notion on grassmannians and Hoffman and Wang in [8] on multiprojective spaces. Costa and Miró-Roig in [3] and [4] give a definition of regularity for coherent sheaves on $n$-dimensional smooth projective varieties with an $n$-block collection. The aim of this note is to introduce a very simple and natural concept of regularity (the Qregularity) on a quadric hypersurface. If we consider the following geometric collection on $\mathbb{P}^{n}$: $(\hbox{${\cal E}$}_{0},\dots,\hbox{${\cal E}$}_{n})=({\mathcal{O}}_{\mathbb{P}^{n}}(-n),{\mathcal{O}}_{\mathbb{P}^{n}}(-n+1),\dots,{\mathcal{O}}_{\mathbb{P}^{n}}),$ we have that a coherent sheaf F on $\mathbb{P}^{n}$ is said to be $m$-regular according to Castelnuovo-Mumford if for any $i=1,\dots,n$ $H^{i}(F(m)\otimes\hbox{${\cal E}$}_{n-i})=0.$ On $\hbox{${\cal Q}$}_{n}$ we consider the following $n$-block collection: $(\hbox{${\cal G}$}_{0},\dots,\hbox{${\cal G}$}_{n})=(\hbox{${\cal G}$}_{0},{\mathcal{O}}_{\hbox{${\cal Q}$}_{n}}(-n+1),\dots,{\mathcal{O}}_{\hbox{${\cal Q}$}_{n}}),$ where $\hbox{${\cal G}$}_{0}=(\Sigma(-n))$ if $n$ is odd or where $\hbox{${\cal G}$}_{0}=(\Sigma_{1}(-n),\Sigma_{2}(-n))$ if $n$ is even and $\Sigma_{*}$ are the spinor bundles (for generalities about spinor bundles see [13]). We say that a coherent sheaf $F$ is $m$-Qregular if for any $i=1,\dots,n$ $H^{i}(F(m)\otimes\hbox{${\cal G}$}_{n-i})=0.$ The interesting fact is that on $\hbox{${\cal Q}$}_{2}\cong\mathbb{P}^{1}\times\mathbb{P}^{1}$ our definition of $m$-Qregularity coincides with the definition of $(m,m)$-regularity on $\mathbb{P}^{1}\times\mathbb{P}^{1}$ by Hoffman and Wang (see [8]). So we use the results on $\mathbb{P}^{1}\times\mathbb{P}^{1}$ as the starting step in order to prove on $\hbox{${\cal Q}$}_{n}$, by induction on $n$, analogs of the classical properties on $\mathbb{P}^{n+1}$. Next we give some equivalent condition of Qregularity. We compare the Qregularity of coherent sheaves on $\hbox{${\cal Q}$}_{n}\subset\mathbb{P}^{n+1}$ with the Castelnuovo-Mumford regularity of their extension by zero in $\mathbb{P}^{n+1}$. We also classify the coherent sheaves with Qregularity $-\infty$ as those with finite support. The second aim to this paper is to apply our notion of Qregularity in order to investigate under what circumstances a vector bundle can be decomposed into a direct sum of line bundles. A well known result by Horrocks (see [9]) characterizes the vector bundles without intermediate cohomology on a projective space as direct sum of line bundles. This criterion fails on more general varieties. In fact there exist non-split vector bundles without intermediate cohomology. This bundles are called ACM bundles. On $\mathbb{P}^{n}$, Evans and Griffith (see [6]) have improved Horrocks’ criterion: ###### Theorem 1.1 (Evans-Griffith). Let $E$ be a vector bundle of rank $r$ on $\mathbb{P}^{n}$, $n\geq 2$, then ${\cal E}$ splits if and only if $\forall i=1,...,r-1$ $\forall k\in\mathbb{Z}$, $h^{i}(E(k))=0.$ On a quadric hypersurface $\hbox{${\cal Q}$}_{n}$ there is a theorem by Knörrer that classifies all the ACM bundles (see [10]) as direct sums of line bundles and spinor bundles (up to a twist). Ottaviani has generalized Horrocks criterion to quadrics and Grassmanniann giving cohomological splitting conditions for vector bundles (see [14]). Our main result is to extend Evans-Griffiths criterion to vector bundles on quadrics. We improve Knörrer’s theorem in the following way: ###### Theorem 1.2. Let $E$ be a rank $r$ vector bundle on $\hbox{${\cal Q}$}_{n}$. Then the following conditions are equivalents: 1. 1. $H^{i}_{*}(E)=0$ for any $i=1,\dots,r-1$ and $H^{n-1}_{*}(E)=0$ , 2. 2. $E$ is a direct sum of line bundles and spinor bundles with some twist. In particular we get a new and simple proof of the Knörrer’s characterization of ACM bundles. The hypothesis $H^{n-1}_{*}(E)=0$ does not appear in the Evans-Griffiths criterion on $\mathbb{P}^{n}$. On $\hbox{${\cal Q}$}_{n}$ it is necessary. In fact we can find many indecomposable bundles with $H^{1}_{*}(E)=\dots=H^{n-2}_{*}(E)=0$ but $H^{n-1}_{*}(E)\not=0$ (see [7] or [11]). Then we specialize to the case: rank $E=2$. We prove that if a Qregular rank $2$ bundle $E$ has $H^{1}(E(-2))=H^{1}(E(c_{1}))=0$, then it is a direct sum of line bundles and spinor bundles with some twist. In particular if $n>4$, $E\cong{\mathcal{O}}\oplus{\mathcal{O}}(c_{1})$. We work over an algebraically closed field with characteristic zero. We only need the characteristic zero assumption to prove Theorem 1.2 and Proposition 4.4 because in their proofs we will use Le Potier vanishing theorem. We thanks E. Arrondo. He showed us the connection between the Qregularity and the splitting criteria for vector bundles. ## 2 $m$-Qregular coherent sheaves: definition and properties Let us consider a smooth quadric hypersurface $\hbox{${\cal Q}$}_{n}$ in $\mathbb{P}^{n+1}$. We use the unified notation $\Sigma_{*}$ meaning that for even $n$ both the spinor bundles $\Sigma_{1}$ and $\Sigma_{2}$ are considered, and for $n$ odd, the spinor bundle $\Sigma$. We follow the notation of [3] so the spinor bundles are twisted by $1$ with respect to those of [13] ($\Sigma_{*}={\mathcal{S}}_{*}(1)$) ###### Definition 2.1. A coherent sheaf $F$ on $\hbox{${\cal Q}$}_{n}$ ($n\geq 2$) is said $m$-Qregular if $H^{i}(F(m-i))=0$ for $i=1\dots n-1$ and $H^{n}(F(m)\otimes\Sigma_{*}(-n))=0$. We will say Qregular in order to $0$-Qregular. ###### Remark 2.2. A coherent sheaf $F$ on $\hbox{${\cal Q}$}_{2}\cong\mathbb{P}^{1}\times\mathbb{P}^{1}$ is $m$-Qregular if $H^{1}(F(m-1,m-1))=0$ and $H^{2}(F(m,m)\otimes\Sigma_{*}(-2,-2))=0$. Since we have $\Sigma_{1}\cong{\mathcal{O}}(1,0)$ and $\Sigma_{2}\cong{\mathcal{O}}(0,1)$, our conditions become $H^{1}(F(m-1,m-1))=0$, $H^{2}(F(m-1,m-2))=0$ and $H^{2}(F(m-2,m-1))=0$. So the definition of $m$-Qregularity coincides with the definition of $(m,m)$-regularity on $\mathbb{P}^{1}\times\mathbb{P}^{1}$ by Hoffman and Wang (see [8]). ###### Proposition 2.3. Let $F$ be an $m$-Qregular coherent sheaf on $\hbox{${\cal Q}$}_{n}$ $(n\geq 2)$ then 1. 1. $F$ is $k$-Qregular for $k\geq m$. 2. 2. $H^{0}(F(k))$ is spanned by $H^{0}(F(k-1))\otimes H^{0}({\mathcal{O}}(1))$ if $k>m$. ###### Proof. We use induction on $n$: for $n=2$, $(1)$ comes from [8] Proposition $2.7.$ and $(2)$ from [8] Proposition $2.8.$ Let us study the case $n=3$. Let $F$ be an $m$-Qregular coherent sheaf on $\hbox{${\cal Q}$}_{3}$. We have the following exact sequence for any integer $k$: $0\rightarrow F(k-1)\rightarrow F(k)\rightarrow F_{|\hbox{${\cal Q}$}_{2}}(k,k)\rightarrow 0.$ In particular we get $0=H^{1}(F(m-1))\rightarrow H^{1}(F_{|\hbox{${\cal Q}$}_{2}}(m-1,m-1))\rightarrow H^{2}(F(m-2))=0.$ This implies that $H^{1}(F_{|\hbox{${\cal Q}$}_{2}}(m-1,m-1))=0$. Let us consider now the exact sequence on $\hbox{${\cal Q}$}_{3}$: $0\rightarrow F(m-2)\otimes\Sigma(-1)\rightarrow F(m-2)^{4}\rightarrow F(m-2)\otimes\Sigma\rightarrow 0.$ Since $H^{2}(F(m-2))=0$ and $H^{3}(F(m)\otimes\Sigma(-3))=0$, we also have $H^{2}(F(m)\otimes\Sigma(-2))=0$. If we tensorize by $F(m)$ the exact sequence $0\rightarrow\Sigma(-3)\rightarrow\Sigma(-2)\rightarrow\Sigma_{|\hbox{${\cal Q}$}_{2}}(-2,-2)\rightarrow 0$ we get $0\rightarrow F(m)\otimes\Sigma(-3)\rightarrow F(m)\otimes\Sigma(-2)\rightarrow(F_{|\hbox{${\cal Q}$}_{2}}(m)\otimes\Sigma_{1}(-2,-2))\oplus(F_{|\hbox{${\cal Q}$}_{2}}(m)\otimes\Sigma_{2}(-2,-2))\rightarrow 0.$ So we obtain $H^{2}(F(m)\otimes\Sigma(-2))\rightarrow H^{2}(F(m)\otimes\Sigma_{1}(-2,-2))\oplus H^{2}(F(m)\otimes\Sigma_{2}(-2,-2))\rightarrow H^{3}(F(m)\otimes\Sigma(-3))$ and hence $H^{2}(F_{|\hbox{${\cal Q}$}_{2}}(m)\otimes\Sigma_{1}(-2,-2))=H^{2}(F_{|\hbox{${\cal Q}$}_{2}}(m)\otimes\Sigma_{2}(-2,-2))=0$. We can conclude that if $F$ is $m$-Qregular on $\hbox{${\cal Q}$}_{3}$, then $F_{|\hbox{${\cal Q}$}_{2}}$ is an $m$-Qregular on $\hbox{${\cal Q}$}_{2}$. We have hence $(1)$ and $(2)$ for $F_{|\hbox{${\cal Q}$}_{2}}$. Moreover from the exact sequence on $\hbox{${\cal Q}$}_{2}$: $0\rightarrow F_{|\hbox{${\cal Q}$}_{2}}(m-1,m-1)\otimes\Sigma_{1}(-1,-1)\rightarrow F_{|\hbox{${\cal Q}$}_{2}}(m-1,m-1)^{2}\rightarrow F_{|\hbox{${\cal Q}$}_{2}}(m-1,m-1)\otimes\Sigma_{2}\rightarrow 0,$ we get $H^{2}(F_{|\hbox{${\cal Q}$}_{2}}(m,m)\otimes\Sigma_{1}(-2,-2))\rightarrow H^{2}(F_{|\hbox{${\cal Q}$}_{2}}(m-1,m-1))^{2}\rightarrow H^{2}(F_{|\hbox{${\cal Q}$}_{2}}(m+1,m+1)\otimes\Sigma_{2}(-2,-2)).$ By $(1)$ for $F_{|\hbox{${\cal Q}$}_{2}}$, the last group is $0$; by $m$-Qregularity the first group is $0$. Therefore $H^{2}(F_{|\hbox{${\cal Q}$}_{2}}(m-1,m-1))=0$. Let us consider now $H^{i}(F(m-i))\rightarrow H^{i}(F(m+1-i))\rightarrow H^{i}(F_{|\hbox{${\cal Q}$}_{2}}(m+1-i,m+1-i)).$ By $(1)$ for $F_{|\hbox{${\cal Q}$}_{2}}$ if $i=1$, or by the above argument if $i=2$, the last group is $0$; by $m$-Qregularity , the first is $0$ if $i=1,2$. Therefore $H^{1}(F(m+1-1))=H^{2}(F(m+1-2))=0$. Moreover from the sequence $H^{3}(F(m)\otimes\Sigma(-3))\rightarrow H^{3}(F(m+1)\otimes\Sigma(-3))\rightarrow 0$ we see that also $H^{3}(F(m+1)\otimes\Sigma(-3))=0$. We can conclude that $F$ is $(m+1)$-Qregular. Continuing in this way we prove $(1)$ for $F$. To get $(2)$ we consider the following diagramm: $\begin{array}[]{ccc}H^{0}(F(k-1))\otimes H^{0}({\mathcal{O}}_{\hbox{${\cal Q}$}_{3}}(1))&\xrightarrow{\sigma}&H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k-1,k-1))\otimes H^{0}({\mathcal{O}}_{\hbox{${\cal Q}$}_{2}}(1,1))\\\ \downarrow\scriptsize{\mu}&&\downarrow\scriptsize{\tau}\\\ H^{0}(F(k))&\xrightarrow{\nu}&H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k,k))\par\end{array}$ Note that $\sigma$ is surjective if $k>m$ because $H^{1}(F(k-2))=0$. Let us prove that if $k>m$ then $\tau$ is surjective: From [8] Proposition $2.8.$ we know that $H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k,k))$ is spanned by $H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k-1,k))\otimes H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(1,0))$ and also by $H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k,k-1))\otimes H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(0,1))$ so the maps $H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k-1,k-1))\otimes H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(1,0))\otimes H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(0,1))\rightarrow H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k-1,k))\otimes H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(1,0))$ and $H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k-1,k))\otimes H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(1,0))\rightarrow H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k,k))$ are both surjective. Hence their composition is surjective. Now since we have the surjection $H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(1,0))\otimes H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(0,1))\rightarrow H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(1,1))$ we have that also $H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k-1,k-1))\otimes H^{0}({\mathcal{O}}_{|\hbox{${\cal Q}$}_{2}}(1,1))\xrightarrow{\tau}H^{0}(F_{|\hbox{${\cal Q}$}_{2}}(k,k)),$ is a surjection. Since both $\sigma$ and $\tau$ are surjective, we can see as in [12] page $100$ that $\mu$ is also surjective. Let assume $(1)$ and $(2)$ on $\hbox{${\cal Q}$}_{2n-1}$. We prove it on $\hbox{${\cal Q}$}_{2n}$. Let $F$ be an $m$-Qregular coherent sheaf on $\hbox{${\cal Q}$}_{2n}$. We have the following exact sequence for any integer $k$: $0\rightarrow F(k-1)\rightarrow F(k)\rightarrow F_{|Q_{2n-1}}(k)\rightarrow 0.$ In particular we get $H^{i}(F(m-i))\rightarrow H^{i}(F_{|Q_{2n-1}}(m-i))\rightarrow H^{i+1}(F(m-i-1)).$ This implies that $H^{i}(F_{|\hbox{${\cal Q}$}_{2n-1}}(m-i))=0$ for $i=1,\dots,2n-2$. Let us consider now the exact sequences on $\hbox{${\cal Q}$}_{2n}$: $0\rightarrow F(m-2n+1)\otimes\Sigma_{1}(-1)\rightarrow F(m-2n+1)^{2^{n}}\rightarrow F(m-2n+1)\otimes\Sigma_{2}\rightarrow 0$ and $0\rightarrow F(m-2n+1)\otimes\Sigma_{2}(-1)\rightarrow F(m-2n+1)^{2^{n}}\rightarrow F(m-2n+1)\otimes\Sigma_{1}\rightarrow 0.$ Since $H^{2n-1}(F(m-2n+1))=0$ and $H^{2n}(F(m)\otimes\Sigma_{*}(-2n))=0$, we also have $H^{2n-1}(F(m)\otimes\Sigma_{*}(-2n+1))=0$. If we tensorize by $F(m)$ the exact sequence $0\rightarrow\Sigma_{1}(-2n)\rightarrow\Sigma_{1}(-2n+1)\rightarrow{\Sigma_{1}}_{|\hbox{${\cal Q}$}_{2n-1}}(-2n+1)\rightarrow 0$ we get $0\rightarrow F(m)\otimes\Sigma_{1}(-2n)\rightarrow F(m)\otimes\Sigma_{1}(-2n+1)\rightarrow F_{|\\\ Q_{2n-1}}(m)\otimes\Sigma_{1}(-2n+1)\rightarrow 0.$ So we obtain $H^{2n-1}(F(m)\otimes\Sigma_{1}(-2n+1))\rightarrow H^{2n-1}(F_{|\hbox{${\cal Q}$}_{2n-1}}(m)\otimes\Sigma_{1}(-2n+1))\rightarrow H^{2n}(F(m)\otimes\Sigma_{1}(-2n))$ and hence $H^{2n-1}(F_{|\\\ Q_{2n-1}}(m)\otimes\Sigma_{1}(-2n+1))=0$. We can conclude that if $F$ is an $m$-Qregular sheaf on $\hbox{${\cal Q}$}_{2n}$, then $F_{|\hbox{${\cal Q}$}_{2n-1}}$ is an $m$-Qregular on $\hbox{${\cal Q}$}_{2n-1}$. We have hence by the induction hypothesis $(1)$ and $(2)$ for $F_{|\hbox{${\cal Q}$}_{2n-1}}$. Moreover from the exact sequence on $\hbox{${\cal Q}$}_{2n-1}$: $0\rightarrow F_{|\hbox{${\cal Q}$}_{2n-1}}(m)\otimes\Sigma(-2n+1)\rightarrow F_{|\hbox{${\cal Q}$}_{2n-1}}(m-2n+2)^{2^{n}}\rightarrow F_{|\hbox{${\cal Q}$}_{2n-1}}(m+1)\otimes\Sigma(-2n+1)\rightarrow 0,$ we get $H^{2n-1}(F_{|\hbox{${\cal Q}$}_{2n-1}}(m)\otimes\Sigma(-2n+1))\rightarrow H^{2n-1}(F_{|\hbox{${\cal Q}$}_{2n-1}}(m-2n+2))^{2^{n}}\rightarrow$ $\rightarrow H^{2n-1}(F_{|\hbox{${\cal Q}$}_{2n-1}}(m+1)\otimes\Sigma(-2n+1)).$ By $(1)$ for $F_{|\hbox{${\cal Q}$}_{2n-1}}$, the last group is $0$; by $m$-Qregularity the first group is $0$. Therefore $H^{2n-1}(F_{|\hbox{${\cal Q}$}_{2n-1}}(m-2n+2))=0$. Let us consider now the exact sequence $H^{i}(F(m-i))\rightarrow H^{i}(F(m+1-i))\rightarrow H^{i}(F_{|\hbox{${\cal Q}$}_{2n-1}}(m+1-i)).$ By $(1)$ for $F_{|\\\ Q_{2n-1}}$ if $i=1,\dots 2n-2$, or by the above argument if $i=2n-1$, the last group is $0$; by $m$-Qregularity , the first is $0$ if $i=1,\dots 2n-1$. Therefore $H^{i}(F(m+1-i))=0$ for $i=1,\dots 2n-1$. Moreover from the exact sequence $H^{2n}(F(m)\otimes\Sigma(-2n))\rightarrow H^{2n}(F(m+1)\otimes\Sigma(-2n))\rightarrow 0$ we also see that $H^{2n}(F(m+1)\otimes\Sigma(-2n))=0$. We can conclude that $F$ is $(m+1)$-Qregular. Continuing in this way we prove $(1)$ for $F$. To get $(2)$ we consider the following diagramm: $\begin{array}[]{ccc}H^{0}(F(k-1))\otimes H^{0}({\mathcal{O}}_{\hbox{${\cal Q}$}_{2n}}(1))&\xrightarrow{\sigma}&H^{0}(F_{|\hbox{${\cal Q}$}_{2n-1}}(k-1))\otimes H^{0}({\mathcal{O}}_{\hbox{${\cal Q}$}_{2n-1}}(1))\\\ \downarrow\scriptsize{\mu}&&\downarrow\scriptsize{\tau}\\\ H^{0}(F(k))&\xrightarrow{\nu}&H^{0}(F_{|\hbox{${\cal Q}$}_{2n-1}}(k))\par\end{array}$ Note that $\sigma$ is surjective if $k>m$ because $H^{1}(F(k-2))=0$. Moreover if $k>m$ also $\tau$ is surjective by $(2)$ for $F_{|\hbox{${\cal Q}$}_{2n-1}}$. Since both $\sigma$ and $\tau$ are surjective we can see as in [12] page $100$ that $\mu$ is also surjective. In a very similar way we can prove $(1)$ and $(2)$ from $\hbox{${\cal Q}$}_{2n}$ to $\hbox{${\cal Q}$}_{2n+1}$. ∎ We can give some equivalent definition of $m$-Qregular coherent sheaves: ###### Proposition 2.4. Let $F$ be a coherent sheaf on $\hbox{${\cal Q}$}_{n}$. The following conditions are equivalent: 1. 1. $F$ is $m$-Qregular. 2. 2. $H^{i}(F(m-i))=0$ for $i=1,\dots,n-1$, $H^{n-1}(F(m)\otimes\Sigma_{*}(-n+1))=0$ and $H^{n}(F(m-n+1))=0$. ###### Proof. We look at the exact sequences $0\rightarrow F(k)\otimes\Sigma_{*}(-1)\rightarrow F(k)^{2^{([n/2]+1)}}\rightarrow F(k)\otimes\Sigma_{*}\rightarrow 0.$ $(1)\Rightarrow(2).$ Let $F$ be $m$-Qregular then by (2.3) is also $(m+1)$-Qregular. We see that $H^{n}(F(m+1)\otimes\Sigma_{*}(-n-1))=H^{n}(F(m+1)\otimes\Sigma_{*}(-n))=0\Rightarrow H^{n}(F(m+1-n))=0$ and $H^{n}(F(m+1)\otimes\Sigma_{*}(-n-1))=H^{n-1}(F(m+1-n))=0\Rightarrow H^{n-1}(F(m+1)\otimes\Sigma_{*}(-n))=0.$ So we have $(2)$. $(2)\Rightarrow(1).$ Let $F$ be a coherent sheaf which satisfies ($2$). From $H^{n-1}(F(m)\otimes\Sigma_{*}(-n+1))=H^{n}(F(m-n+1))=0\Rightarrow H^{n}(F(m)\otimes\Sigma_{*}(-n))=0,$ we see that $F$ is $m$-regular. ∎ Now we show that the Qregular coherent sheaf are globally generated: ###### Proposition 2.5. Any Qregular coherent sheaf $F$ on $\hbox{${\cal Q}$}_{n}$ is globally generated. ###### Proof. We need to prove that the evaluation map $\varphi:H^{0}(F)\otimes{\mathcal{O}}_{Q_{n}}\to F$ is surjective. This is equivalent to prove that its tensor product with $id_{\Sigma}$ is surjective, because this would imply that $\varphi\otimes id_{\Sigma}\otimes id_{\Sigma^{\vee}}:H^{0}(F)\otimes\Sigma\otimes\Sigma^{\vee}\to F\otimes\Sigma\otimes\Sigma^{\vee}$ is surjective, and restricting to the component of endomorphims of $\Sigma$ of trace zero we get that$\varphi$ is surjective. We thus observe that we have a commutative diagram $\begin{matrix}H^{0}(F)\otimes H^{0}(\Sigma)\otimes{\mathcal{O}}_{Q_{n}}&\xrightarrow{\eta}&H^{0}(F\otimes\Sigma)\otimes{\mathcal{O}}_{Q_{n}}\\\ \downarrow&&\downarrow\psi\\\ H^{0}(F)\otimes\Sigma&\xrightarrow{\varphi\otimes id_{\Sigma}}&F\otimes\Sigma\end{matrix}$ so that it is enough to prove that $\eta$ and $\psi$ are surjective. This follows from Proposition 2.4 which implies that $H^{1}(F\otimes\Sigma_{*}(-1)=0$ (hence $\eta$ is surjective) and more generally $F\otimes\Sigma$ is regular, (and hence it is globally generated and $\psi$ is surjective).∎ ## 3 Qregularity on $\hbox{${\cal Q}$}_{n}$ ###### Definition 3.1. Let $F$ be a coherent sheaf on $\hbox{${\cal Q}$}_{n}$. We define the Qregularity of $F$, $Qreg(F)$, as the least integer $m$ such that $F$ is $m$-Qregular. We set $Qreg(F)=-\infty$ if there is no such integer. ###### Remark 3.2. On $\hbox{${\cal Q}$}_{n}$, we show that $Qreg({\mathcal{O}})=Qreg(\Sigma_{*})=0$. ${\mathcal{O}}$ and $\Sigma_{*}$ are ACM bundles. $H^{n}({\mathcal{O}}\otimes\Sigma_{*}(-n))\cong H^{0}(\Sigma^{\vee})=H^{0}(\Sigma(-1))=0$ and $H^{n}(\Sigma_{*}\otimes\Sigma_{*}(-n))\cong H^{0}(\Sigma_{*}\otimes\Sigma_{*}(-2))=0$. So ${\mathcal{O}}$ and $\Sigma_{*}$ are $0$-Qregular. $H^{n}({\mathcal{O}}(-1)\otimes\Sigma_{*}(-n))\cong H^{0}(\Sigma^{\vee}(1))=H^{0}(\Sigma)\not=0$ implies that ${\mathcal{O}}$ is not $(-1)$-Qregular. $H^{n}(\Sigma_{*}(-1)\otimes\Sigma_{*}(-n))\cong H^{0}(\Sigma^{\vee}_{*}(1)\otimes\Sigma_{*}(-1))$. Now from the exact sequence $0\rightarrow\Sigma_{*}^{\vee}\otimes\Sigma_{*}(-1)\rightarrow\Sigma_{*}^{\vee}\otimes{\mathcal{O}}^{2^{n}}\rightarrow\Sigma_{*}^{\vee}\otimes\Sigma_{*}\rightarrow 0,$ since $H^{0}(\Sigma_{*}^{\vee})=H^{1}(\Sigma_{*}^{\vee})$ we see that $H^{0}(\Sigma^{\vee}_{*}(1)\otimes\Sigma_{*}(-1))\cong H^{1}(\Sigma_{*}(-1)\otimes\Sigma_{*}^{\vee})$. But $H^{1}(\Sigma_{*}(-1)\otimes\Sigma_{*}^{\vee})\not=0$ by [5] Lemma $4.1.$ This means that $\Sigma_{*}$ is not $(-1)$-Qregular. ###### Remark 3.3. Let $0\rightarrow F_{1}\rightarrow F_{2}\rightarrow F_{3}\rightarrow 0$ be an exact sequence of coherent sheaves. Then $Qreg(F_{2})\leq\max\\{Qreg(F_{1}),Qreg(F_{3})\\}$. Let $F$ and $G$ be coherent sheaves. Then $Qreg(F\oplus G)=\max\\{Qreg(F),Qreg(G)\\}$. Let $F$ be a coherent sheaf on $\hbox{${\cal Q}$}_{n}$ ($n>1$) and let $i_{*}F$ be its extension by zero in the embedding $\hbox{${\cal Q}$}_{n}\hookrightarrow\mathbb{P}^{n+1}$. We can compare the Qregularity of $F$ with the regularity in the sense of Castelnuovo-Mumford of $i_{*}F$. We recall the definition: ###### Definition 3.4. A coherent sheaf $F$ on $\hbox{${\cal Q}$}_{n}$ is said $m$-regular in the sense of Castelnuovo-Mumford if $H^{i}(\mathbb{P}^{n+1},i_{*}F(m-i))=0$ for $i=1\dots n+1$. $Reg(F)$ is the least integer $m$ such that $F$ is $m$-regular. We set $reg(F)=-\infty$ if there is no such integer. ###### Proposition 3.5. Let $\hbox{${\cal Q}$}_{n}\hookrightarrow\mathbb{P}^{n+1}$ be a quadric hypersurface ($n>1$). For any coherent sheaf $F$ we have: $Qreg(F)\leq Reg(i_{*}F)\leq Qreg(F)+1.$ ###### Proof. We have to prove that: $F$ $m$-regular $\Rightarrow$ $F$ $m$-Qregular and $F$ $(m-1)$-Qregular $\Rightarrow$ $F$ $m$-regular. For any integer $t$ and for any $i>0$ we have: $H^{i}(\mathbb{P}^{n+1},i_{*}F(t))=H^{i}(\hbox{${\cal Q}$}_{n},F(t)),$ so $F$ is $m$-regular if and only if $H^{i}(\hbox{${\cal Q}$}_{n},F(m-i))=0$ for $i=1\dots n$. Let $F$ be $m$-regular we only need to prove that $H^{n}(\hbox{${\cal Q}$}_{n},F(m)\otimes\Sigma_{*}(-n))=0$. From the exact sequence $0\rightarrow F(m)\otimes\Sigma_{*}(-n-1)\rightarrow F(m-n)^{2^{([n/2]+1)}}\rightarrow F(m)\otimes\Sigma_{*}(-n)\rightarrow 0,$ we see that $H^{n}(\hbox{${\cal Q}$}_{n},F(m-n))=0\Rightarrow H^{n}(F(m)\otimes\Sigma_{*}(-n))=0$. Let $F$ be $(m-1)$-Qregular, by (2.3) $F$ is $m$-Qregular, so we only need to prove that $H^{n}(\hbox{${\cal Q}$}_{n},F(m-n))=0$. From the exact sequence $0\rightarrow F(m-1)\otimes\Sigma_{*}(-n)\rightarrow F(m-n)^{2^{([n/2]+1)}}\rightarrow F(m)\otimes\Sigma_{*}(-n)\rightarrow 0,$ we see that $H^{n}(F(m-1)\otimes\Sigma_{*}(-n))=H^{n}(F(m)\otimes\Sigma_{*}(-n))=0\Rightarrow H^{n}(\hbox{${\cal Q}$}_{n},F(m-n))=0$. ∎ ###### Remark 3.6. The above Proposition is optimal because for instance $Qreg({\mathcal{O}})=Qreg(\Sigma_{*})=0$ by Remark 3.2 but $Reg(i_{*}{\mathcal{O}})=1$ and $Reg(i_{*}\Sigma_{*})=0$ on $\hbox{${\cal Q}$}_{n}$ ($n>2$). In fact $H^{n}(\mathbb{P}^{n+1},i_{*}{\mathcal{O}}(t-n))=H^{n}(\hbox{${\cal Q}$}_{n},{\mathcal{O}}(t-n))=0$ if and only if $t\geq 1$ and $H^{n}(\mathbb{P}^{n+1},i_{*}\Sigma_{*}(t-n))=H^{n}(\hbox{${\cal Q}$}_{n},\Sigma_{*}(t-n))=0$ if and only if $t\geq 0$. We are ready to classify the coherent sheaves with Qregularity $-\infty$: ###### Theorem 3.7. Let $F$ be a coherent sheaf on $\hbox{${\cal Q}$}_{n}$ ($n$ even). The following condition are equivalent: 1. 1. $Qreg(F)=-\infty$. 2. 2. $Reg(F)=-\infty$. 3. 3. $\mbox{Supp}(F)$ is finite. Let $F$ be a coherent sheaf on $\hbox{${\cal Q}$}_{n}$ ($n$ odd). Let us consider the geometric collection on $\hbox{${\cal Q}$}_{n}$: $\sigma=({\mathcal{O}},\dots,{\mathcal{O}}(n-1),\Sigma(-n-1)).$ The following condition are equivalent: 1. 1. $Qreg(F)=-\infty$. 2. 2. $Reg(F)=-\infty$. 3. 3. $Reg_{\sigma}(F)=-\infty$. 4. 4. $\mbox{Supp}(F)$ is finite. ###### Proof. Let $n$ be an even integer and let $F$ be a coherent sheaf on $\hbox{${\cal Q}$}_{n}$. By (3.5) $Qreg(F)=-\infty$ if and only if $Reg(F)=-\infty$. By [1] Theorem $1.$ $Reg(F)=-\infty$ if and only if $\mbox{Supp}(F)$ is finite. Let $n$ be an odd integer and let $F$ be a coherent sheaf on $\hbox{${\cal Q}$}_{n}$. By (3.5) $Qreg(F)=-\infty$ if and only if $Reg(F)=-\infty$. By [3] Theorem $4.3.$ $Reg(F)=-\infty$ if and only if $Reg_{\sigma}(F)=-\infty$. By [1] Theorem $1.$ $Reg_{\sigma}(F)=-\infty$ if and only if $\mbox{Supp}(F)$ is finite. ∎ ## 4 Evans-Griffiths criterion on quadrics We use our notion of regularity in order to proving our main result: _Proof of Theorem 1.2._ Let assume that $E$ is Qregular but $E(-1)$ not. Here we use the definition of Qregularity as in Remark 2.4. Since $E$ is Qregular, it is globally generated and $E(1)$ is ample. So, by Le Potier vanishing theorem, we have that $H^{i}(E^{\vee}(-l))=0$ for every $l>0$ and $i=1,\dots,n-r$. So by Serre duality $H^{i}(E(-n+l))=0$ for every $l>0$ and $i=r,\dots,n-1$. In particular $H^{i}(E(-1-i))=0$ for $i=r,\dots,n-2$ and by hypothesis $H^{i}(E(-1-i))=0$ for $i=1,\dots,n-r$ and $H^{n-1}(E(-1-n+1)=0$. We can conclude that $E(-1)$ is not Qregular if and only if $H^{n-1}(E(-1)\otimes\Sigma_{*}(-n+1))\not=0$ or $H^{n}(E(-1-n+1))\not=0$. Let assume first that $H^{n}(E(-1-n+1))\not=0$, this means by Serre duality that $H^{0}(E^{\vee})\not=0$. We have a non zero map $f:E\rightarrow{\mathcal{O}}$ Now, since $E$ is globally generated, we have the exact sequence ${\mathcal{O}}\rightarrow{\mathcal{O}}\otimes H^{0}(E)\rightarrow E\rightarrow{\mathcal{O}}.$ The composition of the maps is not zero so must be the identity and we can conclude that ${\mathcal{O}}$ is a direct summand of $E$. Let assume now that $H^{n-1}(E(-1)\otimes\Sigma_{*}(-n+1))\not=0$ and $H^{n}(E(-1-n+1))=0$. Let see first the even case: let $n=2m$ and $H^{n-1}(E(-1)\otimes\Sigma_{1}(-n+1))\not=0$. We consider the following exact sequences: $0\rightarrow E(k)\otimes\Sigma_{2}(-1)\rightarrow E(k)^{2^{([n/2]+1)}}\rightarrow E(k)\otimes\Sigma_{1}\rightarrow 0.$ Since $H^{n}(E(-1-n+1))=H^{n-1}(E(-1-n+1))=0$, we see that $H^{n-1}(E(-1)\otimes\Sigma_{1}(-n+1))\cong H^{n}(E(-1)\otimes\Sigma_{2}(-n))$ so, by Serre duality $H^{0}(E^{\vee}(1)\otimes\Sigma^{\vee}_{2})\not=0$ and there exists a non zero map $f:E(-1)\rightarrow\Sigma^{\vee}_{2}.$ On the other hand, since for any $j=1,\dots,n-1$ $H^{j}(E(-1-j))=0$, the following maps $H^{0}(E\otimes\Sigma_{2}(-1))\rightarrow H^{1}(E\otimes\Sigma_{1}(-2))\rightarrow\dots\rightarrow H^{n-2}(E\otimes\Sigma_{2}(-n+1))\rightarrow H^{n-1}(E\otimes\Sigma_{1}(-n))$ are all surjective. In particular we have that $H^{0}(E(-1)\otimes\Sigma_{2})\not=0$ and there exists a non zero map $g:\Sigma^{\vee}_{2}\rightarrow E(-1).$ Let us consider the following commutative diagram: $\begin{array}[]{ccc}H^{n-1}(E(-1)\otimes\Sigma_{1}(-n+1))\otimes H^{1}(E^{\vee}(1)\otimes\Sigma^{\vee}_{1}(-1))&\xrightarrow{\sigma}&H^{n}(\Sigma_{1}^{\vee}(-1)\otimes\Sigma_{1}(-n+1))\cong\mathbb{C}\\\ \downarrow&&\downarrow\\\ H^{0}(E(-1)\otimes\Sigma_{2})\otimes H^{1}(E^{\vee}(1)\otimes\Sigma^{\vee}_{1}(-1))&\xrightarrow{\mu}&H^{1}(\Sigma_{1}^{\vee}(-1)\otimes\Sigma_{2})\cong\mathbb{C}\\\ \downarrow&&\downarrow\\\ H^{0}(E(-1)\otimes\Sigma_{2})\otimes H^{0}(E^{\vee}(1)\otimes\Sigma^{\vee}_{2})&\xrightarrow{\tau}&H^{0}(\Sigma_{2}^{\vee}\otimes\Sigma_{2})\cong\mathbb{C}\\\ \uparrow{\cong}&&\uparrow{\cong}\\\ {Hom}(\Sigma_{2}^{\vee},E(-1))\otimes{Hom}(E(-1),\Sigma_{2}^{\vee})&\xrightarrow{\gamma}&{Hom}(\Sigma_{2}^{\vee},\Sigma_{2}^{\vee})\end{array}$ The map $\sigma$ comes from Serre duality and it is not zero, the right vertical map are isomorphisms and the left vertical map are surjective so also the map $\tau$ is not zero. This means that the composition of the maps $f$ and $g$ is not zero so must be the identity and we can conclude that $\Sigma^{\vee}_{2}$ is a direct summand of $E(-1)$. By [13] Theorem $2.8.$ we have $\Sigma^{\vee}_{2}\cong\left\\{\begin{array}[]{cc}\Sigma_{2}(1)&\textrm{if $m\equiv 0$ (mod $4$)}\\\ \Sigma_{1}(1)&\textrm{if $m\equiv 2$ (mod $4$)}\end{array}\right.$ In the same way we can prove that, if $H^{n-1}(E(-1)\otimes\Sigma_{2}(-n+1))\not=0$, $\Sigma^{\vee}_{1}$ is a direct summand of $E(-1)$; or in the odd case that $\Sigma^{\vee}$ is a direct summand of $E(-1)$. By iterating these arguments we have that $E$ is a direct sum of line bundles and spinor bundles with some twist. As a Corollary we get the following splitting criterion: ###### Corollary 4.1. Let $E$ be a rank $r$ vector bundle on $\hbox{${\cal Q}$}_{n}$ such that $H^{i}_{*}(E)=0$ for any $i=1,\dots,r-1$, $H^{n-1}_{*}(E)=0$ and $H^{n-1}_{*}(E\otimes\Sigma_{*})=0$, then $E$ is a direct sum of line bundles. Let $E$ be a rank $r$ vector bundle on $\hbox{${\cal Q}$}_{n}$. Then the following conditions are equivalents: 1. 1. $H^{i}_{*}(E)=0$ for any $i=1,\dots,r-1$, $H^{n-1}_{*}(E)=0$ and $H^{n-1}_{*}(E\otimes\Sigma_{*})=0$, 2. 2. $E$ is a direct sum of line bundles with some twist. ###### Corollary 4.2 (Knörrer). Let $E$ be a rank $r$ vector bundle on $\hbox{${\cal Q}$}_{n}$ such that $H^{i}_{*}(E)=0$ for any $i=1,\dots,n-1$, then $E$ is a direct sum of line bundles and spinor bundles with some twist. ###### Remark 4.3. The hypothesis $H^{n-1}_{*}(E)=0$ does not appear in the Evans-Griffiths criterion on $\mathbb{P}^{n}$. On $\hbox{${\cal Q}$}_{n}$ it is necessary. In fact we can find many indecomposable bundles with $H^{1}_{*}(E)=\dots=H^{n-2}_{*}(E)=0$ but $H^{n-1}_{*}(E)\not=0$. For instance on $\hbox{${\cal Q}$}_{4}$ there is the rank $3$ bundle $P_{4}$ arising from the following exact sequence (see [7] or [11]): $0\to{\mathcal{O}}\rightarrow\Sigma_{1}\oplus\Sigma_{2}\rightarrow P_{4}\to 0.$ On $\hbox{${\cal Q}$}_{5}$ there is the rank $3$ bundle $P_{5}$ arising from the following exact sequence (see [11]): $0\to{\mathcal{O}}\rightarrow\Sigma\rightarrow P_{5}\to 0.$ For rank $2$ bundles, since $E^{\vee}\cong E(c_{1})$, the hypothesis $H^{n-1}_{*}(E)=0$ is not necessary. We can also prove the following result: ###### Proposition 4.4. Let $E$ be a rank $2$ bundle on $\hbox{${\cal Q}$}_{n}$ with $Qreg(E)=0$ and $H^{1}(E(-2))=H^{1}(E(c_{1}))=0$. Then $E$ is a direct sum of line bundles and spinor bundles with some twist. If $n>4$, $E\cong{\mathcal{O}}\oplus{\mathcal{O}}(c_{1})$. ###### Proof. Since $E$ is Qregular, it is globally generated and $E(1)$ is ample. So, by Le Potier vanishing theorem, we have that $H^{i}(E^{\vee}(-l))=0$ for every $l>0$ and $i=1,\dots,n-2$. So by Serre duality $H^{i}(E(-n+l))=0$ for every $l>0$ and $i=2,\dots,n-1$. In particular $H^{i}(E(-1-i))=0$ for $i=r,\dots,n-2$ and by hypothesis $H^{i}(E(-1-i))=0$ for $i=1,\dots,n-r$ and $H^{n-1}(E(-1-n+1))\cong H^{1}(E(c_{1}))=0$. We can conclude that $E(-1)$ is not Qregular if and only if $H^{n-1}(E(-1)\otimes\Sigma_{*}(-n+1))\not=0$ or $H^{n}(E(-1-n+1))\not=0$. Now arguing as in the above theorem we can conclude that $E$ contains ${\mathcal{O}}$ as a direct summand if $n>4$. If $n<5$ $E$, since the rank of the spinor bundles is smaller than $3$, can also contains $\Sigma_{*}$ as a direct summand. ∎ ## References * [1] E. Ballico and F. Malaspina, _ $n$-blocks collections on Fano manifolds and sheaves with regularity $-\infty$,_ arXiv:math/0710.3531 Matematiche (Catania) to appear. * [2] J. V. Chipalkatti, _A generalization of Castelnuovo regularity to Grassmann varieties,_ Manuscripta Math. 102 (2000), no. 4, 447–464. * [3] L. Costa and R. M. Miró-Roig, _Geometric collections and Castelnuovo-Mumford regularity,_ arXiv:math/0609561, Math. Proc. Cambridge Phil. Soc., to appear. * [4] L. Costa and R. M. Miró-Roig, _$m$ -blocks collections and Castelnuovo-Mumford regularity in multiprojective spaces,_ Nagoya Math. J. 186 (2007), 119–155. * [5] L. Costa and R.M. Miró-Roig, _Monads and regularity of vector bundles on projective varieties,_ 2007, Preprint. * [6] E.G. Evans, P. Griffiths, _The syzygy problem_ , 1981, Ann. of Math. 114(2), 323-333. * [7] R. Hernandez, I. Sols, _On a family of rank 3 bundles on $Gr(1,3)$_, 1985, J. reine angew Math. 360, 124-135. * [8] J. W. Hoffman and H. H. Wang, _Castelnuovo-Mumford regularity in biprojective spaces,_ Adv. Geom. 4 (2004), no. 4, 513–536. * [9] G. Horrocks, _Vector bundles on the punctured spectrum of a ring_ , 1964, Proc. London Math. Soc. (3) 14, 689-713. * [10] H. Knörrer, _Cohen-Macaulay modules of hypersurface singularities I_ , 1987, Invent. Math. 88, 153-164. * [11] F. Malaspina, _Monads and rank Three Vector Bundles on Quadrics_ , 2006, arXiv:math/0612515 . * [12] D. Mumford, _Lectures on curves on an algebraic surface, Princeton University Press,_ Princeton, N.J., 1966. * [13] G. Ottaviani, _Spinor bundles on Quadrics_ , 1988, Trans. Am. Math. Soc:, 307, no 1, 301-316. * [14] G. Ottaviani, _Some extension of Horrocks criterion to vector bundles on Grassmannians and quadrics_ , 1989, Annali Mat. Pura Appl. (IV) 155, 317-341.

arxiv-papers

2024-09-04T02:48:53.528540

0802.0454

# Test of a modified BCS theory performance in the Picket Fence Model V.Yu. Ponomarev and A.I. Vdovin Institut für Kernphysik, Technische Universität Darmstadt, D–64289 Darmstadt, Germany Bogoliubov Laboratory of Theoretical Physics, Joint Institute for Nuclear Research, 141980 Dubna, Russia ###### Abstract Analyses of a modified BCS (MBCS) theory performance at finite temperatures in the Picket Fence Model (PFM) for light and heavy systems is presented. Both symmetric, $\Omega=N$ ($N$ particles on $\Omega$ twice-degenerate levels), and asymmetric, $\Omega\neq N$, versions of the PFM are considered. The quantities known exactly from particle-hole symmetry of the $\Omega=N$ PFM are calculated. Starting from very low temperatures, these quantities are found in dramatic deviation from the exact values in MBCS results. Consequences of the MBCS prediction that heating generates a thermal constituent of the pairing gap, are discussed. Thermodynamical inconsistency of the MBCS is also addressed. PACS: 21.60.-n, 24.10.Pa, 24.60.-k, 24.60.Ky ## 1 Introduction A modified BCS (MBCS) theory for treating pairing correlations in atomic nuclei at finite temperatures [1, 2] has been recently tested [3] in the Picket Fence Model (PFM) in which $N$ particles are distributed over $\Omega$ twice-degenerate levels. The PFM with $N=\Omega$ is usually considered in the literature. The MBCS predicts a smooth decreasing behavior for a pairing gap as temperature $T$ increases up to some $T_{M}$ when the theory suddenly breaks down. It was reported [3, 4] that adding one extra level ($\Omega=N+1$) extends the MBCS applicability to much higher temperatures. As is already established [5], the $\Omega=N+1$ case of the PFM is the only example of an exceptional MBCS pairing gap behavior and thus, it could be hardly considered as a typical one. Accordingly, we find it necessary to provide the reader with other $\Omega\neq N$ PFM examples not available in [3]. This is done in Sec. 2. In Sec. 3, the MBCS prediction that not only the pairing force but also the heating itself generates the pairing gap, is discussed. In Sec. 4, we address the key question of the article: whether the MBCS is a reliable theory in the temperature domain where its pairing gap looks reasonable at first glance. For that, we examine some quantities which are known exactly in the $N=\Omega$ PFM because of symmetry. We show that in the MBCS predictions, these quantities dramatically deviate from their exact values starting from very low temperatures. Thermodynamical inconsistency of the MBCS is discussed in Sec. 5. ## 2 MBCS pairing gap in PFM systems with $\Omega=N$ and $\Omega\neq N$ The PFM or Richardson model is widely used as a test model for the pairing problem. It is the pairing Hamiltonian applied to a system of $N$ fermions distributed over $\Omega$ equidistant levels. All levels are twice degenerate for the spin up and down. The levels below (above) the Fermi surface will be referred to as holes (particles) and labeled by “$-i$” (“$i$”). Their single particle energies are $\varepsilon_{-i}=(0.5-i)$ MeV for holes and $\varepsilon_{i}=(-0.5+i)$ MeV for particles, where $i=1,~{}2,~{},\ldots$ (i.e. $\varepsilon_{i}=-\varepsilon_{-i}$). In all calculations presented below (except for the ones in Fig. 6a), the pairing strength parameter $G$ is adjusted so that the pairing gap $\Delta$ equals 1 MeV at zero temperature. In addition to the fact that exact solution of the PFM is possible if $N$ and $\Omega$ are not big, the PFM with $N=\Omega$ (to be referred to as the conventional PFM) possesses internal particle-hole symmetry. This means that at any temperature: a) the energy of the Fermi surface $E_{F}$ equals exactly 0 MeV: $E_{F}\equiv 0~{};$ (1) b) the quasiparticle energies $E_{i}$ for particles and holes should be degenerate: $E_{i}=\sqrt{(\varepsilon_{i}-E_{F})^{2}+\Delta^{2}}\equiv\sqrt{(\varepsilon_{-i}-E_{F})^{2}+\Delta^{2}}=E_{-i}$ (2) because of (1); c) the particle occupation probabilities (Bogoliubov $u_{i}$ and $v_{i}$ coefficients) are related as: $u_{i}=\sqrt{\frac{1}{2}\left(1+\frac{\varepsilon_{i}-E_{F}}{E_{i}}\right)}\equiv\sqrt{\frac{1}{2}\left(1-\frac{\varepsilon_{-i}-E_{F}}{E_{-i}}\right)}=v_{-i}$ (3) because of (1,2); d) the thermal quasiparticle occupation numbers $n_{i}=1/(1+\exp{(E_{i}/T)})$ and quasiparticle-number fluctuations $\delta N_{i}=\sqrt{n_{i}(1-n_{i})}$ should be equal for particles and holes with the same $i$: $n_{i}\equiv n_{-i}~{}~{}~{}~{}~{}{\rm and}~{}~{}~{}~{}~{}\delta N_{i}\equiv\delta N_{-i}~{}$ (4) because of (2). Of course, an asymmetric version of the PFM with $\Omega=N+k$ (where $k=-N/2+1,\ldots,-1,$ $1,\ldots,$ $\infty$) may be considered as well but Eqs. (1-4) are not valid for it. The first test of the MBCS performance in the conventional PFM with $N=\Omega=10$ revealed that at $T\approx 1.75$ MeV the system undertakes a phase transition which manifests itself as a sharp simultaneous increase in the pairing gap, a sharp decrease of the system energy and a discontinuity in the specific heat $C_{V}$ (this phase transition was defined as a superfluid – super-superfluid phase transition) [6]. This critical temperature was denoted as $T_{M}$ [3, 4]. It was found that $T_{M}$ linearly increases with the number of particles $N$ in the conventional PFM. It was also reported that enlarging the space by one more level, $\Omega=N+1$, restores the MBCS applicability to much higher temperatures even for $N\leq 14$ systems [3, 4]. Fig. 1 plots the MBCS pairing gap for $N=14$ particles and $\Omega=N+k$ levels with $k$ changing from $-4$ to 50 (solid curves). Indeed, the MBCS gap above $T_{c}$ is small and almost constant up to $T_{M}=7.2$ MeV when suggested example of extended configuration spaces, $k=1$, is considered. However, the MBCS theory breaks down again at rather modest temperature $T_{M}=2.15$ MeV with adding one more extra level, $k=2$, etc. Thus, the extension of the configuration space beyond $k=1$ makes the MBCS results inappropriate again. Figure 1: MBCS pairing gap $\Delta$ for the system of $N=14$ particles on $\Omega=N+k$ levels (solid curves). The BCS pairing gap is plotted by the dashed curve. The $k$ values are indicated in an oval at each curve. The conventional critical temperature $T_{c}$ is shown in this and all other figures by the vertical arrow. One immediately notices from Fig. 1 that the MBCS theory predicts two typical scenarios for the system evolution with heating. As temperature increases, the system undertakes either a) a superfluid – super-superfluid phase transition (examples with $k\leq 0$ in Fig. 1) or b) a phase transition from a superfluid phase with a positive gap to another superfluid phase but with a negative gap (examples with $k>1$ in Fig. 1), instead of a superfluid – normal phase transition of the conventional BCS. Corresponding BCS pairing gap behavior in the same systems is shown in Fig. 1 by the dashed curve. It is impossible to visually distinguish the results when $k$ value changes from $-4$ to 50. We have performed additional calculations for the asymmetric PFM with $N$ changing from 6 to 100. The results look very similar to the ones in Fig. 1. In all these examples, there exists only a single case $k=1$ with abnormally large $T_{M}$ which grows almost linearly with $N$. For all other $k$ values, the phase transitions of unknown types take place at much lower $T_{M}$ which becomes smaller and smaller as $k$ differs from $1$ larger and larger. However, only the case $\Omega=N+1$ has been selected for presentation in [3, 4] as an example of an extended configuration space ($\Omega>N$). ## 3 MBCS pairing gap: quantal and thermal constituents Although the MBCS performance has been already discussed in several articles, a clear answer to the question what is responsible for phase transitions of unknown types predicted by this theory is not yet established. According to the MBCS founders, there exists a criterion of the MBCS applicability, according to which the line shape of the quasiparticle-number fluctuations $\delta N_{i}$ should be symmetric [3, 4]. And the theory reaches the limit of its applicability at $T_{M}$. The physical origin of this criterion is unclear. The physical spectra are never symmetric, at least in nuclear physics. On the other hand, the single particle spectrum of the conventional PFM is ideally symmetric with respect to a chemical potential, see Eqs. (2,4), but the MBCS itself breaks down this symmetry (see below). The reader of [3] is suggested to notice from Fig. 2 that the line shape becomes rather asymmetric approaching $T_{M}$, thus violating the above- mentioned criterion of the MBCS applicability. We find it difficult to judge of the symmetry of lines in Figs. 2(a-e). On the other hand, it is very easy to quantify the results in this figure by plotting the $\delta N_{i}/\delta N_{-i}$ ratios as a function of temperature for different $i$. This is done in our Fig. 2 for different systems. Figure 2: MBCS predictions for the ratios of quasiparticle-number fluctuations $\delta N_{i}/\delta N_{-i}$ in different $(N,\Omega)$ systems. Exact result $\delta N_{i}/\delta N_{-i}\equiv 1$ in Figs. 2(a,c,d) is shown by the dotted line. Indeed, the asymmetry increases above $T_{M}$ in the $N=\Omega=10$ system, Fig. 2(a), while this does not in more heavy symmetric $N=\Omega$, Figs. 2(c,d), or selected asymmetric $\Omega=N+1$, Figs. 2(b), systems. However, it is difficult to miss that this effect is marginal compared to asymmetry in all systems (symmetric, asymmetric, light, and heavy) at much lower $T$ with the maximum around $T_{c}$. In other words, the most severe violation of the suggested criterion of the MBCS applicability takes place around the critical temperature of the conventional BCS and not at $T_{M}$ where the MBCS breaks down, as stated in [3, 4]. This conclusion does not depend on details of the PFM system being considered. Since the suggested criterion does not help to understand what causes the theory breaking down, we continue our analysis. Let us read once again the paragraph containing Eq. (26) in Ref. [3] that an important feature of the MBCS theory is that the MBCS gap is the sum of a quantal part (which looks the same as in the conventional BCS) and a thermal part $\delta\Delta$: $\displaystyle\delta\Delta_{h}$ $\displaystyle=G\sum_{i}^{\rm holes}(v_{i}^{2}-u_{i}^{2})\delta N_{i}$ $\displaystyle~{}~{}~{}{\rm for~{}holes}$ (5) $\displaystyle\delta\Delta_{p}$ $\displaystyle=G\sum_{i}^{\rm particles}(v_{i}^{2}-u_{i}^{2})\delta N_{i}$ $\displaystyle~{}~{}~{}{\rm for~{}particles}~{}.$ (6) Fig. 3 presents the MBCS gap for (a) light and (b) heavy PFM systems by the thick solid curve. The quantal part of it is shown by the thin solid curve, it becomes very small soon above $T_{c}$. The thermal part for holes (particles) is plotted by the dashed (dot-dashed) curve. The quantity $\delta\Delta_{h}$ is always positive ($v_{i}>u_{i}$) and the quantity $\delta\Delta_{p}$ is always negative ($u_{i}>v_{i}$); their absolute values increase with temperature. Figure 3: MBCS pairing gap (thick solid curve) for the PFM systems with (a) $N=14$, $\Omega=15$ and (b) $N=\Omega=100$ and in calculation with a realistic single particle spectrum for 120Sn: (c) neutron and (d) proton systems. This gap is the sum of a quantal (particles plus holes – thin solid curve) and thermal (holes – dashed curve, particles – dot-dashed curve) parts. The physics of pairing suggested by the conventional BCS is very simple and transparent. Pairing is generated by the pairing force. As temperature increases the thermal scattering of nucleons becomes stronger and stronger and finally destroys the pairing at $T_{c}$. The MBCS suggests another physics, according to which the heating itself generates a thermal constituent of the pairing gap: positive for holes and negative for particles. The heavier is the system, the stronger thermal pairing gap may be generated. A similar phenomenon takes place in calculation with realistic single particle spectra (see Figs. 3(c) and (d) where the pairing gap behavior in 120Sn is presented for neutrons and protons, respectively). Notice, when the pairing strength is weak to generate pairing at zero temperature, the MBCS predicts that the heating develops the pairing gap at finite $T$, as is obvious from Eqs. (5-6). An example of the pairing induced by heating alone in a magic nuclear system is shown in Fig. 3(d). One should not be surprised that the resulting pairing gap is negative in this example. Fig. 3 clearly shows that as temperature increases, the MBCS gap receives main contribution from $\delta\Delta_{h}$ and $\delta\Delta_{p}$ terms and their sum can be equally positive (less particle levels, $k<0$ in Fig. 1) or negative (more particle levels, $k>1$ in Fig. 1). The main goal of the MBCS theory is to mimic the thermal behavior of the pairing gap of a macroscopic treatment [9] that the normal – superfluid phase transition is washed out and soon above $T_{c}$ the gap remains rather small but positively finite. However, Fig. 3 shows that even technically, it is very difficult to achieve the desirable goal with the MBCS equations: it is necessary that two almost linearly growing functions $\delta\Delta_{h}$ and $-\delta\Delta_{p}$ almost cancel each other with a high accuracy in a large temperature interval111Independently from the physical content of the thermal gap in the MBCS, one finds from Eqs. (3-6) that $\delta\Delta_{h}\equiv-\delta\Delta_{p}$ for the conventional PFM and the cancellation between these two terms should be exact at any $T$. This does not happen in MBCS calculations because the MBCS theory violates Eqs. (3,4), see below.. Of course, the final result is very sensitive to tiny details of a single particle spectrum employed. If a spectrum with desirable properties is occasionally found, the theory breaks down anyway with a step aside from it. Fig. 1 clearly demonstrates it. ## 4 On range of the MBCS validity As is established, the MBCS theory is capable of generating the pairing gap behavior which looks reasonable at first glance below $T_{M}$. It is also established that $T_{M}$ grows linearly with the mass of the system and it is possible to find a single exceptional case of abnormally large $T_{M}$. Does this mean that the MBCS theory should be considered as reliable for $T<T_{M}$ and recommended for applications? To answer this question we return to the conventional PFM. Due to its internal particle - hole symmetry, the accuracy of the MBCS predictions for the theory variables can be easily examined even without having exact solutions. Verification of Eq. (4) for the $\delta N_{i}$ quantities in Figs. 2(a,c,d) reveals that this property of the system is enormously violated in the MBCS predictions. The behavior of the ratios $n_{i}/n_{-i}$ is very similar to the one of $\delta N_{i}/\delta N_{-i}$ in Fig. 2. Deviation from the exact result $n_{i}/n_{-i}\equiv 1$ reaches a few thousand per cent near $T_{c}$. The accuracy of the MBCS predictions for the quasiparticle energies, Eq. (2), and for the particle occupation probability, Eq. (3), with $i=1$ is examined in Fig. 4 and Fig. 5(a), respectively. Fig. 5(b) shows that the MBCS theory does not keep the energy of the Fermi surface at zero energy, as it should be, Eq. (1). Figure 4: MBCS predictions for the ratios of quasiparticle energies $(E_{-i}/E_{i})$ in different $N=\Omega$ systems. Exact results $(E_{-i}/E_{i})\equiv 1$ are shown by the dotted line. Figure 5: MBCS predictions for (a) ratios $(v_{-1}/u_{1})$ and (b) the energy of the Fermi level $E_{F}$ in $N=\Omega=$ 10, 30, and 100 systems. Exact results (a) $(v_{-1}/u_{1})\equiv 1$ and (b) $E_{F}\equiv 0$ for these systems are shown by the dotted line. The results in Figs. 2(a,c,d), 4, and 5 obviously demonstrate that as soon as heating starts to play a role, the MBCS predictions are irrelevant to a system which the theory aims to describe, because genuine properties of the system are severely violated in these predictions. It does not matter whether a light or heavy systems is considered, the results are almost independent on $N$. All-in-all, and this is the main conclusion of the article: although the MBCS theory generates a pairing gap which looks reasonable in some temperature interval $[0,T_{M}]$, it is clear from the analysis of other quantities that the theory predictions have nothing to do with the properties of a system under discussion even in this temperature interval. ## 5 Thermodynamic inconsistency of the MBCS theory We remind the reader that expressions for the MBCS theory were copied from zero-temperature BCS expressions by straightforward replacing the BCS $\\{u_{j},v_{j}\\}$ coefficients with the MBCS $\\{\bar{u}_{j},\bar{v}_{j}\\}$ coefficients. It has been already pointed out [6] that a method of mechanical copying of equations from one theory into another one has absolutely no grounds because the properties of $\\{u_{j},v_{j}\\}$ and $\\{\bar{u}_{j},\bar{v}_{j}\\}$ coefficients are simply different. E.g., the thermal part of the pairing gap $\delta\Delta$, discussed in Sec. 3, has appeared in the MBCS theory as a result of such an uncontrolled copying. It is very natural to verify whether expressions for thermodynamical observables obtained by such a method are thermodynamically consistent in the MBCS theory. Fig. 6 compares the system entropy which is calculated according its thermodynamical definition (dashed lines): $S_{\rm th}=\int_{0}^{T}\,\frac{1}{\tau}\,\cdot\,\frac{\partial E_{tot}}{\partial\tau}\,d\tau$ (7) and within a statistical approach (solid lines): $S_{\rm sp}=-\sum_{j}\,(2j+1)\,\left[\,n_{j}\,{\rm ln}\,n_{j}+(1-n_{j})\,{\rm ln}\,(1-n_{j})\right]~{}~{}.$ (8) The latter one is also called the single particle entropy in Refs. [10, 11] or the quasiparticle entropy in Ref. [8]. Figure 6: MBCS predictions for the system entropy calculated as $S_{\rm th}$, Eq. (7), and $S_{\rm sp}$, Eq. (8) in different light and heavy, symmetric and asymmetric PFM systems. Notice the logarithmic $y$ scale. Fig. 6 shows that $S_{\rm th}$ and $S_{\rm sp}$ being considered separately are amazingly close in different PFM systems for $T\leq 1$ MeV but they are dramatically different one from another in the MBCS predictions. The latter fact has been already addressed [5] when $S_{\rm th}^{\rm MBCS}$ and $S_{\rm sp}^{\rm MBCS}$ for the neutron system of 120Sn have been found different by two orders of magnitude. Different entropy-like quantities in nuclear physics have been considered in [10, 11]. It has been found that “the thermodynamic entropy $\ldots$, the information entropy $\ldots$ and the single-particle entropy $\ldots$, all coincide for strong enough interaction but only in the presence of a mean field” [11]. It is not possible to accept interpretation of these results in Ref. [8] as that $S_{\rm th}$ and $S_{\rm sp}$ “are nearly the same only for noninteracting particles”. Notice, correspondence between $S_{\rm th}$ and $S_{\rm sp}$ in a mean field plus residual forces (Figs. 56(IIa, IIc) in [10]) is definitely not worse than in an almost pure mean field approach (Figs. 56(Ia, Ic) in [10]) . Fig. 6(a) also presents the quantities $S_{\rm th}^{\rm BCS}$ and $S_{\rm sp}^{\rm BCS}$ calculated within the conventional BCS theory and $S_{\rm th}^{\rm exact}$ obtained from the exact solution of the PFM for the $N=\Omega=10$ system. One notices that for $T>100$ keV: $S_{\rm th}^{\rm exact}\approx S_{\rm th}^{\rm BCS}\approx S_{\rm sp}^{\rm BCS}\approx S_{\rm sp}^{\rm MBCS}>>S_{\rm th}^{\rm MBCS}~{}~{}.$ (9) An attempt to compare $S_{\rm th}^{\rm MBCS}$ and $S_{\rm sp}^{\rm MBCS}$ with the entropy calculated from the exact solution of the PFM has been made in Fig. 6 of Ref. [8] for the $N=10$, $\Omega=11$ system. First of all, the calculation of $S_{\rm sp}^{\rm exact}$ (thick solid line in Fig. 6(a) of Ref. [8]) is definitely not correct. The author of [8] claims that $S_{\rm sp}^{\rm exact}$ does not vanish at $T=0$ because occupation probabilities $f_{h}<1$ and $f_{p}>0$. The problem with the third law of thermodynamics222“at absolute zero, any part of the body must be in a definite quantum state - namely the ground state $\ldots$ the entropy of the body - the logarithm of its statistical weight - is equal to zero” (see $\S$23, page 66 in [12]). in this calculation is caused by confusion of interacting particles with noninteracting “quasiparticles”: particle levels are not eigen states of the pairing Hamiltonian, their occupation numbers do not obey Fermi-Dirac distribution and because of that, they cannot be used in Eq. (8) which represents the free Fermi-gas combinatorics. Incorrectly calculated $S_{\rm sp}^{\rm exact}$ is also published in Fig. 8 of [3]. One notices that Fig. 6(b) of Ref. [8] lacks comparison between $S_{\rm th}^{\rm MBCS}$ and $S_{\rm th}^{\rm exact}$ quantities333What is plotted by thick solid line in this figure and makes impression of agreement with $S_{\rm th}^{\rm MBCS}$ is not specified.. The latter quantity can be easily calculated from $E_{tot}^{\rm exact}$ (thick solid line in Fig. 4(a) of Ref. [8]). A level of expected disagreement between $S_{\rm th}^{\rm MBCS}$ and $S_{\rm th}^{\rm exact}$ quantities can be seen from our Fig. 6(a) (compare dashed curve with circles). Analysis of the MBCS predictions for the system entropy clearly indicate a problem with the expression for the system energy which enters in Eq. (7). How this expression was obtained, is mentioned in the beginning of the section. In Fig. 9 of Ref. [6] the system energy calculated from this expression has been straightforwardly compared to the system energy calculated as ${\rm<H>=Tr(HD)}$ where $D$ is a density operator and $<\ldots>$ means averaging over the grand canonical ensemble. Dramatic disagreement in two quantities representing the same physical observable has been obtained indicating the same problem. ## 6 Conclusions In this article we continue discussion on the validity of the MBCS theory. The theory performance is examined within the PFM. Our present goal is to allow the reader, who is not familiar with the previous discussion on the subject in Refs. [4, 5, 6, 7], to better understand the results and conclusions in Refs. [3, 8]. We confirm that there exists a single example of the PFM (with the number of levels $\Omega$ equal to the number of particles $N$ plus one) in which the MBCS produces the thermal behavior of the pairing gap similar to the one of a macroscopic theory up to a rather high temperature. In our opinion, this fact alone is not sufficient to conclude on the theory performance without noticing that in all other examples of the PFM with $\Omega\neq N$ the theory predicts phase transitions of unknown types at a much lower temperature. These other examples are neither shown nor discussed in Ref. [3]. Although the MBCS formally yields a visually acceptable pairing gap behavior in some temperature domain, the physical content of its pairing gap is very dubious. The MBCS predicts that heating generates a thermal constituent of the pairing gap which becomes stronger and stronger with temperature in contradiction with generally accepted understanding of the pairing phenomenon in nuclei as a result of a specific particle-particle interaction. We point out that this unphysical constituent of the MBCS pairing gap is responsible for strong sensitivity of the theory predictions to tiny details of single particle spectra. The conventional PFM with $\Omega=N$ possesses internal particle-hole symmetry. Due to it, some quantities in the model are known exactly without any calculations. We have demonstrated that the MBCS predictions deviate from exact results by a few hundred per cent starting from very low temperatures. We also point out that the MBCS is a thermodynamically inconsistent theory. Performing a systematic analysis of the MBCS performance in the PFM we have failed to find a model system and/or temperature range where the MBCS predictions are not questionable. ## Acknowledgments The work was partially supported by the Deutsche Forschungsgemeinschaft (SFB 634). ## References * [1] N.D. Dang, V. Zelevinsky, Phys. Rev. C 64 (2001) 064319. * [2] N.D. Dang, A. Arima, Phys. Rev. C 67 (2003) 014304. * [3] N.D. Dang, Nucl. Phys. A 784 (2007) 147. * [4] N.D Dang, A. Arima, Phys. Rev. C 74 (2006) 059801. * [5] V.Yu. Ponomarev, A.I. Vdovin, Phys. Rev. C 74 (2006) 059802. * [6] V.Yu. Ponomarev, A.I. Vdovin, Phys. Rev. C 72 (2005) 034309. * [7] N.D. Dang, A. Arima, Phys. Rev. C 68 (2003) 014318. * [8] N.D. Dang, Phys. Rev. C 76 (2007) 064320. * [9] L.G. Moretto, Phys. Lett. B 40 (1972) 1. * [10] V. Zelevinsky, B.A. Brown, N. Frazier, and M. Horoi, Phys. Rep. 276 (1996) 85. * [11] V.K. Kota and R. Sahu, Phys. Rev. E 66 (2002) 037103. * [12] L.D. Landau and E.M. Lifshitz, Statistical Physics (Pergamon Press, London-Paris, 1958).

arxiv-papers

2024-09-04T02:48:53.533598

0802.0480

andrea.taroni@fysik.uu.se, s.t.bramwell@ucl.ac.uk, peter.holdsworth@ens- lyon.fr # Universal Window for Two Dimensional Critical Exponents A Taroni1111Present address: Department of Physics, Uppsala University, Box 530, 751 21 Uppsala, Sweden, S T Bramwell2,1 and P C W Holdsworth3 1 University College London, Department of Chemistry, 20 Gordon Street, London WC1H 0AJ, UK 2 London Centre for Nanotechnology, 17-19 Gordon Street, London WC1H 0AH, UK 3 Université de Lyon, Laboratoire de Physique, École Normale Supérieure de Lyon, 46 Allée d’Italie, 69364 Lyon cedex 07, France ###### Abstract Two dimensional condensed matter is realised in increasingly diverse forms that are accessible to experiment and of potential technological value. The properties of these systems are influenced by many length scales and reflect both generic physics and chemical detail. To unify their physical description is therefore a complex and important challenge. Here we investigate the distribution of experimentally estimated critical exponents, $\beta$, that characterize the evolution of the order parameter through the ordering transition. The distribution is found to be bimodal and bounded within a window $\sim 0.1\leq\beta\leq 0.25$, facts that are only in partial agreement with the established theory of critical phenomena. In particular, the bounded nature of the distribution is impossible to reconcile with existing theory for one of the major universality classes of two dimensional behaviour - the XY model with four fold crystal field - which predicts a spectrum of non- universal exponents bounded only from below. Through a combination of numerical and renormalization group arguments we resolve the contradiction between theory and experiment and demonstrate how the “universal window” for critical exponents observed in experiment arises from a competition between marginal operators. ###### pacs: 82B26 ††: J. Phys.: Condens. Matter ## 1 Introduction New types of two-dimensional system on which meaningful physical experiments can be performed include optical lattices of trapped atomic gases [1], magnetic surfaces [2] and “$\delta$-doped” magnetic layers [3]. These add to a list of well established two dimensional systems that includes ultrathin magnetic films [4], atomic monolayers (both physi- and chemisorbed) [5, 6, 7, 8], crystalline surfaces [9], superconducting layers [10] and arrays of interacting Josephson junctions [11]. Recent theoretical developments on the concept of “extended universality” [12], the effects of finite size [13, 14], and the dipolar interaction [15, 16] should be particularly relevant to understanding experiments on these systems, both old and new. The key experiment on two dimensional systems is to test the existence and temperature dependence of a magnetic or crystalline order parameter $m(T)$. In cases where $m$ can be measured experimentally (which excludes, for example, superfluid films [7]), this is invariably found to approximate a power law over a certain range of temperature: $m\sim(T_{\mathrm{c}}-T)^{\beta}$, where $T_{\mathrm{c}}$ is the transition temperature. Theory predicts a limited number of possibilities for the value of the exponent $\beta$, as dictated by the universality class of the system. In two dimensions crystal symmetries and consequent universality classes are relatively few. We show here that the Ising, XY and XY with 4-fold crystal field anisotropy (XY$h_{4}$) are the three main experimentally relevant classes. The three and four state Potts models provide additional universality classes observed in experiments on adsorbed gaseous monolayers [17, 18] and surface reconstruction [9]. For the Ising, three- and four-state Potts models, $\beta=\frac{1}{8},\frac{1}{9},\frac{1}{12}$ respectively. For the XY model, one expects $\tilde{\beta}=0.23$, a universal number that arises in the finite size scaling at the Kosterlitz-Thouless-Berezinskii (KTB) phase transition [19, 20], though not a conventional critical exponent [21]. For XY$h_{4}$, theory predicts a continuously variable critical exponent $\beta\propto 1/h_{4}$ and thus a continuous spectrum of values when sampled over many real systems (see references [22, 23] and this work). We have tested these ideas by means of an extensive survey of experimental two dimensional critical exponents, including data for magnetic ultrathin films, layered magnets that exhibit a temperature regime of two dimensional behaviour [24, 25, 26, 27], order-disorder transitions in adsorbed gaseous monolayers [17, 18], and surface reconstructions [9]. The results are presented in Figure 1 and in the appendix. As observed previously on more limited data sets [4, 21, 28, 29], the distribution of $\beta$’s is distinctly bimodal, with strong peaks at $\beta=0.12$ and $\beta=0.23$, as expected for the Ising and XY models. In several cases ideal Ising [24, 30] and XY [25, 26, 27, 31, 32] behaviour has been confirmed in great detail by measuring thermodynamic quantities other than the magnetization. Likewise there is compelling evidence for Potts universality in several non magnetic systems [17, 18, 33, 34, 9]. However the XY$h_{4}$ universality is more elusive. Of particular relevance to the present discussion are the ferromagnetic monolayer Fe/W(100) [32] and the layered ferromagnets Rb2CrCl4 [25] and K2CuF4 [26], easy plane systems which have been shown to exhibit the full range of ideal XY behaviour despite their 4-fold symmetry. Another very well characterised easy plane system with a 4-fold crystal field is the layered antiferromagnet K2FeF4 [31], but this is not XY-like, with $\beta=0.15$ intermediate between the XY and Ising values. Claims for XY$h_{4}$ universality have been made for the ferromagnetic films Fe/[Au or Pd](100) [35], characterised only to a limited extent, as well as the order-disorder transitions of H/W(011) [36] and O/Mo(110) [37], for which full sets of critical exponents are available. The behaviour of these candidates for XY$h_{4}$ is seen to fall into two categories, which on closer inspection appears to be related to the strength of $h_{4}$: those with weak $h_{4}$ are XY-like with $\beta\approx 0.23$, while those with stronger $h_{4}$ have exponents in between the XY and Ising limits, $0.125\leq\beta\leq 0.23$. Most strikingly, there is no experimental evidence of the divergence of the exponent $\beta$ implied by $\beta\propto 1/h_{4}$. Instead, most experimental data that cannot be ascribed to the Potts classes lies in a “universal window”, bounded by the Ising and XY values. There are exceptions at the upper bound where crossover to three dimensional behaviour may increase the value of $\beta$ upwards from $0.23$ [21, 38]. However, it is clear from the histogram that the majority of systems are indeed encompassed in a limited range between the Ising and XY values. Figure 1: Histogram of $\beta$ values for all two-dimensional systems reported in Tables 4, LABEL:thinfilm_table and 6. The universal window is highlighted by the grey shading. Criteria for inclusion in the data set are discussed in the appendix. ## 2 Universality Classes in Two Dimensions Before we address the main question of why the universal window exists, it is relevant to specify the occurrence and relationships between two dimensional universality classes. Considering first magnetic degrees of freedom, we ignore the possibility of truly Heisenberg behaviour, remarking that the broken translational symmetry inherent to layers or surfaces, combined with a condition of crystal periodicity, means all real systems have at least one $p$-fold axis, which necessarily introduces relevant perturbations. Thus, although pure Heisenberg behaviour may be observable over a restricted temperature range [39, 40], it must give way to behaviour characteristic of the perturbations at temperatures near to the phase transition. These perturbations take the form of axial anisotropy (either easy axis or easy plane) and $p$-fold in-plane anisotropy ($p$ = 1,2,3,4 and 6). Easy-axis systems are generally Ising-like (despite the fact that the normal to the plane is usually a polar axis) while easy-plane systems with $p$ = 2-6 should be described by the XY$h_{p}$ model. XY$h_{2}$ is in the Ising class, whereas XY$h_{3}$ is in the 3-state Potts class, although it is very unlikely in magnetic systems owing to time reversal symmetry (we found no examples). XY$h_{4}$ constitutes a universality class distinct from the 4-state Potts class, while the phase transition in XY$h_{6}$ is in the XY class [22]. Inclusion of the dipolar interaction on lattices other than the square lattice does not add extra universality classes. However the case of the square lattice must be regarded as an unsolved problem: perturbative calculations [41] and numerical results [42, 43, 44] suggest that the square lattice dipolar model belongs to XY$h_{4}$, but the renormalisation group calculations of Maier and Schwabl indicate a different set of critical exponents [15]. The experimental data considered here are consistent with the former result rather than with the latter, but Maier and Schwabl’s prediction could yet be born out on an as yet undiscovered ideal model dipolar system. At least as far as the existing experimental data set is concerned, we conclude that, for magnetic systems, there are only three main universality classes: Ising, XY and XY$h_{4}$. The situation is essentially the same in non-magnetic systems [45, 46] but with the additional possibility of the 3- or 4-state Potts classes due to competing interactions beyond nearest neighbour [46, 47]. Indeed, Schick [46] used arguments from Landau theory to classify the phase transitions of two dimensional adsorbed systems into only four classes: the Ising, XY$h_{4}$, 3- and 4-state Potts. This set is supplemented by a chiral 3-state Potts class which shares conventional exponents with the pure 3-state Potts class [48] (hence for present purposes we shall treat these two cases as a single class). One result of the current work is that the pure XY class is also relevant to order-disorder transitions in adsorbed layers. Combining these observations we have five universality classes for structural systems and three for magnetic systems, as summarised in Table 1. Table 1: Classification of continuous transitions which can be observed in two-dimensional magnetic systems and in structural order-disorder transitions on surfaces. $\bullet$ indicates the occurrence of a particular universality class, whereas $\times$ indicates its absence. The special case of the square lattice dipolar system is discussed in the text. * Universality Class | Magnetic Systems | Adsorbed Systems ---|---|--- Ising | $\bullet$ | $\bullet$ XY | $\bullet$ | $\bullet$ XY$h_{4}$ | $\bullet$ | $\bullet$ 3-state Potts | $\times$ | $\bullet$ 4-state Potts | $\times$ | $\bullet$ ## 3 Calculation of Critical Exponents The relationship between the Ising, XY, XY$h_{p}$ and clock models may be discussed with reference to the following Hamiltonian: $\mathcal{H}_{p}=-J\sum_{\langle i,j\rangle}\cos(\theta_{i}-\theta_{j})-h_{p}\sum_{i}\cos(p\theta_{i}),$ (1) in which the $\theta_{i}$’s are the orientations of classical spins of unit length situated on a square lattice with periodic boundary conditions and confined to the XY plane, $J$ is the coupling constant and $h_{p}$ is the $p$-fold crystal field. It should be noted that unlike real systems, the lattice symmetry in computer simulations does not constrain the spin symmetry, and consequently the adoption of a square lattice does not restrict the generality of our arguments. In the limit $h_{p}\rightarrow\infty$, the Hamiltonian (1) is called a clock model, since $\theta_{i}$ is restricted to discrete values evenly spaced around a circle: $2\pi(n/p),\;n=1,\dots,p-1$. José, Kadanoff, Kirkpatrick and Nelson (JKKN) [22] have shown that for $p>4$, $h_{p}$ is an irrelevant scaling field down to intermediate temperatures, with the result that fluctuations restore the continuous symmetry of the 2dXY model above a threshold temperature, leading to a KTB transition [20] and quasi-long range order over a finite range of temperature. Recently it has been shown [12] that a similar scenario remains valid even for infinitely strong crystal field strength, with the result that fluctuations restore continuous symmetry for $p$-state clock models with $p>4$, although for $4<p\leq 6$ this occurs above the KTB temperature, $T_{\mathrm{KT}}$. For $p=2$ and $3$, $h_{p}$ is relevant, leading to phase transitions in the Ising and $3$-state Potts universality class respectively. $h_{4}$, on the other hand, is a marginal perturbation [22]. A second order phase transition is predicted with non- universal critical exponents depending on the field strength. As $h_{4}\rightarrow\infty$, XY$h_{4}$ crosses over to the $4$-state clock model, which is equivalent to two perpendicular Ising models, and the transition falls into the Ising universality class [49]. The non-universal transition for XY$h_{4}$ is hence bounded by the Ising universality class for large $h_{4}$. The non-universal exponents of XY$h_{4}$ can be calculated analytically within the framework proposed by JKKN. They showed that to describe the evolution of the KTB transition in the presence of a weak $p$-fold field it is sufficient to replace (1) by the generalized Villain Hamiltonian [22, 50] $\displaystyle\frac{\mathcal{H}}{k_{B}T}=$ $\displaystyle-$ $\displaystyle K\sum_{\langle i,j\rangle}\left[1-{1\over{2}}(\theta_{i}-\theta_{j}-2\pi m_{ij})^{2}\right]+\sum_{i}ipn_{i}\theta_{i}$ (2) $\displaystyle+$ $\displaystyle\log(y_{0})\sum_{i}S_{R}^{2}+\log(y_{p})\sum_{i}n_{i}^{2},$ where $K=J/k_{B}T$. The integers $m_{ij}$ maintain the periodicity of the original Hamiltonian, for rotations over an angle $2\pi$. $S_{R}$ is a directed sum of integers $m_{ij}$ around a square plaquette of four sites centred at $\vec{R}$: $S_{R}=m_{41}+m_{12}-m_{32}-m_{43}$, takes values, $S_{R}=0,\pm 1,\pm 2\dots$ and is therefore a quantum number for a vortex of spin circulation centred on the dual lattice site $\vec{R}$. $y_{0}$ is related to the chemical potential $\mu$ and fugacity $y$ for the creation of a vortex anti-vortex pair on neighbouring dual lattice sites: $y=y_{0}\exp(-\beta\mu)\approx y_{0}\exp(-\pi^{2}K/2)$. In the original Villain model $y_{0}=1$ but it is introduced here as a phenomenological small parameter which is renormalized in the subsequent flows. Similarly $y_{p}$ is a fugacity for a locking process of spins along one of the $p$-fold field directions with integer $n_{i}$ being a measure of the strength of this process at site $i$. For weak crystal fields, $y_{p}={1\over{2}}{\tilde{h}}_{p}$ with $\tilde{h}_{p}=(h_{p}/k_{B}T)$, which reproduces the field contribution to the partition function to leading order in $y_{p}$. For strong fields $y_{p}\rightarrow 1$ and (2) transforms into a discrete $p$-state model. Note however that this is not the $p$-state clock model: although the Villain model maintains the global rotational symmetry it destroys the local $O(2)$ symmetry of the pair interaction. The discrete terms $(\theta_{i}-\theta_{j}-2\pi m_{ij})^{2}$, $\theta_{i}=(n/p)2\pi,\;n=0,1,\dots,p-1$ hence do not have this symmetry over the interval $-\pi<(\theta_{i}-\theta_{j}-2\pi m_{ij})<\pi$. For $p=4$ this means that neighbouring spins orientated perpendicularly have an energy less than half that of antiparallel spins and the ordered state has lower lying excitations than the corresponding clock model. It is therefore not clear whether the Villain model falls into the correct universality class in the strong field limit and for quantitative studies one should use Hamiltonian (1) rather than (2). With $y_{p}$ set equal to zero, a direct space renormalization analysis for the spin-spin correlation functions resulting from (2) leads to RG flow equations for an effective coupling constant $K_{\mathrm{eff}}$ and vortex fugacity, $y$. For $K_{\mathrm{eff}}=2/\pi$, $y=0$, the flows yield the KTB transition [51]. In the presence of the $p$-fold field the flow equations are modified and a third equation is generated [22, 52]. For the explicit case with $p=4$, these are $\displaystyle\left(K^{-1}\right)^{\prime}$ $\displaystyle=$ $\displaystyle K^{-1}+4\left(\pi^{3}y_{0}^{2}e^{-\pi^{2}K}-\ 4\pi K^{-2}y_{4}^{2}e^{-4K^{-1}}\right)\ln(b)$ (3a) $\displaystyle y_{0}^{\prime}$ $\displaystyle=$ $\displaystyle y_{0}+(2-\pi K)y_{0}\ln(b)$ (3b) $\displaystyle y_{4}^{\prime}$ $\displaystyle=$ $\displaystyle y_{4}+\left(2-{4K^{-1}\over{\pi}}\right)y_{4}\ln(b),$ (3c) where $b$ is the scale factor and where the equations are valid as $b\rightarrow 1$. This set of equations has fixed points at $K^{\ast}=2/\pi$, $y_{0}^{\ast}=\pm y_{4}^{\ast}$. We can calculate the linearized transformation matrix evaluated at the fixed point, $\ast$: $M_{i,j}={\partial K_{i}\over{\partial K_{j}}}|_{\ast}$, where $K_{i}=K^{-1},y_{0},y_{4}$. Solving for the eigenvalues we find $\lambda=1,1+{\alpha\over{2}}\pm{1\over{2}}\sqrt{4a^{2}+\alpha^{2}},$ (3d) where $\alpha=16\pi^{2}(2\pi-1){\tilde{y}}^{2}e^{-2\pi}\ln(b)$, $a^{2}=2\gamma\delta$, $\delta={4\over{\pi}}{\tilde{y}}\ln(b)$, $\gamma=8\pi^{3}{\tilde{y}}e^{-2\pi}\ln(b)$, and where $y_{0}=y_{4}={\tilde{y}}$. Writing $\lambda=b^{\sigma}$ we extract the three scaling exponents. There is one relevant exponent, which is interpreted as $\sigma_{1}=1/\nu$, the exponent taking the coupling constant away from the critical value at the now regular second order phase transition. There is also one irrelevant variable $\sigma_{2}$, which is interpreted as driving the vortex fugacity to zero. Finally, there is one marginal variable, $\sigma_{3}$, which, as announced, corresponds to the scaling exponent of the $4$-fold crystal field. Taking $h_{4}=0$ all eigenvalues become marginal, consistent with the particular scaling properties of the 2dXY model. In the small field limit, $\sigma_{1}=-\sigma_{3}=4\pi e^{-\pi}\tilde{h}_{4}$ and $\sigma_{2}=0$. This gives the non-universal correlation length exponent [22] $\nu\approx 1.8(k_{B}T_{\mathrm{KT}}/h_{4})$. The strong field limit, $y_{4}=1$ gives $\nu\approx 0.47$, which should be compared with the exact result for the Ising model, $\nu=1$. The agreement is poor, as might be expected given the distortion of the four fold interaction imposed by the Villain model. It is clear from this result that a quantitative calculation for the strong field limit requires a different starting Hamiltonian. In order to calculate $\beta$ from the scaling relations [53], a second relevant scaling exponent is required. In this case the anomalous dimension exponent $\eta$ can be calculated directly from the correlation function [22]. At the KTB transition of the XY model, $\eta=1/4$, giving the universal jump in the effective spin stiffness, $K_{\rm eff}=2/\pi$. It follows from the scaling relation $2\beta=(d-2+\eta)\nu$ that the finite size scaling exponent $\beta/\nu=1/8$, as in the Ising model, despite the fact that here the true $\beta$ and $\nu$ are not defined. This is an example of “weak universality” [54] between the two models. A striking result in the presence of a $4$-fold field is that $\eta$ remains unchanged to lowest order in $h_{4}$ [22], indicating that a weak universal line extends out from the XY model along the $h_{4}$ axis. Here we make the hypothesis that the line extends right to the Ising limit, in which case $\eta=1/4$ for all $h_{4}$. This is clearly a reasonable assumption for the level of calculation made here. It is also an appealing result as other examples of weak universality are far less accessible to experiment [55]. Analysis of the numerical data presented in the next section lends weight to this hypothesis, although the observed behaviour is found to divide into two regimes, depending on the strength of the $h_{4}$ field. From this analysis we therefore predict a range of non-universal magnetization exponents going from $\beta\approx{1\over{8}}\left({1.8k_{B}T_{\mathrm{KT}}\over{h_{4}}}\right)$ (3e) for weak field, to $\beta=1/8$ in the strong field limit. To make quantitative comparison with simulation and experiment we need to estimate $\beta$ as a function of $h_{4}/J$. The critical value $K^{\ast}=2/\pi$ corresponds to a renormalized coupling constant, $J_{\mathrm{eff}}$, valid at large length scale such that $k_{B}T_{\mathrm{KT}}=\pi J_{\mathrm{eff}}/2$. In general $J_{\mathrm{eff}}<J$: for the Villain model $k_{B}T_{\mathrm{KT}}\approx 1.35J$ [56], while for the XY Hamiltonian (1) $k_{B}T_{\mathrm{KT}}/J\approx 0.9$ and is different again for more realistic Hamiltonians. Hence, while we can make a theoretical prediction for the low field behaviour, $\beta={1\over{8}}\left({\alpha J\over{h_{4}}}\right),$ (3f) with $\alpha$ a constant of order unity, scaling equation (3e) by a factor $k_{B}T_{\mathrm{KT}}/J$ will probably not lead to an accurate quantitative estimate for $\alpha$ and the precise value is beyond the scope of the present calculation. ## 4 Competition with Essential Finite Size Effects The survey of the $\beta$ values illustrated in Figure 1 shows a clear discrepancy between theory presented above and experiment: the large values of $\beta$ predicted for small $h_{4}$ do not appear and the range of values is cut off at $\beta\approx 0.23$. As the latter is an effective exponent characteristic of XY criticality up to a finite length scale, it seems clear that the non-universal critical phenomena are suppressed, for weak field, by the exceptional finite size scaling properties of the pure 2dXY model [21, 57, 58]. This hypothesis can be tested by numerical simulation, in which both $h_{4}$ and the system size may be directly controlled. In a real XY system the relevant length scale will in most cases be less than the physical size of the system: for example, it could be a coherence length controlled by defects or dipolar interactions [59], or, in the case of layered systems, a crossover scale to the third dimension [60, 21]. Thus, although real systems might have, for example, $10^{16}$ spins, the relevant scale for XY critical behaviour will typically be much smaller and compatible with the scale of Monte Carlo simulations, where the appropriate length scale is simply the system size. This finite length scale gives rise to a finite magnetization that disappears at the rounded KTB transition. As emphasised in [21], this is perfectly consistent with the Mermin-Wagner theorem [61], which proves that the magnetization will be strictly zero in the thermodynamic limit. It is easy to convince oneself that finite size corrections to the thermodynamic limit are important for any physically realizable cut off length scale. The resulting low temperature magnetization is therefore directly relevant for experiment. Figure 2: Monte Carlo data for the 2dXY model in the presence of a 4-fold crystal field, with $N=10^{4}$. Plot (a) displays the magnetization data of the XY model in the presence of increasingly strong anisotropies. Plot (b) displays the same data as a function of reduced temperature $t$, in a logarithmic scale. In both plots the “universal window” is highlighted by the blue shading. In Figure 2 we show the magnetic order parameter, the thermally averaged magnetic moment normalized to unity, versus temperature, with different $4$-fold field perturbations, for a system with $N=10^{4}$ spins. For $h_{4}=0$ the magnetization is characterized by the effective critical exponent, $\tilde{\beta}\approx 0.23$. A finite size analysis of Kosterlitz’ renormalization group equations shows that in the region of the transition it approaches a universal number $\tilde{\beta}=3\pi^{2}/128\approx 0.23$, in agreement with both experiment and simulation data, such as that shown here. For weak crystal field, $h_{4}$ there is no change in the the region of the transition and the magnetization data coincide with the data for zero field [62]. Only for $h_{4}/J\geq 0.5$ do they leave the zero field data through the transition, approaching results for the $4$-state clock model for large values of $h_{4}/J$. In Figure 2 we show $\log(m)$ against $\log(t)$, where $t=(T-T_{\mathrm{c}})/T_{\mathrm{c}}$. The transition temperature $T_{\mathrm{c}}$ is calculated from a finite size scaling analysis of the fourth order Binder cumulant for $M$ [63, 64, 59] and is an estimate of the value in the thermodynamic limit. The slopes, for small $t$, give a first estimate of the exponent $\beta$, indicating that it lies in the interval $1/8<\beta(h_{4})<0.23$ for all values of $h_{4}$, exactly as observed in experiment. The crossover to Ising behaviour is slow: for $h_{4}/J=1$, $\beta(h_{4})\approx 0.15$ and to approach $\beta\approx 1/8$ requires $h_{4}/J$ in excess of $5$. Hence the data here, as in previous numerical work [62, 65, 64], show evidence for a finite pocket of XY critical behaviour for small values of $h_{4}$. This appears to refute the prediction of JKKN, derived explicitly in the previous section, that the exponents vary continuously with $h_{4}$ [65] (see the further discussion below). For intermediate field strengths, however, the non- universal criticality does appear to hold as can be confirmed by a more detailed finite size scaling analysis. The values of $\beta$ and $\nu$ can be estimated more accurately by collapsing data for various system sizes onto the scaling relation $ML^{\beta/\nu}\sim tL^{1/\nu}$. The best data collapses for $h_{4}/J=1$ and $h_{4}/J=2$, with $T_{\mathrm{c}}$ in each case fixed from the Binder cumulant calculation, are shown in Figure 3. We find $\beta=0.148(5)$ and $\beta=0.136(10)$, in good agreement with the values found from Figure 2, and $\nu=1.19(4)$, $\nu=1.09(8)$. The ratio $\beta/\nu=0.126(4)$ in each case, is in agreement with the weak universality hypothesis. Similar results for $h_{4}/J=0.5$ can be found in [65]. Although these exponent values are not so different from the Ising model values, the data collapse is less satisfactory when Ising exponents are used. Figure 3: Best data collapses for the 2dXY model with 4-fold crystal field $h_{4}/J=1.0$ and $h_{4}/J=2.0$ for different system sizes. Further evidence for weak universality at intermediate field strengths can be found from studying the finite size scaling properties of $m$ at the transition. In Figure 4 we show $\log(m)$ against $\log(L)$ for $h_{4}/J=1$ for a range of temperatures near the transition. At the transition one expects a power law evolution characterised by the finite size scaling exponent $\eta/2={\beta/\nu}$. The best power law occurs at $T_{\mathrm{c}}=1.010(5)J$, which is the same as the value found from the Binder cumulant method. The scaling exponent $\eta/2=0.126(3)$, is the same as that found for the data collapse in Figures 3 and 3. Figure 4: Magnetisation $m$ against system size $L$, in log-log scale, for the 2dXY model with 4-fold crystal field $h_{4}/J=1.0$, for temperatures near the transition. This and previous numerical work [62, 65, 64, 66] are consistent with $h_{4}$ being marginal. In this case the crossover exponent to the new universality class is zero so crossover occurs, at best on exponentially large length scales, as a result of corrections to scaling [62]. Hence, for small and intermediate crystal field strengths the finite size scaling appears compatible with that of the continuous symmetry of the 2dXY model [62], as in the $6$-fold case. In fact the most detailed finite size scaling analysis [65] shows no evidence of such a crossover for small fields. It therefore remains an open question whether the pocket of pure XY behaviour for small $h_{4}/J$ is a pragmatic observation related to excessively slow crossover, or whether it remains right to the thermodynamic limit. In either case this is the main result of this section: large values of $\beta$ are indeed masked by the pocket of 2dXY behaviour, leading to the effective exponent $\tilde{\beta}$ for weak $h_{4}$ and creating a divide between systems with strong and weak $4$-fold fields, with the non-universal character of XY$h_{4}$ only appearing for $\beta(h_{4})<0.23$. The threshold value of $h_{4}$, separating the two regimes can be estimated theoretically by putting $\beta(h_{4})=0.23$ in Equation (3f). Using $k_{B}T_{\mathrm{KT}}/J\approx 0.9$ gives $\alpha=1.6$ and $h_{4}/J\approx 0.9$, a ratio of order unity, in agreement with the above general arguments, but an over estimate compared with numerics, where the change of regime occurs for $h_{4}/J\sim 0.5$, corresponding to $\alpha\approx 1$. Having confirmed that $\eta\approx 0.25$ over the whole range of $h_{4}$, we finally fix $\eta=0.25$ and use our estimates of $\nu(h_{4})$ from the scaling collapse to give a further estimate of the exponents as a function of $h_{4}$. The estimates of $\nu$ and $\beta$, summarised in Table 2, are in good agreement with all previous unconstrained estimates. We also include estimates of $\beta$ derived by a typical experimental analysis of fixing $T_{\mathrm{c}}$ from the maximum in the susceptibility or where the magnetization approaches zero, and deriving $\beta$ from a log-log plot. There is seen to be a systematic error between the different estimates of $\beta$, especially for small values of $h_{4}$. Nevertheless, the experimental exponents are still found to lie in the universal window of values predicted for the “true” exponents of the underlying model. The various critical exponents plotted in Figure 5 are found to be linear in $1/h_{4}$. By fitting to $\beta(h_{4})=0.125+a/h_{4}$, we estimate the constant $a$ to be 0.032 for the “true” exponents, and 0.05 for the experimental exponents. These values are clearly very different from that expected for the constant $\alpha$ in Equation (3f), but once outside the pocket of pure XY behaviour we are no longer in the weak field regime for which Equation (3f) is valid, as was shown in the previous section. Table 2: Critical exponents for the XY$h_{4}$ model, as determined from a finite size scaling analysis, and as measured directly from Monte Carlo magnetization data for a system of size $L=100$. * $h_{4}$ | $\nu$ | $\beta$ | $T_{\mathrm{c}}^{L}$ | $\beta(T_{\mathrm{c}}^{L})$ ---|---|---|---|--- 0.5 | 1.37(6) | 0.171(10) | 1.01(1) | 0.214(9) 1 | 1.19(4) | 0.148(5) | 1.04(1) | 0.196(6) 2 | 1.09(8) | 0.136(10) | 1.08(1) | 0.155(3) 5 | 1.04(6) | 0.130(7) | 1.12(1) | 0.129(3) $\infty$ | 1.00(5) | 0.125(6) | 1.14(1) | 0.123(3) Figure 5: Exponents $\beta$ and $\nu$ measured from a finite size scaling analysis of the Monte Carlo data plotted against $1/h_{4}$. ## 5 The Strong Field–Weak Field Divide Experimental evidence for the strong field–weak field divide comes from making head to head comparisons between systems listed in the appendix. A quantitative comparison is afforded by the ferromagnet Rb2CrCl4 [67] and the antiferromagnet K2FeF4 [31], both quasi two dimensional square lattice systems with $S=2$. In both systems the intra-plane isotropic exchange coupling, $J$ is much bigger than the inter-plane value $J^{\prime}$, giving rise to an extended temperature range with two dimensional critical fluctuations. However, while the ferromagnet shows all the characteristics of the pure XY universality class [25, 68], the antiferromagnet has non-universal exponents, with $\beta=0.15$ [31], which we now see to be consistent with XY$h_{4}$. A realistic model Hamiltonian for either system has the following form $\mathcal{H}=J_{0}\sum_{\langle i,j\rangle}\mathbf{S}_{i}\cdot\mathbf{S}_{j}+D\sum_{i}(S_{i}^{z})^{2}+\frac{1}{2}e(S_{+}^{4}+S_{-}^{4}),$ (3g) where the weak interplane exchange and a weak fourth order axial term are ignored (in the case of Rb2CrCl4 small departures from tetragonal symmetry are neglected for the purpose of this discussion). The crystal field $D$ confines the spins to an easy plane breaking the $O(3)$ rotational symmetry of the Heisenberg exchange and the 4-fold term $e$ breaks symmetry within that plane, putting Hamiltonian (3g) in the XY$h_{4}$ universality class. For both Rb2CrCl4 and K2FeF4 accurate estimates of the Hamiltonian parameters were derived by fitting magnon dispersions measured by neutron scattering to a self-consistent spin wave calculation [31, 67, 69]. However, in order to fit the spectra, the fourth order term was decoupled into an effective second order term, with amplitude $E\approx 6eS^{2}$. In the low temperature limit one can estimate parameters $\tilde{e}=\left|eS^{4}/J_{0}S^{2}\right|$ and $\tilde{D}=\left|DS^{2}/J_{0}S^{2}\right|$. Values are shown in Table 3 for both materials. Table 3: Main parameters for K2FeF4 and Rb2CrCl4, as determined from experiment [31, 67]. * | K2FeF4 | Rb2CrCl4 ---|---|--- $S$ | $2$ | $2$ $J$($\mathrm{K}$) | -15.7 | 15.12 $D$($\mathrm{K}$) | 5.7 | 1.06 $E$($\mathrm{K}$) | -0.49 | 0.123 $\tilde{D}$ | 0.363 | 0.07 $\tilde{e}$ | 0.0052 | 0.0013 $\beta$ | 0.15(1) | 0.230(2) To get an estimate of the 4-fold field that determines the critical exponents, it is tempting to assume that systems with $S=2$ are classical and to associate $\tilde{e}$ with the parameter $h_{4}/J$ arising from Equation (1). The parameter $\tilde{e}=0.0013$ for Rb2CrCl4 and $0.0052$ for K2FeF4, which seem sufficiently small to put both systems into the weak field regime with pure XY universality, as has been directly confirmed by numerical simulation [70]. However, assigning an effective classical Hamiltonian of the form (1) for systems with finite $S$ is not so straightforward: for finite $S$, through the uncertainty principle, out-of-plane and in-plane spin fluctuations are not statistically independent. As a consequence the energy scale for in-plane spin rotations and the consequent effective value for $h_{4}$ depend collectively on $J$ and $D$ as well as on $\tilde{e}$. This can be seen from a detailed consideration of the magnon dispersion arising from (3g). This calculation reveals a distinct difference between the ferromagnetic and antiferromagnetic cases, with the latter retaining strong quantum effects even for $S=2$. For the antiferromagnet one finds two magnon branches which, for $D=e=0$, are degenerate and gapless for zero wavevector and where each mode constitutes a conjugate in-plane and out-of-plane spin fluctuation term of equal amplitude. For finite crystal field strength the degeneracy is lifted, energy gaps appear everywhere in the spectrum and the symmetry is broken between the in-plane and out-of-plane fluctuation amplitudes. In the following we refer to a mode as in-plane or out-of-plane if the conjugate variable with the largest amplitude is in, or out of the plane. To lowest order in $1/S$ the out-of-plane branch develops a gap at zero wave vector: $\Delta_{1}=S\left[2(D+|E|)(2|J|z+4|E|)\right]^{1/2}\approx 2S\sqrt{Dz|J|},$ (3h) while the in-plane branch has $\Delta_{2}=S\left[(4|E|)(2|J|z+2D+2|E|)\right]^{1/2}\approx 2S\sqrt{2|E|z|J|},$ (3i) (here $z$ = 4). These gaps depend on the geometric mean of the exchange field $zJ$ and the crystal field $e$ or $D$ with the result that they are surprisingly large on the scale of $J$, as noticed by Thurlings et al. [31]. For K2FeF4 $\Delta_{1}=70.8$ K, $\Delta_{2}=23.9$ K at $4.2$ K, renormalising only weakly with temperature [31]. $\Delta_{1}$ is larger than the transition temperature, $T_{\mathrm{N}}=63\;K\approx JS^{2}$, so the out-of-plane branch of spin fluctuations will be frozen by quantum effects over the whole of the ordered phase, leaving the predominantly in-plane spin fluctuations only. Interpreting these as the classical fluctuations in an effective plane rotator model with Hamiltonian (1) leads to a crystal field, $h_{4}(\mathrm{eff})$, of the order of $\Delta_{2}$. This gives $h_{4}(\mathrm{eff})/JS^{2}\sim 0.33$ which is the right order of magnitude to fall into the strong field category. The fact that for K2FeF4, $T_{\mathrm{N}}/JS^{2}\approx 1$, as is the case for the model systems with Hamiltonian (1) presented in the previous section, is highly consistent with this interpretation. Higher order terms in $1/S$ renormalize $D$ and $|E|$ such that the values given in table 3 are higher than those predicted by fitting to linear spin wave theory. One can further speculate that quantum fluctuations for the in-plane branch will renormalize the effective $h_{4}$ in (1) [64] to an even higher value. The non-universal exponents observed for K2FeF4 could therefore be examples of the XY$h_{4}$ universality class. For the ferromagnet Rb2CrCl4, $D$ flattens the cone of spin precession giving a range of $q$ values where the energy spectrum varies approximately linearly with wave vector, but does not open a gap. The field $e$ opens a zero wave vector energy gap that varies as $\sqrt{De}$. It is of order $1$ K, decreasing to zero at the transition temperature $T_{\mathrm{c}}=52$ K, and so can hardly affect the thermodynamics in the critical region. Although the effective value of $h_{4}$ depends on the geometric mean of $D$ and $e$ rather than just the bare value of $e$ it is independent of $J$ and hence much smaller than than for the antiferromagnetic case. This places Rb2CrCl4 in the weak field regime, consistent with the observation of XY universality for this material [25, 68]. From this comparison, it seems likely that magnetic systems that show true XY$h_{4}$ universality will mostly be antiferromagnetic. Indeed a similar “spin dimensional reduction” due to quantum suppression of fluctuations has recently been observed in quantum Monte Carlo simulations with a Hamiltonian similar to (3g) [58]. More calculations beyond the spin wave approximation are required to clarify this point. Among non-magnetic systems, oxygen absorbed onto Mo(110) [37] or W(110) [71] and hydrogen on W(110) [36] have both been claimed to fall into the XY$h_{4}$ class, representing four fold equivalents of the two stage melting process for hexagonally coordinated systems [45]. Note that the (110) surface does not have four fold symmetry, but if we adopt these claims as a premise, then a comparison of the two systems is indeed perfectly consistent with XY $h_{4}$ universality and with the preceding arguments about the strong field-weak field divide. Electron hybridization between absorbed and substrate particles will result in the generation of electronic dipoles aligned perpendicularly to the (110) surface. The resulting $1/r^{3}$ interaction between the particles is repulsive and of sufficiently long range to ensure crystalization into a square lattice. The (110) surface provides a substrate potential with 4-fold topology (though not 4-fold symmetry) and which can be made commensurate with the free standing array by tuning the adsorbate density, the clearest example being the $(2\times 2)$ lattice structure [36]. The result is claimed to be in the XY$h_{4}$ universality class and the measured exponents, $\beta\simeq 0.19$ [37, 71], are, in light of the current work, consistent with this. In principle, the same should be true for the $(2\times 2)$ ordering transition for hydrogen on W(110) but the measured $\beta$, 0.25, is consistent with the pure XY model [36]. Hydrogen being so much lighter that oxygen, larger zero point fluctuations should make the substrate potential less effective at pinning the crystal, putting it in the category of systems with a weak field $h_{4}$, consistent with the experimental observation. ## 6 Other Exponents and Scaling Relations Further evidence for the experimental relevance of the finite size effects is found in the behaviour of other critical exponents. The exponent $\eta$, which according to the previous weak universality arguments should be $0.25$ for all $h_{4}$, is only found to closely approximate the theoretical value for model Ising systems such as Rb2CoF4 [72]. For model XY systems the predicted $\eta=0.25$ or $\delta=15$ are always observed at a temperature well below $T_{\mathrm{c}}$ (say $0.9$ $T_{\mathrm{c}}$), with $\eta(T)$ increasing to larger values at $T_{c}$ and $\delta(T)$ decreasing, since $\delta=(4-\eta)/\eta$. For example in the XY layered ferromagnets Rb2CrCl4 and K2CuF4 $\eta(T)$ and $\delta(T)$ have been measured with precision by several different methods [26, 73, 68]: in both cases $\eta$ rises to about $\eta=0.35$ at $T_{\mathrm{c}}$. This is a very strong signature of the finite size scaling properties of the XY model and is consistent with the predicted logarithmic shift in transition temperature, $\left[T_{\mathrm{c}}(L)-T_{KT}\right]\sim{1\over{\log^{2}(L)}}$ [21, 51], for a finite size system [74]. As the measured value of $\eta$ increases continuously through the transition its value at $T_{\mathrm{c}}(L)$ is thus expected to be in excess of $\eta=1/4$. It seems that the anomalous value of $\eta>0.25$ extends to systems with XY$h_{4}$ universality: for example, in K2FeF4 it is estimated to be $\eta\approx 0.35$. This is again consistent with the shift in transition temperature observed in finite size systems. Defining $T_{\mathrm{c}}(L)$ from the maximum susceptibility leads to a shift, $\left[T_{\mathrm{c}}(L)-T_{\mathrm{c}}\right]\sim L^{-1/\nu}$. Here, in the four fold field problem, $\nu>1$ which means that shift remains important even in the intermediate field regime. Referring to Figure 4, one can see that extracting a critical exponent from the initial slope, for $T>T_{\mathrm{c}}$, will lead to an overestimate of $\eta$. As experiments do not, in general, have access to the finite size scaling information available to numerical studies, it seems reasonable that the experimental $\eta$ values are generally larger than the expected thermodynamic limit value. Thus, we propose that $\eta(T_{\mathrm{c}})$ appearing greater that 1/4 remains a finite size effect. Similarly, the measured values of $\nu$ are systematically smaller than unity, while a consequence of weak universality is that $\nu$ should be greater than one for all finite $h_{4}$. For example for K2FeF4 $\nu\approx 0.9$, giving $\beta/\nu=0.16$, greater than the predicted ratio $1/8$, but together with $\gamma\approx 1.5$ the set of exponents do satisfy the hyperscaling relation, $2\beta+\gamma=d\nu$, as well as the relation $\beta/\nu=d-2+\eta/2$. The same holds true for oxygen on W(110) [71], for which $\beta$ and $\gamma$ have been determined to be 0.19 and 1.48, respectively. The shift in $\nu$ is therefore consistent with the shift in $\eta$. It seems reasonable to assume that these changes are also due to finite size effects, which at present prevent the observation of the weak universal line we have shown evidence for, for all values of $h_{4}$. More detailed experimental and numerical studies to clarify this point would be of great interest. ## 7 Conclusions In conclusion, the XY model with four fold crystal field is of relevance to a great number of experimental two dimensional systems. We have focused on the largest experimental data sets, those for two-dimensional magnets, adsorbed gaseous monolayers and in particular on the measured exponent $\beta$. With regard to the histogram in Figure 1, the systems that comprise it can only be fully understood on a case by case basis. However, we show in this paper that the Hamiltonian (1) contains the principle two dimensional universality classes that are relevant to experiment and that a uniform distribution of values $h_{4}$ would, because of the marginal finite size scaling properties of the model, produce a probability density of the same form as Figure 1 with a continuous spectrum bounded by peaks at the Ising and XY limits. This is what we refer to as the universal window for critical exponents. We have further shown that the actual values of the four fold crystal field that occur in real systems are, at first sight, too small to take any system away from the XY limit. However, we have identified at least one mechanism, in antiferromagnets, whereby the four fold field is effectively amplified by quantum confinement of the spins to the easy plane. Other mechanisms of realising XY$h_{4}$ universality are possible in individual cases [41, 47]. We have demonstrated the relevance of finite size scaling corrections to the experimental data set, with the relevant length scale giving a crossover away from XY criticality. Future work should focus on the finite size scaling aspects and on individual systems to see if a more accurate quantitative connection between the physical $h_{4}$ and the observed critical behaviour can be established. Further to this, we propose here that the non-universal exponents of XY$h_{4}$ should satisfy weak universality, with $\beta/\nu=1/8$ for all $h_{4}$ and we have given evidence that this is true in the range of intermediate field values. The robustness of the pocket of true XY behaviour, observed for weak fields [62, 65], in the thermodynamic limit remains an open question. Finally we remark that all evidence confirms that truly two dimensional systems, quasi two dimensional systems and numerical simulations reveal the same syndrome of behaviour, so much can be learned about new two dimensional systems [2, 3] through comparisons with old ones [31, 67]. It is fortunate that there is such an extensive and carefully determined data base. It is a pleasure to thank Maxime Clusel, Martin Greven, Björgvin Hjörvarsson and Marco Picco for useful discussions. PCWH thanks the London Centre for Nanotechnology and the Royal Society for financial support and the Rudolph Peierls Institute for Theoretical Physics, University of Oxford, for hospitality during the completion of this work. AT thanks the EPSRC for a studentship. ## Appendix: Construction of the Histogram of $\beta$ Exponents In constructing the histogram of experimental two dimensional $\beta$ exponents, a number of factors were considered. First, it was crucial to avoid circular logic by excluding those systems which were assigned a dimensionality purely on the basis of their exponent values, rather than on a large body of experimental evidence. Fortunately, we found no such cases in the literature. Therefore all systems included in the histogram are assigned as two dimensional on the basis of compelling experimental evidence of two dimensionality. Likewise we found no examples of systems considered to be two dimensional that exhibit $\beta\approx 1/3$, which might, in the absence of extra evidence, be mistakenly assigned as three dimensional systems and wrongly excluded from the data set. It should be noted that in layered magnets, the crossover from two dimensional to three dimensional exponents is generally very sharply defined so there is no ambiguity in identifying the two dimensional regime. A second criterion for inclusion in the histogram was that the exponents were determined with reasonable precision and accuracy (typically $\Delta\beta<\pm 0.01$). This inevitably necessitated a subjective judgement, but only a few results were excluded on these grounds: typically those exponents determined by powder (rather than single crystal) neutron diffraction, which is generally accepted to be inadequate for the accurate determination of $\beta$. The experimental exponents are generally not asymptotic exponents, but the numerical study presented above reveals that the difference between asymptotic exponents and those determined using finite size scaling techniques at temperatures down to $\sim 0.9$ $T_{\mathrm{c}}^{L}$ is generally negligible at the level of accuracy required for the present purpose. The histogram also excludes a number of interesting systems on the basis of there being legitimate grounds for alternative explanations for their observed critical behaviour. These include metamagnetic materials [75], systems undergoing spin-Peierls transitions [76, 77, 78], and bulk systems undergoing order-disorder transitions [79]. The following tables lists all the systems included in the histogram. Table 4 contains data for layered magnets, and includes examples of molecular magnets [80]. Note that K2MnF4 represents two data points in the histogram as the elegant work of van de Kamp et al. [81] used a magnetic field to tune the system between Ising and XY symmetry, with $\beta$ recorded for both cases. In all other cases the $\beta$’s are determined in zero applied field. Table LABEL:thinfilm_table contains data for ultrathin magnetic films. Although there are several cases in which films of different thicknesses have been measured in order to study crossover to three dimensional behaviour, only the values of $\beta$ in the two dimensional limit are reported here, and are included as only one data point in the histogram. Finally, Table 6 includes data for adsorbed gaseous monolayers, and systems undergoing surface reconstruction and melting processes. Our aim has been to make an exhaustive survey up to the time of publication. We apologise to any authors whose work we may have inadvertently overlooked, but we are confident that these cases would not significantly modify the form of the histogram. Table 4: List of 2d critical exponents $\beta$ for layered magnets reported in the literature, mostly measured by neutron diffraction (F = ferromagnet, A = antiferromagnet, Fo = (HCO2), chdc = _trans_ -1,4-cyclohexanedicarboxylate, 5CAP = 2-amino-5-chloropyridinium). System | $\beta$ | $t$ range | $T_{\mathrm{c}}$ ($\mathrm{K}$) | Type | Reference ---|---|---|---|---|--- Rb2CoF4 | 0.119(8) | 1$\cdot$10${}^{-1}-$2$\cdot$10-4 | 102.96 | A | [82] ErBa2Cu3O7 | 0.122(4) | 0.11$-$1$\cdot$10-4 | 0.618 | A | [83] K2CoF4 | 0.123(8) | 1$\cdot$10${}^{-1}-$8$\cdot$10-4 | 107.85 | A | [24] BaNi2(AsO4)2 | 0.135 | 3$\cdot$10${}^{-1}-$1$\cdot$10-2 | 19.2 | A | [27] Ba2FeF${}_{6}\,^{\ast}$ | 0.135(3) | 7$\cdot$10${}^{-1}-$4$\cdot$10-3 | 47.96(4) | A | [84] K2NiF4 | 0.138(4) | 2$\cdot$10${}^{-1}-$3$\cdot$10-4 | 97.23 | A | [85] K3Mn2F7 | 0.154(6) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | 58.3(2) | A | [86] Rb2MnCl4 (B $<$ 5.8 $\mathrm{T}$) | 0.15(1) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | 54 | A | [81, 87] K2MnF4 | 0.15(1) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | 42.14 | A | [88] K2FeF4 | 0.15(1) | $-$ | 63.0(3) | A | [31] Rb2MnF4 | 0.16(2) | 1$\cdot$10${}^{-1}-$3$\cdot$10-3 | 38.4 | A | [85] Pb2Sr2TbCu3O8 | 0.165(5) | $-$ | 5.30(2) | A | [89] BaFeF4 | 0.17 | 3$\cdot$10${}^{-1}-$1$\cdot$10-2 | | A | [90] Cr2Si2Te6 | 0.17(1) | 6$\cdot$10${}^{-1}-$3$\cdot$10-2 | 32.1(1) | F | [91] CsDy(MoO4)2 | 0.17(1) | $-$ | 1.36 | A | [92] CoCl${}_{2}\cdot$6H2O | 0.18 | 4$\cdot$10${}^{-1}-$4$\cdot$10-2 | 2.29 | A | [90] MnC3H7PO${}_{3}\cdot$H2O† | 0.18(1) | 4$\cdot$10${}^{-1}-$1$\cdot$10-2 | $\sim 15$ | F | [93] MnC4H9PO${}_{3}\cdot$H2O† | 0.18(1) | 4$\cdot$10${}^{-1}-$2$\cdot 10^{-2}$ | $\sim 15$ | F | [93] KFeF4 | 0.185(5) | 3$\cdot$10${}^{-1}-$1$\cdot$10-2 | 137.2(1) | A | [94, 90] Fe(NCS)2(pyrazine)2 | 0.19(2) | 2$\cdot$10${}^{-1}-$3$\cdot$10-2 | 6.8 | A | [95] Rb2FeF4 | 0.2 | 3$\cdot$10${}^{-1}-$2$\cdot$10-3 | 56.3 | A | [85] La2CoO4 | 0.20(2) | $-$ | 274.7(6) | A | [96] MnC2H5PO${}_{3}\cdot$H2O† | 0.21(2) | 6$\cdot$10${}^{-1}-$9$\cdot$10-2 | $\sim 15$ | A | [93] NH4MnPO${}_{4}\cdot$H2O† | 0.21(3) | 8$\cdot$10${}^{-1}-$2$\cdot$10-2 | 17.5(1) | A | [93, 97] K2CuF4 | 0.22 | 3$\cdot$10${}^{-1}-$3$\cdot$10-2 | 6.25 | F | [98] CuFo${}_{2}\cdot$4D2O‡ | 0.22(2) | 5$\cdot$10${}^{-1}-$5$\cdot$10-2 | 16.72 | A | [99] CuFo${}_{2}\cdot$CO(ND2)${}_{2}\cdot$2D2O‡ | 0.22(1) | 4$\cdot$10${}^{-1}-$1$\cdot$10-3 | 15.31 | A | [99] Tanol suberate§ | 0.22 | 7$\cdot$10${}^{-1}-$2$\cdot$10-2 | 0.7 | F | [100] Sr2CuO2Cl2 | 0.22(1) | 2$\cdot$10${}^{-1}-$1$\cdot$10-2 | 265.5(5) | A | [40] MnFo${}_{2}\cdot$2H2O | 0.22(1) | 4$\cdot$10${}^{-1}-$4$\cdot$10-2 | 3.6 | A | [101] La2NiO4 | 0.22(2) | 8$\cdot$10${}^{-2}-$2$\cdot$10-3 | 327.5(5) | A | [102] BaNi2(PO4)2 | 0.23 | 3$\cdot$10${}^{-1}-$2$\cdot$10-2 | 23.5(5) | A | [27] Cu(DCO2)${}_{2}\cdot$4D2O | 0.23(1) | $t>6\cdot$10${}^{-}2$ | 16.54(5) | A | [103] Rb2CrCl4 | 0.230(2) | 2$\cdot$10${}^{-1}-$1$\cdot$10-2 | 52.3 | F | [25] Gd2CuO4 | 0.23 | 7$\cdot$10${}^{-1}-$3$\cdot$10-3 | 6.4 | A | [104] (C6H5CH2NH3)2CrBr${}_{4}\,^{\P}$ | 0.23 | 7$\cdot$10${}^{-1}-$1$\cdot$10-1 | 52.0(1) | F | [105] KMnPO${}_{4}\cdot$H2O† | 0.23(2) | $t>9\cdot$10-2 | $\sim 15$ | A | [93] (CH3NH3)2MnCl4 | 0.23(2) | 1$\cdot$10${}^{-2}-$1$\cdot$10-3 | 44.75 | F | [106] ErCl3 | 0.23(2) | 4$\cdot$10${}^{-1}-$1$\cdot$10-2 | 0.350(5) | A | [107] (d6-5CAP)2CuBr4 | 0.23(4) | 4$\cdot$10${}^{-2}-$6$\cdot$10-3 | 5.18(1) | A | [108] Li2VOSiO${}_{4}\,^{\S}$ | 0.235(9) | 4$\cdot$10${}^{-1}-$2$\cdot$10-2 | 2.85 | A | [109] Li2VOGeO${}_{4}\,^{\S}$ | 0.236 | $-$ | 1.95 | A | [110] La0.04Sr2.96Mn2O${}_{7}\,^{\S}$ | 0.24(2) | $-$ | 145.0(5) | A | [111] La0.525Sr1.475MnO4 | 0.24(3) | $-$ | 110(1) | A | [112] RbFeF4 | 0.245(5) | 6$\cdot$10${}^{-1}-$1$\cdot$10-2 | 133(2) | A | [90] MnPS3 | 0.25(1) | $t>3\cdot$10-2 | 78.6 | A | [113, 114] Co5(OH)8(chdc)$\cdot$4H2O | 0.25(3) | $-$ | 60.5 | F | [115] YBa2Cu3O6+x | 0.26(1) | 5$\cdot$10${}^{-2}-$5$\cdot$10-3 | 410 | A | [116] Rb2MnCl4 (B $>$ 5.8 $\mathrm{T}$) | 0.26(1) | 1$\cdot$10${}^{-1}-$2$\cdot$10-3 | 54 | A | [87, 81] Rb2CrCl3Br | 0.26(1) | 9$\cdot$10${}^{-1}-$1$\cdot$10-2 | 55 | F | [117, 118] Rb2CrCl2Br2 | 0.26(1) | 9$\cdot$10${}^{-1}-$3$\cdot$10-2 | 57 | F | [117, 118] KMnF4 | 0.26(1) | 3$\cdot$10${}^{-1}-$3$\cdot$10-2 | 5.2(1) | A | [119] RbMnF4 | 0.26(1) | 3$\cdot$10${}^{-1}-$3$\cdot$10-2 | 3.7(1) | A | [119] ∗ Studied by Mössbauer spectroscopy. † Studied by bulk magnetometry. ‡ Studied by proton Nuclear Magnetic Resonance (NMR). § Studied by muon Spin Rotation ($\mu$SR). ¶ Studied by ac susceptibility. Table 5: Summary of transition temperatures $T_{\mathrm{c}}$ and magnetisation critical exponents $\beta$ for epitaxial magnetic films grown on a range of substrates. The thickness $d_{\mathrm{min}}$ denotes the thickness at which these quantities were measured and $t$ range indicates the range of reduced temperature over which $\beta$ was measured. The magnetic anisotropy is indicated by the direction of the easy axis, and can either be perpendicular ($\perp$) or parallel ($\parallel$) to the film plane. System | Structure | $d_{\mathrm{min}}$ (ML) | $\beta$ | $t$ range | $T_{\mathrm{c}}$ ($\mathrm{K}$) | Anisotropy | Method111Experimental properties were measured by the following techniques: Electron Capture Spectroscopy (ECS), Magneto Optical Kerr Effect (MOKE), Spin Polarised Low Energy Electron Diffraction (SPLEED), Torsion Oscillation Magnetometry (TOM), Convertion Electron Mössbauer Spectroscopy (CEMS), Linear Magnetic Dichroism in the Angular Distribution of photoelectron intensity (LMDAD) and Superconducting QUantum Interference Device (SQUID). | Reference ---|---|---|---|---|---|---|---|--- Fe on | | | | | | | | Pd(100) | bct, $1\times 1$ | 2.0 | 0.125(10) | $t<3\cdot 10^{-2}$ | 613.1 | $\perp$ | ECS | [120] | | 1.2 | 0.127(4) | 1$\cdot$10${}^{-1}-$3$\cdot$10-3 | $<100$ | $\perp$ | MOKE | [121] Ag(100) | bcc, $1\times 1$ | 2.5-2.7222Exponent determined by averaging over values of a range of films of different thickness. | 0.124(2) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | 324 | $\perp$ | MOKE | [122, 29] W(110) | bcc, $1\times 1$ | 0.8 | 0.124(1) | 1$\cdot$10${}^{-1}-$4$\cdot$10-3 | 221.1(1) | $\parallel$ [$1\overline{1}0$] | SPLEED | [123, 4] | | 1.0 | 0.134(3) | 1$\cdot$10${}^{-1}-$5$\cdot$10-2 | 224 | $\parallel$ [$1\overline{1}0$] | SPLEED | [30] | | 1.7 | 0.13(2) | $-$ | 317 | $\parallel$ [$1\overline{1}0$] | MOKE | [124] Ag(111) | bcc, $1\times 1$ | 1.8 | 0.139(6) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | $\sim 450$ | $\parallel$ | MOKE | [125] | | 2.0 | 0.130(3) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | $\sim 450$ | $\parallel$ | MOKE | [125] Cu(100) | fct, $4\times 1$ | $\sim 2.5$222Exponent determined by averaging over values of a range of films of different thickness. | 0.17(3) | 1$\cdot$10${}^{-1}-$1$\cdot$10-2 | 370 | $\perp$333Reversible spin reorientation transition from $\parallel$ to $\perp$ with increasing $T$. | MOKE | [126, 127, 128] W(110)444Coated with Ag. | bcc, $1\times 1$ | 0.82 | 0.18(1) | 3$\cdot$10${}^{-1}-$1$\cdot$10-2 | 282(3) | $\parallel$ [$1\overline{1}0$] | TOM, CEMS | [129, 130, 131] Cu84Al16(100) | fcc, $1\times 1$ | 4.0 | 0.212(5) | 3$\cdot$10${}^{-1}-$1$\cdot$10-2 | 288(2) | $\parallel$ | LMDAD | [132] W(100) | bcc, $1\times 1$ | 1.6 | 0.217(2) | 1$\cdot$10${}^{-1}-$1$\cdot$10-2 | 207.8(1) | $\parallel$ [$001$] | CEMS, SPLEED | [32, 133] Au(100) | bcc, $1\times 1$ | 1.0 | 0.22(1) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | 300 | $\parallel$ [$001$] | SPLEED | [134] | | 2.0 | 0.25(1) | 2$\cdot$10${}^{-1}-$1$\cdot$10-4 | 290 | $\parallel$ [$001$] | ECS | [135] W(100)444Coated with Ag. | bcc, $1\times 1$ | 1.5 | 0.22(2) | $-$ | 282(1) | $\parallel$ [$001$] | CEMS | [4] V(001) | bcc | 3 | 0.23(1) | 2$\cdot$10${}^{-1}-$2$\cdot$10-2 | $\sim 190$ | $\parallel$ | MOKE | [136] Pd555Pd layers $\delta$-doped with Fe. | $-$ | 0.2-0.4222Exponent determined by averaging over values of a range of films of different thickness. | 0.23(1) | 2$\cdot$10${}^{-1}-$2$\cdot$10-2 | $>50$ | $\parallel$ | MOKE | [3] GaAs(100) | bcc, $2\times 6$ | 3.4 | 0.26(2) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | 254.8(2) | $\parallel$ | MOKE | [137] Co on | | | | | | | | Cu(111) | fcc, $1\times 1$ | 1.0 | 0.125 | $-$ | 433 | $\perp$ | TOM | [138] | | 1.5 | 0.15(8) | $-$ | 460 | $\perp$ | MOKE | [139] Cu(100) | fcc | 2.0 | 0.24 | $-$ | $\sim 240$ | $\parallel$ | MOKE | [140, 141] Ni on | | | | | | | | W(110) | fcc, $7\times 1$ | 2.0 | 0.13(6) | 1$\cdot$10${}^{-1}-$1$\cdot$10-3 | 325 | $\parallel$ [$001$] | FMR | [142] Cu(111) | fcc, $1\times 1$ | 2.0-3.0222Exponent determined by averaging over values of a range of films of different thickness. | 0.24(7) | 3$\cdot$10${}^{-1}-$6$\cdot$10-3 | 380 | $\parallel$ | MOKE | [143] Cu(100) | fcc, $1\times 1$ | 4.1 | 0.23(5) | 3$\cdot$10${}^{-1}-$1$\cdot$10-2 | 284 | $\parallel$ | MOKE | [144, 139] V on | | | | | | | | Ag(100) | bcc, $1\times 1$ | 5.0 | 0.128(10) | 3$\cdot$10${}^{-1}-$2$\cdot$10-4 | 475.1 | $\parallel$ [$001$] | ECS | [145] Gd on | | | | | | | | Y(0001) | hcp | 1.0 | 0.23(5) | 1$\cdot$10${}^{-1}-$8$\cdot$10-3 | 156 | $\parallel$ | MOKE | [146] Mn5Ge3 on | | | | | | | | Ge(111) | | 1.0 | 0.244 | 2$\cdot$10${}^{-1}-$4$\cdot$10-3 | 296 | $\parallel$ | SQUID | [147] CoAl(100) | bcc, $1\times 1$ | | 0.22(2) | 2$\cdot$10${}^{-1}-$7$\cdot$10-3 | $\sim 90$ | $\parallel$ | MOKE | [2] Table 6: Chemisorbed and physisorbed systems displaying two-dimensional phase transitions. System | $\beta$ | Model Ascribed | Method | Reference ---|---|---|---|--- W(011)$p(2\times 1)$-H | 0.13(4) | 2d Ising∗ | LEED† | [36] W(011)$p(2\times 2)$-H | 0.25(7) | 2dXY∗ | LEED | [36] $p(1\times 2)\leftrightarrow(1\times 1)$-Au(110) | 0.13(2) | 2d Ising | LEED | [6] W(112)$p(2\times 1)$-O | 0.13(1) | 2d Ising | LEED | [5] $p(2\times 2)$-O disordering on Ru(001) | 0.13(2) | 3-state Potts | LEED | [18] $p(2\times 1)$-O disordering on Ru(001) | 0.085(15) | 4-state Potts | LEED | [17] $p(2\times 2)$-O disordering on Mo(110) | 0.19(2) | XY$h_{4}^{\ast}$ | LEED | [37] $p(2\times 1)$-O disordering on W(110) | 0.19(5) | XY$h_{4}^{\ast}$ | LEED | [71] Ru(001)$p(2\times 2)$-S | 0.11(2) | 4-state Potts | LEED | [33] Ru(001)$(\sqrt{3}\times\sqrt{3})R30^{\circ}$-S | 0.14(3) | 3-state Potts | LEED | [33] $(3\times 3)$-Sn disordering on Ge(111) | 0.11(1) | 3-state Potts | HAS‡, XRD§ | [34] $(3\times 1)$ reconstruction on Si(113) | 0.11(4) | 3-state Potts | LEED | [9] Xe melting on graphite | 0.23(4) | 2dXY | XRD | [8] ∗ Model not ascribed by original authors. | † LEED: Low Energy Electron Diffraction. | ‡ HAS: Helium diffraction. | § XRD: X-ray diffraction. | ## References * [1] Z. Hadzibabic, P. Krüger, M. Cheneau, B. Battelier, and J. Dalibard. Berezinskii-Kosterlitz-Thouless crossover in a trapped atomic gas. Nature, 441:1118, 2006. * [2] V. Rose, K. Brüggemann, R. David, and R. Franchy. Two-dimensional surface magnetism in the bulk paramagnetic intermetallic alloy CoAl(100). Phys. Rev. Lett., 98:037202, 2007. * [3] M. Pärnaste, M. Marcellini, E. Holmström, N. Bock, J. Fransson, O. Eriksson, and B. Hjörvarsson. Dimensionality crossover in the induced magnetization of Pd layers. J. Phys.: Condens. Matter, 19:246213, 2007. * [4] H-J. Elmers. Ferromagnetic monolayers. Int. J. Mod. Phys. B, 9:3115, 1995. * [5] G.-C. Wang and T.-M. Lu. Physical realization of two-dimensional Ising critical phenomena: Oxygen chemisorbed on the W(112) surface. Phys. Rev. B, 31:5918, 1985. * [6] J. C. Capunzano, M. S. Foster, G. Jennings, R. F. Willis, and W. Unertl. Au(110) (1$\times$2)-to-(1$\times$1) phase transition: A physical realization of the two-dimensional Ising model. Phys. Rev. Lett., 54:2684, 1985. * [7] D. J. Bishop and J. D. Reppy. Study of the superfluid state in two-dimensional 4He films. Phys. Rev. Lett., 40:1727, 1978. * [8] W. J. Nuttall, D. Y. Noh, B. O. Wells, and R. J. Birgeneau. Isothermal melting of near-monolayer xenon on single-crystal graphite. J. Phys.: Condens. Matter, 7:4337, 1995. * [9] Y.-N. Yang, E. D. Williams, R. L. Park, N. C. Bartelt, and T. L. Einstein. Disordering of the (3$\times$1) reconstruction on Si(113) and the chiral three-state Potts model. Phys. Rev. Lett., 64:2410, 1990. * [10] J.-H. Choy, S.-J. Kwon, and G.-S. Park. High-$T_{\mathrm{c}}$ superconductors in the two-dimensional limit: [(Py-CnH2n+1)2HgI4]-Bi2Sr2Cam-1CumOy (m = 1 and 2). Science, 280:1589, 1998. * [11] D. J. Resnik, J. C. Garland, J. T. Boyd, S. Shoemaker, and R. S. Newrock. Kosterlitz-Thouless transition in proximity-coupled superconducting arrays. Phys. Rev. Lett., 47:1542, 1981. * [12] C. M. Lapilli, P. Pfeifer, and C. Wexler. Universality away from critical points in two-dimensional phase transitions. Phys. Rev. Lett., 96:140603, 2006. * [13] S. G. Chung. Spontaneous symmetry breaking in Josephson junction arrays. Phys. Lett. A, 355:394, 2006. * [14] A. Trombettoni, A. Smerzi, and P. Sodano. Observable signature of the Berezinskii-Kosterlitz-Thouless transition in a planar lattice of Bose-Einstein condensates. New J. Phys., 7:57, 2005. * [15] P. G. Maier and F. Schwabl. Ferromagnetic ordering in the two-dimensional dipolar XY model. Phys. Rev. B, 70:134430, 2004. * [16] K. De’Bell, A. B. MacIsaac, and J. P. Whitehead. Dipolar effects in magnetic thin films and quasi-two-dimensional systems. Rev. Mod. Phys., 72:225, 2000. * [17] H. Pfnür and P. Piercy. Critical behaviour of $p$(2$\times$1) oxygen on Ru(001): An example of four-state Potts critical exponents. Phys. Rev. B, 40:2515, 1989. * [18] H. Pfnür and P. Piercy. Oxygen on Ru(001): Critical behaviour of a $p$(2$\times$2) order-disorder transition. Phys. Rev. B, 41:582, 1990. * [19] V. L. Berezinskiĭ. Destruction of long-range order in one-dimensional and two-dimensional systems having a continuous symmetry group I. Classical systems. Sov. Phys. JETP, 32:493, 1970. * [20] J. M. Kosterlitz and D. J. Thouless. Ordering, metastability and phase transitions in two-dimensional systems. J. Phys. C, 6:1181, 1973. * [21] S. T. Bramwell and P. C. W. Holdsworth. Magnetization and universal sub-critical behaviour in two-dimensional XY magnets. J. Phys.: Condens. Matter, 5:L53, 1993. * [22] J. V. José, L. P. Kadanoff, S. Kirkpatrick, and D. R. Nelson. Renormalization, vortices, and symmetry-breaking perturbations in the two-dimensional planar model. Phys. Rev. B, 16:1217, 1977. * [23] P. Calabrese and A. Celi. Critical behaviour of the two-dimensional $N$-component Landau-Ginzburg Hamiltonian with cubic anisotropy. Phys. Rev. B, 66:184410, 2002. * [24] H. Ikeda and K. Hirakawa. Neutron scattering study of 2-dimensional Ising nature of K2CoF4. Solid State Commun., 14:529, 1974. * [25] J. Als-Nielsen, S. T. Bramwell, M. T. Hutchings, G. J. McIntyre, and D. Visser. Neutron scattering investigation of the static critical properties of Rb2CrCl4. J. Phys.: Condens. Matter, 5:7871, 1993. * [26] K. Hirakawa. Kosterlitz-Thouless transition in two-dimensional planar ferromagnet K2CuF4. J. Appl. Phys., 53:1893, 1982. * [27] L. P. Regnault and J. Rossat-Mignod. In L. J. de Jongh, editor, Magnetic Properties of Layered Transition Metal Compounds, pages 271–321. Kluwer Academic Publishers, Dortrecht, 1990. * [28] S. T. Bramwell and P. C. W. Holdsworth. Universality in two-dimensional magnetic systems. J. Appl. Phys., 73:6096, 1993. * [29] Z. Q. Qiu, J. Pearson, and S. D. Bader. Two-dimensional Ising transition in epitaxial Fe grown on Ag(100). Phys. Rev. B, 49:8797, 1994. * [30] H-J. Elmers, J. Hauschild, and U. Gradmann. Critical behavior of the uniaxial ferromagnetic monolayer Fe(110) on W(110). Phys. Rev. B, 54:15224, 1996. * [31] M. P. H. Thurlings, E. Frikkee, and H. W. de Wijn. Spin-wave analysis in the two-dimensional antiferromagnet K2FeF4. I. Neutron scattering. Phys. Rev. B, 25:4750, 1982. * [32] H-J. Elmers, J. Hauschild, G. H. Liu, and U. Gradmann. Critical phenomena in the two-dimensional XY magnet Fe(100) on W(100). J. Appl. Phys., 79:4984, 1996. * [33] M. Sokolowski and H. Pfnür. Continuous order-disorder phase transitions of the $p$(2$\times$2) and $p$($\sqrt{3}\times\sqrt{3}$)${R}30^{\circ}$ superstructures of sulfur on Ru(001): Effective critical exponents and finite size effects. Phys. Rev. B, 49:7716, 1994. * [34] L. Floreano, D. Cvetko, G. Bavdek, M. Benes, and A. Morgante. Order-disorder transition of the (3$\times$3) Sn/Ge(111) phase. Phys. Rev. B, 64:075405, 2001. * [35] C. Rau and M. Robert. Anisotropic XY model for two-dimensional Fe. Mod. Phys. Lett. B, 10:223, 1996. * [36] I. F. Lyuksyutov and A. G. Fedorus. Critical exponents of the H-W(011) system. Sov. Phys. JETP, 53:1317, 1981. * [37] K. Grzelakowski, I. Lyuksyutov, and E. Bauer. Direct observation of scaling by high-resolution low-energy electron diffraction: O on Mo(110). Phys. Rev. Lett., 64:32, 1990. * [38] T. Lancaster, S. J. Blundell, P. J. Baker, W. Hayes, S. R. Giblin, S. E. McLain, F. L. Pratt, Z. Salman, E. A. Jacobs, J. F. C. Turner, and T. Barnes. Intrinsic magnetic order in Cs2AgF4 detected by muon-spin relaxation. Phys. Rev. B, 75:220408, 2007. * [39] H. M. Rønnow, D. F. McMorrow, and A. Harrison. High-temperature magnetic correlations in the 2D $S=1/2$ antiferromagnet copper formate tetradeuterate. Phys. Rev. Lett., 82:3152, 1999. * [40] M. Greven, R. J. Birgeneau, Y. Endoh, M. A. Kastner, M. Matsuda, and G. Shirane. Neutron scattering study if the two-dimensional spin $S$ = 1/2 square-lattice Heisenberg antiferromagnet Sr2CuO2Cl2. Z. Phys. B, 96:465, 1995. * [41] S. Prakash and C. L. Henley. Ordering due to disorder in dipolar magnets on two-dimensional lattices. Phys. Rev. B, 42:6574, 1990. * [42] K. De’Bell, A. B. MacIsaac, I. N. Booth, and J. P. Whitehead. Dipolar-induced planar anisotropy in ultrathin magnetic films. Phys. Rev. B, 55:15108, 1997. * [43] A. Carbognani, E. Rastelli, S. Regina, and A. Tassi. Dipolar interaction and long-range order in the square planar rotator model. Phys. Rev. B, 62:1015, 2000. * [44] J. F. Fernández and J. J. Alonso. Nonuniversal critical behavior of magnetic dipoles on a square lattice. Phys. Rev. B, 76:014403, 2007. * [45] D. R. Nelson and B. I. Halperin. Dislocation-mediated melting in two dimensions. Phys. Rev. B, 19:2457, 1979. * [46] M. Schick. The classification of order-disorder transitions on surfaces. Prog. Surf. Sci., 11:245, 1981. * [47] D. P. Landau and K. Binder. Phase diagrams and critical behavior of Ising square lattices with nearest-, next-nearest-, and third-nearest-neighbor couplings. Phys. Rev. B, 31:5946, 1985. * [48] D. A. Huse and M. E. Fisher. Domain walls and the melting of commensurate surface phases. Phys. Rev. Lett., 49:793, 1982. * [49] D. D. Betts. Exact solution of some lattice statistics models with 4 states per site. Can. J. Phys., 42:1564, 1964. * [50] J. Villain. Theory of one- and two-dimensional magnets with an easy magnetisation plane. II. The planar, classical, two-dimensional magnet. J. Physique, 36:581, 1975. * [51] J. M. Kosterlitz. The critical properties of the two-dimensional xy model. J. Phys. C, 7:1046, 1974. * [52] J. V. José, L. P. Kadanoff, S. Kirkpatrick, and D. R. Nelson. Erratum: Renormalization, vortices, and symmetry-breaking perturbations in the two-dimensional planar model. Phys. Rev. B, 17:1477, 1978. * [53] M. Plischke and B. Bergersen. Equilibrium Statistical Physics. World Scientific, London, 2nd edition, 1994. * [54] M. Suzuki. New universality of critical exponents. Prog. Theor. Phys., 51:1992, 1974. * [55] R. J. Baxter. Exactly Solved Models in Statistical Mechanics. Academic Press, London, 1982. * [56] W. Janke and K. Nather. Numerical evidence for Kosterlitz-Thouless transition in the 2D XY Villain model. Phys. Lett. A, 157:11, 1991. * [57] S. G. Chung. Essential finite-size effect in the two-dimensional XY model. Phys. Rev. B, 60:11761, 1999. * [58] A. Cuccoli, T. Roscilde, R. Vaia, and P. Verrucchi. Detection of XY behaviour in weakly anisotropic quantum antiferromagnets on the square lattice. Phys. Rev. Lett., 90:167205, 2003. * [59] Chenggang Zhou, D. P. Landau, and T. C. Schulthess. Monte Carlo simulations of Rb2MnF4: A classical Heisenberg antiferromagnet in two dimensions with dipolar interaction. Phys. Rev. B, 76:024433, 2007. * [60] S. Hikami and T. Tsuneto. Phase transition of quasi-two dimensional planar system. Prog. Theor. Phys., 63:387, 1980. * [61] N. D. Mermin and H. Wagner. Absence of ferromagnetism or antiferromagnetism in one- or two-dimensional isotropic Heisenberg models. Phys. Rev. Lett., 17:1133, 1966. * [62] S. T. Bramwell, P. C. W. Holdsworth, and J. Rothman. Magnetization in ultrathin films: Critical exponent $\beta$ for the 2D XY model with 4-fold crystal fields. Mod. Phys. Lett. B, 11:139, 1997. * [63] K. Binder. Critical properties from Monte Carlo coarse graining renormalization. Phys. Rev. Lett., 47:693, 1981. * [64] E. Rastelli, S. Regina, and A. Tassi. Monte Carlo simulation for square planar model with small four fold symmetry-breaking field. Phys. Rev. B, 70:174447, 2004. * [65] E. Rastelli, S. Regina, and A. Tassi. Monte Carlo simulation of a planar rotator model with symmetry-breaking fields. Phys. Rev. B, 69:174407, 2004. * [66] T. Schneider and E. Stoll. Molecular-dynamics study of a two-dimensional ferrodistortive XY model with quartic anisotropy. Phys. Rev. Lett., 36:1501, 1976. * [67] M. T. Hutchings, J. Als-Nielsen, P. A. Lindgard, and P. J. Walker. Neutron scattering investigation of the temperature dependence of long-wavelength spin waves in ferromagnetic Rb2CrCl4. J. Phys. C: Solid State Phys., 14:5327, 1981. * [68] S. T. Bramwell, P. C. W. Holdsworth, and M. T. Hutchings. Static and dynamic magnetic properties of Rb2CrCl4: Ideal 2D-XY behaviour in a layered magnet. J. Phys. Soc. Jap., 64:3066, 1995. * [69] T. Oguchi. Theory of spin-wave interactions in ferro- and antiferromagnetism. Phys. Rev., 117:117, 1960. * [70] A. Taroni. Theoretical Investigations of Two-Dimensional Magnets. PhD thesis, University College London, 2007. * [71] D. H. Baek, J. W. Chung, and W. K. Han. Critical behaviour of the $p$(2$\times$1)-O/W(110) system. Phys. Rev. B, 47:8461, 1993. * [72] H. Ikeda, M. Suzuki, and M. T. Hutchings. Neutron scattering investigation of static critical phenomena in the two-dimensional antiferromagnets: Rb2CocMg1-cCl4. J. Phys. Soc. Japan, 46:1153, 1979. * [73] C. A. Cornelius, P. Day, P. J. Fyne, M. T. Hutchings, and P. J. Walker. Temperature and field dependence of the magnetisation of Rb2CrCl4: a two-dimensional easy-plane ionic ferromagnet. J. Phys. C: Solid State, 19:909, 1986. * [74] J. Cardy. Scaling and Renormalization in Statistical Physics. Cambridge University Press, Cambridge, 1996. * [75] A. Rujiwatra, C. J. Kepert, J. B. Claridge, M. J. Rosseinsky, H. Kumagai, and M. Kurmoo. Layered cobalt hydroxysulfates with both rigid and flexible organic pillars: Synthesis, porosity and cooperative magnetism. J. Am. Chem. Soc., 123:10584, 2001. * [76] B. D. Gaulin, M. D. Lumsden, R. K. Kremer, M. A. Lumsden, and H. Dabkowska. Two dimensional ordering and fluctuations in $\alpha^{\prime}$-NaV2O5. Phys. Rev. Lett., 84:3446, 2000. * [77] J. E. Lorenzo, L. P. Regnault, S. Langridge, C. Vettier, C. Sutter, G. Grübel, J. Souletie, J. G. Lussier, J. P. Schoeffel, J. P. Pouget, A. Stunault, D. Wermeille, G. Dhalenne, and A. Revcolevschi. Observation of a Kosterlitz-Thouless state at the spin-Peierls phase transition in CuGuO3. Europhys. Lett., 45:45, 1999. * [78] R. J. Birgeneau, V. Kiryukhin, and Y. J. Wang. Tricritical to mean-field crossover at the spin-Peierls transition in CuGeO3. Phys. Rev. B, 60:14816, 1999. * [79] M. J. Harris. A new explanation for the unusual critical behaviour of calcite and sodium nitrate, NaNO3. American Mineralogist, 84:1632, 1999. * [80] S. J. Blundell and F. L. Pratt. Organic and molecular magnets. J. Phys.: Condens. Matter, 16:R771, 2004. * [81] R. van de Kamp, M. Steiner, and H. Tietze-Jaensch. Study of the phase diagram and the critical behaviour of the 2D Heisenberg antiferromagnet with small uniaxial anisotropy. Physica B, 241:570, 1998. * [82] E. J. Samuelsen. Experimental study of the two-dimensional Ising antiferromagnet Rb2CoF4. Phys. Rev. Lett., 31:936, 1973. * [83] J. W. Lynn, T. W. Clinton, W-H. Li, R. W. Erwin, J. Z. Liu, K. Vandervoort, and R. N. Shelton. 2D and 3D magnetic behavior of Er in ErBa2Cu3O7. Phys. Rev. Lett., 63:2606, 1989. * [84] K. Brennan, C. Hohenemser, and M. Eibschütz. 2D and 3D magnetic behavior of Er in ErBa2Cu3O7. J. Appl. Phys., 73:5500, 1993. * [85] R. J. Birgeneau, H. J. Guggenheim, and G. Shirane. Neutron scattering investigation of phase transitions and magnetic correlations in the two-dimensional antiferromagnets K2NiF4, Rb2MnF4, Rb2FeF4. Phys. Rev. B, 1:2211, 1970. * [86] C. M. J. van Uijen, E. Frikkee, and H. W. de Wijn. Neutron scattering study of magnetic ordering in the double-layer antiferromagnet K3Mn2F7. Phys. Rev. B, 19:509, 1979. * [87] H. Tietze-Jaensch, R. van de Kamp, and W. Schmidt. Magnetic excitation mode splitting and finite size effects in Rb2MnCl4. Physica B, 241:566, 1998. * [88] R. J. Birgeneau, H. J. Guggenheim, and G. Shirane. Spin waves and magnetic ordering in K2MnF4. Phys. Rev. B, 8:304, 1973. * [89] S. Y. Wu, W.-H. Li, K. C. Lee, J. W. Lynn, T. H. Meen, and H. D. Yang. Two- and three-dimensional magnetic correlations of Tb in Pb2Sr2TbCu3O8. Phys. Rev. B, 54:10019, 1996. * [90] L. J. de Jongh and A. R. Miedema. Experiments on simple magnetic model systems. Adv. Phys., 23:6, 1974. * [91] V. Carteaux, F. Moussa, and M. Spiesser. 2d Ising-like ferromagnetic behaviour for the lamellar Cr2Si2Te6 compound: a neutron scattering investigation. Europhys. Lett., 29:245, 1995. * [92] E. N. Khatsko, A. Zheludev, J. M. Tranquada, W. T. Clooster, A. M. Knigavko, and R. C. Srivastava. Neutron scattering study of the layered Ising magnet CsDy(MoO4)4. Low. Temp. Phys., 30:133, 2004. * [93] S. G. Carling, P. Day, and D. Visser. Dimensionality crossovers in the magnetization of the weakly ferromagnetic two-dimensional manganese alkylphosphonate hydrates MnCnH2n+1PO${}_{3}\cdot$H2O, n = 2-4. J. Phys.: Condens. Matter, 7:L109, 1995. * [94] M. L. Eibschütz, H. J. Guggenheim, L. Holmes, and J. L. Burnstein. CsFeF4: A new planar antiferromagnet. Solid State Commun., 11:457, 1972. * [95] H. N. Bordallo, L. Chapon, J. L. Manson, J. Hernàndez-Velasco, D. Ravot, W. M. Reiff, and D. N. Arhyriou. $S$ = 1/2 Ising behavior in the two-dimensional molecular magnet Fe(NCS)2(pyrazine)2. Phys. Rev. B, 69:224405, 2004. * [96] K. Yamada, M. Matsuda, Y. Endoh, B. Keimer, R. J. Birgeneau, S. Onodera, J. Mizusaki, T. Matsuura, and G. Shirane. Successive antiferromagnetic phase transitions in single-crystal La2CoO4. Phys. Rev. B, 39:2336, 1989. * [97] S. G. Carling, P. Day, and D. Visser. Dimensionality crossovers in the magnetization of the canted antiferromagnets NH4MnPO${}_{4}\cdot$H2O. Solid State Commun., 88:135, 1993. * [98] K. Hirakawa and K. Ikeda. Investigation of two-dimensional ferromagnet K2CuF4 by neutron scattering. J. Phys. Soc. Jpn., 35:1328, 1973. * [99] K. Koyama, H. Nobumasa, and M. Matsuura. Spontaneous staggered magnetization of two-dimensional Heisenberg like antiferromagnet with canting interaction. J. Phys. Soc. Jpn., 56:1553, 1987. * [100] S. J. Blundell, A. Hausmann, Th. Jestädt, F. L. Pratt, I. M. Marshall, B. W. Lovett, M. Kurmoo, T. Sugano, and W. Hayes. Muon studies of molecular magnetism. Physica B, 289:115, 2000. * [101] M. Matsuura, K. Koyama, and Y. Murakami. Asymmetric critical phenomena in a quasi two-dimensional-Heisenberg antiferromagnet Mn(HCCO)${}_{2}\cdot$H2O and Rb2CrCl2Br2. J. Phys. Soc. Jpn., 54:2714, 1985. * [102] K. Nakajiama, K. Yamada, S. Hosoya, Y. Endoh, M. Greven, and R. J. Birgeneau. Spin dynamics and spin correlations in the spin $S=1$ two-dimensional square-lattice Heisenberg antiferromagnet La2NiO4. Z. Phys. B, 96:479, 1995. * [103] S. J. Clarke, A. Harrison, T. E. Mason, G. J. McIntyre, and D. Visser. Magnetic ordering and fluctuations in the $S$ = 1/2 square Heisenberg antiferromagnet Cu(DCO2)${}_{2}\cdot$4D2O. J. Phys.: Condens. Matter, 4:L71, 1992. * [104] T. Chattopadhyay, P. J. Brown, A. A. Stepanov, A. I. Zvyagin, S. N. Barilo, and D. I. Zhigunov. Antiferromagnetic ordering in Gd2CuO4. J. Magn. Magn. Mater., 104:607, 1992. * [105] C. Bellitto, P. Filaci, and S. Patrizio. Zero-field magnetic susceptibility study of the magnetic phase transition in the two-dimensional ionic ferromagnet bis(benzylammonium) tetrabromochromate(II), (C6H5CH2NH3)2CrBr4. Inorg. Chem., 26:191, 1987. * [106] A. Paduan-Filho and C. C. Becerra. Magnetic properties and critical behavior of the pure and diluted two-dimensional weak ferromagnet (CH3NH3)2Mn1-xCdxCl4. J. Appl. Phys., 91:8294, 2002. * [107] K. W. Krämer, H. U. Güdel, P. Fischer, F. Fauth, M. T. Fernandez-Diaz, and T. Hauß. Triangular antiferromagnetic order in the honeycomb layer lattice of ErCl3. Eur. Phys. J. B, 18:39, 2000. * [108] F. C. Coomer, V. Bondah-Jagalu, K. J. Grant, A. Harrison, G. J. McIntyre, H. M. Rønnow, R. Feyerherm, T. Wand, M. Meissner, D. Visser, and D. F. McMorrow. Neutron diffraction studies of nuclear and magnetic structures in the $S$ = 1/2 square Heisenberg antiferromagnets (d6-5CAP)2CuX4 (X = Br and Cl). Phys. Rev. B, 75:094424, 2007. * [109] R. Melzi, S. Aldrovandi, F. Tedoldi, P. Carretta, P. Millet, and F. Mila. Magnetic and thermodynamic properties of Li2VOSiO4: A two-dimensional $S$ = 1/2 frustrated antiferromagnet on a square lattice. Phys. Rev. B, 64:024409, 2001. * [110] P. Carretta, N. Papinutto, R. Melzi, P. Millet, S. Gonthier, P. Mendels, and P. Wzietek. Magnetic properties of frustrated two-dimensional $S=1/2$ antiferromagnets on a square lattice. J. Phys.: Condens. Matter, 16:S849, 2004. * [111] A. I. Coldea, S. J. Blundell, C. A. Steer, J. F. Mitchell, and F. L. Pratt. Spin freezing and magnetic inhomogeneities in bilayer manganites. Phys. Rev. Lett., 89:177601, 2002. * [112] S. Larochelle, A. Mehta, L. Lu, P. K. Mang, O. P. Vajk, N. Kaneko, L. Zhou, and M. Greven. Structural and magnetic properties of the single-layer manganese oxide La1-xSr1+xMnO4. Phys. Rev. B, 71:024435, 2005. * [113] H. M. Rønnow, A. R. Wildes, and S. T. Bramwell. Magnetic correlations in the 2D $S$ = $\frac{5}{2}$ honeycomb antiferromagnet MnPS3. Physica B, 276:676, 2000. * [114] A. R. Wildes, H. M. Rønnow, B. Roessli, M. J. Harris, and K. W. Godfrey. Static and dynamic critical properties of the quasi-two-dimensional antiferromagnet MnPS3. Phys. Rev. B, 74:094422, 2006. * [115] M. Kurmoo, H. Kaumagai, S. M. Hughes, and C. J. Kepert. Reversible guest exchange and ferrimagnetism ($T_{\mathrm{c}}$=60.5 K) in a porous cobalt(II)-hydroxide layer structure pillared with trans-1,4-cyclohexanedicarboxylate. Inorg. Chem., 42:6709, 2003. * [116] W. Montfrooij, H. Casalta, P. Schleger, N. H. Andersen, A. A. Zhokiv, and A. N. Christiansen. Dimensional crossover in the XY-compound YBa2Cu3O3+x. Physica B, 241-243:848, 1998. * [117] S. T. Bramwell. Neutron Scattering, Magnetometry and Optical Spectroscopy of Rb2CrCl4, Rb2Cr$Cl_{3}$Br, Rb2CrCl2Br2 and Rb2CrCl2I2. PhD thesis, University of Oxford, 1989. * [118] S. T. Bramwell, P. Day, M. T. Hutchings, J. R. G. Thorne, and D. Visser. Neutron scattering and optical study of the magnetic properties of the two-dimensional ionic ferromagnets Rb2CrCl3Br and Rb2CrCl2Br2. Inorg. Chem., 25:417, 1986. * [119] M. C. Morón, F. Palacio, and J. Rodriguez-Carvajal. Crystal and magnetic structures of RbMnF4 and KMnF4 investigated by neutron powder diffraction: the relationship between structure and magnetic properties in the Mn3+ layered perovskites AMnF4 (A = Na, K, Rb, Cs). J. Phys.: Condens. Matter, 5:4909, 1993. * [120] C. Rau, P. Mahavadi, and M. Lu. Magnetic order and critical behavior at surfaces of ultrathin Fe(100)$p$(1$\times$1) films on Pd(100) substrates. J. Appl. Phys., 73:6757, 1993. * [121] C. Liu and S. D. Bader. Two-dimensional magnetic phase transition of ultrathin iron films on Pd(100). J. Appl. Phys., 67:5758, 1990. * [122] Z. Q. Qiu, J. Pearson, and S. D. Bader. Asymmetry of the spin reorientation transition in ultrathin Fe films and wedges grown on Ag(100). Phys. Rev. Lett., 70:1006, 1993. * [123] H-J. Elmers, J. Hauschild, H. Höche, U. Gradmann, H. Bethge, D. Heuer, and U. Köhler. Submonolayer magnetism of Fe(110) on W(110): Finite width scaling of stripes and percolation between islands. Phys. Rev. Lett., 73:898, 1994. * [124] C. H. Back, Ch. Würsch, A. Vaterlaus, U. Ramsperger, U. Maier, and F. Pescia. Experimental confirmation of universality for a phase transition in two dimensions. Nature, 378:597, 1995. * [125] Z. Q. Qiu, J. Pearson, and S. D. Bader. Magnetic phase transition of ultrathin Fe films on Ag(111). Phys. Rev. Lett., 67:1646, 1991. * [126] Dongqi Li, M. Freitag, J. Pearson, Z. Q. Qiu, and S. D. Bader. Magnetic phases of ultrathin Fe grown on Cu(100) as epitaxial wedges. Phys. Rev. Lett., 72:3112, 1994. * [127] J. Thomassen, F. May, B. Felfmann, M. Wuttig, and H. Ibach. Magnetic live surface layers in Fe/Cu(100). Phys. Rev. Lett., 69:3831, 1992. * [128] D. P. Pappas, K.-P. Kämper, and H. Hopster. Reversible transition between perpendicular and in-plane magnetization in ultrathin films. Phys. Rev. Lett., 64:3179, 1990. * [129] U. Gradmann, M. Przybylski, H-J Elmers, and G. Liu. Ferromagnetism in the thermodynamically stable monolayer Fe(110) on W(110), coated by Ag. Appl. Phys. A, 49:563, 1989. * [130] M. Przybylski and U. Gradmann. Ferromagnetic order in a Fe(110) monolayer on W(110) by mössbauer spectroscopy. Phys. Rev. Lett., 59:1152, 1987. * [131] H-J. Elmers, G. Liu, and U. Gradmann. Magnetometry of the ferromagnetic monolayer Fe(110) on W(110) coated with Ag. Phys. Rev. Lett., 63:566, 1989. * [132] W. A. A. Macedo, F. Sirotti, G. Panaccione, A. Schatz, W. Keune, W. N. Rodrigues, and G. Rossi. Magnetism of atomically thin fcc Fe overlayers on an expanded fcc lattice: Cu84Al16(100). Phys. Rev. B, 58:11534, 1998. * [133] H-J. Elmers and J. Hauschild. Magnetism and growth in psudomorphic Fe films on W(100). Surf. Sci., 320:134, 1994. * [134] W. Dürr, M. Taborelli, O. Paul, R. Germar, W. Gudat, D. Pescia, and M. Landolt. Magnetic phase transition in two-dimensional ultrathin Fe films on Au(100). Phys. Rev. Lett., 62:206, 1989. * [135] C. Rau. Ferromagnetic order and critical behaviour at surfaces of ultrathin epitaxial films. Appl. Phys. A, 49:579, 1989. * [136] M. Pärnaste, M. Marcellini, and B. Hjörvarsson. Oscillatory exchange coupling in the two-dimensional limit. J. Phys.: Condens. Matter., 17:L477, 2005. * [137] F. Bensch, G. Garreau, R. Moosbühler, G. Bayreuther, and E. Beaurepaire. Onset of ferromagnetism on Fe epitaxially grown on GaAs(001) (4$\times$2) and (2$\times$6). J. Appl. Phys., 89:7133, 2001. * [138] J. Kohlepp, H-J. Elmers, S. Cordes, and U. Gradmann. Power laws of magnetization in ferromagnetic monolayers and the two-dimensional Ising model. Phys. Rev. B, 45:12287, 1992. * [139] F. Huang, M. T. Kief, G. J. Mankey, and R. F. Willis. Magnetism in the few monolayer limit: A surface magneto-optic kerr-effect study of the magnetic behavior of ultrathin films of Co, Ni, and Co-Ni alloys on Cu(100) and Cu(111). Phys. Rev. B, 49:3962, 1994. * [140] C. C. Kuo, C. L. Chiu, W. C. Lin, and Minn-Tsong Lin. Dramatic depression of Curie temerature for magnetic Co/Cu(100) ultrathin films upon deposition at elevated temperature. Surf. Sci., 520:121, 2002. * [141] M. Gruyters, T. Bernhard, and H. Winter. Structural effects on the magnetix behaviour of ultrathin Co films on Cu(001) at the $T_{\mathrm{c}}$ jump. J. Magn. Magn. Mater., 292:192, 2005. * [142] Y. Li and K. Baberschke. Dimensional crossover in ultrathin Ni(111) films on W(110). Phys. Rev. Lett., 68:1208, 1992. * [143] C. A. Ballantine, R. L. Fink, J. Araya-Pochet, and J. L. Erskine. Magnetic phase transition in a two-dimensional system: $p$(1$\times$1)-Ni on Cu(111). Phys. Rev. B, 41:2631, 1990. * [144] F. Huang, G. J. Mankey, M. T. Kief, and R. F. Willis. Finite-size scaling behaviour of ferromagnetic thin films. J. Appl. Phys., 73:6760, 1993. * [145] C. Rau, G. Xing, and M. Robert. Ferromagnetic order and critical behavior at surfeces of ultrathin V(100)$p$(1$\times$1) films on Ag(100). J. Vac. Sci. Technol. A, 6:579, 1988. * [146] M. Gajdzik, T. Trappmann, C. Sürgers, and H. v. Löhneysen. Morphology and magnetic properties of submonolayer Gd films. Phys. Rev. B, 57:3525, 1998. * [147] C. Zeng, S. C. Erwin, L. C. Feldman, A. P. Li, R. Jin, Y. Song, J. R. Thompson, and H. H. Weitering. Epitaxial ferromagnetic Mn3Ge3 on Ge(111). Appl. Phys. Lett., 83:5002, 2003.

arxiv-papers

2024-09-04T02:48:53.539951

0802.0485

# Particle motion during the compaction of granular matter Steven Slotterback scsumd@umd.edu Masahiro Toiya Leonard Goff Department of Physics, and IREAP, University of Maryland, College Park, Maryland, 20742 Jack F. Douglas Polymers Division and Center for Theoretical and Computational Materials Science, National Institute for Standards and Technology, Gaithersburg, Maryland Wolfgang Losert wlosert@umd.edu Department of Physics, IPST and IREAP, University of Maryland, College Park, Maryland, 20742 ###### Abstract We track particle motions in a granular material subjected to compaction using a laser scattering based imaging method where compaction is achieved through thermal cycling. Particle displacements in this jammed fluid correlate strongly with rearrangments of the Voronoi cells defining the local spatial partitioning about the particles, similar to previous observations of Rahman on cooled liquids. Our observations provide further evidence of commonalities between particle dynamics in granular matter close to jamming and supercooled liquids. ###### pacs: 45.70.Cc,81.05.Rm,45.70.-n Granular materials are strongly interacting particle systems that admit an admixture of collective particle motions, as in crystals, and random particle motions, as in simple fluids. Under flow or deformation conditions these materials pass intermittently between a solid “jammed” state and a state of flow. When a granular material jams, the individual particles are in stable mechanical equilibrium with their local neighbors O’Hern et al. (2003). Small perturbations such as tapping Knight et al. (1995) or shearing Toiya et al. (2004) can move the particles, and lead to the evolution from one jammed state to another, generally more compact state. The increase in particle density with the number of perturbations is similar to the slow thermally driven dynamics of fluids close to their glass transition. Observations of phenomena of this kind has led to the development of a general framework for the jamming transition, with temperature, density, and external forcing as the three main variables characterizing this dynamical transition O’Hern et al. (2003); Liu and Nagel (1998). Although there have been a number of recent studies of particle trajectories in glass-forming colloidal fluids Weeks and Weitz (2002); Marcus et al. (1999) and other strongly interacting particle systems, little is known about the nature of particle trajectories in granular materials. While recent studies have successfully observed particle positions in a 3D jammed granular material Richard et al. (2003); Aste et al. (2007), particle trajectories were not observed. It seems plausible that the particle motion has features in common with glass-forming liquids, such as the emergence of string-like collective motion Douglas et al. (2006), and based on this hypothesis we examine whether the shape fluctuations in the Voronoi cells environments about the particles in compactified granular media correlate with the direction of particle displacements, as noticed by Rahman Rahman (1966) in his pioneering studies of collective particle motions in cooled liquids. On the other hand, the non- equilibrium nature of these driven particle systems leads us to expect differences from thermalized liquids that require a better understanding. In this paper, we provide the first direct measurements of 3D particle trajectories during compaction, with a focus on the statistics of particle motion. In particular, we study particle rearrangements in a column of granular materials during the course of thermal cycling. We use a laser sheet scanning method to reconstruct the configurations of the system at the end of each cycle. The motion of individual particles is tracked over 10 cycles, and we analyze how the local geometry affects local particle dynamics. Thermal cycling provides a method of granular compaction without exciting strong motions of particles. By thermal cycling we meanalternately heating and cooling the material and container, with one heating and cooling cycle corresponding to one perturbation. This technique leads to exponential compaction of the material with two characteristic timescales Chen et al. (2006) if the granular material and container have different thermal expansion coefficients. While most methods of forcing rearrangements of granular matter, such as tapping of the boundaries or air fluidization agitate individual particles significantly, i.e. provide particles kinetic energy relative to their neighbors, thermal cycling provides a quasi-stationary method of evolving the structure towards a jammed state. Experimental Setup: The granular material used is soda-lime glass beads (diameter $D=3{\rm\>mm}\>\pm\>0.3{\rm\>mm}$), poured into a transparent polymethylpentene (PMP) cylinder of diameter 5cm (16.6 D) to a height of $19.5{\rm\>cm}\>\pm\>0.5{\rm\>cm}$ (63.3 D to 66.6 D). This is the same materials as used in previous work on thermal cycling Chen et al. (2006). The thermal expansion coefficient for PMP is $1.17\times 10^{-4}{\rm\>K^{-1}}$, about one order of magnitude larger than that of the beads ($9\times 10^{-6}{\rm\>K^{-1}}$) Chen et al. (2006). The beads were immersed in index- matching oil (Cargille Labs Type DF 111Identification of a commercial product is made only to facilitate reproducibility and to adequately describe procedure. In no case does it imply endorsement by NIST or imply that it is necessarily the best product for the procedure.) which contains $2.08\>\mu{\rm g/mL}$ laser dye (Nile Blue 690 perchlorate). A transparent weight of $145.5{\rm\>g}$ was placed on top of the particles to monitor the filling height and to apply a controlled vertical force. The cylinder is placed inside a transparent water tank for thermal cycling at a rate of $16^{o}$ C per hour. As the PMP container was heated/cooled the cross-sectional area expanded/contracted by $0.46{\rm\>mm^{2}}$ for every degree heated/cooled ($6.5\>\%$ of a single particle cross-section). While bead expansion was negligible, the sample and oil are index-matched at room temperature. Since the index of refraction changes with temperature differently for glass and oil, we can only image our sample at the end of each thermal cycle. Figure 1: Experimental setup: Glass beads immersed in index matching oil in a cylinder are subjected to thermal cycling via a water bath. The fluorescently dyed fluid is imaged via a laser sheet that is moved across the sample. One of 600 cross-sectional images shown. Three dimensional images of the system at each cycle were obtained using a laser sheet scanning method, first described in Toiya et al. (2007). Briefly, a laser sheet is sent through the specimen at the excitation frequency of the laser dye to produce a cross-sectional image of the granular material in which the particles appear as dark circles within the dyed fluid (see Fig. 1). Images are taken with a high sensitivity cooled CCD camera (Sensicam, PCO). To obtain a three dimensional image, the laser is translated in $100\>\mu{\rm m}$ increments along the axis of the cylinder for a distance of 60 mm (20 D), starting from the top of the pile. The camera is not translated, but the changing object distance is accounted for in the analysis. A 3D bandpass filter is used to reduce noise and to smooth out images of the beads. A particle tracking algorithm by Crocker Crocker and Grier (1996) and Weeks Weeks (2007) detects the centers of $>98\>\%$ of all beads to within $100\>\mu{\rm m}$, which is smaller than the uncertainty in the bead diameters. The remaining $<2\>\%$ of bead positions are detected manually with comparable resolution using an interactive 3D visualization of images with an overlay of extracted particles. This allows us to generate complete maps of 3D jammed states. Particle rearrangmenets are followed through multiple thermal cycles. The particles move by less than 0.5 D in each thermal cycle, a scale small enough to allow for reliable particle tracking. Figure 2: Packing fraction vs. cycle number for various temperature differentials Results: We apply ten thermal cycles for three different temperature differentials: $25{\rm{}^{o}\>C}$, $40{\rm{}^{o}\>C}$, and $60{\rm{}^{o}\>C}$. The average packing fraction is determined from the vertical position of the flat glass weight on top of the pile in the 3D image. The initial conditions are generated by letting particles settle into the oil filled cylinder. After the particles settle, the glass top is allowed to sink onto the surface of the fluid immersed particles. Filling in oil first is necessary to eliminate air bubbles from the sample, but leads to significant variability in the initial packing fraction. Nevertheless, the increase in packing fraction during thermal cycling can be fit with an exponential $\rho(t)=\rho_{0}-A\>e^{t/\tau{}}$ as shown in Fig. 2 with $\rho_{0}=0.619\pm 0.0008$, $A=0.028\pm 0.0007$, and $\tau=4.87\>{\rm cycles}\pm 0.31\>{\rm cycles}$. The compaction rate is slower than the faster of the two time scales, 2.72 cycles, reported by Chen et al. (2006), consistent with a weaker forcing of the particles due to buoyancy forces. We find that the pair correlation function $g(r)$ does not evolve significantly during compaction, as shown in Fig. 3. This is not surprising since the particles only need to move closer by about $0.001\>D$ on average to achieve an increase in packing fraction of $1.5\>\%$. The overall shape of $g(r)$ is consistent with a random arrangement of particles in the presence of some ordering at the boundaries. Figure 3: Pair correlation function $g(r)$ averaged over 10 thermal cycles. The Voronoi construction defines a unique partitioning of space where each particle is assigned a local neighborhood, the Voronoi cell, corresponding to the region of space closest to a given particle. The Voronoi cell structures were determined using the QHULL algorithm Barber et al. (1996). To avoid boundary artifacts, Voronoi cells were computed for all particles, but only the volumes of particles that are not on the boundaries we used in the analysis. The distributions of Voronoi volumes is shown in Fig. 4. The distribution is non-gaussian with a sharp cutoff at small volumes and a broad range of large volumes, similar to distributions observed in simulations in glass-forming liquids Starr et al. (2002) and in recent experiments of granular materials Aste et al. (2007). One possible functional form of the distribution of voronoi volumes proposed by Aste et al. Aste et al. (2007) contains only one fitting parameter $k$. Fitting our Voronoi cell volume distributions to this function yields $k=24$, appreciably larger than the $k$ value, $k=12$, suggested by Aste et al. Aste et al. (2007), however. The larger polydispersity in bead diameters in the experiments presented here ($10\>\%$ as opposed to $2\>\%\>{\rm to}\>5\>\%$), could explain the wider distribution of Voronoi volumes. We find that the distribution of Voronoi volumes does not change significantly as the system is compacted. Figure 4: Histogram of Voronoi volumes Following Rahman’s investigation of cooled liquids, we analyze particle motion and its relation to the Voronoi cell shape. We define the vector $\vec{v}_{i,j}$ to be the displacement of the center of particle $i$ from cycle $j$ to $j+1$ (see Fig. 5). Next, we define a vector describing the Voronoi cell shape relative to the particle center. From the perspective of the particle, the vertices of the vornonoi construction correspond to the direction in which a particle “sees” a void between three neighbors. The mean position of all vertices therefore indicates where a particle would see more local void spaces. The vector $\vec{u}_{i,j}$ defines the displacement from the center of particle $i$ at cycle $j$ to the mean position of its corresponding Voronoi vertices, as shown in Fig. 5. The averages for $|\vec{u}|$ and $|\vec{v}|$ are approximately 1 mm (0.33 D) larger than the accuracy with which we detect particle positions. Figure 5: Definition of particle displacement vector v and u vector for a cross-section of the 3D Voronoi volume. Since this is a cross-section of the 3D Voronoi cell, the particle need not be in contact with any of the edges of the Voronoi volume. To determine the correlation between particle displacement and Voronoi shapes, we then consider the alignment of $\vec{u}_{i,j}$ and $\vec{v}_{i,j}$. Specifically, we calculate $\cos(\theta)$ for each image, where $\theta$ is the angle between $\vec{u}_{i,j}$ and $\vec{v}_{i,j}$ for each particle. The probability distribution of $\cos{\theta}$ averaged over all cycles is shown in Fig. 6. In addition, we show the distribution for the first and the last cycle, and a control for two 3-D images without thermal cycling between the images. We find that the distribution peaks at $\cos{\theta}=1$, indicating that the particle moves toward the center of the vertices of its Voronoi cell. Particle motion and Voronoi shape are indeed correlated. Figure 6: a) Preferred Direction PDFs for system with $\Delta T=40{\rm{}^{o}\>C}$ for each thermal cycle. b) Preferred Direction PDFs for system with $\Delta T=25{\rm{}^{o}\>C}$ for each thermal cycle. c) Preferred Direction PDFs for system at constant temperature d) 3D reconstruction of system from the perspective of a given particle facing five vertices of its Voronoi cell. Notice that the vertex is located in the void between particles. Discussion: By using the laser sheet scanning method, we visualize the internal structure and dynamics of a jammed granular material. Under quasi-static forcing via thermal cycling flows are slow enough (less than $\approx 0.5\>{\rm D}$ between frames), so we can observe the microscopic dynamics of this strongly interacting particle system. While the differences in local structures are too small to be observable in local distributions such as $g(r)$ or the Voronoi volume distribution, the Voronoi volume appears to hold important clues for the future dynamical evolution of the particles. Voronoi reconstruction indicates that particles tend to move toward the centers of Voronoi vertices as long ago observed in Rahman’s pioneering simulations of particle motion in cooled liquids Rahman (1966). This correlated displacement in our granular fluid is independent of the size of the Voronoi volume, the velocity of particles, and the position of particles within the granular column. Rahman Rahman (1966) and others following him Zwanzig and Bishop (1974) inferred from the correlations in the tracer particle displacement and the Voronoi cell shape that the particles were moving in tube-like environments, thus anticipating the string-like motion in cooled liquids that has recently become widely appreciated Donati et al. (1998); Douglas et al. (2006). This raises questions about whether such collective interparticle displacements also characterize particle motion in compactified granular materials. The absence of equilibrium in granular fluids, however, complicates the definition of “mobile” and collectively moving particles in granular materials, since there is no model such as Brownian motion to provide a “baseline” for this comparison. A recent study of a quasi-2D air-driven granular material Keys et al. (2007) indicates the presence of collective string-like motion as in cooled liquids so the general presence of such collective motions in driven granular media seem highly plausible. However, we expect important differences to emerge in systems such as ours where the volume of the material is evolving in time with compaction. We plan to study the evolutionary or aging dynamics of our granular fluid in the future. ###### Acknowledgements. We thank Krisztian Ronaszegi for assistance in data analysis. This work was supported by NSF grants CTS0457431 and CTS0625890. Official contribution of the National Institute of Standards and Technology - Not subject to copyright in the United States. ## References * O’Hern et al. (2003) C. S. O’Hern, L. E. Silbert, A. J. Liu, and S. R. Nagel, Phys. Rev. E 68, 011306 (2003). * Knight et al. (1995) J. B. Knight, C. G. Fandrich, C. N. Lau, H. M. Jaeger, and S. R. Nagel, Phys. Rev. E 51, 3957 (1995). * Toiya et al. (2004) M. Toiya, J. Stambaugh, and W. Losert, Phys. Rev. Lett. 93, 088001 (2004). * Liu and Nagel (1998) A. J. Liu and S. R. Nagel, Nature 396, 21 (1998). * Weeks and Weitz (2002) E. R. Weeks and D. A. Weitz, Phys. Rev. Lett. 89, 095704 (2002). * Marcus et al. (1999) A. H. Marcus, J. Schofield, and S. A. Rice, Phys. Rev. E 60, 5725 (1999). * Richard et al. (2003) P. Richard, P. Philippe, F. Barbe, S. Bourlès, X. Thibault, and D. Bideau, Phys. Rev. E 68, 020301 (2003). * Aste et al. (2007) T. Aste, T. D. Matteo, M. Saadatfar, T. J. Senden, M. Schr ter, and H. L. Swinney, Euro. Phys. Lett. 79, 24003 (2007). * Douglas et al. (2006) J. F. Douglas, J. Dudowicz, and K. F. Freed, The Journal of Chemical Physics 125, 144907 (pages 17) (2006). * Rahman (1966) A. Rahman, J. Chem. Phys. 45, 2585 (1966). * Chen et al. (2006) K. Chen, J. Cole, C. Conger, J. Draskovic, M. Lohr, K. Klein, T. Scheidemantel, and P. Schiffer, Nature 442, 257 (2006). * Toiya et al. (2007) M. Toiya, J. Hettinga, and W. Losert, Granular Matter 9, 323 (2007). * Crocker and Grier (1996) J. Crocker and D. Grier, J. Colloid Interface Sci. 179, 298 (1996). * Weeks (2007) E. Weeks, _Particle tracking using idl_ (2007), URL http://www.physics.emory.edu/~weeks/idl/. * Barber et al. (1996) C. B. Barber, D. P. Dobkin, and H. Huhdanpaa, ACM Trans. Math. Softw. 22, 469 (1996). * Starr et al. (2002) F. W. Starr, S. Sastry, J. F. Douglas, and S. C. Glotzer, Phys. Rev. Lett. 89, 125501 (2002). * Zwanzig and Bishop (1974) R. Zwanzig and M. Bishop, The Journal of Chemical Physics 60, 295 (1974). * Donati et al. (1998) C. Donati, J. F. Douglas, W. Kob, S. J. Plimpton, P. H. Poole, and S. C. Glotzer, Phys. Rev. Lett. 80, 2338 (1998). * Keys et al. (2007) A. S. Keys, A. R. Abate, S. C. Glotzer, and D. J. Durian, Nature 3, 260 (2007).

arxiv-papers

2024-09-04T02:48:53.548302

0802.0564

# Controlling the quantum computational speed N. Metwally, M. Abdel-Aty111E-mail: abdelatyquantum@gmail.com and M. Sebawe Abdalla Mathematics Department, College of Science, Bahrain University, 32038 Kingdom of Bahrain Mathematics Department, Faculty of Science, Sohag University, 82524 Sohag, Egypt Mathematics Department, College of Science, King Saud University, Riyadh 11451, Saudi Arabia ###### Abstract The speed of quantum computation is investigated through the time evolution of the speed of the orthogonality. The external field components for classical treatment beside the detuning and the coupling parameters for quantum treatment play important roles on the computational speed. It has been shown that the number of photons has no significant effect on the speed of computation. However, it is very sensitive to the variation in both detuning and the interaction coupling parameters. ###### pacs: 03.67.-a; 32.80.Pj; 42.50.Ct; 42.65.Yj; 03.75.-b ## 1 Introduction Recently, there are a great interest for developing the computer device. This is mainly due to the complications of the problems which we are usually facing and urgently need to find their solutions [1, 2, 3]. To overcome these problems we have to find a computer device with high capacity and speed or we have to develop a new kind of computer has such properties. This in fact means that we need one has a large memory, an adequate processor and a large hard disk. This stimulated and encouraged the researcher to seek for another kind of computer different to that of the classical one, that is the quantum computer. However, in the quantum information and more precisely in the quantum computer there is an important question would be raised, what is the speed of sending information from nod to the other one that to reach the final output. Since the information is coded in a density operator, therefore we ask how fast the density operator will change its orthogonality. In other words, we search for a minimum time needed for a quantum system to pass from one orthogonal state to another [4]. To perform this task we need qubit contains the information to evolve through a unitary operator where the carrier transforms it from one nod to the other. For entangled qubit pair, one can see the operators cause a decay of entanglement [5, 6, 7]. Further, it has been shown that the classical noise leads to what is called entanglement sudden death [8]. Moreover, the time- dependent interaction of a single qubit with a field can also produce such a phenomena as well as along-lived entanglement[12]. This means that there are different factors would be involved and affect the transmission speed and consequently the information. Recently the efforts in quantum information research are directed towards improving the performance of single qubit interaction. Also, evolution speed (maximum transition rate between orthogonal state) and the time evolution of some models has been discussed in Ref. [9]. The main purpose for the present communication is to consider the interaction between a single qubit and an external field for the classical treatment and the interaction between a single qubit and cavity field for the quantum treatment. This is to shed some light on the general behavior of the interaction process and its relationship with the speed of the computation [4] (maximum number of orthogonal states that the system can pass through per unit time), speed of orthogonality [10] (minimum time for a quantum state $|\psi_{i}\rangle$ to evolve into orthogonal state $|\psi_{f}\rangle$ where $\langle\psi_{i}|\psi_{f}\rangle=0$) and speed of evolution [11] (maximum transition rate between orthogonal state). The paper is organized as follows. In Sec. $2$, we consider the classical treatment, where we calculate the general form of the time evaluation of the density operator. The quantum interaction of the qubit will be considered in Sec. $3$, where we introduce the unitary operator in an adequate form. Also, we obtain the final state by means of the Bloch vectors. Also we study the effect of the field parameters on the speed of the quantum computation. Finally, our conclusion is given in Sec $4.$ ## 2 Classical treatment of Qubit Let us start out with a short reminder on a general form of the density operator of a qubit with the aid of analogs of Pauli’s spin vector operator $\overrightarrow{\sigma}$. This row vector refer to the three dimensional vector [13] $\overrightarrow{\sigma}=\sum_{\alpha=x,y,z}{\overrightarrow{\sigma_{\alpha}}{e_{\alpha}^{\downarrow}}}=(\sigma_{x},\sigma_{y},\sigma_{z})\left(\begin{array}[]{c}\overrightarrow{e}_{x}\\\ \overrightarrow{e}_{y}\\\ \overrightarrow{e}_{z}\end{array}\right),$ (1) where $\overrightarrow{e}_{x,y,z}$ are orthonormal vectors of the three coordinate axes to which the components $\sigma_{i}$ are referred and $\sigma_{i}$ are Pauli matrices satisfying the commutation relation $[\sigma_{i},\sigma_{j}]=2i\sigma_{k},$ where $i,j,k$ form an even permutation of $x,y,z$. In this case, the density operator can be represented in the following form [11] $\displaystyle\rho_{a}$ $\displaystyle=$ $\displaystyle\frac{1}{2}(1+\overrightarrow{s}\otimes{\sigma^{\downarrow}})$ (6) $\displaystyle=$ $\displaystyle\frac{1}{2}(1+s_{x}\sigma_{x}+s_{y}\sigma_{y}+s_{z}\sigma_{z})$ $\displaystyle=$ $\displaystyle\frac{1}{2}\left(\begin{array}[]{c}1+s_{z}\\\ s_{x}-is_{y}\end{array}\begin{array}[]{c}s_{x}+is_{y}\\\ 1-s_{z}\end{array}\right).$ where $\overrightarrow{s}=\langle{\overrightarrow{\sigma}\rangle}$. The effective Hamiltonian for one qubit can be defined as [14] $H=\alpha_{x}\sigma_{x}+\alpha_{y}\sigma_{y}+\alpha_{z}\sigma_{z},$ (7) where $\alpha_{i}$ are the external field components. The unitary evolution operator can be obtained from the Hamiltonian, thus, $\mathcal{U}=\sum_{i}\exp\\{-i\alpha_{i}t\sigma_{i}\\},$ (8) Using the density operator (2), Alice qubit can be transformed as $\rho_{a}\rightarrow\tilde{\rho}_{a}=\frac{1}{2}(1+\tilde{\mathord{\buildrel\lower 3.0pt\hbox{$\scriptscriptstyle\rightarrow$}\over{s}}}\otimes{\sigma^{\raisebox{2.0pt}[0.0pt]{$\scriptstyle\downarrow$}}}),$ (9) where the component of $\tilde{\mathord{\buildrel\lower 3.0pt\hbox{$\scriptscriptstyle\rightarrow$}\over{s}}}$ are given by $\displaystyle\tilde{s_{x}}$ $\displaystyle=$ $\displaystyle\frac{1}{3}\bigl{[}s_{x}(1+\cos(2t\alpha_{2})+\cos(2t\alpha_{3}))-s_{y}\sin(2t\alpha_{3})+s_{z}\sin(2t\alpha_{2})\bigr{]}$ $\displaystyle\tilde{s_{y}}$ $\displaystyle=$ $\displaystyle\frac{1}{3}\bigl{[}s_{y}(1+\cos(2t\alpha_{1})+\cos(2t\alpha_{3}))+s_{x}\sin(2t\alpha_{3})-s_{z}\sin(2t\alpha_{1})\bigr{]},$ $\displaystyle\tilde{s_{z}}$ $\displaystyle=$ $\displaystyle\frac{1}{3}\bigl{[}s_{z}(1+\cos(2t\alpha_{1})+\cos(2t\alpha_{2}))-s_{x}\sin(2t\alpha_{2})+s_{y}\sin(2t\alpha_{1})\bigr{]}.$ Having obtained the above analytical expressions for $\tilde{s_{x}},\tilde{s_{y}}$ and $\tilde{s_{z}}$, we are therefore in position to investigate the speed of the orthogonality and hence the speed of computation. To clarify our idea let us assume that the user Alice has encoded some information in her qubit which is defined by $\rho_{a}=\frac{1}{2}(1+s_{x}\sigma_{x}).$ (11) Using the time evolution of the unitary operator, (8) one can transform $\rho_{a}$ into $\tilde{\rho}_{a}$ from which the new Bloch vectors take the form, $\tilde{s_{x}}=\frac{s_{x}}{3}(1+\cos 2t\alpha_{2}+\cos 2t\alpha_{3}),\quad\tilde{s_{y}}=\frac{s_{x}}{3}\sin 2t\alpha_{3},\quad\tilde{s_{z}}=-\frac{s_{x}}{3}\sin 2t\alpha_{2}.$ (12) Let us assume that Alice has prepared her qubit such as $s_{x}=1$ and $s_{y}=s_{z}=0$. Then the eigenvectors of the (11) state can be written as $v_{1}=[-1,1],\quad\mbox{and\quad}v_{2}=[1,-1].$ (13) Thus, it will be easy to get the eigenvectors for the final state $\tilde{\rho}_{a}$, which is described by Bloch vectors (12). After some algebraic calculations, we can explicitly write $u_{i}$ as $u_{1}=u_{2}=\Gamma\biggl{\\{}[\sin^{2}2t\alpha_{2}-\left(3+\cos 2t\alpha_{2}+\cos 2t\alpha_{3}+\cos 2t\alpha_{2}\cos 2t\alpha_{3}\right)]\biggr{\\}},$ (14) where $\Gamma=(1+\cos 2t\alpha_{2}+\cos 2t\alpha_{3}-i\sin 2t\alpha_{3})^{-1}$. In order to facilitate our discussion let us define the scalar product of the vectors $u_{i}$ and $v_{i}$ such as $Sp_{ij}=\left\langle v_{i}|u_{j}\right\rangle.$ (15) Figure 1: The speed of orthogonality of qubit as a function of the scaled time, where, the component $\left\langle u_{i}|v_{1}\right\rangle$ is represented by the solid curve, while $\left\langle u_{i}|v_{2}\right\rangle$ is represented by the dotted-curve. The other parameters are $s_{x}=1$ and $s_{y}=s_{z}=0$, (a) $\alpha_{1}=\alpha_{2}=\alpha_{3}=\frac{\pi}{2}$, (b) $\alpha_{1}=\alpha_{2}=\alpha_{3}=\frac{\pi}{3}$, (c) $\alpha_{1}=\alpha_{2}=\alpha_{3}=\frac{\pi}{4}$ and (d) $\alpha_{1}=\frac{\pi}{2}$, $\alpha_{2}=\frac{\pi}{3}$ and $\alpha_{3}=\frac{\pi}{4}$. It should be noted that in our calculations we have taken into account all the possible products of $u_{i}$ and $v_{i}$. In figure (1) we have plotted the amplitude values of $Sp_{ij}$ against the scaled time to display its behavior for different values of the control parameter $\alpha_{i}.$ In figure $(1a)$ we have considered the case in which $\alpha_{i}=\pi/2$ where one can see both $\langle v_{1}|u_{j}\rangle$ and $\langle v_{2}|u_{j}\rangle$ are coincides on the horizontal axis at different period of time. However, when we change the value of the parameters $\alpha_{i}$ such as $\alpha_{i}=\pi/3$ it is noted that there is decreasing in the number of coincidences points which refer to reduction in the computation speed, see figure (1b). This phenomenon gets more pronounced for the case in which $\alpha_{i}=\pi/4$, see figure (1c). Thus we may conclude that as the value of the control parameters $\alpha_{i}$ increases as the speed of the computation increases and vise versa. On the other hand when we consider different values for the control parameters $\alpha_{i}$ such that $\alpha_{1}=\frac{\pi}{2},~{}\alpha_{2}=\frac{\pi}{3},~{}\alpha_{3}=\frac{\pi}{4}$, then more decreasing can be seen in the computation speed. In the meantime we can observe irregular fluctuations in both functions $Sp_{1j}$ and $Sp_{2j}$ in addition to the intersection at different points, see figure (1d). This is contrary to the previous cases where regular oscillations can be realized in each case. ## 3 Quantum treatment of Qubit Now let us turn our attention to consider the quantum treatment of the computation speed, taking into account the quantized field interacting with a single qubit. In this case the Hamiltonian can be written as, $\hat{H}_{int}=\lambda(\hat{a}^{\dagger}\sigma_{-}+\hat{a}\sigma_{+})+\frac{\Delta}{2}\sigma_{z},$ (16) where $\hat{a}^{\dagger}$ and $\hat{a}$ are the creation and annihilation operators satisfy the commutation relation $[\hat{a},\hat{a}^{\dagger}]=1.$ We denote by $\lambda$ the coupling constant and $\Delta$ the detuning parameter while $\sigma_{+}(\sigma_{-}),$ $\sigma_{z}$ are usual raising (lowering) and inversion operators for the two-level atomic system, satisfying $[\sigma_{z},\sigma_{\pm}]=\pm 2\sigma_{\pm}$ and $[\sigma_{+},\sigma_{-}]=2\sigma_{z}$ . The time-dependent density operator $\rho(t)$ is given by Figure 2: The same as figure 1, but $n=10,~{}\Delta/\gamma=2$ (a) $\eta=0.05$, (b) $\eta=0.1$, (c) $\eta=0.15$ and (d) $\eta=0.2$. $\rho(t)=\mathcal{U}(t)\rho(0)\mathcal{U}^{\dagger}(t),$ (17) where $\rho(0)=\rho_{a}(0)\otimes\rho_{f}(0)$ is the initial state of the system. We assume that the initial state of the qubit $\rho_{a}(0)$ is given by equation (7) while the field starts from a Fock state. Under the above assumptions the unitary evolution operator $\mathcal{U}(t)$ can be defined as $\mathcal{U}(t)=u_{11}\bigl{|}0\bigr{\rangle}\bigl{\langle}0\bigr{|}+u_{12}\bigl{|}1\bigr{\rangle}\bigl{\langle}0\bigr{|}+u_{21}\bigl{|}0\bigr{\rangle}\bigl{\langle}1\bigr{|}+u_{22}\bigl{|}1\bigr{\rangle}\bigl{\langle}1\bigr{|}$ (18) where $u_{11}=C_{n+1}-i\frac{\Delta}{2}S_{n+1},\quad u_{12}=-i\eta S_{n}a,\quad u_{21}=u_{12}^{\dagger},\quad u_{22}=u_{11}^{\dagger},$ and $S_{n}=\frac{\sin(\mu_{n}\gamma t)}{\mu_{n}},\qquad C_{n}=\cos(\mu_{n}\gamma t),\qquad\mu_{n}=\sqrt{\frac{\Delta^{2}}{4\gamma^{2}}+\eta^{2}n}.$ It should be noted that in the above equations, we have introduced the parameters $\eta$ and $\gamma$ to connect up the coupling parameter $\lambda$ such that $\lambda=\eta\gamma.$ This in fact would enable us to discuss the effect of the coupling parameter using $\eta$ instead of $\lambda$ regarding $\gamma$ as a dimensionless parameter. Using Eqs.(17) and (18) one can obtain the explicitly time-dependent density operator $\rho(t)$ in Bloch vectors representation, thus $\displaystyle\tilde{S_{x}}$ $\displaystyle=$ $\displaystyle-i\eta S_{n+1}\frac{1+s_{z}}{2}\bigl{[}\sqrt{n+1}(C_{n+1}+\frac{\Delta}{2}S_{n+1})+(C_{n+1}-\frac{\Delta}{2}S_{n+1})\bigr{]}$ $\displaystyle+\eta^{2}\sqrt{n+1}\sqrt{n+2}S_{n}S_{n+1}s_{x}+i\eta\sqrt{n+1}\frac{1-s_{z}}{2}S_{n}C_{n+2}$ $\displaystyle+(C_{n+1}^{2}-\frac{\Delta^{2}}{2}S_{n+1}^{2})s_{x}-\Delta S_{n+1}C_{n+1}s_{y},$ $\displaystyle\tilde{S_{y}}$ $\displaystyle=$ $\displaystyle\eta S_{n+1}\frac{1+s_{z}}{2}\bigl{[}\sqrt{n+1}(C_{n+1}+\frac{\Delta}{2}S_{n+1})+(C_{n+1}-\frac{\Delta}{2}S_{n+1})\bigr{]}$ $\displaystyle-\eta^{2}\sqrt{n+1}\sqrt{n+2}S_{n}S_{n+1}s_{y}+i\eta\Delta\sqrt{n+1}\frac{1-s_{z}}{2}S_{n}S_{n+2}$ $\displaystyle+\Delta S_{n+1}C_{n+1}s_{x}+(C_{n+2}^{2}-\frac{\Delta^{2}}{2}S_{n+1}^{2})s_{y},$ $\displaystyle\tilde{S_{z}}$ $\displaystyle=$ $\displaystyle-i\frac{\eta}{2}\bigl{[}(1+\sqrt{n+1})C_{n+1}+\frac{\Delta}{2}(1-\sqrt{n+1})S_{n+1}\biggr{]}s_{x}S_{n+1}$ (19) $\displaystyle+\frac{\eta}{2}\bigl{[}(1-\sqrt{n+1})C_{n+1}+\frac{\Delta}{2}(1+\sqrt{n+1})S_{n+1}\biggr{]}s_{y}S_{n+1}$ $\displaystyle-(C_{n+1}^{2}+\frac{\Delta^{2}}{4}S_{n+1})s_{z}+\eta^{2}S_{n}^{2}[\frac{1}{2}-(n+\frac{1}{2})s_{z}]$ $\displaystyle-i\eta\frac{s_{x}-is_{y}}{2}\sqrt{n+1}S_{n}C_{n+1}.$ The parameters $s_{x},s_{y}$ and $s_{z}$ which appear in the right hand side of equation (15), describe the initial state Bloch vectors. Figure 3: The same as figure (2a), but $n=20$, (a) $\Delta/\gamma=2$, (b) $\Delta/\gamma=1$. In fact, these states are widely used in the quantum information tasks. For example, one may consult a recent applications given in Ref. [15]. In the meantime it would be interesting to employ the Hamiltonian given by equation (12) that to discuss the speed of computation. This is extensively used in the field of quantum information to describe the interaction between field and qubit, particularly for investigating the loss of entanglement after propagation in a quantum noisy channel. Furthermore, compare with the classical treatment the interaction Hamiltonian contains three different parameters to control the dynamics of the system, $\Delta$ the detuning parameter, $\eta$ the coupling constant, and the mean photon number $\langle a^{\dagger}a\rangle$ [16]. These parameters are involved through the Rabi frequency $\lambda_{n}$ as well as in the Bloch vectors themselves. This would give us a wide range of variety to discuss the variation in $Sp_{ij}$ resultant of change one of these parameters. To do so we have numerically calculated the overlap between the initial and the final states of $Sp_{ij}$. For example, to see the effect of the coupling parameter $\eta$we have considered the number of photons $n=10$, and the detuning parameter $\Delta/\gamma=2,$ while $\eta=0.05.$ In this case and from figure (2a) we can see nearly perfect overlap between both of $Sp_{1j}$ and $Sp_{2j}$ as well as coincidences at the horizontal line showing high speed. Increase the value of $\eta$ such that $\eta=0.1$ leads to slight increase in the speed of computation, beside increases in one of the projectors value, see figure (2b). More increases in the coupling parameter $\eta=0.15$ shows increasing in the speed of computation but with less coincidences between the two projectors, see figure (2c). More increasing in the coupling value $\eta=0.2$ leads to more decreasing in the speed of computation but with regular increasing in both projectors value. This means that there is a certain value (critical value) of the coupling parameter where the speed of computation reaches its maximum and then starts to slow down. To examine the effect of the mean photon number we have considered the case in which $n=20,$ keeping the other parameters unchange as in figure (2a). In this case we observe no change in the speed of orthogonality and the behavior in general is the same as before, however, there is increasing in the amplitude for one of the projectors, see figure (3a). However, if we decrease the value of the detuning parameter $\Delta/\gamma=1$ drastic change can be realized. For instance, we can see decreasing in the number of the oscillations period, increasing in one of the projector amplitude, in addition to decrease in the speed of computation, see figure (3b). Thus we come to conclusion if one increases the value of the detuning parameter then the speed of the interaction increases. This result is in agreement with that given by Montangero [17], where they investigated the dynamics of entanglement in quantum computer with imperfections. ## 4 Conclusion In the above sections of the present paper we have considered the problem of speed computation in quantum information. The problem has been handled from two different point of view; where we have considered both of classical and quantum treatments. The main concentration was on how to improve and control the computation’s speed in each case separately. For the classical treatment it has been shown that the speed of computation is proportional with the total value of the external field. However, for quantum treatment we have seen that the speed of computation is sensitive to the variation of the coupling parameter and the detuning parameter. In the meantime we found the mean photon number does not play any role with the speed of computation but it is just effect the amplitude of the projectors. This in fact would turn our attention to look for the atom-atom interaction to be discussed in a forthcoming work. Acknowledgements: One of us (M.S.A) is grateful for the financial support from the project Math 2005/32 of the research center, College of Science, King Saud University. ## References * [1] V. N. Gorbachev, A. I. Trubilko, Laser Phys. Lett. 3, 59 (2005) * [2] C. H. Bennett and P. W. Shor, IEEE Trans. Inf. Theory 44, 2724, (1998). * [3] G. Brida, M. Genovese, M. Gramegna, Laser Phys. Lett. 3, 115 (2005); M. Abdel-Aty, Laser Phys. Lett. 1, 104 (2004). * [4] N. Margolus and L. B. Levitin, Physica D 120, 188 (1998); J. Batle, M. Casas, A. Plastino, A. R. Plastino, Phys. Rev. A 72, 032337 (2005); A. Borrs, M. Casas, A. R. Plastino and A. Plastino, Phys. Rev. A 74, 022326 (2006). * [5] T. Yu, J. H. Eberly, Phys. Rev. B 68, 165322 (2003). * [6] M. Ban, S. Kitajima and F. Shibata, Phys. Lett. A 349, 415 (2006); M. Ban, S. Kitajima and F. Shibata, J. Phys. A 38, 4235 (2005) * [7] M. Ban, Phys. Lett. A 361, 48 (2007). * [8] T. Yu. and J. H. Eberly, Optc. Commu. 264, 393 (2006). * [9] R. F. Sawyer, Phys. Rev. A 70, 022308 (2004) * [10] M.-H. Yung, Phys. Rev. A 74, 030303(R) (2006) * [11] D. de Falco and D. Tamascelli, J. Phys. A: Math. Gen. 39, 5873 (2006). * [12] M. S. Abdalla, M. Abdel-Aty and A.-S. F. Obada, Opt. Commu.211, 225 (2002). * [13] B.-G. Englert and N. Metwally, J. Mod. Opt. 47, 221 (2000); B.-G. Englert and N. Metwally, Appl. Phys. B 72, 35 (2001). * [14] M. Mottonen, R. de Sousa, J. Zhang and K. Whaley Phys. Rev. A 73, 022332 (2006) * [15] J. S. Nielsen, B. M. Nielson, C. Hehich, K. Moelmer and E. S. Polzik, Phys. Rev. Lett. 97, 083604 2006; Qing-Yu Cai and Yong-gang Tan, Phys. Rev. A 73, 032305 (2006); A. Qurjoumtser, R. T.-Brouri and P. Grangier, Phys. Rev. Lett. 96, 213601 (2006). * [16] W. A. Brito and R, V. Ramos, Phys. Lett. A 360, 251 (2006). * [17] S. Montangero, G. Beneti and R. Faazio, Phys. Rev. A 41, 060306 R (2005).

arxiv-papers

2024-09-04T02:48:53.555165

0802.0590

# Positive divisors in symplectic geometry Jianxun Hu & Yongbin Ruan Department of Mathematics Zhongshan University Guangzhou P. R. China stsjxhu@mail.sysu.edu.cn Department of Mathematics University of Michigan Ann Arbor, MI 48109-1109 and Yangtze Center of Mathematics Sichuan University Chengdu, 610064, P.R. China ruan@umich.edu 1Partially supported by the NSFC Grant 10631050, NKBRPC(2006CB805905) and NCET-04-0795 3supported by NSF Grant ###### Contents 1. 1 Introduction 2. 2 Preliminaries 1. 2.1 GW-invariants 2. 2.2 Relative GW-invariants 3. 2.3 Partial orderings on relative GW invariants 4. 2.4 Degeneration formula 3. 3 Relative GW-invariants of ${\mathbb{P}}^{1}$-bundles 1. 3.1 Fiber class invariants 2. 3.2 A vanishing theorem 3. 3.3 A nonvanishing theorem 4. 4 A comparison theorem 5. 5 Rationally connected symplectic divisors 1. 5.1 Rationally connectedness in algebraic geometry 2. 5.2 Rationally connected symplectic divisors ## 1\. Introduction Divisors or codimension two symplectic submanifolds play an important role in Gromov-Witten theory and symplectic geometry. For example, there is now a well-known degeneration operation to decompose a symplectic manifold $X$ into $X_{1}\cup_{Z}X_{2}$ a union of $X_{1},X_{2}$, along a common divisor $Z$. To utilize the above degeneration to compute Gromov-Witten invariants inductively, one needs to develop the relative Gromov-Witten theory of the pair $(X,Z)$. Such a theory and its degeneration formula were first constructed by Li-Ruan [LR] ( see Ionel and Parker [IP] for a different version and J. Li [Li1, Li2] for an algebraic treatment). Maulik and Pandharipande [MP] systematically studied the degeneration formula for the degeneration $X\rightarrow X\cup_{Z}P_{Z}$, where $P_{Z}$ is the projective closure of the normal bundle $N_{Z}$. By introducing a certain partial order on relative invariants, they interpreted the degeneration formula as a “correspondence”, a complicated upper triangular linear map from relative invariants to absolute invariants. One consequence of their correspondence is that the absolute and relative invariants determine each other. Such a correspondence is a very powerful tool in determining the totality of Gromov-Witten theory. But it is not effective in determining any single invariant. A natural question is if a divisor with a stronger condition will give us a much stronger correspondence? We answer this affirmatively for so-called positive symplectic divisors. We call a symplectic divisor $Z$ positive if for some tamed almost complex structure $J$, $C_{1}(N_{Z})(A)>0$ for any $A$ represented by a non-trivial $J$-sphere in $Z$. This is a generalization of ample divisor from algebraic geometry. One of our main theorems is the following comparison theorem. We call $0\neq A\in H_{2}(Z,{\mathbb{Z}})$ stably effective if there is a nonzero genus zero Gromov-Witten invariant of $Z$ with class $A$. Let $\pi:P_{Z}\rightarrow Z$ be the projection. Maulik and Pandharipande show that any nonzero absolute or relative Gromov-Witten invariant of $P_{Z}$ is determined by the Gromov-Witten invariants of $Z$. It is a direct consequence of their argument that if a nonzero genus zero Gromov-Witten or relative Gromov-Witten invariant of $P_{Z}$ has class $B$, then $\pi_{*}(B)=\sum_{i}a_{i}A_{i}$ for stably effective $A_{i}\in H_{2}(Z,{\mathbb{Z}})$ and $a_{i}\in{\mathbb{Z}}_{+}$. Let $V=\min\,\\{C_{1}(N_{Z}(A))\mid A\in H_{2}(Z,{\mathbb{Z}})\,\mbox{is stably effective}\\}$. ###### Theorem 1.1. Suppose that $Z$ is a positive divisor and $V\geq l$. Then for $A\in H_{2}(X,{\mathbb{Z}})$, $\alpha_{i}\in H^{*}(X,{\mathbb{R}})$, $1\leq i\leq\mu$, and $\beta_{j}\in H^{*}(Z,{\mathbb{R}})$, $1\leq j\leq l$, we have $\displaystyle\langle\alpha_{1},\cdots,\alpha_{\mu},\iota^{!}(\beta_{1}),\cdots,\iota^{!}(\beta_{l})\rangle^{X}_{A}$ $\displaystyle=$ $\displaystyle\sum_{\mathcal{T}}\langle\alpha_{1},\cdots,\alpha_{\mu}\mid{\mathcal{T}}\rangle_{A}^{X,Z},$ where $\iota:Z\hookrightarrow X$ and $\iota^{!}=PD\iota_{*}PD$, the summation runs over all possible weighted partitions ${\mathcal{T}}=\\{(1,\gamma_{1})$, $\cdots,(1,\gamma_{q}),(1,[Z]),\cdots,(1,[Z])\\}$ of $Z\cdot A$, where $\gamma_{i}$’s are the products of some $\beta_{j}$ classes. McDuff [M1] also considered the similar comparison result in the some special case. For readers familiar with Maulik-Pandharipande’s relative/absolute correspondence, the above theorem means that there is no lower order term in the degeneration formula. The second motivation comes from symplectic birational geometry. A fundamental problem in symplectic geometry is to generalize birational geometry to symplectic geometry. In the 80’s, Mori introduced a program towards the birational classification of algebraic manifolds of dimension three and up. In 90’s, the last author [R1] speculated that there should be a symplectic geometric program. First of all, since there is no notion of regular or rational function in symplectic geometry, we must therefore first define the appropriate notion of symplectic birational equivalence. For this purpose, in [HLR], the authors proposed to use Guillemin-Sternberg’s birational cobordism to replace birational maps. Secondly, we need to study what geometric properties behave nicely under this birational cobordism. In [HLR], the authors defined the uniruledness of symplecitc manifolds by requiring a nonzero Gromov-Witten invariant with a point insertion and settled successfully the fundamental birational cobordism invariance of uniruledness. McDuff [M2] proved that Hamiltonian $S^{1}$-manifolds are uniruled using the techniques from [HLR]. Furthermore, Tian-Jun Li and the second author [LtjR] investigate the dichotomy of uniruled symplectic divisors. The dichotomy asserts that if the normal bundle is non-negative in some sense, then the ambient manifold is uniruled. It is clear that our stronger comparison theorem should yield stronger results. This is indeed the case. As an application of our comparison theorem, we investigate symplectic rationally connected manifolds. Similar to the case of uniruledness (see [HLR]), we define the notion of $k$-point (strongly) rational connectedness by requiring a non-zero (primary) Gromov-Witten invariant with $k$ point insertions (see section five for the detailed discussion). Of course, one important problem here of interest is whether the notion of symplectic $k$-point rational connectedness is invariant under birational cobordism given in [HLR]. From the blowup formula of [H1, H2, H3, HZ, La], we know that symplectic rational connectedness is invariant under the symplectic blowup along points and some special submanifolds with convex normal bundles. The general case is still unknown. We should mention that a longstanding problem in Gromov-Witten theory is to characterize algebraic rationally connectedness in terms of Gromov-Witten theory. Our second main theorem is the following theorem analogous to the theorem of McDuff [M3, LtjR] for uniruled divisors. ###### Theorem 1.2. Let $(X,\omega)$ be a compact $2n$ dimensional symplectic manifold which contains a submanifold $P$ symplectomorphic to ${\mathbb{P}}^{n-1}$ whose normal Chern number $m\geq 2$, then $X$ is strongly rationally connected. The paper is organized as follows. In section two, we first review Gromov- Witten theory and its degeneration formula to set up the notation. In Section three, we prove some vanishing and non-vanishing results for relative Gromov- Witten invariants of ${\mathbb{P}}^{1}$-bundles. In section four, we prove a comparison theorem between absolute and relative Gromov-Witten invariants. In section five, we generalize the from divisor to ambient space inductive construction of [LtjR] to the case $k$-point rational connectedness. Acknowledgements: Both of the authors would like to thank Zhenbo Qin who took part in the early time of this work. This article was prepared during the first author’s visit to the Department of Mathematics, University of Michigan- Ann Arbor and the Department of Mathematics, University of Missouri-Columbia. The first author is grateful to the departments for their hospitality. The authors would like to thank Tian-Jun Li for his valuable suggestions and telling us about McDuff’s work. Thanks also to A. Greenspoon for his editorial assistance. ## 2\. Preliminaries In this section, we want to review briefly the constructions of virtual integration in the definitions of the absolute and relative GW invariants, which are the main tool of our paper. We refer to [R1, LR] for the details. ### 2.1. GW-invariants Suppose that $(X,\omega)$ is a compact symplectic manifold and $J$ is a tamed almost complex structure. ###### Definition 2.1. A stable $J-$holomorphic map is an equivalence class of pairs $(\Sigma,f)$. Here $\Sigma$ is a connected nodal marked Riemann surface with arithmetic genus $g$, $k$ smooth marked points $x_{1},...,x_{k}$, and $f:\Sigma\longrightarrow X$ is a continuous map whose restriction to each component of $\Sigma$ (called a component of $f$ in short) is $J$-holomorphic. Furthermore, it satisfies the stability condition: if $f|_{S^{2}}$ is constant (called a ghost bubble) for some $S^{2}-$component, then the $S^{2}-$component has at least three special points (marked points or nodal points). $(\Sigma,f)$, $(\Sigma^{\prime},f^{\prime})$ are equivalent, or $(\Sigma,f)\sim(\Sigma^{\prime},f^{\prime})$, if there is a biholomorphic map $h:\Sigma^{\prime}\longrightarrow\Sigma$ such that $f^{\prime}=f\circ h$. An essential feature of Definition 2.1 is that, for a stable $J-$holomorphic map $(\Sigma,f)$, the automorphism group $\mbox{Aut}(\Sigma,f)=\\{h\mid h\circ(\Sigma,f)=(\Sigma,f)\\}$ is finite. We define the moduli space $\overline{\mathcal{M}}^{X}_{A}(g,k,J)$ to be the set of equivalence classes of stable $J-$holomorphic maps such that $[f]=f_{*}[\Sigma]=A\in H_{2}(X,{\bf Z})$. The virtual dimension of $\overline{\mathcal{M}}^{X}_{A}(g,k,J)$ is computed by index theory, $\mbox{virdim}_{\mathbb{R}}\overline{\mathcal{M}}^{X}_{A}(g,k,J)=2c_{1}(A)+2(n-3)(1-g)+2k,$ where $n$ is the complex dimension of $X$. Unfortunately, $\overline{\mathcal{M}}^{X}_{A}(g,k,J)$ is highly singular and may have larger dimension than the virtual dimension. To extract invariants, we use the following virtual neighborhood method. First, we drop the $J$-holomorphic condition from the previous definition and require only that each component of $f$ be smooth. We call the resulting object a stable map or a $C^{\infty}$-stable map. Denote the corresponding space of equivalence classes by $\overline{\mathcal{B}}^{X}_{A}(g,k,J)$. $\overline{\mathcal{B}}^{X}_{A}(g,k,J)$ is clearly an infinite dimensional space. It has a natural stratification given by the topological type of $\Sigma$ together with the fundamental classes of the components of $f$. The stability condition ensures that $\overline{\mathcal{B}}^{X}_{A}(g,k,J)$ has only finitely many strata such that each stratum is a Fréchet orbifold. Further, one can use the pregluing construction to define a topology on $\overline{\mathcal{B}}^{X}_{A}(g,k,J)$ which is Hausdorff and makes $\overline{\mathcal{M}}^{X}_{A}(g,k,J)$ a compact subspace. (see [R1]). We can define another infinite dimensional space $\Omega^{0,1}$ together with a map $\pi:\Omega^{0,1}\longrightarrow\overline{\mathcal{B}}^{X}_{A}(g,k,J)$ such that the fiber is $\pi^{-1}(\Sigma,f)=\Omega^{0,1}(f^{*}TX)$. The Cauchy- Riemann operator is now interpreted as a section of $\pi:\Omega^{0,1}\longrightarrow\overline{\mathcal{B}}^{X}_{A}(g,k,J)$, $\overline{\partial}_{J}:\overline{\mathcal{B}}^{X}_{A}(g,k,J)\to\Omega^{0,1}$ with $\overline{\partial}_{J}^{-1}(0)$ nothing but $\overline{\mathcal{M}}^{X}_{A}(g,k,J)$. At each $(\Sigma,f)\in\overline{\mathcal{M}}^{X}_{A}(g,k,J)$, there is a canonical decomposition of the tangent space of $\Omega^{0,1}$ into the horizontal piece and the vertical piece. Thus we can linearize $\overline{\partial}_{J}$ with respect to deformations of stable maps and project to the vertical piece to obtain an elliptic complex (1) $L_{\Sigma,f}:\Omega^{0}(f^{*}TX)\longrightarrow\Omega^{0,1}(f^{*}TX).$ Several explanations are in order for formula $(\ref{virtual-0})$. Choose a compatible Riemannian metric on $X$ and let $\nabla$ be the Levi-Civita connection. When $\Sigma$ is irreducible, $\nabla$ induces a connection on $f^{*}TX$, still denoted by $\nabla$. Then $L_{\Sigma,f}=\overline{\nabla}$, where $\overline{\nabla}$ is the projection of $\nabla$ onto the $(0,1)$-factor. When $\Sigma$ is reducible, formula (1) is interpreted as follows. For simplicity, suppose that $\Sigma$ is the union of $\Sigma_{1}$ and $\Sigma_{2}$ intersecting at $p\in\Sigma_{1}$ and $q\in\Sigma_{2}$. Let the corresponding maps be $f_{1}$ and $f_{2}$. Then define $\Omega^{0}(f^{*}TX)=\\{(v_{1},v_{2})\in\Omega^{0}(f^{*}_{1}TX)\times\Omega^{0}(f^{*}_{2}TX)\mid v_{1}(p)=v_{2}(q)\\}$ and $\Omega^{0,1}(f^{*}TX)=\Omega^{0,1}(f^{*}_{1}TX)\oplus\Omega^{0,1}(f^{*}_{2}TX).$ $L_{\Sigma,f}$ is then the restriction of $L_{\Sigma_{1},f_{1}}\oplus L_{\Sigma_{2},f_{2}}$ to $\Omega^{0}$. It is clear that $\mbox{Ker}L_{\Sigma,f}=\\{(v_{1},v_{2})\in\mbox{Ker}\,L_{\Sigma_{1},f_{1}}\times\mbox{Ker}\,L_{\Sigma_{2},f_{2}}\mid v_{1}(p)=v_{2}(q)\\}.$ To understand $\mbox{Coker}\,L_{\Sigma,f}$ geometrically, it is convenient to use another elliptic complex. The idea is as follows. We would like to drop the condition $v_{1}(p)=v_{2}(q)$. To keep the index unchanged, we need to enlarge $\Omega^{0,1}(f^{*}TX)$. A standard method motivated by algebraic geometry is to allow a simple pole at the intersection point. This leads to $\tilde{L}_{\Sigma,f}:\tilde{\Omega}^{0}(f^{*}TX)\longrightarrow\tilde{\Omega}^{0,1}(f^{*}TX),$ where $\displaystyle\tilde{\Omega}^{0}(f^{*}TX)$ $\displaystyle=$ $\displaystyle\Omega^{0}(f_{1}^{*}TX)\times\Omega^{0}(f_{2}^{*}TX),$ $\displaystyle\tilde{\Omega}^{0,1}(f^{*}TX)$ $\displaystyle=$ $\displaystyle\\{(v_{1},v_{2})\in\Omega^{0,1}(f_{1}^{*}TX\otimes{\mathcal{O}}(p))\times\Omega^{0,1}(f_{1}^{*}TX\otimes{\mathcal{O}}(q))\mid$ $\displaystyle\mbox{Res}_{p}v_{1}+\mbox{Res}_{q}v_{2}=0\\},$ and $\tilde{L}_{\Sigma,f}=\tilde{L}_{\Sigma_{1},f_{1}}\oplus\tilde{L}_{\Sigma_{2},f_{2}}.$ It is well-known that $\mbox{Ker}\,L_{\Sigma,f}\cong\mbox{Ker}\,\tilde{L}_{\Sigma,f},\hskip 28.45274pt\mbox{Coker}\,L_{\Sigma,f}\cong\mbox{Coker}\,\tilde{L}_{\Sigma,f}.$ Therefore $\displaystyle\mbox{Coker}\,L_{\Sigma,f}=$ $\displaystyle\\{(v_{1},v_{2})\in\mbox{Coker}\,\tilde{L}_{\Sigma_{1},f_{1}}\times\mbox{Coker}\,\tilde{L}_{\Sigma_{2},f_{2}}\mid$ $\displaystyle\mbox{Res}_{p}v_{1}+\mbox{Res}_{q}v_{2}=0\\}.$ If $\Sigma$ has more than two components, the construction above extends in a straightforward fashion. To consider the full linearization of $\overline{\partial}_{J}$, we have to include the deformation space $Def(\Sigma)$ of the nodal marked Riemann surface $\Sigma$. $Def(\Sigma)$ fits into the short exact sequence, $0\longrightarrow H^{1}(T\Sigma)\longrightarrow Def(\Sigma)\longrightarrow T_{p}\Sigma_{1}\otimes T_{q}\Sigma_{2}\longrightarrow 0,$ where the first term represents the deformation space of $\Sigma$ preserving the nodal point and the third term represents the smoothing of the nodal point. Moreover, $H^{1}(T\Sigma)$ is a product, with each factor being the deformation space of a component while treating the nodal point as a new marked point. The full linearization of $\overline{\partial}_{J}$ is given by $L_{\Sigma,f}\oplus\frac{1}{2}Jdf$. Denote $\displaystyle Def(\Sigma,f)$ $\displaystyle=\mbox{Ker}(L_{\Sigma,f}\oplus\frac{1}{2}Jdf)/T_{e}\mbox{Aut}(\Sigma),$ $\displaystyle Obs(\Sigma,f)$ $\displaystyle=\mbox{Coker}(L_{\Sigma,f}\oplus\frac{1}{2}Jdf).$ When $Obs(\Sigma,f)=0$, $(\Sigma,f)$ is a smooth point of the moduli space and $Def(\Sigma,f)$ is its tangent space. Now we choose a nearby symplectic form $\omega^{\prime}$ such that $\omega^{\prime}$ is tamed with $J$ and $[\omega^{\prime}]$ is a rational cohomology class. Using $\omega^{\prime}$, B. Siebert [S1] (see also the appendix in [R1]) constructed a natural finite dimensional vector bundle over $\overline{\mathcal{B}}_{A}(g,k,J)$. It has the property of dominating any local finite dimensional orbifold bundle as follows. Let $U$ be a neighborhood of $(\Sigma,f)\in\overline{\mathcal{B}}_{A}(g,k,J)$ and $F_{U}$ be an orbifold bundle over $U$. Then Siebert constructed a bundle $E$ over $\overline{\mathcal{B}}_{A}(g,k,J)$ such that there is a surjective bundle map $E|_{U}\longrightarrow F_{U}$. For each $(\Sigma,f)\in\overline{\mathcal{M}}_{A}(g,k,J)$, $Obs(\Sigma,f)$ extends to a local orbifold bundle $F(\Sigma,f)$ over a neighborhood $\tilde{\mathcal{U}}_{\Sigma,f}$ of $(\Sigma,f)$. Then we use Siebert’s construction to find a global orbifold bundle ${\mathcal{E}}(\Sigma,f)$ dominating $F(\Sigma,f)$. In fact, any global orbifold bundle with this property will work. We also remark that it is often convenient to replace $Obs(\Sigma,f)$ by $\mbox{Coker}\,L_{\Sigma,f}$ in the construction. Over each $\tilde{\mathcal{U}}_{\Sigma,f}$, by the domination property, we can construct a stabilizing term $\eta_{\Sigma,f}:{\mathcal{E}}(\Sigma,f)\longrightarrow\Omega^{0,1}$ supported in $\tilde{\mathcal{U}}_{\Sigma,f}$ such that $\eta_{\Sigma,f}$ is surjective onto $Obs(\Sigma,f)$ at $(\Sigma,f)$. Obviously, $\eta_{\Sigma,f}$ can be viewed as a map from ${\mathcal{E}}(\Sigma,f)$ to $\Omega^{0,1}$. Then the stabilizing equation $\overline{\partial}_{J}+\eta_{\Sigma,f}:{\mathcal{E}}(\Sigma,f)\to\Omega^{0,1},\quad((\Sigma^{\prime},f^{\prime}),e)\to\overline{\partial}_{J}f^{\prime}+\eta_{\Sigma,f}(e)$ has no cokernel at $(\Sigma,f)$. By semicontinuity, it has no cokernel in a neighborhood ${\mathcal{U}}_{\Sigma,f}\subset\tilde{\mathcal{U}}_{\Sigma,f}$. Since $\overline{\mathcal{M}}^{X}_{A}(g,k,J)$ is compact, there are finitely many ${\mathcal{U}}_{\gamma}={\mathcal{U}}_{\Sigma_{\gamma},f_{\gamma}}$ covering $\overline{\mathcal{M}}^{X}_{A}(g,k,J)$. Let ${\mathcal{U}}=\cup_{\gamma}{\mathcal{U}}_{\gamma},\hskip 14.22636pt{\mathcal{E}}=\oplus_{\gamma}{\mathcal{E}}(\Sigma_{\gamma},f_{\gamma}),\hskip 14.22636pt\eta=\sum_{\gamma}\eta_{\Sigma_{\gamma},f_{\gamma}}.$ Consider the finite dimensional vector bundle over ${\mathcal{U}}$, $p:{\mathcal{E}}|_{\mathcal{U}}\longrightarrow{\mathcal{U}}$. The stabilizing equation $\overline{\partial}_{J}+\eta$ can be interpreted as a section of the bundle $p^{*}\Omega^{0,1}\to{\mathcal{E}}|_{\mathcal{U}}$. By construction, this section $\overline{\partial}_{J}+\eta:{\mathcal{E}}|_{\mathcal{U}}\to p^{*}\Omega^{0,1}$ is transverse to the zero section of $p^{*}\Omega^{0,1}\to{\mathcal{E}}|_{\mathcal{U}}$. The set $U^{X}_{{\mathcal{S}}_{e}}=(\overline{\partial}_{J}+\eta)^{-1}(0)$ is called the virtual neighborhood in [R1]. The heart of [R1] is to show that $U^{X}_{{\mathcal{S}}_{e}}$ has the structure of a $C^{1}-$manifold. Notice that $U^{X}_{{\mathcal{S}}_{e}}\subset{\mathcal{E}}|_{\mathcal{U}}$. Over $U^{X}_{{\mathcal{S}}_{e}}$ there is the tautological bundle ${\mathcal{E}}_{X}=p^{*}({\mathcal{E}}|_{U})|_{U^{X}_{{\mathcal{S}}_{e}}}.$ It comes with the tautological inclusion map $S_{X}:U^{X}_{{\mathcal{S}}_{e}}\longrightarrow{\mathcal{E}}_{X},\quad((\Sigma^{\prime},f^{\prime}),e)\to e,$ which can be viewed as a section of ${\mathcal{E}}_{X}$. It is easy to check that $S^{-1}_{X}(0)=\overline{\mathcal{M}}^{X}_{A}(g,k,J).$ Furthermore, one can show that $S_{X}$ is a proper section. Note that the stratification of $\overline{\mathcal{B}}_{A}(g,k,J)$ induces a natural stratification of ${\mathcal{E}}$. We can define $\eta_{\gamma}=\eta_{\Sigma_{\gamma},f_{\gamma}}$ inductively from lower stratum to higher stratum. For example, we can first define $\eta_{\gamma}$ on a stratum and extend to a neighborhood. Then we define $\eta_{\gamma+1}$ at the next stratum supported away from lower strata. One consequence of this construction is that $U^{X}_{{\mathcal{S}}_{e}}$ has the same stratification as that of $\mathcal{E}$. Namely, if ${\mathcal{B}}_{D^{\prime}}\subset\overline{\mathcal{B}}_{D}$ is a lower stratum, $U^{X}_{{\mathcal{S}}_{e}}\cap{\mathcal{E}}|_{{\mathcal{B}}_{D^{\prime}}}\subset U^{X}_{{\mathcal{S}}_{e}}\cap{\mathcal{E}}|_{\overline{\mathcal{B}}_{D}}$ is a submanifold of codimension at least $2$. There are evaluation maps $ev_{i}:\overline{\mathcal{B}}_{A}(g,k,J)\longrightarrow X,\quad(\Sigma,f)\to f(x_{i}),$ for $1\leq i\leq k$. $ev_{i}$ induces a natural map from $U_{{\mathcal{S}}_{e}}\longrightarrow X^{k}$, which can be shown to be smooth. Let $\Theta$ be the Thom form of the finite dimensional bundle ${\mathcal{E}}_{X}\to U^{X}_{{\mathcal{S}}_{e}}$. ###### Definition 2.2. The (primary) Gromov-Witten invariant is defined as $\langle\alpha_{1},\cdots,\alpha_{k}\rangle^{X}_{g,A}=\int_{U_{{\mathcal{S}}_{e}}}S_{X}^{*}\Theta\wedge\Pi_{i}ev_{i}^{*}\alpha_{i},$ where $\alpha_{i}\in H^{*}(X;{\mathbb{R}})$. For the genus zero case, we also write $\langle\alpha_{1},\cdots,\alpha_{k}\rangle^{X}_{A}=\\\ \langle\alpha_{1},\cdots,\alpha_{k}\rangle^{X}_{0,A}$. ###### Definition 2.3. For each marked point $x_{i}$, we define an orbifold complex line bundle ${\mathcal{L}}_{i}$ over $\overline{\mathcal{B}}^{X}_{A}(g,k,J)$ whose fiber is $T_{x_{i}}^{*}\Sigma$ at $(\Sigma,f)$. Such a line bundle can be pulled back to $U^{X}_{{\mathcal{S}}_{e}}$ (still denoted by ${\mathcal{L}}_{i}$). Denote $c_{1}({\mathcal{L}}_{i})$, the first Chern class of ${\mathcal{L}}_{i}$, by $\psi_{i}$. ###### Definition 2.4. The descendent Gromov-Witten invariant is defined as $\langle\tau_{d_{1}}\alpha_{1},\cdots,\tau_{d_{k}}\alpha_{k}\rangle^{X}_{g,A}=\int_{U^{X}_{{\mathcal{S}}_{e}}}S_{X}^{*}\Theta\wedge\Pi_{i}\psi_{i}^{d_{i}}\wedge ev_{i}^{*}\alpha_{i},$ where $\alpha_{i}\in H^{*}(X;{\mathbb{R}})$. ###### Remark 2.5. In the stable range $2g+k\geq 3$, one can also define non-primary GW invariants (See e.g. [R1]). Recall that there is a map $\pi:\overline{\mathcal{B}}^{X}_{A}(g,k,J)\rightarrow\overline{\mathcal{M}}_{g,k}$ contracting the unstable components of the source Riemann surface. We can introduce a class $\kappa$ from the Deligne-Mumford space via $\pi$ to define the ancestor GW invariants $\langle\kappa\mid\Pi_{i}\alpha_{i}\rangle^{X}_{g,A}=\int_{U^{X}_{{\mathcal{S}}_{e}}}S_{X}^{*}\Theta\wedge\pi^{*}\kappa\wedge\Pi_{i}ev_{i}^{*}\alpha_{i}.$ The primary Gromov-Witten invariants are the special invariants with the point class in $\overline{\mathcal{M}}_{0,k}$. ###### Remark 2.6. For computational purpose we would mention the following variation of the virtual neighborhood construction. Suppose $\iota:D\subset X$ is a submanifold. For $\alpha\in H^{*}(D;{\mathbb{R}})$ we define $\iota^{!}(\alpha)\in H^{*}(X;{\mathbb{R}})$ via the transfer map $\iota^{!}=PD_{X}\circ\iota_{*}\circ PD_{D}$. One can construct Gromov-Witten invariants with an insertion of the form $\iota^{!}(\alpha)$ as follows. Apply the virtual neighborhood construction to the compact subspace $\overline{\mathcal{M}}_{A}(g,k,J)\cap ev_{1}^{-1}(D)$ in $\overline{\mathcal{B}}^{X}_{A}(g,k,J,D)=ev_{1}^{-1}(D)$ to obtain a virtual neighborhood ${\mathcal{U}}_{{\mathcal{S}}_{e}}(D)$ together with the natural map $ev_{D}:{\mathcal{U}}_{{\mathcal{S}}_{e}}(D)\longrightarrow D$. It is easy to show that $\displaystyle\langle\tau_{d_{1}}\iota^{!}(\alpha),\tau_{d_{2}}\beta_{2},\cdots,\tau_{d_{k}}\beta_{k}\rangle^{X}_{g,A}=\int_{U_{{\mathcal{S}}_{e}}(D)}S^{*}\Theta\wedge ev_{D}^{*}\alpha\wedge\prod_{i=2}^{k}\psi_{i}^{d_{i}}ev_{i}^{*}\beta_{i}.$ ###### Remark 2.7. For each $\langle\tau_{d_{1}}\alpha_{1},\cdots,\tau_{d_{k}}\alpha_{k}\rangle^{X}_{g,A}$, we can conveniently associate a simple graph $\Gamma$ of one vertex decorated by $(g,A)$ and a tail for each marked point. We then further decorate each tail by $(d_{i},\alpha_{i})$ and call the resulting graph $\Gamma(\\{(d_{i},\alpha_{i})\\})$ a weighted graph. Using the weighted graph notation, we denote the above invariant by $\langle\Gamma(\\{(d_{i},\alpha_{i})\\})\rangle^{X}$. We can also consider the disjoint union $\Gamma^{\bullet}$ of several such graphs and use $A_{\Gamma^{\bullet}},g_{\Gamma^{\bullet}}$ to denote the total homology class and total arithmetic genus. Here the total arithmetic genus is $1+\sum(g_{i}-1)$. Then, we define $\langle\Gamma^{\bullet}(\\{(d_{i},\alpha_{i})\\})\rangle^{X}$ as the product of Gromov-Witten invariants of the connected components. ### 2.2. Relative GW-invariants In this section, we will review the relative GW-invariants. The readers can find more details in the reference [LR]. Let $Z\subset X$ be a real codimension $2$ symplectic submanifold. Suppose that $J$ is an $\omega-$tamed almost complex structure on $X$ preserving $TZ$, i.e. making $Z$ an almost complex submanifold. The relative Gromov-Witten invariants are defined by counting the number of stable $J-$holomorphic maps intersecting $Z$ at finitely many points with prescribed tangency. More precisely, fix a $k$-tuple $T_{k}=(t_{1},\cdots,t_{k})$ of positive integers, consider a marked pre-stable curve $(C,x_{1},\cdots,x_{m},y_{1},\cdots,y_{k})$ and stable $J-$holomorphic map $f:C\longrightarrow X$ such that the divisor $f^{*}Z$ is $f^{*}Z=\sum_{i}t_{i}y_{i}.$ One would like to consider the moduli space of such curves and apply the virtual neighborhood technique to construct the relative invariants. But this scheme needs modification as the moduli space is not compact. It is true that for a sequence of $J-$holomorphic maps $(\Sigma_{n},f_{n})$ as above, by possibly passing to a subsequence, $f_{n}$ will still converge to a stable $J$-holomorphic map $(\Sigma,f)$. However the limit $(\Sigma,f)$ may have some $Z-$components, i.e. components whose images under $f$ lie entirely in $Z$. To deal with this problem the authors in [LR] adopt the open cylinder model. Choose a Hamiltonian $S^{1}$ function $H$ in a closed $\epsilon-$symplectic tubular neighborhood $X_{0}$ of $Z$ with $H(X_{0})=[-\epsilon,0]$ and $Z=H^{-1}(-\epsilon)$. Next we need to choose an almost complex structure with nice properties near $Z$. An almost complex structure $J$ on $X$ is said to be tamed relative to $Z$ if $J$ is $\omega$-tamed, $S^{1}-$invariant for some $(X_{0},H)$, and such that $Z$ is an almost complex submanifold. The set of such $J$ is nonempty and forms a contractible space. With such a choice of almost complex structure, $X_{0}$ can be viewed as a neighborhood of the zero section of the complex line bundle $N_{Z|X}$ with the $S^{1}$ action given by the complex multiplication $e^{2\pi i\theta}$. Now we remove $Z$. The end of $X-Z$ is simply $X_{0}-Z$. Recall that the punctured disc $D-\\{0\\}$ is biholomorphic to the half cylinder $S^{1}\times[0,\infty)$. Therefore, as an almost complex manifold, $X_{0}-Z$ can be viewed as the translation invariant almost complex half cylinder $P\times[0,\infty)$ where $P=H^{-1}(0)$. In this sense, $X-Z$ is viewed as a manifold with almost complex cylinder end. Now we consider a holomorphic map in the cylinder model where the marked points mapped into $Z$ are removed from the domain surface. Again we can view a punctured neighborhood of each of these marked points as a half cylinder $S^{1}\times[0,\infty)$. With such a $J$, a $J-$holomorphic map of $X$ intersecting $Z$ at finitely many points then exactly corresponds to a $J-$holomorphic map to the open manifold $X-Z$ from a punctured Riemann surface which converges to (a multiple of) an $S^{1}-$orbit at a puncture point. Now we reconsider the convergence of $(\Sigma_{n},f_{n})$ in the cylinder model. The creation of a $Z-$component $f_{\nu}$ corresponds to disappearance of a part of im$(f_{n})$ to infinity. We can use translation to rescale back the missing part of im$(f_{n})$. In the limit, we may obtain a stable map $\tilde{f}_{\nu}$ into $P\times{\mathbb{R}}$. When we obtain $X$ from the cylinder model, we need to collapse the $S^{1}$-action at infinity. Therefore, in the limit, we need to take into account maps into the closure of $P\times{\mathbb{R}}$. Let $Y$ be the projective completion of the normal bundle $N_{Z|X}$, i.e. $Y={\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})$. Then $Y$ has a zero section $Z_{0}$ and an infinity section $Z_{\infty}$. We view $Z\subset X$ as the zero section. One can further show that $\tilde{f}_{\nu}$ is indeed a stable map into $Y$ with the stability specified below. To form a compact moduli space of such maps we thus must allow the target $X$ to degenerate as well (compare with [Li1]). For any non-negative integer $m$, construct $Y_{m}$ by gluing together $m$ copies of $Y$, where the infinity section of the $i^{th}$ component is glued to the zero section of the $(i+1)^{st}$ component for $1\leq i\leq m$. Denote the zero section of the $i^{th}$ component by $Z_{i-1}$, and the infinity section by $Z_{i}$, so Sing $Y_{m}=\cup_{i=1}^{m-1}Z_{i}$. We will also denote $Z_{m}$ by $Z_{\infty}$ if there is no possible confusion. Define $X_{m}$ by gluing $X$ along $Z$ to $Y_{m}$ along $Z_{0}$. Thus Sing $X_{m}=\cup_{i=0}^{m-1}Z_{i}$ and $X_{0}=X$. $X_{0}=X$ will be referred to as the root component and the other irreducible components will be called the bubble components. Let $\mbox{Aut}_{Z}Y_{m}$ be the group of automorphisms of $Y_{m}$ preserving $Z_{0}$, $Z_{m}$, and the morphism to $Z$. And let $\mbox{Aut}_{Z}X_{m}$ be the group of automorphisms of $X_{m}$ preserving $X$ (and $Z$) and with restriction to $Y_{m}$ being contained in $\mbox{Aut}_{Z}Y_{m}$ (so $\mbox{Aut}_{Z}X_{m}=\mbox{Aut}_{Z}Y_{m}\cong({\mathbb{C}}^{*})^{m}$, where each factor of $({\mathbb{C}}^{*})^{m}$ dilates the fibers of the ${\mathbb{P}}^{1}-$bundle $Y_{i}\longrightarrow Z_{i}$). Denote by $\pi[m]:X_{m}\longrightarrow X$ the map which is the identity on the root component $X_{0}$ and contracts all the bubble components to $Z_{0}$ via the fiber bundle projections. Now consider a nodal curve $C$ mapped into $X_{m}$ by $f:C\longrightarrow X_{m}$ with specified tangency to $Z$. There are two types of marked points: (i) absolute marked points whose images under $f$ lie outside $Z$, labeled by $x_{i}$, (ii) relative marked points which are mapped into $Z$ by $f$, labeled by $y_{j}$. A relative $J-$holomorphic map $f:C\longrightarrow X_{m}$ is said to be pre- deformable if $f^{-1}(Z_{i})$ consists of a union of nodes such that for each node $p\in f^{-1}(Z_{i}),i=1,2,\cdots,m$, the two branches at the node are mapped to different irreducible components of $X_{m}$ and the orders of contact to $Z_{i}$ are equal. An isomorphism of two such $J-$holomorphic maps $f$ and $f^{\prime}$ to $X_{m}$ consists of a diagram $\begin{array}[]{ccc}(C,x_{1},\cdots,x_{l},y_{1},\cdots,y_{k})&\stackrel{{\scriptstyle f}}{{\longrightarrow}}&X_{m}\\\ &&\\\ h\downarrow&&\downarrow t\\\ &&\\\ (C^{\prime},x^{\prime}_{1},\cdots,x^{\prime}_{l},y^{\prime}_{1},\cdots,y^{\prime}_{k})&\stackrel{{\scriptstyle f^{\prime}}}{{\longrightarrow}}&X_{m}\end{array}$ where $h$ is an isomorphism of marked curves and $t\in\mbox{Aut}_{Z}(X_{m})$. With the preceding understood, a relative $J-$holomorphic map to $X_{m}$ is said to be stable if it has only finitely many automorphisms. We introduced the notion of a weighted graph in Remark 2.7. We need to refine it for relative stable maps to $(X,Z)$. A (connected) relative graph $\Gamma$ consists of the following data: (1) a vertex decorated by $A\in H_{2}(X;\mathbb{Z})$ and genus $g$, (2) a tail for each absolute marked point, (3) a relative tail for each relative marked point. ###### Definition 2.8. Let $\Gamma$ be a relative graph with $k$ (ordered) relative tails and $T_{k}=(t_{1},\cdots,t_{k})$, a $k-$tuple of positive integers forming a partition of $A\cdot Z$. A relative $J-$holomorphic map to $(X,Z)$ with type $(\Gamma,T_{k})$ consists of a marked curve $(C,x_{1},\cdots,x_{l},y_{1},\cdots,y_{k})$ and a map $f:C\longrightarrow X_{m}$ for some non-negative integer $m$ such that (i) $C$ is a connected curve (possibly reducible) of arithmetic genus $g$, (ii) the map $\pi_{m}\circ f:C\longrightarrow X_{m}\longrightarrow X$ satisfies $(\pi_{m}\circ f)_{*}[C]=A$, (iii) the $x_{i},1\leq i\leq l$, are the absolute marked points, (iv) the $y_{i},1\leq i\leq k$, are the relative marked points, (v) $f^{*}Z_{m}=\sum_{i=1}^{k}t_{i}y_{i}$. Let $\overline{\mathcal{M}}_{\Gamma,T_{k}}(X,Z,J)$ be the moduli space of pre- deformable relative stable $J-$holomorphic maps with type $(\Gamma,T_{k})$. Notice that for an element $f:C\to X_{m}$ in $\overline{\mathcal{M}}_{\Gamma,T_{k}}(X,Z,J)$ the intersection pattern with $Z_{0},...,Z_{m-1}$ is only constrained by the genus condition and the pre- deformability condition. Now we apply the virtual neighborhood technique to construct $U^{X,Z}_{{\mathcal{S}}_{e}}$, $\mathcal{E}_{X,Z}$, $S_{X,Z}$ as in section 2.1. Consider the configuration space $\overline{\mathcal{B}}_{\Gamma}(X,Z,J)$ of equivalence classes of smooth pre-deformable relative stable maps. Here we still take the equivalence class under ${\mathbb{C}}^{*}$-action on the fibers of ${\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})$. In particular, the subgroup of ${\mathbb{C}}^{*}$ fixing such a map is required to be finite. The maps are required to intersect the $Z_{i}$ only at finitely many points in the domain curve. Further, at these points, the map is required to have a holomorphic leading term in the normal Taylor expansion for any local chart of $X$ taking $Z$ to a coordinate hyperplane and being holomorphic in the normal direction along $Z$. Thus the notion of contact order still makes sense, and we can still impose the pre-deformability condition and contact order condition at the $y_{i}$ being governed by $T_{k}$. Next, we can define $\Omega^{1}$ similarly. We also need to understand the $Obs$ space. The discussion is similar to that of stable maps. According to its label, a relative stable map is naturally divided into components of two types: (i) a stable map in $X$ intersecting $Z$ transversely, called a rigid factor; (ii) a stable map in ${\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})$ such that its projection to $Z$ is stable and it intersects $Z_{0}$, $Z_{\infty}$ transversely, called a rubber factor. For each component $(\Sigma,f)$, we also have the extra condition that $f$ intersects $Z_{0}$ or $Z_{\infty}$ with an order fixed by the graph. Suppose that $f(y_{i})\in Z_{0}$ or $Z_{\infty}$ with order $t_{i}$. The analog of (1) is $L^{X,Z}_{\Sigma,f}:\Omega^{0}_{r}\longrightarrow\Omega^{0,1}_{r}.$ Here an element $u\in\Omega^{0}_{r}$ is an element of $\Omega^{0}(f^{*}TX)$ or $\Omega^{0}(f^{*}T{\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}}))$ with the following property: Choose a unitary connection on $N$ so that we can decompose the tangent bundle of ${\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})$ into tangent and normal directions. Near $f(y_{i})$, $u(y_{i})$ can be decomposed into $(u_{Z},u_{N})$ where $u_{Z}$, $u_{N}$ are tangent and normal components, respectively. Now we require that $u_{N}$ vanish at $y_{i}$ with order $t_{i}$. When $\Sigma$ consists of two components joined at one point, we require their $u_{Z}$ components be the same at the intersection point. We can also consider the analog of $\tilde{L}_{\Sigma,f}$, $\tilde{L}^{X,Z}_{\Sigma,f}\oplus\oplus_{i}T_{t_{i}}:\tilde{\Omega}^{0}_{r}\longrightarrow\tilde{\Omega}^{0,1}_{r}\oplus\oplus_{i}{\mathcal{J}}^{t_{i}}_{i}.$ Here $\tilde{\Omega}^{0}_{r}=\\{u\in\Omega^{0}\mid u_{N}(y_{i})=0\\}.$ Also an element $v\in\tilde{\Omega}^{0,1}_{r}$ is required to have a simple pole at each $y_{i}$ such that $\mbox{Res}_{y_{i}}v\in TZ$, and if two components are joined together, we require that the sum of the residues be zero. Each summand ${\mathcal{J}}^{t_{i}}_{i}\cong\oplus^{t_{i}-1}_{j=1}\mbox{Hom}((T_{y_{i}}\Sigma)^{j},N_{f(y_{i})})$ is the $(t_{i}-1)$-jet space, and the map $T_{t_{i}}(f)$ is the $(t_{i}-1)$-jet of $f$ at $y_{i}$, i.e. the first $(t_{i}-1)$ terms of the Taylor polynomial. It is clear that $\mbox{Coker}\,\tilde{L}^{X,Z}_{\Sigma}$ has a similar description as $\mbox{Coker}\,\tilde{L}^{X}_{\Sigma}$, the only difference being that we require the residue at each nodal point be in $TZ$. Moreover, $\tilde{L}^{X,Z}_{\Sigma,f}\oplus\oplus_{i}T_{t_{i}}=\tilde{L}^{X,Z}_{\Sigma,f}\oplus\oplus_{i}T_{t_{i}}\mid_{\mbox{Ker}\,\tilde{L}^{X,Z}_{\Sigma,f}}.$ Finally the process of adding the deformation of a nodal Riemann surface is identical. In addition to the evaluation maps on $\overline{\mathcal{B}}_{\Gamma,T_{k}}(X,Z,J)$, $\begin{array}[]{lllll}ev^{X}_{i}:&\overline{\mathcal{B}}_{\Gamma,T_{k}}(X,Z,J)&\longrightarrow&X,&\quad 1\leq i\leq l,\\\ &(\Sigma,x_{1},\cdots,x_{l},y_{1},\cdots,y_{k},f)&\longrightarrow&f(x_{i}),&\end{array}$ there are also the evaluations maps $\begin{array}[]{lllll}ev^{Z}_{j}:&\overline{\mathcal{B}}_{\Gamma,T_{k}}(X,Z,J)&\longrightarrow&Z,&\quad 1\leq j\leq k,\\\ &(\Sigma,x_{1},\cdots,x_{l},y_{1},\cdots,y_{k},f)&\longrightarrow&f(y_{j}),&\end{array}$ where $Z=Z_{m}$ if the target of $f$ is $X_{m}$. ###### Definition 2.9. Let $\alpha_{i}\in H^{*}(X;{\mathbb{R}})$,$1\leq i\leq l$, $\beta_{j}\in H^{*}(Z;{\mathbb{R}})$, $1\leq j\leq k$. Define the relative Gromov-Witten invariant $\langle\Pi_{i}\tau_{d_{i}}\alpha_{i}\mid\Pi_{j}\beta_{j}\rangle^{X,Z}_{\Gamma,T_{k}}=\frac{1}{|\hbox{Aut}(T_{k})|}\int_{U^{X,Z}_{{\mathcal{S}}_{e}}}S^{*}_{X,Z}\Theta\wedge\Pi_{i}\psi_{i}^{d_{i}}\wedge(ev^{X}_{i})^{*}\alpha_{i}\wedge\Pi_{j}(ev^{Z}_{j})^{*}\beta_{j},$ where $\Theta$ is the Thom class of the bundle ${\mathcal{E}}_{X,Z}$ and Aut$(T_{k})$ is the symmetry group of the partition $T_{k}$. Denote by ${\mathcal{T}}_{k}=\\{(t_{j},\beta_{j})\mid j=1,\cdots,k\\}$ the weighted partition of $A\cdot Z$. If the vertex of $\Gamma$ is decorated by $(g,A)$, we will sometimes write $\langle\Pi_{i}\tau_{d_{i}}\alpha_{i}\mid{\mathcal{T}}_{k}\rangle^{X,Z}_{g,A}$ for $\langle\Pi_{i}\tau_{d_{i}}\alpha_{i}\mid\Pi_{j}\beta_{j}\rangle^{X,Z}_{\Gamma,T_{k}}$. ###### Remark 2.10. In [LR] only invariants without descendent classes were considered. But it is straightforward to extend the definition of [LR] to include descendent classes. We can decorate the tail of a relative graph $\Gamma$ by $(d_{i},\alpha_{i})$ as in the absolute case. We can further decorate the relative tails by the weighted partition ${\mathcal{T}}_{k}$. Denote the resulting weighted relative graph by $\Gamma\\{(d_{i},\alpha_{i})\\}|{\mathcal{T}}_{k}$. In [LR] the source curve is required to be connected. We will also need to use a disconnected version. For a disjoint union $\Gamma^{\bullet}$ of weighted relative graphs and a corresponding disjoint union of partitions, still denoted by $T_{k}$, we use $\langle\Gamma^{\bullet}\\{(d_{i},\alpha_{i})\\}|{\mathcal{T}}_{k}\rangle^{X,Z}$ to denote the corresponding relative invariants with a disconnected domain, which is simply the product of the connected relative invariants. Notice that although we use $\bullet$ in our notation following [MP], our disconnected invariants are different. The disconnected invariants there depend only on the genus, while ours depend on the finer graph data. ### 2.3. Partial orderings on relative GW invariants In [MP], the authors first introduced a partial order on the set of relative Gromov-Witten invariants of a ${\mathbb{P}}^{1}$-bundle. The authors, [HLR], refined their partial order on the set of relative Gromov-Witten invariants of a Blow-up manifold relative to the exceptional divisor, and used this partial order to obtain a Blow-up correspondence of absolute/relative Gromov-Witten theory. In this subsection, we will review the partial order on the set of relative Gromov-Witten invariants. First of all, all Gromov-Witten invariants vanish if $A\in H_{2}(X,{\mathbb{Z}})$ is not an effective curve class. We define a partial ordering on $H_{2}(X,{\mathbb{Z}})$ as follows: $A^{\prime}<A$ if $A-A^{\prime}$ is a nonzero effective curve class. The set of pairs $(m,\delta)$ where $m\in{\mathbb{Z}}_{>0}$ and $\delta\in H^{*}(Z,{\mathbb{Q}})$ is partially ordered by the following size relation (2) $(m,\delta)>(m^{\prime},\delta^{\prime})$ if $m>m^{\prime}$ or if $m=m^{\prime}$ and $\deg(\delta)>\deg(\delta^{\prime})$. Let $\mu$ be a partition weighted by the cohomology of $Z$, i.e., $\mu=\\{(\mu_{1},\delta_{r_{1}}),\cdots,(\mu_{\ell(\mu)},\delta_{r_{\ell(\mu)}})\\}.$ We may place the pairs of $\mu$ in decreasing order by size (2). We define $\deg(\mu)=\sum\deg(\delta_{r_{i}}).$ A lexicographic ordering on weighted partitions is defined as follows: $\mu\stackrel{{\scriptstyle l}}{{>}}\mu^{\prime}$ if, after placing the pairs in $\mu$ and $\mu^{\prime}$ in decreasing order by size, the first pair for which $\mu$ and $\mu^{\prime}$ differ in size is larger for $\mu$. For the nondescendent relative Gromov-Witten invariant $\langle\varpi\mid\mu\rangle^{X,Z}_{g,A},$ denote by $\|\varpi\|$ the number of absolute insertions. ###### Definition 2.11. A partial ordering $\stackrel{{\scriptstyle\circ}}{{<}}$ on the set of nondescendent relative Gromov-Witten invariants is defined as follows: $\langle\varpi^{\prime}\mid\mu^{\prime}\rangle^{X,Z}_{g^{\prime},A^{\prime}}\stackrel{{\scriptstyle\circ}}{{<}}\langle\varpi\mid\mu\rangle^{X,Z}_{g,A}$ if one of the conditions below holds 1. (a) $A^{\prime}<A$, 2. (b) equality in (a) and $g^{\prime}<g$, 3. (c) equality in (a)-(b) and $\|\varpi^{\prime}\|<\|\varpi\|$, 4. (d) equality in (a)-(c) and $\deg(\mu^{\prime})>\deg(\mu)$, 5. (e) equality in (a)-(d) and $\mu^{\prime}\stackrel{{\scriptstyle l}}{{>}}\mu$. ### 2.4. Degeneration formula Now we describe the degeneration formula of GW-invariants under symplectic cutting. As an operation on topological spaces, the symplectic cut is essentially collapsing the circle orbits in the hypersurface $H^{-1}(0)$ to points in $Z$. Suppose that $X_{0}\subset X$ is an open codimension zero submanifold with a Hamiltonian $S^{1}-$action. Let $H:X_{0}\to\mathbb{R}$ be a Hamiltonian function with $0$ as a regular value. If $H^{-1}(0)$ is a separating hypersurface of $X_{0}$, then we obtain two connected manifolds $X_{0}^{\pm}$ with boundary $\partial X_{0}^{\pm}=H^{-1}(0)$. Suppose further that $S^{1}$ acts freely on $H^{-1}(0)$. Then the symplectic reduction $Z=H^{-1}(0)/S^{1}$ is canonically a symplectic manifold of dimension $2$ less. Collapsing the $S^{1}-$action on $\partial X^{\pm}=H^{-1}(0)$, we obtain closed smooth manifolds $\overline{X_{0}}^{\pm}$ containing respectively real codimension $2$ submanifolds $Z^{\pm}=Z$ with opposite normal bundles. Furthermore $\overline{X_{0}}^{\pm}$ admits a symplectic structure $\overline{\omega}^{\pm}$ which agrees with the restriction of $\omega$ away from $Z$, and whose restriction to $Z^{\pm}$ agrees with the canonical symplectic structure $\omega_{Z}$ on $Z$ from symplectic reduction. This is neatly shown by considering $X_{0}\times\mathbb{C}$ equipped with appropriate product symplectic structures and the product $S^{1}$-action on $X_{0}\times\mathbb{C}$, where $S^{1}$ acts on $\mathbb{C}$ by complex multiplication. The extended action is Hamiltonian if we use the standard symplectic structure $\sqrt{-1}dw\wedge d\bar{w}$ or its negative on the $\mathbb{C}$ factor.Then the moment map is $\mu_{+}(u,w)=H(u)+|w|^{2}:X_{0}\times\mathbb{C}\to{\mathbb{R}},$ and $\mu_{+}^{-1}(0)$ is the disjoint union of $S^{1}-$invariant sets $\\{(u,w)|H(u)=-|w|^{2}<0\\}\quad\hbox{and}\quad\\{(u,0)|H(z)=0\\}.$ We define $\overline{X_{0}}^{+}$ to be the symplectic reduction $\mu_{+}^{-1}(0)/S^{1}$. Then $\overline{X}_{0}^{+}$ is the disjoint union of an open symplectic submanifold and a closed codimension 2 symplectic submanifold identified with $(Z,\omega_{Z})$. The open piece can be identified symplectically with $X_{0}^{+}=\\{u\in X_{0}|H(u)<0\\}\subset X_{0}$ by the map $u\to(u,\sqrt{-H(u)}).$ Similarly, if we use $-{\bf i}dw\wedge d\bar{w}$, then the moment map is $\mu_{-}(u,w)=H(u)-|w|^{2}:X_{0}\times\mathbb{C}\to\mathbb{R}$ and the corresponding symplectic reduction $\mu_{-}^{-1}(0)/S^{1}$, denoted by $\overline{X}_{0}^{-}$, is the disjoint union of an open piece identified symplectically with $X_{0}^{-}=\\{u\in X_{0}|H(u)>0\\}$ by the map $\phi_{0}^{-}:u\to(u,\sqrt{H(u)})$, and a closed codimension 2 symplectic submanifold identified with $(Z,\omega_{Z})$. We finally define $\overline{X}^{+}$ and $\overline{X}^{-}$. $\overline{X}^{+}$ is simply $\overline{X_{0}}^{+}$, while $\overline{X}^{-}$ is obtained from gluing symplectically $X^{-}$ and $\overline{X}_{0}^{-}$ along $X_{0}$ via $\phi_{0}^{-}$. Notice that $\overline{X}^{-}=(X^{-}-X_{0})\cup\overline{X_{0}}^{-}$ as a set. The two symplectic manifolds $(\overline{X}^{\pm},\overline{\omega}^{\pm})$ are called the symplectic cuts of $X$ along $H^{-1}(0)$. Thus we have a continuous map $\pi:X\to\overline{X}^{+}\cup_{Z}\overline{X}^{-}.$ As for the symplectic forms, we have $\omega^{+}|_{Z}=\omega^{-}|_{Z}$. Hence, the pair $(\omega^{+},\omega^{-})$ defines a cohomology class of $\overline{X}^{+}\cup_{Z}\overline{X}^{-}$, denoted by $[\omega^{+}\cup_{Z}\omega^{-}]$. It is easy to observe that (3) $\pi^{*}([\omega^{+}\cup_{Z}\omega^{-}])=[\omega].$ Let $B\in H_{2}(X;{\mathbb{Z}})$ be in the kernel of $\pi_{*}:H_{2}(X;{\mathbb{Z}})\longrightarrow H_{2}(\overline{X}^{+}\cup_{Z}\overline{X}^{-};{\mathbb{Z}}).$ By (3) we have $\omega(B)=0$. Such a class is called a vanishing cycle. For $A\in H_{2}(X;{\mathbb{Z}})$ define $[A]=A+\mbox{Ker}(\pi_{*})$ and (4) $\langle\Pi_{i}\tau_{d_{i}}\alpha_{i}\rangle^{X}_{g,[A]}=\sum_{B\in[A]}\langle\Pi_{i}\tau_{d_{i}}\alpha_{i}\rangle^{X}_{g,B}.$ Notice that $\omega$ has constant pairing with any element in $[A]$. It follows from the Gromov compactness theorem that there are only finitely many such elements in $[A]$ represented by $J$-holomorphic stable maps. Therefore, the summation in (4) is finite. The degeneration formula expresses $\langle\Pi_{i}\tau_{d_{i}}\alpha_{i}\rangle^{X}_{g,[A]}$ in terms of relative invariants of $(\overline{X}^{+},Z)$ and $(\overline{X}^{-},Z)$ possibly with disconnected domains. To begin with, we need to assume that the cohomology class $\alpha_{i}$ is of the form $\alpha_{i}=\pi^{*}(\alpha^{+}\cup_{Z}\alpha^{-}).$ Here $\alpha_{i}^{\pm}\in H^{*}(\overline{X}^{\pm};{\mathbb{R}})$ are classes with $\alpha_{i}^{+}|_{Z}=\alpha_{i}^{-}|_{Z}$ so that they give rise to a class $\alpha_{i}^{+}\cup_{Z}\alpha_{i}^{-}\in H^{*}(\overline{X}^{+}\cup_{Z}\overline{X}^{-};{\mathbb{R}})$. Next, we proceed to write down the degeneration formula. We first specify the relevant topological type of a marked Riemann surface mapped into $\overline{X}^{+}\cup_{Z}\overline{X}^{-}$ with the following properties: 1. (i) Each connected component is mapped either into $\overline{X}^{+}$ or $\overline{X}^{-}$ and carries a respective degree 2 homology class; 2. (ii) The images of two distinct connected components only intersect each other along $Z$; 3. (iii) No two connected components which are both mapped into $\overline{X}^{+}$ or $\overline{X}^{-}$ intersect each other; 4. (iv) The marked points are not mapped to $Z$; 5. (v) Each point in the domain mapped to $Z$ carries a positive integer (representing the order of tangency). By abuse of language we call the above data a $(\overline{X}^{+},\overline{X}^{-})-$graph. Such a graph gives rise to two relative graphs of $(\overline{X}^{+},Z)$ and $\overline{X}^{-},Z)$ from (i-iv), each possibly being disconnected. We denote them by $\Gamma^{\bullet}_{+}$ and $\Gamma^{\bullet}_{-}$ respectively. From (v) we also get two partitions $T_{+}$ and $T_{-}$. We call a $(\overline{X}^{+},\overline{X}^{-})-$graph a degenerate $(g,A,l)-$graph if the resulting pairs $(\Gamma^{\bullet}_{+},T_{+})$ and $(\Gamma^{\bullet}_{-},T_{-})$ satisfy the following constraints: the total number of marked points is $l$, the relative tails are the same, i.e. $T_{+}=T_{-}$, and the identification of relative tails produces a connected graph of $X$ with total homology class $\pi_{*}[A]$ and arithmetic genus $g$. Let $\\{\beta_{a}\\}$ be a self-dual basis of $H^{*}(Z;{\mathbb{R}})$ and $\eta^{ab}=\int_{Z}\beta_{a}\cup\beta_{b}$. Given $g,A$ and $l$, consider a degenerate $(g,A,l)-$graph. Let $T_{k}=T_{+}=T_{-}$ and ${\mathcal{T}}_{k}$ be a weighted partition $\\{t_{j},\beta_{a_{j}}\\}$. Let ${\mathcal{T}}^{\prime}_{k}=\\{t_{j},\beta_{a_{j^{\prime}}}\\}$ be the dual weighted partition. The degeneration formula for $\langle\Pi_{i}\tau_{d_{i}}\alpha_{i}\rangle^{X}_{g,[A]}$ then reads as follows, $\begin{array}[]{ll}&\langle\Pi_{i}\tau_{d_{i}}\alpha_{i}\rangle^{X}_{g,[A]}\cr=&\sum\langle\Gamma^{\bullet}\\{(d_{i},\alpha_{i}^{+})\\}|{\mathcal{T}}_{k}\rangle^{\overline{X}^{+},Z}\Delta({\mathcal{T}}_{k})\langle\Gamma^{\bullet}\\{(d_{i},\alpha_{i}^{-})\\}|{\mathcal{T}}^{\prime}_{k}\rangle^{\overline{X}^{-},Z},\cr\end{array}$ where the summation is taken over all degenerate $(g,A,l)-$graphs, and $\Delta({\mathcal{T}}_{k})=\prod_{j}t_{j}|\mbox{Aut}\,(T_{k})|.$ ## 3\. Relative GW-invariants of ${\mathbb{P}}^{1}$-bundles Suppose that $Z$ is a symplectic submanifold of $X$ of codimension 2\. When applying the degeneration formula, we often need to express the absolute Gromov-Witten invariants of $X$ as a summation of products of relative Gromov- Witten invariants of symplectic cuts of $X$. Thus if we want to obtain a comparison theorem of Gromov-Witten invariant by the degeneration formula, the point will be how to compute the relative Gromov-Witten invariants of a ${\mathbb{P}}^{1}$-bundle. In this section, we will prove a vanishing theorem for genus zero relative Gromov-Witten invariants of the ${\mathbb{P}}^{1}$-bundle $Y$ relative to the infinity section and compute some genus zero two-point relative fiber class GW invariants of the ${\mathbb{P}}^{1}$-bundle $Y$. Suppose that $L$ is a line bundle over $Z$ and $Y={\mathbb{P}}(L\oplus{\mathbb{C}})$. Let $D$, $Z$ be the infinity section and zero section of $Y$ respectively. Let $\beta_{1},\cdots,\beta_{m_{Z}}$ be a self-dual basis of $H^{*}(Z,{\mathbb{Q}})$ containing the identity element. We will often denote the identity by $\beta_{\rm id}$. The degree of $\beta_{i}$ is the real grading in $H^{*}(Z,{\mathbb{Q}})$. We view $\beta_{i}$ as an element of $H^{*}(Y,{\mathbb{Q}})$ via the pull-back by $\pi:Y={\mathbb{P}}(L\oplus{\mathbb{C}})\longrightarrow Z$. Let $[Z]$, $[D]\in H^{2}(Y,{\mathbb{Q}})$ denote the cohomology classes associated to the divisors. Define classes in $H^{*}(Y,{\mathbb{Q}})$ by $\displaystyle\gamma_{i}$ $\displaystyle=$ $\displaystyle\beta_{i},$ $\displaystyle\gamma_{m_{Z}+i}$ $\displaystyle=$ $\displaystyle\beta_{i}\cdot[Z],$ $\displaystyle\gamma_{2m_{Z}+i}$ $\displaystyle=$ $\displaystyle\beta_{i}\cdot[D].$ We will use the following notation: $\displaystyle\gamma_{i}^{\beta}$ $\displaystyle=$ $\displaystyle\beta_{i\,\,\mbox{mod}\,\,m_{Z}},$ $\displaystyle\gamma_{i}^{D}$ $\displaystyle=$ $\displaystyle 1,[Z],\mbox{or}[D].$ The second assignment depends upon the integer part of $(i-1)/m_{Z}$. The set $\\{\gamma_{1},\cdots,\gamma_{2m_{Z}}\\}$ determines a basis of $H^{*}(Y,{\mathbb{Q}})$. Since $D\cong Z$ as topological manifolds, $H^{*}(D,{\mathbb{Q}})$ is isomorphic to $H^{*}(Z,{\mathbb{Q}})$. If there is no confusion, we will also use $\beta_{1},\cdots,\beta_{m_{Z}}$ as a self-dual basis of $H^{*}(D,{\mathbb{Q}})$. ### 3.1. Fiber class invariants In this subsection, we mainly compute some genus zero relative GW invariants of ${\mathbb{P}}^{1}$-bundles with a fiber class. According to [MP], we may transfer the computation of this invariant on the ${\mathbb{P}}^{1}$-bundle into that of some associated invariants on ${\mathbb{P}}^{1}$. Suppose that $L$ is a line bundle over $Z$ and $Y={\mathbb{P}}(L\oplus{\mathcal{O}})=\\{(z,l)|z\in Z,l\subset L_{z}\oplus{\bf C}\\},$ where $L_{z}$ is the fiber of $L$ at $z$. Denote by $\pi:Y\longrightarrow Z$ the projection of the ${\mathbb{P}}^{1}$-bundle $Y$. Moreover there are two inclusions (sections) of $Z$ in $Y={\mathbb{P}}(L\oplus{\mathcal{O}})$: 1. (1) the “zero section” $z\longrightarrow(z,0\oplus{\bf C})$, denoted by $Z$, 2. (2) the “section at infinity” $z\longrightarrow(z,L_{z}\oplus 0)$, denoted by $D$. Let $\Gamma$ be the relative graph with the following data 1. (1) a vertex decorated by $A=sF\in H_{2}(Y,{\mathbb{Z}})$ and genus zero; 2. (2) $k$ relative tails; 3. (3) $l$ absolute tails. For any non-negative integer $m$, define $Y_{m}$ by gluing together $m$ copies of $Y$, where the infinity section of the $i^{th}$ component is glued to the zero section of the $(i+1)^{st}$ $(1\leq i\leq m)$ component; see Section 2.2 for details. Denote by $\pi[m]:Y_{m}\longrightarrow Y$ the map which is the identity on the root component $Y_{0}$ and contract all the bubble components to $D_{0}$ via the projection of the fiber bundle of $Y_{i}$. Let $T_{k}=\\{t_{1},\cdots,t_{k}\\}$ be a $k$-tuple of positive integers forming a partition of $s$. Denote by $\overline{\mathcal{M}}_{\Gamma,T_{k}}(Y,D)$ the moduli space of morphisms $f:(C,x_{0},\cdots,x_{l};y_{1},\cdots,y_{k})\longrightarrow(Y_{m},D_{\infty}),$ such that 1. (1) $(C,x_{0},\cdots,x_{l};y_{1},\cdots,y_{k})$ is a prestable curve of genus zero with $l$ absolute marked points $x_{0},\cdots,x_{l}$ and $k$ relative marked points $y_{1},\cdots,y_{k}$;. 2. (2) $f^{-1}(D_{\infty})=\sum t_{i}y_{i}$ as Cartier divisor and $\deg(\pi[m]\circ f)=s$. 3. (3) The predeformability condition: The preimage of the singular locus Sing $Y_{m}=\cup_{i=0}^{m-1}D_{i}$ of $Y_{m}$ is a union of nodes of $C$, and if $p$ is one such node, then the two branches of $C$ at $p$ map into different irreducible components of $Y_{m}$, and their orders of contact with the divisor $D_{i}$ (in their respective components of $Y_{m}$) are equal. The morphism $f$ is also required to satisfy a stability condition that there are no infinitesimal automorphisms of the sequence of maps $(C,x_{0},\cdots,x_{l};y_{1},\cdots,y_{k})\longrightarrow Y_{m}\stackrel{{\scriptstyle\pi[m]}}{{\longrightarrow}}Y$ where the allowed automorphisms of the map from $Y_{m}$ to $Y$ are $\mbox{Aut}_{D}(Y_{m})$. 4. (4) The automorphism group of $f$ is finite. Two such morphisms are isomorphic if they differ by an isomorphism of the domain and an automorphism of $(Y_{m},D_{0},D_{\infty})$. In particular, this defines the automorphism group in the stability condition $(4)$ above. We introduce some notations which are used in [Li1]. For any non-negative integer $m$, let ${\mathbb{P}}^{1}[m]={\mathbb{P}}^{1}_{(0)}\cup{\mathbb{P}}^{1}_{(1)}\cup\cdots\cup{\mathbb{P}}^{1}_{(m)}$ be a chain of $m+1$ copies ${\mathbb{P}}^{1}$, where ${\mathbb{P}}^{1}_{(l)}$ is glued to ${\mathbb{P}}^{1}_{(l+1)}$ at $p_{1}^{(l)}$ for $0\leq l\leq m-1$. The irreducible component ${\mathbb{P}}^{1}_{(0)}$ will also be referred to as the root component and the other irreducible components will be called the bubble components. A point $p_{1}^{(m)}\not=p_{1}^{(m-1)}$ is fixed on ${\mathbb{P}}^{1}_{(m)}$. Denote still by $\pi[m]:{\mathbb{P}}^{1}[m]\longrightarrow{\mathbb{P}}^{1}$ the map which is the identity on the root component and contracts all the bubble components to $p_{1}^{(0)}$. For $m>0$, let ${\mathbb{P}}^{1}(m)={\mathbb{P}}^{1}_{(1)}\cup\cdots\cup{\mathbb{P}}^{1}_{(m)}$ denote the union of bubble components of ${\mathbb{P}}^{1}[m]$. Similar to the case of $Y_{m}$, we may define the associated moduli space $\overline{\mathcal{M}}_{\Gamma}({\mathbb{P}}^{1},p_{1}^{(0)};T_{1})$ of relative stable maps to $({\mathbb{P}}^{1},p_{1}^{(0)})$, see [Li1] for its definition. Next, we first review Maulik-Pandharipande’s algorithm [MP] which reduces the relative Gromov-Witten invariant of $(Y,D)$ of fiber class to that of $({\mathbb{P}}^{1},p_{1})$. Note that the moduli space of stable relative maps $\overline{\mathcal{M}}_{Y}=\overline{\mathcal{M}}_{\Gamma,T_{k}}(Y,D)$ is fibered over $Z$, (5) $\pi:\overline{\mathcal{M}}_{Y}\longrightarrow Z,$ with fiber isomorphic to the moduli space of maps of degree $s$ to ${\mathbb{P}}^{1}$ relative to the infinity point $p_{1}$ with tangency order $s$: $\overline{\mathcal{M}}_{{\mathbb{P}}^{1}}=\overline{\mathcal{M}}_{\Gamma,T_{k}}({\mathbb{P}}^{1},p_{1}).$ In fact, $\overline{\mathcal{M}}_{Y}$ is the fiber bundle constructed from the principal $S^{1}$-bundle associated to $L$ and a standard $S^{1}$-action on $\overline{\mathcal{M}}_{{\mathbb{P}}^{1}}$. The $\pi$-relative obstruction theory of $\overline{\mathcal{M}}_{Y}$ is obtained from the $\overline{\mathcal{M}}_{{\mathbb{P}}^{1}}$-fiber bundle structure over $Z$. The relationship between the $\pi$-relative virtual fundamental class $[\overline{\mathcal{M}}_{\bar{M}^{+}}]^{vir_{\pi}}$ and the virtual fundamental class $[\overline{\mathcal{M}}_{\bar{M}^{+}}]^{vir}$ is given by the equation (6) $[\overline{\mathcal{M}}_{Y}]^{vir}=c_{top}({\mathbb{E}}\otimes TZ)\cap[\overline{\mathcal{M}}_{Y}]^{vir_{\pi}}$ where ${\mathbb{E}}$ is the Hodge bundle. Since we only consider the case of genus zero, (6) can be written as $[\overline{\mathcal{M}}_{Y}]^{vir}=[\overline{\mathcal{M}}_{Y}]^{vir_{\pi}}.$ By integrating along the fiber, we can compute the relative Gromov-Witten invariants of $Y$ by computing the equivariant integrations in the relative Gromov-Witten theory of ${\mathbb{P}}^{1}$; see [MP] for the details. Let ${\mathcal{T}}_{k}=\\{(t_{i},\beta_{i})\\}$ be the cohomology weighted partition of $s$. By definition, we have $\displaystyle\langle\tau_{d_{1}-1}\gamma_{1},\cdots,\tau_{d_{l}-1}\gamma_{l}\mid{\mathcal{T}}_{k}\rangle^{Y,D}_{\Gamma}=\int_{[\overline{\mathcal{M}}_{Y}]^{vir}}\prod_{i=1}^{l}\psi_{i}^{d_{i}-1}ev_{i}^{*}\gamma_{i}\wedge\prod_{j}ev_{j}^{*}\beta_{j}$ $\displaystyle=$ $\displaystyle\frac{1}{|\mbox{Aut}(T_{k})|}\int_{Z}(\prod_{i}\gamma_{i}^{\delta}\prod_{j}\beta_{j}\cap\pi_{*}(\prod_{i}\psi_{i}^{d_{i}-1}ev_{i}^{*}(\gamma_{i}^{D_{0}})\cap[M_{Y}]^{vir_{\pi}})),$ where the interior push-forward $\pi_{*}(\prod_{i}\psi_{i}^{d_{i}-1}ev_{i}^{*}(\gamma_{i}^{D_{0}})\cap[M_{Y}]^{vir_{\pi}})$ is obtained from the corresponding Hodge integral in the equivariant Gromov- Witten theory of $({\mathbb{P}}^{1},p_{1})$ after replacing the hyperplane class on ${\mathbb{C}}{\mathbb{P}}^{\infty}$ by $C_{1}(L)$. Therefore, via (3.1), we may reduce the computation of relative Gromov-Witten invariants $\langle\tau_{d_{1}-1}\gamma_{1},\cdots,\tau_{d_{l}-1}\gamma_{l}\mid{\mathcal{T}}_{k}\rangle^{Y,D}_{\Gamma}$ to that of $\langle\tau_{d_{1}-1}\delta_{1},\cdots,\tau_{d_{l}-1}\delta_{l}\mid pt,\cdots,pt\rangle^{{\mathbb{P}}^{1},p_{1}}_{\Gamma,T_{k}},$ where $\delta_{i}\in H^{*}({\mathbb{P}}^{1},{\mathbb{Q}})$, $1\leq i\leq l$. About the two point genus zero relative Gromov-Witten invariant of $({\mathbb{P}}^{1},p_{1})$, we have ###### Lemma 3.1. Let $\varpi\in H^{2}({\mathbb{P}}^{1},{\mathbb{Q}})$. 1. (i) If $d\not=s$, then $\langle\tau_{d-1}\varpi\mid(s,[pt])\rangle^{{\mathbb{P}}^{1},p_{1}}_{s}=0$. 2. (ii) For $s>0$, we have $\langle\tau_{s-1}\varpi\mid(s,[pt])\rangle^{{\mathbb{P}}^{1},p_{1}}_{s}=\frac{1}{s!}.$ The proof of (i) follows from a simple dimension calculation and (ii) of the lemma is Lemma 1.4 of [OP]. In [HLR], the authors generalized the result to general projective space ${\mathbb{P}}^{n}$ via localization techniques. ###### Proposition 3.2. Let $s>0$. 1. (i) Let ${\mathcal{T}}_{k}=\\{(t_{i},\beta_{i})\\}$ be a cohomology weighted partition of $s$. Then $\langle\pi^{*}\alpha_{1},\cdots,\pi^{*}\alpha_{q},\beta_{1}\cdot[Z],\cdots,\beta_{l}\cdot[Z]\mid{\mathcal{T}}_{k}\rangle^{Y,D}_{sF}=0$ except for $s=k=1$ and $q=0$. 2. (ii) For $s>0$, we have the two-point relative invariant $\langle\tau_{d-1}(\beta_{0}\cdot[Z])\mid(s,\beta_{\infty})\rangle^{Y,D}_{sF}=\left\\{\begin{array}[]{ll}\frac{1}{s!}\int_{Z}\beta_{0}\wedge\beta_{\infty},&d=s\\\ &\\\ 0,&d\not=s\end{array}\right.,$ where $\beta_{0}\in H^{*}(Z,{\mathbb{Q}})$ and $\beta_{\infty}\in H^{*}(D,{\mathbb{Q}})$. 3. (iii) For $s=k=1$, we have $\langle\iota^{!}(\beta_{1}),\cdots,\iota^{!}(\beta_{l})\mid(1,\gamma)\rangle^{Y,D}_{F}=\int_{Z}\beta_{1}\wedge\cdots\wedge\beta_{l}\wedge\gamma.$ ###### Proof. (i). From (3.1), we are reduced to a relative Gromov-Witten invariant of ${\mathbb{P}}^{1}$ of the form $\langle{\mathbb{P}}^{1},\cdots,{\mathbb{P}}^{1},[pt],\cdots,[pt]\mid(t_{1},[pt]),\cdots,(t_{k},[pt])\rangle^{{\mathbb{P}}^{1},p_{1}}_{s}.$ A dimension count shows that this invariant of ${\mathbb{P}}^{1}$ is nonzero only if $s+k=2-q$. Since $s>0$ and $k>0$, the only possibility is $s=k=1$ and $q=0$. The proof of (ii) directly follows from (3.1) and Lemma 3.1. (iii). From (3.1), we have $\displaystyle\langle\iota^{!}(\beta_{1}),\cdots,\iota^{!}(\beta_{l})\mid(1,\gamma)\rangle^{Y,D}_{F}$ $\displaystyle=$ $\displaystyle\int_{Z}\beta_{1}\wedge\cdots\wedge\beta_{l}\wedge\gamma\langle[pt],\cdots,[pt]|(1,[pt])\rangle^{{\mathbb{P}}^{1},p_{1}}_{1}.$ It remains to prove $\langle[pt],\cdots,[pt]|(1,[pt])\rangle^{{\mathbb{P}}^{1},p_{1}}_{1}=1$. In fact, we consider the absolute invariant of ${\mathbb{P}}^{1}$ with $l+1$ point insertions: $\langle[pt],\cdots,[pt]\rangle^{{\mathbb{P}}^{1}}_{1}$. First of all, by divisor axiom, we know that this absolute invariant equals $1$. We apply the degeneration formula to this invariant of ${\mathbb{P}}^{1}$ and distribute one point insertion to one side and other $l$ point insertions to other side. Then we have $\displaystyle 1$ $\displaystyle=$ $\displaystyle\langle[pt],\cdots,[pt]\rangle^{{\mathbb{P}}^{1}}_{1}$ $\displaystyle=$ $\displaystyle\langle[pt],\cdots,[pt]|(1,[pt])\rangle^{{\mathbb{P}}^{1},p_{1}}_{1}\langle[pt]\mid(1,[pt])\rangle^{{\mathbb{P}}^{1},p_{1}}_{1}$ $\displaystyle=$ $\displaystyle\langle[pt],\cdots,[pt]|(1,[pt])\rangle^{{\mathbb{P}}^{1},p_{1}}_{1}.$ In the last equality, we used Lemma 3.1. This proved (iii). ∎ ### 3.2. A vanishing theorem In this subsection, we will prove a vanishing result for some relative Gromov- Witten invariants of ${\mathbb{P}}^{1}$-bundle, in particular, for some non- fiber homology class invariants. Let $\Gamma_{0}$ be a relative graph with the following data: 1. (i) a vertex decorated by a homology class $A\in H_{2}(Y,{\mathbb{Q}})$ and genus zero, 2. (ii) $l+q$ tails associated to $l+q$ absolute marked points, 3. (iii) $k$ relative tails associated to $k$ relative marked points. Denote by $A$ the homology class of the relative stable map $(\Sigma,f)$ to $(Y,D)$ and by $F$ the homology class of a fiber of $Y$. Let $T_{k}=\\{t_{1},\cdots,t_{k}\\}$ be a partition of $D\cdot A$ and $d_{i}$, $1\leq i\leq l$, be positive integers. Denote $d=\sum^{l}_{i=1}d_{i}$. Denote by $\iota:Z\longrightarrow Y$ the inclusion of $Z$ into $Y$ via the zero section of $Y$. Then for any $\beta\in H^{*}(Z,{\mathbb{R}})$, the inclusion map $i$ pushes forward the class $\beta$ to a cohomology class $\iota^{!}(\beta)\in H^{*}(Y,{\mathbb{Q}})$, determined by the pull-back map $\iota^{*}$ and Poincaré duality. ###### Proposition 3.3. Suppose $A\not=sF$ or $k+l+q\geq 3$. Assume that $Z^{*}(A)\geq\sum d_{i}$ and $c_{1}(L)(C)\geq 0$ for any $J$-holomorphic curve $C$ into $Z$. Then for any $\beta_{i}\in H^{*}(Z,{\mathbb{Q}})$, $1\leq i\leq l$, and any weighted partition ${\mathcal{T}}_{k}=\\{(t_{i},\delta_{i})\\}$ of $D\cdot A$, we have $\langle\varpi,\tau_{d_{1}-1}\iota^{!}(\beta_{1}),\cdots,\tau_{d_{l}-1}\iota^{!}(\beta_{l})\mid{\mathcal{T}}_{k}\rangle^{Y,D}_{\Gamma_{0},T_{k}}=0,$ where $\varpi$ consists of insertions of the form $\pi^{*}\alpha_{1},\cdots,\pi^{*}\alpha_{q}$. ###### Proof. The projection $\pi:Y={\mathbb{P}}(L\oplus{\mathbb{C}})\longrightarrow Z$ induces a map between the moduli spaces, denoted also by $\pi$, (8) $\pi:\overline{\mathcal{B}}_{\Gamma_{0},T_{k}}(Y,D,J)\longrightarrow\overline{\mathcal{B}}^{Z}_{\pi_{*}(A)}(0,k+l+q,J),$ where $\pi$ contracts the unstable rational component whose image is a fiber. $\pi$ is well-defined if $A\not=sF$ or $k+l+q\geq 3$. By the definition, $\pi$ commutes with the evaluation map. Furthermore, there is also a natural map(denoted by $\pi$ as well) on $\Omega^{0,1}$ commuting with the map on configuration spaces. Moreover, $\overline{\partial}$ commutes with $\pi$. Hence, it induces a map from $\mbox{Coker}\,L^{Y,D}$ to $\mbox{Coker}\,L^{Z}$. We claim that $\pi$ induces a map on a virtual neighborhood. Let $\omega$ be an integral symplectic form on $Z$. Using Siebert’s construction [S1], we can construct a bundle $\mathcal{E}$ dominating the local obstruction bundle generated by $\mbox{Coker}\,L^{Z}_{\pi(\Sigma,f)}$. $\pi^{*}{\mathcal{E}}$ is a bundle over $\overline{\mathcal{B}}_{\Gamma_{0},T_{k}}(Y,D,J)$. We want to show that $\pi^{*}\mathcal{E}$ dominates its local obstruction bundle. Let $(\Sigma,f)$ be a relative stable map of $(Y,D)$. Then we have ###### Lemma 3.4. $\mbox{Coker}\,L^{Y,D}_{\Sigma,f}$ is isomorphic to $\mbox{Coker}\,L^{Z}_{\pi(\Sigma,f)}$. ###### Proof. It is well-known that a stable map can be naturally decomposed into connected components lying outside of $D$( rigid factors) or completely inside $D$(rubber factors). Let $(\Sigma,f)$ be a rigid factor or a rubber factor with relative marked points $x_{1},\cdots,x_{r}$ such that $f(x_{i})\in Z$ or $D$ with order $k_{i}$. In both cases, it is a stable map into $Y$. We take the complex as $\tilde{L}^{Y,D}_{\Sigma,f}\times\sum T^{k_{i}}_{x_{i}}:\\{u\in\Omega^{0}(f^{*}TY)\mid u(x_{i})\in TZ\\}\longrightarrow\Omega^{0,1}(f^{*}TY\otimes_{i}{\mathcal{O}}_{\Sigma}(x_{i}))\oplus_{i}{\mathcal{J}}^{k_{i}}_{x_{i}}.$ We first study the cohomology $H^{0}_{\tilde{L}}$, $H^{1}_{\tilde{L}}$ of $\tilde{L}^{Y,D}_{\Sigma,f}$. There is a short exact sequence (9) $0\longrightarrow V\longrightarrow TY\longrightarrow\pi^{*}TZ\longrightarrow 0,$ where $V$ is the vertical tangent bundle. It induces a short exact sequence (10) $0\longrightarrow f^{*}V\longrightarrow f^{*}TY\longrightarrow f^{*}\pi^{*}TZ\longrightarrow 0.$ Choose a Hermitian metric and a unitary connection on $L$. It induces a splitting of $(\ref{proof-1})$. We choose a metric of $TY$ as the direct sum of the metric on $V$ and $\pi^{*}TZ$, where the second one is induced from a metric on $Z$. The Levi-Civita connection is a direct sum. Then $f^{*}V$ is a holomorphic line bundle with respect to pullback of the Levi-Civita connection. $(\ref{proof-2})$ induces a long exact sequence in cohomology $\displaystyle 0$ $\displaystyle\longrightarrow$ $\displaystyle H^{0}(f^{*}V)\longrightarrow H^{0}(f^{*}TY)\longrightarrow H^{0}(f^{*}\pi^{*}TZ)$ $\displaystyle\longrightarrow$ $\displaystyle H^{1}(f^{*}V)\longrightarrow H^{1}(f^{*}TY)\longrightarrow H^{1}(f^{*}\pi^{*}TZ)\longrightarrow 0.$ It induces exact sequences $\displaystyle 0\longrightarrow H^{0}_{\tilde{L}}(f^{*}V)\longrightarrow H^{0}_{\tilde{L}}(f^{*}TY)\longrightarrow H^{0}_{\tilde{L}}(f^{*}\pi^{*}TZ),$ $\displaystyle H^{1}_{\tilde{L}}(f^{*}V)\longrightarrow H^{1}_{\tilde{L}}(f^{*}TY)\longrightarrow H^{1}_{\tilde{L}}(f^{*}\pi^{*}TZ)\longrightarrow 0.$ Note that the normal bundle at the zero or infinity section is the restriction of $V$. It is obvious that $H^{1}_{\tilde{L}}(f^{*}\pi^{*}TZ)=H^{1}(f^{*}\pi^{*}TZ)$. An element of $H^{1}_{\tilde{L}}(f^{*}V)$ with residue in $TZ$ must have zero residue. Therefore, (11) $H^{1}_{\tilde{L}}(f^{*}V)=H^{1}(\tilde{f}^{*}V),$ where $(\tilde{\Sigma},\tilde{f})$ is obtained from $(\Sigma,f)$ by dropping the new marked points. For the same reason, $H^{0}_{\tilde{L}}(f^{*}V)=\\{v\in H^{0}(f^{*}V)\mid v(x_{i})=0\\}.$ We claim that $H^{1}(\tilde{f}^{*}V)=0$. Note that since $\Sigma$ is a tree of ${\mathbb{P}}^{1}$’s, we see that $H^{1}(L)=0$ for any line bundle $L$ on $\Sigma$ satisfying $\deg(L|_{\overline{\Sigma}})\geq 0$ for any irreducible component $\overline{\Sigma}$ of $\tilde{\Sigma}$. Now we have $\displaystyle\deg(\tilde{f}^{*}V|_{\overline{\Sigma}})$ $\displaystyle=$ $\displaystyle\tilde{f}_{*}[\tilde{\Sigma}]\cdot c_{1}(V)=\tilde{f}_{*}[\overline{\Sigma}]\cdot(\pi^{*}c_{1}(L)+2\xi)$ $\displaystyle=$ $\displaystyle c_{1}(N)\cdot(\pi\circ\tilde{f})_{*}[\tilde{\Sigma}]+2(\tilde{f}_{*}[\tilde{\Sigma}]\cdot\xi)\geq 0.$ Applying $L=\tilde{f}^{*}V$, we conclude that $H^{1}_{\tilde{L}}(\tilde{f}^{*}V)=0$. Next, we show that (12) $\oplus_{i}T^{k_{i}}_{x_{i}}:H^{0}_{\tilde{L}}(f^{*}TY)\longrightarrow\oplus_{i}{\mathcal{T}}^{k_{i}}_{x_{i}}$ is surjective. It is enough to show that the restriction to $H^{0}_{\tilde{L}}(f^{*}V)$ is surjective. Consider the exact sequence $0\longrightarrow\tilde{f}^{*}V\otimes_{i}{\mathcal{O}}(-k_{i}x_{i})\longrightarrow\tilde{f}^{*}V\longrightarrow\oplus_{i}\tilde{f}^{*}V_{k_{i}x_{i}}\longrightarrow 0.$ It induces a long exact sequence $H^{0}(\tilde{f}^{*}V)\longrightarrow\oplus_{i}H^{0}(\tilde{f}^{*}V_{k_{i}x_{i}})\longrightarrow H^{1}(\tilde{f}V\otimes_{i}{\mathcal{O}}(-k_{i}x_{i})).$ Over each $\overline{\Sigma}$, $\displaystyle\deg\tilde{f}^{*}V\otimes_{i}{\mathcal{O}}(-k_{i}x_{i})\mid_{\overline{\Sigma}}=\tilde{f}_{*}[\overline{\Sigma}]\cdot(\pi^{*}c_{1}(N)+2\xi)$ $\displaystyle=c_{1}(N)\cdot(\pi\circ\tilde{f})_{*}[\overline{\Sigma}]+2(\tilde{f}_{*}[\tilde{\Sigma}])\cdot\xi\geq 0.$ Hence, $H^{1}(\tilde{f}^{*}V\otimes_{i}{\mathcal{O}}(-k_{i}x_{i}))=0$. This implies that $H^{0}(\tilde{f}^{*}V)\longrightarrow\oplus H^{0}(\tilde{f}V_{k_{i}x_{i}})$ is surjective. Now we go back to $f$ and drop the constant term in $H^{0}(\tilde{f}^{*}V_{k_{i}x_{i}})$. $(\ref{proof-4})$ becomes (13) $H^{0}_{\tilde{L}}(f^{*}V)\longrightarrow\oplus_{i}{\mathcal{T}}^{k_{i}}_{x_{i}},$ which is obviously surjective. By $(\ref{proof-3})$, we have proved that $\mbox{Coker}(\tilde{L}^{Y,D}_{\Sigma,f}\times\sum T^{k_{i}}_{x_{i}})$ is isomorphic to $H^{1}(f^{*}\pi^{*}TZ)$. Then, we argue that $H^{1}(f^{*}\pi^{*}TZ)$ is isomorphic to $H^{1}(\pi(f)^{*}TZ)$. This is obvious if $\pi(f)$ contracts an unstable component ${\mathbb{P}}^{1}$, $\pi\circ f({\mathbb{P}}^{1})=$ constant and ${\mathbb{P}}^{1}$ has one or two special points. Moreover, $\pi(\Sigma)$ is obtained by contracting ${\mathbb{P}}^{1}$. Note that $f^{*}\pi^{*}TZ\mid_{{\mathbb{P}}^{1}}$ is trivial. The space of meromorphic $1$-forms on ${\mathbb{P}}^{1}$ with a simple pole at one or two points is zero or $1$-dimensional. If ${\mathbb{P}}^{1}$ has only one special point, the residue at the special point has to be zero. We can simply contract this component and remove the pole at the other component which ${\mathbb{P}}^{1}$ is connected to. If ${\mathbb{P}}^{1}$ has two special points, the residues at the two points have to be the same. Then we can remove this component and the joint residue at the two special points. Then we identify $H^{1}(f^{*}\pi^{*}TZ)$ and $H^{1}(\pi(f)^{*}TZ)$. Suppose that $(\Sigma,f)$ has more than one subfactor. Both $\mbox{Coker}\,L^{Y,D}_{\Sigma,f}$ and $\mbox{Coker}\,L^{Z}_{\pi(\Sigma,f)}$ are obtained by requiring the residues at the new marked points to be opposite to each other. Then our proof also extends to this case. Then we finish the proof of Lemma 3.4. ∎ Next, we continue the proof of Proposition 3.3 . Since we have identified the obstruction spaces, we first choose a stabilization term $\eta_{i}$ on $\overline{\mathcal{B}}^{Z}_{\pi_{*}(A)}(0,k+l+q,J)$ to dominate the local obstruction bundle generated by $\mbox{Coker}\,L^{Z}_{\pi(\Sigma,f)}$. Then, we pull back $\eta_{i}$ over $\overline{\mathcal{B}}_{\Gamma_{0},T_{k}}(Y,D,J)$. By Lemma 3.4, it dominates $\mbox{Coker}\,L^{Y,D}$. This implies that $\pi$ induces a smooth map on virtual neighborhood and a commutative diagram on obstruction bundles (14) $\begin{array}[]{ccccc}&&E_{Y,D}&\longrightarrow&E_{Z}\\\ &&\downarrow&&\downarrow\\\ \pi_{{\mathcal{S}}_{e}}&:&U^{Y,D}_{{\mathcal{S}}_{e}}&\longrightarrow&U^{Z}_{{\mathcal{S}}_{e}}.\end{array}$ Furthermore, the proper sections $S_{Y,D}$, $S_{Z}$ commutes with the above diagram. $\pi_{{\mathcal{S}}_{e}}$ commutes with the evaluation map for those $\beta_{i}$ classes. Choose a Thom form $\Theta$ of $E_{Z}$. Its pullback is the Thom form on $E_{Y,D}$ (still denoted by $\Theta$). It is clear that $\dim D_{i}=\dim U^{Y,D}_{{\mathcal{S}}_{e}}-2.$ By our construction, the $D_{i}$ intersect each other transversely. Note that $\displaystyle\dim U^{Y,D}_{{\mathcal{S}}_{e}}$ $\displaystyle=$ $\displaystyle\mbox{rank}\,E_{Z}+2(c^{Y}_{1}(A)+n-3+l+q+k-\sum t_{i}).$ $\displaystyle\dim U^{Z}_{{\mathcal{S}}_{e}}$ $\displaystyle=$ $\displaystyle\mbox{rank}\,E_{Z}+2(c^{Z}_{1}(\pi_{*}(A))+n-1-3+l+q+k).$ However, $c_{1}^{Y}(A)=c_{1}^{Z}(\pi_{*}(A))+c_{1}(N)(\pi_{*}(A))+2\sum t_{i}=c_{1}^{Z}(\pi_{*}(A))+Z^{*}(A)+\sum t_{i}.$ Hence, $\dim U^{Y,D}_{{\mathcal{S}}_{e}}-\dim U^{Z}_{{\mathcal{S}}_{e}}=2(Z^{*}(A)+1).$ By definition, we have (15) $\displaystyle\deg\Theta+\sum_{i}\deg(\beta_{i})+\deg\varpi+\sum_{j}\deg(\delta_{j})$ $\displaystyle=$ $\displaystyle\dim U^{Y,D}_{{\mathcal{S}}_{e}}-2d>\dim U^{Z}_{{\mathcal{S}}_{e}},$ where $\deg\varpi=\sum\deg\alpha_{j}$. Then, from (15)and $Z^{*}(A)\geq d$, we have $\displaystyle(S_{Y,D})^{*}\Theta\prod\psi_{i}^{d_{i}-1}ev^{*}_{i}\beta_{i}\wedge ev^{*}\varpi\wedge\prod ev^{*}_{j}\delta_{j}$ $\displaystyle=$ $\displaystyle((S_{Y,D})^{*}\Theta\prod\psi_{i}^{d_{i}-1}(\pi_{{\mathcal{S}}_{e}})^{*}(ev^{*}_{i}\beta_{i}\wedge\prod ev^{*}\varpi\wedge\prod ev^{*}_{j}\delta_{j})=0.$ In the last equality, we use $ev^{*}_{i}\beta_{i}\wedge\prod ev^{*}\varpi\wedge\prod ev^{*}_{j}\delta_{j}=0$ on $U^{Z}_{{\mathcal{S}}_{e}}$. Hence, the relative invariant is zero. This completes the proof of Proposition 3.3. ∎ ###### Remark 3.5. McDuff also proved the same result in the case without insertion classes $\beta_{i}$ by a totally different method; see Lemma $1.7$ in [M1]. ### 3.3. A nonvanishing theorem When we apply the degeneration formula, we often need to compute some special terms where the degeneration graph lies completely on the side of the projective bundle. In the previous subsection, we proved a vanishing theorem for some relative invariants of $(Y,D)$. In this subsection, we will consider the case where the relative invariants of $(Y,D)$ with empty relative insertion on $D$ are no longer zero and the invariant of $Z$ will contribute in a nontrivial way. Suppose that $A\in H_{2}(Z,{\mathbb{Z}})$. Denote by $i:Z\longrightarrow Y$ the embedding of $Z$ into $Y$ as the zero section. Consider the relative invariant of $(Y,D)$ $\langle\tau_{i_{1}}(\beta_{1}[Z]),\tau_{i_{2}}(\beta_{2}[Z]),\cdots,\tau_{i_{k}}(\beta_{k}[Z]),\varpi\mid\emptyset\rangle^{Y,D}_{0,A},$ where $\varpi$ consists of insertions of the form $\pi^{*}\alpha_{1},\cdots,\pi^{*}\alpha_{l}$. The dimension condition is $2\sum(i_{t}+1)+\sum\deg\beta_{t}+\deg\varpi=2(C_{1}^{Y}(A)+n-3+k+l).$ The dimension condition of the divisor invariant $\langle\tau_{i_{1}}(\beta_{1}),\cdots,\tau_{i_{k}}(\beta_{k}),i^{*}\varpi\rangle^{Z}_{0,A}$ is $2(C_{1}^{Z}(A)+n-1-3+k+l)=2\sum i_{t}+\sum\deg\beta_{t}+\deg\varpi.$ Since $C_{1}^{Y}(A)=C_{1}^{Z}(A)+Z\cdot A$, so both invariants are well- defined only when $k=Z\cdot A+1$. ###### Theorem 3.6. Let $A\in H_{2}(Z,{\mathbb{Z}})$. Suppose that $k=Z\cdot A+1$ and $C_{1}(L)(C)\geq 0$ for any holomorphic curve $C$ into $Z$. Then $\displaystyle\langle\varpi,\beta_{1}\cdot[Z],\beta_{2}\cdot[Z],\cdots,\beta_{k}\cdot[Z]\mid\emptyset\rangle^{Y,D}_{0,A}$ $\displaystyle=$ $\displaystyle\langle\iota^{*}\varpi,\beta_{1},\cdots,\beta_{k}\rangle^{Z}_{0,A},$ where $\varpi$ consists of insertions of the form $\pi^{*}\alpha_{1},\cdots,\pi^{*}\alpha_{l}$ and $\iota:Z\longrightarrow Y$ is the embedding of $Z$ into $Y$ as the zero section. ###### Proof. Choose a Hermitian metric and a unitary connection on $L$ such that they induce a splitting $0\longrightarrow V\longrightarrow TY\longrightarrow\pi^{*}TZ\longrightarrow 0,$ where $V$ is the vertical tangent bundle. We choose a metric of $TY$ as the direct sum of a metric on $V$ and $\pi^{*}TZ$, where the second one is induced from a metric on $Z$. The Levi-Civita connection is a direct sum. Therefore, we may choose almost complex structures $J_{Z}$ on $TZ$ and $J_{V}$ on $V$ such that we may choose the direct sum $J_{Z}\oplus J_{V}$ as an almost complex structure $J_{Y}$ on $TY$. It is easy to see that $\bar{\partial}$ commutes with $\pi$. ¿From Lemma 3.4, we know that the projection $\pi:Y\longrightarrow Z$ induces a smooth map $\pi_{{\mathcal{S}}_{e}}:U^{Y,D}_{{\mathcal{S}}_{e}}\longrightarrow U^{Z}_{{\mathcal{S}}_{e}}$ on virtual neighborhoods and the obstruction bundle $E_{Y,D}$ over $U^{Y,D}_{{\mathcal{S}}_{e}}$ is the pullback of the obstruction bundle $E_{Z}$ over $U^{Z}_{{\mathcal{S}}_{e}}$. Therefore, by the definition of Gromov-Witten invariants, we have (16) $\displaystyle\langle\varpi\beta_{1}\cdot[Z],\beta_{2}\cdot[Z],\cdots,\beta_{k}\cdot[Z]\mid\emptyset\rangle^{Y,D}_{0,A}$ $\displaystyle=$ $\displaystyle\deg(\pi_{{\mathcal{S}}_{e}})\langle i^{*}\varpi,\beta_{1},\cdots,\beta_{k}\rangle^{Z}_{0,A}.$ We claim that $\deg(\pi_{{\mathcal{S}}_{e}})=1$. In fact, by the construction of virtual neighborhoods, we know that for every generic element $({\mathbb{P}}^{1},x_{1},\cdots,x_{k+l},\tilde{f})\in U^{Z}_{{\mathcal{S}}_{e}}$, there is a section $\nu$ of the obstruction bundle $E_{Z}$ such that $\bar{\partial}_{J_{Z}}\tilde{f}=\nu.$ Suppose that a generic element $({\mathbb{P}}^{1},x_{1},\cdots,x_{k+l},f)\in U^{Y,D}_{{\mathcal{S}}_{e}}$ is a preimage of $({\mathbb{P}}^{1},x_{1},\cdots,x_{k+l},\tilde{f})$ under $\pi_{{\mathcal{S}}_{e}}$. That is, $f$ is a lifting of $\tilde{f}$ to $Y$ vanishing at the marked points $x_{1},\cdots,x_{k}$. Therefore, from the fact that $\bar{\partial}$ commutes with $\pi$, we have that $({\mathbb{P}}^{1},x_{1},\cdots,x_{k+l},f)$ satisfies (17) $\bar{\partial}_{J_{Y}}f=\pi^{*}_{{\mathcal{S}}_{e}}\nu.$ If we choose a local coordinate $(z,s)$ on $Y$, where $s$ is the Euclidean coordinate on the fiber ${\mathbb{P}}^{1}$, then locally we may write $f=(\tilde{f},f^{V})$. Therefore (17) locally can be written as (18) $\left\\{\begin{array}[]{lcl}\bar{\partial}_{J_{Z}}\tilde{f}&=&\nu,\\\ &&\\\ \bar{\partial}_{J_{V}}f^{V}&=&0\end{array}\right.$ Since $\bar{\partial}^{2}=0$ always holds on ${\mathbb{P}}^{1}$, it follows from a well-known fact of complex geometry that $f^{*}L$ is a holomorphic line bundle over ${\mathbb{P}}^{1}$. Moreover, (18) shows that $f^{V}$ gives rise to a holomorphic section of the bundle $f^{*}L$, up to ${\mathbb{C}}^{*}$, which vanishes at the marked points $x_{1},\cdots,x_{k}$. Since $\deg(f^{*}L)=Z\cdot A$, therefore, from our assumption that $k=Z\cdot A+1$, we know that $f^{*}L\otimes(-x_{1}-\cdots-x_{k})$ has no nonzero holomorphic sections. Therefore, $f^{V}\equiv 0$. This says that the only preimage of a generic element $({\mathbb{P}}^{1},x_{1},\cdots,x_{k+l},\tilde{f})$ in $U^{Z}_{{\mathcal{S}}_{e}}$ is itself. This implies $\deg(\pi_{{\mathcal{S}}_{e}})=1$. This proves the theorem. ∎ ## 4\. A comparison theorem Let $X$ be a compact symplectic manifold and $Z\subset X$ be a smooth symplectic submanifold of codimension $2$. $\iota:Z\longrightarrow X$ is the inclusion map. The cohomological push-forward $\iota^{!}:H^{*}(Z,{\mathbb{R}})\longrightarrow H^{*}(X,{\mathbb{R}})$ is determined by the pullback $\iota^{*}$ and Poincaré duality. ###### Definition 4.1. A symplectic divisor $Z$ is said to be positive if for some tamed almost complex structure $J$, $C_{1}(N_{Z})(A)>0$ for any $A$ represented by a non- trivial $J$-sphere in $Z$. This is a generalization of ample divisor from algebraic geometry. Define (19) $V:=\min\\{C_{1}(N_{Z|X})(A)>0\mid A\in H_{2}(Z,{\mathbb{Z}})\,\mbox{is a stably effective class}\\}.$ In [MP], the authors point out that the relative Gromov-Witten theory of $(X,Z)$ does not provide new invariants: the relative Gromov-Witten theory of $(X,Z)$ is completely determined by the absolute Gromov-Witten theory of $X$ and $Z$ in principle. In this section, under some positivity assumptions on the normal bundle of the divisor, we will give an explicit relation between the absolute and relative Gromov-Witten invariants, which we call as a comparison theorem. The main tool of this section is the degeneration formula of Gromov-Witten invariants. The central theorem of this section is ###### Theorem 4.2. Suppose that $Z$ is a positive divisor and $V\geq l$. Then for $A\in H_{2}(X,{\mathbb{Z}})$, $\alpha_{i}\in H^{*}(X,{\mathbb{R}})$, $1\leq i\leq\mu$, and $\beta_{j}\in H^{*}(Z,{\mathbb{R}})$, $1\leq j\leq l$, we have $\displaystyle\langle\alpha_{1},\cdots,\alpha_{\mu},\iota^{!}(\beta_{1}),\cdots,\iota^{!}(\beta_{l})\rangle^{X}_{A}$ $\displaystyle=$ $\displaystyle\sum_{\mathcal{T}}\langle\alpha_{1},\cdots,\alpha_{\mu}\mid{\mathcal{T}}\rangle_{A}^{X,Z},$ where the summation runs over all possible weighted partitions ${\mathcal{T}}=\\{(1,\gamma_{1})$, $\cdots,(1,\gamma_{q}),(1,[Z]),\cdots,(1,[Z])\\}$ where $\gamma_{i}$’s are the products of some $\beta_{j}$ classes. ###### Proof. We perform the symplectic cutting along the boundary of a tubular neighborhood of $Z$. Then we have $\overline{X}^{-}=X$, $\overline{X}^{+}={\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})$. Since $\beta_{i}\in H^{*}(Z,{\mathbb{R}})$, we choose the support of $\iota^{!}(\beta_{i})$ near $Z$. Then, $\iota^{!}(\beta_{i})^{-}=0$, $\iota^{!}(\beta_{i})^{+}=\iota^{!}(\beta_{i})$. Here $\iota$ in the second term is understood as the inclusion map of $Z$ via the zero section into $Y={\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})$. Up to a rational multiple, each $\alpha_{i}$ is Poincaré dual to an immersed submanifold $W_{i}$. We can perturb $W_{i}$ to be transverse to $Z$. In a neighborhood of $Z$, $W_{i}$ is $\pi^{-1}(W_{i}\cap Z)$, where $\pi:N_{Z|X}\longrightarrow Z$ is the projection. Clearly, $\pi$ induces the projection ${\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})\longrightarrow Z$, still denoted by $\pi$. The symplectic cutting naturally decomposes $W_{i}$ into $W^{-}_{i}=W_{i}$, $W^{+}_{i}=h^{*}(\alpha_{i}|_{Z})=h^{*}\tilde{\alpha}_{i}$. In other words, we can choose $\alpha_{i}^{-}=\alpha_{i}$, $\alpha^{+}_{i}=h^{*}(\alpha_{i}|_{Z})$. Now we apply the degeneration formula for invariants $\langle\alpha_{1},\cdots,\alpha_{\mu},\iota^{!}(\beta_{1}),\cdots,\iota^{!}(\beta_{l})\rangle_{A}^{X}$ and express it as a summation of products of relative invariants of $(X,Z)$ and $(Y,D)$. Moreover, from the degeneration formula, each summand $\Psi_{C}$ may consist of a product of relative Gromov-Witten invariants with disconnected domain curves of both $(X,Z)$ and $(Y,D)$. On the side of $Y$, there may be several disjoint components. Let $A^{\prime}$ be the total homology class. Then, from our assumption, we have $Z\cdot A^{\prime}=Z\cdot A\geq d$. Suppose that we have a nonzero summand $\Psi_{C}\not=0$. We claim that each factor from the relative Gromov-Witten invariants of $(Y,D)$ must be in the form $\langle\iota^{!}(\beta_{i_{1}}),\cdots,\iota^{!}(\beta_{i_{t}})|(1,\gamma)\rangle^{Y,D}_{F}$ or $\langle\,|(1,[pt])\rangle^{Y,D}_{F}$,where $F$ is the homology class of a fiber of $Y$ and $\gamma$ is a basis element of $H^{*}(D,{\mathbb{R}})$ such that $\int_{Z}\beta_{i_{1}}\wedge\cdots\wedge\beta_{i_{t}}\wedge\gamma\not=0$. Note that for these components, $Z^{*}(sF)=s$. From our assumption $V\geq d$ and Proposition 3.3, the nonzero factor of the relative Gromov-Witten invariants of $(Y,D)$ must be the fiber class relative invariants. From Proposition 3.2, we know that the nonzero factor must be of the form $\langle\iota^{!}(\beta_{i_{1}}),\cdots,\iota^{!}(\beta_{i_{t}})|(1,\gamma)\rangle^{Y,D}_{F}$ or $\langle\,|(1,[pt])\rangle^{Y,D}_{F}$. Moreover, if some $\beta_{i}=[pt]$, then the marked point must be in a two-point component and the nonzero relative invariant must be $\langle\iota^{!}([pt])|(1,[Z])\rangle^{Y,D}_{F}=1$. Since there are no vanishing two-cycles in this case, we may write down the summation as follows. $\displaystyle\langle\alpha_{1},\cdots,\alpha_{\mu},\iota^{!}(\beta_{1}),\cdots,\iota^{!}(\beta_{l})\rangle^{X}_{A}$ $\displaystyle=$ $\displaystyle\sum_{\mathcal{T}}\langle\alpha_{1},\cdots,\alpha_{\mu}\mid{\mathcal{T}}\rangle_{A}^{X,Z},$ where the summation runs over all possible weighted partitions ${\mathcal{T}}=\\{(1,\gamma_{1})$, $\cdots,(1,\gamma_{q}),(1,[Z]),\cdots,(1,[Z])\\}$ where $\gamma_{i}^{\prime}$ are the product of some $\beta_{j}$ classes. We complete the proof of our comparison theorem. ∎ ###### Corollary 4.3. Under the assumption of Theorem 4.2. If the product of any two $\beta_{j}$ classes vanishes, then we have $\langle\alpha_{1},\cdots,\alpha_{\mu},\iota^{!}(\beta_{1}),\cdots,\iota^{!}(\beta_{l})\rangle^{X}_{A}=\langle\alpha_{1},\cdots,\alpha_{l}\mid{\mathcal{T}}\rangle^{X,Z}_{A},$ where ${\mathcal{T}}=\\{(1,\beta_{1}),\cdots,(1,\beta_{l}),(1,[Z]),\cdots,(1,[Z])\\}$ is a weighted partition of $Z\cdot A$. ## 5\. Rationally connected symplectic divisors ### 5.1. Rationally connectedness in algebraic geometry The basic reference for this subsection is [A]. We refer to [C, D, K, KMM1, KMM2, V] for more details. Let us recall the notion of rational connectedness in algebraic geometry. ###### Definition 5.1. Let $X$ be a smooth complex projective variety of positive dimension. We say that $X$ is rationally connected if one of the following equivalent conditions holds. 1. (1) Any two points of $X$ can be connected by a rational curve (called as rationally connected). 2. (2) Two general points of $X$ can be connected by a chain of rational curves (called as rationally chain-connected). 3. (3) Any finite set of points in $X$ can be connected by a rational curve. 4. (4) Two general points of $X$ can be connected by a very free rational curve. Here we say that a rational curve $C\subset X$ is a very free curve if there is a surjective morphism $f:{\mathbb{P}}^{1}\longrightarrow C$ such that $f^{*}X\cong\oplus_{i=1}^{\dim X}{\mathcal{O}}_{{\mathbb{P}}^{1}}(a_{i}),\,\,\,\,\,\mbox{with all}\,\,a_{i}\geq 1.$ Next let us look at some properties of rationally connected varieties. ###### Proposition 5.2. The following properties of rationally connected manifolds hold: 1. (1) Rationally connectedness is a birational invariant. 2. (2) Rationally connectedness is invariant under smooth deformation. 3. (3) If $X$ is rationally connected, then $H^{0}(X,(\Omega^{1}_{X})^{\otimes m})=0$ for every $m\geq 1$. 4. (4) Fano varieties (i.e., smooth complex projective varieties $X$ for which $-K_{X}$ is ample) are rationally connected. In particular, smooth hypersurfaces of degree $d$ in ${\mathbb{P}}^{n}$ are rationally connected for $d\leq n$. 5. (5) Rationally connected varieties are well behaved under fibration, i.e.: Let $X$ be s smooth complex projective variety. Assume that there exists a surjective morphism $f:X\longrightarrow Y$ with $Y$ and the general fiber of $f$ rationally connected. Then $X$ is rationally connected. An important theorem connecting birational geometry to Gromov-Witten theory is the result of Kollár and Ruan [K], [R1]: a uniruled projective manifold has a nonzero genus zero GW-invariant with a point insertion. A longstanding problem in Gromov-Witten theory is that a similar result with two point insertions should also hold for rationally connected projective manifolds. ### 5.2. Rationally connected symplectic divisors In this subsection, we want to apply our comparison theorem to study the $k$-point rationally connectedness properties. We will show that a symplectic manifold is $k$-point strongly rationally connected if it contains a k-point strongly rationally connected symplecitc divisor with sufficiently positive normal bundle. Our main tool is the degeneration formula of Gromov-Witten invariants and our comparison Theorem 4.2. Before we state our main theorem, we want first to define the notion of $k$-point rational connectedness. ###### Definition 5.3. Let $A\in H_{2}(X,{\mathbb{Z}})$ be a nonzero class. $A$ is said to be a $k$-point rationally connected class if there is a nonzero Gromov-Witten invariant (20) $\langle\tau_{d_{1}}[pt],\cdots,\tau_{d_{k}}[pt],\tau_{d_{k+1}}\alpha_{k+1},\cdots,\tau_{d_{l}}\alpha_{l}\rangle^{X}_{A},$ where $\alpha_{i}\in H^{*}(X,{\mathbb{R}})$ and $d_{1},\cdots,d_{l}$ are non- negative integers. We call a class $A\in H_{2}(X,{\mathbb{Z}})$ a $k$-point strongly rationally connected class if $d_{i}=0$, $1\leq i\leq l$, in (20). ###### Definition 5.4. $X$ is said to be (symplectic) $k$-point (strongly ) rationally connected if there is a $k$-point (strongly) rationally connected class. We simply call a $2$-point (strongly) rationally connected symplectic manifold as (strongly) rationally connected symplectic manifold. ###### Remark 5.5. From the definition of uniruledness of [HLR], a $1$-point rationally connected symplectic manifold is equivalent to a uniruled symplectic manifold. From the definitions, we know that a $k$-point (strongly) rational connected symplectic manifold must be uniruled. ###### Remark 5.6. It is possible that $k$-point rational connectedness is equivalent to $k$-point strongly rational connectedness. We do not know how to prove this. ###### Example 5.7. It is well-known that for any positive integer $k$, the projective space ${\mathbb{P}}^{n}$ is $k$-point strongly rationally connected. ###### Example 5.8. Let $G(k,n)$ be the Grassmannian manifold of $k$-planes in ${\mathbb{C}}^{n}$. It is well-known that the classical cohomology of $Gr(k,n)$ has a basis of Schubert classes $\sigma_{\lambda}$, as $\lambda$ varies over partitions whose Young diagram fits in a $k$ by $n-k$ rectangle. The (complex) codimension of $\sigma_{\lambda}$ is $|\lambda|=\sum\lambda_{i}$, the number of boxes in the Young diagram. The quantum cohomology of the Grassmannian is a free module over the polynomial ring ${\mathbb{Z}}[q]$, with a basis of Schubert classes; the variable $q$ has (complex) degree $n$. The quantum product $\sigma_{\lambda}\star\sigma_{\mu}$ is a finite sum of terms $q^{d}\sigma_{\nu}$, the sum over $d\geq 0$ and $|\nu|=|\lambda|+|\mu|-dn$, each occurring with a nonnegative coefficient (a Gromov-Witten invariant). Denote by $\rho=\sigma_{((n-k)^{k})}$ the class of a point. In [BCF], the authors proved that $\sigma_{\rho}\star\sigma_{\rho}=q^{k}\sigma_{((n-2k)^{k})}$ if $k\leq n-k$, and $\sigma_{\rho}\star\sigma_{\rho}=q^{n-k}\sigma_{((n-k)^{2k-n})}$ if $n-k\leq k$. This means that the Grassmannian $Gr(k,n)$ is symplectic rationally connected. ###### Example 5.9. For any integer $d\geq 0$, Consider the Grassmannian $G(d,2d)$. Buch-Kresch- Tamvakis [BKT] proved that for three points $U,V,W\in G(d,2d)$ pairwise in general position, there is a unique morphism $f:{\mathbb{P}}^{1}\longrightarrow G(d,2d)$ of degree $d$ such that $f(0)=U$, $f(1)=V$ and $f(\infty)=W$. This implies that the Gromov-Witten invariant $\langle[pt],[pt],[pt]\rangle^{G(d,2d)}_{d}=1$. Therefore, $G(d,2d)$ are $3$-point strongly rationally connected. ###### Example 5.10. Let ${\mathbb{H}}=\mbox{Hilb}^{2}({\mathbb{P}}^{1}\times{\mathbb{P}}^{1})$ the Hilbert scheme of points on ${\mathbb{P}}^{1}\times{\mathbb{P}}^{1}$. In [P], the author gave a ${\mathbb{Q}}$-basis for $A*({\mathbb{H}})$ as follows: $T_{0}=[{\mathbb{H}}]$, $T_{1}$, $T_{2}$, $T_{3}$, $T_{4}$, $T_{5}=T_{1}T_{2}$, $T_{6}=T_{1}^{2}$, $T_{7}=T_{2}^{2}$, $T_{8}=T_{1}T_{3}$, $T_{9}=T_{2}T_{3}$, $T_{10}=C_{2}+F$, $T_{11}=C_{1}+F$, $T_{12}=C_{1}+C_{2}+F$ and $T_{13}$ the class of a point. The author also computed the quantum product $T_{4}\star T_{4}=T_{13}+2q_{1}q_{2}q_{3}^{2}T_{0}$ and $T_{4}\star T_{4}\star T_{4}=2q_{1}q_{2}q_{3}^{2}T_{4}$. So we have $T_{13}\star T_{13}=-2q_{1}q_{2}q_{3}^{2}T_{13}\not=0$. This implies that ${\mathbb{H}}$ is symplectic rationally connected. ¿From the computation of [G], it is easy to know that $\mbox{Hilb}^{2}({\mathbb{P}}^{2})$ also is symplectic rationally connected. Since Gromov-Witten invariants are invariant under smooth symplectic deformations, $k$-point rational connectedness is invariant under smooth symplectic deformations. It is not yet known whether a projective rationally connected manifold is symplectic rationally connected. Moreover, so far, we could not show that this notion is invariant under symplectic birational cobordisms defined in [HLR]. We will leave this for future research. However, we would like to mention some partial results along this direction. From the blowup formula of Gromov-Witten invariants in [H1, H2, H3, HZ, La], we have ###### Proposition 5.11. Suppose that $X$ is a $k$-point strongly rationally connected symplectic manifold. Let $\tilde{X}$ be the blowup of $X$ along a finite number of points or some special submanifolds with convex normal bundles (see, [H1, La]). Then $\tilde{X}$ is $k$-point strongly rationally connected. In 1991, McDuff [M3] first observed that a semi-positive symplectic $4$-manifold, which contains a submanifold $P$ symplectomorphic to ${\mathbb{P}}^{1}$ whose normal Chern number is non-negative, must be uniruled. In [LtjR], the authors generalize McDuff’s result to more general situations. More importantly, they gave a rather general from divisor to ambient space inductive construction of uniruled symplectic manifolds. In this subsection, we will generalize their inductive construction to the case of rationally connected symplectic manifolds. Suppose that $X$ is a compact symplectic manifold and $Z\subset X$ is a symplectic submanifold of codimension $2$. Denote by $N_{Z|X}$ the normal bundle of $Z$ in $X$. Denote by $\iota:Z\subset X$ the inclusion of $Z$ into $X$. Let $V$ be the minimal normal Chern number defined in (19). We call a class $A\in H_{2}(Z,{\mathbb{R}})$ a minimal class if $Z\cdot A=V$. ###### Theorem 5.12. Suppose that $X$ is a compact symplectic manifold and $Z\subset X$ is a symplectic submanifold of codimension $2$. If $Z$ is k-point strongly rationally connected and $A\in H_{2}(Z,{\mathbb{Z}})$ is a minimal class such that (21) $\langle\iota^{*}\alpha_{1},\cdots,\iota^{*}\alpha_{l},[pt],\cdots,[pt],\beta_{k+1},\cdots,\beta_{r}\rangle^{Z}_{A}\not=0$ for some $r\leq V+1$, $\beta_{i}\in H^{*}(Z,{\mathbb{R}})$ and $\alpha_{j}\in H^{*}(X,{\mathbb{R}})$, then $X$ is k-point strongly rationally connected. In particular, if $\iota:Z\rightarrow X$ is homologically injective, then $X$ is $k$-point strongly symplectic rational connected if $k\leq V+1$. ###### Proof. Since we can always increase the number of $Z$-insertions by adding divisor insertions in (21), therefore, without loss of generality, we may assume that $r=Z\cdot A+1$. Consider the following Gromov-Witten invariant of $X$: (22) $\langle\alpha_{1},\cdots,\alpha_{l},[pt],\cdots,[pt],\beta_{k+1}\cdot[Z],\cdots,\beta_{r}\cdot[Z]\rangle^{X}_{A}.$ If the invariant (22) is nonzero, then we are done. So in the following we assume that the invariant (22) equals zero. To find a nonzero Gromov-Witten invariant of $X$ with at least $k$ point insertions, we first apply the degeneration formula to the invariant (22) to obtain a nonzero relative Gromov-Witten invariant of $(X,Z)$ with at least $k$ point insertions, then use our comparison theorem to obtain a nonzero Gromov- Witten invariant of $X$. We perform the symplectic cutting along the boundary of a tubular neighborhood of $Z$. Then we have $\overline{X}^{-}=X$, $\overline{X}^{+}=Y={\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})$. Since $\beta_{i}\in H^{*}(Z,{\mathbb{R}})$, we choose the support of $[pt]\cdot Z$, $\beta_{i}\cdot Z$ near $Z$. Then,$([pt])^{-}=0$, $(\beta_{i}\cdot Z)^{-}=0$, $([pt])^{+}=[pt]$, $(\beta_{i}\cdot Z)^{+}=\iota^{!}(\beta_{i})$. Here $\iota$ in the second term is understood as the inclusion map of $Z$ via the zero section into $Y={\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})$. Up to a rational multiple, each $\alpha_{i}$ is Poincaré dual to an immersed submanifold $W_{i}$. We can perturb $W_{i}$ to be transverse to $Z$. In a neighborhood of $Z$, $W_{i}$ is $\pi^{-1}(W_{i}\cap Z)$, where $\pi:N_{Z|X}\longrightarrow Z$ is the projection. Clearly, $\pi$ induces the projection ${\mathbb{P}}(N_{Z|X}\oplus{\mathbb{C}})\longrightarrow Z$, still denoted by $\pi$. The symplectic cutting naturally decomposes $W_{i}$ into $W^{-}_{i}=W_{i}$, $W^{+}_{i}=h^{*}(\alpha_{i}|_{Z})=h^{*}\tilde{\alpha}_{i}$. In other words, we can choose $\alpha_{i}^{-}=\alpha_{i}$, $\alpha^{+}_{i}=h^{*}(\alpha_{i}|_{Z})$. Now we apply the degeneration formula for the invariant $\langle\alpha_{1},\cdots,\alpha_{l},[pt],\cdots,[pt],\beta_{k+1}\cdot[Z],\cdots,\beta_{r}\cdot[Z]\rangle^{X}_{A}.$ and express it as a summation of products of relative invariants of $(X,Z)$ and $(Y,D)$. Moreover, from the degeneration formula, each summand $\Psi_{C}$ may consist of a product of relative Gromov-Witten invariants with disconnected domain curves of both $(X,Z)$ and $(Y,D)$. On the side of $Y$, there may be several disjoint components. We claim that every component on the side of $Y$ is a multiple of the fiber class. Suppose that there is a component on the side of $Y$ which has a homology class $A^{\prime}+\mu F$ where $A^{\prime}\in H_{2}(Z,{\mathbb{Z}})$ and $F$ is the fiber class. Then we have $Z\cdot(A^{\prime}+\mu F)=Z\cdot A^{\prime}+\mu$. Since every component must intersect the infinity section of $Y$, $\mu>0$. Therefore, we have $Z\cdot(A^{\prime}+\mu F)\geq V+1$. From Proposition 3.3, we know that the contribution of this component to the corresponding relative Gromov-Witten invariants of $(Y,D)$ must be zero. So the corresponding summand in the degeneration formula must be zero. From Proposition 3.2 and the same argument as in the proof of Theorem 4.2 , we have $\displaystyle 0$ $\displaystyle=$ $\displaystyle\langle\alpha_{1},\cdots,\alpha_{l},[pt],\cdots,[pt],\beta_{k+1}\cdot[Z],\cdots,\beta_{r}\cdot[Z]\rangle^{X}_{A}$ $\displaystyle=$ $\displaystyle\sum_{\mu}C_{\mu}\langle\alpha_{1},\cdots,\alpha_{l}\mid\mu\rangle^{X,Z}_{A}$ $\displaystyle+\langle i^{*}\alpha_{1},\cdots,i^{*}\alpha_{l},[pt],\cdots,[pt],\beta_{k+1}\cdot[Z],\cdots,\beta_{r}\cdot[Z]\mid\emptyset\rangle^{Y,D}_{A}$ $\displaystyle=$ $\displaystyle\sum_{\mu}C_{\mu}\langle\alpha_{1},\cdots,\alpha_{l}\mid\mu\rangle^{X,Z}_{A}$ $\displaystyle+\langle i^{*}\alpha_{1},\cdots,i^{*}\alpha_{l},[pt],\cdots,[pt],\beta_{k+1},\cdots,\beta_{r}\rangle^{Z}_{A},$ where we used Theorem 3.6 in the last equality and the summation runs over the possible partitions $\mu=\\{(1,[pt]),\cdots,(1,[pt]),(1,\gamma_{1}),\cdots,(1,\gamma_{q}),(1,[Z])$, $\cdots,(1,[Z])\\}$ where $\gamma_{i}$ are the product of some $\beta_{i}$ classes.. From our assumption (21), we have (23) $\sum_{\mu}C_{\mu}\langle\alpha_{1},\cdots,\alpha_{l}\mid\mu\rangle^{X,Z}_{A}\not=0.$ Denote by $\langle\alpha_{1},\cdots,\alpha_{l}\mid\mu_{0}\rangle^{X,Z}_{A}$ the minimal nonzero relative invariant in the summand (23)in the sense of Definition 2.11. Write $\mu_{0}=\\{(1,[pt]),\cdots$, $(1,[pt]),(1,\gamma_{1}),\cdots,(1,\gamma_{q}),(1,[Z]),\cdots,(1,[Z])\\}$. Then it is easy know that the product of any two $\gamma_{i}$ and $\gamma_{j}$ vanishes. Now we consider the following absolute Gromov-Witten invariant $\langle\alpha_{1},\cdots,\alpha_{l},[pt],\cdots,[pt],\gamma_{1}\cdot[Z],\cdots,\gamma_{l}\cdot[Z],[Z],\cdots,[Z]\rangle^{X}_{A}.$ Applying the degeneration formula and always distribute the point insertions to the side of $(Y,D)$. Therefore, from Theorem 4.2, we have $\displaystyle\langle\alpha_{1},\cdots,\alpha_{l},[pt],\cdots,[pt],\gamma_{1}\cdot[Z],\cdots,\gamma_{q}\cdot[Z]\rangle^{X}_{A}$ $\displaystyle=$ $\displaystyle\langle\alpha_{1},\cdots,\alpha_{l}\mid\mu_{0}\rangle^{X,Z}_{A}\not=0.$ This implies that $X$ is $k$-point strongly rationally connected. This proves our theorem. ∎ It is well-known that ${\mathbb{P}}^{n-1}$ is strongly rationally connected. Therefore, from Theorem 5.12, we have ###### Corollary 5.13. Let $(X,\omega)$ be a compact $2n$-dimensional symplectic manifold which contains a submanifold $P$ symplectomorphic to ${\mathbb{P}}^{n-1}$ whose normal Chern number $m\geq 2$. Then $X$ is strongly rationally connected. ## References * [A] C. Araujo, Rationally connected varities, arXiv:math.AG/0503305 * [BCF] A. Bertram, I. Ciocan-Fontanine, W. Fulton, Quantum multiplication of Schur Polynomials, J. Alg., 219(1999), 728-746. * [BKT] A. S. Buch, A. Kresch, H. Tamvakis, Gromov-Witten invariants on Grassmannians, J. Amer. Math. Soc., 16(4)(2003), 901-915. * [C] F. Campana, Connexité tationanelle des variétés de Fano, Ann. Sci. École Norm. Sup. 25(4)(1992), no.5, 539-545. * [D] O. Debarre, Higher-dimensional algebraic geometry, Universitext, Springer-Verlag, New York, 2001. * [G] T. Graber, Enumerative geometry of hyperelliptic plane curves, J. Alg. Geom., 10(2001), 725-755. * [H1] J. Hu, Gromov-Witten invariants of blow-ups along points and curves, Math. Z. 233(2000), 709-739. * [H2] J. Hu, Gromov-Witten invariants of blow-ups along surfaces, Compositio Math., 125(2001),345-352. * [H3] J. Hu, Gromov-Witten invariants of blowups along semi-positive submanifold, Adv. in Math. Research, vol. 3(2003), 103-109. * [HLR] J. Hu, T.-J. Li, Y, Ruan, Birational cobordism invariance of uniruled symplectic manifolds, math.SG/0611592, to appear in Invent. Math. * [HZ] J. Hu, H. Zhang, Elliptic Gromov-Witten invariants of blowups along surfaces, Int. J. Math. Math. Sci, 2005:1(2005), 81-90. * [IP] E. Ionel, T. Parker, Relative Gromov-Witten invariants, Ann. of Math., 157(2)(2003), 45-96. * [K] J. Kollár, Rationally curves on algebraic varieties, Ergebnisse der Mathematik und ihrer Grenzgebiete, vol. 32, Springer-Verlag, Berlin, 1996. * [KMM1] J. Kollár, Y. Miyaoka, S. Mori, Rational connectedness and boundedness of Fano manifolds, J. Differential Geom. 36(3)(1992), 765-779. * [KMM2] J. Kollár, Y. Miyaoka, S. Mori, Rationally connected varieties, J. Algebraic Geom. 1(3)(1992),429-448. * [La] H. Lai, Gromov-Witten invariants of blowups along submanifolds with convex normal bundles, arXiv:math.AG/0710.3968. * [L] E. Lerman, Symplectic cuts, Math. Research Lett. 2(1995), 247-258. * [Li1] J. Li, Stable morphisms to singular schemes and relative stable morphisms, J. Diff. Geom. 57(2001),509-578. * [Li2] J. Li, Relative Gromov-Witten invariants and a degeneration formula of Gromov-Witten invariants, J. Diff. Geom. 60(2002), 199-293 * [LR] A. Li, Y. Ruan, Symplectic surgery and Gromov-Witten invariants of Calabi-Yau 3-folds, Invent. Math. 145(2001), 151-218. * [LtjR] T.-J. Li, Y. Ruan, Uniruled symplectic divisors, arXiv:math.SG/0711.4254 * [MP] D. Maulik, R. Pandharipande, A topological view of Gromov-Witten theory, Topology, 45(5)(2006), 887-918. * [M1] D. McDuff, Comparing absolute and relative Gromov-Witten invariants, preprint. * [M2] D. McDuff, Hamiltonian $S^{1}$-manifolds are uniruled, preprint, 2007 * [M3] D. McDuff, Symplectic manifolds with contact type boundaries, Invent. Math. 103(1991), 651-671. * [OP] A. Okounkov, R. Pandharipande, Gromov-Witten theory, Hurwitz numbers, and completed cycles, Ann. of Math., 163(2)(2006), 517-560. * [P] D. Pontoni, Quantum cohomology of $\mbox{Hilb}^{2}({\mathbb{P}}^{1}\times{\mathbb{P}}^{1})$ and enumerative applications, Trans. Amer. Math. Soc., 359(2007), 5419-5448. * [Q] Z. Qin, private communication. * [R1] Y. Ruan, Virtual neighborhoods and pseudoholomorphic curves, Turkish J. Math., 23(1999), 161-231. * [R2] Y. Ruan, Surgery, Quantum cohomology and birational geometry, Northern California Symplectic Geometry Seminar, 183-198, Amer. Amth. Soc. Transl. Ser.2, 196. Amer. Math. Soc., Providence, RI, 1999\. * [R3] Y. Ruan, Topological sigma model and Donaldson type invariants in Gromov theory, Duke Math. J. 83(2)(1996),461-500. * [S1] N. Siebert, Gromov-Witten invariants for general symplectic manifolds, New trends in algebraic geometry(Warwick, 1996), 375-424, Cambrdge Uni. Press, 1999. * [V] C. Voisin, Rationally connected $3$-folds and symplectic geometry, arXiv:0801.1396.

arxiv-papers

2024-09-04T02:48:53.561023

0802.0628

# Heegaard Floer invariants of Legendrian knots in contact three–manifolds Paolo Lisca Dipartimento di Matematica “L. Tonelli”, Università di Pisa I-56127 Pisa, Italy lisca@dm.unipi.it Peter Ozsváth Department of Mathematics, Columbia University, New York 10027, USA petero@math.columbia.edu András I. Stipsicz Rényi Institute of Mathematics, Budapest, Hungary and Department of Mathematics, Columbia University, New York 10027, USA stipsicz@renyi.hu Zoltán Szabó Department of Mathematics, Princeton University, Princeton, New Jersey 08544, USA szabo@math.princeton.edu ###### Abstract We define invariants of null–homologous Legendrian and transverse knots in contact 3–manifolds. The invariants are determined by elements of the knot Floer homology of the underlying smooth knot. We compute these invariants, and show that they do not vanish for certain non–loose knots in overtwisted 3–spheres. Moreover, we apply the invariants to find transversely non–simple knot types in many overtwisted contact 3–manifolds. ###### keywords: Legendrian knots, transverse knots, Heegaard Floer homology 57M27 57R58 ## 1 Introduction The reformulation of knot Floer homology (introduced originally in [35, 39]) through grid diagrams [26, 27] provided not only a combinatorial way of computing the knot Floer homology groups of knots in $S^{3}$, but also showed a natural way of defining invariants of Legendrian and transverse knots in the standard contact 3–sphere $(S^{3},\xi_{st})$, see [37]. As it is shown in [29, 44], such invariants can be effectively applied to study transverse simplicity of knot types. The definition of these invariants relies heavily on the presentation of the knot through a grid diagram, hence does not generalize directly to Legendrian and transverse knots in other closed contact 3–manifolds. The aim of the present paper is to define invariants for any null–homologous Legendrian (and transverse) knot $L\subset(Y,\xi)$ (resp. $T\subset(Y,\xi)$). As demonstrated by explicit computations, these constructions give interesting invariants for Legendrian and transverse knots, even in cases where the ambient contact structure is overtwisted. Recall [35] that a smooth null–homologous knot $K\subset Y$ in a closed 3–manifold $Y$ gives rise to the _knot Floer homology groups_ $\widehat{\mathrm{HFK}}(Y,K)$ and $\mathrm{HFK}^{-}(Y,K)$ (see also Section 2 for a short review of Heegaard Floer homology). These groups are computed as homologies of appropriate chain complexes, which in turn result from doubly–pointed Heegaard diagrams of the pair $(Y,K)$. For an isotopic pair $K_{1},K_{2}\subset Y$ the corresponding chain complexes are quasi–isomorphic and hence the homologies are isomorphic. In this sense the knot Floer homology groups (as abstract groups) are invariants of the isotopy class of the knot $K$. For the sake of simplicity, throughout this paper we use Floer homology with coefficients in $\mathbb{F}:=\mathbb{Z}/2\mathbb{Z}$. In this paper we define the invariants ${\mathfrak{L}}(L)$ and $\widehat{\mathfrak{L}}(L)$ of an oriented null–homologous Legendrian knot $L\subset(Y,\xi)$. For an introduction to Legendrian and transverse knots see [10]. (We will always assume that $Y$ is oriented and $\xi$ is cooriented.) Recall that the contact invariant $c(Y,\xi)$ of a contact 3–manifold $(Y,\xi)$ — as defined in [34] — is an element (up to sign) of the Floer homology of $-Y$ (rather than of $Y$). In the same vein, the invariants ${\mathfrak{L}}(L)$ and $\widehat{\mathfrak{L}}(L)$ are determined by elements in the Floer homology of $(-Y,L)$. If $Y$ admits an orientation reversing diffeomorphism $\mu$ (like $S^{3}$ does), then the Floer homology of $(-Y,L)$ can be identified with the Floer homology of $(Y,\mu(L))$. More precisely, we will show that, given an oriented, null–homologous Legendrian knot $L$ in a closed contact 3–manifold $(Y,\xi)$, one can choose certain auxiliary data $D$ which, together with $L$, determine a cycle ${\bf{x}}(L,D)$ in a complex defining the $\mathbb{F}[U]$–modules $\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi})$ or $\widehat{\mathrm{HFK}}(-Y,L,\mathbf{t}_{\xi})$ (with trivial $U$–action), where $\mathbf{t}_{\xi}$ is the ${\mathrm{Spin}}^{c}$ structure on $Y$ induced by $\xi$ (see Section 2). A different choice $D^{\prime}$ of the auxiliary data determines $\mathbb{F}[U]$–module automorphisms sending the class $[{\bf{x}}(L,D)]$ to the class $[{\bf{x}}(L,D^{\prime})]$. Formulating this a bit more formally, we can consider the set of pairs $(M,m)$, where $M$ is an $\mathbb{F}[U]$–module and $m\in M$, and introduce an equivalence relation by declaring two pairs $(M,m)$ and $(N,n)$ equivalent if there is an $\mathbb{F}[U]$–module isomorphism $f\colon\thinspace M\to N$ such that $f(m)=n$. Let $[M,m]$ denote the equivalence class of $(M,m)$. Then, our main result can be stated as follows: ###### Theorem 1.1. Let $L$ be an oriented, null–homologous Legendrian knot in the closed contact 3–manifold $(Y,\xi)$, and let $\mathbf{t}_{\xi}$ be the ${\mathrm{Spin}}^{c}$ structure on $Y$ induced by $\xi$. Then, after choosing some suitable auxiliary data, it is possible to associate to $L$ homology classes $\alpha_{{\mathfrak{L}}}(L)\in\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi})$ and $\alpha_{\widehat{\mathfrak{L}}}(L)\in\widehat{\mathrm{HFK}}(-Y,L,\mathbf{t}_{\xi})$ such that ${\mathfrak{L}}(L):=[\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi}),\alpha_{{\mathfrak{L}}}(L)]$ and $\widehat{\mathfrak{L}}(L):=[\widehat{\mathrm{HFK}}(-Y,L,\mathbf{t}_{\xi}),\alpha_{\widehat{\mathfrak{L}}}(L)]$ do not depend on the choice of the auxiliary data, and in fact only depend on the Legendrian isotopy class of $L$. We can define multiplication by $U$ on the set of equivalence classes $[M,m]$ by setting $U\cdot[M,m]:=[M,U\cdot m]$. We will say that ${\mathfrak{L}}(L)$ is _vanishing_ (respectively _nonvanishing_) and write ${\mathfrak{L}}(L)=0$ (respectively ${\mathfrak{L}}(L)\neq 0$) if ${\mathfrak{L}}(L)=[\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi}),0]$ (respectively ${\mathfrak{L}}(L)\neq[\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi}),0]$). Similar conventions will be used for $\widehat{\mathfrak{L}}(L)$. Let $-L$ denote the knot $L$ with reversed orientation. It turns out that the pairs ${\mathfrak{L}}(\pm L)$ and $\widehat{\mathfrak{L}}(\pm L)$ admit properties similar to those of the pair of invariants $\lambda_{\pm}(L)$ of [37]. One useful feature of the invariant ${\mathfrak{L}}(\pm L)$ (shared with $\lambda_{\pm}$ from [37]) is that it satisfies a nonvanishing property, which can be formulated in terms of the contact invariant $c(Y,\xi)$ of the ambient contact 3–manifold $(Y,\xi)$. ###### Theorem 1.2. If the contact Ozsváth–Szabó invariant $c(Y,\xi)\in\widehat{\mathrm{HF}}(-Y,\mathbf{t}_{\xi})$ of the contact 3–manifold $(Y,\xi)$ does not vanish, then for any oriented Legendrian knot $L\subset(Y,\xi)$ we have ${\mathfrak{L}}(L)\neq 0$. If $c(Y,\xi)=0$ then for $d$ large enough $U^{d}\cdot{\mathfrak{L}}(L)$ vanishes. As it will be explained later, such strong nonvanishing property does not hold for $\widehat{\mathfrak{L}}$. Since a strongly symplectically fillable contact 3–manifold has nonzero contact invariant while for an overtwisted structure $c(Y,\xi)=0$, it follows immediately from Theorem 1.2 that ###### Corollary 1.3. If $(Y,\xi)$ is strongly symplectically fillable then for any null–homologous Legendrian knot $L\subset(Y,\xi)$ the invariant ${\mathfrak{L}}(L)$ is nonvanishing. If $(Y,\xi)$ is overtwisted, then for any Legendrian knot $L$ there is $d\geq 0$ such that $U^{d}\cdot{\mathfrak{L}}(L)$ vanishes. ∎ A stronger vanishing theorem holds for loose knots. Recall that a Legendrian knot $L\subset(Y,\xi)$ is _loose_ if its complement contains an overtwisted disk (and hence $(Y,\xi)$ is necessarily overtwisted). A Legendrian knot $L\subset(Y,\xi)$ is _non–loose_ (or _exceptional_ in the terminology of [8]) if $(Y,\xi)$ is overtwisted, but the complement of $L$ is tight. ###### Theorem 1.4. If $L\subset(Y,\xi)$ is an oriented, null–homologous and loose Legendrian knot, then ${\mathfrak{L}}(L)=0$. Transverse knots admit a preferred orientation, and can be approximated, uniquely up to negative stabilization, by oriented Legendrian knots [9, 13]. This fact can be used to define invariants of transverse knots: ###### Theorem 1.5. Suppose that $T$ is a null–homologous transverse knot in the contact 3–manifold $(Y,\xi)$. Let $L$ be a (compatibly oriented) Legendrian approximation of $T$. Then, $\mathfrak{T}(T):={\mathfrak{L}}(L)$ and $\widehat{\mathfrak{T}}(T):=\widehat{\mathfrak{L}}(L)$ are invariants of the transverse knot type of $T$. The proof of this statement relies on the invariance of the Legendrian invariant under negative stabilization. After determining the invariants of stabilized Legendrian unknots in the standard contact 3–sphere and the behaviour of the invariants under connected sum, in fact, we will be able to determine the effect of both kinds of stabilization on the invariant ${\mathfrak{L}}$, leading us to ###### Theorem 1.6. Suppose that $L$ is an oriented Legendrian knot and $L^{+}$, resp. $L^{-}$ denote the oriented positive, resp. negative stabilizations of $L$. Then, ${\mathfrak{L}}(L^{-})={\mathfrak{L}}(L)$ and ${\mathfrak{L}}(L^{+})=U\cdot{\mathfrak{L}}(L)$. The invariants can be effectively used in the study of non–loose Legendrian knots. Notice that the invariant of a non–loose Legendrian knot is necessarily a $U$–torsion element, and can be nonvanishing only if the knot and all its negative stabilizations are non–loose. In Section 6 a family of non–loose torus knots in overtwisted contact $S^{3}$’s are constructed, for which we can determine the invariants ${\mathfrak{L}}$ by direct computation. Recall that a knot type $K$ in a contact three-manifold is said to be transversely non-simple if there are two transversely non-isotopic transverse knots in $Y$ in the topological type of $K$ which have the same self-linking number. (For more background on transverse non-simplicity, see [10, 13].) In an overtwisted contact structure however, this definition admits refinements. (For more on knots in overtwisted contact structures, see [5, 6, 8, 16].) Namely, it is not hard to find examples of pairs of _loose_ and _non–loose_ (Legendrian or transverse) knots with equal ’classical’ invariants. By definition their complements admit different contact structures (one is overtwisted, the other is tight), hence the knots are clearly not Legendrian/transverse isotopic. Examples for this phenomenon will be discussed in Section 8. Finding pairs of Legendrian/transverse knots with equal classical invariants, both loose or both non–loose, is a much more delicate question. Non–loose pairs of Legendrian knots not Legendrian isotopic were found in [12]; by applying our invariants we find transverse knots with similar properties: ###### Theorem 1.7. The knot type $T_{(2,-7)}\\#T_{(2,-9)}\subset S^{3}$ has two non–loose, transversely non–isotopic transverse representatives with the same self–linking number with respect to the overtwisted contact structure $\xi_{12}$ with Hopf invariant $d_{3}(\xi_{12})=12.$ In fact, by further connected sums we get a more general statement: ###### Corollary 1.8. Let $(Y,\xi)$ be a contact 3–manifold with $c(Y,\xi)\neq 0$. Let $\zeta$ be an overtwisted contact structure on $Y$ with $\mathbf{t}_{\zeta}=\mathbf{t}_{\xi}$. Then, in $(Y,\zeta)$ there are null–homologous knot types which admit two non–loose, transversely non–isotopic transverse representatives with the same self–linking number. ###### Remarks 1.9. (a) Notice that to a Legendrian knot $L$ in the standard contact 3–sphere $(S^{3},\xi_{st})$ one can now associate several sets of invariants: $\lambda_{\pm}(L)$ of [37] and ${\mathfrak{L}}(\pm L)$, $\widehat{\mathfrak{L}}(\pm L)$ of the present work. It would be interesting to compare these elements of the knot Floer homology groups. (b) We also note that recent work of Honda-Kazez-Matić [20] provides another invariant for Legendrian knots $L\subset(Y,\xi)$ through the sutured contact invariant of an appropriate complement of $L$ in $Y$, using the sutured Floer homology of Juhász [21]. This invariant seems to have slightly different features than the invariants defined in this paper; the relationships between these invariants have yet to be understood.111Added in proof: the relation between $\widehat{{\mathfrak{L}}}$ and the sutured invariant has been recently worked out in [42]. (c) In [12] arbitrarily many distinct non–loose Legendrian knots with the same classical invariants are constructed in overtwisted contact 3–manifolds. In [14] similar examples were constructed in the standard tight contact $S^{3}$, using connected sums of torus knots. These constructions, however, do not tell us anything about transverse simplicity of the knot types. (d) Note that connected summing preserves transverse simplicity [14] while torus knots are transversely simple with respect to the standard tight contact $S^{3}$ [13, 28]. Therefore, in the standard contact 3–sphere there are no examples such as those of Theorem 1.7. The paper is organized as follows. In Section 2 we recall facts about open books and contact structures, Heegaard Floer groups, and the contact Ozsváth–Szabó invariants. In Section 3 we establish some preliminary results on Legendrian knots and open books, we define our invariants and we prove Theorem 1.1. In Section 4 we compute the invariants of a certain (stabilized) Legendrian unknot in the standard contact $S^{3}$. This calculation is both instructive and useful: it will be used in the proof of the stabilization invariance. In Section 5 we prove Theorems 1.2, 1.4 and 1.5. In Section 6 we determine the invariant $\widehat{\mathfrak{L}}$ for some non–loose Legendrian torus knots in certain overtwisted contact structures on $S^{3}$. In Section 7 we describe the behaviour of the invariants with respect to Legendrian connected sum, and then derive Theorem 1.6. In Section 8 we discuss transverse simplicity in overtwisted contact 3–manifolds, give a refinement of the Legendrian invariant and prove Theorem 1.7 and Corollary 1.8 modulo some technical results which are deferred to the Appendix. Acknowledgements: PSO was supported by NSF grant number DMS-0505811 and FRG-0244663. AS acknowledges support from the Clay Mathematics Institute. AS and PL were partially supported by OTKA T49449 and by Marie Curie TOK project BudAlgGeo. ZSz was supported by NSF grant number DMS-0704053 and FRG-0244663. We would like to thank Tolga Etgü for many useful comments and corrections. ## 2 Preliminaries Our definition of the Legendrian knot invariant relies on the few basic facts listed below. * • There is a one–to–one correspondence between (isotopy classes of) contact structures and open book decompositions (up to positive stabilization), as it was shown by Giroux [2, 11, 18]. * • An idea of Giroux (cf. [2, 11]) allows one to construct an open book decomposition of $Y$ compatible with a contact structure $\xi$, which is also adapted to a given Legendrian knot $L\subset(Y,\xi)$. * • There is a contact invariant $c(Y,\xi)$ of a closed contact 3–manifold $(Y,\xi)$ originally defined in [34] and recently reformulated by Honda–Kazez–Matić [19] (cf. also [20]). We describe these ingredients in more detail in the following subsections. In Subsection 2.1 we recall how contact 3–manifold admit open book decompositions adapted to given Legendrian knots. In Subsection 2.2 (which is not logically required by the rest of this article, but which fits in neatly at this point) we explain how, conversely, an isotopy class of embedded curves in a page of an open book decomposition gives rise to a unique isotopy class of Legendrian knots in the associated contact three–manifold. In Subsection 2.3 we recall the basics of Heegaard Floer homology, mainly to set up notation, and finally in Subsection 2.4 we recall the construction of the contact invariant. ### 2.1 Generalities on open books and contact structures Recall that an open book decomposition of a 3–manifold $Y$ is a pair $(B,\varphi)$ where $B\subset Y$ is a (fibered) link in $Y$ and $\varphi\colon Y-B\to S^{1}$ is a locally trivial fibration such that the closure of each fiber, $S_{t}={\overline{\varphi^{-1}(t)}}$ (a _page_ of the open book), is a Seifert surface for the _binding_ $B$. The fibration $\varphi$ can also be determined by its monodromy $h_{\varphi}\colon S_{+1}\to S_{+1}$, which gives rise to an element of the mapping class group of the page (regarded as a surface with boundary). An open book decomposition $(B,\varphi)$ can be modified by a classical operation called _stabilization_ [41]. The page $S^{\prime}$ of the resulting open book $(B^{\prime},\varphi^{\prime})$ is obtained from the page $S$ of $(B,\varphi)$ by adding a 1–handle $H$, while the monodromy of $(B^{\prime},\varphi^{\prime})$ is obtained by composing the monodromy of $(B,\varphi)$ (extended trivially to $S^{\prime}$) with a Dehn twist $D_{\gamma}$ along a simple closed curve $\gamma\subset S^{\prime}$ intersecting the cocore of $H$ transversely in a single point. Depending on whether the Dehn twist is right– or left–handed, the stabilization is called _positive_ or _negative_. A positive stabilization is also called a _Giroux stabilization_. ###### Definition 2.1. (Giroux) A contact structure $\xi$ and an open book decomposition are _compatible_ if $\xi$, as a cooriented 2–plane field, is the kernel of a contact one–form $\alpha$ with the property that $d\alpha$ is a symplectic form on each page, hence orients both the page and the binding, and with this orientation the binding $B$ is a link positively transverse to $\xi$. In this situation we also say that the contact one–form $\alpha$ is compatible with the open book. A theorem of Thurston and Winkelnkemper [43] can be used to verify that each open book admits a compatible contact structure. Moreover, Giroux [2, 11, 18] proved the following: 1. 1. Each contact structure is compatible with some open book decomposition; 2. 2. Two contact structures compatible with the same open book are isotopic; 3. 3. If a contact structure $\xi$ is compatible with an open book $(B,\varphi)$, then $\xi$ is isotopic to any contact structure compatible with a positive stabilization of $(B,\varphi)$. The construction of an open book compatible with a given contact structure rests on a _contact cell decomposition_ of the contact 3–manifold $(Y,\xi)$, see [11, Section 4] or [2, Subsection 3.4]. In short, consider a $CW$–decomposition of $Y$ such that its 1–skeleton $G$ is a Legendrian graph, each 2–cell $D$ satisfies the property that the twisting of the contact structure along its boundary $\partial D$ (with respect to the framing given by $D$) is $-1$ and the 3–cells are in Darboux balls of $(Y,\xi)$. Then, there is a compact surface $R$ (a ribbon for $G$) such that $R$ retracts onto $G$, $T_{p}R=\xi_{p}$ for all $p\in G$ and $T_{p}R\neq\xi_{p}$ for $p\in R-G$. The 3–manifold $Y$ admits an open book decomposition with binding $\partial R$ and page $R$ which is compatible with $\xi$. We also have the following (see [11, Theorem 4.28] or [2, Theorem 3.4]): ###### Theorem 2.2. (Giroux) Two open books compatible with the same contact structure admit isotopic Giroux stabilizations.∎ The proof of this statement rests on two facts. The first fact is given by ###### Lemma 2.3. ([11, Lemma 4.29] or [2, Proposition 3.7]) Each open book decomposition compatible with $(Y,\xi)$ admits a sequence of Giroux stabilizations such that the resulting open book comes from a contact cell decomposition. ∎ The second fact (see the proof of [11, Theorem 4.28]) is that any two contact cell decompositions can be connected by a sequence of operations of the following types: (1) a subdivision of a 2–cell by a Legendrian arc intersecting the dividing set (for its definition see [13]) of the 2–cell once, (2) an addition of a 1–cell $c^{\prime}$ and a 2–cell $D$ so that $\partial D=c\cup c^{\prime}$, where $c$ is part of the original 1–skeleton and the twisting of the contact structure along $\partial D$ (with respect to $D$) is $-1$, and (3) an addition of a 2–cell $D$ whose boundary is already in the 2–skeleton and satisfies the above twisting requirement. It is not hard to see that operations (1) and (2) induce positive stabilizations on the open book associated to the cell decomposition, while (3) leaves the open book unchanged. Thus, an invariant of open book decompositions which is constant under positive stabilizations is, in fact, an isotopy invariant of the compatible contact structure. Since in the construction of a contact cell decomposition for a contact 3–manifold $(Y,\xi)$ one can choose the $CW$–decomposition of $Y$ in such a way that a Legendrian knot (or link) $L\subset(Y,\xi)$ is contained in its 1–skeleton $G$, we have: ###### Proposition 2.4. ([11, Corollary 4.23]) Given a Legendrian knot $L$ in a closed, contact 3–manifold $(Y,\xi)$, there is an open book decomposition compatible with $\xi$, containing $L$ on a page $S$ and such that the contact framing of $L$ is equal to the framing induced on $L$ by $S$. The open book can be chosen in such a way that $L$ is homologically essential on the page $S$. ∎ Recall (see e.g. [11]) that any open book is obtained via a mapping torus construction from a pair (sometimes called an _abstract open book_) $(S,\varphi)$, where $S$ is an oriented surface with boundary and $\varphi\colon S\to S$ is an orientation–preserving diffeomorphism which restricts as the identity near $\partial S$. Our previous observations amount to saying that any triple $(Y,\xi,L)$, where $L\subset Y$ is a Legendrian knot (or link) is obtained via the standard mapping torus construction from a triple $(S,L,\varphi)$, where $L\subset S$ is a homologically essential simple closed curve. ###### Definition 2.5. Let $(S,\varphi)$ be an abstract open book, and let $L\subset S$ be a homologically essential simple closed curve. We say that the Giroux stabilization $(S^{\prime},R_{\gamma}\circ\varphi)$ is _$L$ –elementary_ if after a suitable isotopy the curve $\gamma$ intersects $L$ transversely in at most one point. The construction of the Legendrian invariant rests on the following: ###### Proposition 2.6. Suppose that $L\subset(Y,\xi)$ is a Legendrian knot in a contact 3–manifold. If the triple $(Y,\xi,L)$ is associated via the mapping torus construction with two different triples ${\cal T}_{i}=(S_{i},L_{i},\varphi_{i})$, $i=1,2$, then $(Y,\xi,L)$ is also associated with a triple $(S,L,\varphi)$, obtained from each of ${\cal T}_{1}$ and ${\cal T}_{2}$ by a finite sequence of $L$–elementary Giroux stabilizations. ###### Proof. The proof of [11, Lemma 4.29] uses $L$–elementary stabilizations only (cf. [11, Figure 12]) and (1)–(2) after Lemma 2.3 determine $L$–elementary stabilizations as well, while (3) leaves the open book decomposition unchanged. ∎ ### 2.2 From curves on a page to Legendrian knots: uniqueness Giroux’s results give a correspondence between Legendrian knots and knots on a page in an open book decomposition. In fact, with a little extra work, this correspondence can be suitably inverted. Although this other direction is not strictly needed for our present applications (and hence, the impatient reader is free to skip the present subsection), it does fit in naturally in the discussion at this point. Specifically, in this subsection, we prove the following: ###### Theorem 2.7. Let $(B,\varphi)$ be an open book decomposition of the closed, oriented 3–manifold $Y$. Let $\xi=\ker(\alpha)$ be a contact structure, with $\alpha$ a contact 1–form compatible with $(B,\varphi)$. Suppose that $K\subset S:=\varphi^{-1}(1)$ is a smooth knot defining a nontrivial homology class in $H_{1}(S;\mathbb{Z})$. Then the smooth isotopy class of $K$ on the page uniquely determines a Legendrian knot in $(Y,\xi)$ up to Legendrian isotopy. The proof of this theorem rests on two technical lemmas. ###### Lemma 2.8. Let $(B,\varphi)$ be an open book decomposition of the closed, oriented 3–manifold $Y$. Let $K_{t}\subset S:=\varphi^{-1}(1)$ be a smooth family of knots which are homologically nontrivial on the page and provide an isotopy from $K_{0}$ to $K_{1}$. Then, there exists a smooth family of contact 1–forms $\alpha_{K_{t}}\in\Omega^{1}(Y)$ compatible with $(B,\varphi)$ and such that the restriction of $\alpha_{K_{t}}$ to $K_{t}$ vanishes. ###### Proof. In view of the argument given in [11, pp. 115–116], it suffices to show that there exists a smooth family $\lambda_{t}\in\Omega^{1}(S)$ such that, for each $t$ 1. 1. $\lambda_{t}=(1+s)d\theta$ near $\partial S$, with coordinates $(s,\theta)\in[0,1]\times S^{1}$ near each boundary component of $S$; 2. 2. $d\lambda_{t}$ is a volume form on $S$; 3. 3. $\lambda_{t}$ vanishes on $K_{t}$. To construct the family $\lambda_{t}$ we proceed as follows. For each $t$, choose a closed collar $U$ around $K_{t}\subset S$, parametrized by coordinates $(s,\theta)\in[-1,1]\times S^{1}$, so that $ds\wedge d\theta$ is a volume form on $U$ with the orientation induced from $S$. Let $\lambda_{1,t}\in\Omega^{1}(S)$ be of the form $(1+s)d\theta$ near $\partial S$ and of the form $sd\theta$ on $U$. We have $\int_{S}d\lambda_{1,t}=\int_{\partial S}\lambda_{1,t}=2\pi|\partial S|.$ Let $\omega$ be a volume form on $S$ such that: * • $\int_{S}\omega=2\pi|\partial S|$; * • $\omega=ds\wedge d\theta$ near $\partial S$ and on $U$. (The first condition can be fulfilled since $K_{t}\subset S$ is nontrivial in homology, hence each component of its complement meets $\partial S$.) Let $U^{\prime}\subset U$ correspond to $[-1/2,1/2]\times S^{1}\subset[-1,1]\times S^{1}$. We have $\int_{\overline{S\setminus U^{\prime}}}\omega-d\lambda_{1,t}=\int_{S}\omega-d\lambda_{1,t}=0.$ Since $\omega-d\lambda_{1,t}=0$ near $\partial(\overline{S\setminus U^{\prime}})$, by de Rham’s theorem there is a compactly supported 1–form $\beta_{t}\in\Omega^{1}({\overline{S\setminus U^{\prime}}})$ such that $d\beta_{t}=\omega-d\lambda_{1,t}$ on ${\overline{S\setminus U^{\prime}}}$. Let $\tilde{\beta}_{t}$ be the extension of $\beta_{t}$ to $S$ by zero. Then, $\lambda_{t}:=\lambda_{1,t}+\tilde{\beta}_{t}\in\Omega^{1}(S)$ satisfies (1), (2) and (3) above, and the dependence on $t$ can be clearly arranged to be smooth. ∎ ###### Lemma 2.9. Let $(B,\varphi)$ be an open book decomposition of the closed, oriented 3–manifold $Y$. Let $\xi=\ker(\alpha)$ be a contact structure, with $\alpha$ a contact 1–form compatible with $(B,\varphi)$. Let $K\subset(Y,\xi)$ be a smooth knot contained in the page $S:=\varphi^{-1}(1)$ with $K$ homologically nontrivial in $S$. Then, the quadruple $(B,\varphi,\alpha,K)$ determines, uniquely up to Legendrian isotopy, a Legendrian knot $L=L(B,\varphi,\alpha,K)\subset(Y,\xi)$ smoothly isotopic to $K$. ###### Proof. Notice first that Giroux’s proof that two contact structures compatible with the same open book are isotopic shows that the space $C\subset\Omega^{1}(Y)$ of contact 1–forms compatible with $(B,\varphi)$ is connected and simply connected. In fact, given $\alpha_{0},\alpha_{1}\in C$, one can first deform each of them inside $C$ to $\alpha_{0,R},\alpha_{1,R}\in C$, where $R\geq 0$ is a constant, so that when $R$ is large enough, the path $(1-s)\alpha_{0,R}+s\alpha_{1,R}$ from $\alpha_{0,R}$ to $\alpha_{1,R}$ is inside $C$ (see [11]). This proves that $C$ is connected. A similar argument shows that $C$ is simply connected. In fact, given a loop $\mathcal{L}=\\{\alpha_{z}\\}\subset C$, $z\in S^{1}$, one can deform it to $\mathcal{L}_{R}=\\{\alpha_{z,R}\\}\subset C$. By compactness, when $R$ is large enough, $\cal L_{R}$ can be shrunk in $C$ onto $\\{\alpha_{1,R}\\}$ by taking convex linear combinations. By our assumptions $\alpha\in C$ and by Lemma 2.8 there is a 1–form $\alpha_{K}\in C$ whose restriction to $K$ vanishes. We can choose a path $P$ in the space $C$ connecting $\alpha_{K}$ to $\alpha$. Then, setting $\xi_{K}:=\ker(\alpha_{K})$, by Gray’s theorem there is a contactomorphism $\Phi=\Phi(\alpha,\alpha_{K},P)\colon\thinspace(Y,\xi_{K})\to(Y,\xi).$ We define $L:=\Phi(K)$. Since $\Phi$ is smoothly isotopic to the identity, $L$ is smoothly isotopic to $K$. Hence, to prove the lemma it suffices to show that changing our choices of $\alpha$, $\alpha_{K}$ or $P$ only changes $L$ by a Legendrian isotopy. Observe that if $P^{\prime}$ is another path from $\alpha_{K}$ to $\alpha$, since $C$ is simply connected there exists a family $P_{t}$ of paths from $\alpha_{K}$ to $\alpha$ connecting $P$ to $P^{\prime}$. This yields a family of contactomorphisms $\Phi_{t}\colon\thinspace(Y,\xi_{K})\to(Y,\xi)$ and therefore a family of Legendrian knots $L_{t}:=\Phi_{t}(K)\subset(Y,\xi)$ with $L_{0}=L$. Thus, $L$ only depends, up to Legendrian isotopy, on the endpoints $\alpha_{K}$ and $\alpha$ of the path $P$. Suppose now that we chose a different 1–form $\alpha^{\prime}_{K}\in C$ whose restriction to $K$ vanishes. Then, we claim that there is a smooth path $\alpha_{K,s}\subset C$ from $\alpha_{K}$ to $\alpha^{\prime}_{K}$ such that the restriction of $\alpha_{K,s}$ to $K$ vanishes for every $s$. In fact, such a path can be found by first deforming each of $\alpha_{K}$ and $\alpha^{\prime}_{K}$ to the forms $\alpha_{K,R}$ and $\alpha^{\prime}_{K,R}$ obtained by adding multiples of the standard angular 1–form suitably modified near the binding, and then taking convex linear combinations (see [11]). Neither of the two operations alters the vanishing property along $K$, therefore this proves the claim. Now we can find a smooth family $Q_{s}$ of paths in $C$, such that $Q_{s}$ joins $\alpha_{K,s}$ to $\alpha$ for every $s$. This produces a family of Legendrian knots $\phi(\alpha,\alpha_{K,s},Q_{s})(K)\subset(Y,\xi)$, thus proving the independence of $L$ from $\alpha_{K}$ up to Legendrian isotopy. The independence from $\alpha$ can be established similarly: if $\alpha^{\prime}\in C$ and $\xi=\ker(\alpha^{\prime})$ then $\alpha_{s}:=(1-s)\alpha+s\alpha^{\prime}\in C$ and $\xi=\ker(\alpha_{s})$ for every $s$. Therefore, we can find a smooth family of paths $R_{s}$ in $C$, with $R_{s}$ joining $\alpha_{K}$ to $\alpha_{s}$ for every $s$, and proceed as before. ∎ ###### Proof of Theorem 2.7. It suffices to show that if $K_{t}\subset S$ is a family of smooth knots which gives a smooth isotopy from $K_{0}$ to $K_{1}$, then, the Legendrian knots $L_{0}=L(B,\varphi,\alpha,K_{0})$ and $L_{1}=L(B,\varphi,\alpha,K_{1})$ determined via Lemma 2.9 are Legendrian isotopic. By Lemma 2.8 there exists a smooth family of contact 1–forms $\alpha_{K_{t}}\in\Omega^{1}(Y)$ compatible with $(B,\varphi)$ and such that the restriction of $\alpha_{K_{t}}$ to $K_{t}$ vanishes. Applying the construction of Lemma 2.9 for each $t$ we obtain the required Legendrian isotopy $L_{t}:=L(B,\varphi,\alpha,K_{t})\subset(Y,\xi).$ ∎ ### 2.3 Heegaard Floer homologies The Heegaard Floer homology groups $\mathrm{HF}^{-}(Y),\widehat{\mathrm{HF}}(Y)$ of a 3–manifold were introduced in [33] and extended in the case where $Y$ is equipped with a null-homologous knot $K\subset Y$ to variants $\mathrm{HFK}^{-}(Y,K),\widehat{\mathrm{HFK}}(Y,K)$ in [35, 40]. For the sake of completeness here we quickly review the construction of these groups, emphasizing the aspects most important for our present purposes. We start with the closed case. An oriented 3–manifold $Y$ can be conveniently presented by a Heegaard diagram, which is an ordered triple $(\Sigma,\mathbb{\alpha},\mathbb{\beta})$, where $\Sigma$ is an oriented genus–$g$ surface, ${\bf{\alpha}}=\\{\alpha_{1},\ldots,\alpha_{g}\\}$ (and similarly ${\bf{\beta}}=\\{\beta_{1},\ldots,\beta_{g}\\}$) is a $g$–tuple of disjoint simple closed curves in $\Sigma$, linearly independent in $H_{1}(\Sigma;{\mathbb{Z}})$. The $\alpha$–curves can be viewed as belt circles of the 1–handles, while the $\beta$–curves as attaching circles of the 2–handles in an appropriate handle decomposition of $Y$. We can assume that the $\alpha$– and $\beta$–curves intersect transversely. Consider the tori ${\mathbb{T}}_{\alpha}=\alpha_{1}\times\ldots\times\alpha_{g}$, ${\mathbb{T}}_{\beta}=\beta_{1}\times\ldots\times\beta_{g}$ in the $g^{th}$ symmetric power $\operatorname{Sym}^{g}(\Sigma)$ of $\Sigma$ and define $\mathrm{CF}^{-}(Y)$ as the free $\mathbb{F}[U]$–module generated by the elements of the transverse intersection ${\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}$. (Recall that in this paper we assume $\mathbb{F}=\mathbb{Z}/2\mathbb{Z}$. The constructions admit a sign refinement to $\mathbb{Z}[U]$, but we do not need this for our current applications.) For appropriate symplectic, and compatible almost complex structures $(\omega,J)$ on $\operatorname{Sym}^{g}(\Sigma)$, ${\bf{x}},{\bf{y}}\in{\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}$, and relative homology class $\phi\in\pi_{2}({\bf{x}},{\bf{y}})$, we define ${\mathfrak{M}}(\phi)$ as the moduli space of holomorphic maps from the unit disk $D\subset{\mathbb{C}}$ to $(\operatorname{Sym}^{g}(\Sigma),J)$ with the appropriate boundary conditions (cf. [33]). Take $\mu(\phi)$ to be the formal dimension of ${\mathfrak{M}}(\phi)$ and ${\widehat{\mathfrak{M}}}(\phi)={\mathfrak{M}}(\phi)/{\mathbb{R}}$ the quotient of the moduli space by the translation action of $\mathbb{R}$. An equivalence class of nowhere zero vector fields (under homotopy away from a ball) on a closed 3–manifold is called a _Spin c structure_. It is easy to see that a cooriented contact structure $\xi$ on a closed 3–manifold naturally induces a ${\mathrm{Spin}}^{c}$ structure: this is the equivalence class of the oriented unit normal vector field of the 2–plane field $\xi$. Fix a point $w\in\Sigma-{\bf{\alpha}}-{\bf{\beta}}$, and for $\phi\in\pi_{2}({\bf{x}},{\bf{y}})$ denote the algebraic intersection number $\\#(\phi\cap\\{w\\}\times\operatorname{Sym}^{g-1}\Sigma)$ by $n_{w}(\phi)$. With these definitions in place, the differential $\partial^{-}\colon\mathrm{CF}^{-}(Y)\to\mathrm{CF}^{-}(Y)$ is defined as $\partial^{-}{\bf{x}}=\sum_{{\bf{y}}\in{\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}}\sum_{\phi\in\pi_{2}({\bf{x}},{\bf{y}}),\mu(\phi)=1}\\#{\widehat{\mathfrak{M}}}(\phi)\cdot U^{n_{w}(\phi)}\cdot{\bf{y}}.$ With the aid of $w$ the elements of ${\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}$ can be partitioned according to the ${\mathrm{Spin}}^{c}$ structures of $Y$, resulting a decomposition $\mathrm{CF}^{-}(Y)=\oplus_{\mathbf{t}\in{\mathrm{Spin}}^{c}(Y)}\mathrm{CF}^{-}(Y,\mathbf{t}),$ and the map $\partial^{-}$ respects this decomposition. If the technical condition of _strong admissibility_ (cf. [33]) is satisfied for the Heegaard diagram $(\Sigma,\alpha,\beta,w)$ of $(Y,\mathbf{t})$, the chain complex $(\mathrm{CF}^{-}(Y,\mathbf{t}),\partial^{-})$ results in a group $\mathrm{HF}^{-}(Y,\mathbf{t})$ which is an invariant of the ${\mathrm{Spin}}^{c}$ 3–manifold $(Y,\mathbf{t})$. Strong admissibility of $(\Sigma,\alpha,\beta,w)$ for $(Y,\mathbf{t})$ can be achieved as follows: consider a collection $\\{\gamma_{1},\ldots,\gamma_{n}\\}$ of curves in $\Sigma$ generating $H_{1}(Y;{\mathbb{Q}})$. It is easy to see that such $\gamma_{i}$ ($i=1,\ldots,n$) can be found for any Heegaard decomposition. Then, by applying sufficiently many times a specific isotopy of the $\beta$–curves along each $\gamma_{i}$ (called ‘spinning’, the exact amount depending on the value of $c_{1}(\mathbf{t})$ on a basis of $H_{2}(Y;{\mathbb{Q}})$, cf. [33]) one can arrange the diagram to be strongly admissible. By specializing the $\mathbb{F}[U]$–module $(\mathrm{CF}^{-}(Y,\mathbf{t}),\partial^{-})$ to $U=0$ we get a new chain complex $(\widehat{\mathrm{CF}}(Y),\widehat{\partial})$, resulting in an invariant $\widehat{\mathrm{HF}}(Y,\mathbf{t})$ of the ${\mathrm{Spin}}^{c}$ 3–manifold $(Y,\mathbf{t})$. In addition, if $c_{1}(\mathbf{t})$ is a torsion class, then the homology groups $\mathrm{HF}^{-}(Y,\mathbf{t})$ and $\widehat{\mathrm{HF}}(Y,\mathbf{t})$ come with a ${\mathbb{Q}}$–grading, cf. [32], and hence split as $\mathrm{HF}^{-}(Y,\mathbf{t})=\oplus_{d}\mathrm{HF}^{-}_{d}(Y,\mathbf{t}),\ \ \ \widehat{\mathrm{HF}}(Y,\mathbf{t})=\oplus_{d}\widehat{\mathrm{HF}}_{d}(Y,\mathbf{t}).$ Since $\widehat{\mathrm{CF}}(Y)$ is generated over $\mathbb{F}$ by the elements of ${\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}$, the Floer homology group $\widehat{\mathrm{HF}}(Y)$ and hence also each $\widehat{\mathrm{HF}}(Y,\mathbf{t})$ are finitely generated $\mathbb{F}$–modules. There is a long exact sequence $\ldots\longrightarrow\mathrm{HF}^{-}_{d}(Y,\mathbf{t})\stackrel{{\scriptstyle\cdot U}}{{\longrightarrow}}\mathrm{HF}^{-}_{d-2}(Y,\mathbf{t})\longrightarrow\widehat{\mathrm{HF}}_{d-2}(Y,\mathbf{t})\longrightarrow\ldots$ which establishes a connection between the two versions of the theory. Suppose now that we fix two distinct points $w,z\in\Sigma-\alpha_{1}-...-\alpha_{g}-\beta_{1}-...-\beta_{g},$ where $(\Sigma,\alpha,\beta,w)$ is a Heegaard diagram for $Y$. The ordered pair of points $(w,z)$ determines an oriented knot $K$ in $Y$ by the following convention. We consider an embedded oriented arc $\zeta$ in $\Sigma$ from $z$ to $w$ in the complement of the $\alpha$–arcs, and let $\eta$ be an analogous arc from $w$ to $z$ in the complement of the $\beta$–arcs. Pushing $\zeta$ and $\eta$ into the $\alpha$– and $\beta$–handlebodies we obtain a pair of oriented arcs $\zeta^{\prime}$ and $\eta^{\prime}$ which meet $\Sigma$ at $w$ and $z$. Their union now is an oriented knot $K\subset Y$. We call the tuple $(\Sigma,\alpha,\beta,w,z)$ a Heegaard diagram compatible with the oriented knot $K\subset Y$. (This is the orientation convention from [35]; it is opposite to the one from [27].) We have a corresponding differential, defined by $\partial^{-}_{K}{\bf{x}}=\sum_{{\bf{y}}\in{\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}}\sum_{\phi\in\pi_{2}({\bf{x}},{\bf{y}}),\mu(\phi)=1,n_{z}(\phi)=0}\\#{\widehat{\mathfrak{M}}}(\phi)\cdot U^{n_{w}(\phi)}\cdot{\bf{y}}.$ Using this map we get a chain complex $(\mathrm{CFK}^{-}(Y),\partial^{-}_{K})$. This group has an additional grading, which can be formulated in terms of relative Spinc structures, which are possible extensions of $\mathbf{t}|_{Y-\nu K}$ to the zero–surgery $Y_{0}(K)$ along the null–homologous knot $K$. Given $K$, there are infinitely many relative ${\mathrm{Spin}}^{c}$ structures with a fixed background ${\mathrm{Spin}}^{c}$ structure on the 3–manifold $Y$. By choosing a Seifert surface $F$ for the null–homologous knot $K$, we can extract a numerical invariant for relative ${\mathrm{Spin}}^{c}$ structures, gotten by half the value of $c_{1}(\mathbf{s})$ on $F$ (which is defined as the integral of the first Chern class of the corresponding ${\mathrm{Spin}}^{c}$ structure of the 0–surgery on the surface ${\hat{F}}$ we get by capping off the surface $F$). Note that the sign of the result depends on the fixed orientation of $K$. The induced $\mathbb{Z}$-grading on the knot Floer complex is called its _Alexander grading_. When $b_{1}(Y)=0$, this integer, together with the background ${\mathrm{Spin}}^{c}$ structure $\mathbf{t}$, uniquely specifies the relative ${\mathrm{Spin}}^{c}$ structure; moreover, the choice of the Seifert surface becomes irrelevant, except for the overall induced orientation on $K$. For a null–homologous knot $K\subset Y$ the homology group $\mathrm{HFK}^{-}(Y,K,\mathbf{s})$ of the above chain complex (with relative ${\mathrm{Spin}}^{c}$ structure $\mathbf{s}$) is an invariant of $(Y,K,\mathbf{s})$, and is called the _knot Floer homology_ of $K$. The specialization $U=0$ of the complex defines again a new complex $(\widehat{\mathrm{CFK}}(Y,K,\mathbf{s}),\widehat{\partial}_{K})$ with homology denoted by $\widehat{\mathrm{HFK}}(Y,K,\mathbf{s})$. The homology groups $\widehat{\mathrm{HFK}}(Y,K,\mathbf{s})$ and $\mathrm{HFK}^{-}(Y,K,\mathbf{s})$ are both finitely generated vector spaces over $\mathbb{F}$. An alternative way to view this construction is the following. Using one basepoint $w$ one can define the chain complex $(\mathrm{CF}^{-}(Y,\mathbf{t}),\partial^{-})$ as before, and with the aid of the other basepoint $z$ one can equip this chain complex with a filtration. As it was shown in [35], the filtered chain homotopy type of the resulting complex is an invariant of the knot, and the Floer homology groups can be defined as the homology of the associated graded object. The restriction of a relative ${\mathrm{Spin}}^{c}$ structure $\mathbf{s}$ to the complement of $K$ extends to a unique ${\mathrm{Spin}}^{c}$ structure $\mathbf{t}$ on $Y$. The map induced by multiplication by $U$ changes the relative ${\mathrm{Spin}}^{c}$ structure, but it preserves the background ${\mathrm{Spin}}^{c}$ structure $\mathbf{t}\in{\mathrm{Spin}}^{c}(Y)$. Thus, we can view $\mathrm{CFK}^{-}(Y,K,\mathbf{t}):=\oplus_{\mathbf{s}\ {\mbox{{\small{restricts to}}}}\ \mathbf{t}}\mathrm{CFK}^{-}(Y,K,\mathbf{s}),$ $\mathrm{HFK}^{-}(Y,K,\mathbf{t}):=\oplus_{\mathbf{s}\ {\mbox{{\small{restricts to}}}}\ \mathbf{t}}\mathrm{HFK}^{-}(Y,K,\mathbf{s})$ and $\widehat{\mathrm{HFK}}(Y,K,\mathbf{t}):=\oplus_{\mathbf{s}\ {\mbox{{\small{restricts to}}}}\ \mathbf{t}}\widehat{\mathrm{HFK}}(Y,K,\mathbf{s})$ as modules over $\mathbb{F}[U]$ (where the $U$–action on $\widehat{\mathrm{HFK}}(Y,K,\mathbf{t})$ is trivial). If $\mathbf{t}$ is torsion, then there is an absolute $\mathbb{Q}$–grading on these modules, as in the case of closed 3–manifolds. As before, the two versions of knot Floer homologies are connected by the long exact sequence $\cdots\longrightarrow\mathrm{HFK}^{-}_{d}(Y,K,\mathbf{t})\stackrel{{\scriptstyle\cdot U}}{{\longrightarrow}}\mathrm{HFK}^{-}_{d-2}(Y,K,\mathbf{t})\longrightarrow\widehat{\mathrm{HFK}}_{d-2}(Y,K,\mathbf{t})\longrightarrow\cdots$ A map $F\colon\mathrm{HFK}^{-}(Y,K,\mathbf{t})\to\widehat{\mathrm{HF}}(Y,\mathbf{t})$ can be defined, which is induced by the map $f\colon\mathrm{CFK}^{-}(Y,K,\mathbf{t})\to\widehat{\mathrm{CF}}(Y,\mathbf{t})$ by setting $U=1$ and taking $z$ to play the role of $w$ in the complex $\widehat{\mathrm{CF}}(Y,\mathbf{t})$. According to the definition, this specialization simply disregards the role of the basepoint $w$. Indeed, the fact that a generator of $\mathrm{CFK}^{-}(Y,K)$ belongs to the summand $\mathrm{CFK}^{-}(Y,K,\mathbf{t})$ is determined only by the point $z$. On the chain level this map fits into the short exact sequence $0\to\mathrm{CFK}^{-}(Y,K,\mathbf{t})\stackrel{{\scriptstyle U-1}}{{\longrightarrow}}\mathrm{CFK}^{-}(Y,K,\mathbf{t})\stackrel{{\scriptstyle f}}{{\longrightarrow}}\widehat{\mathrm{CF}}(Y,\mathbf{t})\to 0,$ (2.1) since $f$ is obviously surjective. ###### Lemma 2.10. Let $F\colon\mathrm{HFK}^{-}(Y,K,\mathbf{t})\longrightarrow\widehat{\mathrm{HF}}(Y,\mathbf{t})$ be map induced on homology by $f$. The kernel of $F$ consists of elements $x$ of the form $(U-1)y$. Moreover, an element $x\in\mathrm{HFK}^{-}(Y,K,\mathbf{t})$ is both in the kernel of $F$ and homogeneous, i.e. contained in the summand determined by a fixed relative ${\mathrm{Spin}}^{c}$ structure, if and only if $x$ satisfies $U^{n}x=0$ for some $n\geq 0$. ###### Proof. The long exact sequence associated to Exact Sequence (2.1) identifies $\ker F$ with $(U-1)\cdot\mathrm{HFK}^{-}(Y,K,\mathbf{t})$. The only statement left to be proved is the characterization of homogeneous elements in $\ker F$. If $U^{n}x=0$ then $x=(1-U^{n})x=(1-U)(x+Ux+\ldots+U^{n-1}x),$ hence $x$ is of the form $x=(1-U)y$. Conversely, if $x=(1-U)y$ we may assume without loss that $y=\sum_{i=0}^{n-1}y_{i}$, where each $y_{i}$ is homogeneous, $y_{0}=x$ and, for $i=0,\ldots,n-1$, $Uy_{i}$ belongs to the same relative ${\mathrm{Spin}}^{c}$ structure as $y_{i+1}$. A simple cancellation argument shows that $y_{1}=Uy_{0},y_{2}=Uy_{1},\ldots$ hence by the finiteness of the sum we get $U^{n}x=0$. ∎ ### 2.4 Contact Ozsváth–Szabó invariants Next we turn to the description of the contact Ozsváth–Szabó invariant of a closed contact 3–manifold as it is given in [19]. (See [34] for the original definition of these invariants.) Suppose that $(B,\varphi)$ is an open book decomposition of the 3–manifold $Y$ compatible with the given contact structure $\xi$. Consider a _basis_ $\\{a_{1},\ldots,a_{n}\\}$ of the page $S_{+1}$, that is, take a collection of disjoint properly embedded arcs $\\{a_{1},\ldots,a_{n}\\}$ such that $S_{+1}-\cup_{i=1}^{n}a_{i}$ is connected and simply–connected (therefore it is homeomorphic to a disk). Let $b_{i}$ be a properly embedded arc obtained by a small isotopy of $a_{i}$ so that the endpoints of $a_{i}$ are isotoped along $\partial S_{+1}$ in the direction given by the boundary orientation, and $a_{i}$ intersects $b_{i}$ in a unique transverse point in int$S_{+1}$, cf. Figure 1 and [19, Figure 2]. 2pt $S$ at 94 94 $a_{i}$ at 165 24 $b_{i}$ at 231 24 Figure 1: The arcs $a_{i}$ and $b_{i}$. Considering ${\overline{a_{i}}}=a_{i}$ and ${\overline{b_{i}}}=h_{\varphi}(b_{i})$ in $S_{-1}={\overline{\varphi^{-1}(-1)}}$ (where $h_{\varphi}$ denotes a diffeomorphism representing the monodromy of the given open book), it is shown in [19] that the triple $(S_{+1}\cup(-S_{-1}),\\{a_{i}\cup{\overline{a_{i}}}\\}_{i=1}^{n},\\{b_{i}\cup{\overline{b_{i}}}\\}_{i=1}^{n})=(S_{+1}\cup(-S_{-1}),\alpha,\beta)$ is a Heegaard diagram for $Y$. (Notice that we have the freedom of choosing $h_{\varphi}$ within its isotopy class; this freedom will be used later, cf. Proposition 2.13.) For technical purposes, however, we consider the triple $(S_{+1}\cup(-S_{-1}),\\{b_{i}\cup{\overline{b_{i}}}\\}_{i=1}^{n},\\{a_{i}\cup{\overline{a_{i}}}\\}_{i=1}^{n})=(S_{+1}\cup(-S_{-1}),\beta,\alpha),$ (2.2) which is now a Heegaard diagram for $-Y$. With this choice, and the careful placement of the basepoint $w$ we can achieve that the proposed chain in the chain complex for defining the Heegaard Floer group is, in fact, a cycle. More formally, put the basepoint $w$ in the disk $S_{+1}-\cup_{i=1}^{n}a_{i}$ outside of the small strips between the $a_{i}$’s and the $b_{i}$’s and consider the element ${\bf{x}}(B,\varphi)=\\{a_{i}\cap b_{i}\\}$ in the chain complex corresponding to the Heegard diagram (2.2) above (pointed by $w$). It is not hard to see (cf. [19]) that with these choices the Heegaard diagram is weakly admissible. In the following we always have to keep in mind the reversal of the $\alpha$– and $\beta$–curves when working with the contact (or Legendrian) invariants. ###### Theorem 2.11 ([19], cf. also [20]). The chain ${\bf{x}}(B,\varphi)$ defined above is closed when regarded as an element of the Heegaard Floer chain complex $\widehat{\mathrm{CF}}(-Y)$. The homology class $[{\bf{x}}(B,\varphi)]\in\widehat{\mathrm{HF}}(-Y)$ defined by ${\bf{x}}(B,\varphi)$ is independent (up to sign) of the chosen basis and compatible open book decomposition. Therefore the homology class $[{\bf{x}}(B,\varphi)]$ is an invariant of the contact structure $(Y,\xi)$. The original definition of this homology class is given in [34], which leads to a different Heegaard diagram. It can be shown that the two invariants are identified after a sequence of handleslides; though one can work directly with the above definition, as in [19]. We adopt this point of view, supplying an alternative proof of the invariance of the contact class (which will assist us in the definition of the Legendrian invariant). ###### Alternative proof of Theorem 2.11.. In computing $\partial{\bf{x}}(B,\varphi)$ we need to encounter holomorphic disks which avoid $w$ but start at ${\bf{x}}(B,\varphi)$. By the chosen order of $a_{i}$ and $b_{i}$ (resulting in a Heegaard diagram of $-Y$ rather than $Y$) we get that such holomorphic disk does not exist. In fact, there are no Whitney disks $\phi\in\pi_{2}({\bf{x}},{\bf{y}})$ for any ${\bf{y}}$ with $n_{w}(\phi)=0$ and whose local multiplicities are all non–negative, cf. [19, Section 3]. Thus, the intersection point represents a cycle in the chain complex. The independence of $[{\bf{x}}(B,\varphi)]$ from the basis is given in [19, Proposition 3.3]. The argument relies on the observation that two bases of $S_{+1}$ can be connected by a sequence of arc slides [19, Section 3.3], inducing handle slides on the corresponding Heegaard diagrams. (We will discuss a sharper version of this argument in Proposition 3.2.) As it is verified in [19, Lemma 3.4], these handle slides map the corresponding ${\bf{x}}$’s into each other. In order to show independence of the chosen open book decomposition, we only need to verify that if $(B^{\prime},\varphi^{\prime})$ is the result of a Giroux stabilization of $(B,\varphi)$, then there are appropriate bases, giving Heegaard decompositions, for which the map $\Phi\colon\widehat{\mathrm{CF}}(B,\varphi)\to\widehat{\mathrm{CF}}(B^{\prime},\varphi^{\prime})$ induced by the stabilization satisfies $\Phi({\bf{x}}(B,\varphi))={\bf{x}}(B^{\prime},\varphi^{\prime})$. Let us assume that the right–handed Dehn twist of the Giroux stabilization is equal to $D_{\gamma}$, where $\gamma$ is a simple closed curve in the page $S^{\prime}$ of $(B^{\prime},\varphi^{\prime})$, and $\gamma_{1}$ is the portion of it inside the page $S$ of $(B,\varphi)$. Our aim is to find a basis $\\{a_{1},\ldots,a_{n}\\}$ for $S$ which is disjoint from $\gamma_{1}$. If $S-\gamma_{1}$ is connected, then choose $a_{1}$ to be (a little push–off of) $\gamma_{1}$, and extend $\\{a_{1}\\}$ to a basis for $S$. If $S-\gamma_{1}$ is disconnected then the union of any bases of the components (possibly after isotoping some endpoints along $\gamma_{1}$) will be appropriate. Now consider the basis for $S^{\prime}$ obtained by extending the above basis for $S$ with the cocore $a_{n+1}$ of the new 1–handle. Since $\gamma$ is disjoint from all $a_{i}$ ($i\leq n$), we clearly get that $\alpha_{n+1}=a_{n+1}\cup{\overline{a_{n+1}}}$ will be intersected only (and in a single point $y_{n+1}$) by $\beta_{n+1}=b_{n+1}\cup{\overline{h_{\varphi}(b_{n+1})}}$. Therefore the map $\Phi$ sending a generator $(y_{1},\ldots,y_{n})$ of $\widehat{\mathrm{CF}}(B,\varphi,\\{a_{1},\ldots,a_{n}\\})$ to $(y_{1},\ldots,y_{n},y_{n+1})$ (with the last coordinate being the unique intersection $\alpha_{n+1}\cap\beta_{n+1}=\\{y_{n+1}\\}$) establishes an isomorphism $\Phi\colon\widehat{\mathrm{CF}}(B,\varphi,\\{a_{1},\ldots,a_{n}\\})\to\widehat{\mathrm{CF}}(B^{\prime},\varphi^{\prime},\\{a_{1},\ldots,a_{n},a_{n+1}\\})$ between the underlying Abelian groups. Since $\alpha_{n+1}$ contains a unique intersection point with all the $\beta$–curves, a holomorphic disk encountered in the boundary map must be constant at $y_{n+1}$, hence $\Phi$ is a chain map. Since it maps ${\bf{x}}(B,\varphi)$ to ${\bf{x}}(B^{\prime},\varphi^{\prime})$, the proof is complete. (See also [20, Section 3].) ∎ ###### Remark 2.12. The basic properties (such as the vanishing for overtwisted and nonvanishing for Stein fillable structures, and the transformation under contact $(+1)$–surgery) can be directly verified for the above construction, cf. [19]. In our later arguments we will need that the Heegaard diagram can be chosen to be strongly admissible, hence we address this issue presently, using an argument which was first used in [38]. ###### Proposition 2.13. For any ${\mathrm{Spin}}^{c}$ structure $\mathbf{t}$ the monodromy map $h_{\varphi}$ of the given open book decomposition can be chosen in its isotopy class in such a way that the Heegaard diagram defined before Theorem 2.11 is strongly admissible for $\mathbf{t}$. ###### Proof. Recall that strong admissibility of a Heegaard diagram for a given ${\mathrm{Spin}}^{c}$ structure $\mathbf{t}$ can be achieved by isotoping the $\beta$–curves through spinning them around a set of curves $\\{\gamma_{1},\ldots,\gamma_{n}\\}$ in the Heegaard surface $\Sigma$ representing a basis of $H_{1}(Y)$. It can be shown that for a Heegaard diagram coming from an open book decomposition, we can choose the curves $\gamma_{i}$ all in the same page, hence all $\gamma_{i}$ can be chosen to be in $S_{-1}$. Since the spinnings are simply isotopies in this page, we can change a fixed monodromy $h_{\varphi}$ within its isotopy class to realize the required spinnings. In this way we get a strongly admissible Heegaard diagram for $(Y,\mathbf{t})$. ∎ ## 3 Invariants of Legendrian knots Suppose now that $L\subset(Y,\xi)$ is a given Legendrian knot, and consider an open book decomposition $(B,\varphi)$ compatible with $\xi$ containing $L$ on a page. To define our Legendrian knot invariants we need to analyze the dependence from the choice of an appropriate basis and from the open book decomposition as in Section 2, but now in the presence of the Legendrian knot. ### Legendrian knots and bases Suppose that $S$ is a surface with $\partial S\neq\emptyset$ and $\\{a_{1},\ldots,a_{n}\\}$ is a basis in $S$. If (after possibly reordering the $a_{i}$’s) the two arcs $a_{1}$ and $a_{2}$ have adjacent endpoints on some component of $\partial S$, that is, there is an arc $\tau\subset\partial S$ with endpoints on $a_{1}$ and $a_{2}$ and otherwise disjoint from all $a_{i}$’s, then define $a_{1}+a_{2}$ as the isotopy class (rel endpoints) of the union $a_{1}\cup\tau\cup a_{2}$. The modification $\\{a_{1},a_{2},\ldots,a_{n}\\}\mapsto\\{a_{1}+a_{2},a_{2},\ldots,a_{n}\\}$ is called an arc slide, cf. [19]. Suppose that $L\subset S$ is a homologically essential simple closed curve. The basis $\\{a_{1},\ldots,a_{n}\\}$ of $S$ is _adapted to $L$_ if $L\cap a_{i}=\emptyset$ for $i\geq 2$ and $L$ intersects $a_{1}$ in a unique transverse point. ###### Lemma 3.1. For any surface $S$ and homologically essential knot $L\subset S$ there is an adapted basis. ###### Proof. The statement follows easily from the fact that $L$ represents a nontrivial class in $H_{1}(S,\partial S)$. ∎ Suppose now that $\\{a_{1},\ldots,a_{n}\\}$ is an adapted basis for $(S,L)$. An arc slide $\\{a_{i},a_{j}\\}\mapsto\\{a_{i}+a_{j},a_{j}\\}$ is called _admissible_ if the arc $a_{i}$ is not slid over the distinguished arc $a_{1}$. The aim of this subsection is to prove the following ###### Proposition 3.2. If $\\{a_{1},\ldots,a_{n}\\}$ and $\\{A_{1},\ldots,A_{n}\\}$ are two adapted bases for $(S,L)$ then there is a sequence of admissible arc slides which trasform $\\{a_{1},\ldots,a_{n}\\}$ into $\\{A_{1},\ldots,A_{n}\\}$. ###### Proof. As a first step in proving the statement on arc slides, we want to show that, up to applying a sequence of admissible arc slides to the $a_{i}$’s, we may assume $(a_{1}\cup\ldots\cup a_{n})\cap(A_{1}\cup\ldots\cup A_{n})=\emptyset$. We start with showing that $a_{1}\cap A_{1}=\emptyset$ can be assumed. Suppose that $a_{1}\cap A_{1}\neq\emptyset$; we will find arc slides reducing $|a_{1}\cap A_{1}|$. To this end, consider the disk $D^{2}$ obtained by cutting $S$ along the $a_{i}$’s. Then, $A_{1}\cap D^{2}$ is a collection of arcs, and (at least) one component intersects $a_{1}$. This component of $A_{1}$ divides $D^{2}$ into two components $D_{1}$ and $D_{2}$, and one of them, say $D_{1}$, contains $a_{1}^{-1}$ (or $a_{1}$). Sliding $a_{1}$ over all the $a_{i}$’s contained in the boundary semicircle of $D_{2}$ (cf. Figure 2(a)) we reduce $|a_{1}\cap A_{1}|$ by one, so ultimately we can assume that $a_{1}\cap A_{1}=\emptyset$. 2pt $a_{1}$ at 9 406 $A_{1}$ at 77 371 $D_{1}$ at 71 313 $D_{2}$ at 94 406 $a_{1}^{-1}$ [lt] at 134 294 $A_{1}$ at 352 343 $L$ at 359 399 $a_{1}$ at 286 408 $a_{1}^{-1}$ at 431 416 $a_{2}$ at 264 354 $a_{1}$ [lb] at 126 196 $a_{1}^{-1}$ [lt] at 130 72 $A_{1}$ at 40 121 $L$ at 90 176 $A_{2}$ at 319 121 $L$ at 360 179 $a_{2}$ [lt] at 391 59 $a_{2}^{-1}$ [b] at 367 211 $a_{1}$ [rb] at 300 180 $a_{1}^{-1}$ [lb] at 422 185 $(a)$ at 79 254 $(b)$ at 359 254 $(c)$ at 79 35 $(d)$ at 363 35 Figure 2: Arc slides. Next we apply further arc slides to achieve $a_{i}\cap A_{1}=\emptyset$ ($i\geq 2$). For this, let us assume that $a_{2}$ is the first arc intersecting $A_{1}$ when traversing $A_{1}$ starting from $\partial S$. (Since $A_{1}$ intersects $L$ exactly once, after possibly starting at the other end of $A_{1}$ we can assume that it first meets $a_{2}$ and then $L$.) As before, we can find a segment of $A_{1}$ intersecting $a_{2}$ and dividing $D^{2}=S-\cup a_{i}$ into two components, one of which contains $a_{1},a_{1}^{-1}$ (since these are connected by $L$, and the segment we chose is disjoint from $L$), cf. Figure 2(b). If $a_{2}^{-1}$ is in the same semicircle as $a_{1}$ (and so $a_{1}^{-1}$) then we can slide $a_{2}$ along the other semicircle to eliminate one intersection point from $a_{2}\cap A_{1}$. If $a_{2}^{-1}$ is on the other semicircle, then we cannot proceed in such a simple way (since $a_{2}$ is not allowed to be slid over neither $a_{1}$ nor $a_{2}^{-1}$). Now consider the continuation of $A_{1}$, which comes out from $a_{2}^{-1}$ and stays in the same component. It might go back to $a_{2}$, and repeat this spiraling some more times, but eventually it will go to another part of the the boundary of $D^{2}$, producing an arc, which starts from $a_{2}$ (or $a_{2}^{-1}$) and divides $D^{2}$ in such a way that $a_{1}$ and $a_{2}^{-1}$ (or $a_{2}$) are on the same side of it. Then, $a_{2}$ can be slid across the opposite side so as to reduce the intersection number in question. After these slides we can assume that $A_{1}\cap(\cup_{i=1}^{n}a_{i})=\emptyset$. This however allows us to slide $a_{1}$ until it becomes isotopic to $A_{1}$, cf. Figure 2(c). Consider now $A_{2}$. If $A_{2}$ intersects some $a_{i}$, then the segment of $A_{2}$ connecting $\partial S$ to the first such intersection (with, say, $a_{2}$) divides $D^{2}$ into two components, and one of them contains both $a_{1}$ and $a_{1}^{-1}$ (since $A_{2}$ is disjoint from $L$). If $a_{2}^{-1}$ is on the same semicircle as $a_{1}$, then sliding over the other semicircle reduces the number of intersections. The other possibility can be handled exactly as before. Finally, we get to the position when $a_{1}$ is isotopic to $A_{1}$ and $(\cup_{i=1}^{n}a_{i})\cap(\cup_{i=1}^{n}A_{i})=\emptyset$. Now we argue as follows: consider $A_{2}$ and choose $a_{i}$ such that $a_{i}$ and $a_{i}^{-1}$ are in different components of $D^{2}-A_{2}$. Such $a_{i}$ exists because $A_{2}$ is nonseparating. Suppose without loss of generality that $i=2$. Then on the side of $A_{2}$ not containing $a_{1}$ (and $a_{1}^{-1}$) we can slide $a_{2}$ until it becomes isotopic to $A_{2}$, see Figure 2(d). Repeating this procedure for each $A_{i}$ the proof is complete. ∎ ### Invariants of Legendrian knots Consider now a null–homologous Legendrian knot $L\subset(Y,\xi)$ and fix an open book decomposition $(B,\varphi)$ compatible with $\xi$ and containing $L$ on the page $S_{+1}:={\overline{\varphi^{-1}(1)}}$. Pick a basis $A=\\{a_{1},\ldots,a_{n}\\}\subset S_{+1}$ such that $a_{1}\cap L$ is a unique point and $a_{i}\cap L=\emptyset$ ($i\geq 2$). Under the above conditions we will say that the triple $(B,\varphi,A)$ is _compatible_ with the triple $(Y,\xi,L)$. Place the basepoint $w$ as before. Putting the other basepoint $z$ between the curves $a_{1}$ and $b_{1}$ we recover a knot in $S_{+1}$ smoothly isotopic to $L$: connect $z$ and $w$ in the complement of $a_{i}$, and then $w$ and $z$ in the complement of $b_{i}$ within the page $S_{+1}$. This procedure (hence the ordered pair $(w,z)$) equips $L$ with an orientation. Moreover, if the point $z$ is moved from one domain between $a_{1}$ and $b_{1}$ to the other, the orientation induced on $L$ gets reversed. Thus, if $L$ is already oriented then there is only one compatible choice of position for $z$. Notice that $z$ and $w$ chosen as above determine a knot in $S_{+1}$, unique up to isotopy in $S_{+1}$. In turn, by Theorem 2.7 the open book decomposition together with such a knot uniquely determines a Legendrian knot (up to Legendrian isotopy) in the corresponding contact structure. In short, $(B,\varphi,S_{+1},A,z,w)$ determines the triple $(Y,\xi,L)$. Recall that when defining the chain complex $\widehat{\mathrm{CF}}$ containing the contact invariant, we reverse the roles of the $\alpha$– and $\beta$–curves, resulting in a Heegaard diagram for $-Y$ rather than for $Y$. For the same reason, we do the switch between the $\alpha$– and $\beta$–curves here as well. According to our conventions, this change would reverse the orientation of the knot $L$ as well; to keep the fixed orientation on $L$, switch the position of the basepoints $w$ and $z$. The two possible locations of $z$ and $w$ we use in the definition of $\widehat{\mathrm{CF}}(-Y,L)$ are illustrated in Figure 3; the orientation of $L$ specifies the location of $w$. 2pt $b_{1}$ [b] at 76 69 $a_{1}$ [b] at 100 69 $b_{1}$ [b] at 298 69 $a_{1}$ [b] at 322 69 $L$ at 36 47 $w$ at 87 38 $w$ at 310 56 $L$ at 358 55 $z$ at 147 32 $z$ at 371 32 or at 198 51 Figure 3: There are two regions between $a_{1}$ and $b_{1}$ in $S_{+1}$; the placement of $w$ is determined by the orientation of $L$. With ${\bf{x}}={\bf{x}}(B,\varphi)=(a_{i}\cap b_{i})$ as before, it is easy to see that there are no nonnegative Whitney disks $\phi\in\pi_{2}({\bf{x}},{\bf{y}})$ with $n_{z}(\phi)=0$ for any ${\bf{y}}$, hence (as in [19, Section 3]) the intersection point ${\bf{x}}$ can be viewed as a cycle in both $\mathrm{CFK}^{-}(-Y,L,\mathbf{t})$ and $\widehat{\mathrm{CFK}}(-Y,L,\mathbf{t})$ for some ${\mathrm{Spin}}^{c}$ structure $\mathbf{t}$ on $Y$. As observed in [35, Subsection 2.3], the ${\mathrm{Spin}}^{c}$ structure $\mathbf{t}$ is determined by the point $z$ only, being equal to $s_{z}({\bf{x}})$ (see [35] for notation). This shows that the contact invariant $c(Y,\xi)$ lives in the summand of $\widehat{\mathrm{HF}}(-Y)$ corresponding to $\mathbf{t}$. On the other hand, it is known [34] that $c(Y,\xi)\in\widehat{\mathrm{HF}}(-Y,\mathbf{t}_{\xi})$, therefore we conclude $\mathbf{t}=\mathbf{t}_{\xi}$. Since $L$ is null–homologous, one needs to make sure that the Heegaard diagram is strongly admissible for $(Y,\mathbf{t}_{\xi})$; this fact follows from Proposition 2.13. Next we will address invariance properties of the knot Floer homology class represented by ${\bf{x}}(B,\varphi)$. ###### Proposition 3.3. Let $L\subset(Y,\xi)$ be a Legendrian knot. Let $(B,\varphi,A,w,z)$ and $(B^{\prime},\varphi^{\prime},A^{\prime},w^{\prime},z^{\prime})$ be two open books compatible with $(Y,\xi,L)$ and endowed with bases and basepoints adapted to $L$. Then, there are isomorphisms of $\mathbb{F}[U]$–modules $\Phi^{-}\colon\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi})\to\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi})$ and $\widehat{\Phi}\colon\widehat{\mathrm{HFK}}(-Y,L,\mathbf{t}_{\xi})\to\widehat{\mathrm{HFK}}(-Y,L,\mathbf{t}_{\xi})$ such that $\Phi^{-}([{\bf{x}}(B,\varphi,A)])=[{\bf{x}}(B^{\prime},\varphi^{\prime},A^{\prime})]$ and $\widehat{\Phi}([{\bf{x}}(B,\varphi,A)])=[{\bf{x}}(B^{\prime},\varphi^{\prime},A^{\prime})].$ Recall from Section 1 that, if $M$ and $N$ are $\mathbb{F}[U]$–modules, we consider $(M,m)$ and $(N,n)$ to be equivalent, and we write $[M,m]=[N,n]$, if there is an isomorphism of $\mathbb{F}[U]$–modules $f\colon\thinspace M\to N$ such that $f(m)=n$. In view of Proposition 3.3, we introduce the following ###### Definition 3.4. Let $L\subset(Y,\xi)$ be an oriented Legendrian knot, $(B,\varphi,A,w,z)$ an open book decomposition compatible with $(Y,\xi,L)$ with an adapted basis and basepoints. Then, we define ${\mathfrak{L}}(L):=\left[\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi}),[{\bf{x}}(B,\varphi,A)]\right].$ Similarly, we define $\widehat{\mathfrak{L}}(L):=\left[\widehat{\mathrm{HFK}}(-Y,L,\mathbf{t}_{\xi}),[{\bf{x}}(B,\varphi,A)]\right].$ ###### Proof of Proposition 3.3. Suppose for the moment that $(B^{\prime},\varphi^{\prime})=(B,\varphi)$. Then, Proposition 3.2 provides a sequence of arc slides transforming $A^{\prime}$ to the chosen $A$. As it is explained in [19], arc slides induce handle slides on the associated Heegaard diagrams, and the invariance of the Floer homology element under these handle slides is verified in [19, Lemma 3.4]. Notice that since we do not slide over the arc intersecting the Legendrian knot $L$ (and hence over the second basepoint defined by this arc), actually all the handle slides induce $\mathbb{F}[U]$–module isomorphisms on the knot Floer groups, cf. [35]. This argument proves the statement in the special case $(B^{\prime},\varphi^{\prime})=(B,\varphi)$. We now consider the special case when $(B^{\prime},\varphi^{\prime})$ is an $L$–elementary stabilization of $(B,\varphi)$. Suppose that the Dehn twist of the stabilization is along $\gamma\subset S^{\prime}$, with $\gamma_{1}\subset S$. Since we work with an $L$–elementary stabilization, $|\gamma_{1}\cap L|\leq 1$. If $\gamma_{1}$ is separating then $\gamma_{1}\cap L=\emptyset$ and we choose the bases of $S$ and $S^{\prime}$ as in the proof of Theorem 2.11. If $\gamma_{1}$ is nonseparating and $\gamma_{1}\cap L=\emptyset$ then choose $a_{2}=\gamma_{1}$ and extend it to an appropriate basis. For $\gamma_{1}\cap L=\\{{\mbox{pt.}}\\}$ we take $a_{1}=\gamma_{1}$ and extend it further. Denote by $T$ and $T^{\prime}$ the resulting bases of $S$ and $S^{\prime}$ respectively. The proof of Theorem 2.11 now applies verbatim to show the existence of automorphisms of $\mathrm{HFK}^{-}(-Y,L,\mathbf{t}_{\xi})$ and $\widehat{\mathrm{HFK}}(-Y,L,\xi)$ mapping $[{\bf{x}}(B,\varphi,T)]$ to $[{\bf{x}}(B^{\prime},\varphi^{\prime},T^{\prime})]$. On the other hand, by the first part of the proof we know that there are other automorphisms sending $[{\bf{x}}(B,\varphi,A)]$ to $[{\bf{x}}(B,\varphi,T)]$ and $[{\bf{x}}(B^{\prime},\varphi^{\prime},T^{\prime})]$ to $[{\bf{x}}(B^{\prime},\varphi^{\prime},A^{\prime})]$. This proves the statement when $(B,\varphi^{\prime})$ is an $L$–elementary stabilization of $(B,\varphi)$. In the general case, since $(B,\varphi)$ and $(B^{\prime},\varphi^{\prime})$ are two open books compatible with $(Y,\xi,L)$, by Proposition 2.6 we know that there is a sequence of $L$–elementary stabilizations which turns each of them into the same stabilization $(B^{\prime\prime},\varphi^{\prime\prime})$. Thus, applying the previous special case the required number of times, the proof is complete. ∎ ###### Remark 3.5. Let $L\subset(Y,\xi)$ be a Legendrian knot and $(B,\varphi,A,w,z)$ a compatible open book with adapted basis and basepoints. Then, it follows immediately from the definitions that the map $f$ from $\mathrm{HFK}^{-}(-Y,L,\mathbf{s})$ to $\widehat{\mathrm{HFK}}(-Y,L,\mathbf{s})$ induced by setting $U=0$ sends the class of ${\bf{x}}(B,\varphi,A)$ in the first group to the class of ${\bf{x}}(B,\varphi,A)$ in the second group. Moreover, the chain map inducing $f$ can be viewed as the canonical map from the complex $\mathrm{CFK}^{-}(-Y,L,\mathbf{s})$ onto its quotient complex $\widehat{\mathrm{CFK}}(-Y,L,\mathbf{s})$. As such, it is natural with respect to the transformations of the two complexes induced by the isotopies, stabilizations and arc slides used in the proof of Proposition 3.3. Thus, it makes sense to write $f({\mathfrak{L}}(L))=\widehat{\mathfrak{L}}(L)$. Therefore $\widehat{\mathfrak{L}}(L)\neq 0$ readily implies ${\mathfrak{L}}(L)\neq 0$, although the converse does not necessarily hold: a nonvanishing invariant ${\mathfrak{L}}(L)$ determined by a class which is in the image of the $U$–map gives rise to vanishing $\widehat{\mathfrak{L}}(L)$. As we will see, such examples do exist. ###### Corollary 3.6. Let $L_{1},L_{2}\subset(Y,\xi)$ be oriented Legendrian knots. Suppose that there exists an isotopy of oriented Legendrian knots from $L_{1}$ to $L_{2}$. Then, ${\mathfrak{L}}(L_{1})={\mathfrak{L}}(L_{2})$ and $\widehat{\mathfrak{L}}(L_{1})=\widehat{\mathfrak{L}}(L_{2})$. ###### Proof. Let $(B,\varphi,A)$ be an open book compatible with $(Y,\xi,L_{1})$ with an adapted basis, and let $f_{1}$ be the time–1 map of the isotopy. Then, the triple $(f_{1}(B),\varphi\circ f_{1}^{-1},f_{1}(A))$ is compatible with and adapted to $(Y,\xi,L_{2})$. The induced map on the chain complexes maps ${\bf{x}}(B,\varphi,A)$ to ${\bf{x}}(f_{1}(B),\varphi\circ f_{1}^{-1},f_{1}(A))$, verifying the last statement. ∎ ###### Remark 3.7. In fact, we only used the fact that $f_{1}\colon(Y,\xi)\to(Y,\xi)$ is a contactomorphism mapping $L_{1}$ into $L_{2}$ (respecting their orientation). In conclusion, Legendrian knots admitting such an identification have the same Legendrian invariants. The existence of $f_{1}$ with these properties and the isotopy of the two knots is equivalent in the standard contact 3–sphere, but the two conditions are different in general. ###### Proof of Theorem 1.1. The theorem follows immediately from Proposition 3.3 and Corollary 3.6. ∎ ## 4 An example Suppose that $L\subset(S^{3},\xi_{st})$ is the Legendrian unknot with Thurston–Bennequin invariant $-1$ in the standard tight contact 3–sphere. It is easy to see that the positive Hopf link defines an open book on $S^{3}$ which is compatible with $\xi_{st}$ and it contains $L$ on a page. A basis in this case consists of a single arc cutting the annulus. The corresponding genus–1 Heegaard diagram has now a single intersection point, which gives the generator of $\mathrm{HFK}^{-}(-S^{3},L)$ (the 3–sphere $S^{3}$ has a unique ${\mathrm{Spin}}^{c}$ structure, so we omit it from the notation). The two possible choices $w_{1}$ and $w_{2}$ for the position of $w$, corresponding to the two choices of an orientation for $L$, give the same class defining ${\mathfrak{L}}(L)$, because in this case $w_{1}$ and $w_{2}$ are in the same domain, cf. Figure 4$(i)$. Let $L^{\prime}$ denote the stabilization of $L$. The knot $L^{\prime}$ then can be put on the page of the once stabilized open book, depicted (together with the monodromies) by Figure 4$(ii)$, where the unknot is represented by the curve with $L^{\prime}$ next to it, and the thin circles represent curves along which Dehn twists are to be performed to get the monodromy map. 2pt $L$ [lb] at 12 85 $L^{\prime}$ [lb] at 244 113 $b$ [rt] at 96 106 $a$ [lt] at 63 106 $w_{1}$ [B] at 79 116 $w_{2}$ [B] at 79 146 $z$ [lb] at 126 113 $b_{1}$ [t] at 326 88 $a_{1}$ [t] at 300 88 $z$ at 358 110 $b_{2}$ [t] at 329 165 $a_{2}$ [t] at 295 164 $w_{1}$ [B] at 311 188 $w_{2}$ [B] at 309 212 $(i)$ at 79 0 $(ii)$ at 311 0 Figure 4: The page of the open book and the basis for the stabilized unknot. The page on the left represents the open book decomposition for the Hopf band, while the page on the right is its stabilization. The corresponding Heegaard diagram (with the use of the adapted basis of Figure 4$(ii)$) is shown in Figure 5. 2pt $P$ [t] at 98 104 $w_{1}$ [B] at 66 127 $V$ [ll] at 225 139 $T$ [ll] at 261 142 $X_{1}$ [tl] at 334 99 $X_{2}$ [B] at 353 178 $z$ at 262 199 $w_{2}$ at 140 155 Figure 5: Heegaard diagram for the unknot. We record both possible choices of $w$ by putting a $w_{1}$ and a $w_{2}$ on the diagram — the two choices correspond to the two possible orientations of $L$. Incidentally, these two choices also correspond to the two possible stabilizations with respect to a given orientation, since positive stabilization for an orientation is exactly the negative stabilization for the reversed orientation. It still needs to be determined whether $w_{2}$ gives a positive or negative stabilization. ###### Lemma 4.1. The oriented knot determined by the pair $z$ and $w_{2}$ represents the negative stabilization of the oriented unknot (i.e. the stabilization with $\operatorname{rot}=-1$). ###### Proof. In order to compute the rotation number of the stabilization given by $w_{2}$, first we construct a Seifert surface for $L^{\prime}$. To this end, consider the loop $A_{1}$ given by the upper part of $L^{\prime}$ together with the dashed line $C_{1}$ of Figure 6. 2pt $L^{\prime}$ at 33 149 $R$ at 30 104 $C_{1}$ at 138 133 $C_{2}$ at 138 75 Figure 6: Computation of the rotation of $L^{\prime}$. This loop bounds a disk $D_{1}$ in the 3–manifold given by the open book decomposition, and the tangent vector field along $A_{1}$ obviously extends as a nonzero section of $\xi$ to $D_{1}$, since $D_{1}$ can be regarded as an appropriate Seifert surface for the unknot $L$ before stabilization (cf. Figure 4). Define $A_{2}$ and $D_{2}$ similarly, now using the lower part of $L^{\prime}$. A Seifert surface for $L^{\prime}$ then can be given by the union of $D_{1},D_{2}$ and the region $R$ of Figure 6. Extending the tangent vector field along $L^{\prime}$ to a section of $\xi$ over $C_{1},C_{2}$ first, the above observation shows that the rotation number of $L^{\prime}$ is the same as the obstruction to extending the above vector field to $R$ as a section of $\xi$. Notice that along $R$ we have that $\xi=TR$. The region $R$ (with the given vector field on its boundary) can be embedded into the disk with the tangent vector field along its boundary, hence a simple Euler characteristic computation shows that the obstruction we need to determine is equal to $-1$, concluding the proof. ∎ Notice that the monodromy is pictured on the page $S_{+1}$ (which also contains the knot), but its effect is taken into account on $S_{-1}$, which comes in the Heegaard surface with its orientation reversed. Therefore when determining the $\alpha$– and $\beta$–curves in the Heegaard diagrams, right–handed Dehn twists of the monodromy induce left–handed Dehn twists on the diagram and vice versa. The three generators of the chain complex corresponding to the Heegaard diagram of Figure 5 in the second symmetric product of the genus–2 surface are the pairs $PX_{1},PX_{2}$ and $TV$. It is easy to see that there are holomorphic disks connecting $TV$ and $PX_{2}$ (passing through the basepoint $w_{1}$) and $TV$ and $PX_{1}$ (passing through $w_{2}$); and these are the only two possible holomorphic disks not containing $z$. When we use $w_{1}$ as our second basepoint, the two disks out of $TV$ show that the class represented by $PX_{1}$, and the class of $PX_{2}$ when muliplied by $U$ are homologous: $\partial^{-}(TV)=PX_{1}+U\cdot PX_{2}.$ In conclusion, in this case $[PX_{2}]$ generates $\mathrm{HFK}^{-}(-S^{3},L^{\prime})$ and the invariant $[PX_{1}]$ is determined by the class of $U$–times the generator. When using $w_{2}$ as the second basepoint, we see that the class $[PX_{2}]$ will be equal to $U\cdot[PX_{1}]$, hence in this case $[PX_{1}]$ is the generator. Since $PX_{1}$ represents the Legendrian invariant, we conclude that in this case the equivalence class modulo automorphisms of the generator of $\mathrm{HFK}^{-}(-S^{3},L^{\prime})$ (over $\mathbb{F}[U]$) is equal to the Legendrian invariant of $L^{\prime}$. In summary, we get ###### Corollary 4.2. Suppose that $L\subset(S^{3},\xi)$ is an oriented Legendrian unknot with $\operatorname{tb}(L)=-1$. Then ${\mathfrak{L}}(L)$ is represented by the generator of the $\mathbb{F}[U]$–module $\mathrm{HFK}^{-}(L)=\mathbb{F}[U]$. If $L^{-}$ is the negative stabilization of $L$ then ${\mathfrak{L}}(L^{-})={\mathfrak{L}}(L)$, while for the positive stabilization $L^{+}$ we have ${\mathfrak{L}}(L^{+})=U\cdot{\mathfrak{L}}(L)$. ∎ ## 5 Basic properties of the invariants ### Nonvanishing and vanishing results The invariant ${\mathfrak{L}}$ admits a nonvanishing property provided the contact invariant $c(Y,\xi)$ of the ambient 3–manifold is nonzero (which holds, for example, when the ambient contact structure is strongly fillable). When $c(Y,\xi)=0$ (e.g., if $(Y,\xi)$ is overtwisted) then ${\mathfrak{L}}(L)$ is a $U$–torsion element. ###### Proof of Theorem 1.2. Consider the natural chain map $f\colon\mathrm{CFK}^{-}(-Y,L)\to\widehat{\mathrm{CF}}(-Y)$ given by setting $U=1$, cf. Lemma 2.10. Let $(B,\varphi,A,w,z)$ be an open book compatible with $(Y,\xi,L)$ with an adapted basis and basepoints. Since the map $F$ induced by $f$ on the homologies sends $[{\bf{x}}(B,\varphi,A)]$ to $c(Y,\xi)$, the nonvanishing of ${\mathfrak{L}}(L)$ when $c(Y,\xi)\neq 0$ obviously follows. If the above map sends $[{\bf{x}}(B,\varphi,A)]$ to zero, then by Lemma 2.10 (and the fact that $[{\bf{x}}(B,\varphi,A)]$ is homogeneous) we get that $U^{d}\cdot{\mathfrak{L}}(L)=0$ for some $d\geq 0$, verifying Theorem 1.2. ∎ A vanishing theorem can be proved for a _loose_ knot, that is, for a Legendrian knot in a contact 3–manifolds with overtwisted complement. Before this result we need a preparatory lemma from contact topology: ###### Lemma 5.1. Suppose that $L\subset(Y,\xi)$ is a Legendrian knot such that $(Y,\xi)$ contains an overtwisted disk in the complement of $L$. Then, the complement $(Y-\nu L,\xi|_{Y-\nu L})$ admits a connected sum decomposition $(Y-\nu L,\xi_{1})\\#(S^{3},\xi_{2})$ with the property that $\xi_{1}$ coincides with $\xi|_{Y-\nu L}$ near $\partial(Y-\nu L)$ and $\xi_{2}$ is overtwisted. ###### Proof. Let us fix an overtwisted disk $D$ disjoint from the knot $L$ and consider a neighbourhood $V$ (diffeomorphic to $D^{3}$) of $D$, with the property that $V$ is still disjoint from $L$. By the classification of overtwisted contact structures on $D^{3}$ with a fixed characteristic foliation on the boundary [7, Theorem 3.1.1], we can take $\xi_{1}$ on $Y-\nu L$ and $\xi_{2}$ on $S^{3}$ such that $\xi_{1}\\#\xi_{2}=\xi|_{Y-\nu L}$ and $\xi_{1}$ is equal to $\xi$ near $\partial(Y-\nu L)$, while $\xi_{2}$ is overtwisted. The statement of the lemma then follows at once. ∎ ###### Proof of Theorem 1.4. Let us fix a decomposition of $(Y,\xi)$ as before, that is, $(Y,\xi)=(Y,\xi_{1})\\#(S^{3},\xi_{2})$ with the properties that $L\subset(Y,\xi_{1})$ and $\xi_{2}$ is overtwisted on $S^{3}$. Consider open book decompositions compatible with $(Y,\xi_{1})$ and $(S^{3},\xi_{2})$. Assume furthermore that the first open book has a basis adapted to $L$, while the second open book has a basis containing an arc which is displaced to the left by the monodromy. (The existence of such a basis is shown in the proof of [19, Lemma 3.2].) The Murasugi sum of the two open books and the union of the two bases provides an open book decomposition for $(Y,\xi)$, adapted to the knot $L$, together with an arc disjoint from $L$ which is displaced to the left by the monodromy. Since the basepoint $w$ in the Heegaard diagram is in the strip determined by the arc intersecting the knot, the holomorphic disk appearing in the proof of [19, Lemma 3.2] avoids both basepoints and shows the vanishing of ${\mathfrak{L}}(L)$. ∎ ### Transverse knots Next we turn to the verification of the formula relating the invariants of a negatively stabilized Legendrian knot to the invariants of the original knot. We will spell out the details for ${\mathfrak{L}}$ only. Then we will discuss the implication of the stabilization result regarding invariants of transverse knots. The effect of more general stabilizations on the invariant will be addressed later using slightly more complicated techniques. ###### Proposition 5.2. Suppose that $L$ is an oriented Legendrian knot and $L^{-}$ denotes the oriented negative stabilization of $L$. Then, ${\mathfrak{L}}(L^{-})={\mathfrak{L}}(L)$. ###### Proof. The proof relies on the choice of a convenient open book decomposition. To this end, fix an open book decomposition $(B,\varphi,A)$ compatible with $(Y,\xi)$ and with a basis adapted to $L$. Place $w$ according to the given orientation. As shown in [11], after an appropriate Giroux stabilization the open book also accomodates the stabilization of $L$. The new open book with adapted basis $(B^{-},\varphi^{-},A^{-})$ together with the new choice of $w$ (denoted by $w^{-}$) is illustrated by Figure 7. 2pt $a$ [t] at 84 22 $b$ [t] at 119 24 $a$ [t] at 328 22 $b$ [t] at 363 24 $w$ [b] at 346 48 $w$ at 101 56 $z$ at 150 47 $z$ at 404 49 $w^{-}$ [b] at 348 113 $a^{\prime}$ [b] at 361 127 $b^{\prime}$ [b] at 329 127 $c$ [tl] at 372 71 Figure 7: Change of the open book after negative stabilization of $L$. The monodromy changes by a right–handed Dehn twist along the curve $c$. As in Lemma 4.1, we can easily see that this choice provides the negative stabilization $L^{-}$. (Recall that the stabilization changed the monodromy of the open book by multiplying it with the right–handed Dehn twist $D_{c}$.) In the new page the stabilization of $L$ is determined up to isotopy simply by changing the basepoint from $w$ to $w^{-}$. (Notice that by placing $w^{-}$ in the other domain in the strip between $a^{\prime}$ and $b^{\prime}$ the orientation of the stabilized knot would be incorrect.) Now the corresponding portion of the Heegaard diagram has the form shown by Figure 8. 2pt $a$ [t] at 174 22 $b$ [t] at 217 24 $w$ at 194 72 $a^{\prime}$ [tl] at 219 197 $w^{-}$ at 194 182 $b^{\prime}$ [tr] at 173 197 $C_{1}$ [l] at 219 105 $C_{2}$ [l] at 202 255 $\beta$ [l] at 251 280 $z$ at 302 116 Figure 8: Stabilization in the Heegaard diagram. The top and bottom boundary components of the surface and the circles $C_{1}$ and $C_{2}$ should be thought of as identified via a reflection across the middle (dotted) line of the picture. In the picture, the top and bottom boundary components of the surface and the circles $C_{1}$ and $C_{2}$ should be thought of as identified via a reflection across the middle (dotted) line of the picture. Moreover, the curve $\beta$ is only partially represented, due to the action of the monodromy. As before, in the diagram $w$ gives rise to $L$ while $w^{-}$ to $L^{-}$ (together with the common $z$). It is straightforward from the picture that $w$ and $w^{-}$ are in the same domain, hence the statement follows. ∎ It follows from Proposition 5.2 that the invariant ${\mathfrak{L}}$ of a Legendrian approximation provides an invariant for a transverse knot. ###### Proof of Theorem 1.5. Fix a transverse knot $T$ and consider a Legendrian approximation $L$ of $T$. By [9, 13], up to negative stabilizations the Legendrian knot $L$ only depends on the transverse isotopy class of $T$. Therefore by Proposition 5.2 the equivalence classes $\mathfrak{T}(T)={\mathfrak{L}}(L)$ and $\widehat{\mathfrak{T}}(T)=\widehat{{\mathfrak{L}}}(L)$ are invariants of the transverse isotopy class of the knot $T$, and hence the theorem follows. ∎ ## 6 Non–loose torus knots in $S^{3}$ In this section we describe some examples where the invariants defined in the paper are explicitly determined. Some interesting consequences of these computations will be drawn in the next section. For the sake of simplicity, we will work with the invariant $\widehat{\mathfrak{L}}$. ### Positive Legendrian torus knots $T_{(2,2n+1)}$ in overtwisted contact $S^{3}$’s Let us consider the Legendrian knot $L(n)$ given by the surgery diagram of Figure 9. 2pt $L(n)$ [l] at 453 55 $+1$ [l] at 458 82 $+1$ [l] at 454 107 $-1$ [l] at 453 131 $-1$ [l] at 514 154 $-1$ [l] at 513 201 $n$ [r] at 55 177 Figure 9: Non–loose Legendrian torus knot $T_{(2,2n+1)}$ in $S^{3}$ ($n\geq 1$). The meaning of the picture is that we perform contact $(\pm 1)$–surgeries along the given Legendrian knots, the result being a contact 3–manifold containing the unframed knot $L(n)$ as a Legendrian knot. (For contact $(\pm 1)$–surgeries and surgery presentations see [3, 30].) ###### Lemma 6.1. The contact structure $\xi_{n}$ defined by the surgery diagram of Figure 9 is the overtwisted contact structure on $S^{3}$ with Hopf invariant $d_{3}(\xi_{n})=1-2n$. The Legendrian knot $L(n)$ is smoothly isotopic to the torus knot $T_{2,2n+1}$ and is non–loose. ###### Proof. Figure 10 gives a smooth surgery diagram corresponding to Figure 9. 2pt $-2$ at 28 52 $-1$ at 139 53 $-3$ [l] at 75 55 $-4$ [lt] at 51 80 $-4$ [rb] at 35 96 $0$ [r] at 203 53 $0$ [r] at 232 53 $L(n)$ [l] at 266 56 $(n)$ at 37 81 Figure 10: A smooth version of Figure 9. The knot type of $L(n)$ and the underlying 3–manifold can be easily identified. The Kirby calculus moves of Figure 11 show that Figure 10 is equivalent to the left–hand side of Figure 12. 2pt $-1$ at 97 214 $-2$ at 20 213 $-4$ at 44 233 $-4$ at 22 252 $-3$ at 58 214 $0$ at 139 214 $0$ at 159 214 $(n)$ at 27 234 $L(n)$ at 190 239 $+1$ at 397 178 $L(n)$ at 434 225 $(n)$ at 263 236 $+1$ at 392 243 $+1$ at 379 229 $-1$ at 284 235 $-1$ at 261 255 $+1$ at 236 232 $+1$ at 234 216 $-2$ at 308 231 $-1$ at 367 27 $L(n)$ at 428 83 $+1$ at 235 85 $+1$ at 231 66 $(n)$ at 263 89 $-1$ at 288 91 $-1$ at 267 111 $-2$ at 371 88 $-2$ at 39 71 $-3$ at 82 68 $-2$ at 83 84 $-2$ at 81 150 $n-1$ at 152 114 $L(n)$ at 168 70 $-1$ at 97 14 Figure 11: Kirby moves on Figure 10. 2pt $-1$ [b] at 76 6 $-2$ [rb] at 29 61 $-3$ [r] at 70 61 $-2$ [r] at 70 77 $-2$ [r] at 70 141 $n-1$ [l] at 125 107 $L(n)$ [b] at 139 61 $n$ at 285 55 $L(n)$ [b] at 286 102 Figure 12: $L(n)$ is the torus knot $T_{2,2n+1}$. Applying a number of “blow–downs” yields the right–hand side of Figure 12, veryfing that $L(n)$ is isotopic to the positive torus knot $T_{(2,2n+1)}$ in $S^{3}$. Figures 9 and 10 can be used to determine the signature $\sigma(X)$ and the Euler characteristic $\chi(X)$ of the 4–manifold obtained by viewing the integral surgeries as 4–dimensional 2–handle attachments to $S^{3}\times[0,1]$. (As it is customary in Heegaard Floer theory, the 4-manifold $X$ denotes the cobordism between $S^{3}$ and the 3–manifold we get after performing the prescribed surgeries.) In addition, the rotation numbers define a second cohomology class $c\in H^{2}(X;{\mathbb{Z}})$, and a simple computation shows $\sigma(X)=-n-1,\ \ \chi(X)=n+3,\ \ c^{2}=-9n-1.$ Since we apply two $(+1)$–surgeries, the formula $d_{3}(\xi)=\frac{1}{4}(c^{2}-3\sigma(X)-2\chi(X))+q$ (with $q$ denoting the number of contact $(+1)$–surgeries) computes $d_{3}(\xi_{n})$ of the contact structure, providing $1-2n<0$ for all $n\in{\mathbb{N}}$. Since the unique tight contact structure $\xi_{st}$ on $S^{3}$ has vanishing Hopf invariant $d_{3}(\xi_{st})$, we get that $\xi_{n}$ is overtwisted. Applying contact $(-1)$–surgery along $L(n)$ we get a tight contact structure, since this $(-1)$–surgery cancels one of the $(+1)$–surgeries, and a single contact $(+1)$–surgery along the Legendrian unknot provides the Stein fillable contact structure on $S^{1}\times S^{2}$. Therefore there is no overtwisted disk in the complement of $L(n)$ (since such a disk would persist after the surgery), consequently $L(n)$ is non–loose. ∎ As it is explained in [25, Section 6], the Legendrian link underlying the surgery diagram for $\xi_{n}$ (together with the Legendrian knot $L(n)$) can be put on a page of an open book decomposition with planar pages, which is compatible with the standard contact structure $\xi_{st}$ on $S^{3}$. This can be seen by considering the annular open book decomposition containing the Legendrian unknot (and its Legendrian push–offs), and then applying the stabilization method described in [11] for the stabilized knots. The monodromy of this open book decomposition can be computed from the Dehn twists resulting from the stabilizations, together with the Dehn twists (right–handed for $(-1)$ and left–handed for $(+1)$) defined by the surgery curves. Notice that one of the left–handed Dehn twists is cancelled by the monodromy of the annular open book decomposition we started our procedure with. This procedure results in the monodromies given by the curves of Figure 13. We perform right–handed Dehn twists along solid curves and a left–handed Dehn twist along the dashed one. 2pt $L(n)$ [l] at 187 159 $(n)$ at 95 243 Figure 13: Monodromy of the open book decomposition compatible with $L(n)$. Solid curves represent right–handed Dehn twists, while the dashed one (parallel to $L(n)$) represents a left–handed Dehn twist. The application of the lantern relation simplifies the monodromy factorization to the one shown in Figure 14. 2pt $L(n)$ at 149 137 $(n)$ [b] at 30 118 Figure 14: Simplified monodromy. Notice that in the monodromy factorization given by Figure 13 there are Dehn twists along intersecting curves, hence these elements of the mapping class group do not commute. Therefore, strictly speaking, an order of the Dehn twists should be specified. Observe, however, that although the elements do not commute, the fact that there is only one such pair of intersecting curves implies that the two possible products are conjugate, and therefore give the same open book decomposition, allowing us to suppress the specification of the order. Figure 15 helps to visualize the curves on ‘half’ of the Heegaard surface, and also indicates the chosen basis. 2pt $L(n)$ at 58 77 $a_{4}$ at 75 90 $a_{3}$ at 174 90 $a_{2}$ at 270 90 $a_{1}$ at 370 90 $(n)$ at 408 102 Figure 15: Another view of the monodromy. The diagram also indicates the chosen basis. (Notice that here $a_{4}$ intersects $L(n)$ in a unique point and not $a_{1}$.) The open book decomposition found above equips $S^{3}$ with a Heegaard decomposition compatible with $L(n)$. The $\alpha$– and $\beta$–curves of this decomposition are given in Figure 16. 0pt $L(n)$ at 79 39 $\alpha_{4}$ at 227 58 $\alpha_{3}$ at 218 193 $\alpha_{2}$ at 222 236 $\alpha_{1}$ at 223 372 $\beta_{1}$ at 109 406 $\beta_{2}$ at 90 299 $\beta_{3}$ at 93 202 $\beta_{4}$ at 88 100 $n$ at 161 434 $w$ at 128 18 $z$ at 50 211 $L$ at 141 116 $M$ at 155 98 $B_{3}$ at 205 211 $X_{1}$ at 137 200 $C_{2}$ at 157 199 $X_{2}$ at 194 199 $Y_{2}$ at 167 218 $B_{4}$ at 149 218 $Y_{1}$ at 121 217 $B_{1}$ at 86 257 $P$ at 166 303 $B_{2}$ at 218 277 $Q$ at 124 311 $A_{1}$ at 88 381 $A_{2}$ at 133 392 $A_{n}$ at 176 405 $A_{n+1}$ [l] at 218 329 $C_{1}$ at 87 167 $D=D_{1}$ [r] at 91 68 Figure 16: The Heegaard decomposition compatible with $L(n)$. Recall that we get the $b_{i}$ arcs by the usual perturbation of the $a_{i}$’s and the action of the monodromy yields a Heegaard decomposition for $S^{3}$ with the distinguished point ${\bf{x}}$ in ${\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}$ determining the Legendrian invariant $\widehat{\mathfrak{L}}(L(n))$. A warning about orientations is in place. We illustrated the monodromy curves on the page $S_{+1}$ containing the Legendrian knot $L(n)$, but their action must be taken into account on the page $S_{-1}$, which comes in the Heegaard surface with its opposite orientation; hence right–handed Dehn twists take curves to the left on the Heegaard surface and vice versa. The basepoint $z$ is placed in the ‘large’ region of the page $S_{+1}$, while the point $w$ is in the strip between $a_{1}$ and $b_{1}$ shown in Figure 16. This choice determines the orientation on $L(n)$. Recall that since $L(n)$ is isotopic to the positive torus knot $T_{(2,2n+1)}$, the Legendrian invariant is given by an element of $\widehat{\mathrm{HFK}}(-S^{3},T_{(2,2n+1)})$, which is isomorphic to $\widehat{\mathrm{HFK}}(S^{3},T_{(2,-(2n+1))})$. This latter group is determined readily from the Alexander polynomial and the signature of the knot (since it is alternating), cf. [36, 39]. In fact, we have $\widehat{\mathrm{HFK}}_{n+s}(S^{3},T_{(2,-(2n+1))},s)\cong\left\\{\begin{array}[]{ll}\mathbb{F}&{\text{if $|s|\leq n$}}\\\ 0&{\text{otherwise.}}\end{array}\right.$ (6.1) After these preparations we are ready to determine the invariants of the Legendrian knots discussed above. ###### Proposition 6.2. The homology class $\widehat{\mathfrak{L}}(L(n))$ is determined by the unique nontrivial homology class in Alexander grading $1-n$ in $\widehat{\mathrm{HFK}}(S^{3},T_{(2,-2n-1)})$. ###### Proof. We claim that in the Heegaard diagram of Figure 16 the point $A_{1}B_{1}C_{1}D_{1}$ (representing the Legendrian invariant $\widehat{\mathfrak{L}}(L(n))$) is the only intersection point in Alexander grading $1-n$. We sketch the argument establishing this. We can orient every $\alpha$– and $\beta$–curve in the diagram so that their intersection matrix (whose $(i,j)^{th}$ entry is the algebraic intersection of $\alpha_{i}$ with $\beta_{j}$) is $M=\left(\begin{array}[]{rrrr}n+1&-1&0&0\\\ -1&4&-2&0\\\ 0&-2&2&-1\\\ 0&0&-1&1\end{array}\right).$ Note that for this particular diagram, the absolute values of the algebraic intersection number and the geometric intersection numbers coincide. A simple calculation shows that there are $16n+19$ intersecton points in ${\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}$. To calculate the relative Alexander gradings of intersection points, it is convenient to organize them into types: specifically, the intersection points of $\mathbb{T}_{\alpha}$ and $\mathbb{T}_{\beta}$ correspond to permutations $\sigma$ and then $4$-tuples in $\prod\alpha_{i}\cap\beta_{\sigma(i)}$, which in terms of the notation of Figure 16 are given by quadruples of letters; specifically, intersection points all have one of the following five types $\begin{array}[]{lllll}ABCD&PQCD&AXYD&ABLM&PQLM.\end{array}$ We begin by calculating relative Alexander gradings of intersection points of the same type. Consider first the relative gradings between points of the form $A_{i}\mbox{$\star\star\star$}$ and $A_{i+1}\mbox{$\star\star\star$}$, where now $\star\star\star$ is a fixed triple of intersection points. Start from an initial path $\delta$ which travels from $A_{i}$ to $A_{i+1}$ along $\alpha_{1}$, then back from $A_{i+1}$ to $A_{i}$ along $\beta_{1}$. We add to this cycle copies of $\alpha_{i}$ and $\beta_{j}$ to obtain a null-homologous cycle. This can be done since the $\alpha_{i}$’s and $\beta_{j}$’s span $H_{1}(\Sigma)$, which follows from the fact that the ambient 3–manifold is $S^{3}$. Concretely, we need to solve the expression $\sum n_{i}[\alpha_{i}]+\sum m_{j}[\beta_{j}]+[\delta]=0$ for $n_{i}$ and $m_{j}$. In fact, all we are concerned about is the difference in local multiplicities at $z$ and $w$ in the null-homology of the above expression. Since $z$ and $w$ are separated by $\alpha_{4}$, and our initial curve $\delta$ is supported on $\alpha_{1}$ and $\beta_{1}$, this difference is given by the multiplicity $n_{4}$ of $\alpha_{4}$ in the above expression. In view of the fact that $\\#\beta_{i}\cap\beta_{j}=0$, $n_{4}$ can be obtained as the last coefficient of $M^{-1}v$, where $v$ is the vector whose $i^{th}$ coordinate is $\\#\delta\cap\beta_{j}$ and $M$ is the incidence matrix. We can find paths of the above type $\delta$ connecting $A_{i}$ and $A_{i+1}$ whose intersection numbers with the $\beta_{j}$’s are given by the vector $(1,0,0,0)$, hence giving that the Alexander grading of $A_{i}\mbox{$\star\star\star$}$ is two smaller than the Alexander grading of $A_{i+1}\mbox{$\star\star\star$}$. We express this by saying that the relative gradings of the $A_{i}$ are given by $2i$. Repeating this procedure for the other curves, we find that relative gradings of $B_{1},...,B_{4}$ (completed by a fixed triple to quadruples of intersection points) are given by $0$, $2n$, $4n+2$, and $2n+1$ respectively; for $C_{1}$ and $C_{2}$ the relative gradings are $0$ and $2n+1$ respectively, for $X_{1}$ and $X_{2}$ they are $0$ and $2n+1$, and they are also $0$ and $2n+1$ for $Y_{1}$ and $Y_{2}$. Putting these together, we can calculate the relative Alexander gradings of any two intersection points of the same type. To calculate the relative Alexander gradings of points of different types, we use three rectangles. Specifically, intersection points of the form $PQ\mbox{$\star\star$}$ and $A_{n+1}B_{1}\mbox{$\star\star$}$ (where here $\star\star$ denotes some fixed pair of intersection points) have equal Alexander gradings, shown by the rectangle with vertices $PB_{1}QA_{n+1}$, avoiding both $w$ and $z$. Similar argument relates points of type $X_{1}Y_{1}\mbox{$\star\star$}$ to the ones of type $B_{4}C_{1}\mbox{$\star\star$}$. The rectangle $LDMC_{1}$ contains $w$ once, showing that the Alexander gradings of elements of the form $LM\mbox{$\star\star$}$ are one less than the Alexander gradings of elements of the form $C_{1}D\mbox{$\star\star$}$. This allows us to calculate the relative Alexander gradings of any two intersection points. Given this information, it is now straightforward to see that there are no other intersection points in the same Alexander grading as $A_{1}B_{1}C_{1}D_{1}$, and hence that it represents a homologically nontrivial cycle in $\widehat{\mathrm{HFK}}(S^{3},T_{(2,-(2n+1))})$. Thus, we have shown that for all $n\in{\mathbb{N}}$, the class $\widehat{\mathfrak{L}}(L(n))$ is a nontrivial generator in $\widehat{\mathrm{HFK}}(S^{3},T_{(2,-(2n+1))})$. In fact, the absolute Alexander grading of generators is pinned down by the following symmetry property: the Alexander grading is normalized so that the parity of the number of points of Alexander grading $i$ coincides with the parity of the number of points of Alexander grading $-i$. Using this property, one finds that $A_{1}B_{1}C_{1}D_{1}$ is supported in Alexander grading $1-n$. ∎ ###### Remark 6.3. Recall (cf. [36, 39]) that $\mathrm{HFK}^{-}(S^{3},T_{(2,-(2n+1))})\cong\mathbb{F}^{n}\oplus\mathbb{F}[U],$ (6.2) where the top generator of the free $\mathbb{F}[U]$–module is at $(A=n,M=2n)$ while the $n$ generators of the $\mathbb{F}^{n}$ summand are of bi–degrees $(A=n-1-2i,\quad M=2n-1-2i),\quad i=0,\ldots,n-1.$ It is not hard to show that the above computation implies that ${\mathfrak{L}}(L(n))$ is determined by the unique nonzero $U$–torsion element of $\mathrm{HFK}^{-}(S^{3},T_{(2,-(2n+1))})$ in Alexander grading $1-n$. ### Negative Legendrian torus knots $T_{(2,-(2n-1))}$ in overtwisted contact $S^{3}$’s For $k,l\geq 0$ let us consider the knot $L_{k,l}$ in the contact 3–manifold $(Y_{k,l},\xi_{k,l})$ given by the surgery presentation of Figure 17. Let $n=k+l+2$. 2pt $L_{k,l}$ [l] at 453 55 $+1$ [l] at 458 82 $+1$ [l] at 454 107 $-1$ [l] at 453 131 $-1$ [l] at 514 154 $-1$ [l] at 513 201 $l$ at 469 391 $k$ at 46 343 Figure 17: Non–loose Legendrian $T_{(2,-(2n-1))}$ torus knot in $S^{3}$. ###### Proposition 6.4. The contact 3–manifold specified by the surgery diagram of Figure 17 is $(S^{3},\xi_{k,l})$ with $d_{3}(\xi_{k,l})=2l+2$, hence $\xi_{k,l}$ is overtwisted. The Legendrian knot $L_{k,l}$ as a smooth knot is isotopic to the negative $(2,2n-1)$ torus knot $T_{(2,-(2n-1))}$, and it is non–loose in $(S^{3},\xi_{k,l})$. ###### Proof. The simple Kirby calculus argument illustrated in Figure 18 2pt $L_{k,l}$ at 187 255 $-1$ at 106 227 $0$ at 142 226 $0$ at 161 226 $-3$ at 15 200 $-3$ at 38 209 $-n$ at 29 232 $-n$ at 230 258 $-2$ at 265 253 $L_{k,l}$ at 399 249 $-2$ at 295 214 $1$ at 323 196 $1$ at 361 258 $1$ at 343 253 $-n$ at 38 158 $-2$ at 86 167 $-1$ at 115 107 $L_{k,l}$ at 194 157 $-2$ at 133 168 $-n$ at 257 150 $-1$ at 294 108 $1$ at 328 125 $L_{k,l}$ at 381 147 $-1$ at 375 110 $-n$ at 84 68 $0$ at 63 25 $L_{k,l}$ at 182 68 $T_{2,-(2n-1)}\subset S^{3}$ (after two slides) [l] at 233 39 Figure 18: Kirby moves on the diagrams of Figure 17. shows that the 3–manifold $Y_{k,l}$ is $S^{3}$, while the formula $d_{3}(\xi_{k,l})=\frac{1}{4}(c^{2}-3\sigma-2b_{2})+q$ of [4] computes the Hopf invariant of $\xi_{k,l}$. As always, $\sigma$ and $b_{2}$ denote the signature and the second Betti number of the 4–manifold $X$ specified by the underlying smooth surgery diagram, while $c\in H^{2}(X;{\mathbb{Z}})$ is specified by the rotation numbers of the contact surgery curves, and $q$ is the number of $(+1)$–surgeries. Simple algebra verifies that $d_{3}(\xi_{k,l})=2l+2\geq 2,$ and since the unique tight structure on $S^{3}$ has vanishing Hopf invariant, we get that $\xi_{k,l}$ is overtwisted. Following the Kirby moves of Figure 18 with the knot $L_{k,l}$ we arrive to the last surgery picture of Figure 18, and by sliding $L_{k,l}$ twice over the $(-n)$–framed unknot and cancelling the pair we see that $L_{k,l}$ is, in fact, isotopic to the negative $(2,2n-1)$ torus knot $T_{(2,-(2n-1))}\subset S^{3}$. If contact $(-1)$–surgery on $L_{k,l}$ provides a tight contact structure, then $L_{k,l}$ is obviously non–loose, since any overtwisted disk in its complement would give an overtwisted disk in the surgered manifold. In our case, however, contact $(-1)$–surgery simply cancels one of the $(+1)$–surgeries defining $\xi_{k,l}$, and since a single contact $(+1)$–surgery on the Legendrian unknot provides the contact boundary of the Stein 1–handle (cf. [23]), we get that $(-1)$–surgery along $L$ provides a Stein fillable contact structure. ∎ ###### Remark 6.5. In fact, by stabilizing $L_{k,l}$ on the _left_ and then performing a contact $(-1)$–surgery we still get a tight contact 3–manifold: it will be Stein fillable if we perform only one stabilization, and not Stein fillable but tight for more stabilizations. The tightness of the result of these latter surgeries were verified in [17] by computing the contact Ozsváth–Szabó invariants of the resulting contact structures. Notice that this observation implies that after arbitrary many left stabilizations $L_{k,l}$ remains non–loose, which, in view of Proposition 5.2, is a necessary condition for ${\mathfrak{L}}(L_{k,l})$ to be nonvanishing. (Finally note that performing contact $(-1)$–surgery on $L_{k,l}$ after a single _right_ stabilization provides an overtwisted contact structure (cf. now [17, Section 5])). In contrast, for the non–loose knots $L(n)$ of the previous subsection (also having nontrivial ${\mathfrak{L}}$–invariants) the same intuitive argument does not work, since some negative surgery on the knot $L(n)$ will produce a contact structure on the 3–manifold $S^{3}_{2n-1}(T_{2,2n+1})$ and since this 3–manifold does not admit any tight contact structure [24], the result of the surgery will be overtwisted independently on the chosen stabilizations. Nevertheless, the overtwisted disk in the 3–manifold obtained by only negative stabilizations cannot be in the complement of the knot $L(n)$, since such stabilizations are still non–loose (shown by the nonvanishing of the invariant $\widehat{{\mathfrak{L}}}(L(n))$). Next we want to determine the classical invariants of $L_{k,l}$. Considering the problem slightly more generally, and let $L\subset Y$ be a null–homologous Legendrian knot in a contact 3–manifold $(Y,\xi)$. Assume furthermore that $Y$ is a _rational homology 3–sphere_. The knot $L$ has two “classical invariants”: the Thurston–Bennequin and the rotation numbers $\operatorname{tb}(L),\operatorname{rot}(L)$. Recall that the Thurston–Bennequin invariant of a Legendrian knot is defined as the framing induced by the contact 2–plane field distribution on the knot, hence, given an orientation for the knot, it naturally gives rise to a homology class $\operatorname{TB}\in H_{1}(Y-L;\mathbb{Z})$ (supported in a tubular neighborhood of $L$). Fixing a Seifert surface $F$ for $L$, the oriented intersection number of $\operatorname{TB}$ with $F$ provides a numerical invariant, called the _Thurston–Bennequin number_ $\operatorname{tb}(L)\in\mathbb{Z}$. This intersection number is independent of the choice of the Seifert surface; moreover, the Thurston–Bennequin number is independent of the orientation of $L$ since the orientation of $\operatorname{TB}$ and the orientation of $F$ both depend on this choice. Analogously, the rotation $\operatorname{rot}$ of $L$ is the relative Euler class of $\xi$ when restricted to $Y-L$, with the trivialization given by the tangents of $L$. Again, a Seifert surface for $L$ can be used to turn this class uniquely into an integer, also denoted by $\operatorname{rot}(L)$. Note that unlike the Thurston–Bennequin number, the sign of the rotation number does depend on the orientation of $L$. Suppose that ${\mathbb{S}}={\mathbb{S}}_{+}\cup{\mathbb{S}}_{-}\subset(S^{3},\xi_{st})$ is a contact $(\pm 1)$–surgery presentation of the contact 3–manifold $(Y,\xi)$ and $L$ is a Legendrian knot in $(S^{3},\xi_{st})$ disjoint from ${\mathbb{S}}$, null–homologous in $Y$. The Thurston–Bennequin and rotation numbers of $L$ in $(Y,\xi)$ can be obtained from the Thurston–Bennequin and rotation numbers of the individual components of ${\mathbb{S}}$ and $L$ through the following data. Let $\operatorname{tb}_{0}$ denote the Thurston–Bennequin number of $L$ as a knot in the standard contact 3–sphere (which, in terms of a front projection, is equal to the writhe minus half the number of cusps). Writing ${\mathbb{S}}=\cup_{i=1}^{n}L_{i}$, let $a_{i}$ be the integral surgery coefficient on the link component $L_{i}$; i.e. $a_{i}=\operatorname{tb}(L_{i})\pm 1$ if $L_{i}\in{\mathbb{S}}_{\pm}$. Define the linking matrix $M(a_{0},a_{1},...,a_{n})=\left(m_{i,j}\right)_{i,j=0}^{n}$ where $m_{i,j}:=\left\\{\begin{array}[]{ll}a_{i}&{\text{if $i=j$}}\\\ {\mathrm{lk}}(L_{i},L_{j})&{\text{if $i\neq j$,}}\end{array}\right.$ with the convention that $L=L_{0}$ and ${\mathbb{S}}=\cup_{i=1}^{n}L_{i}$. Similarly, let $M(a_{1},...,a_{n})$ denote the matrix $\left(m_{i,j}\right)_{i,j=1}^{n}$. Consider the integral rotation numbers $r_{0},...,r_{n}$ obtained from our Legendrian knot $L$ and Legendrian presentation as a link in $S^{3}$. ###### Lemma 6.6. Suppose that ${\mathbb{S}}={\mathbb{S}}_{+}\cup{\mathbb{S}}_{-}\subset(S^{3},\xi_{st})$ is a contact $(\pm 1)$–surgery presentation of the contact 3–manifold $(Y,\xi)$ and $L$ is a Legendrian knot in $(S^{3},\xi_{st})$ disjoint from ${\mathbb{S}}$, null–homologous in $Y$. Then the Thurston-Bennequin and rotation numbers $\operatorname{tb}(L)$ and $\operatorname{rot}(L)$ can be extracted from the above data by the formulae: $\operatorname{tb}(L)=\operatorname{tb}_{0}+\frac{\det(M(0,a_{1},...,a_{n}))}{\det(M(a_{1},...,a_{n}))}$ (6.3) and $\operatorname{rot}(L)=r_{0}-\langle\left(\begin{array}[]{c}r_{1}\\\ \vdots\\\ r_{n}\end{array}\right),M^{-1}\cdot\left(\begin{array}[]{c}\operatorname{lk}(L_{0},L_{1})\\\ \vdots\\\ \operatorname{lk}(L_{0},L_{n}),\end{array}\right)\rangle,$ (6.4) where $M=M(a_{1},...,a_{n})$. ###### Proof. We turn to the verification of Equation (6.4), after a few preliminary observations. Let $\mu_{i}\subset S^{3}-(L\cup{\mathbb{S}})$ be a meridian for $L_{i}\subset L\cup{\mathbb{S}}$, and $\lambda_{i}$ be its corresponding longitude. Recall that $H_{1}(S^{3}-(L\cup{\mathbb{S}}))$ is a free $\mathbb{Z}$-module, generated by the meridians $\mu_{i}$ (we continue with the convention that the $i=0$ component of the link $L\cup{\mathbb{S}}$ is $L$). We can express the homology class of $\lambda_{i}$ in terms of the other meridians by the expression $\lambda_{i}=\sum_{j\neq i}\operatorname{lk}(L_{i},L_{j})\cdot\mu_{j}.$ The homology groups of the surgered manifold are gotten from the homology groups of the link complement by dividing out by the relations $a_{i}\mu_{i}+\lambda_{i}=0$ for $i=1,...,n$; more precisely, $H_{1}(S^{3}-(L\cup{\mathbb{S}}))$ is freely generated by $\mu_{0},...,\mu_{n}$, while $H_{1}(Y-L)\cong\mathbb{Z}$ is obtained from this free group by dividing out the $n$ relations $a_{i}\cdot\mu_{i}+\sum_{j\neq i}\operatorname{lk}(L_{i},L_{j})\cdot\mu_{j}=0,\quad i=1,\ldots,n.$ The rotation numbers $r_{i}$ can be thought of as follows. Let $e(\xi,L\cup{\mathbb{S}})\in H^{2}(S^{3},L\cup{\mathbb{S}})$ denote the relative Euler class of $\xi$ relative to the trivialization it inherits along $L\cup{\mathbb{S}}$. Then, the rotation numbers are the coefficients in the expansion of the Poincaré dual $\operatorname{PD}[e(\xi,L\cup{\mathbb{S}})]\in H_{1}(S^{3}-(L\cup{\mathbb{S}}))$ in terms of the basis of meridians $\operatorname{PD}[e(\xi,L\cup{\mathbb{S}})]=\sum_{i=0}^{n}r_{i}\cdot\mu_{i}$. Similarly, $H_{1}(Y-L)\cong\mathbb{Z}$ is generated by $\mu_{0}$, the meridian of $L$, and the rotation number $\operatorname{rot}(L)$ is calculated by $\operatorname{PD}(e(\xi,L))=\operatorname{rot}(L)\cdot\mu_{0}$. Note also that $\operatorname{PD}[e(\xi,L)]$ is the image of $\operatorname{PD}[e(\xi,L\cup{\mathbb{S}})]$ under the inclusion $j\colon S^{3}-(L\cup{\mathbb{S}})\subset Y-L$. Thus, to find $\operatorname{rot}(L)$, it suffices to express $j_{*}(\mu_{i})$ in terms of $\mu_{0}$. Write $\displaystyle\Lambda=\left(\begin{array}[]{c}\operatorname{lk}(L_{0},L_{1})\\\ \vdots\\\ \operatorname{lk}(L_{0},L_{n}),\end{array}\right)$ and $\displaystyle R=\left(\begin{array}[]{c}r_{1}\\\ \vdots\\\ r_{n}\end{array}\right).$ In view of our presentation for $H_{1}(S^{3}-L)$, we see that for all $i>0$ we have $\mu_{i}=c_{i}\cdot\mu_{0}$, where $c_{i}$ is the $i^{th}$ entry in the vector $-M^{-1}\cdot\Lambda$. It follows that $\sum_{i=0}^{n}r_{i}\cdot\mu_{i}=(r_{0}-\langle R,M^{-1}\cdot\Lambda\rangle)\cdot\mu_{0},$ estabilishing Equation (6.4). We turn now to Equation (6.3). Choose $a_{0}$ so that the curve $\lambda^{\prime}=a_{0}\cdot\mu_{0}+\lambda_{0}$ is null–homologous in $Y-L$. Letting $\operatorname{TB}$ denote the Thurston–Bennequin framing curve of $L=L_{0}$, we have that $\operatorname{TB}=\operatorname{tb}_{0}\cdot\mu_{0}+\lambda_{0}=\operatorname{tb}(L)\cdot\mu_{0}+\lambda^{\prime};$ thus $\operatorname{tb}(L)-\operatorname{tb}_{0}=-a_{0}$. In view of our presentation for this first homology group, we see that $a_{0}$ is determined by the condition that $\displaystyle 0$ $\displaystyle=$ $\displaystyle\left|\begin{array}[]{llll}a_{0}&\operatorname{lk}(L_{0},L_{1})&...&\operatorname{lk}(L_{0},L_{n})\\\ \operatorname{lk}(L_{0},L_{1})&a_{1}&...&\operatorname{lk}(L_{0},L_{1})\\\ \vdots&\vdots&\ddots&\vdots\\\ \operatorname{lk}(L_{n},L_{0})&\operatorname{lk}(L_{n},L1)&\ldots&a_{n}\end{array}\right|$ $\displaystyle=$ $\displaystyle a_{0}\cdot\det(M(a_{1},...,a_{n}))+\det(M(0,a_{1},...,a_{n})).$ Equation (6.3) follows. ∎ ###### Lemma 6.7. With the orientation given by Figure 17, the Thurston–Bennequin and rotation numbers of the knot $L_{k,l}$ are given by $\operatorname{tb}(L_{k,l})=-4(k+l)-6\quad{\mbox{and}}\quad\operatorname{rot}(L_{k,l})=-6l-2k-7.$ ###### Proof. Applying Formulae (6.3) and (6.4) to the surgery diagrams defining $L_{k,l}$, simple algebra gives the statement of the lemma. ∎ As it is explained in the previous subsection (resting on observations from [25, Section 6]), the surgery diagram for $\xi_{k,l}$ (together with the Legendrian knot $L_{k,l}$) can be put on a page of an open book decomposition with planar pages, which is compatible with the standard contact structure $\xi_{st}$ on $S^{3}$. The diagram for all choices of $k$ and $l$ is less apparent, hence we restrict our attention first to $k=0$. In this case we get the monodromies defined by the curves of Figure 19. 2pt $L_{0,l}$ at 195 158 $l$ at 97 104 Figure 19: Monodromy of the open book decomposition compatible with $L_{0,l}$. The application of the lantern relation simplifies the monodromy factorization to the one illustrated in Figure 20. 2pt $L_{0,l}$ at 155 140 $l$ at 389 155 Figure 20: The simplified monodromy of the open book for $L_{0,l}$. Figure 21 helps in visualizing the curves on ‘half’ of the Heegaard surface. The diagram also indicates the chosen basis. 2pt $L_{0,l}$ at 47 70 $l+1$ at 349 154 Figure 21: Another view of the open book for $L_{0,l}$. Notice that in the monodromy illustrated in Figure 19 there are Dehn twists with intersecting curves, hence these elements of the mapping class group do not commute. As before, the two products are conjugate, hence there is no need to record the order. In a similar spirit the general diagrams could be deduced — since we will not use them in our computations, we will be content with working on a further special case when $k=1$ and $l\geq 1$; the result is given by Figure 22. 2pt $L(1,l)$ [lb] at 110 152 $l-1$ at 378 136 Figure 22: The monodromy factorization for $L(1,l)$ with $l\geq 1$. We just note here that in these cases we need to apply the lantern relation twice to get commuting Dehn twists as illustrated in Figure 22. For this reason, these monodromy factorizations contain two left–handed Dehn twists. In Figure 23 the monodromy for $L(1,l)$ with $l\geq 1$ is represented on ’half’ of the Heegaard surface. 2pt $L(1,l)$ at 39 74 $l-1$ at 185 151 Figure 23: The open book for $L(1,l)$ with $l\geq 1$. The open book decomposition found above equips $S^{3}$ with a Heegaard decomposition compatible with $L_{k,l}$; the $\alpha$– and $\beta$–curves of this decomposition for $k=0$ are given by Figure 24 when $l=0$ and by Figure 25 when $l>0$. 2pt $L_{0,0}$ at 76 41 $w$ at 130 18 $z$ at 44 161 Figure 24: The Heegaard decomposition for $L(0,0)$. 2pt $z$ at 45 306 $w$ at 145 18 $L_{0,l}$ at 81 43 $A$ at 78 661 $B$ at 153 710 $C$ at 153 622 $C^{\prime}$ at 206 622 Figure 25: The Heegaard decomposition for $L(0,l)$, $l>0$. Recall that we get the $b_{i}$ curves by the usual perturbation of the $a_{i}$’s and the action of the monodromy yields a Heegaard decomposition for $S^{3}$ with the distinguished point ${\bf{x}}$ in ${\mathbb{T}}_{\alpha}\cap{\mathbb{T}}_{\beta}$ determining the Legendrian invariant. As usual, the basepoint $z$ is placed in the ‘large’ region of the page $S_{+1}$, while the point $w$ giving rise to $\widehat{\mathfrak{L}}$ is in the strip between $a_{1}$ and $b_{1}$, as indicated in the pictures. It is easy to see that moving $w$ to the other domain in the strip gives $\widehat{\mathfrak{L}}=0$, because there is a holomorphic disk running into ${\bf{x}}$. Recall that $L_{k,l}$ is isotopic to the negative torus knot $T_{(2,-(2n-1))}$. The group $\widehat{\mathrm{HFK}}(-S^{3},T_{(2,-(2n-1))})$ is isomorphic to the Floer homology $\widehat{\mathrm{HFK}}(S^{3},T_{(2,2n-1)})$ of the positive torus knot $T_{(2,2n-1)}$. In this case we have the following (compare Equation (6.1)): $\widehat{\mathrm{HFK}}_{s-n+1}(S^{3},T_{(2,2n-1)},s)\cong\left\\{\begin{array}[]{ll}\mathbb{F}&{\text{if $|s|\leq n-1$}}\\\ 0&{\text{otherwise.}}\end{array}\right.$ (6.7) After these preparations we are ready to determine the invariants of some of the Legendrian knots discussed above. The computation admits a relatively simple scheme when $k=0$, so we start with that case. ###### Theorem 6.8. Let $L_{0,l}$ be given the orientation specified by the diagram of Figure 24 or Figure 25. Then, $\widehat{\mathfrak{L}}(L_{0,l})\neq 0$. ###### Proof. We show that the intersection point ${\bf{x}}_{0}\in\widehat{\mathrm{CFK}}$, which determines $\widehat{\mathfrak{L}}(L_{0,l})$, is alone in its Alexander grading. Claim: Any other intersection point ${\bf{y}}$ in the Heegaard decomposition given by Figure 24 or 25 has Alexander grading $A({\bf{y}})$ strictly less than $A({\bf{x}}_{0})$. The claim, in turn, will be proved by induction on $l$. For $l=0$ (which means that we are computing Legendrian invariants of a Legendrian left–handed trefoil knot in an overtwisted $S^{3}$) the explicit determination of the Alexander gradings of each of the 19 intersection points verifies the statement. To do this we only need to find domains connecting any two intersection points, and their relative Alexander grading is the multiplicity of the domain containing the basepoint $w$. A straightforward linear algebra argument similar to the one outlined in the proof of Proposition 6.2 shows that all top three degress contain unique intersection points (giving rise to homology elements), while the remaining 16 intersection points are of strictly smaller Alexander grading. The argument for the inductive step proceeds as follows: Notice that by deleting the top–most $\alpha$– and $\beta$–curves (containing the point $A$ of Figure 25) in the Heegaard decomposition ${\mathfrak{H}}_{1}$ (and destabilizing the Heegaard surface) we get a Heegaard decomposition ${\mathfrak{H}}_{0}$ adapted to $T_{(2,2n-3)}$. Repeating this procedure one more time we get a Heegaard decomposition ${\mathfrak{H}}_{-1}$, which will be a decomposition adapted to $T_{(2,2n-5)}$. Note that the intersection matrix of the $\alpha$– and $\beta$–curves has the following form $\left(\begin{array}[]{rrrrrrrr}-2&1&0&0&0&0&0\\\ 1&-2&1&0&0&0&0\\\ 0&1&\ddots&\ddots&0&0&0\\\ 0&0&\ddots&-2&1&0&0\\\ 0&0&0&1&-4&2&0\\\ 0&0&0&0&2&-2&1\\\ 0&0&0&0&0&1&-1\end{array}\right)$ Now let us group the intersection points of the original diagram into types $A,B$ and $C$ depending on their coordinate on the top–most $\beta$–curve. It is easy to see that points in each $A$ and $B$ are in 1-1 correspondence with intersections in ${\mathfrak{H}}_{0}$, while points in $C$ are in 1-1 correspondence with points in ${\mathfrak{H}}_{-1}$. From the form of this matrix it is easy to see that the relative gradings within the groups coincide with the relative gradings in their respective Heegaard diagrams ${\mathfrak{H}}_{0}$ and ${\mathfrak{H}}_{-1}$. As usual, let ${\bf{x}}_{0}$ denote the intersection point representing the Legendrian invariant, $B_{{\bf{x}}_{0}}$ the point with the same coordinates except on the top-most $\alpha$–circle, where $A$ is substituted with $B$. Finally, $C_{{\bf{x}}_{0}}$ will denote the intersection point where on the top–most $\beta$–circle we choose $C$ (and so on the top–most $\alpha$–circle we should take $C^{\prime}$) and otherwise we take the same intersection points as in ${\bf{x}}_{0}$. Our inductive assumption is that ${\bf{x}}_{0}$, $B_{{\bf{x}}_{0}}$ and $C_{{\bf{x}}_{0}}$ are of highest Alexander gradings in the groups $A,B,C$, respectively. Therefore to conclude the argument we only need to compare the gradings of these intersection points. This relative computation can be performed locally near the top–most $\alpha$– and $\beta$–curves and we get that $A({\bf{x}})-A(B_{{\bf{x}}})=2$ and $A({\bf{x}})-A(C_{{\bf{x}}})=2$. By the inductive hypothesis this shows that ${\bf{x}}_{0}$ is the unique cycle in the top Alexander grading, concluding the proof. ∎ Note that the relation found in the proof of Theorem 6.8 amongst the Alexander gradings of the intersection points of the diagrams ${\mathfrak{H}}_{i}$ ($i=\pm 1,0$) is a manifestation of the fact that the symmetrized Alexander polynomial $\Delta_{n}(t):=\Delta_{T_{(2,2n-1)}}(t)$ satisfies the identity $\Delta_{n}(t)=(t-t^{-1})\Delta_{n-1}(t)-\Delta_{n-2}(t).$ We consider two more cases, which can be handled by explicit methods: $L_{1,1}$ and $L_{1,2}$. In these cases the Alexander grading alone is not sufficient to calculate the homology class; we need a little analysis of holomorphic disks. We begin with the first of these: ###### Theorem 6.9. Let $L_{1,1}$ be given the orientation specified by the diagram of Figure 26. Then, $\widehat{\mathfrak{L}}(L_{1,1})\neq 0$. Recall that $L_{1,1}$ represents the negative torus knot $T_{(2,-7)}$. ###### Proof. Consider the adapted Heegaard diagram for $L_{1,1}$ exhibited in Figure 26. 2pt $w$ at 141 18 $z$ at 31 258 $L_{1,1}$ at 77 41 $E$ at 82 68 $R$ at 157 99 $S$ at 139 117 $D_{1}$ at 82 165 $V_{1}$ at 142 199 $D_{2}$ at 172 200 $V_{2}$ at 192 198 $C_{1}$ at 78 257 $C_{2}$ at 148 218 $M_{1}$ at 99 210 $M_{2}$ at 166 218 $C_{3}$ at 184 217 $C_{4}$ at 220 261 $C_{5}$ at 218 273 $C_{6}$ at 215 289 $P_{2}$ at 176 304 $C_{7}$ at 159 293 $P_{1}$ at 136 293 $B_{3}$ at 156 314 $B_{2}$ at 134 313 $Q_{1}$ at 113 312 $X$ at 169 399 $Y$ at 131 409 $B_{1}$ at 85 392 $A_{1}$ at 76 449 $A_{2}$ at 148 497 Figure 26: Adapted Heegaard diagram for $L_{1,1}$. Analyzing this diagram as in the other cases, we see that there are $149$ elements in $\mathbb{T}_{\alpha}\cap\mathbb{T}_{\beta}$, seven of which with the same Alexander grading as the point ${\bf{x}}_{0}$ determining $\widehat{\mathfrak{L}}(L(1,1))$. Indeed, in the notation suggested by that diagram, the seven intersection points in Alexander grading $A=1$ are $\begin{array}[]{llll}A_{2}B_{2}C_{7}D_{1}E,&A_{2}D_{1}EP_{2}Q_{1},&A_{2}D_{1}EP_{1}Q_{1},&A_{2}B_{1}C_{6}D_{1}E,\\\ XYC_{6}D_{1}E,&A_{1}B_{2}C_{5}D_{1}E,&A_{1}B_{1}C_{1}D_{1}E={\bf{x}}_{0}.&\end{array}$ Since ${\bf{x}}_{0}$ is a cycle, it can be thought of as determining a subcomplex of $\widehat{\mathrm{CFK}}(L_{1,1},1)$. The quotient complex then is generated by the six remaining generators in the given Alexander gradings. Indeed, it is straightforward to find the positive homotopy classes connecting these six generators with Maslov index one. There are six, and five of these are rectangles, and hence admit holomorphic representatives. The sixth (connecting $XYC_{6}D_{1}E$ to $A_{1}B_{2}C_{5}D_{1}E$) is also a planar surface, and it is not difficult to verify directly that it, too, always has a holomorphic representative. We assemble this information of the chain complex in Figure 27. $A_{2}D_{1}EP_{2}Q_{1}$$A_{2}B_{1}C_{6}D_{1}E$$A_{2}B_{2}C_{7}D_{1}E$$A_{2}D_{1}EP_{1}Q_{2}$$A_{1}B_{2}C_{5}D_{1}E$$XYC_{6}D_{1}E$ Figure 27: Part of the chain complex for $L_{1,1}$. This is the quotient of $\widehat{\mathrm{CFK}}(L_{1,1},1)$ by the generator $A_{1}B_{1}C_{1}D_{1}E$. Arrows indicate differentials. It is straightforward to verify that this complex is acyclic. It follows at once that ${\bf{x}}_{0}$ represents the nontrivial homology class in $\widehat{\mathrm{HFK}}(L_{1,1},1)$. ∎ In a similar vein, we have the following: ###### Theorem 6.10. Let $L_{1,2}$ be given the orientation specified by the diagram of Figure 28. Then, $\widehat{\mathfrak{L}}(L_{1,2})\neq 0$. Recall that $L_{1,2}$ corresponds to $T_{(2,-9)}$. ###### Proof. The adapted Heegaard diagram is pictured in Figure 28. 2pt $w$ at 141 18 $z$ at 31 258 $L_{1,2}$ at 77 41 $F$ at 84 68 $T$ at 160 98 $S$ at 139 117 $E_{1}$ at 86 165 $R_{1}$ at 142 199 $E_{2}$ at 176 199 $R_{2}$ at 194 198 $D_{5}$ at 148 218 $N_{1}$ at 99 210 $N_{2}$ at 169 217 $D_{4}$ at 187 217 $D_{3}$ at 223 260 $D_{2}$ at 213 294 $V$ at 170 294 $M$ at 155 309 $C_{4}$ at 179 310 $D_{1}$ at 85 257 $C_{1}$ at 123 308 $P_{1}$ at 127 396 $C_{2}$ at 162 388 $P_{2}$ at 180 401 $C_{3}$ at 216 382 $B_{1}$ at 104 491 $Q_{1}$ at 107 396 $B_{2}$ at 138 409 $B_{3}$ at 160 409 $X$ at 176 495 $Y$ at 131 494 $A_{1}$ at 79 544 $A_{2}$ at 153 592 $1$ at 251 452 $2$ at 220 529 $3$ at 146 397 $4$ at 94 381 $5$ at 148 413 $6$ at 172 413 $7$ at 113 410 $8$ at 118 502 $9$ at 100 291 $10$ at 191 295 Figure 28: Adapted Heegaard diagram for $L_{1,2}$. The domains (in the complement of the $\alpha$– and $\beta$–circles) labeled by $1,...,10$ play a role in Figure 29. There are $347$ intersection points, $13$ of which are in the same Alexander grading as ${\bf{x}}_{0}$. In terms of the numbering conventions from Figure 28, these are $\begin{array}[]{llll}A_{2}B_{1}C_{3}D_{1}E_{1}F&A_{2}B_{2}C_{2}D_{1}E_{1}F&A_{1}B_{2}C_{4}D_{1}E_{1}F&A_{1}B_{2}C_{1}D_{2}E_{1}F\\\ A_{1}B_{1}C_{2}D_{2}E_{1}F&A_{2}B_{2}C_{3}D_{2}E_{1}F&A_{1}B_{2}MVE_{1}F&A_{1}D_{2}E_{1}P_{1}Q_{1}F\\\ A_{2}D_{1}E_{1}P_{2}E_{1}F&A_{2}D_{1}E_{1}P_{1}Q_{2}F&A_{2}D_{2}E_{1}P_{2}Q_{2}F&XYC_{3}D_{1}E_{1}F\\\ A_{1}B_{1}C_{1}D_{1}E_{1}F={\bf{x}}_{0}.&&&\end{array}$ We can draw a graph whose vertices consist of these generators, and its arrows denote positive Whitney disks with Maslov index one, as pictured in Figure 29. $B_{2}C_{3}$$P_{2}Q_{2}$$B_{1}C_{2}$$P_{1}Q_{1}$$B_{2}C_{1}$$A_{1}B_{2}MV$$A_{2}B_{2}C_{4}D_{1}E_{1}$$XYC_{3}$$A_{2}B_{1}C_{3}$$P_{2}Q_{1}$$B_{2}C_{2}$$P_{1}Q_{2}$$B_{1}C_{1}$$6$$1+2$$7$$4$$10$$9$$6+8$$2$$6+3$$5$$3$$1+2+3+4$$5+6$$1+2+4+5$ Figure 29: Complex for $L_{1,2}$. We have displayed here the $13$ generators in $A=-3$. Arrows represent domains connecting generators. All the thicker arrows actually represent quadrilaterals; there are only four which are not: those containing $1$, and those containing $6+8$. Those domains are written as sums of the $10$ domains labelled in Figure 28. Note that in that figure, we have abbreviated the names of generators (for example, dropping $F$, which occurs in each). In that picture, all but four of the arrows represent rectangles (and hence, immediately, nontrivial differentials in the chain complex). Inspection of this graph immediately establishes that there are no other non–negative Whitney disks among the given generators with Maslov index one (for example, the domain from $P_{2}Q_{1}=A_{2}D_{1}E_{1}P_{2}Q_{1}F$ to $B_{2}C_{4}=A_{2}B_{2}C_{4}D_{1}E_{1}$ is gotten as $6+3-2+8+6$, which contains $2$ with multiplicity $-1$). To show that $B_{1}C_{1}=A_{1}B_{1}C_{1}D_{1}F$ represents a homologically nontrivial cycle, it suffices to show that the positive domain $6+8$ supports holomorphic representatives, which can be done by direct means. (Note that it is an annulus with an appropriate cut.) ∎ ###### Remark 6.11. Recall that $\mathrm{HFK}^{-}(S^{3},T_{(2,2n-1)})\cong\mathbb{F}^{n-1}\oplus\mathbb{F}[U],$ where the top generator of the free $\mathbb{F}[U]$–module is at $(A=-(n-1),\quad M=-2(n-1)),$ while the $n-1$ generators of the $\mathbb{F}^{n-1}$ summand are of bi–degrees $(A=n-1-2i,\quad M=-2i),\quad i=0,\ldots,n-2.$ (This follows readily from the calculation of $\widehat{\mathrm{HFK}}$, stated in Equation (6.7).) Once again, the computation above can be adapted to show that the homogeneous $U$–torsion elements ${\mathfrak{L}}(L_{0,l}),{\mathfrak{L}}(L_{1,1})$ and ${\mathfrak{L}}(L_{1,2})$ in $\mathrm{HFK}^{-}(S^{3},T_{(2,2n-1)})$ are also nontrivial. Notice that there is an obvious bijection between the homogeneous $U$–torsion elements of $\mathrm{HFK}^{-}(S^{3},T_{(2,2n-1)})$ and the Legendrian knots $L_{k,l}$ with $k+l=n-2$ constructed above. Moreover, this bijection can be chosen in such a way that if $L_{k,l}$ corresponds to $x_{k,l}\in\mathrm{HFK}^{-}(S^{3},T_{(2,2n+1)})$ then $d_{3}(\xi_{k,l})=2A(x_{k,l})-M(x_{k,l}).$ It is reasonable to expect that the Legendrian invariants $\widehat{\mathfrak{L}}(L_{k,l})$ and ${\mathfrak{L}}(L_{k,l})$ are nonzero for all $k,l\geq 0$, and hence that ${\mathfrak{L}}(L_{k,l})$ is determined by $x_{k,l}$. We will not address this general computation in the present paper. ## 7 Connected sums Suppose that $L_{i}\subset(Y_{i},\xi_{i})$ are oriented Legendrian knots in the contact 3–manifolds $(Y_{i},\xi_{i})$, $i=1,2$. We want to relate the invariants of $L_{1}$ and $L_{2}$ with the invariants of the connected sum $L_{1}\\#L_{2}\subset(Y_{1}\\#Y_{2},\xi_{1}\\#\xi_{2})$ (see [14] for the definition of the connected sum operation in the contact setting). Fix open book decompositions $(B_{i},\varphi_{i})$ adapted to $L_{i}\subset(Y_{i},\xi_{i})$ (with pages $S_{i}$) for $i=1,2$. We may assume that suitable portions of the open books and the adapted bases appear as in the left–hand picture of Figure 30, where, as usual, the ordered pairs $(z_{i},w_{i})$ determine the orientations of the knots. We can now perform the Legendrian connected sum of $(Y_{1},\xi_{1},L_{1})$ and $(Y_{2},\xi_{2},L_{2})$ in such a way that we glue the open books as well as the contact structures and the knots. Specifically, we take the Murasugi sum of the two open books in the domains containing $z_{2}$ and $w_{1}$, and then we drop these two basepoints. We can make sure that the resulting oriented knot $L_{1}\\#L_{2}$ is smoothly determined on a page of the resulting open book by the ordered pair $(z_{1},w_{2})$, as illustrated in the central picture of Figure 30. The bases of arcs can also be arranged to be the same as the ones illustrated. There is a corresponding map of chain complexes $\Phi\colon\mathrm{CFK}^{-}(-Y_{1},L_{1})\otimes_{\mathbb{F}[U]}\mathrm{CFK}^{-}(-Y_{2},L_{2})\to\mathrm{CFK}^{-}(-(Y_{1}\\#Y_{2}),L_{1}\\#L_{2})$ (7.1) which (denoting the intersection point determining ${\mathfrak{L}}(L_{i})$ by ${\bf{x}}(L_{i})$) maps the intersection point ${\bf{x}}(L_{1})\otimes{\bf{x}}(L_{2})$ to the generator $({\bf{x}}(L_{1}),{\bf{x}}(L_{2}))$ of the chain complex of the connected sum. According to [35, Section 7] the above map induces an isomorphism on homology. The union of the two bases is not adapted to the Legendrian knot $L_{1}\\#L_{2}$, since each of the two bases contains an arc intersecting it. Nevertheless, as it will be shown in the proof of Theorem 7.1, there is a sequence of arc slides which carries the new basis of curves into an adapted basis for $L_{1}\\#L_{2}$, inducing handleslides on the underlying Heegaard diagram for $(-(Y_{1}\\#Y_{2}),L_{1}\\#L_{2})$. These handleslides induce a map of chain complexes by counting holomorphic triangles (cf. [33, Section 9]) $\Psi\colon\mathrm{CFK}^{-}(-(Y_{1}\\#Y_{2}),L_{1}\\#L_{2})\longrightarrow\mathrm{CFK}^{-}(-(Y_{1}\\#Y_{2}),L_{1}\\#L_{2}),$ where the second chain complex now denotes the chain complex with respect to the adapted Heegaard diagram. ###### Theorem 7.1. Suppose that $L_{1}\\#L_{2}\subset(Y_{1}\\#Y_{2},\xi_{1}\\#\xi_{2})$ is the oriented connected sum of the oriented Legendrian knots $L_{i}\subset(Y_{i},\xi_{i})$, $(i=1,2)$. Then, there is a quasi–isomorphism $F\colon\mathrm{CFK}^{-}(-Y_{1},L_{1})\otimes_{\mathbb{F}[U]}\mathrm{CFK}^{-}(-Y_{2},L_{2})\to\mathrm{CFK}^{-}(-(Y_{1}\\#Y_{2}),L_{1}\\#L_{2})$ which maps ${\bf{x}}(L_{1})\otimes{\bf{x}}(L_{2})$ to ${\bf{x}}(L_{1}\\#L_{2})$. ###### Proof. There are three Heegaard diagrams for $(-(Y_{1}\\#Y_{2}),L_{1}\\#L_{2})$ coming into play: the connected sum diagram (whose $\mathbb{\alpha}$– and $\mathbb{\beta}$–circles are gotten by doubling the initial bases), an intermediate diagram gotten by sliding the $\alpha$-circles as dictated by the arcslides in the middle diagram in Figure 30 (whose attaching circles we denote $\alpha^{\prime}$ and $\beta$), and the final one gotten by performing handleslides on the $\beta$-circles, as dictated by arcslides of the $b_{i}$ as in the rightmost diagram in Figure 30 (whose attaching cicles are $\alpha^{\prime}$ and $\beta^{\prime}$). Recall, that the $a_{i}$ arcs determine the $\alpha$–circles while the $b_{i}$’s give rise to the $\beta$–circles, and we are examining the Heegaard diagrams $(\Sigma,\beta,\alpha),(\Sigma,\beta,\alpha^{\prime})$ and $(\Sigma,\beta^{\prime},\alpha^{\prime})$. 2pt $z_{1}$ at 20 40 $z_{2}$ at 20 114 $a_{2}$ at 35 15 $b_{2}$ at 61 14 $w_{1}$ at 47 51 $w_{2}$ at 43 105 $b_{1}$ at 32 142 $a_{1}$ at 56 143 $b_{2}$ at 184 12 $a_{2}$ at 135 10 $a^{\prime}_{2}$ at 146 11 $b_{1}$ at 180 115 $a_{1}$ at 205 124 $a^{\prime}_{1}$ at 192 119 $z_{1}$ at 121 34 $w_{2}$ at 208 77 $b_{2}$ at 319 9 $b^{\prime}_{2}$ at 299 8 $a^{\prime}_{2}$ at 282 8 $b_{1}$ at 315 110 $b^{\prime}_{1}$ at 329 116 $a^{\prime}_{1}$ at 340 120 $z_{1}$ at 260 28 $w_{2}$ at 341 71 Figure 30: Connected sums. Starting from two knots as pictured on the left, we form their connected sum to obtain a Heegaard diagram as in the middle, where we have three sets of arcs, two sets of which are isotopic to one another, and the third is obtained as an arcslide. In the middle picture, the dark dots represent the initial generator ${\bf{x}}$ (which represents the product of the two Legendrian invariants), the white dots represent the canonical generator for $\mathbb{T}_{\alpha}\cap{\mathbb{T}}_{\alpha^{\prime}}$, and the grey dots represent the intermediate generator ${\bf{x}}^{\prime}$; whereas on the picture on the right, the dark dots represent ${\bf{x}}^{\prime}$, the white dots represent the canonical generator of $\mathbb{T}_{\beta}\cap{\mathbb{T}}_{\beta^{\prime}}$, and the grey dots represent the final intersection point ${\bf{x}}^{\prime\prime}$ (which represents the Legendrian invariant for the connected sum). In all three of these diagrams there is a unique intersection point of the $\alpha$– and $\beta$–tori which is supported in $S_{+1}$: the first, ${\bf{x}}$, represents the product of ${\bf{x}}(L_{1})$ and ${\bf{x}}(L_{2})$ under the connected sum map (denoted above by $\Phi$), the second is denoted ${\bf{x}}^{\prime}$, and the third, ${\bf{x}}^{\prime\prime}$, clearly determines the Legendrian invariant for $L_{1}\\#L_{2}$. We claim that the three generators are mapped to one another under the maps induced by handleslides. As a warm-up, we argue first that ${\bf{x}}^{\prime}$ is, indeed, a cycle. As for the contact class, we argue that for any ${\bf{y}}\in{\mathbb{T}}_{\beta^{\prime}}\cap\mathbb{T}_{\alpha}$, if $\phi\in\pi_{2}({\bf{x}}^{\prime},{\bf{y}})$ is any Whitney disk with all non–negative local multiplicities, then ${\bf{x}}^{\prime}={\bf{y}}$ and $\phi$ is the constant disk. The argument is similar to the argument from [19] recalled in the proof of Theorem 2.11 with one slight difference: now the arcs in the basis disconnect $S_{+1}$ into two regions, only one of which contains $z$. However, it is still easy to see (by another look at Figure 30) that any postive domain flowing out of ${\bf{x}}^{\prime}$ which has positive multiplicity on the other region (not containing $z$) must also have positive multiplicity at $z$. We now turn to the verification of the claim about the triangle maps. The Heegaard triple in this case is $(\Sigma,\beta,\alpha,\alpha^{\prime},w,z)$, and the diagram $(\Sigma,\alpha,\alpha^{\prime},w,z)$ represents an unknot in the $g$–fold connected sum of $S^{2}\times S^{1}$. Moreover, in that diagram there is a unique intersection point $\Theta\in\mathbb{T}_{\alpha}\cap{\mathbb{T}}_{\alpha^{\prime}}$ representing the top–dimensional Floer homology class. The handleslide map $\Psi_{1}\colon\mathrm{CFK}^{-}(\Sigma,\beta,\alpha,w,z)\longrightarrow\mathrm{CFK}^{-}(\Sigma,\beta,\alpha^{\prime},w,z)$ is defined by $\Psi_{1}({\mathbf{u}})=\sum_{{\bf{y}}\in\mathbb{T}_{\beta}\cap{\mathbb{T}}_{\alpha^{\prime}}}\sum_{\\{\psi\in\pi_{2}({\mathbf{u}},\Theta,{\bf{y}})\big{|}\mu(\psi)=0,n_{z}(\psi)=0\\}}\\#{\mathfrak{M}}(\psi)\cdot U^{n_{w}(\psi)}{\bf{y}},$ where, as usual, $\pi_{2}({\mathbf{u}},\Theta,{\bf{y}})$ denotes the space of homology classes of Whitney triangles at ${\mathbf{u}}$, $\Theta$, and ${\bf{y}}$. The claim that $\Psi_{1}({\bf{x}})={\bf{x}}^{\prime}$ follows from the facts that * • there is a triangle $\psi_{0}\in\pi_{2}({\bf{x}},\Theta,{\bf{x}}^{\prime})$, gotten as a disjoint union of the obvious small triangles in $S_{+1}$, and * • any other triangle $\psi\in\pi_{2}({\bf{x}},\Theta,{\bf{y}})$ with $n_{z}(\psi)=0$ has a negative local multiplicity somewhere. The first is gotten by glancing at Figure 30. To see the second, notice that any homology class $\psi\in\pi_{2}({\bf{x}}^{\prime},\Theta,{\bf{y}})$ can be decomposed as $\psi_{0}*\phi$, where $\phi\in\pi_{2}({\bf{x}}^{\prime},{\bf{y}})$. Moreover, if $\psi$ has only positive local multiplicities, then the same follows for $\phi$; also, since $n_{z}(\psi)=0=n_{z}(\psi_{0})$, it follows that $n_{z}(\phi)=0$, as well. But by our above argument that ${\bf{x}}^{\prime}$ is a cycle, it now follows that $\phi$ is constant. Consider next the handleslide map $\Psi_{2}\colon(\Sigma,\beta,\alpha^{\prime},w,z)\longrightarrow(\Sigma,\beta^{\prime},\alpha^{\prime},w,z).$ This is defined by pairing with an intersection point ${\Theta^{\prime}}\in{\mathbb{T}}_{\beta^{\prime}}\cap\mathbb{T}_{\beta}$ representing the top–dimensional nontrivial homology. The argument that $\Psi_{2}({\bf{x}}^{\prime})={\bf{x}}^{\prime\prime}$, where ${\bf{x}}^{\prime\prime}$ determines the Legendrian invariant for the connected sum, follows through a similar argument. The composition $\Psi_{2}\circ\Psi_{1}\circ\Phi$ now induces an isomorphism on homology, carrying ${\bf{x}}(L_{1})\otimes{\bf{x}}(L_{2})$ to ${\bf{x}}(L_{1}\\#L_{2})$, which completes the proof of the theorem. ∎ The above result quickly leads to the following observation, providing ###### Proof of Theorem 1.6. Interpret stabilization as the connected sum with the stabilized Legendrian unknot in the standard contact 3–sphere, use the connected sum formula and the model calculation given in Section 4. This gives quasi–isomorphisms $\Psi^{-}\colon\thinspace\mathrm{CFK}^{-}(-Y,L)\to\mathrm{CFK}^{-}(-Y,L^{-})$ and $\Psi^{+}\colon\thinspace\mathrm{CFK}^{-}(-Y,L)\to\mathrm{CFK}^{-}(-Y,L^{+})$ such that $\Psi^{-}({\bf{x}}(L))={\bf{x}}((L^{-})\quad\text{and}\quad U\cdot\Psi^{+}({\bf{x}}(L))={\bf{x}}(L^{+}),$ concluding the proof of the theorem. ∎ Note that Proposition 5.2 is a special case of Theorem 1.6; we gave a separate proof of it earlier, since that proof is somewhat more direct. As an application of our previous computations we can now prove the following result: ###### Theorem 7.2. Let $\xi_{i}$ denote the overtwisted contact structure on $S^{3}$ with Hopf invariant $d_{3}(\xi_{i})=i\in{\mathbb{Z}}$. Then, $\xi_{i}$ contains non–loose Legendrian knots with non–vanishing Legendrian invariants $\widehat{\mathfrak{L}}$ for each $i\in{\mathbb{Z}}$. ###### Proof. By Proposition 6.4 and Lemma 6.1, for $i>0$ even the Legendrian knot $L_{0,i-1}$, while for $i<0$ odd the knot $L(i)$ satisfies the claim. If $i=2j-1>0$ is odd then by the connected sum formula $L_{0,j-1}\\#L(1)$ is a good choice. For $i=2j<0$ even take $L(j)\\#L(1)$ and finally for $i=0$ the knot $L_{0,0}\\#L(1)\\#L(1)$ will do. ∎ ###### Corollary 7.3. Suppose that $(Y,\xi)$ is a contact 3–manifold with nonvanishing contact invariant $c(Y,\xi)$ and $\zeta$ is an overtwisted contact structure on $Y$ with $\mathbf{t}_{\zeta}=\mathbf{t}_{\xi}$. Then $(Y,\zeta)$ contains Legendrian (transverse) knots with nonvanishing Legendrian (resp. transverse) invariants. ###### Proof. Consider a Legendrian unknot $L$ (in a Darboux chart) in $(Y,\xi)$ with $c(Y,\xi)\neq 0$. It is not difficult to find, possibly after stabilization, an open book decomposition compatible with $(Y,\xi,L)$ together with an adapted basis, in such a way that the two basepoints $z$ and $w$ determining $L$ are, in fact, in the same domain. This observation immediately implies $\widehat{\mathrm{HFK}}(-Y,L)=\widehat{\mathrm{HF}}(-Y)$, and the Legendrian invariant $\widehat{\mathfrak{L}}$ of $L$ is easily seen to be nonzero in $\widehat{\mathrm{HFK}}(-Y,L)$, being equal to $c(Y,\xi)$. Any overtwisted contact structure $\zeta$ on $Y$ with $\mathbf{t}_{\zeta}=\mathbf{t}_{\xi}$ can be given as $\xi\\#\xi_{i}$, where $\xi_{i}$ is the overtwisted contact structure on $S^{3}$ with $d_{3}(\xi_{i})=i$. This fact easily follows from Eliashberg’s classification of overtwisted contact structures, together with the classification of homotopy types of 2–plane fields. Now the connected sum of the Legendrian unknot with the Legendrian knot having nontrivial Legendrian knot invariant in $\xi_{i}$ (provided by Theorem 7.2) has nontrivial Legendrian invariant by the connected sum formula. This knot, and each of its negative stabilizations are non–loose, and hence its transverse push–off is non–loose and has nontrivial transverse invariant. ∎ ## 8 Transversely non–simple knots A knot type $\mathcal{K}$ in $S^{3}$ is traditionally called _transversely simple_ if two transverse knots $T_{1},T_{2}$ in the standard contact structure $\xi_{st}$ both of knot type $\mathcal{K}$ and equal self–linking number are transversely isotopic. Transversely non–simple knot types were recently found by Etnyre-Honda, Birman-Menasco [15, 1], cf. also [29, 44]. The notion of transverse simplicity obviously generalizes to any null–homologous knot type (for the self–linking number to be well–defined) in any contact 3–manifold $(Y,\xi)$. In this form of the definition, however, it is easy to find transversely non–simple knot types provided $\xi$ is overtwisted: The binding of any open book decomposition compatible with $\xi$ is a null–homologous, non–loose transverse knot, while the same binding in $(Y\\#S^{3},\xi\\#\xi_{0})$ (where $\xi_{0}$ is the overtwisted contact structure on $S^{3}$ homotopic to $\xi_{st}$ and the connected sum is taken away from the binding) is loose. Since $\xi$ and $\xi\\#\xi_{0}$ are homotopic as 2–plane fields, hence isotopic as contact structures, the binding gives rise to two transverse knots in $\xi$. The self–linking numbers of these two representatives are clearly equal, while one has tight, the other overtwisted complement; consequently the two knots cannot be transversely isotopic. In conclusion, for overtwisted contact structures in the definition of transverse simplicity we should require that the two knots are either both non–loose or both loose (besides being smoothly isotopic and having identical self–linking). Two transverse loose knots with equal knot type and self–linking number, and with a _common_ overtwisted disk in their complement can be shown to be isotopic. Without the existence of the common overtwisted disk, however, this question is surprisingly subtle. With the aid of the transverse invariant $\mathfrak{T}$ and the model calculations for $L_{1,1}$ and $L_{1,2}$ discussed in the previous section (together with a refinement of the gradings on Heegaard Floer homology, deferred to Section 9) we get ###### Theorem 8.1. Let ${\mathcal{K}}$ denote the knot type obtained by the connected sum of the negative torus knots $T_{(2,-7)}$ and $T_{(2,-9)}$. Then, in the overtwisted contact 3–manifold $(S^{3},\xi_{12})$ with Hopf invariant $d_{3}(\xi_{12})=12$ there are two non–loose Legendrian knots representing ${\mathcal{K}}$, with equal Thurston–Bennequin and rotation numbers, which are not Legendrian isotopic and stay nonisotopic after arbitrarily many negative stabilizations. The proof of Theorem 8.1 proceeds by considering a pair of appropriate Legendrian representatives of ${\mathcal{K}}$, the computation of their invariants, and finally the distinction of these elements. Our candidate Legendrian knots are $L_{1}=L_{0,2}\\#L_{1,2}\quad{\mbox{and}}\quad L_{2}=L_{1,1}\\#L_{0,3}.$ The knots $L_{1}$ and $L_{2}$ are obviously smoothly isotopic to $T_{(2,-7)}\\#T_{(2,-9)}$, and their Thurston–Bennequin and rotation numbers can be easily deduced from their definition: ###### Lemma 8.2. The Thurston–Bennequin and rotation numbers of $L_{1}$ and $L_{2}$ are $\operatorname{tb}(L_{1})=\operatorname{tb}(L_{2})=-31\quad{\mbox{and}}\quad\operatorname{rot}(L_{1})=\operatorname{rot}(L_{2})=-40,$ with the orientations specified by Figure 17. ###### Proof. The formulae follow from the additivity of rotation numbers and the identity $\operatorname{tb}(L\\#L^{\prime})=\operatorname{tb}(L)+\operatorname{tb}(L^{\prime})+1$ (cf. [14, Lemma 3.3]), together with the computations of Lemma 6.7. ∎ Since $L_{1}$ and $L_{2}$ are knots in the connected sums of the corresponding overtwisted contact structures, the claim about the resulting contact structure follows. (Recall that on $S^{3}$ an overtwisted contact structure is determined up to isotopy by its Hopf invariant, which is additive under connected sum.) The connected sum formula of Theorem 7.1 implies that the Legendrian invariant of both knots is determined by the tensor product of the invariants of the summands. This argument shows that $\widehat{\mathfrak{L}}(L_{1})$ and $\widehat{\mathfrak{L}}(L_{2})$ are non–vanishing, hence the knots $L_{1},L_{2}$ are both non–loose. (Notice that, in general, the connected sum of two non–loose knots might become loose.) The connected sum formula for knot Floer homology shows that in the Alexander and Maslov grading $(A=5,M=-2)$ of the elements $\widehat{\mathfrak{L}}(L_{i})$ the knot Floer homology group $\widehat{\mathrm{HFK}}(S^{3},T_{(2,7)}\\#T_{(2,9)})$ is isomorphic to $\mathbb{F}\oplus\mathbb{F}\oplus\mathbb{F}$. (Only $\mathbb{F}\oplus\mathbb{F}$ comes from the $\mathrm{HFK}^{-}$–theory, hence capable of containing $\widehat{\mathfrak{L}}$.) Therefore to conclude our proof for Theorem 8.1 we need to show that the knots $L_{1}$ and $L_{2}$ are not isotopic. This step requires an auxiliary result from knot Floer homology, showing that elements of the knot Floer group of a connected sum do remember the Alexander and Maslov gradings of their components in the summands. This leads to a refinement of the invariant, which we explain now. (The necessary Heegaard Floer theoretic result will be given in Section 9.) Recall that the invariant ${\mathfrak{L}}(L)$ was defined as an element of $\mathrm{HFK}^{-}(-Y,L)$, determined up to any $\mathbb{F}[U]$–module isomorphism of this module. The ambiguity stems from the fact that for (oriented) Legendrian isotopic knots $L_{1},L_{2}$ the isotopy is not canonical, therefore the isomorphism between the knot Floer homologies $\mathrm{HFK}^{-}(-Y,L_{1})$ and $\mathrm{HFK}^{-}(-Y,L_{2})$ will depend on the chosen isotopy. To prove vanishing and non–vanishing results, it was sufficient to mod out by all the module automorphisms. When we want to distinguish knots based on their invariants, however, we might need to understand the necessary equivalence relations a little better. Notice that an isotopy between $L_{1}$ and $L_{2}$ induces a filtered chain homotopy between the filtered chain complexes $\mathrm{CFK}^{-}(-Y,L_{i})$ (where the filtration comes from the second basepoint), hence any algebraic structure of the filtered chain complex provides further restrictions on the automorphisms of $\mathrm{HFK}^{-}(-Y,L)$ we need to take into account for the invariance property to hold. For example, the _length_ of a homogeneous element $x$ (that is, the minimal $n$ for which $d_{n}(x)$ vanishes, where $d_{n}$ is the $n^{th}$ higher derivative on the spectral sequence associated to the filtered chain complex $\mathrm{CFK}^{-}(-Y,L)$) is such a further structure, as it is exploited in [29]. In a slightly different direction, in [31] two of the authors of the present paper have shown that the _mapping class group_ ${\rm{MCG}}(Y,L)={\rm{Diff}}^{+}(Y,L)/{\rm{Diff}}^{+}_{0}(Y,L)$ of the complement of the knot admits a natural action on $\mathrm{HFK}^{-}(-Y,L)$, and it is not hard to see that the Legendrian invariant can be defined as an orbit of the action of ${\rm{MCG}}(Y,L)$ rather than of the full automorphism group. In many cases the mapping class group of a knot is relatively small, hence this refinement provides a significant sharpening of the invariant — without even understanding the explicit action of the mapping class group on the knot Floer homologies. In some cases, however, the mapping class group can be large (although might admit a fairly trivial representation on the Floer homology), in which case the description of the action cannot be avoided. For example, connected sums of knots have infinite mapping class groups. For connected sums, therefore, we describe a slightly different refinement of the invariant. For the sake of simplicity, we will be content with formulating the relevant results for the $\widehat{\mathrm{HFK}}$–theory only. ###### Definition 8.3. The $\mathbb{F}$–vector space $M=\oplus_{s}M_{*}(s)$ is _Alexander bigraded_ if it admits a splitting $M=\oplus_{s}(\oplus_{\\{s_{1},s_{2}\mid s_{1}+s_{2}=s\\}}M_{*}(s_{1},s_{2})).$ The pair $(M,m)$ of an Alexander bigraded vector space $M$ and an element $m\in M$ defines an equivalence class $[[M,m]]$ of such objects by saying that $(M,m)$ and $(N,n)$ are equivalent if there is an isomorphism $f\colon M\to N$ with the property that $f(m)=n$ and $f(M_{*}(s_{1},s_{2}))=N(s_{1},s_{2})$. In other words, we require the isomorphisms to respect the Alexander bigrading of the vector space. For a null–homologous Legendrian knot $L$ in the contact 3–manifold $(Y,\xi)$ consider the homology class $\alpha_{\widehat{\mathfrak{L}}}(L)\in\widehat{\mathrm{HFK}}(-Y,L,\mathbf{t}_{\xi})$ given by the cycle ${\bf{x}}(B,\varphi,A)$ as in Definition 3.4. By slightly modifying the definition of Theorem 1.1, consider ${\widetilde{{\mathfrak{L}}}}(L):=[[\widehat{\mathrm{HFK}}(-Y,L,\mathbf{t}_{\xi}),\alpha_{\widehat{\mathfrak{L}}}(L)]].$ ###### Theorem 8.4. Suppose that $L$ is a given null–homologous Legendrian knot in the contact 3–manifold $(Y,\xi)$ representing the knot type ${\mathcal{K}}$, which is the connected sum of two knot types ${\mathcal{K}}_{1}$ and ${\mathcal{K}}_{2}$. Then for any knot $K$ of type ${\mathcal{K}}$ the knot Floer homology $\widehat{\mathrm{HFK}}(-Y,K)$ naturally admits an Alexander bigrading and ${\widetilde{{\mathfrak{L}}}}(L)$ is an invariant of its oriented Legendrian isotopy class. In particular, if $L,L^{\prime}$ are two Legendrian representatives of ${\mathcal{K}}$ and ${\widetilde{{\mathfrak{L}}}}(L)\neq{\widetilde{{\mathfrak{L}}}}(L^{\prime})$ then $L$ and $L^{\prime}$ are not Legendrian isotopic. ###### Proof. The Künneth formula [35] provides an Alexander bigrading for the knot Floer homology of $K=K_{1}\\#K_{2}$ by using the Alexander gradings of the knot Floer groups of $K_{1}$ and $K_{2}$. Theorem 9.1 now shows that the map $f_{1}$ appearing in the proof of Corollary 3.6 induces an isomorphism on the knot Floer homology which respects this Alexander bigrading. Therefore the proof of the corollary, in fact, shows that for connected sums the refined equivalence class ${\widetilde{{\mathfrak{L}}}}(L)$ is an invariant of the Legendrian isotopy class of $L$. ∎ ###### Proof of Theorem 8.1. As we saw above, the Legendrian knots $L_{1}$ and $L_{2}$ both represent the knot type of ${\mathcal{K}}$, and have equal Thurston–Bennequin invariants and rotation numbers. Now $\widehat{\mathrm{HFK}}(S^{3},m(K),5)=\widehat{\mathrm{HFK}}(S^{3},T_{(2,7)},3)\otimes\widehat{\mathrm{HFK}}(S^{3},T_{(2,9)},2)\oplus$ $\widehat{\mathrm{HFK}}(S^{3},T_{(2,7)},2)\otimes\widehat{\mathrm{HFK}}(S^{3},T_{(2,9)},3)\oplus\widehat{\mathrm{HFK}}(S^{3},T_{(2,7)},1)\otimes\widehat{\mathrm{HFK}}(S^{3},T_{(2,9)},4),$ and the cycle which determines $\widehat{\mathfrak{L}}(L_{1})$ is in the first summand, while the cycle which determines $\widehat{\mathfrak{L}}(L_{2})$ is in the last. (The image of the quotient map from $\mathrm{HFK}^{-}$ to $\widehat{\mathrm{HFK}}$, which contains ${\widehat{\mathfrak{L}}}$, is disjoint from the middle summand.) In other words, the Alexander bigrading of $\widehat{\mathfrak{L}}(L_{1})$ is $(3,2)$, while for $\widehat{\mathfrak{L}}(L_{2})$ this quantity is $(1,4)$. This readily implies that the refined invariants ${\widetilde{{\mathfrak{L}}}}(L_{1})$ and ${\widetilde{{\mathfrak{L}}}}(L_{2})$ are distinct, and so by Theorem 8.4 the knots $L_{1}$ and $L_{2}$ are not Legendrian isotopic. ∎ ###### Proof of Theorem 1.7. Let $K$ denote the connected sum $T_{(2,-7)}\\#T_{(2,-9)}$. Consider the positive transverse push–offs $T_{1}$ and $T_{2}$ of the Legendrian knots $L_{1}$ and $L_{2}$. By defining the refined transverse invariant ${\tilde{\mathfrak{T}}}(T_{i})$ as ${\tilde{{\mathfrak{L}}}}(L_{i})$, the proof of Theorem 8.1 implies that $T_{1}$ and $T_{2}$ are transversely nonisotopic. Since their self–linking numbers $s\ell(T_{i})$ can be easily computed from the Thurston–Bennequin and rotation numbers of the Legendrian knots $L_{i}$, we get that $s\ell(T_{1})=s\ell(T_{2})$. Therefore $K$ as defined above is a transversely non–simple knot type in the overtwisted contact structure $\xi_{12}$, concluding the proof. ∎ Notice that if the nontriviality of the Legendrian invariants of all Legendrian knots $L_{i,j}$ can be established, the argument used in the proof of Theorem 1.7 actually provides arbitrarily many transversely distinct transverse knots with the same classical invariants: consider the connected sums $L_{i,j}\\#L_{k,l}$ with $i+j=n-2$ and $k+l=m-2\geq n-2$ fixed (hence fixing the knot type and the Thurston–Bennequin numbers of the knots) and take only those knots which satisfy $i+k=n-2$. It is not hard to see that these knots will have the same rotation numbers, but the argument given in the proof of Theorem 1.7, assuming that the invariants for the $L_{i,j}$ are nontrivial, shows that the transverse invariants of their positive transverse push–offs do not agree. This would be a way of constructing arbitrarily many distinct transverse knots with the same classical invariants in some overtwisted contact $S^{3}$. ###### Proof of Corollary 1.8.. Consider the overtwisted contact structure $\zeta$ on $Y$ with $\mathbf{t}_{\zeta}=\mathbf{t}_{\xi}$ and write it as $\zeta^{\prime}\\#\xi_{12}$, where $\zeta^{\prime}$ is an overtwisted contact structure on $Y$ with $\mathbf{t}_{\zeta^{\prime}}=\mathbf{t}_{\xi}$. (Again, by simple homotopy theoretic reasons and the classification of overtwisted contact structures, the above decomposition is possible.) The connected sum of the Legendrian knot $L$ in $\zeta^{\prime}$ having nontrivial invariant ${\tilde{{\mathfrak{L}}}}(L)$ with $L_{1}$ and $L_{2}$ of Theorem 1.7 obviously gives a pair of Legendrian knots with equal classical invariants in $\zeta$, but with the property that their transverse push–offs have distinct transverse invariants. The distinction of these elements relies on a straightforward modification of Theorem 9.1, where the components in a fixed connected sum decomposition are not necessarily prime knots, but we fix the isotopy class of the embedded sphere separating the knot into two connected components. This construction concludes the proof. ∎ ## 9 Appendix: On knot Floer homology of connected sums Recall from [35] that if $(Y_{1},K_{1})$ and $(Y_{2},K_{2})$ are two three–manifolds equipped with knots, then we can form their connected sum $(Y_{1}\\#Y_{2},K_{1}\\#K_{2})$. In this case, the knot Floer homology for the connected sum can be determined by the knot Floer homology of its summands by a Künneth formula. Our aim here is to investigate naturality properties of this decomposition. Specifically, we prove the following ###### Theorem 9.1. Let $K$ be a knot obtained as a connected sum of two distinct prime knots $K_{1}$ and $K_{2}$. Then there is an additional intrinsic grading on the knot Floer homology of $K$ with the property that $\widehat{\mathrm{HFK}}(K,s)=\bigoplus_{\\{s_{1},s_{2}\big{|}s_{1}+s_{2}=s\\}}\widehat{\mathrm{HFK}}(K,s_{1},s_{2}),$ where $\widehat{\mathrm{HFK}}(K,s_{1},s_{2})=\widehat{\mathrm{HFK}}(K_{1},s_{1})\otimes\widehat{\mathrm{HFK}}(K_{2},s_{2})$. More precisely, if $D(K_{1})$ and $D(K_{2})$ are diagrams for $K_{1}$ and $K_{2}$ respectively, then there is an induced diagram $D(K_{1})\\#D(K_{2})$ for $K_{1}\\#K_{2}$, together with an isomorphism $\Phi\colon H_{*}(\widehat{\mathrm{CFK}}(D(K_{1})))\otimes H_{*}(\widehat{\mathrm{CFK}}(D(K_{2})))\longrightarrow H_{*}(\widehat{\mathrm{CFK}}(D(K_{1})\\#D(K_{2}))).$ If $D^{\prime}(K_{1})$ and $D^{\prime}(K_{2})$ are a pair of different diagrams for $K_{1}$ and $K_{2}$, and $D^{\prime}(K_{1})\\#D^{\prime}(K_{2})$ denotes the induced sum of diagrams for $K_{1}\\#K_{2}$, then there is an isomorphism $\Psi\colon H_{*}(\widehat{\mathrm{CFK}}(D(K_{1})\\#D(K_{2})))\longrightarrow H_{*}(\widehat{\mathrm{CFK}}(D^{\prime}(K_{1})\\#D^{\prime}(K_{2})))$ such that the diagram $\begin{CD}H_{*}(\widehat{\mathrm{CFK}}(D(K_{1})))\otimes H_{*}(\widehat{\mathrm{CFK}}(D(K_{2})))@>{\Phi}>{}>H_{*}(\widehat{\mathrm{CFK}}(D(K_{1})\\#D(K_{2})))\\\ @V{\psi_{1}\otimes\psi_{2}}V{}V@V{\Psi}V{}V\\\ H_{*}(\widehat{\mathrm{CFK}}(D^{\prime}(K_{1})))\otimes H_{*}(\widehat{\mathrm{CFK}}(D^{\prime}(K_{2})))@>{\Phi^{\prime}}>{}>H_{*}(\widehat{\mathrm{CFK}}(D^{\prime}(K_{1})\\#D^{\prime}(K_{2})))\end{CD}$ (9.1) commutes. Let $Y$ be a closed, oriented three–manifold, $K\subset Y$ be an oriented knot, and $S\subset Y$ meeting $K$ transversely in exactly two points. Call $S$ a splitting sphere for $K$. A splitting sphere expresses $K$ as a connected sum of two knots $K_{1}$ and $K_{2}$. In the case where one of the two summands is unknotted, we call the splitting sphere trivial. In this language, a knot $K$ is prime if every splitting sphere for $K$ is trivial. Recall the following classical result [22]. ###### Theorem 9.2. If $K$ is the connected sum of two prime knots $K_{1}$ and $K_{2}$, then there is, up to isotopy, a unique nontrivial splitting sphere $S$ for $K$. ∎ Let $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$ be a doubly-pointed Heegaard diagram equipped with a curve $\gamma\subset\Sigma$ which is disjoint from all $\alpha_{i}$, $\beta_{j}$, $w$, and $z$. Suppose moreover that $\gamma$ is a separating curve, dividing $\Sigma$ into two components $F_{1}$ and $F_{2}$, so that $w\in F_{1}$ and $z\in F_{2}$. This decorated Heegaard diagram determines the following data: * • a three–manifold $Y$ (gotten from the Heegaard diagram) * • an oriented knot $K\subset Y$ (determined by $w$ and $z$) * • an embedded two–sphere $S$ meeting the Heegaard surface along $\gamma$ (consisting of all Morse flows between index zero and index three critical points passing through $\gamma$); i.e. a decorated Heegaard diagram determines a knot $K$ in $Y$ together with a splitting sphere $S$. We call $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$ a decorated Heegaard diagram compatible with $(Y,K,S)$. ###### Definition 9.3. A decorated Heegaard move on $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$ is a move of one of the following types: * • Isotopies. Isotopies of $\mathbb{\alpha}$ and $\mathbb{\beta}$, preserving the conditions that all curves among the $\mathbb{\alpha}$ and $\mathbb{\beta}$ are disjoint, and disjoint from $w$, $z$, and $\gamma$; isotopy of $\gamma$, preserving the condition that it is disjoint from $\mathbb{\alpha}$, $\mathbb{\beta}$, $w$, and $z$. * • Handleslides. Handleslides among the $\mathbb{\alpha}$ or $\mathbb{\beta}$, supported in the complement of $w$, $z$, and away from $\gamma$. * • Stabilizations/destabilizations. Stabilization is obtained by forming the connected sum of $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$ with a genus one surface equipped with a pair of curves $\alpha_{g+1}$ and $\beta_{g+1}$ meeting transversely in a single point; destabilization is the inverse operation. ###### Proposition 9.4. Suppose that $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$ and $(\Sigma^{\prime},\mathbb{\alpha}^{\prime},\mathbb{\beta}^{\prime},w^{\prime},z^{\prime},\gamma^{\prime})$ are two decorated Heegaard diagrams compatible with $(Y,K,S)$ and $(Y^{\prime},K^{\prime},S^{\prime})$ respectively. If $(Y,K,S)$ is diffeomorphic to $(Y^{\prime},K^{\prime},S^{\prime})$, then the decorated Heegaard diagram $(\Sigma^{\prime},\mathbb{\alpha}^{\prime},\mathbb{\beta}^{\prime},w^{\prime},z^{\prime},\gamma^{\prime})$ is diffeomorphic to one obtained from $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$ after a finite sequence of decorated Heegaard moves. ###### Proof. Fix a Morse function $f_{0}$ compatible with $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$. Let $f_{1}$ be a different Morse function compatible with $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$, which agrees with $f_{0}$ in a neighborhood of $S\cup K$. Then we can connect them by a generic one-parameter family $f_{t}$, wherein they undergo isotopies, handleslides and stabilizations/destabilizations. Since all functions have the prescribed behavior at $w$, $z$, and $\gamma$, the Heegaard moves will be supported away from $w$, $z$, and $\gamma$. Changing $f$ near $S\cup K$ has the effect of isotopies of $\Sigma$ supported near $w$, $z$, and $\gamma$. ∎ ###### Proof of Theorem 9.1. Let $(\Sigma,\mathbb{\alpha},\mathbb{\beta},w,z,\gamma)$ be a Heegaard diagram for $K=K_{1}\\#K_{2}$. Then, our additional grading is defined (up to an additive constant) as follows. Fix a point $m$ near $\gamma$. We define $\mathfrak{s}_{1}^{\prime}({\bf{x}})-\mathfrak{s}_{1}^{\prime}({\bf{y}})=n_{m}(\phi)-n_{w}(\phi)$ where $\phi\in\pi_{2}({\bf{x}},{\bf{y}})$ is any homotopy class; similarly, $\mathfrak{s}_{2}^{\prime}({\bf{x}})-\mathfrak{s}_{2}^{\prime}({\bf{y}})=n_{z}(\phi)-n_{m}(\phi)$. Since the handleslides between compatible decorated projections cannot cross $\gamma$ (or $w$ or $z$), clearly, the triangle maps induced by handleslides preserve these gradings (at least in the relative sense). In fact, if $D(K_{1})$ and $D(K_{2})$ are diagrams for $K_{1}$ and $K_{2}$, we can form the connected sum diagram by connecting $z_{1}\in D(K_{1})$ with $w_{2}\in D(K_{2})$, dropping these two basepoints, and using only $w_{1}$ and $z_{2}$ in $D(K_{1})\\#D(K_{2})$. This is the doubly-pointed Heegaard diagram for the connected sum, and if we draw $\gamma$ around the connected sum annulus, then it is decorated so as to be compatible with the nontrivial sphere $S$ splitting $K$. Indeed, the usual proof of the Künneth principle shows that the map $\widehat{\mathrm{CFK}}(D(K_{1}))\otimes\widehat{\mathrm{CFK}}(D(K_{2}))\longrightarrow\widehat{\mathrm{CFK}}(D(K_{1})\\#D(K_{2}))$ is a quasi-isomorphism. It is also easy to see that $(\mathfrak{s}_{1}({\bf{x}}),\mathfrak{s}_{2}({\bf{x}}))=(\mathfrak{s}_{1}^{\prime}({\bf{x}}),\mathfrak{s}_{2}^{\prime}({\bf{x}})),$ at least up to an overall additive constant. Indeed, we can now define $\mathfrak{s}_{1}^{\prime}$ and $\mathfrak{s}_{2}^{\prime}$ to be normalized so that they agree with $\mathfrak{s}_{1}$ and $\mathfrak{s}_{2}$. Finally, if we have two diagrams which differ by decorated Heegaard moves, then the triangle maps clearly induce isomorphisms required in Equation (9.1). The statement follows at once since $\Psi$ preserves the bigrading (in an absolute sense). ∎ ## References * [1] J. Birman and W. Menasco, A note on transversal knots which are closed 3–braids, arXiv:math/0703669. * [2] V. Colin, Livres Ouverts en Géométrie de Contact [d’après Emmanuel Giroux], Séminaire Bourbaki, 59ème année, 2006–2007, no 969, to appear in Asterisque. * [3] F. Ding and H. Geiges, A Legendrian surgery presentation of contact 3-manifolds, Math. Proc. Cambridge Philos. Soc. 136 (2004) 583–598. * [4] F. Ding, H. Geiges and A. Stipsicz, Surgery diagrams for contact 3–manifolds, Turkish J. Math. 28 (2004) 41–74. * [5] K. Dymara, Legendrian knots in overtwisted contact structures on $S^{3}$, Ann. Global Anal. Geom. 19 (2001) 293–305. * [6] K. Dymara, Legendrian knots in overtwisted contact structures, arXiv:math/0410122. * [7] Y. Eliashberg, Classification of overtwisted contact structures on 3–manifolds, Invent. Math. 98 (1989) 623–637. * [8] Y. Eliashberg and M. Fraser, Topologically trivial Legendrian knots, arXiv:0801.2553 * [9] J. Epstein, D. Fuchs and M. Meyer, Chekanov–Eliashberg invariants and transverse approximations of Legendrian knots, Pacific J. Math. 201 (2001) 89–106. * [10] J. Etnyre, Legendrian and transverse knots, in _Handbook of knot theory_ , pages 105–185. Elsevier B. V., Amsterdam, 2005. * [11] J. Etnyre, Lectures on open book decompositions and contact structures, in _Floer homology, gauge theory, and low-dimensional topology_ , 103–141, Clay Math. Proc. 5, Amer. Math. Soc., Providence, RI, 2006. * [12] J. Etnyre, On contact surgery, Proc. Amer. Math. Soc. 136 (2008) 3355–3362. * [13] J. Etnyre and K. Honda, Knots and contact geometry, I. Torus knots and the figure eight knot, J. Symplectic Geom. 1 (2001) 63–120. * [14] J. Etnyre and K. Honda, On connected sums and Legendrian knots, Adv. Math. 179 (2003) 59–74. * [15] J. Etnyre and K. Honda, Cabling and transverse simplicity, Ann. Math. 162 (2005) 1305–1333. * [16] J. Etnyre and T. Vogel, in preparation * [17] P. Ghiggini, P. Lisca and A. Stipsicz, Classification of tight contact structures on some small Seifert fibered 3–manifolds, Amer. J. Math. 129 (2007) 1403–1447. * [18] E. Giroux, Géométrie de contact: de la dimension trois vers les dimensions supérieures, Proceedings of the International Congress of Mathematicians, Vol. II (Beijing, 2002) 405–414. * [19] K. Honda, W. Kazez and G. Matić, On the contact class in Heegaard Floer homology, arXiv:math.GT/0609734. * [20] K. Honda, W. Kazez and G. Matić, The contact invariant in sutured Floer homology, arXiv:0705.2828 * [21] A. Juhász, Holomorphic disks and sutured manifolds, Algebr. Geom. Topol. 6 (2006) 1429-1457. * [22] R. Lickorish, An introduction to knot theory, Graduate Texts in Math. 175, Springer–Verlag, 1997. * [23] P. Lisca and A. Stipsicz, Seifert fibered contact three–manifolds via surgery, Algebr. Geom. Topol. 4 (2004) 199–217. * [24] P. Lisca and A. Stipsicz, Ozsváth–Szabó invariants and tight contact 3–manifolds, II, J. Differential Geom. 75 (2007) 109–141. * [25] P. Lisca and A. Stipsicz, Ozsváth–Szabó invariants and tight contact 3–manifolds, III, J. Symplectic Geometry 5 (2007) 357–384. * [26] C. Manolescu, P. Ozsváth and S. Sarkar, A combinatorial description of knot Floer homology, arXiv:math.GT/0607691. * [27] C. Manolescu, P. Ozsváth, Z. Szabó and D. Thurston, On combinatorial link Floer homology, Geom. Topol. 11 (2007) 2339–2412. * [28] W. Menasco, On iterated torus knots and transversal knots, Geom. Topol. 5 (2001) 651–682. * [29] L. Ng, P. Ozsváth and D. Thurston, Transverse knots distinguished by knot Floer homology, J. Symplectic Geom. 6 (2008) 461–490. * [30] B. Ozbagci and A. Stipsicz, Surgery on contact 3–manifolds and Stein surfaces, Bolyai Society Mathematical Studies 13, Springer Verlag 2004. * [31] P. Ozsváth and A. Stipsicz, Contact surgeries and the transverse invariant in knot Floer homology, arXiv:0803.1252 * [32] P. Ozsváth and Z. Szabó, Absolutely graded Floer homologies and intersection forms for four-manifolds, Adv. Math. 173 (2003) 179–261. * [33] P. Ozsváth and Z. Szabó, Holomorphic disks and topological invariants of closed three–manifolds, Ann. of Math. 159 (2004) 1159–1245. * [34] P. Ozsváth and Z. Szabó, Heegaard Floer homology and contact structures, Duke Math. J. 129 (2005) 39–61. * [35] P. Ozsváth and Z. Szabó, Holomorphic disks and knot invariants, Adv. Math. 186 (2004) 58–116. * [36] P. Ozsváth and Z. Szabó, Heegaard Floer homology and alternating knots, Geom. Topol. 7 (2003) 225–254. * [37] P. Ozsváth, Z. Szabó and D. Thurston, Legendrian knots, transverse knots and combinatorial Floer homology, Geom. Topol. 12 (2008) 941–980. * [38] O. Plamenevskaya, A combinatorial description of the Heegaard Floer contact invariant, Algebr. Geom. Topol. 7 (2007) 1201–1209. * [39] J. Rasmussen, Floer homology of surgeries on two-bridge knots, Algebr. Geom. Topol. 2 (2002) 757–789. * [40] J. Rasmussen, Floer homology and knot complements, Ph.D Thesis, Harvard (2003), arXiv:math/0306378. * [41] J. Stallings, Construction of fibered knots and links, Algebraic and Geometric Topology (Stanford 1976), Part 2, 55 60, Proc. Sympos. Pure Math. 32, Amer. Math. Soc. 1978. * [42] A. Stipsicz and V. Vértesi, On invariants for Legendrian knots, Pacific J. Math. 239 (2009) 157–177. * [43] W. Thurston and H. Winkelnkemper, On the existence of contact forms, Proc. Amer. Math. Soc. 52 (1975) 345–347. * [44] V. Vértesi, Transversely non simple knots, Algebr. Geom. Topol. 8 (2008) 1481–1498.

arxiv-papers

2024-09-04T02:48:53.574430

0802.0669

# Deeply Virtual Compton Scattering on nucleons and nuclei in generalized vector meson dominance model K. Goeke Klaus.Goeke@tp2.rub.de Institut für Theoretische Physik II, Ruhr- Universität-Bochum, D-44780 Bochum, Germany V. Guzey vguzey@jlab.org Theory Center, Jefferson Lab, Newport News, VA 23606, USA M. Siddikov Marat.Siddikov@tp2.rub.de Institut für Theoretische Physik II, Ruhr- Universität-Bochum, D-44780 Bochum, Germany Theoretical Physics Dept,Uzbekistan National University, Tashkent 700174, Uzbekistan ###### Abstract We consider Deeply Virtual Compton Scattering (DVCS) on nucleons and nuclei in the framework of generalized vector meson dominance (GVMD) model. We demonstrate that the GVMD model provides a good description of the HERA data on the dependence of the proton DVCS cross section on $Q^{2}$, $W$ (at $Q^{2}=4$ GeV2) and $t$. At $Q^{2}=8$ GeV2, the soft $W$-behavior of the GVMD model somewhat underestimates the $W$-dependence of the DVCS cross section due to the hard contribution not present in the GVMD model. We estimate $1/Q^{2}$ power-suppressed corrections to the DVCS amplitude and the DVCS cross section and find them large. We also make predictions for the nuclear DVCS amplitude and cross section in the kinematics of the future Electron-Ion Collider. We predict significant nuclear shadowing, which matches well predictions of the leading-twist nuclear shadowing in DIS on nuclei. DVCS, nuclear DVCS, generalized vector dominance ###### pacs: 12.40.Vv,13.60.Hb,25.30.Rw ††preprint: JLAB-THY-08-774 ## I Introduction During the last decade, one of main focuses of hadronic physics has been the study of the hadronic structure using hard exclusive reactions, such as Deeply Virtual Compton Scattering (DVCS), $\gamma^{\ast}N\to\gamma N$, and hard exclusive meson production (HEMP), $\gamma^{\ast}N\to MN$. These processes have been the subject of intensive theoretical and experimental investigations Mueller:1998fv ; Ji:1996nm ; Ji:1998pc ; Radyushkin:1996nd ; Radyushkin:1997ki ; Radyushkin:2000uy ; Collins:1998be ; Collins:1996fb ; Brodsky:1994kf ; Goeke:2001tz ; Diehl:2000xz ; Belitsky:2001ns ; Diehl:2003ny ; Belitsky:2005qn . In addition, there were investigated ”inverse” hard exclusive reactions such as $\gamma N\to\gamma^{\ast}N\to l^{+}l^{-}N$ Berger:2001xd and $\pi N\to\gamma^{\ast}N\to l^{+}l^{-}N$ Berger:2001zn , and ”$u$-channel” reactions such as $\gamma^{\ast}\gamma\to\pi\pi$ Diehl:2000uv . The interest to the DVCS and HEMP reactions is motivated by the fact that in the Bjorken limit (large $Q^{2}$), the corresponding amplitudes factorize Collins:1998be ; Collins:1996fb in convolution of perturbative (hard) coefficient functions with nonperturbative (soft) matrix elements, which are parameterized in terms of generalized parton distributions (GPDs). GPDs are universal (process-independent) functions that contain information on parton distributions, form factors and correlations in hadrons. GPDs also parameterize parton correlations in matrix elements describing transitions between two different hadrons, which appear in reactions such as e.g. $\gamma^{\ast}p\to\pi^{+}n$ Mankiewicz:1998kg ; Frankfurt:1999fp . While the description of DVCS and HEMP based on the factorization approach is most general, in experiments the values of the virtualities $Q^{2}$ are below the range required for the validity of the factorization theorem Airapetian:2001yk ; Airapetian:2006zr ; Ye:2006gza ; Stepanyan:2001sm ; Munoz Camacho:2006hx ; Girod:2007jq . Hence, contributions of higher-twist effects might be substantial (it is an open issue how large these effects are), which will affect the extraction of GPDs from the data. Therefore, it is important to have an effective model for the DVCS and HEMP amplitudes, which would interpolate between the photoproduction ($Q^{2}\approx 0$) and deep inelastic ($Q^{2}\sim{\cal O}({10})$ GeV2) regimes. In this paper, using the generalized vector meson dominance (GVMD) model Fraas:1974gh ; Ditsas:1975vd ; Shaw:1993gx , which is consistent with perturbative QCD at small transverse distances Frankfurt:1997zk , we derive expressions for the amplitudes of DVCS on nucleons and nuclei, which are valid at high energies and which are applicable over a wide range of $Q^{2}$. We show that the resulting cross section of DVCS on nucleons compares well to the HERA data Chekanov:2003ya ; Aktas:2005ty . In particular, the dependence of the DVCS cross section on $Q^{2}$, $W$ (at $Q^{2}$=4 GeV2) and $t$ are reproduced rather well; the $W$-dependence of the cross section at $Q^{2}$=8 GeV2 is somewhat underestimated, which can be interpreted as due to the onset of the hard regime beyond the soft dynamics of the GVMD model. We also estimate the relative contribution of $1/Q^{2}$-corrections, which correspond to the higher-twist corrections in perturbative QCD Kivel:2000fg ; Freund:2003qs ; Radyushkin:2000ap . We show that these corrections are large: the contribution of the $1/Q^{2}$-corrections to the DVCS amplitude at $t=t_{{\min}}$ is 20% at $Q^{2}=2$ GeV2, 11% at $Q^{2}=4$ GeV2 and 6% at $Q^{2}=8$ GeV2; the contribution of the $1/Q^{2}$-corrections to the $t$-integrated DVCS cross section is 56% at $Q^{2}=2$ GeV2, 32% at $Q^{2}=4$ GeV2 and 17% at $Q^{2}=8$ GeV2. We also make predictions for the DVCS cross section on nuclear targets, which are relevant for the physics program of the future Electron-Ion Collider. We predict significant nuclear shadowing, which matches well predictions of the leading-twist nuclear shadowing in DIS on nuclei Frankfurt:2003zd . The hypothesis of vector meson dominance (VMD) Feynman assumes a definite relation between the amplitude of the photon (real or virtual)-hadron interaction, ${\cal A}(\gamma_{{\rm tr}}^{\ast}+T\to\dots)$, and a linear combination of the amplitudes of the corresponding strong production by transversely polarized vector mesons, ${\cal A}(V_{{\rm tr}}+T\to\dots)$, ${\cal A}(\gamma_{{\rm tr}}^{\ast}+T\to\dots)=\sum_{V=\rho,\omega,\phi}\frac{e}{f_{V}}\frac{m_{V}^{2}}{m_{V}^{2}+Q^{2}}{\cal A}(V_{{\rm tr}}+T\to\dots)\,,$ (1) where $f_{V}$ is the coupling constant determined from the $V\to e^{+}e^{-}$ decay; $m_{V}$ is the vector meson mass; $Q^{2}$ is the virtuality of the photon; $T$ denotes any hadronic target. Note that Eq. (1) is written for the transversely polarized photons. In Eq. (1), we took into account only the contribution of the $\rho^{0}$, $\omega$ and $\phi$ mesons. The VMD model and its generalizations explain a large wealth of data on the real and virtual ($Q^{2}<1$ GeV2) photon-hadron scattering, which include the pion electric form factor, total cross sections of photon-nucleon and photon- nucleus scattering (inclusive structure functions), exclusive production of vector mesons on nucleons and nuclei, exclusive production of pseudoscalar mesons, for a review, see Bauer:1977iq . As the virtuality of the photon increases, $Q^{2}>1$ GeV2, the simple VMD model, see Eq. (1), becomes inadequate since it leads to the violation of the approximate Bjorken scaling. In order to restore the approximate Bjorken scaling, the simple VMD model can be generalized Fujikawa:1972ux . This can be done using the model-independent method of mass-dispersion representation for the virtual photon-hadron scattering amplitude Gribov:1968gs . In order to illustrate the approach, let us consider the forward virtual photon-hadron scattering amplitude. The dispersion representation for the imaginary part of ${\cal A}(\gamma^{\ast}_{{\rm tr}}+T\to\gamma^{\ast}_{{\rm tr}}+T)$ reads $\Im m\,{\cal A}(\gamma_{{\rm tr}}^{\ast}+T\to\gamma_{{\rm tr}}^{\ast}+T)_{|t=0}=\int\frac{dM^{2}M^{2}}{M^{2}+Q^{2}}\frac{dM^{\prime 2}M^{\prime 2}}{M^{\prime 2}+Q^{2}}\frac{e}{f_{V}}\sigma_{VV^{\prime}}\frac{e}{f_{V^{\prime}}}\,,$ (2) where $\sigma_{VV^{\prime}}$ is the $V+T\to V^{\prime}+T$ scattering cross section (spectral function) which weakly depends on the masses $M$ and $M^{\prime}$. The main idea of the GVMD model Fraas:1974gh ; Ditsas:1975vd ; Shaw:1993gx is to approximate Eq. (2) by an infinite series of (ficticious) vector mesons of ever increasing mass, allowing for both diagonal ($V+T\to V+T$) and non-diagonal ($V+T\to V^{\prime}+T$) transitions. The role of the non-diagonal transitions is to partially cancel the diagonal transitions so that, effectively, $\sigma_{VV^{\prime}}\propto 1/M^{2}$ for large $M^{2}$. This softens the spectral function and leads to the approximate Bjorken scaling of the inclusive structure function $F_{2}(x,Q^{2})$, see Frankfurt:1997zk for the discussion. In the language of the color dipole model, the fact that $\sigma_{VV^{\prime}}\propto 1/M^{2}$ for large $M^{2}$ means that besides dipoles of large transverse sizes, the virtual photon also contains small transverse-size dipoles. The latter fact is called color transparency. One should note that, while the simple vector meson dominance model fails to reproduce the approximate scaling of the inclusive structure function $F_{2}(x,Q^{2})$ (see above), the simple VMD model predicts the correct $Q^{2}$-behavior of cross sections of exclusive reactions, such as $\gamma^{\ast}p\to\pi^{+}n$ Fraas:1971hk ; Dar:1971wh . This is also true for DVCS: Even the simple VMD model provides the correct $Q^{2}$-behavior of the DVCS cross section (up to logarithmic corrections). The structure of this paper is the following. In Sect. II, we explain main assumptions of the GVMD model, which we further generalize to take into account a non-zero momentum transfer $t\neq 0$. We derive the expression for the amplitude of DVCS on the nucleon and make predictions for the DVCS cross section. We demonstrate that the GVMD model provides a good description of the HERA data on the $W$, $Q^{2}$ and $t$-dependence of the cross section of DVCS on the proton Chekanov:2003ya ; Aktas:2005ty . In this section, we also estimate $1/Q^{2}$-corrections to the DVCS amplitude and the DVCS cross section. Predictions for the nuclear DVCS amplitude and for the cross section of DVCS on nuclei in the collider kinematics are presented in Sect. III. In Sect. IV, we summarize and discuss our results. ## II DVCS on the nucleon In this section, we extend the generalized vector meson dominance (GVMD) model Fraas:1974gh ; Ditsas:1975vd ; Shaw:1993gx to the off-forward case and apply it to Deeply Virtual Compton Scattering (DVCS) on the nucleon. ### II.1 DVCS amplitude The GVMD model assumes that the virtual (real) photon interacts with the hadronic target by fluctuating into a coherent and infinite sum of ficticious vector mesons $V_{n}$. Then, the DVCS amplitude at the photon level, ${\cal A}(\gamma^{\ast}p\to\gamma p)$, can be graphically presented as depicted in Fig. 1. Figure 1: The DVCS amplitude in the GVMD model, see Eq. (3). In the GVMD model, the DVCS amplitude for transversely polarized virtual photons reads ${\cal A}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)=\sum_{n,n^{\prime}=0}^{\infty}\frac{e}{f_{n}}\frac{M_{n}^{2}}{M_{n}^{2}+Q^{2}}\Sigma_{n,n^{\prime}}(W,t)\frac{e}{f_{n^{\prime}}}\,,$ (3) where $W^{2}=(p_{\gamma^{\ast}}+p)^{2}$ with $p_{\gamma^{\ast}}$ the momentum of the initial photon and $p$ the momentum of the initial proton; $t=(p^{\prime}-p)^{2}$ with $p^{\prime}$ the momentum of the final proton. The masses $M_{n}$ and the coupling constants $f_{n}$ are connected by the following relations, $\frac{M_{n}^{2}}{M_{0}^{2}}=\frac{f_{n}^{2}}{f_{0}^{2}}=(1+2n)\,,$ (4) where $M_{0}=m_{\rho}$ and $f_{0}=f_{\rho}$ refer to the physical $\rho^{0}$ meson. Note that the relation of the vector mesons $V_{n}$ conventionally used in the GVMD model to physical $J^{P}=1^{-}$ vector mesons found in the Review of particle physics Yao:2006px is not direct. The motivation for this is that while the vector meson masses are known with a reasonable accuracy up to $M_{\rho}\gtrsim 2$ GeV, there is no accurate data on the partial decay width $\Gamma_{e^{+}e^{-}}$ for mesons heavier than $\rho(1450)$. On the other hand, the parameterization (4) provides reasonable results for physical observables. One can check that the linear $n$-dependence of the ratio $M_{n}^{2}/M_{0}^{2}$ in Eq. (4) is confirmed experimentally for large-$n$ Yao:2006px . However, the slope of the $n$-dependence is underestimated by approximately a factor of two. It is important to point out that, at high energies, the DVCS cross section at the photon level is dominated by the contribution of the transversely polarized virtual photons due to the helicity conservation Aktas:2005ty . In the language of the color dipole model, this dominance is explained by the dominance of the large transverse-size dipoles over the small transverse-size dipoles, see Diehl:2003ny for the discussion. Therefore, Eq. (3) gives the complete description of the DVCS amplitude. The quantity $\Sigma_{n,n^{\prime}}(t)$ is the $V_{n,{\rm tr}}+p\to V_{n^{\prime},{\rm tr}}+p$ scattering amplitude, see Fig. 1. The matrix $\Sigma_{n,n^{\prime}}(t)$ is assumed to have a tri-diagonal form with the following non-zero elements, $\displaystyle\Sigma_{n,n}(W,t)=i\sigma_{\rho p}(W^{2})(1-i\eta)F_{n}(t)\,,$ $\displaystyle\Sigma_{n,n+1}(W,t)=\Sigma_{n+1,n}(W,t)=-\frac{1}{2}\frac{M_{n}}{M_{n+1}}\left(1-2\,\delta\frac{m^{2}_{\rho}}{M_{n}^{2}}\right)\Sigma_{n,n}(W,t)\,,$ (5) where $\sigma_{\rho p}$ is the $\rho$ meson-proton cross section for the transversely polarized meson; $\eta$ is the ratio of the real to imaginary parts of the $\rho$ meson-proton scattering amplitude; $\delta=0.2$ is the parameter of the model. The function $F_{n}(t)$ models the $t$-dependence of $\Sigma_{n,n^{\prime}}(t)$, which goes beyond the original formulation of the GVMD model Fraas:1974gh ; Ditsas:1975vd ; Shaw:1993gx , which addressed only the forward $t=0$ limit. In our analysis, we use the following form of $F_{n}(t)$, $F_{n}(t)=\exp\left(-\frac{1}{2}\left[\frac{1}{n+1}B_{1}+\frac{n}{n+1}B_{2}\right]|t|\right)\,,$ (6) where $B_{1}=11$ GeV-2 and $B_{2}=4.3$ GeV-2. The choice of the slopes $B_{1}$ and $B_{2}$ is motivated as follows. For the moment, let us replace the final real photon by the $\rho$ meson. In the photoproduction limit, the $\gamma p\to\rho p$ cross section measured at HERA by the H1 collaboration was fitted to the form $\exp(-B|t|)$ with the slope $B=(10.9\pm 2.4\pm 1.1)$ GeV-2 Aid:1996bs . The ZEUS measurement gives essentially the same value of $B$ Breitweg:1997ed . In electroproduction, the slope of the exponential fit to the $\gamma^{\ast}p\to\rho p$ cross section is much smaller than in photoproduction: It decreases from $B=(8.0\pm 0.5\pm 0.6)$ GeV-2 at $Q^{2}=1.8$ GeV2 to $B=(4.7\pm 1.0\pm 0.7)$ GeV-2 at $Q^{2}=21.2$ GeV2 Adloff:1999kg . This decrease of the slope of the $t$-dependence with increasing $Q^{2}$ is effectively parameterized by Eq. (6) as a decrease of the slope with the increasing number of the vector meson $n$. Indeed, close to the photoproduction limit, the dominant contribution to the sum in Eq. (3) comes from the $n=0$ term. In the opposite limit of large $Q^{2}$, terms with large $n$, up to $M_{n}^{2}\sim Q^{2}$, contribute to the sum. Choosing $Q^{2}=0$ and $Q^{2}=21.2$ GeV2 as reference points, we determine the values of the slopes $B_{1}$ and $B_{2}$ in Eq. (6). It is clear from the above discussion that while the value of the slope $B_{1}$ in Eq. (6) is model-independent, the value of the slope $B_{2}$ is somewhat more uncertain. We have chosen not to introduce $Q^{2}$-dependent slopes $B_{1}$ and $B_{2}$ since this would contradict the spirit of the VMD model: The $W$ and $t$-dependence of the DVCS amplitude is determined solely by the vector meson-proton scattering amplitudes; the vector meson propagators provide the $Q^{2}$-dependence. It is important to note that the non-diagonal terms, $\Sigma_{n,n^{\prime}}$ with $n\neq n^{\prime}$, are essential in the GVMD model: The infinite series in Eq. (3) would have been divergent without the non-diagonal transitions. Also, the non-diagonal terms provide the correct scaling of the total $\gamma^{\ast}p$ cross section. Using Eqs. (4) and (5), the DVCS amplitude in Eq. (3) can be written in the following form $\displaystyle{\cal A}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)$ $\displaystyle=$ $\displaystyle i\frac{2(1+\delta)e^{2}}{f_{\rho}^{2}}\sigma_{\rho p}(W^{2})(1-i\eta)\sum_{n=0}^{\infty}\frac{F_{n}(t)}{(\frac{Q^{2}}{m_{\rho}^{2}}+1+2n)(\frac{Q^{2}}{m_{\rho}^{2}}+3+2n)}$ (7) $\displaystyle\times$ $\displaystyle\left[1+\frac{Q^{2}}{2m_{\rho}^{2}(1+\delta)(3+2n)}\left(1+4\delta\frac{1+n}{(1+2n)}\right)\right]\,.$ Equation (7) involves four quantities, $f_{\rho}$, $\sigma_{\rho p}$, $\eta$ and $\delta$, which are known with a certain degree of uncertainty. One can reduce this uncertainty by expressing the DVCS amplitude in terms of the total $\gamma p$ cross section, $\displaystyle\sigma_{{\rm tot}}^{\gamma p}(W^{2})$ $\displaystyle=$ $\displaystyle\Im m\,{\cal A}(\gamma+p\to\gamma+p)_{|t=0}=\frac{2(1+\delta)e^{2}}{f_{\rho}^{2}}\sigma_{\rho p}(W^{2})\sum_{n=0}^{\infty}\frac{1}{(1+2n)(3+2n)}$ (8) $\displaystyle=$ $\displaystyle\frac{(1+\delta)e^{2}}{f_{\rho}^{2}}\sigma_{\rho p}(W^{2})\,.$ Therefore, the final expression for the DVCS amplitude reads $\displaystyle{\cal A}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)$ $\displaystyle=$ $\displaystyle i2\,\sigma_{{\rm tot}}^{\gamma p}(W^{2})(1-i\eta)\sum_{n=0}^{\infty}\frac{F_{n}(t)}{(\frac{Q^{2}}{m_{\rho}^{2}}+1+2n)(\frac{Q^{2}}{m_{\rho}^{2}}+3+2n)}$ (9) $\displaystyle\times$ $\displaystyle\left[1+\frac{Q^{2}}{2m_{\rho}^{2}(1+\delta)(3+2n)}\left(1+4\delta\frac{1+n}{(1+2n)}\right)\right]\,.$ One should also note that another advantage of expressing the DVCS amplitude in terms of $\sigma_{{\rm tot}}^{\gamma p}$ is that Eq. (9) effectively takes into account the contributions of the $\omega$ and $\phi$ vector mesons, which enter through the phenomenological parameterization of $\sigma_{{\rm tot}}^{\gamma p}$. In our analysis, we use the ZEUS parameterization of $\sigma_{{\rm tot}}^{\gamma p}(W^{2})$ Chekanov:2001gw $\sigma_{{\rm tot}}^{\gamma p}(W^{2})=57\,W^{0.2}+121\,W^{-0.716}\,,$ (10) where the cross section is in $\mu$b and $W$ is in GeV. The ratio of the real to imaginary parts of the $V+p\to V+p$ scattering amplitude, $\eta$, is found using the Gribov-Migdal relation Gribov:1968uy , $\eta\approx\frac{\pi}{2}\,\frac{p}{2}\approx 0.16\,,$ (11) where $p=0.2$ was used, which corresponds to the power of the $W$-dependence of $\sigma_{{\rm tot}}^{\gamma p}(W^{2})$ at large $W$ in Eq. (10). The remaining parameter in Eq. (9) is $\delta$, for which we use $\delta=0.2$ Fraas:1974gh ; Ditsas:1975vd ; Shaw:1993gx ; Pautz:1997eh . However, the exact numerical value of $\delta$ affects weakly our numerical predictions. One of simplest DVCS observables is the skewing factor $R$, which is defined as the ratio of the DVCS to the DIS amplitudes and which was recently extracted from the HERA DVCS and DIS data Schoeffel:2007dt , $R(t)\equiv\frac{\Im m{\cal A}(\gamma^{\ast}_{{\rm tr}}+p\to\gamma+p)_{|t}}{\Im m{\cal A}(\gamma^{\ast}_{{\rm tr}}+p\to\gamma^{\ast}_{{\rm tr}}+p)_{|t=0}}\,.$ (12) Note that we generalized the ratio $R$ originally defined at $t=t_{{\rm min}}$ Schoeffel:2007dt to any value of $t$. At high energies, the minimal momentum transfer $|t_{{\rm min}}|\approx x_{B}^{2}m_{N}^{2}\approx 0$, where $x_{B}$ is the Bjorken variable; $m_{N}$ is the nucleon mass. The GVMD model makes an unambiguous prediction for the ratio $R$, $\displaystyle R(t)$ $\displaystyle=$ $\displaystyle\sum_{n=0}^{\infty}\frac{F_{n}(t)}{(\frac{Q^{2}}{m_{\rho}^{2}}+1+2n)(\frac{Q^{2}}{m_{\rho}^{2}}+3+2n)}\left[1+\delta+\frac{Q^{2}}{2m_{\rho}^{2}(3+2n)}\left(1+4\delta\frac{1+n}{(1+2n)}\right)\right]$ (13) $\displaystyle\Big{/}$ $\displaystyle\left(\sum_{n=0}^{\infty}\frac{1+2n}{(\frac{Q^{2}}{m_{\rho}^{2}}+1+2n)^{2}(\frac{Q^{2}}{m_{\rho}^{2}}+3+2n)}+\frac{1}{2}\frac{\delta}{1+\frac{Q^{2}}{m_{\rho}^{2}}}\right)\,.$ Figure 2 presents the GVMD predictions for the ratio $R$ as a function of $Q^{2}$ for three values of $t$: $t=t_{{\rm min}}\approx 0$ (evaluated with $x_{B}=0.001$), $t=-0.1$ GeV2 and $t=-0.2$ GeV2. Note that in the GVMD model, the ratio $R$ does not depend on $W$ or $x_{B}$ at given $Q^{2}$ and $t$. Figure 2: The GVMD prediction for the ratio $R(t)$ of the DVCS and DIS amplitudes, see Eq. (12), as a function of $Q^{2}$ for three values of $t$. A comparison of the solid curve in Fig. 2 to the experimental results for the ratio $R$, see Fig. 4 of Ref. Schoeffel:2007dt , reveals that the GVMD model provides a good description of the data for $Q^{2}<5$ GeV2. For higher values of $Q^{2}$, the GVMD model overestimates the experimental $R$. Therefore, the GVMD model and similar models can be used to reliably determine DVCS observables and generalized parton distributions at $Q^{2}$ of the order of a few GeV2. This can be used as an input for QCD evolution to higher $Q^{2}$ scales. An example of such an approach, which uses the align-jet model to construct input GPDs and which excellently compares to the HERA data on the DVCS cross section and on the ratio $R$, was worked out in Freund:2002qf . ### II.2 DVCS cross section The DVCS amplitude in Eq. (9) is normalized such that in the $Q^{2}\to 0$ limit, the imaginary part of the $\gamma p\to\gamma p$ amplitude is equal to the total photoabsorption cross section, see Eq. (8). With such a normalization, the differential and integrated DVCS cross sections at the photon level read $\displaystyle\frac{d\sigma_{{\rm DVCS}}}{dt}(W,Q^{2},t)=\frac{1}{16\pi}|{\cal A}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)|^{2}\,,$ $\displaystyle\sigma_{{\rm DVCS}}(W,Q^{2})=\frac{1}{16\pi}\int_{-1\ {\rm GeV}^{2}}^{t_{\rm min}}dt\,|{\cal A}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)|^{2}\,.$ (14) where $t_{{\rm min}}\approx-x_{B}^{2}m_{N}^{2}$ ($t_{\rm min}\approx 0$ in the HERA kinematics). In order to compare the GVMD model predictions to the data on the DVCS cross section at the photon level Chekanov:2003ya ; Aktas:2005ty , one needs to make sure that one compares the same quantities. Using the classic result of L. N. Hand Hand:1963bb , one can readily see that the HERA DVCS cross section at the photon level is indeed a properly defined and normalized cross section of the $\gamma^{\ast}p\to\gamma p$ reaction. As a byproduct of the above mentioned exercise, one establishes the connection between the GVMD and GPD descriptions of the DVCS cross section: $\displaystyle|{\cal A}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)|^{2}$ $\displaystyle=$ $\displaystyle\frac{e^{4}x_{B}^{2}}{Q^{4}\sqrt{1+\epsilon^{2}}}\Big{(}(1-\xi^{2})(|{\cal H}|^{2}+|\tilde{{\cal H}}|^{2})-\xi^{2}({\cal H}^{\ast}{\cal E}+{\cal H}{\cal E}^{\ast}$ (15) $\displaystyle+$ $\displaystyle\tilde{{\cal H}^{\ast}}\tilde{{\cal E}}+{\tilde{\cal H}}\tilde{{\cal E}^{\ast}})-|{\cal E}|^{2}(\frac{t}{4m_{N}^{2}}+\xi^{2})-\xi^{2}\frac{t}{4m_{N}^{2}}|\tilde{{\cal E}}|^{2}\Big{)}\,,$ where $x_{B}$ is the Bjorken variable; $\xi=x_{B}/(2-x_{B})$; $\epsilon^{2}=4x_{B}^{2}m_{N}^{2}/Q^{2}$. The quantities ${\cal H}$, ${\cal E}$, $\tilde{{\cal H}}$ and ${\tilde{\cal E}}$ are the so-called Compton form factors of the corresponding proton GPDs Belitsky:2001ns . It is important to have the connection between the GVMD-based and the GPD-based descriptions of the DVCS cross section since the both approaches have an overlapping region of applicability, namely, $1<Q^{2}<5$ GeV2. The simple expression for the DVCS amplitude in the GVMD model (9) allows one to examine the relative contribution of $1/Q^{2}$-corrections, which correspond to higher-twist corrections in perturbative QCD. To this end, let us expand the DVCS amplitude in Eq. (9) in terms of $1/Q^{2}$ and let us call the leading contribution, which behaves as $1/Q^{2}$, ${\cal A}^{{\rm LO}}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)$. The corresponding $t$-integrated cross section is denoted as $\sigma_{{\rm DVCS}}^{{\rm LO}}$. We quantify the contribution of $1/Q^{2}$-corrections to the DVCS amplitude and to the DVCS cross section by introducting the ratios $R_{{\rm ampl}}^{{\rm HT}}$ and $R_{\sigma}^{{\rm HT}}$, $\displaystyle R_{{\rm ampl}}^{{\rm HT}}(Q^{2})$ $\displaystyle=$ $\displaystyle 1-\frac{{\cal A}^{{\rm LO}}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)_{|t=t_{{\rm min}}}}{{\cal A}(\gamma_{{\rm tr}}^{\ast}+p\to\gamma+p)_{|t=t_{{\rm min}}}}\,,$ $\displaystyle R_{\sigma}^{{\rm HT}}(Q^{2})$ $\displaystyle=$ $\displaystyle 1-\frac{\sigma_{{\rm DVCS}}^{{\rm LO}}}{\sigma_{{\rm DVCS}}}\,.$ (16) The ratios $R_{{\rm ampl}}^{{\rm HT}}$ and $R_{\sigma}^{{\rm HT}}$ as functions of $Q^{2}$ are summarized in Table 1. Note that these ratios do not depend on $W$ in the chosen model. Table 1: The $1/Q^{2}$-corrections to the DVCS amplitude and to the $t$-integrated DVCS cross section as functions of $Q^{2}$, see Eq. (16). $Q^{2}$ [GeV2] | $R_{{\rm ampl}}^{{\rm HT}}(Q^{2})$ | $R_{\sigma}^{{\rm HT}}(Q^{2})$ ---|---|--- 2 | 0.20 | 0.56 4 | 0.11 | 0.32 8 | 0.058 | 0.17 As one can see from Table 1, the $1/Q^{2}$-corrections are large. Moreover, $R_{\sigma}^{{\rm HT}}(Q^{2})>2R_{{\rm ampl}}^{{\rm HT}}(Q^{2})$ due to the enhancement of the heavy vector meson contributions to $\sigma_{{\rm DVCS}}$ because of the decreasing slope of the $t$-dependence of the DVCS amplitude with increasing $n$, see Eq. (6). ### II.3 Comparison to the HERA DVCS data Using Eqs. (9) and (14), we make predictions for the DVCS cross section and compare our findings to the HERA data Chekanov:2003ya ; Aktas:2005ty . Figure 3 presents the $W$-dependence of the DVCS cross section at $Q^{2}=4$ GeV2 and $Q^{2}=8$ GeV2. The solid curves correspond to the GVMD calculations; the experimental points are those from the H1 Aktas:2005ty and ZEUS Chekanov:2003ya measurements. The error bars correspond to the statistical and systematic errors added in quadrature. The ZEUS data taken at $Q^{2}=9.6$ GeV2 have been interpolated to $Q^{2}=8.0$ GeV2 using the fit to the $Q^{2}$-dependence of $\sigma_{{\rm DVCS}}$, $\sigma_{{\rm DVCS}}\sim 1/(Q^{2})^{n}$ with $n=1.54$ Chekanov:2003ya . Figure 3: The DVCS cross section as a function of $W$. The GVMD model results (solid curves) are compared to the H1 Aktas:2005ty and ZEUS Chekanov:2003ya data. The error bars correspond to the statistical and systematic errors added in quadrature. We shall discuss the left and right panels of Fig. 3 separately. As seen from the left panel of Fig. 3, the GVMD model reproduces both the absolute value and the $W$-dependence of $\sigma_{{\rm DVCS}}$ sufficiently well. The latter fact signifies that, at $Q^{2}=4$ GeV2, the DVCS cross section is still dominated by soft physics. At $Q^{2}=4$ GeV2, the $W$-behavior of $\sigma_{{\rm DVCS}}$ is consistent with that predicted by the GVMD model, $\sigma_{{\rm DVCS}}\sim W^{0.4}$. Turning to the right panel of Fig. 3, we observe that while the GVMD model compares fairly with the H1 data, the model underestimates the slope of the $W$-dependence of $\sigma_{{\rm DVCS}}$ for the ZEUS data set, which has smaller error bars. In particular, the predicted $\sigma_{{\rm DVCS}}\sim W^{0.4}$ behavior is much slower than that given by the fit to the ZEUS data points, $\sigma_{{\rm DVCS}}\sim W^{\delta}$ with $\delta=0.75\pm 0.15$ Chekanov:2003ya . This indicates the onset of the hard regime in the total DVCS cross section at $Q^{2}=8$ GeV2, where the GVMD model becomes inadequate. Figure 4 presents the $Q^{2}$-dependence of the DVCS cross section at $W=82$ GeV. The solid curve corresponds to the GVMD model; the experimental points come from the H1 Aktas:2005ty and ZEUS Chekanov:2003ya experiments. The error bars correspond to the statistical and systematic errors added in quadrature. The ZEUS data taken at $W=89$ GeV have been extrapolated to $W=82$ GeV using the fitted $W$-dependence of $\sigma_{{\rm DVCS}}$, $\sigma_{{\rm DVCS}}\sim W^{0.75}$ Chekanov:2003ya . Figure 4: The DVCS cross section as a function of $Q^{2}$. The GVMD model result (solid curve) is compared to the H1 Aktas:2005ty and ZEUS Chekanov:2003ya data. The error bars correspond to the statistical and systematic errors added in quadrature. One sees from Fig. 4 that the GVMD model reproduces the $Q^{2}$-dependence of $\sigma_{{\rm DVCS}}$ over a very wide range of $Q^{2}$, $3\leq Q^{2}\leq 85$ GeV2. This is quite a remarkable result that the model, which was initially developed for photoproduction and was later extended to electroproduction with $Q^{2}$ of the order of a few GeV2, provides a quantitative description for such large values of $Q^{2}$. In other words, at fixed $W$, the GVMD model correctly reproduces the $Q^{2}$-scaling of $\sigma_{{\rm DVCS}}$. Figure 5 presents the comparison of the $t$-dependence of the GVMD model calculations (solid curves) to the H1 data on the differential DVCS cross section $d\sigma_{{\rm DVCS}}/dt$ Aktas:2005ty . The error bars are the statistical and systematic errors added in quadrature. Figure 5: The differential DVCS cross section $d\sigma_{{\rm DVCS}}/dt$ as a function of $t$. The GVMD model calculation (solid curves) is compared to the H1 data Aktas:2005ty . The error bars are the statistical and systematic errors added in quadrature. As one sees from Fig. 5, the GVMD model describes $d\sigma_{{\rm DVCS}}/dt$ well. This result is not quite trivial. In order to achieve this within the framework of the GVMD model, one has to assume that either all vector meson $V_{n}$-nucleon cross sections have the same $Q^{2}$-dependent slope of the $t$-dependence or that the slope decreases with increasing $n$, see Eq. (6). While the value of the slope $B_{1}=11$ GeV-2 is fixed by photoproduction of $\rho$ mesons, the value of the slope $B_{2}$ is model-dependent. The values $B_{2}=4\div 5$ GeV-2 provide a good agreement with the H1 data (see Fig. 5), which were fitted to the exponential form, $d\sigma_{{\rm DVCS}}/dt\sim e^{-B|t|}$ with $B=(6.66\pm 0.54\pm 0.43)$ GeV-2 at $Q^{2}=4$ GeV2 and $B=(5.82\pm 0.59\pm 0.60)$ GeV-2 at $Q^{2}=8$ GeV2 Aktas:2005ty . ### II.4 DVCS cross section in Jefferson Lab kinematics The $\vec{e}p\to ep\gamma$ cross section in the DVCS regime was recently measured by the Hall A collaboration at Jefferson Laboratory (JLab) Munoz Camacho:2006hx . The cross section involves the contributions of the Bethe- Heitler (BH) amplitude squared, the DVCS amplitude squared and the interference of the BH and DVCS amplitudes. Based on the kinematics of the experiment, in the analysis of the data, the contribution of the DVCS amplitude squared was neglected compared to the other two contributions Belitsky:2001ns , which allowed for the extraction of the so-called Compton form factors of the proton. In this subsection, we check the validity of the assumption that the contribution of the DVCS amplitude squared is negligibly small by explicitly calculating the DVCS cross section within the GVMD model in the Jefferson Lab kinematics. The DVCS cross section at the lepton level reads, see e.g. Belitsky:2001ns , $\frac{d^{4}\sigma}{dQ^{2}dx_{B}dtd\phi}=\frac{\alpha_{{\rm e.m.}}(1-y+y^{2}/2)}{\pi Q^{2}x_{B}}\frac{1}{2\pi}\frac{d\sigma_{{\rm DVCS}}(W,Q^{2},t)}{dt}\,,$ (17) where $\alpha_{{\rm e.m.}}$ is the fine-structure constant; $\phi$ is the angle between the lepton and production planes; $d\sigma_{{\rm DVCS}}(W,Q^{2},t)/dt$ is the DVCS cross section at the proton level defined by Eq. (14). The extra factor $1/(2\pi)$ in the right-hand side of Eq. (17) takes into account the fact that the integration over the angle $\phi$ is included in the definition of $d\sigma_{{\rm DVCS}}(W,Q^{2},t)/dt$. The DVCS cross section does not depend on $\phi$ when one neglects the photon helicity changing transitions Diehl:2000xz . Using Eqs. (9) and (14), we evaluate the DVCS cross section at the lepton level $d^{4}\sigma/(dQ^{2}dx_{B}dtd\phi)$ in the kinematics of the Hall A experiment, $E=5.75$ GeV (the energy of the lepton beam), $Q^{2}=2.3$ GeV, $t=-0.28$ GeV and $x_{B}=0.36$, $\frac{d^{4}\sigma}{dQ^{2}dx_{B}dtd\phi}=0.0022\ {\rm nb}/{\rm GeV}^{4}\,.$ (18) This value is an order of magnitude smaller than the sum of the BH and interference contributions to the $ep\to ep\gamma$ cross section, which confirms the assumption that, in the JLab kinematics, the contribution of the DVCS amplitude squared to the $ep\to ep\gamma$ cross section can be safely neglected. ## III DVCS on nuclei In this section, we derive the expression for the DVCS amplitude on a nucleus using the GVMD model for the photon-nucleon interactions and the generalized Glauber formalism Bauer:1977iq in order to account for the multiple rescattering of the vector mesons inside the nucleus. Using the obtained amplitude, we make predictions for the nuclear DVCS cross section at the photon level in the collider kinematics. At high energies, in the GVMD model, photons (real and virtual) interact with hadrons by fluctuating into an infinite sum of vector mesons. When the involved hadron is a nucleus, each vector meson undergoes multiple interactions with the nucleons of the nucleus, which leads to the attenuation (decrease) of the vector meson-nucleus cross section compared to the sum of free vector meson-nucleon cross section. As a consequence, the resulting photon-nucleus cross section is smaller than the sum of the corresponding photon-nucleon cross sections. This phenomenon is called nuclear shadowing. It has been observed in various reactions with nuclei induced by real and virtual photons, see Bauer:1977iq ; Piller:1999wx for review. In the GVMD model, the nuclear DVCS amplitude can be organized as a multiple scattering series (Glauber series), where each term corresponds to the number of interactions of the vector mesons with the nucleons. This is schematically presented in Fig. 6, where the interactions with one, two and three nucleons are depicted. The dashed lines correspond to the vector mesons (note that the GVMD model allows for non-diagonal vector meson-nucleon transitions); the solid lines correspond to the nucleons involved in the interactions; the nuclear part is denoted by ovals with legs corresponding to the initial and final nucleus. Figure 6: A schematic representation of the multiple scattering (Glauber) series for the nuclear DVCS amplitude in the GVMD model. The dashed lines correspond to vector mesons; the solid lines correspond to nucleons; the ovals with legs correspond to the final and initial nucleus. Using the standard technique Bauer:1977iq , one can readily write down the expression for the nuclear DVCS amplitude ${\cal A}(\gamma^{\ast}+A\to\gamma+A)$, $\displaystyle{\cal A}(\gamma^{\ast}+A\to\gamma+A)$ $\displaystyle=$ $\displaystyle\sum_{n,m=0}^{\infty}\frac{e}{f_{n}}\frac{M_{n}^{2}}{M_{n}^{2}+Q^{2}}\frac{e}{f_{m}}\Bigg{[}AF_{A}(t)\Sigma_{n,m}(W,t)$ (19) $\displaystyle-$ $\displaystyle\frac{A(A-1)}{2i}\int_{-\infty}^{\infty}dz_{1}\int_{z_{1}}^{\infty}dz_{2}\int d^{2}\vec{b}\,e^{i\vec{q}_{t}\cdot\vec{b}}\rho(b,z_{1})\rho(b,z_{2})e^{iz_{1}(k_{\gamma^{\ast}}-k_{V_{n}})}$ $\displaystyle\times$ $\displaystyle\tilde{\Sigma}_{n,n^{\prime}}(W,z_{1})\Big{(}\delta_{n^{\prime},n^{\prime\prime}}-\frac{A-2}{2i}\,\tilde{\Sigma}_{n^{\prime},n^{\prime\prime}}(W,z^{\prime})\Theta(z_{1}\leq z^{\prime}\leq z_{2})\rho(b,z^{\prime})$ $\displaystyle+$ $\displaystyle\dots\Big{)}\tilde{\Sigma}_{n^{\prime\prime},m}(W,z_{2})e^{iz_{2}(k_{V_{m}}-k_{\gamma})}\Bigg{]}\,,$ where $A$ is the number of nucleons in the nucleus (we do not distinguish protons and neutrons); $F_{A}(t)$ is the nuclear form factor ($F_{A}(0)=1$); $\rho(r)$ is the density of nucleons in the nucleus DeJager:1987qc ; $\vec{b}$ is the two-dimensional vector (impact parameter) in the plane perpendicular to the direction of the incoming photon, whose momentum is assumed to be along the $z$-direction; $z_{i}$ are longitudinal positions of the nucleons of the nucleus involved in the interaction; $\vec{q}_{t}$ is the transverse component of the momentum transfer. Since we neglected the $t$-dependence of the elementary vector meson-nucleon scattering amplitudes compared to the steep $t$-dependence of the nuclear form factor, all scatterings of the vector mesons in Eq. (19) occur at the same impact parameter $\vec{b}$. In Eq. (19), $\tilde{\Sigma}_{n,m}(W,z)=e^{iz(k_{V_{n}}-k_{V_{m}})}\Sigma_{n,m}(W,t=0)\,,$ (20) where $\Sigma_{n,m}$ is defined by Eq. (5). In Eqs. (19) and (20), the exponential factors (except for the $\exp(i\vec{q}_{t}\cdot\vec{b})$ factor) arise due to the non-zero longitudinal momentum transfer associated with non- diagonal in mass transitions. At high energies, $\displaystyle k_{\gamma^{\ast}}-k_{V_{n}}=\sqrt{\nu^{2}+Q^{2}}-\sqrt{\nu^{2}-M_{n}^{2}}=\frac{Q^{2}+M_{n}^{2}}{2\nu}=x_{B}m_{N}\left(1+\frac{M_{n}^{2}}{Q^{2}}\right)\,,$ $\displaystyle k_{V_{m}}-k_{\gamma}=\sqrt{\nu^{\prime 2}-M_{n}^{2}}-\nu^{\prime}=-\frac{M_{m}^{2}}{2\nu^{\prime}}\approx-\frac{M_{m}^{2}}{2\nu}=-x_{B}m_{N}\frac{M_{m}^{2}}{Q^{2}}\,,$ $\displaystyle k_{V_{n}}-k_{V_{m}}=\frac{M_{m}^{2}-M_{n}^{2}}{2\nu}\,,$ (21) where $\nu$ is the energy of the incoming virtual photon in the laboratory reference frame; $\nu^{\prime}$ is the energy of the final real photon. We also used that $\nu^{\prime}=\nu+t/(2m_{N})\approx\nu$ for the small momentum transfer $t$. In Eq. (19), the first term corresponds to the left graph in Fig. 6, which describes the interaction with one nucleon of the nucleus (the Born term). The second term in Eq. (19) corresponds to the middle graph in Fig. 6, which describes the interaction of hadronic fluctuations of the involved photons with two nucleons of the nucleus. Those nucleons are located at the points $\vec{r}_{1}=(\vec{b},z_{1})$ and $\vec{r}_{2}=(\vec{b},z_{2})$. This graph leads to the attenuation (nuclear shadowing) of the Born term. The third term corresponds to the interaction with three nucleons of the nucleus. The dots in Eq. (19) denote higher rescattering terms not shown in Fig. 6. Equation (19) is rather general and, because of the non-diagonal $V_{n}\to V_{n^{\prime}}$ transitions, the direct calculation of the nuclear DVCS amplitude for heavy nuclei using Eq. (19) is impossible. Therefore, for our numerical predictions, we make an approximation and ignore the non-diagonal transitions for the interactions with three and more nucleons (this does affect the convergence of the series), $\displaystyle\delta_{n^{\prime},n^{\prime\prime}}-\frac{A-2}{2i}\,\tilde{\Sigma}_{n^{\prime},n^{\prime\prime}}(W,z^{\prime})\Theta(z_{1}\leq z^{\prime}\leq z_{2})\rho(b,z^{\prime})+\dots$ (22) $\displaystyle\to$ $\displaystyle\delta_{n^{\prime},n^{\prime\prime}}\left(1-\frac{A-2}{2i}\,\Sigma_{n^{\prime},n^{\prime}}(W,t=0)\Theta(z_{1}\leq z^{\prime}\leq z_{2})\rho(b,z^{\prime})+\dots\right)$ $\displaystyle=$ $\displaystyle\delta_{n^{\prime},n^{\prime\prime}}\,e^{-\frac{A}{2}\sigma_{\rho p}(W^{2})(1-i\eta)\int^{z_{2}}_{z_{1}}dz^{\prime}\rho(b,z^{\prime})}\,,$ where in the last line we used the large-$A$ approximation. Therefore, Eq. (19) now reads $\displaystyle{\cal A}(\gamma^{\ast}+A$ $\displaystyle\to$ $\displaystyle\gamma+A)=\sum_{n,m=0}^{\infty}\frac{e}{f_{n}}\frac{M_{n}^{2}}{M_{n}^{2}+Q^{2}}\frac{e}{f_{m}}\Bigg{[}AF_{A}(t)\Sigma_{n,m}(W,t)$ (23) $\displaystyle-$ $\displaystyle\frac{A(A-1)}{2i}\int_{-\infty}^{\infty}dz_{1}\int_{z_{1}}^{\infty}dz_{2}\int d^{2}\vec{b}\,e^{i\vec{q}_{t}\cdot\vec{b}}\rho(b,z_{1})\rho(b,z_{2})e^{iz_{1}(k_{\gamma^{\ast}}-k_{V_{n}})}$ $\displaystyle\times$ $\displaystyle\tilde{\Sigma}_{n,n^{\prime}}(W,z_{1})e^{-\frac{A}{2}\sigma_{\rho p}(W^{2})(1-i\eta)\int^{z_{2}}_{z_{1}}dz^{\prime}\rho(b,z^{\prime})}\tilde{\Sigma}_{n^{\prime},m}(W,z_{2})e^{iz_{2}(k_{V_{m}}-k_{\gamma})}\Bigg{]}\,.$ For comparison of nuclear shadowing in DVCS and DIS, we also give the expression for the forward nuclear DIS amplitude, which can be readily obtained from Eq. (23), $\displaystyle{\cal A}(\gamma^{\ast}+A$ $\displaystyle\to$ $\displaystyle\gamma^{\ast}+A)_{|t=0}=\sum_{n,m=0}^{\infty}\frac{e}{f_{n}}\frac{M_{n}^{2}}{M_{n}^{2}+Q^{2}}\frac{e}{f_{m}}\frac{M_{m}^{2}}{M_{m}^{2}+Q^{2}}\Bigg{[}A\Sigma_{n,m}(W,t=0)$ (24) $\displaystyle-$ $\displaystyle\frac{A(A-1)}{2i}\int_{-\infty}^{\infty}dz_{1}\int_{z_{1}}^{\infty}dz_{2}\int d^{2}\vec{b}\,\rho(b,z_{1})\rho(b,z_{2})e^{iz_{1}(k_{\gamma^{\ast}}-k_{V_{n}})}$ $\displaystyle\times$ $\displaystyle\tilde{\Sigma}_{n,n^{\prime}}(W,z_{1})e^{-\frac{A}{2}\sigma_{\rho p}(W^{2})(1-i\eta)\int^{z_{2}}_{z_{1}}dz^{\prime}\rho(b,z^{\prime})}\tilde{\Sigma}_{n^{\prime},m}(W,z_{2})e^{iz_{2}(k_{V_{m}}-k_{\gamma^{\ast}})}\Bigg{]}\,.$ We quantify predictions of the GVMD model for the nuclear DVCS and DIS amplitudes by considering the ratios $R_{{\rm ampl}}^{{\rm Im}}$, $R_{{\rm ampl}}^{{\rm Re}}$ and $R_{{\rm ampl}}^{{\rm DIS}}$, $\displaystyle R_{{\rm ampl}}^{{\rm Im}}$ $\displaystyle=$ $\displaystyle\frac{\Im m\,{\cal A}(\gamma^{\ast}+A\to\gamma+A)}{\Im m\,{\cal A}^{{\rm Born}}(\gamma^{\ast}+A\to\gamma+A)}\,,$ $\displaystyle R_{{\rm ampl}}^{{\rm Re}}$ $\displaystyle=$ $\displaystyle\frac{\Re e\,{\cal A}(\gamma^{\ast}+A\to\gamma+A)}{\Re e\,{\cal A}^{{\rm Born}}(\gamma^{\ast}+A\to\gamma+A)}\,,$ $\displaystyle R_{{\rm ampl}}^{{\rm DIS}}$ $\displaystyle=$ $\displaystyle\frac{\Im m\,{\cal A}(\gamma^{\ast}+A\to\gamma^{\ast}+A)_{|t=0}}{\Im m\,{\cal A}^{{\rm Born}}(\gamma^{\ast}+A\to\gamma^{\ast}+A)|_{t=0}}\,,$ (25) where ${\cal A}(\gamma^{\ast}+A\to\gamma+A)$ is the nuclear DVCS amplitude of Eq. (23); ${\cal A}^{{\rm Born}}(\gamma^{\ast}+A\to\gamma+A)$ is the first term (Born contribution) of Eq. (23); ${\cal A}^{{\rm Born}}(\gamma^{\ast}+A\to\gamma^{\ast}+A)_{|t=0}$ is the first term of Eq. (26). Note that $R_{{\rm ampl}}^{{\rm DIS}}$ is nothing but the the ratio of the nuclear to the nucleon inclusive structure functions, $R_{{\rm ampl}}^{{\rm DIS}}=F_{2A}(x,Q^{2})/[AF_{2N}(x,Q^{2})]$. Figure 7 presents the ratios $R_{{\rm ampl}}^{{\rm Im}}$ (solid curves), $R_{{\rm ampl}}^{{\rm Re}}$ (dotted curves) and $R_{{\rm ampl}}^{{\rm DIS}}$ (dot-dashed curves) at $Q^{2}=1$ GeV2 and $t=t_{{\rm min}}\approx- x_{B}^{2}m_{N}^{2}$ as functions of $x_{B}$. The left panel corresponds to the nucleus of 40Ca; the right panel corresponds to 208Pb. Figure 7: The ratios $R_{{\rm ampl}}^{{\rm Im}}$ (solid), $R_{{\rm ampl}}^{{\rm Re}}$ (dotted) and $R_{{\rm ampl}}^{{\rm DIS}}$ (dot-dashed) of Eq. (25) at $Q^{2}=1$ GeV2 and $t=t_{{\rm min}}$ as functions of Bjorken $x_{B}$. The left panel is for 40Ca; the right panel is for 208Pb. Let us now discuss the results presented in Fig. 7 in detail. The solid and dot-dashed curves coincide for $x_{B}<0.01$ and deviate only slightly for $0.01<x_{B}<0.1$, which means that the amount of nuclear shadowing is the same in the imaginary parts of the DVCS and DIS amplitudes. This observation agrees with the results obtained within the framework of a different approach to nuclear GPDs at small-$x_{B}$, when the latter are modeled using the align-jet model for the nucleon GPDs and a parameterization of usual nuclear PDFs Freund:2003wm . Moreover, the amount of nuclear shadowing predicted by our calculations in the GVMD model matches very well the leading-twist predictions for $F_{2A}(x,Q^{2})/[AF_{2N}(x,Q^{2})]$ made at somewhat higher $Q^{2}$ Frankfurt:2003zd . This is a consequence of the fact the GVMD model predicts a significant amount of inclusive diffraction in $\gamma^{\ast}p$ scattering at all $Q^{2}$, which controls the size of nuclear shadowing in the leading-twist theory of nuclear shadowing Frankfurt:2003zd . As the value of $x_{B}$ is increased (at fixed $Q^{2}$), the shadowing correction decreases due to the decrease of $\sigma_{\rho p}(W)$ and due to the increasingly destructive role of the $e^{iz_{1}(k_{\gamma^{\ast}}-k_{V_{n}})}$ and $e^{iz_{2}(k_{V_{m}}-k_{\gamma})}$ factors in Eq. (23). For the ratio $R_{{\rm ampl}}^{{\rm Re}}$ of the real parts (dotted curves), at small-$x_{B}$, the shadowing correction is approximately two times as large as for the ratio of the imaginary parts because of the fact that $(1-\eta)^{2}=1-\eta^{2}-2i\eta$, see Eqs. (5) and (23). As $x_{B}$ increases, the real part of the shadowing correction receives a large contribution from the $e^{iz_{1}(k_{\gamma^{\ast}}-k_{V_{n}})}$ and $e^{iz_{2}(k_{V_{m}}-k_{\gamma})}$ factors, and, as a result, it steadily grows and becomes larger than the Born contribution. This behavior of $R_{{\rm ampl}}^{{\rm Re}}$ is similar to that observed in Freund:2003wm . Note, however, that since the effect of $t_{{\rm min}}$ (the factor $F_{A}(t_{{\rm min}})$ in the Born term) was not included in the analysis of Freund:2003wm , the agreement could be coincidental. We also examined nuclear shadowing in DVCS with nuclear targets as a function of the momentum transfer $t$. Figure 8 presents $R_{{\rm ampl}}^{{\rm Im}}$ at $Q^{2}=1$ GeV2 as a function of $t$. Figure 8 demonstrates that the shadowing correction to the DVCS amplitude has the $t$-dependence which is slower than that of the Born term. As one increases $|t|$, the negative nuclear shadowing correction decreases slower than the Born term, which leads to a decrease of $R_{{\rm ampl}}^{{\rm Im}}$. The ratio $R_{{\rm ampl}}^{{\rm Re}}$ follows the similar trend. Figure 8: The ratio $R_{{\rm ampl}}^{{\rm Im}}$ at $Q^{2}=1$ GeV2 as a function of $t$. The left panel is for 40Ca; the right panel is for 208Pb. Equation (23) presents the $\gamma^{\ast}A\to\gamma A$ scattering amplitude as a function of $W$, $Q^{2}$ and $t$. It also allows for the representation of the scattering amplitude as a function of $W$, $Q^{2}$ and $\vec{b}$, where $\vec{b}$ is the impact parameter conjugate to $\vec{q}_{t}$, $\displaystyle{\cal A}(\gamma^{\ast}+A$ $\displaystyle\to$ $\displaystyle\gamma+A)=\sum_{n,m=0}^{\infty}\frac{e}{f_{n}}\frac{M_{n}^{2}}{M_{n}^{2}+Q^{2}}\frac{e}{f_{m}}\Bigg{[}A\int^{\infty}_{-\infty}dze^{ix_{B}m_{N}z}\rho(b,z)\Sigma_{n,m}(W,0)$ (26) $\displaystyle-$ $\displaystyle\frac{A(A-1)}{2i}\int_{-\infty}^{\infty}dz_{1}\int_{z_{1}}^{\infty}dz_{2}\,\rho(b,z_{1})\rho(b,z_{2})e^{iz_{1}(k_{\gamma^{\ast}}-k_{V_{n}})}$ $\displaystyle\times$ $\displaystyle\tilde{\Sigma}_{n,n^{\prime}}(W,z_{1})e^{-\frac{A}{2}\sigma_{\rho p}(W^{2})(1-i\eta)\int^{z_{2}}_{z_{1}}dz^{\prime}\rho(b,z^{\prime})}\tilde{\Sigma}_{n^{\prime},m}(W,z_{2})e^{iz_{2}(k_{V_{m}}-k_{\gamma})}\Bigg{]}\,.$ In the first term in Eq. (26), we took into account the non-zero longitudinal momentum transfer, $k_{\gamma^{\ast}}-k_{\gamma}=x_{B}m_{N}$, see Eq. (21), and also neglected the $t$-dependence of $\Sigma_{n,m}(t)$ compated to $F_{A}(t)$. Using Eq. (26), the nuclear DVCS cross section can be expressed in the following compact form $\sigma_{{\rm DVCS}}(W,Q^{2})=\frac{1}{4}\int d^{2}\vec{b}\,|{\cal A}(\gamma^{\ast}+A\to\gamma+A)|^{2}\,.$ (27) In order to quantify predictions of the GVMD model for nuclear DVCS cross sections, we introduce the ratio $R_{{\rm cs}}$, $R_{{\rm cs}}=\frac{\sigma_{{\rm DVCS}}(W,Q^{2})}{\sigma_{{\rm DVCS}}^{{\rm Born}}(W,Q^{2})}\,,$ (28) where the numerator is calculated using Eq. (27) and the complete expression for the nuclear DVCS amplitude (26); the denominator is calculated using only the first term (Born contribution) in Eq. (26). Predictions of the GVMD model for the ratio $R_{{\rm cs}}$ at $Q^{2}=1$ GeV2 as a function of $x_{B}$ are presented in Fig. 9. The solid curve corresponds to 40Ca; the dotted curve corresponds to 208Pb. Figure 9: The ratio $R_{{\rm cs}}=\sigma_{{\rm DVCS}}/\sigma_{{\rm DVCS}}^{{\rm Born}}$ of Eq. (28) as a function of Bjorken $x_{B}$ at $Q^{2}=1$ GeV2. The solid curve corresponds to 40Ca; the dotted curve is for 208Pb. As one can see from Fig. 9, the predicted amount of nuclear shadowing at small-$x_{B}$ is very large. Since the $t$-dependence of the nuclear DVCS amplitude is very steep, the dominant contribution to the $t$-integrated cross sections entering $R_{{\rm cs}}$ comes from the $t\approx t_{{\rm min}}$ region. Therefore, the amount of nuclear shadowing for $R_{{\rm cs}}$ is equal roughly twice the amount of nuclear shadowing for $R_{{\rm ampl}}^{\rm Im}$, see Fig. 7. Finally, we would also like to point out that for nuclear DVCS, the ratio of the imaginary parts of the DVCS and DIS amplitudes, see Eq. (12) and Fig. 2, is quite similar to the free nucleon case. This is a mere consequence of the fact that the structure of the $Q^{2}$-dependence of the ratio is essentially the same in the DVCS on the nucleon and on nuclei. ## IV Conclusions and Discussion We considered Deeply Virtual Compton Scattering (DVCS) on nucleons and nuclei in the framework of generalized vector meson dominance (GVMD) model. We extended the original GVMD model, which was applied to forward amplitudes of Deep Inelastic Scattering (DIS), to the non-forward $t\neq 0$ case. We introduced the $W$-dependence of the DVCS amplitude through the $W$-dependence of the elementary vector meson-nucleon amplitude, which was taken to be proportional to $W^{0.2}$ at high-$W$. We compared our predictions to the HERA data on DVCS on the proton with the following results. The GVMD model describes well the dependence of the DVCS cross section on $Q^{2}$, $W$ (at $Q^{2}$=4 GeV2) and $t$. At $Q^{2}$=8 GeV2, the $W$-dependence of the cross section is somewhat underestimated, which can be interpreted as due to the onset of the hard regime beyond the soft dynamics of the GVMD model. We estimated the relative contribution of $1/Q^{2}$-corrections to the DVCS amplitude and the DVCS cross section. We found that these corrections are large: the contribution of the $1/Q^{2}$-corrections to the DVCS amplitude at $t=t_{{\min}}$ is 20% at $Q^{2}=2$ GeV2, 11% at $Q^{2}=4$ GeV2 and 6% at $Q^{2}=8$ GeV2; the contribution of the $1/Q^{2}$-corrections to the $t$-integrated DVCS cross section is 56% at $Q^{2}=2$ GeV2, 32% at $Q^{2}=4$ GeV2 and 17% at $Q^{2}=8$ GeV2. We also made predictions for the DVCS amplitude and the DVCS cross section on nuclear targets, which are relevant for the physics program of the future Electron-Ion Collider. We predicted significant nuclear shadowing, which matches well predictions of the leading-twist nuclear shadowing in DIS on nuclei. Our analysis allows us to argue that the GVMD model provides a reliable parameterization of the DVCS amplitude and the DVCS cross section with nucleons and nuclei in a wide range of kinematics. At fixed values of $Q^{2}$, which should not be too large, $Q^{2}\lesssim 5$ GeV2, the GVMD model is applicable starting from $W=2$ GeV (JLab), towards $W\approx 80$ GeV (HERA) and beyond (real photons at the LHC). At fixed $W$, the GVMD model is applicable from the photoproduction limit up to the values of $Q^{2}$, where perturbative QCD can already be used, $0\leq Q^{2}\lesssim 5$ GeV2. In addition, due to the correct $1/Q^{2}$-scaling, predictions of the GVMD model can be extrapolated to much higher values of $Q^{2}$ such that the range of applicability of the GVMD model becomes very wide, $0\leq Q^{2}<80$ GeV2. The model can be applied for a wide range of $t$: $0<ŧ|t|<1$ GeV2. ###### Acknowledgements. The authors would like to thank C. Weiss for suggesting the research topic of the present paper and for useful discussions. We also thank M. Strikman for reading the manuscript and useful comments. The work has been partially supported by the Verbundforschung (”Hadronen und Kerne”) of the BMBF and by the Transregio/SFB Bonn-Bochum-Giessen. Notice: Authored by Jefferson Science Associates, LLC under U.S. DOE Contract No. DE-AC05-06OR23177. The U.S. Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce this manuscript for U.S. Government purposes. ## References * (1) D. Mueller, D. Robaschik, B. Geyer, F. M. Dittes and J. Horejsi, Fortsch. Phys. 42, 101 (1994) [arXiv:hep-ph/9812448]. * (2) X. D. Ji, Phys. Rev. D 55, 7114 (1997). * (3) X. D. Ji, J. Phys. G 24, 1181 (1998) [arXiv:hep-ph/9807358]. * (4) A. V. Radyushkin, Phys. Lett. B 380, 417 (1996) [arXiv:hep-ph/9604317]. * (5) A. V. Radyushkin, Phys. Rev. D 56, 5524 (1997). * (6) A. V. Radyushkin, arXiv:hep-ph/0101225. * (7) J. C. Collins and A. Freund, Phys. Rev. D 59, 074009 (1999). * (8) J. C. Collins, L. Frankfurt and M. Strikman, Phys. Rev. D 56, 2982 (1997). * (9) S. J. Brodsky, L. Frankfurt, J. F. Gunion, A. H. Mueller and M. Strikman, Phys. Rev. D 50, 3134 (1994). * (10) K. Goeke, M. V. Polyakov and M. Vanderhaeghen, Prog. Part. Nucl. Phys. 47, 401 (2001) [arXiv:hep-ph/0106012]. * (11) M. Diehl, T. Feldmann, R. Jakob and P. Kroll, Nucl. Phys. B 596, 33 (2001) [Erratum-ibid. B 605, 647 (2001)] [arXiv:hep-ph/0009255]. * (12) A. V. Belitsky, D. Mueller and A. Kirchner, Nucl. Phys. B 629, 323 (2002) [arXiv:hep-ph/0112108]. * (13) M. Diehl, Phys. Rept. 388, 41 (2003) [arXiv:hep-ph/0307382]. * (14) A. V. Belitsky and A. V. Radyushkin, Phys. Rept. 418, 1 (2005) [arXiv:hep-ph/0504030]. * (15) E. R. Berger, M. Diehl and B. Pire, Eur. Phys. J. C 23, 675 (2002) [arXiv:hep-ph/0110062]. * (16) E. R. Berger, M. Diehl and B. Pire, Phys. Lett. B 523, 265 (2001) [arXiv:hep-ph/0110080]. * (17) M. Diehl, T. Gousset and B. Pire, Phys. Rev. D 62, 073014 (2000) [arXiv:hep-ph/0003233]. * (18) L. Mankiewicz, G. Piller and A. Radyushkin, Eur. Phys. J. C 10, 307 (1999) [arXiv:hep-ph/9812467]. * (19) L. L. Frankfurt, P. V. Pobylitsa, M. V. Polyakov and M. Strikman, Phys. Rev. D 60, 014010 (1999). * (20) A. Airapetian et al. [HERMES Collaboration], Phys. Rev. Lett. 87, 182001 (2001). * (21) A. Airapetian et al. [HERMES Collaboration], Phys. Rev. D 75, 011103 (2007). * (22) Z. Ye [HERMES Collaboration], arXiv:hep-ex/0606061. * (23) S. Stepanyan et al. [CLAS Collaboration], Phys. Rev. Lett. 87, 182002 (2001). * (24) C. Munoz Camacho et al. [Jefferson Lab Hall A Collaboration], Phys. Rev. Lett. 97, 262002 (2006). * (25) F. X. Girod et al. [CLAS Collaboration], arXiv:0711.4805 [hep-ph]. * (26) H. Fraas, B. J. Read and D. Schildknecht, Nucl. Phys. B 86, 346 (1975). * (27) P. Ditsas, B. J. Read and G. Shaw, Nucl. Phys. B 99, 85 (1975). * (28) G. Shaw, Phys. Rev. D 47, 3676 (1993). * (29) L. Frankfurt, V. Guzey and M. Strikman, Phys. Rev. D 58, 094039 (1998). * (30) S. Chekanov et al. [ZEUS Collaboration], Phys. Lett. B 573, 46 (2003) [arXiv:hep-ex/0305028]. * (31) A. Aktas et al. [H1 Collaboration], Eur. Phys. J. C 44, 1 (2005) [arXiv:hep-ex/0505061]. * (32) N. Kivel, M. V. Polyakov and M. Vanderhaeghen, Phys. Rev. D 63, 114014 (2001). * (33) A. Freund, Phys. Rev. D 68, 096006 (2003). * (34) A. V. Radyushkin and C. Weiss, Phys. Rev. D 63, 114012 (2001). * (35) L. Frankfurt, V. Guzey and M. Strikman, Phys. Rev. D 71, 054001 (2005). * (36) R. P. Feynman, Photon-Hadron Interactions (Benjamin, Reading, 1972). * (37) T. H. Bauer, R. D. Spital, D. R. Yennie and F. M. Pipkin, Rev. Mod. Phys. 50, 261 (1978) [Erratum-ibid. 51, 407 (1979)]. * (38) K. Fujikawa, Phys. Rev. D 4, 2794 (1971). * (39) V. N. Gribov, Sov. Phys. JETP 30 (1970) 709 [Zh. Eksp. Teor. Fiz. 57 (1969) 1306]. * (40) H. Fraas and D. Schildknecht, Phys. Lett. B 35, 72 (1971). * (41) A. Dar, Annals Phys. 65, 324 (1971). * (42) W. M. Yao et al. [Particle Data Group], J. Phys. G 33, 1 (2006). * (43) S. Aid et al. [H1 Collaboration], Nucl. Phys. B 463, 3 (1996) [arXiv:hep-ex/9601004]. * (44) J. Breitweg et al. [ZEUS Collaboration], Eur. Phys. J. C 2, 247 (1998) [arXiv:hep-ex/9712020]. * (45) C. Adloff et al. [H1 Collaboration], Eur. Phys. J. C 13, 371 (2000) [arXiv:hep-ex/9902019]. * (46) A. Pautz and G. Shaw, Phys. Rev. C 57, 2648 (1998) [arXiv:hep-ph/9710235]. * (47) S. Chekanov et al. [ZEUS Collaboration], Nucl. Phys. B 627, 3 (2002) [arXiv:hep-ex/0202034]. * (48) V. N. Gribov and A. A. Migdal, Sov. J. Nucl. Phys. 8, 583 (1969) [Yad. Fiz. 8, 1002 (1968)]. * (49) L. Schoeffel, arXiv:0706.3488 [hep-ph]. * (50) A. Freund, M. McDermott and M. Strikman, Phys. Rev. D 67, 036001 (2003). * (51) L. N. Hand, Phys. Rev. 129, 1834 (1963). * (52) G. Piller and W. Weise, Phys. Rept. 330, 1 (2000) [arXiv:hep-ph/9908230]. * (53) C. W. De Jager, H. De Vries and C. De Vries, Atom. Data Nucl. Data Tabl. 36, 495 (1987). * (54) A. Freund and M. Strikman, Phys. Rev. C 69, 015203 (2004).

arxiv-papers

2024-09-04T02:48:53.587521

0802.0765

On the local time of the asymmetric Bernoulli walk Dedicated to Professor Sándor Csörgő on his sixtieth birthday Endre Csáki111Research supported by the Hungarian National Foundation for Scientific Research, Grant No. K 61052 and K 67961. Alfréd Rényi Institute of Mathematics, Hungarian Academy of Sciences, Budapest, P.O.B. 127, H-1364, Hungary. E-mail address: csaki@renyi.hu Antónia Földes222Research supported by a PSC CUNY Grant, No. 66494-0035. Department of Mathematics, College of Staten Island, CUNY, 2800 Victory Blvd., Staten Island, New York 10314, U.S.A. E-mail address: foldes@mail.csi.cuny.edu Pál Révész1 Institut für Statistik und Wahrscheinlichkeitstheorie, Technische Universität Wien, Wiedner Hauptstrasse 8-10/107 A-1040 Vienna, Austria. E-mail address: reveszp@renyi.hu Abstract: We study some properties of the local time of the asymmetric Bernoulli walk on the line. These properties are very similar to the corresponding ones of the simple symmetric random walks in higher ($d\geq 3$) dimension, which we established in the recent years. The goal of this paper is to highlight these similarities. AMS 2000 Subject Classification: Primary 60G50; Secondary 60F15, 60J55. Keywords: transient random walk, local time, occupation time, strong theorems. ## 1\. Introduction The study of the local time (number of visits) of transient random walks started with the landmark papers of Dvoretzky and Erdős [8] and Erdős and Taylor [9], who investigated the properties of simple symmetric random walk in dimension $d\geq 3$, in which case the random walk is transient by Pólya theorem. For further results we mention the books by Feller [10], [11], Spitzer [15] and Révész [14]. In the recent years our investigations were concentrated on some of the fine properties of the local and occupation times of these walks. It is well known that the simple asymmetric Bernoulli walk on the line is also transient and as such it behaves similarly to other transient walks. The goal of the present paper to put into evidence that many of the fine properties of the local and occupation times which we studied for the $d(\geq 3)$ dimensional transient symmetric walks are really shared by the asymmetric one dimensional Bernoulli walk. Here we would like to discuss only the following three major topics. * • limit theorems for multiple visited points * • joint behavior of local and occupation times * • the local time around frequently visited points These results in higher dimension were presented in our papers [2], [3], [4]. In [5], a recent survey paper on these topics some of our present results were given without proof. In this paper we would like to collect all the results which we are having so far on the asymmetric Bernoulli walk. Clearly to give full proofs for all these results are very tedious but saying only, that proofs are similar to the symmetric $d$-dimensional walk case is unfair. So we take the middle way, namely we give some of the proofs with an emphasis on the differences between the two situations. The organization of the paper is as follows. In Section 2 we collect the relevant results on the simple symmetric $d$-dimensional walk. In Section 3 we present the new results on the local and occupation times of the asymmetric one dimensional Bernoulli walk. In Section 4 we will present some lemmas needed later in the proofs. In Section 5 we prove Theorem 3.2. The proofs of Theorems 3.3 and 3.4 will be given in Section 6, while in Section 7 the proofs of Theorems 3.5 and 3.6 are presented. Finally, Section 8 contains some remarks. ## 2\. Random walk in higher dimension Let $\\{\mathbf{S}_{n}\\}_{n=1}^{\infty}$ be a symmetric random walk starting at the origin $\mathbf{0}$ on the $d$-dimensional integer lattice $\mathcal{Z}_{d}$ where $d\geq 3$, i.e. $\mathbf{S}_{0}=\mathbf{0}$, $\mathbf{S}_{n}=\sum_{k=1}^{n}\mathbf{X}_{k}$, $n=1,2,\dots$, where $\mathbf{X}_{k},\,k=1,2,\dots$ are i.i.d. random variables with distribution $\mathbf{P}(\mathbf{X}_{1}=\mathbf{e}_{i})=\frac{1}{2d},\qquad i=1,2,\ldots,2d$ and $\\{\mathbf{e}_{1},\mathbf{e}_{2},...\mathbf{e}_{d}\\}$ is a system of orthogonal unit vectors in $\mathcal{Z}_{d}$ and $\mathbf{e}_{d+j}=-\mathbf{e}_{j},$ $j=1,2,\ldots,d.$ Define the local time of the walk by $\xi(\mathbf{z},n):=\\#\\{k:\,\,0<k\leq n,\,\,\,\mathbf{S}_{k}=\mathbf{z}\\},\quad n=1,2,\ldots,$ (2.1) where $\mathbf{z}$ is any lattice point of $\mathcal{Z}_{d}.$ Let $\xi(\mathbf{z},\infty):=\lim_{n\to\infty}\xi(\mathbf{z},n)$ be the total local time at $\mathbf{z}$ of the infinite path. The maximal local time of the walk up to time $n$ is defined as $\xi(n):=\max_{\mathbf{z}\in\mathcal{Z}_{d}}\xi(\mathbf{z},n),\quad n=1,2,\ldots$ (2.2) Define also the following quantities: $\eta(n):=\max_{0\leq k\leq n}\xi(\mathbf{S}_{k},\infty),\quad n=1,2,\ldots$ (2.3) Denote by $\gamma(n)=\gamma(n;d)$ the probability that in the first $n-1$ steps the $d$-dimensional path does not return to the origin. Then $1=\gamma(1)\geq\gamma(2)\geq...\geq\gamma(n)\geq...>0.$ (2.4) It was proved in [8] that Theorem A (Dvoretzky and Erdős [8]) For $d\geq 3$ $\lim_{n\to\infty}\gamma(n)=\gamma=\gamma(\infty;d)>0,$ (2.5) and ${\gamma}<\gamma(n)<{\gamma}+O(n^{1-d/2}).$ (2.6) Consequently $\mathbf{P}(\xi(\mathbf{0},n)=0,\,\xi(\mathbf{0},\infty)>0)=O\left(n^{1-d/2}\right)$ (2.7) as $n\to\infty$. So $\gamma$ is the probability that the $d$-dimensional simple symmetric random walk never returns to its starting point. For $d\geq 3$ (see Erdős and Taylor [9]) $\,\xi(\mathbf{0},\infty)$ has geometric distribution: $\mathbf{P}(\xi(\mathbf{0},\infty)=k)={\gamma}(1-{\gamma})^{k},\qquad k=0,1,2,\ldots$ (2.8) Erdős and Taylor [9] proved the following strong law for the maximal local time: Theorem B (Erdős and Taylor [9]) For $d\geq 3$ $\lim_{n\to\infty}\frac{\xi(n)}{\log n}=\lambda\hskip 28.45274pt\mathrm{a.s.},$ (2.9) where $\lambda=\lambda_{d}=-\frac{1}{\log(1-{\gamma})}.$ (2.10) We remark that (2.9) is also true if $\xi(n)$ is replaced by $\eta(n).$ Now we present some of our results for the local and occupation times for simple symmetric random walk in ${\cal Z}_{d}$, $d\geq 3$. We note however that Theorems E and H and the Proposition below are true for more general symmetric aperiodic random walk in ${\cal Z}_{d}$, $d\geq 3$. Erdős and Taylor [9] also investigated the properties of $Q(k,n):=\\#\\{\mathbf{z}:\ \mathbf{z}\in\mathcal{Z}_{d},\ \xi(\mathbf{z},n)=k\\},$ i.e. the cardinality of the set of points visited exactly $k$ times in the time interval $[1,n]$. Theorem C (Erdős and Taylor [9]) For $d\geq 3$ and for any $k=1,2,\ldots$ $\lim_{n\rightarrow\infty}{\frac{Q(k,n)}{n}}=\gamma^{2}(1-\gamma)^{k-1}\hskip 28.45274pt\mathrm{a.s.}$ (2.11) This was extended in [2] to a uniform law of large numbers: Theorem D (Csáki, Földes and Révész [2]) Let $d\geq 3$, and define $\displaystyle\mu=\mu(t)$ $\displaystyle:=$ $\displaystyle\gamma(1-\gamma)^{t-1},$ (2.12) $\displaystyle t_{n}$ $\displaystyle:=$ $\displaystyle[\lambda\log n-\lambda B\log\log n],\quad B>2.$ (2.13) Then we have $\lim_{n\rightarrow\infty}\max_{t\leq t_{n}}\left|{\frac{Q(t,n)}{n\gamma\mu(t)}}-1\right|=0\hskip 28.45274pt\mathrm{a.s.}$ (2.14) We introduce the following notations. For $\mathbf{z}\in\mathcal{Z}_{d}$ let $T_{\mathbf{z}}$ be the first hitting time of $\mathbf{z}$, i.e. $T_{\mathbf{z}}:=\min\\{i\geq 1:\mathbf{S}_{i}=\mathbf{z}\\}$ with the convention that $T_{\mathbf{z}}=\infty$ if there is no $i$ with $\mathbf{S}_{i}=\mathbf{z}$. Let $T=T_{\bf 0}$. In general, for a subset $A$ of $\mathcal{Z}_{d}$, let $T_{A}$ denote the first time the random walk visits $A$, i.e. $T_{A}:=\min\\{i\geq 1:\,\mathbf{S}_{i}\in A\\}=\min_{\mathbf{z}\in A}T_{\mathbf{z}}$. Let $\mathbf{P}_{\mathbf{z}}(\cdot)$ denote the probability of the event in the brackets under the condition that the random walk starts from $\mathbf{z}\in\mathcal{Z}_{d}$. We denote $\mathbf{P}(\cdot)=\mathbf{P}_{\mathbf{0}}(\cdot)$. Define $\gamma_{\mathbf{z}}:={\bf P}(T_{\mathbf{z}}=\infty).$ (2.15) Let $B(r)$ be the sphere of radius $r$ centered at the origin, i.e. $B(r):=\\{\mathbf{z}\in\mathcal{Z}_{d}:\|\mathbf{z}\|=r\\},$ and $B:=B(1)$ where $\|\cdot\|$ is the Euclidean norm. Introduce further $\mathbf{p}:=\mathbf{P}_{\mathbf{e}_{1}}(T_{B}<T).$ (2.16) In words, $\mathbf{p}$ is the probability that the random walk, starting from $\mathbf{e}_{1}$ (or any other points of $B$), returns to $B$ before reaching $\mathbf{0}$ (including the case $T_{B}<T=\infty$). It is not hard to show that $\mathbf{p}=1-\frac{1}{2d(1-\gamma)}.$ (2.17) For a set $A\subset\mathcal{Z}_{d}$ the occupation time of $A$ is defined by $\Xi(A,n):=\sum_{\mathbf{z}\in A}\xi(\mathbf{z},n).$ (2.18) Consider the translates of $A$, i.e. $A+\mathbf{u}=\\{\mathbf{z}+\mathbf{u}:\,\mathbf{z}\in A\\}$ with $\mathbf{u}\in\mathcal{Z}_{d}$ and define the maximum occupation time by $\Xi^{*}(A,n):=\max_{\mathbf{u}\in\mathcal{Z}_{d}}\Xi(A+\mathbf{u},n).$ (2.19) It was shown in [6] Theorem E (Csáki, Földes, Révész, Rosen and Shi [6]) For $d\geq 3$ and for any fixed finite set $A\subset\mathcal{Z}_{d}$ $\lim_{n\to\infty}\frac{\Xi^{*}(A,n)}{\log n}=\frac{-1}{\log(1-1/\Lambda_{A})}\qquad\hskip 28.45274pt\mathrm{a.s.},$ (2.20) where $\Lambda_{A}$ is the largest eigenvalue of the $|A|\times|A|$ matrix with elements $G(\mathbf{z}-\mathbf{u}),\qquad\mathbf{z},\mathbf{u}\in A,$ and $G(\mathbf{z})=\sum_{i=0}^{\infty}{\bf P}(\mathbf{S}_{i}=\mathbf{z}),\qquad\mathbf{z}\in{\cal Z}_{d}$ is the Green function of the walk. As a major tool for the above result it was proved that Proposition A (Csáki, Földes, Révész, Rosen and Shi [6]) ${\bf P}(\Xi(A,\infty)>k)=\sum_{j}h_{j}\left(\frac{\lambda_{j}-1}{\lambda_{j}}\right)^{k},\quad k=0,1,\ldots,$ (2.21) where $\lambda_{j}$ are the eigenvalues of the matrix $G_{A}$ and $h_{j}$ are certain coefficients calculated in terms of the eigenvectors. In particular, it was shown in [6] $\lim_{n\to\infty}\frac{\max_{{\bf z}\in{\cal Z}_{d}}\Xi({\bf z},n)}{\log n}=\frac{-1}{\log\left(\mathbf{p}+\frac{1}{2d}\right)}=:\kappa\hskip 28.45274pt\mathrm{a.s.},$ (2.22) where $\Xi({\bf z},n):=\Xi(B+{\bf z},n),$ (2.23) i.e. the occupation time of the unit sphere centered at ${\bf z}$. Note that in this notation ${\bf z}$ stands for the center of the unit sphere not for the one element set $\\{{\bf z}\\}.$ Furthermore if $A=\\{\mathbf{0},\mathbf{z}\\}$ is a two-point set, then the constant in (2.20) of Theorem E is $c_{A}=\frac{-1}{\log(1-1/\Lambda_{A})}=\frac{-1}{\log\left(1-\frac{\gamma}{2-\gamma_{\mathbf{z}}}\right)},$ where $\gamma_{\mathbf{z}}$ is the probability that the random walk, starting from zero, never visits $\mathbf{z}$. It can be seen that in this case $c_{A}<2\lambda$, showing that for large $n$ any point with fixed distance from a maximally visited point can not be maximally visited. This suggests to investigate the behavior of local and occupation times around frequently visited points. Naturally it would be interesting to investigate the joint behavior of the local time and/or occupation time of two sets in general. However this is a very difficult proposition. From the two special cases we discussed in [3] we mention here the following one. Consider the joint behavior of local time of a point and the occupation time of the unit sphere centered at the point. Define the set $\mathcal{B}$ on the plane as $\mathcal{B}:=\\{(x,y):\,y\geq x\geq 0;\,\,-y\log y+x\log(2dx)+(y-x)\log((y-x)/\mathbf{p})\leq 1\\},$ (2.24) where $\mathbf{p}$ was defined in (2.16) and its value in terms of $\gamma$ is given by (2.17). Theorem F (Csáki, Földes and Révész [3]) Let $d\geq 4$. For each $\varepsilon>0$ with probability 1 there exists an $n_{0}=n_{0}(\varepsilon)$ such that if $n\geq n_{0}$ then * • (i) $(\xi(\mathbf{z},n),\Xi(\mathbf{z},n))\in((1+\varepsilon)\log n)\mathcal{B},\quad\quad\forall\mathbf{z}\in\mathcal{Z}_{d}$ * • (ii) for any $(k,\ell)\in((1-\varepsilon)\log n)\mathcal{B}\cap\mathcal{Z}_{2}$ there exists a random $\mathbf{z}\in\mathcal{Z}_{2}$ for which $(\xi(\mathbf{z},n),\Xi(\mathbf{z},n))=(k,\ell+1).$ Theorem G (Csáki, Földes and Révész [3]) Let $d\geq 3$. For each $\varepsilon>0$ with probability 1 there exists an $n_{0}=n_{0}(\varepsilon)$ such that if $n\geq n_{0}$ then * • (i) $(\xi(\mathbf{S}_{j}+{\bf e}_{i},\infty),\Xi(\mathbf{S}_{j}+{\bf e}_{i},\infty))\in((1+\varepsilon)\log n)\mathcal{B},\quad\forall j=1,2,\ldots,n,\quad\forall i=1,2,\ldots,2d$ * • (ii) for any $(k,\ell)\in((1-\varepsilon)\log n)\mathcal{B}\cap\mathcal{Z}_{2}$ and for arbitrary $i\in\\{1,2,\ldots,2d\\}$ there exists a random integer $j=j(k,\ell)\leq n$ for which $(\xi(\mathbf{S}_{j}+{\bf e}_{i},\infty),\Xi(\mathbf{S}_{j}+{\bf e}_{i},\infty))=(k,\ell+1).$ It follows from these results that if the local time of a point is close to $\lambda\log n$, then the local times of each of its neighbors should be asymptotically equal to $\lambda(1-\gamma)\log n$ which is strictly less than $\lambda\log n$. In [4] we investigated whether similar results are true in a wider neighborhood, i.e. whether the local times of points on a certain sphere centered at a heavy point are asymptotically determined. As a positive answer we proved Theorem H (Csáki, Földes and Révész [4]) Let $d\geq 5$ and $k_{n}=(1-\delta_{n})\lambda\log n$. Let $r_{n}>0$ and $\delta_{n}>0$ be selected such that $\delta_{n}$ is non-increasing, $r_{n}$ is non-decreasing, and for any $c>0$ $r_{[cn]}/r_{n}<C$ with some $C>0$ and for $\beta_{n}:=r_{n}^{2d-4}\frac{\log\log n}{\log n}$ (2.25) $\lim_{n\to\infty}\beta_{n}=0,\qquad\lim_{n\to\infty}\delta_{n}r_{n}^{2d-4}=0.$ (2.26) Define the random set of points ${\cal A}_{n}=\\{\mathbf{z}\in Z^{d}:\xi(\mathbf{z},n)\geq k_{n}\\}.$ (2.27) Then we have $\lim_{n\to\infty}\max_{\mathbf{z}\in{\cal A}_{n}}\max_{\mathbf{u}\in S(r_{n})}\left|\frac{\xi(\mathbf{z}+\mathbf{u},n)}{m_{\mathbf{u}}\lambda\log n}-1\right|=0\quad{\rm a.s.},$ (2.28) where $S(r):=\\{\mathbf{u}\in{\cal Z}_{d}:\,\|\mathbf{u}\|\leq r\\}\quad{\it and}\quad m_{\mathbf{u}}:={\bf E}(\xi(\mathbf{u},\infty)\mid T<\infty)=\frac{(1-\gamma_{\mathbf{u}})^{2}}{1-\gamma}.$ Theorem I (Csáki, Földes and Révész [4]) Let $d\geq 3$ and $k_{n}=(1-\delta_{n})\lambda\log n$. Let $r_{n}>0$ and $\delta_{n}>0$ be selected such that $\delta_{n}$ is non-increasing, $r_{n}$ is non-decreasing, and for any $c>0$ $r_{[cn]}/r_{n}<C$ for some $C>0$ and for $\beta_{n}:=r_{n}^{2d-4}\frac{\log\log n}{\log n}$ (2.29) $\lim_{n\to\infty}\beta_{n}=0,\qquad\lim_{n\to\infty}\delta_{n}r_{n}^{2d-4}=0.$ (2.30) Define the random set of indices ${\cal B}_{n}=\\{j\leq n:\xi(\mathbf{S}_{j},\infty)\geq k_{n}\\}.$ (2.31) Then we have $\lim_{n\to\infty}\max_{j\in{\cal B}_{n}}\max_{\mathbf{u}\in S(r_{n})}\left|\frac{\xi(\mathbf{S}_{j}+\mathbf{u},\infty)}{m_{\mathbf{u}}\lambda\log n}-1\right|=0\quad{\rm a.s.}$ (2.32) ## 3 Simple asymmetric random walk on the line Consider a simple asymmetric random walk on the line $\\{S_{n}\\}_{n=0}^{\infty}$ starting at the origin, i.e. $S_{0}:=0$, $S_{n}:=\sum_{k=1}^{n}X_{k}$, $n=1,2,\dots$, where $X_{k},\,k=1,2,\dots$ are i.i.d. random variables with distribution ${\bf P}(X_{1}=1)=p\,\,\,{\rm and}\,\,\,{\bf P}(X_{1}=-1)=q\,(=1-p).$ (3.1) Without restricting generality we will suppose throughout the paper that $p>q\qquad{\rm and\,\,introduce}\quad h:=\frac{q}{p}.$ As it is well-known, this random walk is transient, i.e. with probability one we have $\lim_{n\to\infty}S_{n}=\infty$. There is a huge literature on such transient random walk. Some basic results are given in Feller [10], [11], Jordan [12], Spitzer [15], etc., some of these will be given in the next section. Let ${\cal Z}={\cal Z}_{1}$, i.e. the set of integers on the line. We define the local time by $\xi(z,n):=\\#\\{k:\,\,0<k\leq n,\,\,\,S_{k}=z\\},\quad z\in{\cal Z},\quad n=1,2,\ldots,$ (3.2) $\xi(z,\infty):=\lim_{n\to\infty}\xi(z,n),\quad\xi(n):=\max_{z\in{\cal Z}}\xi(z,n),\quad\eta(n):=\max_{0\leq j\leq n}\xi(S_{j},\infty),$ and the occupation time of a set $A\subset{\cal Z}$ by $\Xi(A,n):=\\#\\{k:\,\,0<k\leq n,\,\,\,S_{k}\in A\\}=\sum_{z\in A}\xi(z,n),\quad n=1,2,\ldots$ Concerning limit theorems for the local time, the analogue of Theorem A is simple, it will be given as Fact 2 in Section 4. It seems that no analogue of Theorem B can be found in the literature, though the following result can be proved from (4.11)- (4.13) just as Theorem B above of Erdős and Taylor. It will be also a trivial consequence of Theorem 3.3 and Theorem 3.4. ###### Theorem 3.1. For the simple asymmetric random walk $\lim_{n\to\infty}\frac{\xi(n)}{\log n}=\lim_{n\to\infty}\frac{\eta(n)}{\log n}=\frac{-1}{\log(2q)}=:\lambda_{0}\quad{\rm a.s.}$ (3.3) We do not know analogues of (2.20) and (2.21), but in certain particular cases the distribution can be expressed in a simple form. In the next section we give a version of the joint distribution of $\xi(z,\infty)$ and $\xi(0,\infty)$ and the distribution of $\Xi(\\{0,z\\},\infty)$ (Proposition 4.1). Furthermore, we present the joint distribution of $\xi(0,\infty)$ and $\Xi(0,\infty)$ (Proposition 4.2). Let $\Xi^{*}(A,n):=\max_{a\in\mathcal{Z}}\Xi(A+a,n).$ From (4.18) in Section 4, similarly to the proof of (2.20) via (2.21) in [6], we will show the following result. ###### Theorem 3.2. For $z>0$ integer $\lim_{n\to\infty}\frac{\Xi^{*}(\\{0,z\\},n)}{\log n}=\frac{-1}{\log\left(\frac{2q+h^{z/2}}{1+h^{z/2}}\right)}\quad{\rm a.s.}$ (3.4) Concerning Theorem C, define $\widetilde{Q}(k,n):=\\#\\{z\in\mathcal{Z}:\ \xi(z,n)=k\\}.$ We have from Pitt [13] that Theorem K (Pitt [13]) For $k=1,2,\ldots$ $\lim_{n\to\infty}\frac{\widetilde{Q}(k,n)}{n}=(1-2q)^{2}(2q)^{k-1}\qquad{\rm a.s.}$ The analogue of Theorem D, i.e. uniform law of large numbers for $\widetilde{Q}(k,n)$ remains an open problem. Next we formulate two theorems which correspond to Theorems F and G for the transient walk on the line. Let $B:=\\{-1,1\\}$, the one dimensional unit sphere around the origin. Just like in the higher dimensional situation we will denote $\Xi(z,n):=\Xi(B+z,n)=\Xi(\\{z-1,z+1\\},n)=\xi(z-1,n)+\xi(z+1,n),$ i.e. the occupation time of the unit sphere centered at $z\in\mathcal{Z}$. Introduce $g(x,y):=x\log x-y\log y+(y-x)\log(y-x)-x\log(2p)-y\log q$ (3.5) and define the set ${\cal D}$ by ${\cal D}:=\\{(x,y):y\geq x\geq 0;\,\,g(x,y)\leq 1\\}.$ (3.6) ###### Theorem 3.3. For each $\varepsilon>0$ with probability 1 there exists an $n_{0}=n_{0}(\varepsilon)$ such that if $n\geq n_{0}$ then (i) $(\xi(z,n),\Xi(z,n))\in((1+\varepsilon)\log n)\mathcal{D},\quad\quad\forall z\in\mathcal{Z}$ (ii) for any $(k,\ell)\in((1-\varepsilon)\log n)\mathcal{D}\cap\mathcal{Z}_{2}$ there exists a random $z\in\mathcal{Z}$ for which $(\xi(z,n),\Xi(z,n))=(k+1,\ell+2).$ ###### Theorem 3.4. For each $\varepsilon>0$ with probability 1 there exists an $n_{0}=n_{0}(\varepsilon)$ such that if $n\geq n_{0}$ then (i) $(\xi(S_{j},\infty),\Xi(S_{j},\infty))\in((1+\varepsilon)\log n)\mathcal{D},\quad\forall j=1,2,\ldots,n$ (ii) for any $(k,\ell)\in((1-\varepsilon)\log n)\mathcal{D}\cap\mathcal{Z}_{2}$ there exists a random integer $j=j(k,\ell)\leq n$ for which $(\xi(S_{j},\infty),\Xi(S_{j},\infty))=(k+1,\ell+2).$ From Theorem 3.3 the following consequence is easily obtained. ###### Corollary 3.1 With probability 1 for all possible sequence of integers $\\{z_{n}\\}$ the set of all possible limit points of $\left(\frac{\xi(z_{n},n)}{\log n},\frac{\Xi(z_{n},n)}{\log n}\right),\qquad n\to\infty$ is equal to ${\cal D}$. Finally, we state the following analogues of Theorems H and I: ###### Theorem 3.5. Define the random set of indices ${\cal A}_{n}:=\\{u\in{\cal Z}:\quad\xi(u,n)\geq(1-\delta_{n})\lambda_{0}\log n\\}.$ (3.7) Let $\alpha=\log(1/h),$ and select $c>0$ such that $\alpha c<1.$ If $\lim_{n\to\infty}\delta_{n}(\log n)^{\alpha c}=0,$ then we have $\lim_{n\to\infty}\max_{u\in{\cal A}_{n}}\max_{|z|\leq c\log\log n}\left|\frac{\xi(u+z,n)}{m_{z}\lambda_{0}\log n}-1\right|=0\quad{\rm a.s.},$ (3.8) where $m_{z}:=\left\\{\begin{array}[]{ll}&\frac{h^{|z|}}{2q}\ \quad{\rm if}\quad z\neq 0,\\\ &1\quad\quad\,\,{\rm if}\quad z=0.\end{array}\right.$ (3.9) ###### Theorem 3.6. Define the random set of indices ${\cal B}_{n}:=\\{j\leq n:\xi(S_{j},\infty)\geq(1-\delta_{n})\lambda_{0}\log n\\}.$ (3.10) Let $\alpha=\log(1/h),$ and select $c>0$ such that $\alpha c<1.$ If $\lim_{n\to\infty}\delta_{n}(\log n)^{\alpha c}=0,$ then we have $\lim_{n\to\infty}\max_{j\in{\cal B}_{n}}\max_{|z|\leq c\log\log n}\left|\frac{\xi(S_{j}+z,\infty)}{m_{z}\lambda_{0}\log n}-1\right|=0\quad{\rm a.s.},$ (3.11) where $m_{z}$ is defined in (3.9). ###### Corollary 3.2 Let $A\subset{\cal Z}$ be a fixed set. (i) If $u_{n}\in{\cal A}_{n}$, $n=1,2,\ldots$, then $\lim_{n\to\infty}\frac{\Xi(A+u_{n},n)}{\log n}=\lim_{n\to\infty}\frac{\sum_{x\in A}\xi(x+u_{n},n)}{\log n}=\lambda_{0}\sum_{x\in A}m_{x}\quad{\rm a.s.}$ (ii) If $j_{n}\in{\cal B}_{n}$, $n=1,2,\ldots$, then $\lim_{n\to\infty}\frac{\Xi(A+S_{j_{n}},\infty)}{\log n}=\lim_{n\to\infty}\frac{\sum_{x\in A}\xi(x+S_{j_{n}},\infty)}{\log n}=\lambda_{0}\sum_{x\in A}m_{x}\quad{\rm a.s.}$ ## 4 Preliminary facts and results Fact 1. For the probability of no return we have (cf. Feller [10]) ${\bf P}(S_{i}\neq 0,\,i=1,2,\ldots)=1-2q=p-q=:\gamma_{0}$ Let $T_{z}:=\min\\{i\geq 1:\,S_{i}=z\\},\quad T_{0}=:T,\quad z\in{\cal Z},$ the first hitting time of $z$. Denote by $\gamma_{0}(n)$ the probability that in the first $n-1$ steps $S_{n}$ does not return to the origin. Then just like in (2.4), we have $1=\gamma_{0}(1)\geq\gamma_{0}(2)\geq...\geq\gamma_{0}(n)\geq...>\gamma_{0}>0.$ (4.1) Fact 2. For $T$, the first return time to $0$ we have (cf. Feller [10]) ${\bf P}(T=2n)={2n\choose n}\frac{1}{2n-1}(pq)^{n}\sim\frac{(4pq)^{n}}{2\sqrt{\pi}n^{3/2}},\quad n\to\infty,$ (4.2) from which one can easily obtain that $\gamma_{0}(n)-\gamma_{0}={\bf P}(n\leq T<\infty)=O\left(\frac{(4pq)^{n/2}}{n^{3/2}}\right),\quad n\to\infty.$ (4.3) Remark: This is the analogue of Theorem A. It can be seen furthermore that ${\bf P}(T=2n,S_{1}=1)={\bf P}(T=2n,S_{1}=-1)=\frac{1}{2}{2n\choose n}\frac{1}{2n-1}(pq)^{n}=\frac{1}{2}{\bf P}(T=2n),$ (4.4) from which one easily obtains ${\bf P}(T<\infty,S_{1}=1)={\bf P}(T<\infty,S_{1}=-1)=\frac{1}{2}{\bf P}(T<\infty)=q,$ (4.5) and ${\bf P}(n\leq T<\infty,S_{1}=1)={\bf P}(n\leq T<\infty,S_{1}=-1)=\frac{1}{2}{\bf P}(n\leq T<\infty).$ (4.6) Now introduce $q(n):={\bf P}(T<n,S_{1}=1)=\mathbf{P}(T<\infty,S_{1}=1)-\mathbf{P}(n\leq T<\infty,S_{1}=1)$ $={\bf P}(T<n,S_{1}=-1)=\mathbf{P}(T<\infty,S_{1}=-1)-\mathbf{P}(n\leq T<\infty,S_{1}=-1).$ (4.7) Then we have $0<q-q(n)=O\left(\frac{(4pq)^{n/2}}{n^{3/2}}\right),\quad n\to\infty$ (4.8) as well. Recall the notation $h=q/p(<1).$ Fact 3. (see e.g. Feller [10]) For $z\in{\cal Z}$ we have ${\bf P}(T_{z}<\infty)=\left\\{\begin{array}[]{lll}&h^{-z}\ \quad{\rm if}\quad z<0,\\\ &2q\quad\quad{\rm if}\quad z=0,\\\ &1\quad\quad\,\ {\rm if}\quad z>0.\end{array}\right.$ (4.9) Fact 4. (see e.g. Spitzer [15], page 10) For the Green function $G(z)$ we have for $z\in{\cal Z}$ : $G(z)=\sum_{i=0}^{\infty}{\bf P}(S_{i}=z)=\left\\{\begin{array}[]{ll}&\frac{1}{p-q}h^{-z}\quad\,\,{\rm if}\quad z\leq 0,\\\ &\frac{1}{p-q}\quad\quad\quad{\rm if}\quad z>0.\end{array}\right.$ (4.10) ###### Lemma 4.1. For $n\geq 1$, $|j|\leq n$ we have $\mathbf{P}(S_{n}=j)\leq C_{1}\exp(-C_{2}n+C_{3}j),$ where the constants $C_{i}>0$, $i=1,2,3$, depend only on $p.$ Proof. Clearly $\mathbf{P}(S_{n}=j)$ differs from $0$ only if $j$ and $n$ have the same parity. So we will suppose that in the proof. $\displaystyle\mathbf{P}(S_{n}=j)$ $\displaystyle=$ $\displaystyle{n\choose\frac{n+j}{2}}p^{\frac{n+j}{2}}q^{\frac{n-j}{2}}\leq{n\choose[n/2]}(pq)^{n/2}\left(\frac{p}{q}\right)^{j/2}$ $\displaystyle\leq$ $\displaystyle C_{1}(4pq)^{n/2}\left(\frac{p}{q}\right)^{j/2}=C_{1}\exp(-C_{2}n+C_{3}j),$ where $C_{2}=-\frac{1}{2}\log(4pq),\qquad C_{3}=\frac{1}{2}\log\frac{p}{q}.$ $\Box$ For the distribution of the local time we have Fact 5. (cf. Dwass [7]) ${\bf P}(\xi(0,\infty)=k)=(2q)^{k}(1-2q),\qquad k=0,1,2,\ldots$ (4.11) For $z>0$ integer ${\bf P}(\xi(z,\infty)=k)=(2q)^{k-1}(1-2q),\qquad k=1,2,\ldots$ (4.12) and ${\bf P}(\xi(-z,\infty)=k)=\left\\{\begin{array}[]{ll}&1-h^{z}\quad\quad\qquad\qquad{\rm if}\quad k=0,\\\ &h^{z}(2q)^{k-1}(1-2q)\quad{\rm if}\quad k=1,2,\ldots\end{array}\right.$ (4.13) For the joint distribution of $\xi(z,\infty)$ and $\xi(0,\infty)$ we have ###### Proposition 4.1. For $z>0$, $k\geq 0$ integers we have ${\bf E}\left(e^{v\xi(z,\infty)},\,\xi(0,\infty)=k\right)=(1-2q)(2q)^{k}\varphi^{k}(v)\psi(v),$ (4.14) ${\bf E}\left(e^{v\xi(-z,\infty)},\,\xi(0,\infty)=k\right)=(1-2q)(2q)^{k}\varphi^{k}(v),\quad$ (4.15) for $v<-\log\left(1-\frac{1-2q}{1-h^{z}}\right),$ where $\varphi(v):=\frac{1-\frac{4q^{2}-h^{z}}{2q(1-2q)}\left(e^{v}-1\right)}{1-\frac{2q-h^{z}}{(1-2q)}\left(e^{v}-1\right)},$ (4.16) $\psi(v):=\frac{e^{v}}{1-\frac{2q-h^{z}}{(1-2q)}\left(e^{v}-1\right)}.$ (4.17) Moreover, ${\bf P}(\Xi(\\{0,z\\},\infty)=k)=\frac{1-2q}{2h^{z/2}}\left(\left(\frac{2q+h^{z/2}}{1+h^{z/2}}\right)^{k}-\left(\frac{2q-h^{z/2}}{1-h^{z/2}}\right)^{k}\right),$ (4.18) $k=1,2,\ldots$ ${\bf P}(\Xi(\\{0,-z\\},\infty)=k)=\frac{1-2q}{2}\left(\left(\frac{2q+h^{z/2}}{1+h^{z/2}}\right)^{k}+\left(\frac{2q-h^{z/2}}{1-h^{z/2}}\right)^{k}\right),$ (4.19) $k=0,1,2,\ldots$ Proof. Let us recall the gambler ruin (cf. Feller [10] or Jordan [12]): for $0\leq a<b<c$ $\mathbf{P}_{b}(T_{a}<T_{c})=1-\frac{1-h^{b-a}}{1-h^{c-a}}.$ (4.20) Let $z>0$ be an integer. Then by (4.20) $s_{z}:={\bf P}(T_{z}<T)=p\,{\bf P}_{1}(T_{z}<T)=p\frac{1-h}{1-h^{z}}=:P_{z}.$ On the other hand, $s_{-z}={\bf P}(T_{-z}<T)={\bf P}_{z}(T<T_{z})=q\frac{h^{z-1}-h^{z}}{1-h^{z}}=h^{z}P_{z}.$ Similarly, a simple calculation shows $q_{z}:={\bf P}(T<T_{z})={\bf P}_{z}(T_{z}<T)=1-P_{z}=:Q_{z},\qquad{\rm and}\qquad q_{-z}={\bf P}(T<T_{-z})=q_{z}.$ Let $Z(A)$ be the number of visits in the set $A\subset{\cal Z}$ in the first excursion away from $0$. In particular, for the one point set $\\{z\\}$ $Z(\\{z\\})=\xi(z,T).$ Note that $T=\infty$ is possible. Fact 6. (Baron and Rukhin [1]) For $z>0$ integer ${\bf P}(Z(\\{z\\})=j,\,T<\infty)={\bf P}(Z(\\{-z\\})=j,\,T<\infty)=\left\\{\begin{array}[]{ll}&Q_{z}\quad\quad\quad\quad{\rm if}\quad j=0,\\\ &h^{z}P_{z}^{2}Q_{z}^{j-1}\quad{\rm if}\quad j=1,2,\ldots\end{array}\right.$ (4.21) ${\bf P}(Z(\\{z\\})=j,\,T=\infty)=(1-2q)P_{z}Q_{z}^{j-1},\qquad j=1,2,\ldots$ (4.22) ${\bf P}(Z(\\{-z\\})=0,\,T=\infty)=(1-2q).\qquad$ (4.23) It can be seen furthermore that ${\bf E}(\xi(z,T),\,T<\infty)={\bf E}(Z(\\{z\\}),\,T<\infty)={\bf E}(Z(\\{-z\\}),\,T<\infty)=h^{z},$ hence ${\bf E}(\xi(z,T)|T<\infty)={\bf E}(Z(\\{z\\})|T<\infty)={\bf E}(Z(\\{-z\\})|T<\infty)=m_{z},$ where $m_{z}$ is defined by (3.9). Now (4.14) can be calculated from (4.21) and (4.22) by using that ${\bf E}\left(e^{v\xi(z,\infty)},\xi(0,\infty)=k\right)=\left({\bf E}(e^{vZ(\\{z\\})},\,T<\infty)\right)^{k}{\bf E}(e^{vZ(\\{z\\})},\,T=\infty).$ It is easy to see that ${\bf E}(e^{vZ(\\{z\\})},\,T<\infty)=Q_{z}+\frac{h^{z}P^{2}_{z}e^{v}}{1-Q_{z}e^{v}}=2q\varphi(v),$ and ${\bf E}(e^{vZ(\\{z\\})},\,T=\infty)=(1-2q)\frac{e^{v}P_{z}}{1-Q_{z}e^{v}}=(1-2q)\psi(v).$ To ease the computation we remark that $h^{z}P^{2}_{Z}-Q^{2}_{z}=\frac{h^{z}-4q^{2}}{1-h^{z}.}$ Similarly we get (4.15) from (4.21) and (4.23). From (4.14) - (4.15), substituting $w=e^{v}$, one can find ${\bf E}(w^{\xi(0,\infty)+\xi(z,\infty)})=\frac{(1-2q)P_{z}w}{1-2Q_{z}w+(Q_{z}^{2}-h^{z}P_{z}^{2})w^{2}},$ (4.24) and ${\bf E}(w^{\xi(0,\infty)+\xi(-z,\infty)})=\frac{(1-2q)(1-Q_{z}w)}{1-2Q_{z}w+(Q_{z}^{2}-h^{z}P_{z}^{2})w^{2}},$ (4.25) consequently (4.18) and (4.19) can be obtained by expanding the right-hand side into powers of $w$. $\Box$ The next result concerns the joint distribution of the local time of the origin and the occupation time of the unit sphere $B$. ###### Proposition 4.2. For $K=0,1,\ldots,\,L=K+1,K+2,\ldots$ we have ${\bf P}(\Xi(0,\infty)=L,\xi(0,\infty)=K)={L-1\choose K}(2p)^{K}q^{L-1}p(1-2q)=:p(L,K).$ (4.26) ${\bf P}(\Xi(0,n)=L,\xi(0,n)=K)\leq p(L,K),$ (4.27) ${\bf P}(\Xi(0,\infty)=L)=(q+2pq)^{L-1}p(1-2q),\quad L=1,2,\ldots$ (4.28) Moreover, for the occupation time of the set $A_{0}:=\\{-1,0,1\\}$ we have ${\bf P}(\Xi(A_{0},\infty)=\ell)=p(1-2q)\left(\frac{q}{2}\right)^{\ell-1}\frac{(1+\beta)^{\ell}-(1-\beta)^{\ell}}{2\beta},\qquad\ell=1,2,\ldots,$ (4.29) where $\beta=\sqrt{1+\frac{8p}{q}}.$ Proof. The probability that the walk goes to $B$ and returns to $0$ immediately, is $2pq.$ However, before returning, the walk can make one or more outward excursions. If the walk starts outward from $B$, it returns with probability $q,$ independently whether it starts from $1$ or $-1.$ Altogether we need $K$ trips from zero to $B$ and back to $0$ and $L-K-1$ outward excursions. Every such arrangement, independently of the order of occurrences, has probability $(2pq)^{K}q^{L-K-1}.$ The number of ways how we can order the outward excursions and the inward ones is ${L-1\choose K}.$ After the $K$-th return to zero the walk must go to $1$ (some of the outward excursion might happen at this time) and then to infinity which has probability $p(1-2q)$, proving our first statement. Now (4.27) follows from (4.26) and (4.8). With a similar argument one can show that if we start the walk at $-1$, then for $K=1,2,\ldots,L=K,K+1,\ldots$ $p_{-1}(L,K):=\mathbf{P}_{-1}(\Xi(0,\infty)=L,\xi(0,\infty)=K)={L\choose K}(2p)^{K-1}q^{L-1}p^{2}(1-2q)$ (4.30) (which does not include the very first visit at $-1$). Similarly for $K=1,2,\ldots,L=K,K+1,\ldots$ $p_{1}(L,K):=\mathbf{P}_{1}(\Xi(0,\infty)=L,\xi(0,\infty)=K)={L\choose K}(2p)^{K-1}q^{L}p(1-2q)$ (4.31) (which does not include the very first visit at $1$), and for $L=0,1,\ldots$ $p_{1}(L,0):=\mathbf{P}_{1}(\Xi(0,\infty)=L,\xi(0,\infty)=0)=q^{L}(1-2q).$ (4.32) Finally, we get (4.29) from ${\bf E}(w^{\Xi(A_{0},\infty)})={\bf E}(w^{(\Xi(0,\infty)+\xi(0,\infty))})=\frac{p(1-2q)w}{1-qw-2pqw^{2}}$ which follows from ${\bf E}(e^{vZ(B)},T<\infty)=\frac{2pqe^{v}}{1-qe^{v}}$ and ${\bf E}(e^{vZ(B)},T=\infty)=p(1-2q)\frac{e^{v}}{1-qe^{v}},$ similarly as we got (4.18) in Proposition 4.1. $\Box$ We will need some basic observations about the reversed walk. By the reversed path of $(S_{0},S_{1},\ldots,S_{j})$ we mean the path $(0,S_{j-1}-S_{j},S_{j-2}-S_{j},\ldots,S_{0}-S_{j})=:(S_{0}^{*},S_{1}^{*},\ldots,S_{j}^{*})$, e.g. $X_{i}^{*}=S_{j-i}-S_{j-i+1}$ for $1\leq i\leq j.$ Then of course $X_{i}^{*},\quad 1\leq i\leq j$ are i.i.d. random variables with ${\bf P}(X^{*}_{1}=1)=q\,\,\,{\rm and}\,\,\,P(X^{*}_{1}=-1)=p=1-q,\quad p>q.$ (4.33) If needed this can be extended to an infinite path $(S_{0}^{*},S_{1}^{*},\ldots,S_{n}^{*},\ldots)$, i.e. $\\{S^{*}_{n}\\}_{n=0}^{\infty}$ is a walk starting at the origin $S^{*}_{0}=0$, with $S^{*}_{n}=\sum_{k=1}^{n}X^{*}_{k}$, $n=1,2,\ldots$, where $X^{*}_{k},\,$ are defined above for $k\leq j$ and for $k>j$ they are an arbitrary sequence of i.i.d. random variables (also independent from the previous ones) with the above distribution. Then we clearly have for all $z$ Fact 7. $\mathbf{P}(\xi^{*}(-z,\infty)=k)=\mathbf{P}(\xi(z,\infty)=k).$ (4.34) Consequently, under the conditions of Proposition 4.1 we have for $z=1,2,\ldots$, $k=0,1,2,\ldots$ and $v<-\log Q_{z}$ ${\bf E}\left(e^{v\xi^{*}(-z,\infty)},\,\xi^{*}(0,\infty)=k\right)=(2q)^{k}(1-2q)\varphi^{k}(v)\psi(v),\quad$ (4.35) ${\bf E}\left(e^{v\xi^{*}(z,\infty)},\,\xi^{*}(0,\infty)=k\right)=(2q)^{k}(1-2q)\varphi^{k}(v).\quad$ (4.36) ## 5 Proof of Theorem 3.2 ###### Lemma 5.1. Let $\theta=-\log\frac{2q+h^{z/2}}{1+h^{z/2}}.$ There exist $u_{0}>0$, $c_{1}>0$, $c_{2}>0$ such that for all $u\geq u_{0}$, $n\geq u^{2}$ we have $c_{1}e^{-\theta u}\leq{\bf P}(\Xi(\\{0,z\\},n)\geq u)\leq{\bf P}(\Xi(\\{0,z\\},\infty)\geq u)\leq c_{2}e^{-\theta u}.$ (5.1) Proof. The second inequality in (5.1) is obvious, and since from (4.18) we can see ${\bf P}(\Xi(\\{0,z\\},\infty)\geq u)\sim ce^{-\theta u},\qquad u\to\infty,$ with some constant $c$, we have also the third inequality in (5.1). To show the first inequality in (5.1), we note that ${\bf P}(\Xi(\\{0,z\\},\infty)\geq u)\leq{\bf P}(\Xi(\\{0,z\\},n)\geq u)+{\bf P}(\cup_{k=n+1}^{\infty}\\{S_{k}=0\\})+{\bf P}(\cup_{k=n+1}^{\infty}\\{S_{k}=z\\}).$ By Lemma 4.1 for $j=0,z$ ${\bf P}(\cup_{k=n+1}^{\infty}\\{S_{k}=j\\})\leq C_{1}e^{-Cn},$ with some $C>0,\,C_{1}>0$. Thus, by (4.18) we have ${\bf P}(\Xi(\\{0,z\\},n)\geq u)\geq{\bf P}(\Xi(\\{0,z\\},\infty)\geq u)-2C_{1}e^{-Cn}\geq ce^{-\theta u}-2C_{1}e^{-Cu^{2}}\geq C_{2}e^{-\theta u}$ for large enough $u$ with some $C_{2}>0$. $\Box$ Now we turn to the proof of Theorem 3.2. First we prove an upper bound in (3.4). Since $\sum_{a=-\infty}^{z}\xi(a,\infty)<\infty\qquad{\rm and}\qquad\sum_{a=n+1}^{\infty}\xi(a,n)=0$ almost surely, it suffices to show the upper bound with $\Xi^{*}(\\{0,z\\},n)$ replaced by $\widetilde{\Xi}^{*}(\\{0,z\\},n):=\max_{0\leq a\leq n}\Xi(\\{a,a+z\\},n).$ Since ${\bf P}(\Xi(\\{a,a+z\\},n)>u)\leq{\bf P}(\Xi(\\{0,z\\},n)>u-1)$ for $a=1,2,\ldots,$ we have by Lemma 5.1 for large enough $n$ ${\bf P}\left(\widetilde{\Xi}^{*}(\\{0,z\\},n)\geq\frac{1+\varepsilon}{\theta}\log n\right)\leq(n+1){\bf P}\left(\Xi(\\{0,z\\},n)\geq\frac{1+\varepsilon}{\theta}\log n-1\right)$ $\leq(n+1){\bf P}\left(\Xi(\\{0,z\\},\infty)\geq\frac{1+\varepsilon}{\theta}\log n-1\right)\leq\frac{c}{n^{\varepsilon}}.$ Applying this for the subsequence $n_{k}=k^{2/\varepsilon}$, using Borel- Cantelli lemma and the monotonicity of $\Xi^{*}(\\{0,z\\})$, $\varepsilon$ being arbitrary, we obtain $\limsup_{n\to\infty}\frac{\widetilde{\Xi}^{*}(\\{0,z\\},n)}{\log n}\leq\frac{-1}{\log\left(\frac{2q+h^{z/2}}{1+h^{z/2}}\right)}\quad{\rm a.s.}$ implying the upper bound in (3.4). Now we show the lower bound in (3.4). Let $k(n)=[\log n]^{3}$, $N_{n}=[n/k(n)]$, $t_{i,n}=ik(n)$, $i=0,1,\ldots,N_{n}-1$. We have $\Xi^{*}(\\{0,z\\},n)\geq\max_{0\leq i\leq N_{n}-1}Z_{i},$ where $Z_{i}:=\Xi(\\{S_{t_{i,n}},S_{t_{i,n}}+z\\},t_{i+1,n})-\Xi(\\{S_{t_{i,n}},S_{t_{i,n}}+z\\},t_{i,n}).$ $Z_{i},\,i=0,1,\ldots N_{n}-1$ are i.i.d. random variables, distributed as $\Xi(\\{0,z\\},k(n))$, so we get by Lemma 5.1 ${\bf P}\left(\Xi^{*}(\\{0,z\\},n)\leq\frac{1-\varepsilon}{\theta}\log n\right)\leq{\bf P}\left(\max_{0\leq i\leq N_{n}-1}Z_{i}\leq\frac{1-\varepsilon}{\theta}\log n\right)$ $\leq\left(1-{\bf P}\left(\Xi(\\{0,z\\},k(n))\geq\frac{1-\varepsilon}{\theta}\log n\right)\right)^{N_{n}}\leq\left(1-c_{1}e^{-(1-\varepsilon)\log n}\right)^{N_{n}}\leq e^{-c_{1}n^{\varepsilon}/k(n)}.$ The lower bound in (3.4) follows by Borel-Cantelli lemma. $\Box$ ## 6 Proof of Theorems 3.3 and 3.4 From (4.26) (4.30), (4.31) and Stirling formula we conclude the following limit relations. ###### Lemma 6.1. For $y\geq x\geq 0$ we have $\lim_{n\to\infty}\frac{\log p([y\log n]+1,[x\log n])}{\log n}=\lim_{n\to\infty}\frac{\log p_{1}([y\log n],[x\log n])}{\log n}$ $=\lim_{n\to\infty}\frac{\log p_{-1}([y\log n]+1,[x\log n]+1)}{\log n}=-g(x,y),$ (6.1) where $g(x,y)$ is defined by (3.5). Consequently, the probability ${\bf P}(\Xi(0,\infty)=[y\log n]+1,\,\xi(0,\infty)=[x\log n])$ is of order $1/n$, if $(x,y)$ satisfies the basic equation $g(x,y)=1,\qquad y\geq x\geq 0.$ (6.2) The following lemma describes the main properties of the boundary of the set ${\cal D}.$ ###### Lemma 6.2. (i) For the maximum value of $x,y$, satisfying (6.2), we have $\displaystyle x_{\max}$ $\displaystyle=$ $\displaystyle{\frac{-1}{\log(2q)}=\lambda_{0},}$ (6.3) $\displaystyle y_{\max}$ $\displaystyle=$ $\displaystyle{\frac{-1}{\log(q(1+2p))}}=:\kappa_{0}.$ (6.4) (ii) If $x=x_{\max}=\frac{-1}{\log(2q)}$, then $y=\frac{-1}{p\log(2q)}$. If $y=y_{\max}=\kappa_{0}$, then $x=(2\kappa_{0}p)/(2p+1)$. If $x=0$, then $y=-1/\log q$. (iii) For given $x$, the equation (6.2) has one solution in $y$ for $0\leq x<-1/\log(2pq)$ and for $x=\lambda_{0}$, and two solutions in $y$ for $\,-1/\log(2pq)\leq x<\lambda_{0}$. Proof. (i) First consider $x$ as a function of $y$ satisfying (6.2). We seek the maximum, where the derivative $x^{\prime}(y)=0$. Differentiating (6.2) and putting $x^{\prime}=0$, a simple calculation leads to $-\log y+\log(y-x)-\log q=0,$ i.e. $y=x/p.$ It can be seen that this is the value of $y$ when $x$ takes its maximum. Substituting this into (6.2), we get $x_{\max}=\frac{-1}{\log(2q)},$ verifying (6.3). Next consider $y$ as a function of $x$ and maximize $y$ subject to (6.2). Again, differentiating (6.2) with respect to $x$ and putting $y^{\prime}=0$, we get $-\log(y-x)+\log x-\log(2p)=0$ from which $x=(2py)/(1+2p)$. Substituting in (6.2) we get $y_{\max}=\kappa_{0}$. This completes the proof of Lemma 6.2(i) and the first two statements in Lemma 6.2(ii). A simple calculation shows that if $x=0$ then $y=-1/\log q$. Now we turn to the proof of Lemma 6.2(iii). For given $0\leq x\leq\lambda_{0}$ consider $g(x,y)$ as a function of $y$. We have ${\frac{\partial g}{\partial y}}=\log{\frac{y-x}{qy}}$ and this is equal to zero if $y=x/p$. It is easy to see that $g$ takes a minimum here and is decreasing if $y<x/p$ and increasing if $y>x/p$. Moreover, ${\frac{\partial^{2}g}{\partial y^{2}}}={\frac{1}{y-x}}-{\frac{1}{y}}>0,$ hence $g$ is convex from below. We have for $0<x<\lambda_{0},$ that this minimum is $g\left(x,\frac{x}{p}\right)=x\log(1/(2q))=\frac{x}{\lambda_{0}}<1,$ and $g(x,x)=-x\log(2pq)\left\\{\begin{array}[]{ll}&<1\,\,\mathrm{if}\,\,x<-1/\log(2pq),\\\ &=1\,\,\mathrm{if}\,\,x=-1/\log(2pq),\\\ &>1\,\,\mathrm{if}\,\,x>-1/\log(2pq).\end{array}\right.$ This shows that equation (6.2) has one solution if $0\leq x<-1/\log(2pq)$ and two solutions if $-1/\log(2pq)\leq x<\lambda_{0}$. For $x=\lambda_{0}$, it can be seen that $y=\lambda_{0}/p$ is the only solution of $g(x,y)=1$. The proof of Lemma 6.2 is complete. $\Box$ Proof of Theorem 3.4(i). Obviously, for $z=1,2,\ldots$ we have for $K=1,2,\ldots,\,L=K+1,K+2,\ldots$ ${\bf P}(\Xi(z,\infty)=L,\,\xi(z,\infty)=K)={\bf P}_{-1}(\Xi(0,\infty)=L-1,\,\xi(0,\infty)=K)$ and for $K=0,1,\ldots,\,L=K+1,K+2,\ldots$ ${\bf P}(\Xi(-z,\infty)=L,\,\xi(-z,\infty)=K)\leq{\bf P}_{1}(\Xi(0,\infty)=L-1,\,\xi(0,\infty)=K).$ Hence for $(k,\ell)\notin((1+{\varepsilon})\log n){\cal D}$ and $z\in{\cal Z}$, as $g(cx,cy)=cg(x,y)$ for any $c>0$, we have by (4.26)–(4.31) and Lemma 6.1 ${\bf P}(\xi(z,\infty)=k,\Xi(z,\infty)=\ell)\leq\frac{c}{n^{1+{\varepsilon}}}.$ (6.5) Consequently, by Fact 5, (4.28), (6.5) we have $\mathbf{P}(\xi(S_{j},\infty),\,\Xi({S_{j}},\infty))\not\in((1+\varepsilon)\log n)\mathcal{D})$ $\leq\sum_{{(k,\ell)\not\in((1+\varepsilon)\log n){\cal D}\atop k\leq(1+\varepsilon)\lambda_{0}\log n}\atop\ell\leq(1+\varepsilon)\kappa_{0}\log n}\mathbf{P}(\xi(S_{j},\infty)=k,\Xi(S_{j},\infty)=\ell)$ $+\sum_{k>(1+\varepsilon)\lambda_{0}\log n}\mathbf{P}(\xi(S_{j},\infty)=k)+\sum_{\ell>(1+\varepsilon)\kappa_{0}\log n}\mathbf{P}(\Xi(S_{j},\infty)=\ell)$ $\leq\frac{c\log^{2}n}{n^{1+\varepsilon}}+\sum_{k>(1+\varepsilon)\lambda_{0}\log n}c(2q)^{k}+\sum_{\ell>(1+\varepsilon)\kappa_{0}\log n}c(q+2pq)^{\ell}\leq\frac{c}{n^{1+\varepsilon/2}}.$ where in the above computation $c$ is an unimportant constant. We continue denoting such constants by the same letter $c$, the value of which might change from line to line. Selecting $n_{r}=r^{4/\varepsilon},$ we have $\mathbf{P}(\cup_{j\leq n_{r+1}}\\{(\xi(S_{j},\infty),\Xi(S_{j},\infty))\not\in((1+\varepsilon)\log n_{r})\mathcal{D}\\})\leq{\frac{c}{n_{r}^{\varepsilon/2}}}=\frac{c}{r^{2}}.$ (6.6) This combined with the Borel-Cantelli lemma shows that with probability 1 for all large $r$ and $j\leq n_{r+1}$ $(\xi(S_{j},\infty),\Xi(S_{j},\infty))\in((1+\varepsilon)\log n_{r})\mathcal{D}.$ It follows that with probability 1 there exists an $n_{0}$ such that if $n\geq n_{0}$ then $(\xi(S_{j},\infty),\Xi(S_{j},\infty))\in((1+\varepsilon)\log n)\mathcal{D}$ for all $j\leq n$. This proves (i) of Theorem 3.4. $\Box$ Proof of Theorem 3.3(i). The proof is similar to that of Theorem 1.1(i) in [3]. Define the following events for $j\leq n$: $\displaystyle B(j,n)$ $\displaystyle:=$ $\displaystyle\\{(\xi(S_{j},n),\Xi(S_{j},n))\notin((1+\varepsilon)\log n)\mathcal{D}\\},$ (6.7) $\displaystyle B^{*}(j,n)$ $\displaystyle:=$ $\displaystyle\\{(\xi(S_{j},j),\Xi(S_{j},j))\notin((1+\varepsilon)\log n)\mathcal{D}\\},$ (6.8) $\displaystyle C(j,n)$ $\displaystyle:=$ $\displaystyle\\{S_{m}\neq S_{j},m=j+1,\ldots,n\,\\},$ (6.9) $\displaystyle D(j,n)$ $\displaystyle:=$ $\displaystyle\\{\Xi(S_{j},\infty)>\Xi(S_{j},n)\\}.$ (6.10) Considering the reverse random walk starting from $S_{j}$, i.e. $S_{r}^{\prime}=S_{j-r}-S_{j}$, $r=0,1,\ldots,j$, we remark $\xi(S_{j},j)=\xi^{\prime}(0,j)$, if $S_{j}=0$, $\xi(S_{j},j)=\xi^{\prime}(0,j)+1$, if $S_{j}\neq 0$, $\Xi(S_{j},j)=\Xi^{\prime}(0,j),$ where $\Xi^{\prime}$ is the occupation time of the unit sphere of the random walk $S^{\prime}$. From this we can follow the proof of Theorem 1.1(i) in [3], using (4.27) and (6.1) instead of (2.18) and (3.1) in [3] and applying Theorem 3.4(i) instead of Theorem 1.2(i) in [3]. $\Box$ Proof of Theorems 3.3(ii) and 3.4(ii). We will say that $S_{i}$ is new if $\max_{0\leq m<i}S_{m}<S_{i}.$ (6.11) ###### Lemma 6.3. Let $\nu_{n}$ denote the number of new points up to time n. Then $\lim_{n\to\infty}\frac{\nu_{n}}{n}=1-2q\qquad{\rm a.s.}$ Proof: Let $Z_{i}:=\left\\{\begin{array}[]{ll}1&\,{\rm if}\,\,S_{i}\,\,{\rm is}\,\,{\rm new}\\\ 0&\,\mathrm{otherwise.}\end{array}\right.$ Then $\nu_{n}=\sum_{i=1}^{n}Z_{i}$. $\mathbf{E}(\nu_{n}^{2})=\mathbf{E}\left(\sum_{j=1}^{n}\sum_{i=1}^{n}Z_{j}Z_{i}\right)=\mathbf{E}\left(\sum_{j=1}^{n}Z_{j}\right)+2\mathbf{E}\left(\sum_{j=1}^{n}\sum_{i=1}^{j-1}Z_{j}Z_{i}\right)$ $\leq n+2\sum_{j=1}^{n}\sum_{i=1}^{j-1}\mathbf{P}(Z_{i}=1)\mathbf{P}(Z_{j-i}=1).$ Considering the reverse random walk from $S_{i}$ to $S_{0}=0$, we see that the event $\\{Z_{i}=1\\}$ is equivalent to the event that this reversed random walk starting from $0$ does not return to $0$ in time $i.$ We remark that for the reversed walk the probability of stepping to the right is $q$ and stepping to left is $p.$ Using (4.3) and observing that it remains true for the reversed random walk as well, we get $\mathbf{P}(Z_{i}=1)=\gamma_{0}+O((4pq)^{i/2})=1-2q+O((4pq)^{i/2}).$ Hence $\mathbf{E}(\nu_{n}^{2})\leq n+2\sum_{j=1}^{n}\sum_{i=1}^{j-1}\left(1-2q+O(4pq)^{i/2}\right)\left((1-2q)+O(4pq)^{(j-i)/2}\right)$ $=n(n-1)(1-2q)^{2}+O(n).$ As ${\bf E}(\nu_{n})=n(1-2q)+O(1)$, we have $Var(\nu_{n})=O(n).$ By Chebyshev’s inequality we get that $P|\nu_{n}-n(1-2q)|>{\varepsilon}n)\leq O\left(\frac{1}{n}\right).$ Considering the subsequence $n_{k}=k^{2}$ and using the Borel-Cantelli lemma and monotonicity of $\nu_{n}$, we obtain the lemma. $\Box$ To show Theorem 3.4(ii), let $\\{a_{n}\\}$ and $\\{b_{n}\\}$ ($a_{n}\log n\ll b_{n}\ll n$ ) be two sequences to be chosen later. Define $\theta_{1}=\min\\{i>b_{n}:S_{i}\,\,{\rm is}\,\,{\rm new}\\},$ $\theta_{m}=\min\\{i>\theta_{m-1}+b_{n}:S_{i}\,\,{\rm is}\,\,{\rm new}\\},\quad m=2,3,\dots$ and let $\nu_{n}^{\prime}$ be the number of $\theta_{m}$ points up to time $n-b_{n}$. Obviously $\nu_{n}^{\prime}(b_{n}+1)\geq\nu_{n}$, hence $\nu_{n}^{\prime}\geq\nu_{n}/(b_{n}+1)$ and it follows from Lemma 6.3 that for $c<1-2q$, we have with probability 1 that $\nu_{n}^{\prime}>u_{n}:=cn/(b_{n}+1)$ except for finitely many $n$. Recall that $B=\\{-1,1\\}$ denotes the unit sphere around $0$. Let $\rho_{0}^{i}=0,\qquad\rho_{h}^{i}=\min\\{j>\rho_{h-1}^{i}:S_{\theta_{i}+j}\in S_{\theta_{i}}+B\\},\quad h=1,2,\ldots,$ i.e. $\rho_{h}^{i},\,h=1,2,\ldots$ are the times when the random walk visits the unit sphere around $S_{\theta_{i}}$. For a fixed pair of integers $(k,\ell)$ define the following events: $\displaystyle A_{i}$ $\displaystyle:=$ $\displaystyle A_{i}(k,\ell)=\\{\xi(S_{\theta_{i}}+1,\theta_{i}+\rho_{\ell+1}^{i})=k+1,\Xi(S_{\theta_{i}}+1,\theta_{i}+\rho_{\ell+1}^{i})=\ell+2,$ $\displaystyle\rho_{h}^{i}-\rho_{h-1}^{i}<a_{n},\,h=1,\ldots,\ell+1,\,S_{j}\not\in S_{\theta_{i}}+B,\,j=\theta_{i}+\rho_{\ell+1}^{i}+1,\ldots,\theta_{i}+b_{n}\\},$ $\displaystyle B_{i}$ $\displaystyle:=$ $\displaystyle B_{i}(k,\ell)=\\{S_{j}\not\in S_{\theta_{i}}+B,\,\,j>\theta_{i}+b_{n}\\},$ $\displaystyle C_{n}$ $\displaystyle:=$ $\displaystyle C_{n}(k,\ell)=A_{1}B_{1}+\overline{A_{1}}A_{2}B_{2}+\overline{A_{1}}\,\,\overline{A_{2}}A_{3}B_{3}+\ldots+\overline{A_{1}}\ldots\overline{A_{u_{n}-1}}A_{u_{n}}B_{u_{n}},$ where $\overline{A}$ denotes the complement of $A$. Then we have $\mathbf{P}(A_{i})=\mathbf{P}(A_{1})$ and $\mathbf{P}(A_{i}B_{i})=\mathbf{P}(A_{1}B_{1})$, $i=2,3,\dots$ and $\mathbf{P}(C_{n})=\mathbf{P}(A_{1}B_{1})\sum_{j=0}^{u_{n}-1}(1-\mathbf{P}(A_{1}))^{j}=\frac{\mathbf{P}(A_{1}B_{1})}{\mathbf{P}(A_{1})}(1-(1-\mathbf{P}(A_{1}))^{u_{n}}),$ $\mathbf{P}(\overline{C_{n}})\leq 1-\frac{\mathbf{P}(A_{1}B_{1})}{\mathbf{P}(A_{1})}+e^{-u_{n}\mathbf{P}(A_{1})}.$ $A_{1}B_{1}$ is the event that starting from the new point $S_{\theta_{1}}$, the random walk visits $S_{\theta_{1}}+1$ exactly $k+1$ times, while it visits the unit sphere around this point exactly $\ell+2$ times (including the initial visit at $S_{\theta_{1}}$) and all the time intervals between consecutive visits are less than $a_{n}$. Since the return to the sphere via its center takes only 2 steps, we have to control only the returns from outside. Similarly to (4.30), one can see $\mathbf{P}(A_{1}B_{1})={{\ell+1}\choose{k+1}}(q(a_{n}))^{\ell-k}(2pq)^{k}p^{2}\left(1-2q\right),$ $={{\ell+1}\choose{k+1}}(q+O((4pq)^{a_{n}}))^{\ell-k}(2pq)^{k}p^{2}\left(1-2q\right)$ and $\mathbf{P}(A_{1})={{\ell+1}\choose{k+1}}(2pq)^{k}(q+O((4pq)^{a_{n}}))^{\ell-k}p^{2}\left(1-2q+O((4pq)^{b_{n}-\ell a_{n}})\right),$ where $q(n)$ is given in (4.7). Using $a_{n}\log n\ll b_{n}\ll n$ $\frac{\mathbf{P}(A_{1}B_{1})}{\mathbf{P}(A_{1})}=1+O\left((4pq)^{c_{1}b_{n}/2}\right),$ for some $c_{1}>0$ depending only on $p,$ hence $\mathbf{P}(\overline{C_{n}})\leq e^{-c_{1}b_{n}}+e^{-cn\mathbf{P}(A_{1})/b_{n}}.$ For fixed $\varepsilon>0$ introduce the notation ${\cal G}_{n}=((1-\varepsilon)\log n){\cal D}\cap{\cal Z}_{2}$. Choosing $b_{n}=n^{\delta/2}$, $a_{n}=n^{\delta/4}$, we can prove using Stirling formula that for $(k,\ell)\in{\cal G}_{n}$ $\mathbf{P}(A_{1})\geq\frac{1}{n^{1-\delta}}$ for some $\delta>0$. Since the cardinality of ${\cal G}_{n}$ is $O(\log^{2}n)$, we can verify that $\sum_{n}\sum_{(k,\ell)\in{\cal G}_{n}}\mathbf{P}(\overline{C}_{n})<\infty.$ By Borel-Cantelli lemma, with probability 1, $\cap_{(k,\ell)\in{\cal G}_{n}}C_{n}(k,\ell)$ occurs for all but finitely many $n$. This completes the proof of the statements (ii) of both Theorems 3.3 and 3.4. $\Box$ ## 7 Proof of Theorems 3.5 and 3.6 We start with the proof of Theorem 3.6 which is similar to the proof of Theorem I. So we do not give all the details. Recall the notations of the theorem, Proposition 4.1 and Fact 6. For $m_{z}$, given in (3.9), we have $m_{z}={\bf E}(\xi(z,T)|\,\,T<\infty).$ (7.1) ###### Lemma 7.1. For $\log(1-2q(1-2q))<v<\log(1+2q(1-2q))$ we have $\varphi(v)=\exp(m_{z}(v+O(v^{2})),\qquad v\to 0,$ (7.2) where O is uniform in $z,$ and $\psi(v)<\frac{1+|e^{v}-1|}{1-\frac{|e^{v}-1|}{1-2q}}.$ (7.3) Proof: The proof of this lemma is based on Proposition 4.1 and goes along the same lines as Lemma 2.3 in [4]. $\Box$ Let $k_{n}:=(1-\delta_{n})\lambda_{0}\log n\sim\lambda_{0}\log n,$ $r_{n}:=c\log\log n,$ and $I(r):=[-r,r].$ Furthermore, let $n_{\ell}=[e^{\ell}]$, $\xi(z)=\xi(z,\infty)$ and define the events $A_{j}=\left\\{\xi(S_{j})\geq k_{n_{\ell}},\,\max_{x\in I(r_{n_{\ell+1}})}\left(\frac{\xi(S_{j}+x)}{m_{x}k_{n_{\ell}}}-1\right)\geq\varepsilon\right\\}.$ Then ${\bf P}\left(\bigcup_{j=0}^{n_{\ell+1}}A_{j}\right)\leq\sum_{j=0}^{n_{\ell+1}}{\bf P}(A_{j})\leq\sum_{j=0}^{n_{\ell+1}}\sum_{x\in I(r_{n_{\ell+1}})}{\bf P}(A_{j}^{(x)}),$ where $A_{j}^{(x)}=\left\\{\xi(S_{j})\geq k_{n_{\ell}},\,\xi(S_{j}+x)\geq(1+\varepsilon)m_{x}k_{n_{\ell}}\right\\}.$ Consider the random walk obtained by reversing the original walk at $S_{j}$, i.e. let $S_{i}^{\prime}:=S_{j-i}-S_{j}$, $i=0,1,\ldots,j$ and extend it to infinite time, and also the forward random walk $S_{i}^{\prime\prime}:=S_{j+i}-S_{j}$, $i=0,1,2,\ldots$ Then $\\{S_{0}^{\prime},S_{1}^{\prime},\ldots\\}$ and $\\{S_{0}^{\prime\prime},S_{1}^{\prime\prime},\ldots\\}$ are independent random walks and so are their respective local times $\xi^{\prime}$ and $\xi^{\prime\prime}$. Moreover, $\xi(S_{j})=\xi^{\prime\prime}(0)+\xi(S_{j},j)\leq\xi^{\prime\prime}(0)+\xi^{\prime}(0)+1,$ $\xi(S_{j}+x)=\xi^{\prime\prime}(x)+\xi(S_{j}+x,j)\leq\xi^{\prime\prime}(x)+\xi^{\prime}(x).$ Here $\xi^{\prime}$ and $\xi^{\prime\prime}$ are independent and $\xi^{\prime}$ has the same distribution as $\xi^{*}$ (see Fact 7) and $\xi^{\prime\prime}$ has the same distribution as $\xi$. Hence $\displaystyle{\bf P}(A_{j}^{(x)})$ $\displaystyle\leq$ $\displaystyle{\bf P}(\xi^{\prime\prime}(0)+\xi^{\prime}(0)\geq k_{n_{\ell}}-1,\,\xi^{\prime\prime}(x)+\xi^{\prime}(x)\geq(1+\varepsilon)m_{x}k_{n_{\ell}})$ $\displaystyle=$ $\displaystyle\sum{\bf P}(\xi^{\prime\prime}(0)=k_{1},\,\xi^{\prime}(0)=k_{2},\,\xi^{\prime\prime}(x)+\xi^{\prime}(x)\geq(1+\varepsilon)m_{x}k_{n_{\ell}}),$ where the summation goes for $\\{(k_{1},k_{2}):\,k_{1}+k_{2}\geq k_{n_{\ell}}-1\\}$. Using exponential Markov inequality, Proposition 4.1, Fact 7, the independence of $\xi^{\prime\prime}$ and $\xi^{\prime}$ and elementary calculus, we get $\displaystyle{\bf P}(A_{j}^{(x)})$ $\displaystyle\leq$ $\displaystyle\sum{\bf E}\left(e^{v(\xi^{\prime\prime}(x)+\xi^{\prime}(x))},\xi^{\prime\prime}(0)=k_{1},\xi^{\prime}(0)=k_{2}\right)e^{-v(1+\varepsilon)m_{x}k_{n_{\ell}}}$ (7.4) $\displaystyle=$ $\displaystyle\sum(\varphi(v))^{k_{1}+k_{2}}(1-2q)^{2}(2q)^{k_{1}+k_{2}}\psi(v)e^{-v(1+\varepsilon)m_{x}k_{n_{\ell}}}$ $\displaystyle=$ $\displaystyle(1-2q)^{2}\psi(v)e^{-v(1+\varepsilon)m_{x}k_{n_{\ell}}}\sum(2q\varphi(v))^{k_{1}+k_{2}}$ $\displaystyle=$ $\displaystyle(1-2q)^{2}\psi(v)e^{-v(1+\varepsilon)m_{x}k_{n_{\ell}}}(2q\varphi(v))^{k_{n_{\ell}}}$ $\displaystyle\times$ $\displaystyle\left(\frac{k_{n_{\ell}}}{2q\varphi(v)(1-2q\varphi(v))}+\frac{1}{(1-2q\varphi(v))^{2}}\right).$ Observe that even though the moment generating functions in Proposition 4.1 and Fact 7 are slightly different for positive and negative values of $x$, in (7.4) we get the same expression while working with $\xi^{\prime}+\xi^{\prime\prime}$. By (7.2), we obtain for all $j\geq 0$ $\displaystyle{\bf P}(A_{j}^{(x)})$ $\displaystyle\leq$ $\displaystyle(1-2q)^{2}\psi(v)\left(\frac{k_{n_{\ell}}}{2q\varphi(v)(1-2q\varphi(v))}+\frac{1}{(1-2q\varphi(v))^{2}}\right)$ $\displaystyle\times$ $\displaystyle e^{-m_{x}vk_{n_{\ell}}(\varepsilon+O(v))}(2q)^{k_{n_{\ell}}}.$ Choose $v_{0}>0$ small enough such that $\varepsilon+O(v_{0})>0,\quad e^{v_{0}}<1+2q(1-2q),\quad\frac{1}{2}<\varphi(v_{0})<\frac{1}{2q}.$ Using $x\in I(r_{n_{\ell+1}})$, we get $m_{x}k_{n_{\ell}}=\frac{h^{|x|}}{2q}(1-\delta_{n_{\ell}})\lambda_{0}\log n_{\ell}\geq\frac{h^{r_{n_{\ell+1}}}}{2q}(1-\delta_{n_{\ell}})\lambda_{0}\log n_{\ell}.$ In the sequel $C_{1},C_{2},\ldots$ denote positive constants whose values are unimportant in our proofs. By the above assumptions $\displaystyle{\bf P}(A_{j}^{(x)})$ $\displaystyle\leq$ $\displaystyle C_{2}k_{n_{\ell}}e^{-m_{x}v_{0}k_{n_{\ell}}(\varepsilon+O(v_{0}))}(2q)^{k_{n_{\ell}}}$ $\displaystyle\leq$ $\displaystyle C_{2}k_{n_{\ell}}\exp\left(-(1-\delta_{n_{\ell}})\log n_{\ell}\left(C_{3}h^{r_{n_{\ell+1}}}+1\right)\right).$ Hence $\displaystyle\sum_{j=0}^{n_{\ell+1}}\sum_{x\in I(r_{n_{\ell+1}})}{\bf P}(A_{j}^{(x)})$ $\displaystyle\leq$ $\displaystyle C_{4}n_{\ell+1}r_{n_{\ell+1}}k_{n_{\ell}}\exp\left(-(1-\delta_{n_{\ell}})\log n_{\ell}\left(C_{3}h^{r_{n_{\ell+1}}}+1\right)\right)$ $\displaystyle\leq$ $\displaystyle C_{5}\frac{n_{\ell+1}}{n_{\ell}}k_{n_{\ell}}r_{n_{\ell+1}}\exp\left(-C_{6}h^{r_{n_{\ell+1}}}\log n_{\ell}+\delta_{n_{\ell}}\log n_{\ell}\right)$ $\displaystyle=$ $\displaystyle C_{5}\frac{n_{\ell+1}}{n_{\ell}}k_{n_{\ell}}r_{n_{\ell+1}}\exp\left(-h^{r_{n_{\ell}}}\log n_{\ell}\left(C_{6}h^{r_{n_{\ell+1}}-r_{n_{\ell}}}-\frac{\delta_{n_{\ell}}}{h^{r_{n_{\ell}}}}\right)\right)$ $\displaystyle\leq$ $\displaystyle C_{5}\frac{n_{\ell+1}}{n_{\ell}}k_{n_{\ell}}r_{n_{\ell+1}}\exp\left(-C_{7}h^{r_{n_{\ell}}}\log n_{\ell}\right)$ $\displaystyle\leq$ $\displaystyle C_{8}(\log n_{\ell})\log\log n_{\ell}\exp(-C_{7}(\log{n_{\ell})^{1-\alpha c}}),$ where in the last two lines we used the conditions of the Theorem. Consequently, ${\bf P}(\bigcup_{j=0}^{n_{\ell+1}}A_{j})\leq\sum_{j=0}^{n_{\ell+1}}\sum_{x\in I(r_{n_{\ell+1}})}{\bf P}(A_{j}^{(x)})\leq C_{8}\ell\log\ell\exp(-C_{7}\ell^{1-\alpha c})$ for large enough $\ell$, which is summable in $\ell$ when $\alpha c<1$. By Borel-Cantelli lemma for large $\ell$ if $\xi(S_{j})\geq k_{n_{\ell}}$, then $\xi(S_{j}+x)\leq(1+\varepsilon)m_{x}k_{n_{\ell}}$ for all $x\in I(r_{n_{\ell+1}})$. Let now $n_{\ell}\leq n<n_{\ell+1}$ and $x\in I(r_{n_{\ell+1}})$. $\xi(S_{j})\geq k_{n},j\leq n$ implies $\xi(S_{j})\geq k_{n_{\ell}}$, i.e. $\xi(S_{j}+x)\leq(1+\varepsilon)m_{x}k_{n_{\ell}}\leq(1+\varepsilon)m_{x}k_{n}.$ (7.5) The lower bound is similar, with slight modifications. However we do not present it. The interested reader should look at the corresponding proof of Theorem 1.2 in [4]. ${\Box}$ The proof of Theorem 3.5 again goes similarly to the proof of Theorem H. As the main ingredient is the following lemma, which is somewhat different from the $d$-dimensional situation, we give a complete proof. ###### Lemma 7.2. Let $0<\alpha<1$, $j\leq n-n^{\alpha}$, $|x|\leq c\log n$ with any $c>0$. Then with probability 1 there exists an $n_{0}(\omega)$ such that for $n\geq n_{0}$ we have $\xi(S_{j}+x,n)=\xi(S_{j}+x,\infty).$ Proof. Let $A_{n}=\bigcup_{j\leq n-n^{\alpha}}\quad\bigcup_{\ell\geq n}\quad\bigcup_{|x|\leq c\log n}\\{S_{\ell}-S_{j}=x\\}.$ By our Lemma 4.1 ${\bf P}(S_{\ell}-S_{j}=x)={\bf P}(S_{\ell-j}=x)\leq C_{1}\exp(-C_{2}(\ell-j)+C_{3}x).$ (7.6) Consequently, ${\bf P}(A_{n})\leq C_{1}\sum_{j\leq n-n^{\alpha}}\sum_{\ell\geq n}\sum_{|x|\leq c\log n}\exp(-C_{2}\ell+C_{2}j+C_{3}x)$ $\leq C\exp(-C_{2}n+C_{2}(n-n^{\alpha})+C_{3}c\log n)=C_{4}n^{C_{5}}\exp(-C_{2}n^{\alpha}).$ Since this is summable, we have the lemma. $\Box$ To prove Theorem 3.5, observe that it suffices to consider points visited before time $n-n^{\alpha}$, ($0<\alpha<1$), since in the time interval $(n-n^{\alpha},n)$ the maximal local time is less than $\alpha(1+\varepsilon)\lambda_{0}\log n$, hence this point cannot be in ${\cal A}_{n}$. Consequently, Theorem 3.5 follows from Theorem 3.6 and Lemma 7.2. $\Box$ ## 8 Concluding remarks First observe that the following points are on the curve $g(x,y)=1.$ $\left(0,\frac{1}{\log(1/q)}\right),\quad\left(-\frac{1}{\log(2pq)},-\frac{1}{\log(2pq)}\right),\quad\left(\frac{2\kappa_{0}p}{2p+1},\kappa_{0}\right),\quad\left(\lambda_{0},\frac{\lambda_{0}}{p}\right).$ (8.1) Consequently, there are points $x_{n}$ such that $\xi(x_{n},n)=1\quad{\rm and}\quad\Xi(x_{n},n)\sim\frac{\log n}{\log(1/q)}$ which in fact means that $\Xi(x_{n},n)=\xi(x_{n}+1,n)\sim\frac{\log n}{\log(1/q)}.$ On the other hand, if for a point $x_{n}$, $\Xi(x_{n},n)>(1+\epsilon)\frac{\log n}{\log(1/q)},$ then we have $\xi(x_{n},n)>c\log n$ with some $c>0.$ If $\xi(x_{n},n)\sim\lambda_{0}\log n$ then for the unit sphere centered at $x_{n}$, that is to say for its two neighbors we have $\Xi(x_{n},n)\sim\frac{\lambda_{0}}{p}\log n.$ Since $m_{-1}=m_{1},$ it follows from Corollary 3.1 that the two neighbors of the nearly maximally visited points have asymptotically equal local time. On the other hand, if the occupation time is asymptotically maximal, $\Xi(x_{n},n)\sim\kappa_{0}\log n\quad{\rm then}\quad\xi(x_{n},n)\sim\frac{2p}{2p+1}\kappa_{0}\log n.$ With some extra calculation one can find the maximal weight of the unit ball: $w(x_{n},n):=\xi(x_{n},n)+\Xi(x_{n},n),\quad w(n):=\max_{x_{n}\in\mathcal{Z}}(\xi(x_{n},n)+\Xi(x_{n},n)).$ We get $\lim_{n\to\infty}\frac{w(n)}{\log n}=\frac{2\,\beta}{\log\left(\left(\frac{2q}{\beta+1}\right)^{\beta+1}\left(p(\beta-1)\right)^{\beta-1}\right)}=:\frac{2\,\beta}{C},\qquad\hskip 28.45274pt\mathrm{a.s.}$ where $\beta$ is the constant defined in (4.29). In this case we have $\xi(x_{n},n)\sim\frac{\beta-1}{C}\log n,\quad{\rm and}\quad\Xi(x_{n},n)\sim\frac{\beta+1}{C}\log n.$ As a final conclusion if any of the three quantities of $\xi(x_{n},n),\,\,\Xi(x_{n},n),\,\,w(x_{n},n),$ is asymptotically maximal, it uniquely determines the asymptotic values of the other two quantities, an interesting phenomenon which we proved for $d>4$ in the symmetric walk case. ## References * [1] Baron, M. and Rukhin, A.L.: Distribution of the number of visits of a random walk. Commun. Statist.-Stochastic Models 15(3) (1999), 593–597. * [2] Csáki, E., Földes, A. and Révész, P.: Heavy points of a $d$-dimensional simple random walk. Statist. Probab. Lett. 76 (2006), 45–57. * [3] Csáki, E., Földes, A. and Révész, P.: Joint asymptotic behavior of local and occupation times of random walk in higher dimensions. Studia Sci. Math. Hungar. 44 (2007), 535–563. * [4] Csáki, E., Földes, A. and Révész, P.: On the behavior of random walk around heavy points. J. Theor. Probab. 20 (2007), 1041–1057. * [5] Csáki, E., Földes, A. and Révész, P.: On the local times of transient random walks. Acta Applicandae Mathematicae 96 (2007), 147–158. * [6] Csáki, E., Földes, A., Révész, P., Rosen, J. and Shi, Z.: Frequently visited sets for random walks. Stoch. Process. Appl. 115 (2005), 1503–1517. * [7] Dwass, M.: Simple random walk and rank order statistics. Ann. Math. Statist. 38 (1967), 1042–1053. * [8] Dvoretzky, A. and Erdős, P.: Some problems on random walk in space. Proc. Second Berkeley Symposium (1951), 353–367. * [9] Erdős, P. and Taylor, S.J.: Some problems concerning the structure of random walk paths. Acta Math. Acad. Sci. Hung. 11 (1960), 137–162. * [10] Feller, W.: An Introduction to Probability Theory and Its Applications, Vol. I, 3rd ed. John Wiley, New York, 1968. * [11] Feller, W.: An Introduction to Probability Theory and Its Applications, Vol. II, 2nd ed. John Wiley, New York, 1971. * [12] Jordan, K.: Chapters on the Classical Calculus of Probability, Disquisitiones Mathematicae Hungaricae, No. 4. Akadémiai Kiadó, Budapest, 1972. * [13] Pitt, J.H.: Multiple points of transient random walk. Proc. Amer. Math. Soc. 43 (1974), 195-199. * [14] Révész, P.: Random Walk in Random and Non-Random Environments, 2nd ed. World Scientific, Singapore, 2005. * [15] Spitzer, F.: Principles of Random Walk, 2nd. ed. Van Nostrand, Princeton, 1976.

arxiv-papers

2024-09-04T02:48:53.595331

0802.0778

TRANSIENT NEAREST NEIGHBOR RANDOM WALK AND BESSEL PROCESS Endre Csáki111Research supported by the Hungarian National Foundation for Scientific Research, Grant No. K 61052 and K 67961. Alfréd Rényi Institute of Mathematics, Hungarian Academy of Sciences, Budapest, P.O.B. 127, H-1364, Hungary. E-mail address: csaki@renyi.hu Antónia Földes222Research supported by a PSC CUNY Grant, No. 69020-0038. Department of Mathematics, College of Staten Island, CUNY, 2800 Victory Blvd., Staten Island, New York 10314, U.S.A. E-mail address: foldes@mail.csi.cuny.edu Pál Révész1 Institut für Statistik und Wahrscheinlichkeitstheorie, Technische Universität Wien, Wiedner Hauptstrasse 8-10/107 A-1040 Vienna, Austria. E-mail address: reveszp@renyi.hu Abstract: We prove strong invariance principle between a transient Bessel process and a certain nearest neighbor (NN) random walk that is constructed from the former by using stopping times. It is also shown that their local times are close enough to share the same strong limit theorems. It is shown furthermore, that if the difference between the distributions of two NN random walks are small, then the walks themselves can be constructed so that they are close enough. Finally, some consequences concerning strong limit theorems are discussed. AMS 2000 Subject Classification: Primary 60F17; Secondary 60F15, 60J10, 60J55, 60J60. Keywords: transient random walk, Bessel process, strong invariance principle, local time, strong theorems. Running head: NN random walk and Bessel process. ## 1\. Introduction In this paper we consider a nearest neighbor (NN) random walk, defined as follows: let $X_{0}=0,\ X_{1},X_{2},\ldots$ be a Markov chain with $\displaystyle E_{i}$ $\displaystyle:=$ $\displaystyle{\bf P}(X_{n+1}=i+1\mid X_{n}=i)=1-{\bf P}(X_{n+1}=i-1\mid X_{n}=i)$ $\displaystyle=$ $\displaystyle\left\\{\begin{array}[]{ll}1&{\rm if}\quad i=0\\\ 1/2+p_{i}&{\rm if}\quad i=1,2,\ldots,\end{array}\right.$ where $-1/2\leq p_{i}\leq 1/2,\ i=1,2,\ldots$. In case $0<p_{i}\leq 1/2$ the sequence $\\{X_{i}\\}$ describes the motion of a particle which starts at zero, moves over the nonnegative integers and going away from 0 with a larger probability than to the direction of 0. We will be interested in the case when $p_{i}\sim B/4i$ with $B>0$ as $i\to\infty$. We want to show that in certain sense, this Markov chain is a discrete analogue of continuous Bessel process and establish a strong invariance principle between these two processes. The properties of the discrete model, often called birth and death chain, connections with orthogonal polynomials in particular, has been treated extensively in the literature. See e.g. the classical paper by Karlin and McGregor [12], or more recent papers by Coolen-Schrijner and Van Doorn [6] and Dette [9]. In an earlier paper [7] we investigated the local time of this Markov chain in the transient case. There is a well-known result in the literature (cf. e.g. Chung [5]) characterizing those sequences $\\{p_{i}\\}$ for which $\\{X_{i}\\}$ is transient (resp. recurrent). Theorem A: ([5], page 74) Let $X_{n}$ be a Markov chain with transition probabilities given in (1.) with $-1/2<p_{i}<1/2$, $i=1,2,\ldots$ Define $U_{i}:={\frac{1-E_{i}}{E_{i}}}={\frac{1/2-p_{i}}{1/2+p_{i}}}$ (1.4) Then $X_{n}$ is transient if and only if $\sum_{k=1}^{\infty}\prod_{i=1}^{k}U_{i}<\infty.$ As a consequence, the Markov chain $(X_{n})$ with $p_{R}\sim B/4R,\,R\to\infty$ is transient if $B>1$ and recurrent if $B<1$. The Bessel process of order $\nu$, denoted by $Y_{\nu}(t),\,t\geq 0$ is a diffusion process on the line with generator $\frac{1}{2}\frac{d^{2}}{dx^{2}}+\frac{2\nu+1}{2x}\frac{d}{dx}.$ $d=2\nu+2$ is the dimension of the Bessel process. If $d$ is a positive integer, then $Y_{\nu}(\cdot)$ is the absolute value of a $d$-dimensional Brownian motion. The Bessel process $Y_{\nu}(t)$ is transient if and only if $\nu>0$. The properties of the Bessel process were extensively studied in the literature. Cf. Borodin and Salminen [2], Revuz and Yor [19], Knight [14]. Lamperti [15] determined the limiting distribution of $X_{n}$ and also proved a weak convergence theorem in a more general setting. His result in our case reads as follows. Theorem B: ([15]) Let $X_{n}$ be a Markov chain with transition probabilities given in (1.) with $-1/2<p_{i}<1/2$, $i=1,2,\ldots$ If $\,\,\lim_{R\to\infty}\,Rp_{R}=B/4>-1/4$, then the following weak convergence holds: $\frac{X_{[nt]}}{\sqrt{n}}\Longrightarrow Y_{(B-1)/2}(t)$ in the space D[0,1]. In particular, $\lim_{n\to\infty}{\bf P}\left(\frac{X_{n}}{\sqrt{n}}<x\right)=\frac{1}{2^{B/2-1/2}\Gamma(B/2+1/2)}\int_{0}^{x}u^{B}e^{-u^{2}/2}\,du.$ In Theorems A and B values of $p_{i}$ can be negative. In the sequel however we deal only with the case when $p_{i}$ are non-negative, and the chain is transient, which will be assumed throughout without mentioning it. Let $D(R,\infty):=1+\sum_{j=1}^{\infty}\prod_{i=1}^{j}U_{R+i},$ (1.5) and define $p_{R}^{*}:=\frac{\frac{1}{2}+p_{R}}{D(R,\infty)}=1-q^{*}_{R}$ (1.6) Now let $\xi(R,\infty)$, $R=0,1,2,\ldots$ be the total local time at $R$ of the Markov chain $\\{X_{n}\\}$, i.e. $\xi(R,\infty):=\\#\\{n\geq 0:X_{n}=R\\}.$ (1.7) Theorem C: ([7]) For a transient NN random walk ${\bf P}(\xi(R,\infty)=k)=p_{R}^{*}(q_{R}^{*})^{k-1},\qquad k=1,2,\ldots$ (1.8) Moreover, $\eta(R,t)$, $R>0$ will denote the local time of the Bessel process, i.e. $\eta(R,t):=\lim_{\varepsilon\to 0}\frac{1}{2\varepsilon}\int_{0}^{t}I\\{Y_{\nu}(s)\in(R-\varepsilon,R+\varepsilon)\\}\,ds,\qquad\eta(R,\infty):=\lim_{t\to\infty}\eta(R,t).$ It is well-known that $\eta(R,\infty)$ has exponential distribution (see e.g. [2]). ${\bf P}(\eta(R,\infty)<x)=1-\exp\left(-\frac{\nu}{R}\,x\right).$ (1.9) For $0<a<b$ let $\tau:=\tau(a,b)=\min\\{t\geq 0:\,Y_{\nu}(t)\notin(a,b)\\}.$ (1.10) Then we have (cf. Borodin and Salminen [2], Section 6, 3.0.1 and 3.0.4). Theorem D: For $0<a<x<b$ we have ${\bf P}_{x}(Y_{\nu}(\tau)=a)=1-{\bf P}_{x}(Y_{\nu}(\tau)=b)=\frac{x^{-2\nu}-b^{-2\nu}}{a^{-2\nu}-b^{-2\nu}},$ (1.11) ${\bf E}_{x}e^{-\alpha\tau}=\frac{S_{\nu}(b\sqrt{2\alpha},x\sqrt{2\alpha})+S_{\nu}(x\sqrt{2\alpha},a\sqrt{2\alpha})}{S_{\nu}(b\sqrt{2\alpha},a\sqrt{2\alpha})},$ (1.12) where $S_{\nu}(u,v)=(uv)^{-\nu}(I_{\nu}(u)K_{\nu}(v)-K_{\nu}(u)I_{\nu}(v)),$ (1.13) $I_{\nu}$ and $K_{\nu}$ being the modified Bessel functions of the first and second kind, resp. Here and in what follows ${\bf P}_{x}$ and ${\bf E}_{x}$ denote conditional probability, resp. expectation under $Y_{\nu}(0)=x.$ For simplicity we will use ${\bf P}_{0}={\bf P},$ and ${\bf E}_{0}={\bf E}.$ Now consider $Y_{\nu}(t),\,t\geq 0$, a Bessel process of order $\nu$, $Y_{\nu}(0)=0$, and let $X_{n},\,n=0,1,2,\ldots$ be an NN random walk with $p_{0}=p_{1}=1/2$, $p_{R}=\frac{(R-1)^{-2\nu}-R^{-2\nu}}{(R-1)^{-2\nu}-(R+1)^{-2\nu}}-\frac{1}{2},\qquad R=2,3,\ldots$ (1.14) Our main results are strong invariance principles concerning Bessel process, NN random walk and their local times. ###### Theorem 1.1. On a suitable probability space we can construct a Bessel process $\\{Y_{\nu}(t),\,t\geq 0\\},$ $\nu>0$ and an NN random walk $\\{X_{n},\,n=0,1,2,\ldots\\}$ with $p_{R}$ as in (1.14) such that for any $\varepsilon>0$, as $n\to\infty$ we have $Y_{\nu}(n)-X_{n}=O(n^{1/4+\varepsilon})\qquad{\rm a.s.}$ (1.15) Our strong invariance principle for local times reads as follows. ###### Theorem 1.2. Let $Y_{\nu}(t)$ and $X_{n}$ as in Theorem 1.1 and let $\eta$ and $\xi$ their respective local times. As $R\to\infty$, we have $\xi(R,\infty)-\eta(R,\infty)=O(R^{1/2}\log R)\quad{\rm a.s.}$ (1.16) We prove the following strong invariance principle between two NN random walks. ###### Theorem 1.3. Let $\\{X_{n}^{(1)}\\}_{n=0}^{\infty}$ and $\\{X_{n}^{(2)}\\}_{n=0}^{\infty}$ be two NN random walk with $p_{j}^{(1)}$ and $p_{j}^{(2)}$, resp. Assume that $\left|p_{j}^{(1)}-\frac{B}{4j}\right|\leq\frac{C}{j^{\gamma}}$ (1.17) and $\left|p_{j}^{(2)}-\frac{B}{4j}\right|\leq\frac{C}{j^{\gamma}}$ (1.18) $j=1,2,\ldots$ with $B>1$, $1<\gamma\leq 2$ and some non-negative constant $C$. Then on a suitable probability space one can construct $\\{X_{n}^{(1)}\\}$ and $\\{X_{n}^{(2)}\\}$ such that as $n\to\infty$ $|X_{n}^{(1)}-X_{n}^{(2)}|=O((X_{n}^{(1)}+X_{n}^{(2)})^{2-\gamma})=O((n\log\log n)^{1-\gamma/2})\quad{\rm a.s.}$ The organization of the paper is as follows. In Section 2 we will present some well-known facts and prove some preliminary results. Sections 3-5 contain the proofs of Theorems 1.1-1.3, respectively. In Section 6 we prove strong theorems (most of them are integral tests) which easily follow from Theorems 1.1 and 1.2 and the corresponding results for Bessel process. In Section 7, using our Theorem 1.3 in both directions, we prove an integral test for the local time of the NN-walk, and a strong theorem for the speed of escape of the Bessel process. ## 2\. Preliminaries ###### Lemma 2.1. Let $Y_{\nu}(\cdot)$ be a Bessel process starting from $x=R$ and let $\tau$ be the stopping time defined by (1.10) with $a=R-1$ and $b=R+1$. Let $p_{R}$ be defined by (1.14). Then as $R\to\infty$ $p_{R}=\frac{2\nu+1}{4R}+O\left(\frac{1}{R^{2}}\right),$ (2.1) ${\bf E}_{R}(\tau)=1+O\left(\frac{1}{R}\right),$ (2.2) $Var_{R}(\tau)=O(1).$ (2.3) Proof: For $\nu=1/2$, i.e. for $d=3$-dimensional Bessel process, in case $x=R$, $a=R-1$, $b=R+1$ we have ${\bf E}_{R}(e^{\lambda\tau})=\frac{1}{\cos(\sqrt{2\lambda})}$ which does not depend on $R$. We prove that this holds asymptotically in general, when $\nu>0$. Using the identity (cf. [2], page 449 and [21], page 78) $\displaystyle K_{\nu}(x)$ $\displaystyle=$ $\displaystyle\left\\{\begin{array}[]{ll}\displaystyle{\frac{\pi}{2\sin(\nu\pi)}(I_{-\nu}(x)-I_{\nu}(x))}\quad{\rm if}\,\,\nu\,\,{\rm is\,\,not\,\,an\,\,integer}\\\ &\\\ \lim_{\mu\to\nu}K_{\mu}(x)\quad{\rm if\,\,}\nu\,\,{\rm is\,\,an\,\,integer}\end{array}\right.$ and the series expansion $I_{\nu}(x)=\sum_{k=0}^{\infty}\frac{(x/2)^{\nu+2k}}{k!\Gamma(\nu+k+1)},$ one can see that the coefficient of $-\alpha$ in the Taylor series expansion of the Laplace transform (1.12) is ${\bf E}_{x}(\tau)=\frac{1}{2(\nu+1)}\frac{(b^{2}-x^{2})a^{-2\nu}+(x^{2}-a^{2})b^{-2\nu}-(b^{2}-a^{2})x^{-2\nu}}{a^{-2\nu}-b^{-2\nu}}$ from which by putting $x=R$, $a=R-1$, $b=R+1$, we obtain ${\bf E}_{R}(\tau)=\frac{1}{2(\nu+1)}\frac{(2R+1)(R-1)^{-2\nu}+(2R-1)(R+1)^{-2\nu}-4R^{1-2\nu}}{(R-1)^{-2\nu}-(R+1)^{-2\nu}}$ giving (2.2) after some calculations. (2.3) can also be obtained similarly, but it seems quite complicated. A simpler argument is to use moment generating function and expansion of the Bessel functions for imaginary arguments near infinity. Put $\alpha=-\lambda$ into (1.12) to obtain ${\bf E}_{x}(e^{\lambda\tau})=\frac{S_{\nu}(ib\sqrt{2\lambda},ix\sqrt{2\lambda})+S_{\nu}(ix\sqrt{2\lambda},ia\sqrt{2\lambda})}{S_{\nu}(ib\sqrt{2\lambda},ia\sqrt{2\lambda})},$ (2.7) where $i=\sqrt{-1}$. We use the following asymptotic expansions (cf. Erdélyi et al. [11], page 86, or Watson [21], pages 202, 219) $I_{\nu}(z)=(2\pi z)^{-1/2}\left(e^{z}+ie^{-z+i\nu\pi}+O(|z|^{-1})\right),$ $K_{\nu}(z)=\left(\frac{\pi}{2z}\right)^{1/2}\left(e^{-z}+O(|z|^{-1})\right).$ Hence one obtains for $\lambda>0$ fixed, and $x<b,\,$ $S_{\nu}(ib\sqrt{2\lambda},ix\sqrt{2\lambda})=(-2\lambda bx)^{-\nu}(I_{\nu}(ib\sqrt{2\lambda})K_{\nu}(ix\sqrt{2\lambda})-I_{\nu}(ix\sqrt{2\lambda})K_{\nu}(ib\sqrt{2\lambda}))$ $=\frac{1}{2}(-2\lambda bx)^{-\nu-1/2}\left(e^{i(b-x)\sqrt{2\lambda}}-e^{-i(b-x)\sqrt{2\lambda}}+O\left(\frac{1}{x}\right)\right),\quad x\to\infty.$ One can obtain asymptotic expansions similarly for $S_{\nu}(ix\sqrt{2\lambda},ia\sqrt{2\lambda})$, $S_{\nu}(ib\sqrt{2\lambda},ia\sqrt{2\lambda})$. Putting these into (2.7), with $x=R$, $a=R-1$, $b=R+1$, we get as $R\to\infty$ ${\bf E}_{R}(e^{\lambda\tau})=\frac{(R^{2}+R)^{-\nu-1/2}+(R^{2}-R)^{-\nu-1/2}}{(R^{2}-1)^{-\nu-1/2}}\,\,\frac{e^{i\sqrt{2\lambda}}-e^{-i\sqrt{2\lambda}}+O\left(\frac{1}{R}\right)}{e^{2i\sqrt{2\lambda}}-e^{-2i\sqrt{2\lambda}}+O\left(\frac{1}{R}\right)}$ $=\frac{1}{\cos(\sqrt{2\lambda})}+O\left(\frac{1}{R}\right).$ Hence putting $\lambda=1$, there exists a constant $C$ such that ${\bf E}_{R}(e^{\tau})\leq C$ for all $R=1,2,\ldots$ By Markov’s inequality we have ${\bf P}_{R}(\tau>t)={\bf P}_{R}(e^{\tau}>e^{t})\leq Ce^{-t},$ from which ${\bf E}_{R}(\tau^{2})\leq 2C$, implying (2.3). $\Box$ Here and throughout $C,C_{1},C_{2},\ldots$ denotes unimportant positive (possibly random) constants whose values may change from line to line. Recall the definition of the upper and lower classes for a stochastic process $Z(t),\,t\geq 0$ defined on a probability space $(\Omega,{\cal F},P)$ (cf. Révész [18], p. 33). The function $a_{1}(t)$ belongs to the upper-upper class of $Z(t)$ ($a_{1}(t)\in{\rm UUC}(Z(t)$) if for almost all $\omega\in\Omega$ there exists a $t_{0}(\omega)>0$ such that $Z(t)<a_{1}(t)$ if $t>t_{0}(\omega).$ The function $a_{2}(t)$ belongs to the upper-lower class of $Z(t)$ ($a_{1}(t)\in{\rm ULC}(Z(t)$) if for almost all $\omega\in\Omega$ there exists a sequence of positive numbers $0<t_{1}=t_{1}(\omega)<t_{2}=t_{2}(\omega)<\ldots$ with $\lim_{i\to\infty}t_{i}=\infty$ such that $Z(t_{i})\geq a_{2}(t_{i})$, $(i=1,2,\ldots).$ The function $a_{3}(t)$ belongs to the lower-upper class of $Z(t)$ ($a_{3}(t)\in{\rm LUC}(Z(t)$) if for almost all $\omega\in\Omega$ there exists a sequence of positive numbers $0<t_{1}=t_{1}(\omega)<t_{2}=t_{2}(\omega)<\ldots$ with $\lim_{i\to\infty}t_{i}=\infty$ such that $Z(t_{i})\leq a_{3}(t_{i})$, $(i=1,2,\ldots).$ The function $a_{4}(t)$ belongs to the lower-lower class of $Z(t)$ ($a_{4}(t)\in{\rm LLC}(Z(t)$) if for almost all $\omega\in\Omega$ there exists a $t_{0}(\omega)>0$ such that $Z(t)>a_{4}(t)$ if $t>t_{0}(\omega).$ The following lower class results are due to Dvoretzky and Erdős [10] for integer $d=2\nu+2$. In the general case when $\nu>0$, the proof is similar (cf. also Knight [14] and Chaumont and Pardo [4] in the case of positive self- similar Markov processes). Theorem E: Let $\nu>0$ and let $b(t)$ be a non-increasing, non-negative function. * • $t^{1/2}b(t)\in{\rm LLC}(Y_{\nu}(t))$ if and only if $\displaystyle{\int_{1}^{\infty}(b(2^{t}))^{2\nu}\,dt<\infty.}$ It follows e.g. that in case $\nu>0$, for any $\varepsilon>0$ we have $Y_{\nu}(t)\geq t^{1/2-\varepsilon}$ (2.8) almost surely for all sufficiently large $t$. In fact, from our invariance principle it will follow that the integral test in Theorem E holds also for our Markov chain $(X_{n})$. In the proof however we need an analogue of (2.8) for $X_{n}$. One can easily calculate the exact distribution of $\xi(R,\infty),$ the total local time of $X_{n}$ of Theorem 1.1 according to Theorem C. Lemma A: If $p_{R}$ is given by (1.14), then $\xi(R,\infty)$ has geometric distribution (1.8) with $p_{R}^{*}=\frac{\frac{1}{2}+p_{R}}{D(R,\infty)}=\frac{(\frac{1}{2}+p_{R})((R+1)^{2\nu}-R^{2\nu})}{(R+1)^{2\nu}}=\frac{\nu}{R}+O\left(\frac{1}{R}\right).$ (2.9) ###### Lemma 2.2. For any $\delta>0$ we have $X_{n}\geq n^{1/2-\delta}$ almost surely for all large enough $n$. Proof: From Lemma A it is easy to conclude that almost surely for some $R_{0}>0$ $\xi(R,\infty)\leq CR\log R$ if $R\geq R_{0}$, with some random positive constant $C$. Hence the time $\displaystyle{\sum_{R=1}^{S}}\xi(R,\infty)$ which the particle spent up to $\infty$ in $[1,S]$ is less than $\sum_{R=1}^{R_{0}-1}\xi(R,\infty)+C\sum_{R=R_{0}}^{S}R\log R\leq C_{1}S^{2+\delta}$ with some (random) $C_{1}>0$. Consequently, after $C_{1}S^{2+\delta}$ steps the particle will be farther away from the origin than $S$. Let $n=[C_{1}S^{2+\delta}],$ then $S\geq\left(\frac{n}{C_{1}}\right)^{1/(2+\delta)}$ and hence $X_{n}\geq\left(\frac{n}{C_{1}}\right)^{1/(2+\delta)}\geq n^{1/2-\delta}$ for $n$ large enough. This proves the Lemma. $\Box$ ## 3\. Proof of Theorem 1.1 Define the sequences $(\tau_{n})$, $t_{0}=0,$ $t_{n}:=\tau_{1}+\ldots+\tau_{n}$ as follows: $\displaystyle\tau_{1}$ $\displaystyle:=$ $\displaystyle\min\\{t:\ t>0,\ Y_{\nu}(t)=1\\},$ $\displaystyle\tau_{2}$ $\displaystyle:=$ $\displaystyle\min\\{t:\ t>0,\ Y_{\nu}(t+t_{1})=2\\},$ $\displaystyle\tau_{n}$ $\displaystyle:=$ $\displaystyle\min\\{t:\ t>0,\ |Y_{\nu}(t+t_{n-1})-Y_{\nu}(t_{n-1})|=1\\}\quad{\rm for}\quad n=3,4,\ldots$ Let $X_{n}=Y_{\nu}(t_{n})$. Then (cf. (1.14)) it is an NN random walk with $p_{0}=p_{1}=1/2$, $p_{R}=\frac{(R-1)^{-2\nu}-R^{-2\nu}}{(R-1)^{-2\nu}-(R+1)^{-2\nu}}-\frac{1}{2},\qquad R=2,3,\ldots$ Let ${\cal F}_{n}$ be the $\sigma$-algebra generated by $(\tau_{k},\ Y_{\nu}(\tau_{k}))_{k=1}^{n}$ and consider $M_{n}:=\sum_{i=1}^{n}(\tau_{i}-{\bf E}(\tau_{i}\mid{\cal F}_{i-1})).$ Then the sequence $(M_{n})_{n\geq 1}$ is a martingale with respect to $({\cal F}_{n})_{n\geq 1}$. It follows from (2.2) of Lemma 2.1 that for $i=2,3,\ldots$ we have ${\bf E}(\tau_{i}\mid{\cal F}_{i-1})={\bf E}(\tau_{i}\mid Y_{\nu}(t_{i-1}))=1+O\left(\frac{1}{Y_{\nu}(t_{i-1})}\right).$ Hence $|t_{n}-n|\leq|M_{n}|+|\tau_{1}-1|+C_{1}\sum_{i=2}^{n}\frac{1}{Y_{\nu}(t_{i-1})}=|M_{n}|+|\tau_{1}-1|+C_{1}\sum_{i=2}^{n}\frac{1}{X_{i-1}}$ with some (random) constant $C_{1}$. By (2.3) of Lemma 2.1 we have ${\bf E}M_{n}^{2}\leq Cn$. Let $\varepsilon>0$ be arbitrary and define $n_{k}=[k^{1/\varepsilon}]$. From the martingale inequality we get ${\bf P}\left(\max_{n_{k-1}\leq n\leq n_{k}}|M_{n}|\geq C_{1}n_{k-1}^{1/2+\varepsilon}\right)\leq\frac{C_{2}}{n_{k}^{2\varepsilon}},$ hence we obtain by Borel-Cantelli lemma $\max_{n_{k-1}\leq n\leq n_{k}}|M_{n}|\leq C_{1}n_{k-1}^{1/2+\varepsilon}$ almost surely for large $k$. Hence we also have $|M_{n}|=O(n^{1/2+\varepsilon})\qquad{\rm a.s.}$ By Lemma 2.2 $\sum_{i=2}^{n}\frac{1}{X_{i-1}}=O(n^{1/2+\varepsilon})\qquad{\rm a.s.},$ consequently $|t_{n}-n|=O(n^{1/2+\varepsilon})\qquad{\rm a.s.}$ (3.1) It is well-known (cf. [2], p. 69) that $Y_{\nu}(t)$ satisfies the stochastic differential equation $dY_{\nu}(t)=dW(t)+\frac{2\nu+1}{2Y_{\nu}(t)}dt,$ (3.2) where $W(t)$ is a standard Wiener process. Hence $X_{n}-Y_{\nu}(n)=Y_{\nu}(t_{n})-Y_{\nu}(n)=W(t_{n})-W(n)+\int_{t_{n}}^{n}\frac{2\nu+1}{2Y_{\nu}(s)}\,ds,$ consequently, $|X_{n}-Y_{\nu}(n)|\leq|W(t_{n})-W(n)|+\frac{(2\nu+1)|t_{n}-n|}{2}\max_{\min(n,t_{n})\leq t\leq\max(n,t_{n})}\frac{1}{Y_{\nu}(t)}.$ Now by (3.1) and (2.8) the last term is $O(n^{2\varepsilon})$ almost surely and since for the increments of the Wiener process (cf. [8], page 30) $|W(t_{n})-W(n)|=O(n^{1/4+\varepsilon})\qquad{\rm a.s.}$ as $n\to\infty$, we have (1.15) of Theorem 1.1. $\Box$ ## 4\. Proof of Theorem 1.2 For $R>0$ integer define $\displaystyle\kappa_{1}$ $\displaystyle:=$ $\displaystyle\min\\{t\geq 0:\,Y_{\nu}(t)=R\\},$ $\displaystyle\delta_{1}$ $\displaystyle:=$ $\displaystyle\min\\{t\geq\kappa_{1}:\,Y_{\nu}(t)\notin(R-1,R+1)\\},$ $\displaystyle\kappa_{i}$ $\displaystyle:=$ $\displaystyle\min\\{t\geq\delta_{i-1}:\,Y_{\nu}(t)=R\\},$ $\displaystyle\delta_{i}$ $\displaystyle:=$ $\displaystyle\min\\{t\geq\kappa_{i}:\,Y_{\nu}(t)\notin(R-1,R+1)\\},$ $\displaystyle\kappa^{*}$ $\displaystyle:=$ $\displaystyle\max\\{t\geq 0\,:Y_{\nu}(t)=R\\},$ $i=2,3,\ldots$ Consider the local times at $R$ of the Bessel process during excursions around $R$, i.e. let $\zeta_{i}:=\eta(R,\delta_{i})-\eta(R,\kappa_{i}),\quad i=1,2,\ldots,$ $\tilde{\zeta}:=\eta(R,\infty)-\eta(R,\kappa^{*}).$ We have $\eta(R,\infty)=\sum_{i=1}^{\xi(R,\infty)-1}\zeta_{i}+\tilde{\zeta}.$ ###### Lemma 4.1. ${\bf E}\left(e^{\lambda\eta(R,\infty)}\right)=\frac{p^{*}_{R}\,\varphi(\lambda)}{1-q_{R}^{*}\,\varphi(\lambda)},$ (4.1) where $p_{R}^{*}=\frac{A_{R}}{A_{R}+B_{R}}\,\frac{(R+1)^{2\nu}-R^{2\nu}}{(R+1)^{2\nu}},\quad q_{R}^{*}=1-p_{R}^{*},$ (4.2) $\varphi(\lambda)=\frac{\nu(A_{R}+B_{R})}{\nu(A_{R}+B_{R})-\lambda R^{2\nu+1}A_{R}B_{R}},$ (4.3) and $A_{R}=(R-1)^{-2\nu}-R^{-2\nu},\qquad B_{R}=R^{-2\nu}-(R+1)^{-2\nu}.$ (4.4) Proof: By ([2], p. 395, 3.3.2) $\zeta_{i}$ are i.i.d. random variables having exponential distribution with moment generating function $\varphi(\lambda)$ given in (4.3). Moreover, it is obvious that $\tilde{\zeta}$ is independent from $\sum_{i=1}^{\xi(R,\infty)-1}\zeta_{i}.$ Furthermore, $\tilde{\zeta}$ is the local time of $R$ under the condition that starting from $R,$ $Y_{\nu}(t)$ will reach $R+1$ before $R-1.$ Hence its distribution can be calculated from formula 3.3.5(b) of [2], and its moment generating function happens to be equal to $\varphi(\lambda)$ of (4.3). $\Box$ We can see $\theta:={\bf E}(\zeta_{i})={\bf E}(\tilde{\zeta})=\frac{\nu(A_{R}+B_{R})}{R^{2\nu+1}A_{R}B_{R}}=1+O\left(\frac{1}{R}\right),\quad R\to\infty.$ ${\bf P}(|\eta(R,\infty)-\xi(R,\infty)|\geq u)={\bf P}\left(\left|\sum_{i=1}^{\xi(R,\infty)-1}(\zeta_{i}-\theta)+\tilde{\zeta}-\theta\right|\geq u\right)$ $\leq{\bf P}(\xi(R.\infty)>N)+{\bf P}\left(\max_{k\leq N}\left|\sum_{i=1}^{k}(\zeta_{i}-\theta)\right|\geq u\right)$ $\leq(q^{*}_{R})^{N}+e^{-\lambda u}\left(\left(\frac{e^{\lambda\theta}}{1+\lambda\theta}\right)^{N}+\left(\frac{e^{-\lambda\theta}}{1-\lambda\theta}\right)^{N}\right).$ In the above calculation we used the common moment generating function (4.3) of $\zeta_{i}$ and $\tilde{\zeta}$, the exact distribution of $\xi(R,\infty)$ (see (1.8)) and the exponential Kolmogorov inequality. Estimating the above expression with standard methods and selecting $N=CR\log R,\quad u=CR^{1/2}\log R,\quad\lambda=\frac{u}{\theta^{2}N}$ we conclude that ${\bf P}(|\eta(R,\infty)-\xi(R,\infty)|\geq CR^{1/2}\log R)\leq C_{1}\exp{\left(-\frac{C\log R}{2\theta}\right)}.$ With a big enough $C$ the right hand side of the above inequality is summable in $R,$ hence Theorem 1.2 follows by the Borel-Cantelli lemma. $\Box$ ## 5\. Proof of Theorem 1.3 Let $p_{j}^{(1)}$ and $p_{j}^{(2)}$ as in Theorem 1.3. Define the two- dimensional Markov chain $(X_{n}^{(1)},X_{n}^{(2)})$ as follows. If $p_{j}^{(1)}\geq p_{k}^{(2)}$, then let $\displaystyle{\bf P}\left((X_{n+1}^{(1)},X_{n+1}^{(2)})=(j+1,k+1)\mid(X_{n}^{(1)},X_{n}^{(2)})=(j,k)\right)$ $\displaystyle=$ $\displaystyle\frac{1}{2}+p_{k}^{(2)}$ $\displaystyle{\bf P}\left((X_{n+1}^{(1)},X_{n+1}^{(2)})=(j+1,k-1)\mid(X_{n}^{(1)},X_{n}^{(2)})=(j,k)\right)$ $\displaystyle=$ $\displaystyle p_{j}^{(1)}-p_{k}^{(2)}$ $\displaystyle{\bf P}\left((X_{n+1}^{(1)},X_{n+1}^{(2)})=(j-1,k-1)\mid(X_{n}^{(1)},X_{n}^{(2)})=(j,k)\right)$ $\displaystyle=$ $\displaystyle\frac{1}{2}-p_{j}^{(1)}.$ If, however $p_{j}^{(1)}\leq p_{k}^{(2)}$, then let $\displaystyle{\bf P}\left((X_{n+1}^{(1)},X_{n+1}^{(2)})=(j+1,k+1)\mid(X_{n}^{(1)},X_{n}^{(2)})=(j,k)\right)$ $\displaystyle=$ $\displaystyle\frac{1}{2}+p_{j}^{(1)}$ $\displaystyle{\bf P}\left((X_{n+1}^{(1)},X_{n+1}^{(2)})=(j-1,k+1)\mid(X_{n}^{(1)},X_{n}^{(2)})=(j,k)\right)$ $\displaystyle=$ $\displaystyle p_{k}^{(2)}-p_{j}^{(1)}$ $\displaystyle{\bf P}\left((X_{n+1}^{(1)},X_{n+1}^{(2)})=(j-1,k-1)\mid(X_{n}^{(1)},X_{n}^{(2)})=(j,k)\right)$ $\displaystyle=$ $\displaystyle\frac{1}{2}-p_{k}^{(2)}.$ Then it can be easily seen that $X_{n}^{(1)}$ and $X_{n}^{(2)}$ are two NN random walks as desired. Consider the following 4 cases. * • (i) $p_{j}^{(1)}\leq p_{k}^{(2)}$, $j\leq k$, * • (ii) $p_{j}^{(1)}\leq p_{k}^{(2)}$, $j\geq k$, * • (iii) $p_{j}^{(1)}\geq p_{k}^{(2)}$, $j\leq k$, * • (iv) $p_{j}^{(1)}\geq p_{k}^{(2)}$, $j\geq k$. In case (i) from (1.17) and (1.18) we obtain $\frac{B}{4j}-\frac{C}{j^{\gamma}}\leq\frac{B}{4k}+\frac{C}{k^{\gamma}}\leq\frac{B}{4k}+\frac{C}{kj^{\gamma-1}},$ implying $k-j\leq\frac{2Cj^{2-\gamma}}{B/4-Cj^{1-\gamma}}=O(j^{2-\gamma})$ if $j\to\infty$. So in this case if $X_{n}^{(1)}=j$ and $X_{n}^{(2)}=k$, then we have $0\leq X_{n}^{(2)}-X_{n}^{(1)}=O((X_{n}^{(1)})^{2-\gamma})$ if $n\to\infty$. In case (ii) either $X_{n+1}^{(1)}-X_{n+1}^{(2)}=X_{n}^{(1)}-X_{n}^{(2)}$, or $X_{n+1}^{(1)}-X_{n+1}^{(2)}=X_{n}^{(1)}-X_{n}^{(2)}-2$, so that we have $-2\leq X_{n+1}^{(1)}-X_{n+1}^{(2)}\leq X_{n}^{(1)}-X_{n}^{(2)}.$ Similar procedure shows that in case (iii) $-2\leq X_{n+1}^{(2)}-X_{n+1}^{(1)}\leq X_{n}^{(2)}-X_{n}^{(1)}$ and in case (iv) $0\leq X_{n}^{(1)}-X_{n}^{(2)}=O((X_{n}^{(2)})^{2-\gamma}).$ Hence Theorem 1.3 follows from the law of the iterated logarithm for $X_{n}^{(i)}$ (cf. [3]). $\Box$ ## 6\. Strong theorems As usual, applying Theorem 1.1 and Theorem 1.3, we can give limit results valid for one of the processes to the other process involved. In this section we denote $Y_{\nu}(t)$ by $Y(t)$ and define the following related processes. $M(t):=\max_{0\leq s\leq t}Y(s),\qquad Q_{n}:=\max_{1\leq k\leq n}X_{k}.$ The future infimums are defined as $I(t):=\inf_{s\geq t}Y(s),\qquad J_{n}:=\inf_{k\geq n}X_{k}.$ Escape processes are defined by $A(t):=\sup\\{s:\,Y(s)\leq t\\},\qquad G_{n}:=\sup\\{k:\,X_{k}\leq n\\}.$ Laws of the iterated logarithm are known for Bessel processes (cf. [2]) and NN random walks (cf. [3]) as well. Upper class results for Bessel process read as follows (cf. Orey and Pruitt [16] for integral $d$, and Pardo [17] for the case of positive self-similar Markov processes). Theorem F: Let $a(t)$ be a non-decreasing non-negative continuous function. Then for $\nu\geq 0$ $\displaystyle{t^{1/2}a(t)\in{\rm UUC}(Y(t))\qquad if\,\,and\,\,only\,\,if\qquad\int_{1}^{\infty}\frac{(a(x))^{2\nu+2}}{x}e^{-a^{2}(x)/2}\,dx<\infty.}$ Now Theorems 1.1, 1.3 and Theorems E and F together imply the following result. ###### Theorem 6.1. Let $\\{X_{n}\\}$ be an NN random walk with $p_{R}$ satisfying $p_{R}=\frac{B}{4R}+O\left(\frac{1}{R^{1+\delta}}\right),\quad R\to\infty$ with $B>1$ and for some $\delta>0$. Let furthermore $a(t)$ be a non-decreasing non-negative function. Then $\displaystyle{n^{1/2}a(n)\in{\rm UUC}(X_{n})\qquad if\,\,and\,\,only\,\,if\qquad\sum_{k=1}^{\infty}\frac{(a(k))^{B+1}}{k}e^{-a^{2}(k)/2}<\infty.}$ If $b(t)$ is a non-increasing non-negative function, then $\displaystyle{n^{1/2}b(n)\in{\rm LLC}(X_{n})\qquad if\,\,and\,\,only\,\,if\qquad\sum_{k=1}^{\infty}(b(2^{k}))^{B-1}<\infty.}$ Next we prove the following invariance principles for the processes defined above. ###### Theorem 6.2. Let $Y(t)$ and $X_{n}$ as in Theorem 1.1. Then for any $\varepsilon>0$ we have $|M(n)-Q_{n}|=O(n^{1/4+\varepsilon})\quad{\rm a.s.}$ (6.1) and $|I(n)-J_{n}|=O(n^{1/4+\varepsilon})\quad{\rm a.s.}$ (6.2) Proof: Define $\tilde{s},s^{*},\tilde{k},k^{*}$ by $Y(\tilde{s})=M(n),\quad Y(s^{*})=I(n),\quad X_{\tilde{k}}=Q_{n},\quad X_{k^{*}}=J_{n}.$ Then as $n\to\infty$, we have almost surely $Q_{n}-M(n)=X_{\tilde{k}}-Y(\tilde{s})\leq X_{\tilde{k}}-Y(\tilde{k})=O(n^{1/4+\varepsilon})$ and $M(n)-Q_{n}=Y(\tilde{s})-X_{\tilde{k}}=Y(\tilde{s})-Y([\tilde{s}])-(X_{[\tilde{s}]}-Y([\tilde{s}]))+X_{[\tilde{s}]}-X_{\tilde{k}}$ $\leq Y(\tilde{s})-Y([\tilde{s}])-(X_{[\tilde{s}]}-Y([\tilde{s}])=Y(\tilde{s})-Y([\tilde{s}])+O(n^{1/4+\varepsilon})$ By (3.2) and recalling the results on the increments of the Wiener process (see [8] page 30) we get $Y(\tilde{s})-Y([\tilde{s}])=W(\tilde{s})-W([\tilde{s}])+\int_{[\tilde{s}]}^{\tilde{s}}\frac{2\nu+1}{2Y(s)}\,ds$ $\leq\sup_{0\leq t\leq n}\sup_{0\leq s\leq 1}|W(t+s)-W(t)|+\frac{2\nu+1}{2}\max_{[\tilde{s}]\leq t\leq\tilde{s}}\frac{1}{Y(t)}=O(\log n)\quad{\rm a.s.},$ since $Y(t)$ in the interval $([\tilde{s}],\tilde{s})$ is bounded away from zero. Hence (6.1) follows. To show (6.2), note that $n\leq s^{*}\leq n^{1+\alpha}$ and $n\leq k^{*}\leq n^{1+\alpha}$ for any $\alpha>0$ almost surely for all large $n$. Then as $n\to\infty$ $I(n)-J_{n}\leq Y(k^{*})-X_{k^{*}}=O((k^{*})^{1/4+\varepsilon})=O(n^{(1+\alpha)(1/4+\varepsilon)})\quad{\rm a.s.}$ On the other hand, $J_{n}-I(n)\leq X_{k^{*}}-Y([s^{*}])+Y([s^{*}])-Y(s^{*})=O(n^{(1+\alpha)(1/4+\varepsilon)})+Y([s^{*}])-Y(s^{*}).$ By (3.2), taking into account that when applying this formula the integral contribution is negative, and recalling again the results on the increments of the Wiener process, we get $Y([s^{*}])-Y(s^{*})\leq W([s^{*}])-W(s^{*})\leq\sup_{0\leq t\leq n^{1+\alpha}}\sup_{0\leq s\leq 1}|W(t+s)-W(t)|=O(\log n)\quad{\rm a.s.}$ as $n\to\infty$. Hence $|I(n)-J_{n}|=O(n^{(1+\alpha)(1/4+\varepsilon)})\quad{\rm a.s.}$ Since $\alpha>0$ and $\varepsilon>0$ are arbitrary, (6.2) follows. This completes the proof of Theorem 6.2. $\Box$ ###### Theorem 6.3. Let $X_{n}^{(1)}$ and $X_{n}^{(2)}$ as in Theorem 1.3 and let $Q_{n}^{(1)}$ and $Q_{n}^{(2)}$ be the corresponding maximums, while let $J_{n}^{(1)}$ and $J_{n}^{(2)}$ be the corresponding future infimum processes. Then for any $\varepsilon>0$, as $n\to\infty$ we have $|Q_{n}^{(1)}-Q_{n}^{(2)}|=O(n^{1-\gamma/2+\varepsilon})\quad{\rm a.s.}$ (6.3) and $|J_{n}^{(1)}-J_{n}^{(2)}|=O(n^{1-\gamma/2+\varepsilon})\quad{\rm a.s.}$ (6.4) Proof: Define $\tilde{k}_{i},k_{i}^{*},\,i=1,2$ by $X^{(i)}_{\tilde{k}_{i}}=Q_{n}^{(i)},\quad X_{k_{i}^{*}}^{(i)}=J_{n}^{(i)}.$ Then $|Q_{n}^{(1)}-Q_{n}^{(2)}|\leq\max(X_{\tilde{k}_{1}}^{(1)}-X_{\tilde{k}_{1}}^{(2)},X_{\tilde{k}_{2}}^{(1)}-X_{\tilde{k}_{2}}^{(2)})=O((n\log\log n)^{1-\gamma/2})\quad{\rm a.s.},$ proving (6.3). Moreover, for any $\alpha>0$, $n\leq k_{i}^{*}\leq n^{1+\alpha}$ almost surely for large $n$, hence we have $|J_{n}^{(1)}-J_{n}^{(2)}|\leq\max(X_{k_{1}^{*}}^{(1)}-X_{k_{1}^{*}}^{(2)},X_{k_{2}^{*}}^{(1)}-X_{k_{2}^{*}}^{(2)})=O((n\log\log n)^{(1+\alpha)(1-\gamma/2)})\quad{\rm a.s.}$ Since $\alpha$ is arbitrary, (6.4) follows. This completes the proof of Theorem 6.3. $\Box$ Khoshnevisan et al. [13] (for $I(t)$ and $A(t)$), Adelman and Shi [1], and Shi [20] (for $Y(t)-I(t)$) proved the following upper and lower class results. Theorem G: Let $\varphi(t)$ be a non-increasing, and $\psi(t)$ be a non- decreasing function, both non-negative. Then for $\nu>0$ * • $\displaystyle{t^{1/2}\psi(t)\in{\rm UUC}(I(t))}$ if and only if $\displaystyle{\int_{1}^{\infty}\frac{(\psi(x))^{2\nu}}{x}e^{-\psi^{2}(x)/2}\,dx<\infty,}$ * • $\displaystyle{t^{2}\varphi(t)\in{\rm LLC}(A(t))}$ if and only if $\displaystyle{\int_{1}^{\infty}\frac{1}{x\varphi^{\nu}(x)}e^{-1/2\varphi(x)}\,dx<\infty.}$ * • $\displaystyle{t^{1/2}\psi(t)\in{\rm UUC}(Y(t)-I(t))}$ if and only if $\displaystyle{\int_{1}^{\infty}\frac{1}{x\psi^{2\nu-2}(x)}e^{-\psi^{2}(x)/2}\,dx<\infty,}$ Theorem H: Let $\rho(t)>0$ be such that $(\log\rho(t))/\log t$ is non- decreasing. Then * • $\displaystyle{1/\rho(t)\in{\rm LLC}(M(t)-I(t))}$ if and only if $\displaystyle{\int_{1}^{\infty}\frac{dx}{x\log\rho(x)}<\infty.}$ Taking into account that $J_{n}$ and $G_{n}$ are inverses of each other, immediate consequences of Theorems F, G, H, Theorems 6.2 and 6.3 are the following upper and lower class results. ###### Theorem 6.4. Let $X_{n}$ be as in Theorem 6.1 and let $\varphi(t)$ be a non-increasing and $\psi(t)$ be a non-decresing function, both non-negative. Then * • $\displaystyle{n^{1/2}\psi(n)\in{\rm UUC}(J_{n})}$ if and only if $\displaystyle{\sum_{k=1}^{\infty}\frac{(\psi(k))^{B-1}}{k}e^{-\psi^{2}(k)/2}<\infty,}$ * • $\displaystyle{n^{2}\varphi(n)\in{\rm LLC}(G_{n})}$ if and only if $\displaystyle{\sum_{k=1}^{\infty}\frac{1}{k\varphi^{(B-1)/2}(k)}e^{-1/2\varphi(k)}<\infty.}$ * • $\displaystyle{n^{1/2}\psi(n)\in{\rm UUC}(X_{n}-J_{n})}$ if and only if $\displaystyle{\sum_{k=1}^{\infty}\frac{1}{k\psi^{B-3}(k)}e^{-\psi^{2}(k)/2}<\infty,}$ ###### Theorem 6.5. Let $\rho(t)>0$ be such that $(\log\rho(t))/\log t$ is non-decreasing. * • $\displaystyle{1/\rho(n)\in{\rm LLC}(Q_{n}-J_{n})}$ if and only if $\displaystyle{\sum_{k=2}^{\infty}\frac{1}{k\log\rho(k)}<\infty.}$ ## 7\. Local time We will need the following result from Yor [22], page 52. Theorem J: For the local time of a Bessel process of order $\nu$ we have $\eta(R,\infty)\stackrel{{\scriptstyle{\cal D}}}{{=}}(2\nu)^{-1}R^{1-2\nu}Y_{0}^{2}(R^{\,2\nu}),$ where $Y_{0}$ is a two-dimensional Bessel process and $\stackrel{{\scriptstyle{\cal D}}}{{=}}$ means equality in distribution. Hence applying Theorem F for $\nu=0$, we get Theorem K: If $f(x)$ is non-decreasing, non-negative function, then * • $\displaystyle{Rf(R)\in{\rm UUC}(\eta(R,\infty))}$ if and only if $\displaystyle{\int_{1}^{\infty}\frac{f(x)}{x}e^{-\nu f(x)}\,dx<\infty}.$ From this and Theorem 1.2 we get the following result. ###### Theorem 7.1. If $f(x)$ is non-decreasing, non-negative function, then * • $\displaystyle{Rf(R)\in{\rm UUC}(\xi(R,\infty))}$ if and only if $\displaystyle{\sum_{k=1}^{\infty}\frac{f(k)}{k}e^{-\nu f(k)}<\infty.}$ In [7] we proved the following result. Theorem L: Let $\displaystyle{p_{R}=\frac{B}{4R}+O\left(\frac{1}{R^{\gamma}}\right)}$ with $B>1$, and $\gamma>1$. Then with probability $1$ there exist infinitely many $R$ for which $\xi(R+j,\infty)=1$ for each $j=0,1,2,\ldots,[\log\log R/\log 2]$. Moreover, with probability $1$ for each $R$ large enough and $\varepsilon>0$ there exists an $R\leq S\leq{\frac{(1+\varepsilon)\log\log R}{\log 2}}$ such that $\xi(S,\infty)>1.$ Remark 1: In fact in [7] we proved this result in the case when $p_{R}=B/{4R}$ but the same proof works also in the case of Theorem L. This theorem applies e.g. for the case when $p_{R}$ is given by (1.14), which in turn, gives the following result for the Bessel process. Let * (i) $\kappa(R):=\inf\\{t:\ Y_{\nu}(t)=R\\}$, * (ii) $\kappa^{*}(R):=\sup\\{t:\ Y_{\nu}(t)=R\\}$, * (iii) $\Psi(R)$ be the largest integer for which the event $A(R)=\bigcap_{j=-1}^{\Psi(R)}\\{\kappa^{*}(R+j)<\kappa(R+j+1)\\}$ occurs. $A(R)$ means that $Y_{\nu}(t)$ moves from $R$ to $R+1$ before returning to $R-1$, it goes from $R+1$ to $R+2$ before returning to $R$, $\ldots$ and also from $R+\Psi(R)$ to $R+\Psi(R)+1$ and it never returns to $R+\Psi(R)-1$. We say that the process $Y_{\nu}(t)$ escapes through $(R,R+\Psi(R))$ with large velocity. ###### Theorem 7.2. $\limsup_{R\to\infty}{\frac{\Psi(R)}{\log\log R}}={\frac{1}{\log 2}}\quad{\rm a.s.}$ Remark 2: The statement of Theorem 7.2 (for integral $d=2\nu+2$) was formulated in [18], p. 291 as a Conjecture. ## References * [1] ADELMAN, O. and SHI, Z.: The measure of the overlap of past and future under a transient Bessel process. Stochastics Stochastics Rep. 57 (1996), 169–183. * [2] BORODIN, A.N. and SALMINEN, P.: Handbook of Brownian Motion – Facts and Formulae. Birkhäuser, Basel, (1996). * [3] BRÉZIS, H., ROSENKRANTZ, W. and SINGER, B.: An extension of Khintchine’s estimate for large deviations to a class of Markov chains converging to a singular diffusion. Comm. Pure Appl. Math. 24 (1971), 705–726. * [4] CHAUMONT, L. and PARDO, J.C.: The lower envelope of positive self-similar Markov Processes. Electr. J. Probab. 11 (2006), 1321–1341. * [5] CHUNG, K.L.: Markov Chains with Stationary Transition Probabilities. 2nd ed. Springer-Verlag, New York, 1967\. * [6] COOLIN-SCHRIJNER, P. and VAN DOORN, E.A.: Analysis of random walks using orthogonal polynomials. J. Comput. Appl. Math. 99 (1998), 387–399. * [7] CSÁKI, E., FÖLDES, A. and RÉVÉSZ, P.: Transient nearest neighbor random walk on the line. J.Theor. Probab., to appear. * [8] CSÖRGŐ, M. and RÉVÉSZ, P.: Strong Approximations in Probability and Statistics. Academic Press, New York, 1981. * [9] DETTE, H.: First return probabilities of birth and death chains and associated orthogonal polynomials. Proc. Amer. Math. Soc. 129 (2001), 1805–1815. * [10] DVORETZKY, A. and ERDŐS, P.: Some problems on random walk in space. Proc. Second Berkeley Symposium on Mathematical Statistics and Probability, 1950. pp. 353–367. University of California Press, Berkeley and Los Angeles, 1951. * [11] ERDÉLYI, A. MAGNUS, W. OBERHETTIGER, F. and TRICOMI, F.G.: Higher Transcendental Functions. Vol. 2, McGraw - Hill, New York, 1953. * [12] KARLIN, S. and McGREGOR, J.: Random walks. Illinois J. Math. 3 (1959), 66–81. * [13] KHOSHNEVISAN, D., LEWIS, M.L. and SHI, Z.: On a problem of Erdős and Taylor. Ann. Probab. 24 (1996), 761–787. * [14] KNIGHT, F.B.: Essentials of Brownian Motion and Diffusion. Am. Math. Soc., Providence, R.I., 1981. * [15] LAMPERTI, J.: A new class of probability limit theorems. J. Math. Mech. 11 (1962), 749–772. * [16] OREY, S. and PRUITT, W.E.: Sample functions of the $N$-parameter Wiener process. Ann. Probab. 1 (1973), 138–163. * [17] PARDO, J.C.: The upper envelope of positive self-similar Markov processes. arXiv:math.PR/0703071 * [18] RÉVÉSZ, P.: Random Walk in Random and Non-Random Environments. 2nd ed. World Scientific, Singapore, 2005\. * [19] REVUZ, D. and YOR, M.: Continuous Martingales and Brownian Motion. 3rd ed. Springer-Verlag, Berlin, 1999. * [20] SHI, Z.: How long does it take a transient Bessel process to reach its future infimum? Séminaire de Probabilités, XXX, Lecture Notes in Math., 1626, Springer, Berlin, 1996, 207–217, * [21] WATSON, G.N.: A Treatise on the Theory of Bessel Functions. 2nd ed. Cambridge University Press, Cambridge, 1944. * [22] YOR, M.: Some Aspects of Brownian Motion. Part I: Some Special Functionals. Birkhäuser, Basel, 1992.

arxiv-papers

2024-09-04T02:48:53.603053

0802.0847

# Collective behavior of ”electronic fireflies” M. Ercsey-Ravasz1,2, Zs. Sárközi1, Z. Néda1, A. Tunyagi1, and I. Burda1 1 Babeş-Bolyai University, Faculty of Physics, RO-400084, Cluj, Romania 2 Péter Pázmány Catholic University, Faculty of Information Technology, HU-1083, Budapest, Hungary ###### Abstract A simple system composed of electronic oscillators capable of emitting and detecting light-pulses is studied. The oscillators are biologically inspired, their behavior is designed for keeping a desired light intensity, $W$, in the system. From another perspective, the system behaves like modified integrate and fire type neurons that are pulse-coupled with inhibitory type interactions: the firing of one oscillator delays the firing of all the others. Experimental and computational studies reveal that although no driving force favoring synchronization is considered, for a given interval of $W$ phase-locking appears. This weak synchronization is sometimes accompanied by complex dynamical patterns in the flashing sequence of the oscillators. ###### pacs: 05.45.Xt, 89.75.Fb, 05.45.-a ††preprint: APS/123-QED ## I Introduction Synchronization of quasi-identical coupled oscillators is one of the oldest and most fascinating problems in physics strogatz ; stewart ; strogatz2 . Its history goes back to C. Huygens who first noticed the synchronization of pendulum clocks hanging on the same wall. Besides mechanical or electric oscillator systems, nature is also full with several amazing examples in this sense glass ; winfree ; neda1 . Synchronization in all these systems appears as a result of some specific coupling between the units. This coupling can be local or global, and can be realized through a phase-difference minimizing force kuramoto ; gomez or through the pulses emitted and detected by the oscillators bottani ; pikovsky . In most of these synchronizing systems there is a clear driving force favoring synchronization, and in such way the appearance of this collective behavior is somehow trivial. In the present work however, a nontrivial synchronization will be presented. This weak synchronization (phase-locking) appears as a co-product of a simple collective optimization rule. One well-known phenomena which inspired us in this work is the collective behavior and synchronization of fireflies fireflies . Although our aim here is not to model fireflies, the oscillators (”electronic fireflies”) considered in our system are somehow similar to them: they are capable of emitting light- pulses and detecting the light-pulse of the others. In this sense our system is similar to an ensemble of fireflies although the coupling between the units is different. From another perspective, the oscillators behave like pulse- coupled ”integrate and fire” type neurons bottani ; pikovsky ; neda2 . Contrary to the classical integrate and fire oscillators, in the considered system an inhibitory type global interaction is considered. This means that the firing of one oscillator delays (and not advances) the phase of all the others. This system does not necessarily favor synchronization, it is rather designed to keep a desired $W$ light intensity in the system. This light intensity is controlled by a firing threshold parameter $G$ imposed globally on the oscillators. Surprisingly, as a co-product of this simple rule, for certain region of the firing threshold parameter phase-locking and complex patterns in the flashing sequence of the oscillators will appear. We believe that such dynamical laws could be realistic for many biological systems. The studied system will be described in more details in the following section. The used electronic device will be briefly presented and the obtained non- trivial collective behavior will be studied. In order to get more confidence in the observed non-trivial results computer simulations were also performed. ## II The experimental setup Figure 1: Experimental setup. The photo on the left shows the ”electronic fireflies” (oscillators) placed on the circuit board. The photo on the right shows one oscillator. The constructed units are integrate and fire type oscillators bottani with a modified interaction rule. Their coupling and communication is through light, the units are capable of emitting and detecting light-pulses. The oscillators are practically realized by a relatively simple circuit, the main active elements being a photoresistor and a Light Emitting Diode (LED). Each oscillator, $i$, has a characteristic voltage $U_{i}$, which depends on the resistance, $R_{i}$, of its photoresistor. The global light intensity influences the value of $R_{i}$ in the following sense: when the light intensity increases $R_{i}$ decreases, leading to a decrease in $U_{i}$. In the system there is a global controllable parameter $G$, identical for all oscillators. By changing the parameter $G$, one can control the average light intensity output, $W$, of the whole system. If the voltage of the oscillator grows above this threshold ($U_{i}>G$) the oscillator will fire, this meaning its LED will flash. This flash occurs only if a minimal time period $T_{min_{i}}$ has expired since the last firing. The oscillator has also a maximal period, meaning if no flash occurred in time $T_{max_{i}}$, then the oscillator will surely fire. In laymen terms firing is favored by darkness and the value of the controllable $G$ parameter characterizes the ”darkness level” at which firing should occur. Through this simple rule the $G$ parameter controls the average light intensity output of the system. The technical realization of the above dynamics is illustrated in Fig. 2. After the system has fired the $22\ \mu$F capacitor is completely discharged by the negative pulse from the inverted output of the monostable. As soon as the light flash ended, the same capacitor will start charging from the current flow through the $270$ K$\Omega$ resistor. The IC1B comparator will trigger another flash as soon as the potential on the mentioned capacitor will overcome the value fixed by the group of three resistors on its positive input (the firing threshold). Two of the resistors are connected to constant potentials (ground and +5 V), the third resistor is connected to the output of the second comparator IC1A which will have a value depending on the ratio between the reference voltage and a certain amount of light measured by the photo resistor. The flash time is determined by the second capacitor together with the $12$ K$\Omega$ resistor connected to the monostable. The photoresistor has a relatively low reaction time around $40$ ms, while the minimal and maximal period of firing are around $800$ ms and $2700$ ms. The time of one flash is around $200$ ms. Figure 2: Circuit diagram of one oscillator. The oscillators are placed on a circuit board in the form of a square lattice (see Figure 1). The maximal number of oscillators which can be included are $24$. A computer interface and program controls the $G$ threshold parameter and allows us to get information automatically about the states of all oscillators. The state of an oscillator is recorded as $0$ if the oscillator does not emit light and $1$ when the oscillator fires (emits light). Whenever the state of the oscillator system changes, the program writes in a file the corresponding time with a precision of milliseconds and the new states of the units. To obtain an enhanced global interaction the whole system is placed inside a closed box. The box has mat glass mirror walls to uniformly disperse the light-pulses in the box. A graphical interface allows to visually control the state of the units. In order to fully understand the behavior of the system one has to accept that the coupling between pairs of oscillators are not exactly of the same strength. Also, the characteristic electronic parameters differ slightly ($2-10\%$) among the units. ## III Collective behavior At constant light intensity one unit behaves as a simple stochastic oscillator. Whenever the $G$ threshold is under a given $G_{c}$ value the oscillator will fire with its minimal period and above $G_{c}$ with its maximal period. $G_{c}$ depends of course on the imposed light intensity. Considering more oscillators ($i=1,\dots,n$) and by letting them interact, interesting collective behavior appears for a certain range of the $G$ threshold parameter. Due to the inhibitory nature of the considered interaction, during the firing of oscillator $i$ the characteristic voltages of the others ($U_{j},j\neq i$) will decrease. If the $G$ parameter is so small, that under this condition the other oscillators can still fire ($U_{j}>G$), than all oscillators will fire in an uncorrelated manner. Each of them will be firing at its own $T_{min_{j}}$ period and the interaction is thus not efficient. In such case no collective behavior can be observed. Increasing the value of $G$ will make the pulse-like interaction efficient. The oscillators will avoid firing simultaneously and a simple phase-locking phenomenon appears. The pulse of one unit (let us assume $i$) delays the firing of the others by decreasing their voltages below the threshold: $U_{j}<G,j\neq i$. Due to the tiny differences in the coupling between the pairs (caused for example by different distances) and in the parameters of the electronic elements, the $U_{j}$ voltages are different. The immediate consequence of this is that the next firing will occur most probably in the oscillator with the highest voltage (counting of course only those oscillators, which are already capable of firing). This oscillator is the one which was influenced the less by the light-pulse of the previous firing. If the total combined time of firing for the $n$ oscillators is smaller than the period $T_{max}$ the result is that after very short time phase-locking appears and a firing chain (with period $T\in[T_{min},T_{max}]$) will form, each oscillator firing in a well-defined sequence. For a given system and a fixed $G$ threshold this stable pattern is always the same. If the total time of firing of the $n$ oscillators exceeds $T_{max}$, the firing pattern will be much longer and more complex. Increasing further and over a limit the $G$ threshold parameter the previously discussed weak synchronization (phase-locking) disappears. In this case the voltages of all oscillators are much smaller than the threshold value $U_{i}<G$, so the firing of a unit can not influence the others. All oscillators will fire with their own $T_{max_{i}}$ period and no interesting collective behavior is observed. Again, the interaction is not efficient. The collective behavior of the system can be easily analyzed by plotting a kind of phase-histogram for the oscillator ensemble. Choosing a reference oscillator, the relative phases of all the others are defined by measuring the time difference between their pulse and the last pulse emitted by the reference oscillator. Studying these time-delays during a longer time period a histogram is constructed for their distribution. This histogram shows how frequently a given time-delay occurred and gives thus a hint whether a constant firing pattern is formed or not. Experimental and computer simulated results for the phase-histogram confirm the above presented scenario of the collective behavior. As an example, on Fig. 3, results obtained on a relatively small system with $n=5$ oscillators are shown. In the first column of Fig. 3 (figures a, b, c and d), experimental results for four different values of the $G$ threshold are plotted. For a small threshold parameter ($G=500$ mV), no self-organization appears (Fig. 3a). Due to the fact that the characteristic time-periods of the oscillators are slightly different, almost all values will occur with the same probability in the phase-histogram. Beginning with $G=1300$ mV a kind of order begins to emerge, and a trend towards the self-organization of the oscillator pulses is observed (e.g. Fig. 3b for $G=2000$ mV). In the neighborhood of $G=3000$ mV threshold value (Fig. 3c) clear phase-locking appears. One can observe that a stable firing pattern has formed, each oscillator has an almost exact phase relative to the reference oscillator. For an even higher value (e.g. $G=4200$ mV), disorder sets in again, phase-locking disappears and all oscillators fire independently with their own maximal period (Fig. 3d). Figure 3: Relative phase histogram for $n=5$ oscillators. Experimental results are in the first column, and the corresponding simulation results are in the second column. Figures a) and e) are for $G=500$ mV; figure b) and f) are for $G=2000$ mV; figures c) and g) are for $G=3000$ mV, and figures d) and h) are for $G=4200$ mV. In the second column of Fig. 3 we present the corresponding simulation results. In simulations the parameters of the oscillators are defined as following: the average minimal time period is $T_{min_{i}}=900$ ms, the average maximal period $T_{max_{i}}=2700$ ms, and the average flashing time $T_{flash}=200$ ms. For an easier comparison, the values are chosen to be similar with the real experimental data. We considered a uniform distribution of the oscillators parameter around these average values using a $\pm 50$ ms interval for $T_{min}$ and $T_{max}$ and a $\pm 20$ ms interval for $T_{flash}$. One could argue of course that a Gaussian distribution would be much more appropriate, but given the fact that we simulate here relatively small number of oscillators the exact statistics is irrelevant. Considering some deviations from the average is however important in order to reproduce the collective behavior of the system. An uncorrelated noise in time is also considered. This will randomly shift the $T_{min}$, $T_{max}$ and $T_{flash}$ periods of each oscillator at each cycle. Again, a uniform distribution on a $\pm 20$ ms interval was considered. The characteristic voltages of the oscillators are set to be in the interval $4100\pm 100$ mV in dark, $2100\pm 100$ mV when one single LED is flashing and $1050\pm 100$ mV when two LEDs are flashing simultaneously. Whenever $k$ LEDs are simultaneously flashing the characteristic voltages of the others are considered to be $2100/k\pm 100$ mV, however for $n=5$ oscillators only very rarely happens to have more than two oscillators simultaneously firing. This values were chosen to approximately match the experimental ones, and we do not try here to give a theoretical model for the nonlinear behavior of the photoresistor. Fluctuations in time and among the parameters of the oscillators are again included. Differences in the strength of the coupling between pairs of oscillators are however neglected. Using these parameters, it is assumed that each oscillator can flash whenever its voltage exceeds the threshold $G$. The flashing cannot occur earlier than $T_{min_{i}}$ or later than $T_{max_{i}}$ relatively to its last firing. On Fig. 3e.,f.,g. and h., the simulated phase-histograms of the oscillators are plotted and compared with the corresponding experimental data. The observed experimental results, including the non-trivial synchronization (phase-locking), were successfully reproduced. Figure 4: Order parameters calculated from experimental (circles) and simulation (dashed line) results plotted as a function of the $G$ threshold. Systems with $n=3,5,7,9$ oscillators are considered. It is also possible to define a kind of order-parameter that characterizes the observed synchronization level. Our method for calculating this is the following: 1) A reference oscillator $k$ is chosen and the phases of all oscillators are calculated relative to this oscillator. 2) Let $h_{i}(f)$ denote the value of the normalized phase-histogram for oscillator $i$ ($i={1,\dots,n},i\neq k$) corresponding to phase difference (time difference) value $f$. Since we have a normalized histogram, $h_{i}(f)\in[0,1]$ gives the occurrence probability of phase difference value $f$ during the measurement ($\sum_{f}{h_{i}(f)}=1$). 3) A window of width $a$ is defined (we have chosen $a=30$ ms). Shifting the window with $\Delta f=1$ ms step, for each discretized value of $f$ the sum $H_{i}(f)=\sum_{j=f-a/2}^{f+a/2}h_{i}(j)$ is calculated for each oscillator $i$. 4) Let $r_{k}$ denote the difference between the maximum and minimum value of $H_{i}(f)$ averaged over all oscillators: $r_{k}=\frac{1}{n-1}\sum_{i=1,(i\neq k)}^{n}max(H_{i})-min(H_{i})$. 5) Items 1-4 are repeated considering each oscillator in the system as reference oscillator. Finally, an averaging is performed over all the obtained $r_{k}$ values ($k={1,\dots,n}$). The final order parameter is calculated thus as $r=\langle r_{k}\rangle_{k}$. Averaging as a function of the reference oscillator is beneficial in order to get a smoother curve when only partial phase locking is detected (Figure 3b.). In such cases the phase-diagrams are very sensible on the choice of the reference oscillator. On Fig. 4 the $r$ order parameter is plotted as a function of the $G$ threshold value. Systems with $n=3,5,7$ and $9$ oscillators are considered. Experimental (circles) and simulation results (dashed line) are again in good agreement. The figure also illustrates that for an intermediate $G$ interval value phase-locking appears. This weak synchronization is better ($r$ is bigger) when there are less units in the system. One obvious reason for this is that by increasing $n$ the total time of firing of the oscillators will increase and slowly exceed the value $T_{max}$. As a result of this the firing pattern will change from a simple ”firing chain” to a much longer and more complicated pattern, decreasing the value of the order parameter. From Fig. 4 it is also observable that the experimental results show more intensive fluctuations. The reason for this is probably the complex noise present in the system. ## IV Conclusion A system of electronic oscillators communicating through light-pulses was studied. The units were designed to optimize the average light intensity of the emitted light-pulses, and no direct driving force favoring synchronization was considered. Although our experiments focused on relatively small systems (up to $24$ oscillators) interesting and rich collective behavior was observed. As a nontrivial result it was found that the inhibitory coupling induced a partial phase-locking for a certain interval of the controllable threshold parameter. This weak synchronization was realized by complex flashing patterns of the units. We believe that this study inspires further interesting research projects in which separately programmable oscillators will be studied with various interaction rules. Many other interesting collective behaviors can be obtained by controlling individually the parameters of the units, their interaction rule and the type of coupling between them. Such systems could also yield a new approach to unconventional computing, being in many sense similar with the presently developed CNN computers marek . Acknowledgments. Work supported from a Romanian CNCSIS No.1571 research grant (contract 84/2007) and a Hungarian ONR grant (N00014-07-1-0350). ## References * (1) S.H. Strogatz, Sync: The Emerging Science of Spontaneous Order (Hyperion, New York, 2003) * (2) S.H. Strogatz and I. Stewart, Sci. Am. (Int. Ed.) 269(6), 102 (1993) * (3) S.H. Strogatz, Lect. Notes Biomath., 100 (1993) * (4) L. Glass and M.C. Mackey, From Clocks to Chaos: The Rhythms of Life (Princeton University Press, Princeton, NJ, 1988) * (5) A.T. Winfree, The Geometry of Biological Time (Springer-Verlag, New-York, 1990); A.T. Winfree, J. Theor. Biol. 16, 15 (1967) * (6) Z. Néda, E. Ravasz, Y. Brechet, T. Vicsek and A.L. Barabási, Nature (London) 403, 849 (2000) * (7) Y. Kuramoto and I. Nishikava, J. Stat. Phys. 49, 569 (1987) * (8) J. Gómez-Gardenes, Y. Moreno and A. Arenas, Phys. Rev. Lett. 98, 034101 (2007) * (9) S. Bottani, Phys. Rev. 54, 2334 (1997) * (10) A.S. Pikovsky and J. Kurths, Phys. Rev. Lett. 78, 775 (1997) * (11) J. Buck and E. Buck, Sci. Am., 234(5), 74 (1976) * (12) A. Nikitin, Z. Néda and T. Vicsek, Phys. Rev. Lett. 87, 024101 (2001) * (13) M. Ercsey-Ravasz, T. Roska, and Z. Néda, Eur. Phys. J. B 51, 407 (2006)

arxiv-papers

2024-09-04T02:48:53.609384

0802.0863

# Nucleon structure with two flavors of dynamical domain-wall fermions Huey-Wen Lin hwlin@jlab.org Jefferson Laboratory, 12000 Jefferson Avenue, Newport News, VA 23606 RIKEN-BNL Research Center, Brookhaven National Laboratory, Upton, NY 11973 Tom Blum tblum@phys.uconn.edu Physics Department, University of Connecticut, Storrs, CT 06269-3046 RIKEN-BNL Research Center, Brookhaven National Laboratory, Upton, NY 11973 Shigemi Ohta shigemi.ohta@kek.jp Institute of Particle and Nuclear Studies, KEK, Tsukuba, 305-0801, Japan RIKEN-BNL Research Center, Brookhaven National Laboratory, Upton, NY 11973 Physics Department, Sokendai Graduate U. Adv. Studies, Hayama, Kanagawa 240-0193, Japan Shoichi Sasaki ssasaki@phys.s.u-tokyo.ac.jp Department of Physics, University of Tokyo, Hongo 7-3-1, Bunkyo-ku, Tokyo 113, Japan Takeshi Yamazaki yamazaki@phys.uconn.edu Physics Department, University of Connecticut, Storrs, CT 06269-3046 RIKEN-BNL Research Center, Brookhaven National Laboratory, Upton, NY 11973 (Feb. 4, 2008) ###### Abstract We present a numerical lattice quantum chromodynamics calculation of isovector form factors and the first few moments of the isovector structure functions of the nucleon. The calculation employs two degenerate dynamical flavors of domain-wall fermions, resulting in good control of chiral symmetry breaking. Non-perturbative renormalization of the relevant quark currents is performed where necessary. The DBW2 gauge action is used to further improve the chiral behavior while maintaining a reasonable physical lattice volume. The inverse lattice spacing, $a^{-1}$, is approximately 1.7 GeV. Degenerate up and down dynamical quark masses of approximately 1, 3/4 and 1/2 times the strange quark mass are used. The physical volume of the lattice is about $(1.9\mbox{ fm})^{3}$. The ratio of the isovector vector to axial charges, $g_{A}/g_{V}$, tends to a somewhat lower value than the experimental value as the quark mass is reduced toward the physical point. Momentum-transfer dependences of the isovector vector, axial, induced tensor and induced pseudoscalar form factors are calculated. The Goldberger-Treiman relation holds at low momentum transfer and yields an estimation of the pion-nucleon coupling, $g_{\pi NN}=15.5(1.4)$, where the quoted error is only statistical. We find that the flavor non- singlet quark momentum fraction $\langle x\rangle_{u-d}$ and quark helicity fraction $\langle x\rangle_{\Delta u-\Delta d}$ overshoot their experimental values after linear chiral extrapolation. We discuss possible systematic errors for this discrepancy. An estimate for transversity, $\langle 1\rangle_{\delta u-\delta d}=0.93(6)$ in $\overline{\rm MS}$ at 2 GeV is obtained and a twist-3 polarized moment, $d_{1}$, appears small, suggesting that the Wandzura-Wilczek relation holds approximately. We discuss in detail the systematic errors in the calculation, with particular attention paid to finite volume, excited state contamination, and chiral extrapolations. ###### pacs: 11.15.Ha, 11.30.Rd, 12.38.Aw, 12.38.-t 12.38.Gc ††preprint: RBRC-675,KEK-TH-1150 ## I Introduction In this paper we report numerical lattice quantum chromodynamics (QCD) calculations of the isovector form factors and moments of the isovector structure functions of the nucleon on gauge field configurations with two degenerate flavors of dynamical domain-wall fermions (DWF) Kaplan (1992); Shamir (1993); Furman and Shamir (1995) generated by the RIKEN-BNL-Columbia (RBC) Collaboration. Four isovector form factors parameterize neutron $\beta$ decay: the vector and induced tensor form factors from the vector current, $\langle p|V^{+}_{\mu}(x)|n\rangle=\bar{u}_{p}\left[\gamma_{\mu}g_{V}(q^{2})-q_{\lambda}\sigma_{\lambda\mu}{g_{T}(q^{2})}\right]u_{n}e^{iq\cdot x},$ (1) and the axial and induced pseudoscalar form factors from the axial current, $\langle p|A^{+}_{\mu}(x)|n\rangle=\bar{u}_{p}\left[\gamma_{\mu}\gamma_{5}g_{A}(q^{2})+iq_{\mu}\gamma_{5}{g_{P}(q^{2})}\right]u_{n}e^{iq\cdot x},$ (2) which are given here in the Euclidean metric convention. Thus, $q^{2}$ as used in this paper stands for Euclidean four-momentum squared, and corresponds to the time-like momentum squared as $q_{M}^{2}=-q^{2}<0$ in Minkowski space. We also note a slight change in the sign convention of the pseudoscalar form factor from our earlier publications, such as Ref. Sasaki et al. (2003). Here $q=p_{n}-p_{p}$ is the momentum transfer between the proton ($p$) and neutron ($n$). In the limit $|{\vec{q}}|\rightarrow 0$, the momentum transfer should be small because the mass difference of the neutron and proton is only about 1.3 MeV. This makes the limit $q^{2}\rightarrow 0$, where the vector and axial form factors dominate, a good approximation. Their values in this limit are called the vector and axial charges of the nucleon: $g_{V}=g_{V}(q^{2}=0)$ and $g_{A}=g_{A}(q^{2}=0)$. Experimentally, $g_{V}=\cos\theta_{C}$ (with the Cabibbo mixing angle $\theta_{C}$), and $g_{A}=1.2695(29)\times g_{V}$ Yao et al. (2006). These form factors are calculable on the lattice Liu et al. (1995); Gockeler et al. (2005); Alexandrou et al. (2006); Hagler et al. (2007); Alexandrou et al. (2007); Gockeler et al. (2007a); Yamazaki and Ohta (2007); Sasaki and Yamazaki (2007) but quite often are prohibitively complicated if one uses conventional staggered or Wilson fermions. The staggered fermions, with their complicated flavor/taste structure, make even definition of baryon operators difficult. The Wilson fermions make the necessary current renormalization complicated due to large explicit violation of chiral symmetry. The domain- wall fermions (DWF) Kaplan (1992, 1993); Shamir (1993); Furman and Shamir (1995), with their exponential suppression of chiral symmetry breaking, make such renormalizations easy. In particular, currents that are connected by chiral transformation such as vector and axial ones should share the same renormalization. Thus, the ratio $g_{A}/g_{V}$ is naturally renormalized in DWF lattice calculations Sasaki et al. (2003). Earlier (mostly quenched) lattice calculations Fukugita et al. (1995); Liu et al. (1994); Gockeler et al. (1996) of this ratio gave significant underestimations by up to 20%. In a quenched calculation with DWF we found much of this deficit comes from the small lattice volumes employed in those earlier studies Sasaki et al. (2003). With the sufficiently large lattice volume of about $(2.4\mbox{ fm})^{3}$ used in our quenched calculation with DWF and the improved DBW2 gauge action Takaishi (1996); de Forcrand et al. (2000), we observed that the ratio does not depend strongly on the quark mass and obtained a value of $1.212\pm 0.027({\rm stat})\pm 0.024({\rm norm})$ in the chiral limit. The value is almost consistent with experiment, and combined with the very weak dependence on the quark mass, does not require any chiral extrapolation but a linear one in quark mass. Thus an obvious question arises: do these observations hold when the quarks are treated dynamically? There has been recent interest in the pseudoscalar form factor, $g_{P}(q^{2})$, due to the MuCap Collaboration’s new high-precision experiment studying ordinary muon capture (OMC) by protons, $\mu^{-}p\to\nu_{\mu}n$ Andreev et al. (2007). The OMC experiments determine the induced pseudoscalar coupling $g_{P}=m_{\mu}g_{P}(q^{2})$ at $q^{2}=0.88m_{\mu}^{2}$, where $m_{\mu}$ is the muon mass. The new experiment yields $g_{P}=7.3\pm 1.1$, which is consistent with the value predicted by chiral perturbation theory, $g_{P}^{\rm ChPT}=8.26\pm 0.16$ Bernard et al. (2002), but much smaller than the last world average for OMC, $(g_{P}^{\rm OMC})_{\rm ave}=10.5\pm 1.8$ given in Ref. Gorringe and Fearing (2004), and the value from a single experiment of radiative muon capture (RMC), $\mu^{-}p\to\nu_{\mu}n\gamma$, $g_{P}^{\rm RMC}=12.4\pm 1.0$ Jonkmans et al. (1996). Recently, a quenched DWF calculation Sasaki and Yamazaki (2007) reported a result, $g_{P}=8.15\pm 0.54({\rm stat})\pm 0.16({\rm norm})$, consistent with the MuCap experiment. It is interesting to see what value is obtained in a $n_{f}=2$ dynamical DWF calculation. In this paper we also report on this form factor, $g_{P}(q^{2})$. It is the induced part of Eq. (2), directly related to the pion-nucleon coupling, $g_{\pi NN}$, which should satisfy $\displaystyle m_{N}g_{A}=F_{\pi}g_{{\pi NN}};$ at finite momentum transfer, $2m_{N}g_{A}(q^{2})-q^{2}{g_{P}(q^{2})}=\frac{2g_{{\pi NN}}F_{\pi}m_{\pi}^{2}}{q^{2}+m_{\pi}^{2}}$ (3) with a more traditional convention of $F_{\pi}=f_{\pi}/\sqrt{2}\sim 92$ MeV. These should hold up to residual dependence on the momentum transfer $q^{2}$; the pion-nucleon coupling at high momentum transfer may significantly differ from its value near zero momentum transfer. On the lattice there may also arise some corrections from the finite lattice volume. Since we study the dependence of these form factors at relatively low momentum transfer, we can try to extract corresponding mean square radii, $\langle r^{2}\rangle_{{V,T,A}}$, defined by $\displaystyle g_{V,T,A}(q^{2})/g_{V,T,A}(0)\sim 1-\frac{1}{6}q^{2}\langle r^{2}\rangle_{{V,T,A}}+\cdots$. The structure functions are measured in deep inelastic scattering of leptons from nucleons Breidenbach et al. (1969); Friedman (1991); Kendall (1991); Taylor (1991); Gluck et al. (1996); Gehrmann and Stirling (1996); Lai et al. (1997); Adams et al. (1997); Adeva et al. (1998); Gluck et al. (1998); Ackerstaff et al. (1999); Martin et al. (2002), the cross section of which is factorized in terms of leptonic and hadronic tensors: $\displaystyle k_{0}^{\prime}\frac{d\sigma}{d^{3}k^{\prime}}=\frac{2M}{s-M^{2}}\frac{\alpha^{2}}{(q^{2})^{2}}l^{\mu\nu}W_{\mu\nu}.$ (4) The leptonic tensor is known to be $l_{\mu\nu}(k,k^{\prime})=2\left(k_{\mu}k^{\prime}_{\nu}+k^{\prime}_{\mu}k_{\nu}-\frac{1}{2}Q^{2}g_{\mu\nu}\right).$ (5) Hence, the cross section provides us with structure information about the target nucleon through the hadronic tensor, $W_{\mu\nu}$, which is decomposed into symmetric unpolarized and antisymmetric polarized parts: $W^{\\{\mu\nu\\}}(x,Q^{2})=\left(-g^{\mu\nu}+\frac{q^{\mu}q^{\nu}}{q^{2}}\right){F_{1}(x,Q^{2})}+\left(P^{\mu}-\frac{\nu}{q^{2}}q^{\mu}\right)\left(P^{\nu}-\frac{\nu}{q^{2}}q^{\nu}\right)\frac{F_{2}(x,Q^{2})}{\nu}$ (6) $W^{[\mu\nu]}(x,Q^{2})=i\epsilon^{\mu\nu\rho\sigma}q_{\rho}\left(\frac{S_{\sigma}}{\nu}({g_{1}(x,Q^{2})}\right.+\left.{g_{2}(x,Q^{2})})-\frac{q\cdot SP_{\sigma}}{\nu^{2}}{g_{2}(x,Q^{2})}\right).$ (7) $\nu=q\cdot P$, $S^{2}=-M^{2}$, and $x=Q^{2}/2\nu$, and $Q^{2}=|q^{2}|$. The unpolarized structure functions are $F_{1}(x,Q^{2})$ and $F_{2}(x,Q^{2})$, and the polarized, $g_{1}(x,Q^{2})$ and $g_{2}(x,Q^{2})$. Their moments are described in terms of Wilson’s operator product expansion: $\displaystyle 2\int_{0}^{1}dx\,x^{n-1}{F_{1}(x,Q^{2})}$ $\displaystyle=$ $\displaystyle\sum_{q=u,d}c^{(q)}_{1,n}(\mu^{2}/Q^{2},g(\mu))\>\langle x^{n}\rangle_{q}(\mu)+{O(1/Q^{2})},$ $\displaystyle\int_{0}^{1}dx\,x^{n-2}{F_{2}(x,Q^{2})}$ $\displaystyle=$ $\displaystyle\sum_{f=u,d}c^{(q)}_{2,n}(\mu^{2}/Q^{2},g(\mu))\>\langle x^{n}\rangle_{q}(\mu)+{O(1/Q^{2})},$ $\displaystyle 2\int_{0}^{1}dx\,x^{n}{g_{1}(x,Q^{2})}$ $\displaystyle=$ $\displaystyle\sum_{q=u,d}e^{(q)}_{1,n}(\mu^{2}/Q^{2},g(\mu))\>\langle x^{n}\rangle_{\Delta q}(\mu)+{O(1/Q^{2})},$ $\displaystyle 2\int_{0}^{1}dx\,x^{n}{g_{2}(x,Q^{2})}$ $\displaystyle=$ $\displaystyle\frac{1}{2}\frac{n}{n+1}\sum_{q=u,d}\left[e^{q}_{2,n}(\mu^{2}/Q^{2},g(\mu))\>d_{n}^{q}(\mu)-2e^{q}_{1,n}(\mu^{2}/Q^{2},g(\mu))\>\langle x^{n}\rangle_{\Delta q}(\mu)\right]+{O(1/Q^{2})},$ (8) where $c_{1}$, $c_{2}$, $e_{1}$, and $e_{2}$ are perturbatively known Wilson coefficients and ${\langle x^{n}\rangle_{q}(\mu)}$, ${\langle x^{n}\rangle_{\Delta q}(\mu)}$ and $d_{n}(\mu)$ are calculable on the lattice as forward nucleon matrix elements of certain local operators. Again, the conventional staggered or Wilson fermions would complicate such lattice calculations for the same reasons as discussed for the form factors. The DWF calculations are simpler because of easier renormalizations due to good chiral symmetry. In particular the first moments $\langle x\rangle_{u-d}$ (quark momentum fraction) and $\langle x\rangle_{\Delta u-\Delta d}$ (quark helicity fraction) share a common renormalization, and so their ratio is naturally renormalized in DWF calculations Orginos et al. (2006). During the last few years lattice computations have provided many interesting results for these structure function moments Gockeler et al. (1996, 1999a); Detmold et al. (2001); Dolgov et al. (2002); Gockeler et al. (2002); Orginos (2002); Hagler et al. (2003); Ohta and Orginos (2004, 2005); Sasaki et al. (2003); LHPC et al. (2004); Gockeler et al. (2004a); Khan et al. (2004); Gockeler et al. (2004b); Orginos et al. (2006); Edwards et al. (2006); Gockeler et al. (2007b); Lin and Ohta (2006); Orginos (2006); Schroers (2007); Hagler et al. (2007); Lin (2007a, b); Yamazaki and Ohta (2007); Hagler (2007), in both quenched and full QCD. These calculations provide first-principles values for the moments of structure functions at leading twist. One of the major unresolved issues in these previous calculations is significant overestimation of the moments $\langle x\rangle_{u-d}$ and $\langle x\rangle_{\Delta u-\Delta d}$ compared with results from fits to the experimental data Gockeler et al. (1996); Dolgov et al. (2002); Gockeler et al. (2004b); Orginos et al. (2006). In our quenched calculation Orginos et al. (2006) of these quantities we encountered similar overestimations. Interestingly, however, the ratio of these quantities, $\langle x\rangle_{u-d}/\langle x\rangle_{\Delta u-\Delta d}$ showed very weak dependence on the quark mass and agreed well with the experimental ratio in the chiral limit. Again an obvious question is whether this behavior survives when the quarks are treated dynamically. We address these questions with a lattice QCD calculation using two degenerate dynamical flavors of domain-wall fermions and DBW2 rectangle-improved gauge action Aoki et al. (2005). Three bare sea quark masses of 0.02, 0.03 and 0.04 in lattice units (corresponding to pion masses of about 0.5, 0.6 and 0.7 GeV) are used with about 200 gauge configurations each. The lattice cutoff is $a^{-1}\sim 1.7\mbox{ GeV}$, and the spatial volume is about $(1.9\mbox{ fm})^{3}$. The lattice cutoff is sufficiently high to allow us to take advantage of various benefits of DWF such as good chiral and flavor symmetries in performing fully-nonperturbative renormalization for our nucleon observables. Unlike our past quenched calculations Sasaki et al. (2003); Orginos et al. (2006) which did not show significant dependence on quark mass in most of the observables, we sometimes observe significant deviations at the lightest quark mass from the heavier quark mass results. The pion mass times lattice extent, $m_{\pi}L$, is slightly less than 5 at the lightest quark mass, which may be problematic for calculations involving a large hadron such as the nucleon. We must therefore perform quark-mass extrapolation on a case- by-case basis and with great caution. The rest of the paper is organized as follows. In Section II, we first briefly summarize the numerical ensembles used in the calculation. Then we discuss the choice of nucleon source and sink and the operators used to calculate the form factors and moments of the structure functions. Finally we briefly summarize the numerical nonperturbative renormalization methods. In Section III we begin with discussion of the form factors at zero momentum transfer. We show the conservation of the isovector vector current is under good control and then give an evaluation of the nucleon axial charge. Then we discuss momentum dependence of various form factors. In particular the Dirac mean-squared charge radius is extracted. The Goldberger-Treiman relation is shown to hold and provides an estimate of the pion-nucleon coupling. Finally, we describe our calculations of some low moments of the structure functions. We summarize our conclusions and future plans in Section IV. ## II Numerical Method We use DWF lattice ensembles generated by the RBC Collaboration Aoki et al. (2005). These were generated with two degenerate flavors of dynamical quarks described by the domain-wall fermion Kaplan (1992, 1993); Shamir (1993); Furman and Shamir (1995) action and gluons described by the doubly-blocked Wilson (DBW2) gauge action Takaishi (1996); de Forcrand et al. (2000). There are three such ensembles with sea quark masses of $m_{\rm sea}=0.02$, 0.03, and 0.04 in lattice units, which respectively correspond to about 1/2, 3/4 and 1 times the strange quark mass, or degenerate pseudoscalar meson masses of about 500, 600 and 700 MeV. The inverse lattice spacing is about 1.7 GeV, set from the $\rho$-meson mass, yielding a physical volume of $(1.9\mbox{ fm})^{3}$. Table 1: Summary of simulation parameters and the numbers of configurations. The pion mass, the nucleon mass and the pion decay constant ($F_{\pi}$) are also tabulated. $m_{f}$ | 0.02 | 0.03 | 0.04 ---|---|---|--- $t_{\rm snk}-t_{\rm src}$ | 12 | 10 | 10 | 10 $t_{\rm src}$ | $0,9,18$ | $0,15$ | $0,15$ | $0,15$ $r_{\rm gauss}$ | 8 | 4.35 | 4.35 | 4.35 # of conf. | 185 each | 220 each | 220 each | 220 each $m_{\pi}$ (GeV)Aoki et al. (2005) | 0.493(5) | 0.607(4) | 0.695(4) $m_{N}$ (GeV)Aoki et al. (2005) | 1.28(2) | 1.43(2) | 1.55(2) $F_{\pi}$ (GeV)Aoki et al. (2005) | 0.1141(7) | 0.1232(7) | 0.1329(7) We study nucleon matrix elements using quark propagators obtained with Gaussian gauge-invariant sources with radius ($r_{\rm gauss}$) of 4.35 for all three quark masses and additionally with Gaussian smearing radius of 8 for the lightest quark mass. The former parameter was previously used for the neutron dipole moments calculation on the $m_{\rm sea}=0.03$ and 0.04 ensembles Berruto et al. (2006). The latter is a better size for overlap with the ground state of the nucleon. For details of our source definition see Ref. Berruto et al. (2006). We use only unitary valence quark mass values, $m_{\rm valence}=m_{\rm sea}=m_{f}$, when computing quark propagators. We use two different sequential sources generated with source-sink locations, $[t_{\rm src},t_{\rm sink}]=[0,10]$ and [15, 25], where the source-sink separation $t_{\rm sep}$ is fixed as 10 lattice units, on a given gauge configuration for all three quark masses with the smaller $r_{\rm gauss}$. As for the larger $r_{\rm gauss}$, we use a different choice of source-sink separation, $t_{\rm sep}=12$. The longer separation causes the larger statistical noise. Therefore, we use three sources placed at $[t_{\rm src},t_{\rm sink}]=[0,12]$, [9, 21] and [18, 30] to increase the statistics. Later, we will discuss the possibility of excited-state contaminations in our calculations by comparing results obtained from two sets of parameters, $\\{t_{\rm sep},r_{\rm gauss}\\}$, on the lightest quark ensemble. The parameter values and the number of configurations used from each ensemble are summarized in Table 1. We also compile some basic physics results from Ref. Aoki et al. (2005) in the same table. We define our nucleon two-point functions with a nucleon interpolation field $\chi$ and smearing parameters $A$ and $B$ as follows: $\langle\chi_{A}(t,{\vec{p}})\overline{\chi}_{B}(0,-{\vec{p}})\rangle=\sum_{s}\langle 0|\chi_{A}|p,s\rangle\langle p,s|\overline{\chi}_{B}|0\rangle e^{-E(p)t}+\cdot\cdot\cdot=\frac{E({\vec{p}})\gamma^{t}-i\vec{\gamma}\cdot\vec{p}+m_{N}}{2E}\sqrt{z_{A}(p)z_{B}(p)}\;e^{-E({\vec{p}})t}+\cdot\cdot\cdot$ (9) with normalized states defined as $\displaystyle\langle 0|\chi_{A}|p,s\rangle$ $\displaystyle=$ $\displaystyle\sqrt{z_{A}({\vec{p}})}u_{s}(\vec{p}),$ (10) and spinors satisfying $\displaystyle\sum_{s}u_{s}(\vec{p})\bar{u}_{s}(\vec{p})$ $\displaystyle=$ $\displaystyle E(\vec{p})\gamma^{t}-i\vec{\gamma}\cdot\vec{p}+m_{N},$ (11) where $E({\vec{p}})=\sqrt{m_{N}^{2}+{\vec{p}}^{2}}$. The ellipsis in Eq. (9) denotes excited-state contributions, which can be ignored in the case of $t\gg 1$. We often use the projector $\displaystyle{\cal P}_{+}=\frac{1+\gamma_{t}}{2}$ to the positive-parity states, $\Gamma^{(2)}_{AB}(t_{\rm snk},t_{\rm src};\vec{p})=\frac{1}{4}{\rm Tr}({\cal P}_{+}\langle\chi_{A}(t_{\rm snk},{\vec{p}})\overline{\chi}_{B}(t_{\rm src},-{\vec{p}})\rangle)=\frac{E({\vec{p}})+m_{N}}{2E}\sqrt{z_{A}({\vec{p}})z_{B}({\vec{p}})}\;e^{-E({\vec{p}})(t_{\rm snk}-t_{\rm src})}+\cdot\cdot\cdot.$ (12) These two-point functions provide appropriate normalization factors when we extract matrix elements from three-point correlation functions, as well as estimates for the nucleon mass. Now we define the three-point functions: $\displaystyle G_{\mu,{AB}}(t_{\rm src},t,t_{\rm snk})$ $\displaystyle=$ $\displaystyle\langle\chi_{A}(t_{\rm snk},\vec{p}_{\rm snk})\,J_{\mu}(t,\vec{q})\,\overline{\chi}_{B}(t_{\rm src},\vec{p}_{\rm src})\rangle$ (13) $\displaystyle=$ $\displaystyle\frac{E^{\prime}\gamma^{t}+i\vec{\gamma}\cdot\vec{p}^{\prime}+m_{N}}{2E^{\prime}}\cdot\Sigma_{\mu}\cdot\frac{E\gamma^{t}-i\vec{\gamma}\cdot\vec{p}+m_{N}}{2E}\,\sqrt{z_{A}({\vec{p}}^{\prime})z_{B}({\vec{p}})}\;e^{-E^{\prime}(t_{\rm snk}-t)}e^{-E(t-t_{\rm src})}+\cdots,$ where ${\vec{q}}={\vec{p}}^{\prime}-{\vec{p}}$, ${\vec{p}}={\vec{p}}_{\rm src}$, ${\vec{p}}^{\prime}={\vec{p}}_{\rm snk}$, $E=E({\vec{p}}_{\rm src})$ and $E^{\prime}=E(\vec{p}_{\rm snk})$. The operator $\Sigma_{\mu}$ is appropriately selected for each observable of interest discussed in the following subsections. We calculate the nucleon isovector form factors and some low moments of structure functions, namely the isovector vector ($g_{V}$), induced tensor ($g_{T}$), isovector axial ($g_{A}$), and induced pseudoscalar ($g_{P}$), form factors at both zero and finite momentum transfer, and structure function moments corresponding to the momentum fraction $\langle x\rangle_{u-d}$, helicity fraction $\langle x\rangle_{\Delta u-\Delta d}$, a twist-3 moment $d_{1}$, and transversity $\langle 1\rangle_{\delta u-\delta d}$. The respective choice of the lattice operators for these observables are the same as in earlier RBC reports Sasaki et al. (2002, 2003); Orginos et al. (2006); Sasaki and Yamazaki (2007) on quenched calculations, and are briefly summarized in what follows. Before we move on to discuss the operator choices for our observables of interest, we would like to briefly explain our different choice of source-sink separation, $t_{\rm sep}$, and Gaussian smearing radius, $r_{\rm gauss}$, for the lightest quark mass ensemble (see Lin and Ohta (2006); Ohta and Orginos (2005); Orginos (2003) for the earlier calculations). We started our calculation by setting $t_{\rm sep}=10$ time units and the Gaussian source smearing radius to 4.35, parameters previously used for the neutron dipole moments calculation on the $m_{f}=0.03$ and 0.04 ensemblesBerruto et al. (2006). Although the new results of the Gaussian source in the heaver quark- mass region agree with the heavier quark-mass points of the box source, for the same calculation with $m_{f}=0.02$, we find a significant decrease in the axial charge compared to our previous box-source calculation, which used a separation of 12 time units as shown in Fig. 1. This discrepancy prompted us to examine whether there is a systematic error due to excited-state contamination in the new results. The effective masses plotted in Fig. 2 show that the Gaussian source with smearing radius 4.35 has a ground state plateau that begins at $t-t_{\rm src}=7$ or 8 (top figure), while the box source appears to plateau sooner (bottom figure). Thus one may suspect that excited-state contamination accounts for the effect observed in Fig. 1. Since the major difference between the two calculations is the source type and separation, we extend the source- sink separation to $t_{\rm sep}=12$ for the Gaussian source (the same as in the previous box-source calculation) and also increase its radius to 8 in order to reduce excited-state contamination. Indeed, the Gaussian source with radius 8 provides similar quality of plateau in the effective mass plot, as shown in the middle panel of Fig. 2. However, as shown in Fig. 1, the resulting $g_{A}/g_{V}$ with the second parameter set $\\{t_{\rm sep},r_{\rm gauss}\\}=\\{12,8\\}$ is consistent with that of the first one $\\{10,4.35\\}$ rather than the previous result of the box source. So, the discrepancy between the results is probably not caused by excited-state contamination. As we mentioned before, to reduce statistical fluctuations, we use multiple sources in the present studies. This is another difference from the previous calculation, where only a single source was utilized. To make this point clear, we show the dependence of bare $g_{V}$ and $g_{A}$ on the location of the current insertion $t$ for each choice of the source location in Fig. 3. In the top panels, all calculations of the bare $g_{V}$ show clear plateaus between source and sink locations. Although the slight dependence on the source location may be observed among the three sources in the $\\{t_{\rm sep},r_{\rm gauss}\\}=\\{12,8\\}$ case, all the values of these plateaus agree with one another within the statistical errors. We see no evidence of excited- state contamination in the bare $g_{V}$. On the other hand, for the bare $g_{A}$ (bottom panels), both the results obtained from box and Gaussian sources with $t_{\rm sep}=12$ exhibit larger fluctuations and less clear plateaus, while two consistent plateaus clearly appear in the cases of $t_{\rm sep}=10$. This indicates that the larger separation causes the larger fluctuations, since the absolute values of three- point functions are exponentially suppressed as a function of $t_{\rm sep}$. Indeed, the $t_{\rm sep}=12$ Gaussian source results obtained from a single source-sink location $[t_{\rm src},t_{\rm sink}]=[9,21]$ agree with those of the box source due to the large statistical errors, while the average from all three sources for $\\{t_{\rm sep},r_{\rm gauss}\\}=\\{12,8\\}$ provides good agreement with the results of $\\{t_{\rm sep},r_{\rm gauss}\\}=\\{10,4.35\\}$. In this context, the lower value of the bare $g_{A}$ obtained from the present studies is more statistically significant, since the box source results have been obtained only for a single source. Although a systematic error from excited-state contaminations in the result of $\\{t_{\rm sep},r_{\rm gauss}\\}=\\{10,4.35\\}$ might be hidden, owing to the large fluctuations in that of $\\{t_{\rm sep},r_{\rm gauss}\\}=\\{12,8\\}$, we may choose to live with the former for the final result. This is mainly because the finite momentum calculation suffers much from such large fluctuations. We obtain results for the form factors only from the case of $t_{\rm sep}=10$. Figure 1: Renormalized isovector axial charge $g_{A}/g_{V}$ calculated with box Ohta and Orginos (2005) and Gaussian smeared sources with various source-sink separations ($t_{\rm sep}$). The value from the Gaussian source with $t_{\rm sep}=10$ and smearing radius 4.35 at the lightest quark mass of $m_{f}=0.02$ deviates significantly from the previous box-source calculation. In that case the box size is chosen for better overlap with the ground state of the nucleon and the longer source-sink separation ($t_{\rm sep}=12$) is adopted to avoid possible excited-state contaminations. This discrepancy is not resolved by using a Gaussian source of an radius 8 and the same source-sink separation. As discussed in the text, this is likely caused by the larger statistical fluctuation in the longer source-sink separation. | | ---|---|--- Figure 2: Nucleon effective mass plots from two-point functions. Gaussian source with smearing radius, $r_{\rm gauss}=4.35$ (left), 8 (center), and box source (right) from the calculation in Ref. Ohta and Orginos (2005). | | ---|---|--- | | Figure 3: Comparison of the bare isovector vector charge $g_{V}$ (top) and axial charge $g_{A}$ (bottom) from source-sink separation $t_{\rm sep}=10$ Gaussian source (left panels), $t_{\rm sep}=12$ Gaussian source (middle panels) and $t_{\rm sep}=12$ box source (right panels). ### II.1 Form factors For numerical convenience, we redefine the form factors as dimensionless quantities. Thus, the isovector vector current in Eq. (1) is rewritten as $\langle p|V^{+}_{\mu}(0)|n\rangle=\bar{u}_{p}\left[\gamma_{\mu}G_{V}(q^{2})-q_{\lambda}\sigma_{\lambda\mu}\frac{G_{T}(q^{2})}{2m_{N}}\right]u_{n}$ (14) in terms of the dimensionless vector $G_{V}$ and induced-tensor $G_{T}$ form factors. Here the $\Sigma_{\mu}$ in Eq. (13) is $\gamma_{\mu}G_{V}(q^{2})-q_{\lambda}\sigma_{\lambda\mu}\frac{G_{T}(q^{2})}{2m_{N}}$. Likewise the isovector axial current in Eq. (1) is rewritten as $\langle p|A^{+}_{\mu}(0)|n\rangle=\bar{u}_{p}\left[\gamma_{\mu}\gamma_{5}G_{A}(q^{2})+iq_{\mu}\gamma_{5}\frac{G_{P}(q^{2})}{2m_{N}}\right]u_{n}$ (15) with the dimensionless axial $G_{A}$ and induced-pseudoscalar $G_{P}$ form factors. Note that the latter is normalized with twice the nucleon mass $2m_{N}$ unlike in some muon-capture literature where it is normalized with the muon mass $m_{\mu}$. This is for numerical convenience in this paper, where the nucleon mass calculated on the lattice is heavier than its physical value. Here the $\Sigma_{\mu}$ in Eq. (13) is $\gamma_{\mu}\gamma_{5}G_{A}(q^{2})+iq_{\mu}\gamma_{5}\frac{G_{P}(q^{2})}{2m_{N}}$. (In the above, $g_{V,A}(q^{2})\equiv G_{V,A}(q^{2})$.) The right-hand sides of the above two equations have the most general form consistent with Lorentz covariance. The momentum transfer $q=p_{n}-p_{p}$ becomes very small in the forward limit, because of the small mass difference between the neutron and proton. We use two projection operators to help us extract the momentum dependence of the form factor: ${\Gamma^{(3),{\cal P}}_{\mu,{AB}}(t_{\rm snk},t;p_{\rm src},p_{\rm snk})}=\frac{1}{4}{\rm Tr}({\cal P}G_{\mu,{AB}}(t_{\rm snk},t;p_{\rm src},p_{\rm snk})),$ (16) where we choose ${\cal P}={\cal P}_{A_{z}}={\cal P}_{+}\gamma_{5}\gamma_{z}$ for both axial and vector currents and ${\cal P}={\cal P}_{+}$ for vector current. We combine three- and two-point functions to remove time dependence and redundant source or sink normalization ($z$) factors Hagler et al. (2003): $R_{j_{\mu}}=\frac{\Gamma^{(3),{\cal P}}_{\mu,GG}(t_{\rm src},t,t_{\rm snk},\vec{p}_{\rm src},\vec{p}_{\rm snk})}{\Gamma^{(2)}_{GG}(t_{\rm src},t_{\rm snk},\vec{p}_{\rm snk})}\times\left(\frac{\Gamma^{(2)}_{LG}(t,t_{\rm snk},\vec{p}_{\rm src})\Gamma^{(2)}_{GG}(t_{\rm src},t,\vec{p}_{\rm snk})\Gamma^{(2)}_{LG}(t_{\rm src},t_{\rm snk},\vec{p}_{\rm snk})}{\Gamma^{(2)}_{LG}(t,t_{\rm snk},\vec{p}_{\rm snk})\Gamma^{(2)}_{GG}(t_{\rm src},t,\vec{p}_{\rm src})\Gamma^{(2)}_{LG}(t_{\rm src},t_{\rm snk},\vec{p}_{\rm src})}\right)^{\frac{1}{2}}$ (17) where $L$ denotes a local (point) source or sink and $G$ stands for a Gaussian-smeared one. In this work, we fix the sink momentum to zero. Therefore, in the axial-current case, Eq. (17) gives $\displaystyle{\cal R}_{\rm A_{i}}$ $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{2E(\vec{p})(E(\vec{p})+m_{N})}}$ (18) $\displaystyle\times$ $\displaystyle\left[\delta_{iz}(E(\vec{p})+m_{N})G_{A}-\frac{p_{z}p_{i}}{2m_{N}}G_{P}\right]$ with $i\in\\{x,y,z\\}$ for corresponding insertion of operator ${A_{i}}$. In the vector current case, we use the projection operator $P_{+}$ on $V_{4}$ and $P_{A_{z}}$ on $V_{1}$ and $V_{2}$ to extract $G_{V}$ and induced $G_{T}$; the necessary equations are respectively $\displaystyle\sqrt{\frac{E(\vec{p})+m_{N}}{2E(\vec{p})}}\left(G_{V}-\frac{E(\vec{p})-m_{N}}{2m_{N}}G_{T}\right),$ (19) $\displaystyle\frac{-ip_{y}}{\sqrt{2E(\vec{p})(E(\vec{p})+m_{N})}}(G_{V}+G_{T}),$ (20) $\displaystyle\frac{+ip_{x}}{\sqrt{2E(\vec{p})(E(\vec{p})+m_{N})}}(G_{V}+G_{T}).$ (21) One can in principle solve for $G_{A}(q^{2})$, $G_{P}(q^{2})$, $G_{V}(q^{2})$, and $G_{T}(q^{2})$ with overconstrained data for each $q^{2}$. ### II.2 Moments of structure functions Of the moments of the structure functions summarized in Eq. (I), we calculate those which do not require finite momentum transfer: the quark momentum fraction $\langle x\rangle_{q}$, helicity fraction $\langle x\rangle_{\Delta q}$, transversity $\langle 1\rangle_{\delta q}$ and the twist-3 moment $d_{1}$. Of these we calculate the isovector contribution for the former three: $q=u-d$. This simplifies their renormalization as the flavor-singlet contribution drops out. Thus we fully nonperturbatively renormalize these three so the results are compared with the experiments under an assumption of good isospin symmetry. On the other hand we do not renormalize the twist-3 moment: here our interest is to look at its individual up and down components to see if the Wandzura-Wilczek relation Wandzura and Wilczek (1977) holds. Our choice of operators ($\Sigma_{\mu}$ in Eq. (13)) for the three-point functions follow an earlier RBC paper Orginos et al. (2006) and are summarized in Table 2. Table 2: Operators used in our structure-function moments calculations, including the notation for the operator, the explicit operator form, the hypercubic group representation, the correlator ratios and the projection operators used in the numerical nonperturbative renormalization in Eq. (26). quark momentum fraction $\langle x\rangle_{q}$ --- $\Sigma_{\mu}$ | $\displaystyle{\cal O}^{q}_{44}=\overline{q}\left[\gamma_{4}\stackrel{{\scriptstyle\leftrightarrow}}{{D_{4}}}-\frac{1}{3}\sum_{k}\gamma_{k}\stackrel{{\scriptstyle\leftrightarrow}}{{D_{k}}}\right]q$ hypercubic group rep. | ${\bf 3}^{+}_{1}$ correlator ratio | $\displaystyle R_{\langle x\rangle_{q}}=\frac{C^{\Gamma,{\cal O}^{q}_{44}}_{\rm 3pt}}{C_{\rm 2pt}}=m_{N}\langle x\rangle_{q}$ NPR projection | ${{\cal P}^{q}_{44}}^{-1}=\gamma_{4}p_{4}-\frac{1}{3}\sum_{i=1,3}\gamma_{i}p_{i}$ quark helicity fraction $\langle x\rangle_{\Delta q}$ $\Sigma_{\mu}$ | $\displaystyle{\cal O}^{5q}_{\\{34\\}}=i\overline{q}\gamma_{5}\left[\gamma_{3}\stackrel{{\scriptstyle\leftrightarrow}}{{D_{4}}}+\gamma_{4}\stackrel{{\scriptstyle\leftrightarrow}}{{D_{3}}}\right]q$ hypercubic group rep. | ${\bf 6}^{-}_{3}$ correlator ratio | $\displaystyle R_{\langle x\rangle_{\Delta q}}=\frac{C^{\Gamma,{\cal O}^{5q}_{\\{34\\}}}_{\rm 3pt}}{C_{\rm 2pt}}=m_{N}\langle x\rangle_{\Delta q}$ NPR projection | ${{\cal P}^{5q}_{34}}^{-1}=i\gamma_{5}\left(\gamma_{3}p_{4}+\gamma_{4}p_{3}\right)$ transversity $\langle 1\rangle_{\delta q}$ $\Sigma_{\mu}$ | $\displaystyle{\cal O}^{\sigma q}_{34}=\overline{q}\gamma_{5}\sigma_{34}q$ hypercubic group rep. | ${\bf 6}^{+}_{1}$ correlator ratio | $\displaystyle R_{\langle 1\rangle_{\delta q}}=\frac{C^{\Gamma,{\cal O}^{\sigma q}_{\\{34\\}}}_{\rm 3pt}}{C_{\rm 2pt}}=\langle 1\rangle_{\delta q}$ NPR projection | ${{\cal P}^{\sigma q}_{34}}^{-1}=\gamma_{5}\sigma_{34}$ twist-3 matrix element $d_{1}$ $\Sigma_{\mu}$ | $\displaystyle{\cal O}^{5q}_{[34]}=i\overline{q}\gamma_{5}\left[\gamma_{3}\stackrel{{\scriptstyle\leftrightarrow}}{{D_{4}}}-\gamma_{4}\stackrel{{\scriptstyle\leftrightarrow}}{{D_{3}}}\right]q$ hypercubic group rep. | ${\bf 6}^{+}_{1}$ correlator ratio | $\displaystyle R_{d_{1}}=\frac{C^{\Gamma,{\cal O}^{5q}_{[34]}}_{\rm 3pt}}{C_{\rm 2pt}}=d_{1}$ NPR projection | ${{\cal P}^{5q}_{[34]}}^{-1}=i\gamma_{5}\left(\gamma_{3}p_{4}-\gamma_{4}p_{3}\right)$ ### II.3 Nonperturbative renormalization In order to compare our calculation with the experimental values, we need to establish the proper connection to the continuum through renormalization. Fortunately, the well-preserved chiral and flavor symmetries of the domain- wall fermions (DWF) make this task much easier than in the cases of more conventional fermions Blum et al. (2003, 2002); Dawson (2003); Aoki et al. (2004). For the form factors, the chiral symmetry of DWF assures the isovector vector and axial local currents which are used in the present calculation share a common renormalization: $Z_{V}=Z_{A}$, up to higher order discretization errors, $O(a^{2})$. Further, since the vector current is conserved, the vector renormalization is calculated from the vector charge, $g_{V}$, as $Z_{V}=1/g_{V}$. These will be demonstrated in the numerical results section. Note the induced tensor and pseudoscalar form factors share this common renormalization as well and do not require any additional calculation for renormalization. For structure function moments, we follow Ref. Martinelli et al. (1995) and implement the regularization-independent momentum-subtraction (RI/MOM) scheme to nonperturbatively renormalize the operators. This was first applied to structure functions by Gimenez et al. Gimenez et al. (1998). The procedure may be described as follows. First, the Fourier transform of the Green function with our operator of interest is constructed: $\displaystyle G_{O_{\Gamma}}(p;a)=\sum_{x,y}e^{-ip\cdot(x-y)}\langle\psi(x){O_{\Gamma}}(0)\overline{\psi}(y)\rangle,$ (22) where $O_{\Gamma}$ is one of ${\cal O}^{q}_{44}$, ${\cal O}^{\sigma q}_{34}$, ${\cal O}^{5q}_{\\{34\\}}$ or ${\cal O}^{5q}_{[34]}$ for the first-moment operator renormalization. In the lattice calculation, we need to produce Fourier-transformed point-source $S(pa;0)$ and point-split–source $D_{\mu}S(pa;0)$ propagators, $\displaystyle S(pa;0)$ $\displaystyle=$ $\displaystyle\sum_{x}e^{-ip\cdot x}S(x;0)$ (23) $\displaystyle D_{\mu}S(pa;0)$ $\displaystyle=$ $\displaystyle\sum_{x}\frac{1}{2}e^{-ip\cdot x}\left[S(x;-\hat{\mu})U_{\mu}(-\hat{\mu})\right.$ (24) $\displaystyle-$ $\displaystyle\left.S(x;\hat{\mu})U_{\mu}^{\dagger}(0)\right],$ to construct the Green function in Eq. (22). In this work, we use momenta $(p_{x},p_{y},p_{z},p_{t})$ ranging from $(0,0,0,0)$ to $(2,2,3,5)$ in units of $2\pi/L$. The next step is to truncate the external legs from the Green function: $\displaystyle\Lambda_{O_{\Gamma}}(p;a)=S(p;a)^{-1}G_{O_{\Gamma}}(p;p^{\prime};a)S(p^{\prime};a)^{-1},$ (25) where $Z^{\rm RI}$ is obtained after the projection $\displaystyle Z_{O_{\Gamma}}(\mu;a)^{-1}Z_{q}(\mu;a)=\frac{1}{12}{{\rm Tr}\left(\Lambda_{O_{\Gamma}}(p;a)P_{\Gamma}\right)}|_{p^{2}=\mu^{2}}.$ (26) The projection operators for various structure function operators are listed in Table 2. Note that $\mu$ must fall inside the renormalization window $\Lambda_{\rm QCD}\ll\mu\ll 1/a$. The following steps allow us to convert the renormalization constants into the continuum $\overline{\rm MS}$ scheme at 2 GeV. See Appendix .1 for specific details on the strong-coupling scheme-matching coefficients, RGI running equation and anomalous dimensions for various operators. 1. 1. Obtain $Z^{\rm RI}(\mu)$: The ratio of $Z_{O_{\Gamma}}(\mu;a)/Z_{q}(\mu;a)$ to $Z_{A}/Z_{q}(\mu;a)$ is computed and yields $Z_{O_{\Gamma}}(\mu;a)/Z_{A}$. Each of the factors in the ratio is first exptrapolated to the chiral limit, $m_{f}=-m_{\rm res}$, at fixed momentum. Using $Z_{V}=1/g_{V}$, we can determine $Z_{O_{\Gamma}}(\mu;a)$ in Eq. (26), the renormalization constant in the RI scheme, which we denote as $Z^{\rm RI}$. (See the lightly-filled circles in Fig. 4.) Figure 4: Renormalization constants in the chiral limit. The lightly-filled circles are the renormalization constants in RI-MOM scheme, and the diamonds are $\overline{\rm MS}$-scheme at 2 GeV. The fits (solid lines) are used to remove residual $(ap)^{2}$ artifacts. 2. 2. Convert to $\overline{\rm MS}$ scheme: We are interested in continuum quantities, mostly calculated in the $\overline{\rm MS}$ scheme. The conversion factors between RI and $\overline{\rm MS}$ schemes for the operators discussed here have been calculated in Refs. Gockeler et al. (1999b); Floratos et al. (1977). To get $Z^{\rm\overline{MS}}(\mu)$, we use the three-loop $\alpha_{s}(\mu)$ running coupling constants defined in Ref. Blum et al. (2002) with $\Lambda(n_{f}=2)=300$ MeVIzubuchi (2004). 3. 3. Remove $(ap)^{2}$ lattice artifacts: To remove ${\cal O}((ap)^{2})$ errors that might mimic continuum scale dependence, we first divide out the continuum running factor. The resulting renormalization factors should be scale-independent, so any remaining $(ap)^{2}$ dependence must be a lattice artifact. Hence, $Z^{\overline{\rm MS},{\rm RGI}}$ can be obtained from a fit to the form $f=A(pa)^{2}+B$. (See Fig. 4.) 4. 4. Running to 2 GeV: Finally, we use the RGI formula to obtain $Z^{\rm\overline{MS}}(2\mbox{ GeV})$. Table 3 summarizes the renormalization factors in $\overline{\rm MS}$ scheme at 2 GeV for each operator. Table 3: Summary of renormalization factors in $\overline{\rm MS}$ scheme at 2 GeV in the chiral limit. The number of configurations is about 100. $m_{f}$ | ${\cal O}^{q}_{44}$ | ${\cal O}^{5q}_{34}$ | ${\cal O}^{\sigma q}_{34}$ ---|---|---|--- $-m_{\rm res}$ | 1.28(7) | 1.23(7) | 0.776(14) These renormalization constants will be applied in Section III.3. ## III Numerical Results ### III.1 Vector and axial charge The isovector vector and axial charges, $g_{V}$ and $g_{A}$, are defined as the zero-momentum-transfer limits of the corresponding form factors, $G_{V}(q^{2})$ and $G_{A}(q^{2})$. Because of chiral symmetry, a Takahashi- Ward identity ensures that the two currents, which are related by chiral transformation, share a common renormalization: $Z_{A}=Z_{V}$ up to a lattice discretization error of $O(a^{2},m_{f}^{2}\,a^{2})$. Since the vector current is conserved, its renormalization is easily obtained as the inverse of the vector charge $g_{V}=G_{V}(q^{2}=0)$. Thus, by calculating the ratio of the three-point functions for $g_{A}/g_{V}$, we get the renormalized axial charge, $(g_{A})^{\rm ren}$Sasaki et al. (2003). Let us first discuss the vector charge. We note, however, that since the vector charge is conserved, it can not provide a transition matrix element from one state to another, in particular from any excited state to the ground state. Thus the precocious plateau does not necessarily mean that excited state contamination is absent in the signal after one time unit. As mentioned earlier, to avoid excited-state contributions at the lightest quark mass, we have lengthened the source-sink separation to 12 time units and used a larger smearing radius for the Gaussian-smeared source and sink. From the fitting ranges of 3–9 for $t_{\rm sep}=12$ and 3–7 for $t_{\rm sep}=10$ (see also Fig. 5), Figure 5: Bare vector charge as a function of the current insertion time. Results are averaged over all sources. The top figure depicts $t_{\rm sep}=12$ calculation at the lightest quark mass, $m_{f}=0.02$. The other three figures are obtained from $t_{\rm sep}=10$ calculations at all three quark masses (in order of increasing mass from top to bottom). The fit ranges are shown by horizontal lines. we obtain the plot in Fig. 6. Figure 6: The isovector vector-charge $(g_{V})^{\rm bare}$ as a function of the pion mass squared. This figure shows that larger fluctuations appear in the case of the longer time separation. By linearly extrapolating three points of $t_{\rm sep}=10$ to zero quark mass, we obtain an estimate of $g_{V}=1.308(9)$ and $Z_{V}=0.764(5)$. The resulting value agrees reasonably well with the value of $Z_{A}=0.75734(55)$ found in Ref. Aoki et al. (2005), which was determined from the ratio of two-point meson correlation functions. Thus we conclude that excited state contamination in our value of $g_{V}$ is negligible. In Fig. 7 Figure 7: The bare axial charge as in Fig. 5 we present the raw results for the bare axial charge. After taking an average of all sources, the a plateau emerges, even in the case of $t_{\rm sep}=12$ (top figure), albeit with larger statistical errors. As can be seen from this figure, the plateaus settle in after about three time slices for both $t_{\rm sep}$ cases at the lightest quark mass. This indicates that excited-state contamination is negligible compared to the statistical error. Corresponding renormalized values are obtained from the ratio of correlation functions, and are presented in Fig. 8. Figure 8: The axial-to-vector charge ratio as in Fig. 5 As expected, we also find that plateaus settle in after about three time slices for all cases. By using the fitting ranges shown in this figure, we obtain the values in Table 4 Table 4: Bare isovector vector and axial charges, $g_{V}^{\rm bare}$ and $g_{A}^{\rm bare}$ and their ratio $(g_{A}/g_{V})^{\rm bare}$ which is by-definition renormalized, extracted from data summarized in Figs. 5, 7 and 8. $m_{f}$ | $\\{t_{\rm sep},r_{\rm gauss}\\}$ | $g_{V}^{\rm bare}$ | $g_{A}^{\rm bare}$ | $(g_{A}/g_{V})^{\rm ren}$ ---|---|---|---|--- 0.02 | {12, 8} | 1.28(3) | 1.28(12) | 1.01(10) 0.02 | {10, 4.35} | 1.299(6) | 1.27(4) | 0.97(4) 0.03 | {10, 4.35} | 1.291(3) | 1.53(3) | 1.19(3) 0.04 | {10, 4.35} | 1.2865(13) | 1.51(2) | 1.171(16) and Fig. 9. Figure 9: The axial-to-vector charge ratio as a function of the pion mass squared. The lightest mass points with either $t_{\rm sep}=10$ or $t_{\rm sep}=12$ deviate significantly from the heavier mass points. Omitting the lightest mass point, a constrained linear fit is performed. We observe that our axial charge value at the lightest quark mass deviates significantly from the heavier quark mass points. This deviation causes the linear extrapolation to the physical point to deviate from experiment by more than five standard deviations. (The obtained value is given as $g_{A}/g_{V}=0.89(6)$ with a poor value of $\chi^{2}/{\rm dof}=11.3$) Omitting the lightest mass point, a constrained linear fit gives $g_{A}/g_{V}=1.23(12)$ at the physical point, in good agreement with experiment. As mentioned in Section II, our physical volume is about $(1.9\mbox{ fm})^{3}$. In light of our previous quenched calculation Sasaki et al. (2003) where we saw that a small volume resulted in a small axial charge, one may worry that a similar finite-volume effect occurs here. Since our quenched calculation Sasaki et al. (2003) was published, a way to interpret the finite-volume effect has been proposed by Beane and Savage Beane and Savage (2004) within the small scale expansion (SSE) scheme Hemmert et al. (1998), which is one possible extension of HBChPT with explicit $\Delta$ degrees of freedom. The SSE scheme can provide milder quark-mass dependence for the axial charge, which seems to be consistent with lattice results, while either leading order (LO) or next-to-leading order (NLO) HBChPT shows strong quark-mass dependence in the vicinity of the chiral limit Procura et al. (2007). There is a caveat that we have to arbitrarily fix at least one of the four parameters in SSE. However, if one considers only the finite-volume correction $\delta g_{A}$ (which is $g_{A}(L)-g_{A}(L=\infty)$) within this model, $\delta g_{A}$ depends only on two phenomenological parameters, the $N$-$\Delta$ coupling ($c_{A}$) and $\Delta$-$\Delta$ coupling ($g_{1}$) 111 The correspondence between the SSE notation and the parameters used in Ref. Beane and Savage (2004) can be found in Ref. Khan et al. (2006). Here, we prefer to use the original SSE notation. , which can be barely fixed. Furthermore, if the $N$-$\Delta$ coupling is set to zero, the Beane-Savage formula for the finite-volume correction to $g_{A}$ reduces to that of HBChPT at leading order. In the original paper Beane and Savage (2004), $\delta g_{A}$ is predicted to be positive. This remains true in the case of $c_{A}=0$, where SSE reduces to LO HBChPT. In this sense, this formula fails to account for the negative $\delta g_{A}$ observed in the quenched DWF calculation. In Ref. Khan et al. (2006), a finite-volume study of $g_{A}$ has been done with $n_{f}=2$ dynamical clover simulations. They also observe negative $\delta g_{A}$ in both quenched and dynamical simulations. Although their lattice data points are outside of the range of applicability of SSE (or LO HBChPT), they can fit their data using the SSE formula for $g_{A}$ with the finite-volume corrections, which should be negative. They insisted that the parameters adopted in the Beane-Savage paper do not fully reflect decoupling constraints, which are guaranteed to reduce SSE to LO HBChPT when $c_{A}=0$. Indeed, their adopted parameter set yields $\delta g_{A}<0$. Apart from the concept of SSE, this suggests that the sign of $\delta g_{A}$ predicted from SSE is very sensitive to adopted parameters. We find that the choice of $c_{A}$ strongly affects the sign of $\delta g_{A}$, as shown in Fig. 10. Figure 10: Finite volume correction of the axial charge $g_{A}$ is predicted within ${\cal O}(\varepsilon^{3})$ SSE. The sign of $\delta g_{A}$ is very sensitive to the choice of $c_{A}$. We fix the chiral limit value of the axial charge $g_{A}$, the pion decay constant $F_{\pi}$ and the $N$-$\Delta$ mass splitting $\Delta_{0}$ to their physical values and set $g_{1}=3.0$ as in Ref Beane and Savage (2004). We also set $L=1.9$ fm. We vary the remaining parameter of the $N$-$\Delta$ coupling $c_{A}$ from 0 to 1.5. Here, we recall that $c_{A}=1.5$ is adopted in Ref. Khan et al. (2006), while $c_{A}=1$ in Ref. Beane and Savage (2004). The predicted correction at our simulation point, less than 5%, is negligible. Furthermore, the sign of $\delta g_{A}$ seems to be positive, against our expectation. Therefore, the finite-volume correction from SSE can not account for our data either qualitatively or quantitatively. Again, it is likely that the heavy quark masses in our simulation do not allow use of such formulae. We will come back to this finite-volume question when we discuss the momentum dependence of the form factors and quark momentum and helicity fractions. ### III.2 Momentum dependence of form factors As described in the previous subsection, data sets with $\\{t_{\rm sep},r_{\rm gauss}\\}=\\{10,4.35\\}$ do not seem to suffer from excited-state contamination. Therefore, we focus on those data sets for analysis of the form factors at finite momentum transfer in this subsection. #### III.2.1 Vector current Under the present assumption of exact isospin symmetry that ignores strange quark contribution, a straight-forward exercise in SU(2) Lie algebra leads to the following well known relations between the weak vector-current matrix elements in neutron beta decay and the isovector part of the nucleon electromagnetic current matrix elements: $\langle p|\bar{u}\gamma_{\alpha}d|n\rangle=\langle p|\bar{u}\gamma_{\alpha}u-\bar{d}\gamma_{\alpha}d|p\rangle=\langle p|j_{\alpha}^{\rm em}|p\rangle-\langle n|j_{\alpha}^{\rm em}|n\rangle$ (27) where $j_{\alpha}^{\rm em}=\frac{2}{3}\bar{u}\gamma_{\alpha}u-\frac{1}{3}\bar{d}\gamma_{\alpha}d$. This relates the weak vector and induced tensor form factors with the isovector part of the nucleon electromagnetic form factors: $\displaystyle F_{1}^{p}(q^{2})-F_{1}^{n}(q^{2})$ $\displaystyle=$ $\displaystyle G_{V}(q^{2})$ (28) $\displaystyle F_{2}^{\it p}(q^{2})-F_{2}^{\it n}(q^{2})$ $\displaystyle=$ $\displaystyle G_{T}(q^{2}).$ (29) These are respectively are called the Dirac and Pauli form factors. They are related to the more conventional Sachs electric $G_{E}$ and magnetic $G_{M}$ form factors: $\displaystyle G_{E}^{N}(q^{2})$ $\displaystyle=$ $\displaystyle F_{1}^{N}(q^{2})-\frac{q^{2}}{4m^{2}_{N}}F_{2}^{N}(q^{2})$ (30) $\displaystyle G_{M}^{N}(q^{2})$ $\displaystyle=$ $\displaystyle F_{1}^{N}(q^{2})+F_{2}^{N}(q^{2})$ (31) where $N$ represents $p$ (proton) or $n$ (neutron). The isovector ones $G_{E,M}^{v}$ are defined by $G_{E,M}^{v}=G_{E,M}^{p}-G_{E,M}^{n}$. As such we can compare our lattice results computed at finite momentum transfer with the experiments. The easiest quantity for such a comparison is the charge radius, which is conventionally defined as $\langle r^{2}\rangle=-6{dF_{1}}/{dq^{2}}$. The proton mean-squared charge radius is known experimentally Yao et al. (2006) as $0.7656(119)\mbox{ fm}^{2}$ and the neutron, -0.1161(22), resulting in an estimate of $0.636(12)\mbox{ fm}^{2}$ for the mean-squared “Dirac” radius of the nucleon. (For details, see Appendix A of Ref Sasaki and Yamazaki (2007)). The raw results from the lattice are presented in Fig. 11. Figure 11: Bare Dirac isovector form factor as a function of the current insertion time at the heaviest pion mass ($m_{f}=0.04$). The horizontal lines represent the average value (solid lines) and their one standard deviations (dashed lines) in the fitted range. From these we extract the values of the Dirac isovector form factor, normalized by the value at $q^{2}=0$, as listed in Table 5. Table 5: Momentum transfer dependence of the isovector form factors. $n^{2}$ refers to the momenta carried by the vertex in units of $(2\pi/L)^{2}$. | $m_{f}$ | $n^{2}=0$ | $n^{2}=1$ | $n^{2}=2$ | $n^{2}=3$ | $n^{2}=4$ ---|---|---|---|---|---|--- | 0.02 | 1.299(6) | 0.86(3) | 0.68(3) | 0.63(7) | 0.51(17) $G_{V}$ | 0.03 | 1.291(3) | 0.882(18) | 0.73(3) | 0.62(4) | 0.48(5) | 0.04 | 1.2865(13) | 0.930(12) | 0.725(17) | 0.61(3) | 0.58(5) | 0.02 | N/A | 2.13(15) | 1.53(11) | 1.28(18) | 1.1(4) $G_{T}$ | 0.03 | N/A | 2.67(13) | 2.05(10) | 1.57(12) | 1.11(13) | 0.04 | N/A | 2.81(9) | 2.04(7) | 1.69(8) | 1.18(13) | 0.02 | 1.27(4) | 1.09(4) | 0.92(5) | 0.87(9) | 0.58(19) $G_{A}$ | 0.03 | 1.53(3) | 1.16(2) | 0.97(4) | 0.87(5) | 0.69(8) | 0.04 | 1.51(2) | 1.21(2) | 0.99(3) | 0.88(4) | 0.82(8) | 0.02 | N/A | 9.3(8) | 4.9(5) | 3.3(5) | 1.8(7) $G_{P}$ | 0.03 | N/A | 9.5(6) | 6.1(4) | 4.2(4) | 2.6(4) | 0.04 | N/A | 9.7(5) | 6.1(4) | 4.4(3) | 3.1(4) In Fig. 12, Figure 12: Momentum-transfer dependence of vector form factors. A dashed curve corresponds to the dipole form with the empirical value of the dipole mass $M_{V}=0.857(8)$ GeV, which is evaluated from the electric charge and magnetization radii of the proton and neutron. we plot all three $m_{\rm f}$ results for $G_{V}(q^{2})$ as a function of Euclidean four-momentum squared together with the dipole form of $(1+q^{2}/M^{2})^{-2}$ with the empirical value of the isovector Dirac dipole mass $M=0.857(8)$ GeV, which is evaluated from the electric charge and magnetization radii of the proton and neutron. Different symbols represent the values obtained from different quark masses. As can be seen, there is no large quark-mass dependence. Our calculation points are located far from the empirical curve. The slope of the form factor at $q^{2}=0$ determines the mean-squared radius, which can be related to the corresponding dipole mass $M$ as $\langle r^{2}\rangle=12/M^{2}$. To extract the mean-squared radius from our data, we simply adopt the dipole form for fitting three lower $q^{2}$ points including the $q^{2}=0$ value. The fitted values of the Dirac dipole mass $M_{V}$ and corresponding mean-squared Dirac radii $\langle r_{V}^{2}\rangle$ are listed in Table 6. Table 6: Dirac mean-squared charge radius obtained from fitting to the conventional dipole form. $m_{f}$ | $M_{V}$ (GeV) | $\langle r_{V}^{2}\rangle({\rm fm}^{2})$ ---|---|--- 0.02 | 1.40(5) | 0.239(19) 0.03 | 1.49(4) | 0.209(12) 0.04 | 1.57(3) | 0.190(8) Expt. | 0.857(8) | 0.636(12) We show the quark-mass dependence of measured mean square Dirac radius in Fig. 13. Figure 13: The mean-squared Dirac radii $\langle r_{V}^{2}\rangle$ from $G_{V}(q^{2})$. A simple linear extrapolation with respect to the pion mass squared underestimates the experimental point (asterisk). A dotted curve is the predicted quark-mass dependence of mean-squared Dirac radius from LO HBChPT in the vicinity of the physical point. Although it is difficult to extrapolate the lattice estimates at the three heavy pion masses to the physical pion mass, the mild $m_{f}$ dependence in our observed range of $m_{\pi}^{2}$ allows us to perform a simple linear extrapolation. Then we obtain the mean-squared radius of the vector form factor at the physical point: $\langle r_{V}^{2}\rangle=0.281(32)\mbox{ fm}^{2}$, which is much smaller than the experimental value of $0.636(12)\mbox{ fm}^{2}$. However, it is well known that this particular quantity, as well as the pion charge radius, has a logarithmic divergence in heavy baryon chiral perturbation theory (HBChPT) as one approaches the chiral limit Beg and Zepeda (1972). The dotted curve plotted in Fig. 13, is the expected quark-mass dependence of $\langle r_{V}^{2}\rangle$ in the vicinity of the physical point within leading order (LO) HBChPT. (See Appendix .2.2 for more details.) As can be seen, the very steep $m_{\pi}^{2}$ dependence, which is associated with the logarithmic divergence, near the physical point is predicted by HBChPT at leading order. This may account for the smaller values of our measured mean- squared radius, which are calculated in the heavy quark mass region ($m_{\pi}>0.49$ GeV). Thus, we may expect that there is non-linear behavior in terms of $m_{\pi}^{2}$ in the vicinity of the physical point. This expectation may look like somewhat contradictory to what we argue for $g_{A}$ with HBChPT and SSE. However, we recall that the logarithmic divergence can not be cured by higher- order corrections, whereas the strong quark-mass dependence of $g_{A}$, which simultaneously implies slower convergence of the chiral expansion, should be modified by higher-order corrections. Indeed, both the one-loop effective field theory approach with explicit $\Delta$ degrees of freedom (SSE) Procura et al. (2007) and two loop HBChPT calculation Bernard and Meissner (2006) present flat quark-mass dependence at least down to $m_{\pi}\sim 0.3$ GeV. To observe this, we have to at least push the pion mass down to 0.3 GeV, which is approximately the location of the intersection between the LO HBChPT curve and the extrapolated line obtained from our three data points in Fig. 13. We leave this issue to future calculations. Next, we extract the induced-tensor (Pauli) form factor, $G_{T}$, from the vector current; the raw data is listed in Table 5, and an example of $m_{f}=0.04$ at various transfer momenta is shown in Fig. 14. Figure 14: The bare Pauli isovector form factor as in Fig. 11 This corresponds to the combination $F_{2}^{p}-F_{2}^{n}$ of the nucleon electromagnetic form factors. The renormalized Pauli form factor can be obtained from $G_{T}^{\rm ren}(q^{2})=G_{T}(q^{2})/G_{V}(0)$ with vector current renormalization $Z_{V}=1/G_{V}(0)$. Fig. 15 Figure 15: Momentum-transfer dependence of the induced-tensor form factors. The dashed curve corresponds to the dipole form with the experimental value of the dipole mass $M_{T}=0.778(23)$ GeV and nucleon magnetic moment $F_{2}(0)=3.70589$. shows a comparison of our data and the dipole form $G_{T}^{\rm ren}(0)/(1+q^{2}/M_{T}^{2})^{2}$ with the experimental values of $G_{T}^{\rm ren}(0)=F_{2}(0)=3.70589$ and $M_{T}=0.778(23)$ GeV, as described in Appendix A of Ref Sasaki and Yamazaki (2007). The heaviest two pion mass points are almost degenerate, as was the case with the Dirac form factor. However, the data from lightest pion mass here is closer to the experimental values. Again, this could either be interpreted as a trend toward the experimental values with decreasing pion mass or merely finite-volume effects. Furthermore, we fit our $G_{T}^{\rm ren}(q^{2})$ to the dipole form to extract the Pauli mean squared radius, which is related to the corresponding dipole mass $M_{T}$ by $\langle r_{T}^{2}\rangle=12/M_{T}^{2}$. In contrast to the dipole fit on $G_{V}(q^{2})$, it is a two-parameter fit since we do not have data on the value of $G_{T}^{\rm ren}(q^{2})$ at $q^{2}=0$ without the $q^{2}$ extrapolation. Here, the value of the $G_{T}^{\rm ren}(0)$ is associated with the difference between the proton and neutron magnetic moments $\mu_{p}-\mu_{n}=1+F_{2}(0)$. The fitted results are summarized in Table 7. Here we perform the dipole fit with the two lower $q^{2}$ points. Table 7: Pauli mean-squared charge radius obtained from fitting to the conventional dipole form and the extrapolated value of the ratio $G_{M}^{v}(q^{2})/G_{E}^{v}(q^{2})$ at $q^{2}=0$ with the linear $q^{2}$ fitting form. $m_{f}$ | $F_{2}(0)$ | $M_{T}({\rm GeV})$ | $\langle r_{T}^{2}\rangle({\rm fm}^{2})$ | $G^{v}_{M}(0)/G^{v}_{E}(0)$ ---|---|---|---|--- 0.02 | 2.57(37) | 1.29(16) | 0.28(7) | 3.69(41) 0.03 | 2.89(22) | 1.53(12) | 0.20(3) | 4.23(26) 0.04 | 3.32(17) | 1.36(6) | 0.25(2) | 4.24(17) Expt. | 3.70589 | 0.778(23) | 0.773(32) | 4.70589 In Fig. 16, we plot values of $\langle r_{T}^{2}\rangle$ and $F_{2}(0)$ as a functions of pion mass squared Figure 16: The mean-squared Pauli radius $\langle r_{T}^{2}\rangle$ (top) and the value of $G_{T}^{\rm ren}(0)=F_{2}(0)$, which is associated with the difference between the proton and neutron magnetic moments, $\mu_{p}-\mu_{n}=1+F_{2}(0)$, (bottom) and as a function of pion mass squared. A simple linear extrapolation with respect to the pion mass squared underestimates the experimental point (asterisk). along with a naive linear extrapolation to the physical pion mass. In the upper panel of Fig. 16, we linearly extrapolate through all three points, finding $\langle r_{T}^{2}\rangle=0.17(11)\mbox{ fm}^{2}$. This is much smaller than the experimental measurement, $0.773(32)\mbox{ fm}^{2}$, and similar to what we observed for the Dirac mean-squared radius. Lighter pions in future measurements would be desirable to see whether there is an increase in radius as we approach the physical pion mass. In the lower panel of Fig. 16, although the value measured at the heaviest point is close to the experimental value, the extrapolated value of $F_{2}(0)$ at the physical point also tends to somewhat underestimate the experimental value. However, we recall that the value of $F_{2}(0)$ is highly dependent on our adopted fitting form. For example, a monopole fit yields a larger $F_{2}(0)$ which lies closer to the experimental value. Therefore, our estimation of $F_{2}(0)$ should carry a large systematic uncertainty due to the $q^{2}$ extrapolation. We also have an alternative way to evaluate $F_{2}(0)=\mu_{p}-\mu_{n}-1$. The ratio of the isovector magnetic form factor $G^{v}_{M}(q^{2})$ and the isovector electric form factor $G^{v}_{E}(q^{2})$ provides the difference between the proton and neutron magnetic moments in the forward limit, $\mu_{p}-\mu_{n}=G^{v}_{M}(0)/G_{E}^{v}(0)$, which is related to $1+F_{2}(0)$. Experimentally, it is known that this ratio shows no $q^{2}$ dependence at low $q^{2}$ since both form factors are well fitted by the dipole form with the comparable dipole masses. Therefore, this ratio may have milder $q^{2}$ dependence than $G_{T}^{\rm ren}(q^{2})$ Sasaki and Yamazaki (2007). Some RBC results on related ratios for the heavier quark masses were reported in Ref. Berruto et al. (2006). In Fig. 17, we show the $q^{2}$ dependence of the ratio $G^{v}_{M}(q^{2})/G_{E}^{v}(q^{2})$, which clearly exhibits mild dependence. We therefore use a simple linear fitting form with respect to $q^{2}$ for an alternative evaluation of the value $\mu_{p}-\mu_{n}$. In Fig. 18, we plot values of $\mu_{p}-\mu_{n}$, from two determinations, $1+F_{2}(0)$ and $G_{M}(0)/G_{E}(0)$, as functions of pion mass squared. Both determinations are roughly consistent with each other within the statistical error. Weak $m_{\pi}^{2}$ dependence is observed; the heavier two points in the case of $G_{M}(0)/G_{E}(0)$ are the same within error, while slight downward $m_{\pi}^{2}$ dependence appears in the results of $1+F_{2}(0)$. The simple linear extrapolation of $G_{M}(0)/G_{E}(0)$ yields $\mu_{p}-\mu_{n}=3.4(7)$, which gives the better agreement with the experimental value. We may quote this value for our final value of $\mu_{p}-\mu_{n}$, since the latter approach appears to have smaller systematic error in the $q^{2}$-extrapolation. Figure 17: The $q^{2}$ dependence of the ratio $G^{v}_{M}(q^{2})/G^{v}_{E}(q^{2})$. A simple linear extrapolation with respect to $q^{2}$ is utilized for an alternative evaluation of $\mu_{p}-\mu_{n}$ thanks to its mild $q^{2}$ dependence. Figure 18: Comparison of two determinations of $\mu_{p}-\mu_{n}$. There is no appreciable $m_{\pi}^{2}$ dependence for the results from $G^{v}_{M}(0)/G^{v}_{E}(0)$. A simple linear extrapolation is applied to them. The resulting extrapolated value at the physical point shows the better agreement with the experimental value (asterisk). #### III.2.2 Axial-vector current In Fig. 19 Figure 19: The bare axial form factor as in Fig. 11 we present the raw data for the axial form factor at finite momentum transfer. These allow us to extract the momentum-transfer dependence of the form factor as listed in Table 5 and plotted in the top of Fig. 20. Figure 20: (Top) Momentum-transfer dependence of vector form factors. The dashed curve corresponds to the dipole form with the experimental value of the dipole mass $M_{A}=1.026(21)$ GeV, evaluated from pion electroproduction experiments. (Bottom) Mean-square axial radii $\langle r^{2}_{A}\rangle$ from $G_{A}(q^{2})$. The shaded regions are naive linear extrapolations (with errors) of our data points to the physical pion mass. An unexpected reduction of the measured $\langle r^{2}_{A}\rangle$ at the lightest quark mass is observed. The axial form factor is phenomenologically fitted with the dipole form, at least at low $q^{2}$; so are the Dirac and Pauli form factors Bernard et al. (2002). The dashed curve in the top of Fig. 20 shows the dipole form with an experimental value of the axial dipole mass $M_{A}=1.026(21)$ GeV Bernard et al. (2002). Our lattice data lie above the experimental curve. Notice that the likely finite volume effect observed earlier for the axial charge leads to the non-monotonic behavior of the form factor with quark mass, and leads to the downward curvature of the mean-square axial radius. To extract the mean-squared radius of the axial vector form factor, we perform the dipole fit to the three lowest $q^{2}$ points and convert the axial dipole mass $M_{A}$ into the mean-squared axial radius $\langle r_{A}^{2}\rangle$. The obtained values are listed in Table 8 and the bottom of Fig. 20. Table 8: Dipole fits for momentum-transfer dependence of the axial form factor. $m_{f}$ | $M_{A}$ (GeV) | $\langle r_{A}^{2}\rangle$ $({\rm fm}^{2})$ ---|---|--- 0.02 | 2.26(25) | 0.091(20) 0.03 | 1.72(7) | 0.158(11) 0.04 | 1.90(7) | 0.129(9) Expt. | 1.026(21) | 0.444(19) We also plot the mean-squared axial radius as a function of the pion mass squared. As can be seen, the value of $\langle r_{A}^{2}\rangle$ at the lightest pion mass shows a large reduction beyond statistical fluctuations. This large reduction in the axial radius suggests finite-volume effects could be significant at the lightest quark mass. There is renewed interest in the pseudoscalar form factor, $G_{P}(q^{2})$, due to the recent MuCap Collaboration Andreev et al. (2007) high-precision experiment studying ordinary muon capture by protons, $\mu^{-}p\rightarrow\nu_{\mu}n$, and because of improved electroweak radiative correction calculationsCzarnecki et al. (2007) that allow precise extraction of the form factor from these experiments. Using the new MuCap results, a value of $g_{P}=\frac{m_{\mu}}{2m_{N}}G_{P}(0.88m_{\mu}^{2})=7.3\pm 1.1$ is foundAndreev et al. (2007); Czarnecki et al. (2007) which, using PCAC and chiral perturbation theory, is now in good agreement with QCD, $g_{P}=8.26\pm 0.12$ Bernard et al. (2002). Earlier, a TRIUMF group obtained a value of $12.4\pm 1.0$ Jonkmans et al. (1996). Clark et al. Clark et al. (2006) got $g_{P}=10.6\pm 1.1$ after reanalyzing the TRIUMF data. Including the new MuCap result, the “world average” is $8.7\pm 1.0$ Czarnecki et al. (2007). The induced pseudoscalar form factor $G_{P}$ is obtained from Eq. (18), and is shown in Fig. 21. Figure 21: The bare pseudoscalar form factor as in Fig. 11 This allows us to extract its momentum dependence which is listed in Table 5 and plotted in Fig. 22. Figure 22: Momentum-transfer dependence of the induced pseudoscalar form factor compared with experiments. As is noted in the previous section, we use a dimensionless definition for this, as well as the other three form factors, and use twice the nucleon mass $2m_{N}$ estimated for each ensemble of configurations. This is different from some experimental analyses where the muon mass is used for normalization. Momentum-transfer dependence of the pseudoscalar form factor $G_{P}(q^{2})$ has so far only been studied by one pion electroproduction experiment at Saclay in 1993 Choi et al. (1993) with range below $0.2\mbox{ GeV}^{2}$. Fig. 22 shows the $q^{2}$ dependence of the ratio of the induced pseudoscalar to axial form factors, $G_{P}(q^{2})/G_{A}(q^{2})$. Though our lattice momenta transfer are much higher than experimentally explored values, they trend upward toward the experiments. Another topic regarding the induced pseudoscalar form factor is its relation with the axial form factor through the PCAC relation and pion-pole dominance (PPD) model Braun et al. (2006): $G^{\rm PPD}_{P}(q^{2})=\frac{4m_{N}^{2}G_{A}(q^{2})}{q^{2}+m_{\pi}^{2}}.$ (32) To see how the pion-pole behavior is preserved in measured $G_{P}(q^{2})$, we consider the following ratio: $\alpha_{{}_{\rm PPD}}=\frac{G_{P}(q^{2})}{G^{\rm PPD}_{P}(q^{2})},$ (33) which is inspired by the above PCAC prediction. If the measured $G_{P}(q^{2})$ has exactly the same form described in Eq. (32), this ratio yields unity in the entire $q^{2}$ region. In Fig. 23, Figure 23: The ratio of $G_{P}(q^{2})$ and $G_{P}^{\rm PPD}(q^{2})$ as a function of four-momentum squared $q^{2}$. Deviation of the ratio from unity indicated by dashed line is the deviation from the PPD model. However, the deviation is rather modest. At least, the $q^{2}$ dependence of the PPD model approximately accounts for that of measured $G_{P}(q^{2})$. we plot the above ratio as a function of four-momentum squared $q^{2}$. There is no appreciable $q^{2}$ dependence. Thus, the $q^{2}$ dependence of the PPD model approximately accounts for that of measured $G_{P}(q^{2})$. Four different $q^{2}$ points of the ratio $\alpha_{{}_{\rm PPD}}$ reveal a $q^{2}$ independent plateau within the statistical errors. We simply take the weighted average of $\alpha_{{}_{\rm PPD}}$ within four measured $q^{2}$ points, then plot them against the pion mass squared. As shown in Fig. 24, Figure 24: The quenching factor $\alpha_{{}_{\rm PPD}}$ is plotted as a function of the pion mass squared. the average values of $\alpha_{{}_{\rm PPD}}$ gradually approach unity as the pion mass decreases. A simple linear extrapolation of the quenching factor $\alpha_{{}_{\rm PPD}}$ to the chiral limit yields 0.85(8), which is only about 2 standard deviations away from the theoretical expectation. One can obtain $g_{P}$ with input from $\alpha_{{}_{\rm PPD}}$: $g_{P}=\alpha_{{}_{\rm PPD}}\frac{g_{A}}{(1+q_{0}^{2}/M_{A}^{2})^{2}}\frac{2m_{N}m_{\mu}}{(m_{\pi}^{2}+q_{0}^{2})},$ (34) where $q_{0}^{2}=0.88m_{\mu}^{2}$ with $m_{\mu}=0.10568$ GeV and $m_{\pi}=0.13957$ GeV. $M_{A}$ and $g_{A}$ are 1.38(16) GeV and 1.23(12) respectively from naive linear fit using two heavier pion mass points. We found $g_{P}=7.68\pm 1.03$ (35) with statistical error evaluated by jackknife analysis. This is consistent with the “world average” value of $8.7\pm 1.0$ Czarnecki et al. (2007). The pseudoscalar form factor $G_{P}$ is related to the pion-nucleon coupling, $g_{\pi NN}$, through the Goldberger-Treiman relation Goldberger and Treiman (1958): $2m_{N}G_{A}(q^{2})-q^{2}\frac{G_{P}(q^{2})}{2m_{N}}=\frac{2g_{{\pi NN}}F_{\pi}m_{\pi}^{2}}{q^{2}+m_{\pi}^{2}}.$ (36) At zero momentum transfer, the relation reduces to the following form: $m_{N}g_{A}=F_{\pi}g_{\pi NN}.$ (37) Testing the relation experimentally in this zero-momentum form is a little tricky because the value of the pion-nucleon coupling, $g_{\pi NN}$, is not known at zero momentum transfer, but at the pion pole. Worse, the pion-pole value of the coupling varies from analysis to analysis. A conventional partial-wave analysis of the pion-nucleon elastic scattering data to 2 GeV provides $13.75\pm 0.15$ Arndt et al. (1994) while a more recent reanalysis in Ref. Bugg (2004) gives $g_{\pi NN}=13.169\pm 0.057$. Substituting the latter, we find there is a slight discrepancy from the Goldberger-Treiman relation at zero momentum transfer, $\Delta_{\rm GT}$, Coon and Scadron (1981) defined as $\Delta_{\rm GT}=1-{m_{N}g_{A}}/{F_{\pi}g_{\pi NN}}$ (38) which is $(2.259\pm 0.591)\%$ if we use the mean nucleon mass $m_{N}=938.9$ MeV. Testing the relation at finite momenta transfer is more difficult, as the pion-nucleon coupling is even more poorly known from experiment. Thus it is useful to extract the pion-nucleon coupling, $g_{{\pi NN}}$, from our lattice data: $g_{\pi NN,{\rm lat}}=\frac{m_{N,{\rm lat}}g_{A,{\rm lat}}}{F_{\pi,{\rm lat}}}.$ (39) See Fig. 25 Figure 25: Lattice estimate for the pion-nucleon coupling, $g_{\pi NN}^{\rm lat}$, from the Goldberger-Treiman relation as defined in Eq. (39). Omitting the lightest mass point, a constrained linear fit is performed. for $g_{\pi NN,{\rm lat}}$ results. A linear fit is performed without the lightest mass point to give an estimate of $g_{\pi NN}=15.5\pm 1.4$ at the physical point. This should be compared with the experimental estimates at the pion pole such as $13.75\pm 0.15$ or $13.169\pm 0.057$ in the above. Now let us examine the Goldberger-Treiman relation by looking at the discrepancy $\Delta_{\rm GT}(q^{2})$ defined by the lattice quantities as follows: $1-\frac{q^{2}+m_{\pi,{\rm lat}}^{2}}{2m_{N,{\rm lat}}}\frac{4m_{N,{\rm lat}}^{2}G_{A,{\rm lat}}(q^{2})-q^{2}G_{P,{\rm lat}}(q^{2})}{2g_{{\pi NN,{\rm lat}}}F_{\pi,{\rm lat}}m_{\pi,{\rm lat}}^{2}};$ (40) see Fig. 26 Figure 26: Momentum-transfer dependence of the Goldberger-Treiman discrepancy as defined in Eq. (40). for the value of $\Delta_{\rm GT}(q^{2})$. Note again that we are using the pion-nucleon coupling value determined at zero-momentum transfer, and cannot account for its variation at finite momentum transfer. From our data for the two heavier quark mass values, we observe the discrepancy is strongly dependent on the momentum transfer beyond about $q^{2}\approx 0.5~{}\mbox{ GeV}^{2}$. This suggests the pion-nucleon coupling at such high momentum transfer values is very much different from the low momentum-transfer region. Note our pion mass squared is roughly $m_{\pi}^{2}$ = 0.25, 0.36 and $0.49\mbox{ GeV}^{2}$, or below the lowest $q^{2}$ in the plot, and the physical pion mass is even lower. Thus our data also suggest the Goldberger- Treiman relation holds at low momentum transfer, $q^{2}\leq 0.5\mbox{ GeV}^{2}$. In that region the lightest quark mass result is in broad agreement with the heavier masses, albeit with large statistical error. It is desirable to investigate this further at lower momentum transfer. ### III.3 Momentum and helicity fractions Figure 27: Bare momentum fraction as a function of the operator insertion time. Results are averaged over all sources. The top figure depicts the $t_{\rm sep}=12$ calculation at the lightest quark mass, $m_{f}=0.02$. The other three figures are obtained from $t_{\rm sep}=10$ calculations at all three quark masses (in order of increasing mass from top to bottom). The fit ranges are shown by horizontal lines. Let us now turn our attention from the elastic form factors to the deep inelastic structure functions. As is well known and has been summarized in the introduction, what are calculable on the lattice in regard to the structure functions are their low-order moments. In this work we limit ourselves to the lowest order non-trivial moments that are calculable with zero momentum transfer. As in previous RBC reports, the operator renormalizations at the chiral limit are obtained nonperturbatively using the RI/MOM scheme as described in Sec. II.3. We first discuss the quark momentum fraction, $\langle x\rangle_{u-d}$, or the first moment of the unpolarized structure function. The bare lattice three- point correlators are shown in Fig. 27 and listed in Table 9. The results for $t_{\rm sep}=12$ and $t_{\rm sep}=10$ at the lightest pion mass differ by about two standard deviations. This may suggest excited state contamination in this quantity, even though its plateau looks very nice and flat over the range $3\leq t\leq 7$ for $t_{\rm sep}=10$. From the top panel in Fig. 27, the plateau for $t_{\rm sep}=12$ is not so clear or flat, owing to the much larger statistical fluctuations in the data for this larger source-sink separation. However, since it is low on the left and high on the right, the average value is insensitive to the choice of (a symmetric) fit range. In the earlier calculation, using a box source on a single time slice and a source-sink separation of 12 Ohta and Orginos (2005); Orginos (2003), the momentum fraction was consistent, within large statistical errors, with the $t_{\rm sep}=10$ result. In this study, we can not conclude whether this is a true systematic effect or just statistics, especially given the quality of the $t_{\rm sep}=12$ plateau. We show the results for all the measurements but only use the result from $t_{\rm sep}=10$ at the lightest pion mass for extrapolation. Similar results would be obtained using the $t_{\rm sep}=12$ result at the lightest mass since the larger error on this point does not tightly constrain the fit. Using the fit ranges indicated in the figures and the renormalizations presented in subsection II.3, we arrive at the values of $\langle x\rangle_{u-d}$ shown in Fig. 28. Figure 28: Renormalized first moment of the unpolarized structure functions or the quark momentum fraction, $\langle x\rangle_{u-d}$. Note that the circles indicate the data at each pion mass with two errorbars: the inner are statistical errorbars and the outer are errorbars propagated with renormalization factors. The solid square is the extrapolated point; the star is the experimental value; and the band shows the chiral extrapolation with jackknife-calculated uncertainty. Note that there is only small pion-mass dependence; therefore, we use a simple linear fit. The quality of this fit in terms of $\chi^{2}$ per degree of freedom is reasonable at 0.28. The chiral extrapolated value is $\langle x\rangle_{u-d}=0.282(19)$, overshooting the experimental value, 0.154(3) Lai et al. (1997); Gluck et al. (1998); Martin et al. (2002), by five to six standard deviations. Here, the trend of the $t_{\rm sep}=12$ data point is suggestive, but again, not conclusive because of the relatively large statistical error. The quark helicity fraction, $\langle x\rangle_{\Delta u-\Delta d}$, is the first moment of the polarized structure function. The bare three-point correlator values are shown in Fig. 29 and the values are listed in Table 9. Similar remarks and conclusions about the plateaus for $t_{\rm sep}=10$ and 12 as for the momentum fraction hold here as well. Figure 29: The bare helicity fraction; symbols as in Fig. 27. Figure 30: Renormalized first moment of the polarized structure functions or quark helicity fraction, $\langle x\rangle_{\Delta u-\Delta d}$, as a function of $m_{\pi}^{2}$. (Similar graphics description as in Fig. 28.) The renormalization factors at the chiral limit for the helicity fraction can be found in Sec. II.3. Fig. 30 shows the renormalized quantities at each pion mass point. A linear fit to these values versus $m_{\pi}^{2}$ gives a chiral extrapolation of 0.286(25), overshooting the experimental value of 0.196(4) Gluck et al. (1996); Gehrmann and Stirling (1996); Dolgov et al. (2002) by almost four standard deviations. The quality of the fit is poor, as $\chi^{2}$ per degree of freedom is 6.78. In contrast to the momentum fraction, the result at the light mass deviates from the heavier pion mass points, even at $t_{\rm sep}=10$, a situation similar to the deviation observed in the axial charge. This trend is welcome since the experimental value apparently must be approached from above and suggests trying fits different than linear, to which we turn below, after discussing the ratio of the momentum fraction to the helicity fraction. In an earlier quenched calculation Orginos et al. (2006) we reported that the ratio of the momentum and helicity fractions, $\langle x\rangle_{u-d}/\langle x\rangle_{\Delta u-\Delta d}$, agrees well with the experimental value, 0.78(2). The quark mass dependence of this ratio is presented in Fig. 31. Unlike the quenched case where the ratio is only weakly dependent on the quark mass and is in rough agreement with experiment, there is a noticeable deviation for the lightest pion mass. A naive linear extrapolation gives very poor fit quality, $\chi^{2}/{\rm dof}=15.7$, and a value at the physical pion mass of $0.93(7)$, which is about two standard deviations above the experimental one. This may be due to systematic effects like finite volume, or excited state contamination; the $t_{\rm sep}=12$ point is consistent with experiment and shows the weak mass dependence observed in the quenched case. Figure 31: Renormalized ratio $\langle x\rangle_{\Delta u-\Delta d}/\langle x\rangle_{u-d}$ of the quark momentum and helicity fraction. A linear fit is shown with an error estimate at the physical pion mass. Experiment gives 0.78(2). Table 9: Bare quark momentum and helicity fractions and their naturally renormalized ratio at each sea quark mass. $m_{f}$ | $t_{\rm sep}$ | $\langle x\rangle_{u-d}$ | $\langle x\rangle_{\Delta u-\Delta d}$ | $\langle x\rangle_{u-d}/\langle x\rangle_{\Delta u-\Delta d}$ ---|---|---|---|--- 0.02 | 12 | 0.195(17) | 0.22(3) | 0.99(14) 0.02 | 10 | 0.236(9) | 0.255(11) | 0.93(4) 0.03 | 10 | 0.242(6) | 0.306(6) | 0.790(17) 0.04 | 10 | 0.253(4) | 0.306(6) | 0.827(12) Since our pion masses are rather heavy (700, 600 and 500 MeV), it is not surprising that the chiral extrapolation to the physical pion mass (140 MeV) sometimes misses the experimental values. Also our heavy pion masses probably invalidate the application of a leading-order chiral form. The chiral extrapolation should ultimately be studied with more realistic lattice QCD ensembles such as the 2+1-flavor ones being generated jointly by RBC and UKQCD CollaborationsAntonio et al. (2007); Allton et al. (2007), a work which is in progress. Nevertheless, it is a worthwhile exercise to try to extend the extrapolations beyond linear forms. We now discuss how some such attempts fair with our data. We begin with a parametrization by Chen, et al. Chen and Savage (2002); Chen and Ji (2001); Detmold et al. (2002) (the relevant equations are summarized in appendix .2.3.) Their prescription is to fix three parameters, the axial charge $g_{A,{\rm exp}}$, pion decay constant $F_{\pi,{\rm exp}}$ and the scale $\mu$ (set to the pion mass), to their experimental values and fit the coefficients $C$ and $e(\mu^{2})$. The results are summarized in Fig. 32. The fit yields momentum fraction 0.147(9) after extrapolation, which is consistent with the experimental value. However, the quality of the fit is poor, $\chi^{2}/{\rm dof}=2.19$; this is caused by a strong downward curvature in the fit form due to log terms. The corresponding helicity fraction extrapolation results in a value of 0.170(14), again, consistent with the experimental value. The fit for the helicity fraction, while still poor, is improved over the linear one, with a $\chi^{2}$ per degree of freedom of 3.57. Figure 32: Renormalized quark momentum fraction, $\langle x\rangle_{u-d}$, and helicity fraction, $\langle x\rangle_{\Delta u-\Delta d}$. The phenomenological fit (brown band) is according to Refs. Chen and Savage (2002); Chen and Ji (2001); Detmold et al. (2002). The dashed lines are the linear extrapolation for comparison. (Graphics conventions as in Fig. 28.) Another alternative is to consider the dependence in quantities on the dimensionless ratio $m_{\pi}^{2}/F_{\pi,{\rm lat}}^{2}$, as shown in Fig. 33. Similar analyses have been done before, for example, in Refs. Beane et al. (2006); Edwards et al. (2006). (Note that $F_{\pi,{\rm lat}}$ is replaced by $f_{\pi,{\rm lat}}(m_{f})/\sqrt{2}$ at each $m_{f}$ to be consistent with the chiral extrapolation formulation used in Refs. Chen and Savage (2002); Chen and Ji (2001); Detmold et al. (2002).) Since this is a dimensionless quantity, there is no systematic error coming from the lattice scale determination; therefore, it is a good way to find out whether we have control over lattice artifacts. For the momentum fraction, $\langle x\rangle_{u-d}$, the fit yields an extrapolated value of 0.260(30) with an acceptable $\chi^{2}/{\rm dof}=0.48$. That this extrapolation is consistent with the earlier ones indicates no further uncontrollable systematics other than the extrapolation itself. The value is about three standard deviations above experiment. Note the larger error on the extrapolated point is caused by the “re-scaling” of the extrapolation range. Similar conclusions are drawn for the helicity fraction, $\langle x\rangle_{\Delta u-\Delta d}$: the updated extrapolation brings the value to 0.224(41), within two standard deviations of experiment. However, the $\chi^{2}$ per degree of freedom for this fit is still large at 5.17. The fit results from the non-linear chiral extrapolations (see Appendix .2.3 for more details) are summarized in Fig. 33. The momentum fraction is 0.171(19) at the physical point and agrees with experiment within one standard deviation. The fit has an acceptable $\chi^{2}$ per degree of freedom of 0.85. The helicity fraction extrapolates to 0.163(28) and is consistent both with the extrapolation in terms of $m_{\pi}$ and the experimental number. However the fit quality is poor: $\chi^{2}/{\rm dof}=4.57$. These values are consistent with the earlier fits. Figure 33: Renormalized quark momentum fraction, $\langle x\rangle_{u-d}$, and helicity fraction, $\langle x\rangle_{\Delta u-\Delta d}$, in terms of $m_{\pi}^{2}/F_{\pi,{\rm lat}}^{2}$. The blue band indicates the linear extrapolation and the brown band is the phenomenological fit as described in Refs. Chen and Savage (2002); Chen and Ji (2001); Detmold et al. (2002). (Graphics conventions as in Fig. 28.) The trial fits and extrapolations are summarized in Table 10. In the cases of both the quark momentum and helicity fraction, the extrapolation with the chiral form seems to give us consistent values with the experiments. However, these extrapolations require a 50% drop from the magnitude of our measured points and do not always give the smallest $\chi^{2}/{\rm dof}$. We simply must measure at lower pion masses before such extrapolations can be deemed trustworthy. Also note that the extrapolations in terms of $m_{\pi}^{2}$ or the dimensionless quantity $m_{\pi}^{2}/F_{\pi,{\rm lat}}^{2}$ are consistent with each other. Table 10: Summary of the extrapolations of the renormalized first moment of the unpolarized distribution. The chiral perturbation theory formulation can be found in Ref. Chen and Savage (2002); Chen and Ji (2001); Detmold et al. (2002) or appendix .2.3. The fit parameters are in the order of fit slope, intersection for the linear fits and $C$, $e$ for the chiral formulation. | $\langle x\rangle_{u-d}$ | $\chi^{2}/{\rm d.o.f.}$ | fit parameters ---|---|---|--- Linear vs. $m_{\pi}^{2}$ | 0.282(19) | 0.28 | {0.09(5), 0.280(19)} Linear vs. $m_{\pi}^{2}/F_{\pi,{\rm lat}}^{2}$ | 0.260(30) | 0.48 | {0.0024(12), 0.26(3)} ChPT vs. $m_{\pi}^{2}$ | 0.147(9) | 2.19 | {0.109(8), 2.64(11)} ChPT vs. $m_{\pi}^{2}/F_{\pi,{\rm lat}}^{2}$ | 0.171(19) | 0.85 | {0.139(17), 3.74(22)} experiment | 0.154(3) | | | $\langle x\rangle_{\Delta u-\Delta d}$ | $\chi^{2}/{\rm d.o.f.}$ | fit parameters Linear vs. $m_{\pi}^{2}$ | 0.286(25) | 6.78 | {0.21(6), 0.281(27)} Linear vs. $m_{\pi}^{2}/F_{\pi,{\rm lat}}^{2}$ | 0.224(41) | 5.17 | {0.0061(17), 0.22(4)} ChPT vs. $m_{\pi}^{2}$ | 0.170(14) | 3.57 | {0.134(12),0.25(12)} ChPT vs. $m_{\pi}^{2}/F_{\pi,{\rm lat}}^{2}$ | 0.163(28) | 4.57 | { 0.135(27), 3.31(20)} experiment | 0.196(4) | | We emphasize that the data in general is not consistent with a simple linear form, but that it does trend toward experiment as the quark mass is reduced. ### III.4 Transversity Table 11: Bare transversity. $m_{f}$ | $t_{\rm sep}$ | $\langle 1\rangle_{\delta u-\delta d}$ ---|---|--- 0.02 | 12 | 1.29(11) 0.02 | 12 | 1.25(4) 0.03 | 10 | 1.43(3) 0.04 | 10 | 1.41(4) Another interesting quantity regarding the spin structure of the nucleon is its transversityJaffe and Ji (1993); Grosse Perdekamp (2002), $\langle 1\rangle_{\delta q}$. Here again we have calculated only the isovector quantity. The raw data are shown in Fig. 34, together with the fit range we use; Table 11 summarizes the bare values. Again, at the $m_{f}=0.02$ point, we use two different source-sink separations to control possible excited state contributions; in this case the two choices give results that are comfortably consistent. Therefore, we use the value obtained from $t_{\rm sep}=10$ for all further analysis and discussion. Figure 34: The bare values of the transversity; symbols as in Fig. 27. Using these fit values and the renormalization at the chiral limit listed in Table 3 we obtain the quark-mass dependence as shown in Fig. 35. Figure 35: Nonperturbatively renormalized transversity with linear extrapolation to the physical pion mass. (Graphic conventions as in Fig. 28.) Similar to the axial charge radius, we observe a significant deviation at the lightest quark mass. We extrapolate linearly and obtain $\langle 1\rangle_{\delta u-\delta d}=0.93(6)+0.37(17)m_{\pi}^{2}$ with a $\chi^{2}$ per degree of freedom of 7.3. This translates into a value at the physical point of 0.93(6). It will be interesting to compare this result with experiment when the latter becomes available. ### III.5 Twist-3 moment We also calculated the twist-3 first moment of the polarized structure function, $d_{1}$. The good chiral symmetry of DWF prevents lower-dimensional operators from spoiling the calculation, unlike in calculations with fermion discretizations that violate chiral symmetry, such as Wilson fermions. The calculated bare values are summarized in Fig. 36 and Table 12. Figure 36: The bare twist-3 moment $d_{1}$; symbols as in Fig. 27. The quark-mass dependence of the bare values is plotted in Fig. 37. (Note that we do not intend to compare these calculations with experiment so do not renormalize them.) A linear fit yields $-0.0059(16)+0.126(4)m_{\pi}^{2}$ and an extrapolated value $d_{1}^{\rm bare}=-0.0035(13)$ at the physical pion mass, which is about three standard deviations away from zero, but small compared to the values at non-zero mass. This suggests the Wandzura-Wilczek relation Wandzura and Wilczek (1977), which need not hold in a confining theory, nevertheless holds in QCD. Figure 37: Bare values of the twist-3 moment $d_{1}$, with linear extrapolation to the physical pion point. Up quark (up triangles), down quark (down triangles), and the isovector combination (circles) are shown. Table 12: Bare values of the twist-3 moment $d_{1}$ of the polarized structure function. $m_{f}$ | $t_{\rm sep}$ | $d_{1}^{u-d}$ | $d_{1}^{u}$ | $d_{1}^{d}$ ---|---|---|---|--- 0.02 | 12 | 0.0240(25) | 0.0185(19) | $-$0.0055(13) 0.02 | 10 | 0.0246(8) | 0.0189(7) | $-$0.0056(4) 0.03 | 10 | 0.0411(8) | 0.0325(6) | $-$0.0085(3) 0.04 | 10 | 0.0544(7) | 0.0428(6) | $-$0.0117(3) ## IV Conclusion and Outlook In this work, we have presented numerical lattice QCD calculations using two degenerate flavors of dynamical (DWF) quarks. We calculated isovector vector, axial, tensor and pseudoscalar form factors and some low-order moments of the unpolarized and polarized structure functions. We found that the ratio of the axial charge to vector charge, $g_{A}/g_{V}$, has a significant drop at the lightest pion mass point (about 500 MeV) which may be a sign of finite-volume effect; the heavier pion mass points (around 600 and 700 MeV) are in rough agreement with experiment and show small dependence on the quark mass. A linear extrapolation in pion mass-squared gives 0.89(6), which deviates from experiment by more than five standard deviations; leaving out the lightest point gives 1.23(12). If the downward trend at the lightest point, away from experiment, is indeed a finite-volume effect, then even $m_{\pi}L\approx 5$ is not large enough for nucleon calculations with dynamical fermions, in contrast to quenched calculations Sasaki et al. (2003)222This effect may in fact be illusory due to the presence of quenched chiral logs as recently argued in Yamazaki et al. (2008). We studied the momentum-transfer dependence of the Dirac, Pauli, axial and induced pesudoscalar form factors. Using a conventional dipole extrapolation, we found masses $M_{V}$, $M_{T}$ and $M_{A}$ at each pion mass points that are more than 50% higher than the experimental ones, while the mean-square charge radii, $\langle r_{V}^{2}\rangle$, $\langle r_{T}^{2}\rangle$, $\langle r_{A}^{2}\rangle$ are far below experiment. Similar effects have been observed in the past dynamical calculations Edwards et al. (2006). Most phenomenological studies expect to see a dramatic increase at lower pion mass; although this trend is not observed in our current study, it may be resolved in the future with lighter pion calculations. The magnetic moments of nucleon were calculated in two different ways: first, directly from extrapolation of the Pauli form factor and second, by looking at the ratio of the isovector electric and magnetic moments $G^{v}_{M}(0)/G_{E}^{v}(0)$ in the forward limit; the latter expression has milder momentum dependence. Using the ratio method, we see decreasing $m_{\pi}^{2}$ dependence, and with a simple linear extrapolation we find $\mu_{p}-\mu_{n}=3.4(7)$, roughly consistent with the experimental value. In studying the axial current form factors, we find that the renormalized axial form factor at finite momenta appears to be in broad agreement with experimental values from neutrino scattering. The Goldberger-Treiman relation appears to hold for low momentum transfer, $q^{2}\leq 0.5\mbox{ GeV}^{2}$. Assuming the relation, we obtain an estimate for the pion-nucleon coupling of $g_{\pi NN}=15.5(1.4)$ and $g_{P}=\frac{m_{\mu}}{2m_{N}}G_{P}(0.88m_{\mu}^{2})=7.68\pm 1.03$, which is consistent with experiment. For the structure functions, the renormalized values of $\langle x\rangle_{u-d}$ and $\langle x\rangle_{\Delta u-\Delta d}$ show a trend toward their respective experimental values at the lightest quark mass point in our study. The quark momentum fraction, $\langle x\rangle_{u-d}(2\mbox{ GeV})$, is 0.282(19) by simple linear chiral extrapolation, which overshoots the experimental value by more than three standard deviations. The quark helicity fraction, $\langle x\rangle_{\Delta u-\Delta d}(2\mbox{ GeV})$, is 0.286(25) by simple linear extrapolation, which overshoots the experimental value by almost four standard deviations. The former does not deviate significantly from a linear ansatz when the source-sink separation is set to 10 time slices. When increased to 12, the lightest point lies below the straight line fit, suggesting possible excited state contamination for $t_{\rm sep}=10$, and perhaps the desired physical quark mass dependence for $t_{\rm sep}=12$. However, the statistics for the larger separation are relatively poor, and definitive conclusions can not be drawn. A similar pattern holds for the helicity fraction, except that at $t_{\rm sep}=10$ a non-linear quark mass dependence is already observed. Non-linear fits motivated by chiral perturbation theory were also used to fit the data. Again, these indicated favorable trends in some cases, but definitive results require simulations with lighter quark masses and multiple volumes, work that is now well underway. We also calculated transversity and twist-3 matrix elements. The transversity renormalized at 2 GeV, $\langle 1\rangle_{\delta u-\delta d}(2\mbox{ GeV})$, is 0.93(6) by simple linear extrapolation in $m_{\pi}^{2}$. However, there is noticeable non-linearity in the data. A linear extrapolation to the physical point yields $d_{1}^{\rm bare}=-0.0035(13)$ for the twist-3 operator, which is about three standard deviations away from zero, indicating only small breaking of the Wandzura-Wilczek relation Wandzura and Wilczek (1977). Unfortunately, even with our dynamical calculations, we cannot resolve the long-standing differences between the predictions of QCD and experimental measurements, in such quantities as the charge radii, and quark momentum and helicity fractions. If both QCD and the relevant experiments are correct, we expect to see a dramatic shift in these values as the constraints of high quark mass, small volume and coarse lattice spacing are lifted. The limitations of the current gauge ensembles do not allow such a study at this stage. However, we are currently studying these quantities with a new series of ensembles produced by the RBC and UKQCD Collaborations Antonio et al. (2007); Allton et al. (2007). And another ensemble is being generated by these groups and LHPC. They are respectively at $a^{-1}=1.7$ and 2.1 GeV lattice cutoff with up to ($2.75\mbox{ fm})^{3}$ volumes; they feature physical dynamical strange quark mass and two degenerate up and down quarks as light as 1/7 the strange mass. This will give us better control of the systematic errors in a calculation with much lighter pion masses in the valence and the sea sectors; whether we will start to see the curvature suggested by the chiral perturbation theory or other phenomenological models will be very interesting to discover. ## ACKNOWLEDGMENTS The authors would like to acknowledge S. Choi for his private communication providing actual values of the form factor $g_{P}(q^{2})$ in his experiment and Kostas Orginos for his useful discussions on this work. We thank RBC members for physics discussions and RIKEN, Brookhaven National Laboratory and the U.S. Department of Energy for providing the facilities essential for the completion of this work. In addition, TB and TY were partially funded by US Department of Energy Outstanding Junior Investigator Grant DE-FG02-92ER40716; SS is supported by the JSPS for a Grant-in-Aid for Scientific Research (C) (No. 19540265); HWL is supported by DOE contract DE-AC05-06OR23177 under which the Jefferson Science Associates, LLC operates the Thomas Jefferson National Accelerator Facility. ## Appendices ### .1 Nonperturbative renormalization In this appendix, we describe the formulation we used in the nonperturbative renormalization in more detail. In subsection II.3, we obtain renormalization constants in RI/MOM scheme. In order to compare our results with the experimental ones, we need to further to convert the renormalization constant to $\overline{\rm MS}$ scheme. In the case of the first moments of the momentum fraction and helicity operators, one finds the conversion equations in the continuum via Gockeler et al. (1999b) $\displaystyle\frac{Z^{{\rm\overline{MS}}}}{Z^{{\rm RI}}}$ $\displaystyle=$ $\displaystyle 1+\frac{g^{2}}{16\pi^{2}}C_{F}\Bigg{[}G_{n}+(1-\xi)S_{n-1}$ $\displaystyle+$ $\displaystyle\left(-\frac{4}{n+1}+(1-\xi)\frac{2}{n}\right)\frac{\left(\sum_{\mu}p_{\mu}h_{\mu}(p)\right)^{2}}{p^{2}\sum_{\mu}h_{\mu}(p)^{2}}\Bigg{]}$ with $\displaystyle G_{n}$ $\displaystyle=$ $\displaystyle\frac{2}{n(n+1)}\left(-3-S_{n-1}+2S_{n+1}\right)$ (42) $\displaystyle+$ $\displaystyle\frac{2}{n+1}-4\sum_{j=2}^{n}\frac{1}{j}\left(2S_{j}-S_{j-1}\right)-1\,,$ $\displaystyle S_{n}$ $\displaystyle=$ $\displaystyle\sum_{j=1}^{n}\frac{1}{j}\,,$ (43) $\displaystyle h_{\mu}(p)$ $\displaystyle=$ $\displaystyle\sum_{\mu_{2},\ldots,\mu_{n}}c_{\mu\mu_{2}\ldots\mu_{n}}p_{\mu_{2}}\cdots p_{\mu_{n}}\,,$ (44) where $\xi=0$ in Landau gauge. $c_{\mu\mu_{2}\ldots\mu_{n}}=c_{\mu\nu}$ are $\delta_{\mu 4}\delta_{\nu 4}-\frac{1}{3}\sum_{k=1}^{3}\delta_{\mu k}\delta_{\nu k}$ for momentum fraction and $\delta_{\mu 3}\delta_{\nu 4}+\delta_{\mu 4}\delta_{\nu 3}$ for helicity operators. In the tensor-current case, we use the matching factor calculated from Ref. Aoki et al. (2007): $\displaystyle\frac{Z^{{\rm\overline{MS}}}}{Z^{{\rm RI}}}$ $\displaystyle=$ $\displaystyle 1+\left(\frac{\alpha_{s}}{4\pi}\right)^{2}\frac{1}{81}(-4866+1656\zeta(3)+259n_{f}).$ We use the three-loop solution for the running of $\alpha_{s}$ Gimenez et al. (1998): $\displaystyle\frac{\alpha_{s}}{4\pi}$ $\displaystyle=$ $\displaystyle\frac{1}{\beta_{0}\ln\left(\mu^{2}/\Lambda^{2}_{\rm QCD}\right)}-\frac{\beta_{1}\ln\ln\left(\mu^{2}/\Lambda^{2}_{\rm QCD}\right)}{\beta_{0}^{3}\ln^{2}\left(\mu^{2}/\Lambda^{2}_{\rm QCD}\right)}$ (46) $\displaystyle+$ $\displaystyle\frac{1}{\beta_{0}^{5}\ln^{3}\left(\mu^{2}/\Lambda^{2}_{\rm QCD}\right)}\left\\{\beta_{1}^{2}\ln^{2}\ln\left(\mu^{2}/\Lambda^{2}_{\rm QCD}\right)\right.$ $\displaystyle-$ $\displaystyle\left.\beta_{1}^{2}\ln\ln\left(\mu^{2}/\Lambda^{2}_{\rm QCD}\right)+\beta_{2}\beta_{0}-\beta_{1}^{2}\right\\}$ where $\alpha=\frac{g^{2}}{4\pi}\,$ and $C_{F}=\frac{N_{C}^{2}-1}{2N_{C}}$, $\Lambda_{\rm QCD}=300$ MeVIzubuchi (2004). The various $\beta_{i}$ are $\displaystyle\beta_{0}$ $\displaystyle=$ $\displaystyle\frac{11N_{c}-2n_{f}}{3}$ $\displaystyle\beta_{1}$ $\displaystyle=$ $\displaystyle\frac{34N_{c}^{2}}{3}-\frac{10}{3}N_{c}n_{f}-\frac{(N_{c}^{2}-1)n_{f}}{N_{c}}$ $\displaystyle\beta_{2}$ $\displaystyle=$ $\displaystyle\frac{2857}{54}N_{c}^{3}+\frac{(N_{c}^{2}-1)^{2}}{4N_{c}^{2}}n_{f}-\frac{205}{36}(N_{c}^{2}-1)n_{f}$ (47) $\displaystyle-$ $\displaystyle\frac{1415}{54}N_{c}^{2}n_{f}$ $\displaystyle+$ $\displaystyle\frac{11}{18}\frac{N_{c}^{2}-1}{N_{c}}n_{f}^{2}+\frac{79}{54}N_{c}n_{f}^{2}\quad\mbox{in $\overline{\rm MS}$ scheme},$ where $N_{c}$ is the number of colors and $n_{f}$ is the number of quark flavors. In the quenched approximation, $n_{f}=0$; in this work, $n_{f}=2$. The running of the $Z$-factors to two loops Buras (1998) is $\displaystyle C(\mu^{2})=\alpha_{s}(\mu)^{\overline{\gamma}_{0}}\left\\{1+\frac{\alpha_{s}(\mu)}{4\pi}\left(\overline{\gamma}_{1}-\overline{\beta}_{1}\overline{\gamma}_{0}\right)\right\\},$ (48) where $\overline{\gamma}_{i}$ is defined by the anomalous dimensions $\gamma_{i}$ divided by $2\beta_{0}$. The anomalous dimensions for twist-two operators of interest were calculated by Floratos et al. Floratos et al. (1977): $\displaystyle\gamma_{0}$ $\displaystyle=$ $\displaystyle 2C_{F}\left(1-\frac{2}{n(n+1)}+4\sum_{j=2}^{n}\frac{1}{j}\right)$ $\displaystyle\gamma_{1}$ $\displaystyle=$ $\displaystyle(-8.30C_{F}^{2}+27.85C_{F}C_{A}-9.48C_{F}T_{F}),$ (49) for twist-${n}$ operator. (Note that where $T_{F}(n_{f})=\frac{n_{f}}{2}$, $C_{F}=\frac{N_{c}-1}{2N_{c}}$ and $C_{A}=N_{c}$.) The anomalous dimensions for the operator, such as ${\cal O}_{34}^{\sigma q}$, can be found in Ref. Broadhurst and Grozin (1995) $\displaystyle\gamma_{0}$ $\displaystyle=$ $\displaystyle 2C_{F}$ $\displaystyle\gamma_{1}$ $\displaystyle=$ $\displaystyle\frac{C_{F}}{9}(257C_{A}-171C_{F}-52T_{F}).$ (50) ### .2 Chiral extrapolation models In this appendix, we collect some relevant formulae used in this paper from various models proposed to conduct chiral extrapolations for baryonic observables. #### .2.1 Finite-volume correction on $g_{A}$ The small-scale expansion scheme Hemmert et al. (1998) tries to use explicit degrees of freedom from the pion, nucleon and $\Delta(1232)$ resonance, expanding in terms of $\Delta_{0}$, the mass splitting between the $N$ and $\Delta$ in the chiral limit. This splitting is treated as a small parameter, of the order $O(\epsilon)$. They define $g_{A}^{0}$ and $F_{\pi}^{0}$ as the value of the axial charge $g_{A}$ and the pion decay constant $F_{\pi}$ in the chiral limit, $c_{A}$ the $N\Delta$ and $g_{1}$ the $\Delta\Delta$ axial coupling constants respectively. We simply abbreviate $\Delta_{0}$, $g_{A}^{0}$ and $F_{\pi}^{0}$ as $\Delta$, $g_{A}$ and $F_{\pi}^{0}$ in the following formula. At the leading order $O(\epsilon^{3})$, the finite-volume correction to the nucleon axial charge, $g_{A}$, was proposed in Ref. Beane and Savage (2004) within the SSE scheme: $\displaystyle\delta g_{A}$ $\displaystyle=$ $\displaystyle{m_{\pi}^{2}\over 3\pi^{2}F_{\pi}^{2}}\left[\ g_{A}^{3}{\bf F_{1}}+2c_{A}^{2}\left(g_{A}-{25\over 81}g_{1}\right){\bf F_{2}}\right.$ (51) $\displaystyle+$ $\displaystyle\left.g_{A}{\bf F_{3}}+2c_{A}^{2}g_{A}{\bf F_{4}}\right],$ where $\displaystyle{\bf F_{1}}(m,L)$ $\displaystyle=$ $\displaystyle\sum_{{\bf n}\neq{\bf 0}}\left[\ K_{0}(mL|{\bf n}|)-{K_{1}(mL|{\bf n}|)\over mL|{\bf n}|}\ \right];$ $\displaystyle{\bf F_{2}}(m,\Delta,L)$ $\displaystyle=$ $\displaystyle-\sum_{{\bf n}\neq{\bf 0}}\left[\ {K_{1}(mL|{\bf n}|)\over mL|{\bf n}|}+{\Delta^{2}-m^{2}\over m^{2}}K_{0}(mL|{\bf n}|)-\ {\Delta\over m^{2}}\int_{m}^{\infty}d\beta\;{2\beta K_{0}(\beta L|{\bf n}|)+(\Delta^{2}-m^{2})L|{\bf n}|K_{1}(\beta L|{\bf n}|)\over\sqrt{\beta^{2}+\Delta^{2}-m^{2}}}\right];$ $\displaystyle{\bf F_{3}}(m,L)$ $\displaystyle=$ $\displaystyle-{3\over 2}\sum_{{\bf n}\neq{\bf 0}}{K_{1}(mL|{\bf n}|)\over mL|{\bf n}|};$ $\displaystyle{\bf F_{4}}(m,\Delta,L)$ $\displaystyle=$ $\displaystyle{8\over 9}\sum_{{\bf n}\neq{\bf 0}}\left[\ {K_{1}(mL|{\bf n}|)\over mL|{\bf n}|}-{\pi e^{-mL|{\bf n}|}\over 2\Delta L|{\bf n}|}-{\Delta^{2}-m^{2}\over m^{2}\Delta}\int_{m}^{\infty}d\beta\ {\beta\ K_{0}(\beta L|{\bf n}|)\over\sqrt{\beta^{2}+\Delta^{2}-m^{2}}}\right],$ (52) with modified Bessel function of the second kind $K_{\alpha}(z)$. (Note that the $m$ in the above equations are short for $m_{\pi}$.) This formula can be reduced to that of LO HBChPT with $c_{A}=0$. #### .2.2 The mean-squared Dirac radius in HBChPT Heavy baryon chiral perturbation theory (HBChPT)Bernard et al. (1992) involves only the pion and nucleon, expanding in terms of the momentum $p$. At leading order (LO), once we use the physical values of $g_{A}$, $F_{\pi}$ and $\langle r_{V}^{2}\rangle$ instead of their values in the chiral limit, the HBChPT formula provides a closed form for the $m_{\pi}^{2}$ dependence of the Dirac mean-squared radius without unknown parameters Beg and Zepeda (1972); Bernard et al. (1992): $\langle r_{V}^{2}\rangle(m_{\pi,{\rm lat}})=\langle r_{V}^{2}\rangle_{\rm exp}-\frac{1+5g^{2}_{A,{\rm exp}}}{(4\pi F_{\pi,{\rm exp}})^{2}}{\rm ln}\left(\frac{m^{2}_{\pi,{\rm lat}}}{\mu^{2}}\right)$ (53) where the scale $\mu$ should be set by the physical value of the pion mass as $\mu=0.139$ GeV 333A missing factor of 2 in the prefactor of $\ln m_{\pi}^{2}$ term, which can be found in Ref. Leinweber and Cohen (1993), causes an accidental agreement between this leading one-loop form and lattice data obtained in the heavy quark mass region. . #### .2.3 Chen, et al. Chen et al. Chen and Ji (2001); Chen and Savage (2002); Detmold et al. (2002) calculate a chiral perturbation expression in the continuum for the moments of the unpolarized and helicity distributions: $\displaystyle\langle x\rangle_{u-d}$ $\displaystyle=$ $\displaystyle C\left[1-\frac{3g_{A,{\rm exp}}^{2}+1}{(4\pi F_{\pi,{\rm exp}})^{2}}m_{\pi,{\rm lat}}^{2}\ln\left(\frac{m_{\pi,{\rm lat}}^{2}}{\mu^{2}}\right)\right]+e(\mu^{2})\frac{m_{\pi,{\rm lat}}^{2}}{(4\pi F_{\pi,{\rm exp}})^{2}}$ (54) $\displaystyle\langle x\rangle_{\Delta u-\Delta d}$ $\displaystyle=$ $\displaystyle\tilde{C}\left[1-\frac{2g_{A,{\rm exp}}^{2}+1}{(4\pi F_{\pi,{\rm exp}})^{2}}m_{\pi,{\rm lat}}^{2}\ln\left(\frac{m_{\pi,{\rm lat}}^{2}}{\mu^{2}}\right)\right]+\tilde{e}(\mu^{2})\frac{m_{\pi,{\rm lat}}^{2}}{(4\pi F_{\pi,{\rm exp}})^{2}}.$ (55) We use $\mu=0.139$ MeV and $F_{\pi,{\rm exp}}=92.4$ MeV in our extrapolation. There is a variation of this formulation replacing the $F_{\pi,{\rm exp}}$ with $F_{\pi,{\rm lat}}$ and changing the scale $\mu$ to $F_{\pi,{\rm lat}}$. This provides a dimensionless quantity $\frac{m_{\pi,{\rm lat}}^{2}}{F_{\pi,{\rm lat}}^{2}}$ in the extrapolation, which is independent of the scale setting in the lattice calculation. ## References * Kaplan (1992) D. B. Kaplan, Phys. Lett. B288, 342 (1992), eprint hep-lat/9206013. * Shamir (1993) Y. Shamir, Nucl. Phys. B406, 90 (1993), eprint hep-lat/9303005. * Furman and Shamir (1995) V. Furman and Y. Shamir, Nucl. Phys. B439, 54 (1995), eprint hep-lat/9405004. * Sasaki et al. (2003) S. Sasaki, K. Orginos, S. Ohta, and T. Blum (the RIKEN-BNL-Columbia-KEK), Phys. Rev. D68, 054509 (2003), eprint hep-lat/0306007. * Yao et al. (2006) W. M. Yao et al. (Particle Data Group), J. Phys. G33, 1 (2006). * Sasaki and Yamazaki (2007) S. Sasaki and T. Yamazaki (2007), eprint arXiv:0709.3150 [hep-lat]. * Hagler et al. (2007) P. Hagler et al. (LHPC) (2007), eprint arXiv:0705.4295 [hep-lat]. * Yamazaki and Ohta (2007) T. Yamazaki and S. Ohta (RBC and UKQCD), PoS LAT2007, 165 (2007), eprint arXiv:0710.0422 [hep-lat]. * Liu et al. (1995) K. F. Liu, S. J. Dong, T. Draper, and W. Wilcox, Phys. Rev. Lett. 74, 2172 (1995), eprint hep-lat/9406007. * Gockeler et al. (2005) M. Gockeler et al. (QCDSF), Phys. Rev. D71, 034508 (2005), eprint hep-lat/0303019. * Alexandrou et al. (2006) C. Alexandrou, G. Koutsou, J. W. Negele, and A. Tsapalis, Phys. Rev. D74, 034508 (2006), eprint hep-lat/0605017. * Alexandrou et al. (2007) C. Alexandrou, G. Koutsou, T. Leontiou, J. W. Negele, and A. Tsapalis (2007), eprint arXiv:0706.3011 [hep-lat]. * Gockeler et al. (2007a) M. Gockeler et al. (QCDSF/UKQCD), PoS LAT2007, 161 (2007a), eprint arXiv:0710.2159 [hep-lat]. * Kaplan (1993) D. B. Kaplan, Nucl. Phys. Proc. Suppl. 30, 597 (1993). * Fukugita et al. (1995) M. Fukugita, Y. Kuramashi, M. Okawa, and A. Ukawa, Phys. Rev. Lett. 75, 2092 (1995), eprint [http://arXiv.org/abs]hep-lat/9501010. * Liu et al. (1994) K. F. Liu, S. J. Dong, T. Draper, J. M. Wu, and W. Wilcox, Phys. Rev. D49, 4755 (1994), eprint [http://arXiv.org/abs]hep-lat/9305025. * Gockeler et al. (1996) M. Gockeler et al., Phys. Rev. D53, 2317 (1996), eprint [http://arXiv.org/abs]hep-lat/9508004. * Takaishi (1996) T. Takaishi, Phys. Rev. D54, 1050 (1996). * de Forcrand et al. (2000) P. de Forcrand et al. (QCD-TARO), Nucl. Phys. B577, 263 (2000), eprint arXiv:hep-lat/9911033. * Andreev et al. (2007) V. A. Andreev et al. (2007), eprint arXiv:0704.2072 [nucl-ex]. * Bernard et al. (2002) V. Bernard, L. Elouadrhiri, and U. G. Meissner, J. Phys. G28, R1 (2002), eprint hep-ph/0107088. * Gorringe and Fearing (2004) T. Gorringe and H. W. Fearing, Rev. Mod. Phys. 76, 31 (2004), eprint nucl-th/0206039. * Jonkmans et al. (1996) G. Jonkmans et al., Phys. Rev. Lett. 77, 4512 (1996), eprint nucl-ex/9608005. * Breidenbach et al. (1969) M. Breidenbach et al., Phys. Rev. Lett. 23, 935 (1969). * Friedman (1991) J. I. Friedman, Rev. Mod. Phys. 63, 615 (1991). * Kendall (1991) H. W. Kendall, Rev. Mod. Phys. 63, 597 (1991). * Taylor (1991) R. E. Taylor, Rev. Mod. Phys. 63, 573 (1991). * Gluck et al. (1996) M. Gluck, E. Reya, M. Stratmann, and W. Vogelsang, Phys. Rev. D53, 4775 (1996), eprint hep-ph/9508347. * Gehrmann and Stirling (1996) T. Gehrmann and W. J. Stirling, Phys. Rev. D53, 6100 (1996), eprint hep-ph/9512406. * Lai et al. (1997) H. L. Lai et al., Phys. Rev. D55, 1280 (1997), eprint hep-ph/9606399. * Adams et al. (1997) D. Adams et al. (Spin Muon (SMC)), Phys. Rev. D56, 5330 (1997), eprint hep-ex/9702005. * Adeva et al. (1998) B. Adeva et al. (Spin Muon), Phys. Lett. B420, 180 (1998), eprint hep-ex/9711008. * Gluck et al. (1998) M. Gluck, E. Reya, and A. Vogt, Eur. Phys. J. C5, 461 (1998), eprint hep-ph/9806404. * Ackerstaff et al. (1999) K. Ackerstaff et al. (HERMES), Phys. Lett. B464, 123 (1999), eprint hep-ex/9906035. * Martin et al. (2002) A. D. Martin, R. G. Roberts, W. J. Stirling, and R. S. Thorne, Eur. Phys. J. C23, 73 (2002), eprint hep-ph/0110215. * Orginos et al. (2006) K. Orginos, T. Blum, and S. Ohta, Phys. Rev. D73, 094503 (2006), eprint hep-lat/0505024. * Gockeler et al. (1999a) M. Gockeler et al. (1999a), eprint [http://arXiv.org/abs]hep-ph/9909253. * Dolgov et al. (2002) D. Dolgov et al. (LHPC), Phys. Rev. D66, 034506 (2002), eprint [http://arXiv.org/abs]hep-lat/0201021. * Gockeler et al. (2002) M. Gockeler, R. Horsley, D. Pleiter, P. E. L. Rakow, and G. Schierholz, Nucl. Phys. Proc. Suppl. (Lattice 2002) (2002), eprint [http://arXiv.org/abs]hep-lat/0209151. * Orginos (2002) K. Orginos (RBC), Nucl. Phys. Proc. Suppl. (Lattice 2002) (2002), eprint [http://arXiv.org/abs]hep-lat/0209137. * Hagler et al. (2003) P. Hagler et al. (LHPC), Phys. Rev. D68, 034505 (2003), eprint hep-lat/0304018. * Ohta and Orginos (2004) S. Ohta and K. Orginos (RBCK), Nucl. Phys. Proc. Suppl. 129, 296 (2004), eprint hep-lat/0309113. * Ohta and Orginos (2005) S. Ohta and K. Orginos (RBCK), Nucl. Phys. Proc. Suppl. 140, 396 (2005), eprint hep-lat/0411008. * LHPC et al. (2004) LHPC et al. (LHPC), Phys. Rev. Lett. 93, 112001 (2004), eprint hep-lat/0312014. * Gockeler et al. (2004a) M. Gockeler et al. (QCDSF), Phys. Rev. Lett. 92, 042002 (2004a), eprint hep-ph/0304249. * Khan et al. (2004) A. A. Khan et al. (2004), eprint hep-lat/0409161. * Detmold et al. (2001) W. Detmold, W. Melnitchouk, J. W. Negele, D. B. Renner, and A. W. Thomas, Phys. Rev. Lett. 87, 172001 (2001), eprint hep-lat/0103006. * Gockeler et al. (2004b) M. Gockeler, R. Horsley, D. Pleiter, P. E. L. Rakow, and G. Schierholz (QCDSF) (2004b), eprint hep-ph/0410187. * Lin and Ohta (2006) H.-W. Lin and S. Ohta, PoS LAT2006, 118 (2006), eprint hep-lat/0610028. * Edwards et al. (2006) R. G. Edwards et al. (2006), eprint hep-lat/0610007. * Gockeler et al. (2007b) M. Gockeler et al., Eur. Phys. J. A32, 445 (2007b), eprint hep-lat/0609001. * Schroers (2007) W. Schroers, Eur. Phys. J. A31, 784 (2007), eprint hep-lat/0701003. * Lin (2007a) H.-W. Lin (RBC and UKQCD), AIP Conf. Proc. 915, 498 (2007a). * Lin (2007b) H.-W. Lin (2007b), eprint arXiv:0707.3844 [hep-lat]. * Orginos (2006) K. Orginos, PoS LAT2006, 018 (2006). * Hagler (2007) P. Hagler, PoS LAT2007, 013 (2007), eprint arXiv:0711.0819 [hep-lat]. * Aoki et al. (2005) Y. Aoki et al., Phys. Rev. D72, 114505 (2005), eprint hep-lat/0411006. * Berruto et al. (2006) F. Berruto, T. Blum, K. Orginos, and A. Soni, Phys. Rev. D73, 054509 (2006), eprint hep-lat/0512004. * Sasaki et al. (2002) S. Sasaki, T. Blum, and S. Ohta, Phys. Rev. D65, 074503 (2002), eprint hep-lat/0102010. * Orginos (2003) K. Orginos (RBC), AIP Conf. Proc. 675, 166 (2003), eprint hep-lat/0211025. * Wandzura and Wilczek (1977) S. Wandzura and F. Wilczek, Phys. Lett. B72, 195 (1977). * Blum et al. (2003) T. Blum et al. (RBC), Phys. Rev. D68, 114506 (2003), eprint hep-lat/0110075. * Blum et al. (2002) T. Blum et al., Phys. Rev. D66, 014504 (2002), eprint hep-lat/0102005. * Dawson (2003) C. Dawson (RBC), Nucl. Phys. Proc. Suppl. 119, 314 (2003), eprint hep-lat/0210005. * Aoki et al. (2004) Y. Aoki et al., Phys. Rev. D69, 074504 (2004), eprint hep-lat/0211023. * Martinelli et al. (1995) G. Martinelli, C. Pittori, C. T. Sachrajda, M. Testa, and A. Vladikas, Nucl. Phys. B445, 81 (1995), eprint hep-lat/9411010. * Gimenez et al. (1998) V. Gimenez, L. Giusti, F. Rapuano, and M. Talevi, Nucl. Phys. B531, 429 (1998), eprint hep-lat/9806006. * Gockeler et al. (1999b) M. Gockeler et al., Nucl. Phys. B544, 699 (1999b), eprint hep-lat/9807044. * Floratos et al. (1977) E. G. Floratos, D. A. Ross, and C. T. Sachrajda, Nucl. Phys. B129, 66 (1977), errarum, ibid., B139, 545-546 (1978). * Izubuchi (2004) T. Izubuchi (RBC), Nucl. Phys. Proc. Suppl. 129, 266 (2004), eprint hep-lat/0310058. * Beane and Savage (2004) S. R. Beane and M. J. Savage, Phys. Rev. D70, 074029 (2004), eprint hep-ph/0404131. * Hemmert et al. (1998) T. R. Hemmert, B. R. Holstein, and J. Kambor, J. Phys. G24, 1831 (1998), eprint hep-ph/9712496. * Procura et al. (2007) M. Procura, B. U. Musch, T. R. Hemmert, and W. Weise, Phys. Rev. D75, 014503 (2007), eprint hep-lat/0610105. * Khan et al. (2006) A. A. Khan et al., Phys. Rev. D74, 094508 (2006), eprint hep-lat/0603028. * Beg and Zepeda (1972) M. A. B. Beg and A. Zepeda, Phys. Rev. D6, 2912 (1972). * Bernard and Meissner (2006) V. Bernard and U.-G. Meissner, Phys. Lett. B639, 278 (2006), eprint hep-lat/0605010. * Czarnecki et al. (2007) A. Czarnecki, W. J. Marciano, and A. Sirlin (2007), eprint arXiv:0704.3968 [hep-ph]. * Clark et al. (2006) J. H. D. Clark et al., Phys. Rev. Lett. 96, 073401 (2006), eprint nucl-ex/0509025. * Choi et al. (1993) S. Choi et al., Phys. Rev. Lett. 71, 3927 (1993). * Braun et al. (2006) V. M. Braun, A. Lenz, and M. Wittmann, Phys. Rev. D73, 094019 (2006), eprint hep-ph/0604050. * Goldberger and Treiman (1958) M. L. Goldberger and S. B. Treiman, Phys. Rev. 110, 1178 (1958). * Arndt et al. (1994) R. A. Arndt, R. L. Workman, and M. M. Pavan, Phys. Rev. C49, 2729 (1994). * Bugg (2004) D. V. Bugg, Eur. Phys. J. C33, 505 (2004). * Coon and Scadron (1981) S. A. Coon and M. D. Scadron, Phys. Rev. C23, 1150 (1981). * Antonio et al. (2007) D. J. Antonio et al. (RBC and UKQCD), Phys. Rev. D75, 114501 (2007), eprint hep-lat/0612005. * Allton et al. (2007) C. Allton et al. (RBC) (2007), eprint hep-lat/0701013. * Chen and Savage (2002) J.-W. Chen and M. J. Savage, Nucl. Phys. A707, 452 (2002), eprint nucl-th/0108042. * Chen and Ji (2001) J.-W. Chen and X.-d. Ji, Phys. Lett. B523, 107 (2001), eprint hep-ph/0105197. * Detmold et al. (2002) W. Detmold, W. Melnitchouk, and A. W. Thomas, Phys. Rev. D66, 054501 (2002), eprint hep-lat/0206001. * Beane et al. (2006) S. R. Beane, P. F. Bedaque, K. Orginos, and M. J. Savage (NPLQCD), Phys. Rev. D73, 054503 (2006), eprint hep-lat/0506013. * Jaffe and Ji (1993) R. L. Jaffe and X.-D. Ji, Phys. Rev. Lett. 71, 2547 (1993), eprint hep-ph/9307329. * Grosse Perdekamp (2002) M. Grosse Perdekamp, Nucl. Phys. Proc. Suppl. 105, 71 (2002). * Yamazaki et al. (2008) T. Yamazaki et al. (2008), eprint arXiv:0801.4016 [hep-lat]. * Aoki et al. (2007) Y. Aoki et al. (2007), eprint arXiv:0712.1061 [hep-lat]. * Buras (1998) A. J. Buras (1998), eprint hep-ph/9806471. * Broadhurst and Grozin (1995) D. J. Broadhurst and A. G. Grozin, Phys. Rev. D52, 4082 (1995), eprint hep-ph/9410240. * Bernard et al. (1992) V. Bernard, N. Kaiser, J. Kambor, and U. G. Meissner, Nucl. Phys. B388, 315 (1992). * Leinweber and Cohen (1993) D. B. Leinweber and T. D. Cohen, Phys. Rev. D47, 2147 (1993).

arxiv-papers

2024-09-04T02:48:53.615954

0802.0920

# Fermi Surface Effect on Lorenz Number of Correlated Metal Takuya Okabe Faculty of Engineering, Shizuoka University, 3-5-1 Johoku, Hamamatsu 432-8561,Japan ###### Abstract We investigate an effect that an ideal Lorenz number $L_{\rm i}$ of correlated metal shows peculiar Fermi surface dependence, which is caused by the onset of a particular channel of Umklapp scattering. We evaluate $L_{\rm i}$ for some simple models and transition metals, and note that $L_{\rm i}$ for NaxCoO2 decreases sensitively as $x$ approaches an Umklapp threshold around $x_{c}\simeq 0.6$. ###### pacs: 74.25.Fy, 71.10.Ay, 71.10.Fd, 71.20.Be ## I Introduction The discovery of large thermopower in NaxCoO2Terasaki et al. (1997) has prompted quest for related good thermoelectric (TE) materials and encouraged experimental and theoretical research on TE properties of strongly correlated electron systems. In practical TE application, the important material parameter is the dimensionless figure of merit $ZT=S^{2}/(\kappa/\sigma T)$, where $S$ is the Seebeck coefficient, $\kappa$ the thermal conductivity, $\sigma$ the electrical conductivity, and $T$ the absolute temperature. Hence it is equally as important to enhance $S$ as it is to decrease the Lorenz number $L$ defined by $L=\kappa/\sigma T$Mahan (1998). The latter however is usually hampered by the Wiedemann-Franz (WF) law, according to which $L$ should be a universal constant, e.g., $L=\frac{\pi^{2}}{3}(k/e)^{2}$ for impure metals. In fact, normal metals, which generally have low $S$, will not be good TE devices with $ZT>1$, unless the WF law is overcome. It would thus be interesting to investigate a possible material dependence of $L$ to see if it is a controllable variable in principle. We discuss the ideal value $L_{\rm i}$ (which is simply denoted as $L$ below) of a correlated electron system on a rigid lattice without impurity in the low temperature limit for the purpose of elucidating its material specific dependence. Indeed, for pure transition metals, the Lorenz numbers have been observed to vary from metal to metalSchriempf (1968); White and Tainsh (1967); Anderson et al. (1968); Beitchman et al. (1970); Wagner et al. (1971). Theoretically, Herring had derived a universal constantHerring (1967), while the problem had also been addressed specifically by treating $s$-$d$ hybridization as perturbation, namely, by a two band model of conductive $s$ electrons scattering off localized $d$ statesRice (1968); Schriempf et al. (1969). Herring’s argument based on the Fermi liquid theory should be the proper approach to the problem at low temperatures. We derive a dimensionless factor (Eq. (10)) so as to modify his constant result, with which we investigate the Fermi surface (FS) dependence of $L$ for some typical cases concretely. After showing nontrivial results obtained for simple models, we give results of numerical evaluation of the Lorenz numbers of some transition metals. Lastly, we find it interesting to investigate a model of the sodium cobalt oxide NaxCoO2, because a nontrivial $x$ dependence of $L$ is expected theoretically owing to its simple cylindrical FSSingh (2000). ## II Fermi Liquid theory We begin with the linearized transport equation of a Fermi liquid under the temperature gradient $\nabla T$ and the electric field ${\bf E}$, $\frac{\partial n_{p}}{\partial\varepsilon_{p}}\left(-\frac{\varepsilon_{p}-\mu}{T}\nabla T\cdot{\bf v}_{p}+e{\bf E}\cdot{\bf v}_{p}\right)=I[\varphi],$ (1) $\displaystyle TI[\varphi]$ $\displaystyle=$ $\displaystyle\sum_{p^{\prime},k}W^{pp^{\prime}}_{k}n_{p}n_{p^{\prime}}(1-n_{p-k})(1-n_{p^{\prime}+k})(\varphi_{p}+\varphi_{p^{\prime}}$ $\displaystyle-\varphi_{p-k}-\varphi_{p^{\prime}+k})\delta(\varepsilon_{p}+\varepsilon_{p^{\prime}}-\varepsilon_{p-k}-\varepsilon_{p^{\prime}+k}),$ where $n_{p}=n(\varepsilon_{p})=1/({\rm e}^{(\varepsilon_{p}-\mu)/T}+1)$ is the Fermi distribution function, and $W^{pp^{\prime}}_{k}$ represents the transition probability of quasiparticle scattering; $p,p^{\prime}\rightarrow p-k,p^{\prime}+k$. In terms of the solution $\varphi_{p}$ of Eq. (1), the electric and heat currents carried by quasiparticles are respectively given by ${\bf J}=2e\sum_{p}{\bf v}_{p}\frac{\partial n}{\partial\varepsilon_{p}}\varphi_{p},\quad{\bf Q}=2\sum_{p}{\bf v}_{p}(\varepsilon_{p}-\mu)\frac{\partial n}{\partial\varepsilon_{p}}\varphi_{p},$ where $\varepsilon_{p}$ and ${\bf v}_{p}$ are energy and velocity of quasiparticles, respectively. In comparison with the phenomenological formulae ${\bf J}=\sigma{\bf E}-\sigma S\nabla T$ and ${\bf Q}=T\sigma S{\bf E}-\kappa_{0}\nabla T,$ we obtain the transport coefficients $\sigma$, $S$, and $\kappa_{0}$, the thermal conductivity at zero electric field. The thermal conductivity at zero current, $\kappa$, is given by $\kappa=\kappa_{0}-T\sigma S^{2}$Mahan (1998). Formally, the above results expressed in terms of the renormalized quantities $\varepsilon_{p}$ and ${\bf v}_{p}$ bear resemblance to those of a weakly interacting Fermi gas. Nevertheless, many body effects are included not only in the renormalization for the individual quasiparticle, but also in the field induced shift of the quasiparticle distribution $\varphi_{p}$, which implicitly includes a collective Fermi liquid effect depending on Landau parametersPines and Nozières (1966); Okabe (1998). We do not write down the explicit expressions for them as they are irrelevant for our purposes in what follows. For definiteness, let us assume that the currents ${\bf J}$ and ${\bf Q}$ flow in the $x$-direction. Then, for the two functions $l_{1}$ and $l_{2}$ defined by $\varphi_{p}=-eE_{x}l_{1}+{\partial_{x}T}l_{2}$, we obtain the equations $-\frac{\partial n_{p}}{\partial\varepsilon_{p}}{v}_{px}=I[l_{1}],$ (2) and $-\frac{\partial n_{p}}{\partial\varepsilon_{p}}t_{p}{v}_{px}=I[l_{2}],$ (3) where $t_{p}=(\varepsilon_{p}-\mu)/T$. For an isotropic system, the collision integral $I$ has been evaluated analytically by an elaborate approximationAbrikosov and Khalatnikov (1959). As we cannot assume the predominance of normal scattering processes in general cases of anisotropic Fermi liquids of our concern, we have to make approximations in another way. To simplify the multiple momentum sum in $I$, we replace $(1-n(\varepsilon_{p^{\prime}+k}))\delta(\varepsilon_{p}+\varepsilon_{p^{\prime}}-\varepsilon_{p-k}-\varepsilon_{p^{\prime}+k})$ with $(1-n(\varepsilon_{p}+\varepsilon_{p^{\prime}}-\varepsilon_{p-k}))\delta(\varepsilon_{p^{\prime}+k}-\mu),$ for significant contributions to the collision term should come from a thermal neighborhood of the FS in any case. Moreover, as in the isotropic case, to describe the momentum dependence of the solutions $l_{i}$, we decouple the crystal momentum variable ${\bf p}$ into the radial (energy) direction $t_{p}=(\varepsilon_{p}-\mu)/T$ and the perpendicular component $\bar{\Omega}_{p}$, and set $l_{i,p}=M_{i}(\bar{\Omega}_{p})N_{i}(t_{p})$ ($i=1,2$). Then the momentum sum is written as $\sum_{p}=T\int\rho{\rm d}t\int{\rm d}\bar{\Omega}_{p},$ where $\rho$ represents the density of quasiparticle states (DOS) under the normalization condition $\int{\rm d}\bar{\Omega}_{p}=1$. By integrating over the energy variables, we obtain equations for $M_{i}(\bar{\Omega}_{p})$. First we note that we can reproduce the same conductivity formula as derived and discussed previouslyYamada and Yosida (1986); Okabe (1998). By setting $N_{1}(t_{p})$ as a constant, the energy integrals which appear in the four terms in $I[l_{1}]$ of Eq. (2) can be calculated analytically. As a result, we obtain $\displaystyle v_{px}$ $\displaystyle=$ $\displaystyle\frac{\pi^{2}}{2}\rho^{2}T^{2}\int{{\rm d}\bar{\Omega}_{p^{\prime}}}\int{{\rm d}\bar{\Omega}_{p^{\prime\prime}}}W^{pp^{\prime}}_{p-p^{\prime\prime}}({M}_{1,p}+{M}_{1,p^{\prime}}-{M}_{1,p^{\prime\prime}}-{M}_{1,p+p^{\prime}-p^{\prime\prime}})\delta(\varepsilon_{p+p^{\prime}-p^{\prime\prime}}-\mu)$ (4) $\displaystyle=$ $\displaystyle\frac{\pi^{2}T^{2}}{2}\sum_{p^{\prime},p^{\prime\prime}}W^{pp^{\prime}}_{p-p^{\prime\prime}}({M}_{1,p}+{M}_{1,p^{\prime}}-{M}_{1,p^{\prime\prime}}-{M}_{1,p+p^{\prime}-p^{\prime\prime}})\rho_{p^{\prime}}\rho_{p^{\prime\prime}}\rho_{p+p^{\prime}-p^{\prime\prime}},$ and $\sigma={2e^{2}}\rho\int{v}_{px}M_{1}(\bar{\Omega}_{p}){{\rm d}\bar{\Omega}_{p}}.$ (5) In Eq. (4), we used $\rho_{p}=\delta(\varepsilon_{p}-\mu)$. Assuming the angular dependence ${M}_{1,p}\propto v_{px}$, we obtain the resistivity coefficient $A=\frac{\pi^{2}\sum_{p,p^{\prime},p^{\prime\prime}}W^{pp^{\prime}}_{p-p^{\prime\prime}}\rho_{p}\rho_{p^{\prime}}\rho_{p^{\prime\prime}}\rho_{p+p^{\prime}-p^{\prime\prime}}v_{p_{x}}({v}_{p{x}}+{v}_{p^{\prime}{x}}-{v}_{p^{\prime\prime}{x}}-{v}_{p+p^{\prime}-p^{\prime\prime}x})}{4e^{2}\left(\sum_{p}\rho_{p}{v}_{px}^{2}\right)^{2}}$ (6) for the electrical resistivity $\sigma^{-1}=AT^{2}$. On the other hand, we have to make a further approximation for $\kappa_{0}$. By taking the $t$-derivative at $t_{p}=0$ of Eq. (3), while adopting $N_{2}(t)\propto t/(t^{2}+\pi^{2})$Herring (1967), we finally obtain similar equations, namely, $\displaystyle v_{px}$ $\displaystyle=$ $\displaystyle\rho^{2}T^{2}\int{{\rm d}\bar{\Omega}_{p^{\prime}}}\int{{\rm d}\bar{\Omega}_{p^{\prime\prime}}}W^{pp^{\prime}}_{p-p^{\prime\prime}}({M}_{2,p}-c({M}_{2,p^{\prime}}$ (7) $\displaystyle+{M}_{2,p^{\prime\prime}}+{M}_{2,p+p^{\prime}-p^{\prime\prime}}))\delta(\varepsilon_{p+p^{\prime}-p^{\prime\prime}}-\mu),$ where $c=\int_{0}^{\infty}{\rm d}t\left({t}\coth\frac{t}{2}-2\right)\frac{2t/(t^{2}+\pi^{2})}{\sinh{t}}\simeq 0.162,$ and $\kappa_{0}=2\pi^{2}\rho T\int_{-\infty}^{\infty}\frac{\partial n}{\partial t}\frac{t^{2}{\rm d}t}{t^{2}+\pi^{2}}\int{v}_{px}M_{2}(\bar{\Omega}_{p}){{\rm d}\bar{\Omega}_{p}},$ from which we obtain the thermal resistivity $\kappa_{0}^{-1}=BT$. It is remarked that the difference in the integrand kernels of Eqs. (4) and (7) stems from the fact that $N_{2}(t)$ for $\kappa_{0}$ is an odd function, while $N_{1}(t)$ for $\sigma$ is even. To evaluate the Lorenz number $L={\kappa}/{\sigma T}$ concretely, assuming $M_{2,p}\propto v_{px}$ as above, we obtain $L<L_{0}\equiv\frac{\kappa_{0}}{\sigma T}=\frac{\pi^{2}}{12}(12-\pi^{2}){\cal L}\left(\frac{k}{e}\right)^{2},$ (8) where the Boltzmann constant $k$ is written explicitly111 In the same manner, we get $\sigma S=\frac{2\pi^{2}e}{3}T\int\frac{{\rm d}(\rho v_{px})}{{\rm d}\varepsilon}M_{1}{\rm d}\bar{\Omega}_{p}.$ For the parabolic band, we obtain $S=\frac{2\pi^{2}}{3e}\frac{\rho^{\prime}}{\rho}T=\frac{\pi^{2}}{3e}\frac{T}{\varepsilon_{F}}$, the same result as that for free electrons.. The constant prefactor in Eq. (8) is separated as it comes from the energy integral, and corresponds to the result first derived by HerringHerring (1967), though our result differs by $\pi^{2}/12$. In addition, the dimensionless factor ${\cal L}$ in Eq. (8) originates from the directional dependence of scattering, and is given explicitly by $\displaystyle{\cal L}=\frac{A}{B}$ $\displaystyle=$ $\displaystyle\frac{\displaystyle\int{{\rm d}\bar{\Omega}_{1}}\int{{\rm d}\bar{\Omega}_{2}}\int{{\rm d}\bar{\Omega}_{3}}W^{p_{1}p_{2}}_{p_{1}-p_{3}}v_{p_{1}x}({v}_{p_{1}x}+{v}_{p_{2}x}-{v}_{p_{3}x}-{v}_{p_{1}+p_{2}-p_{3},x})\delta(\varepsilon_{p_{1}+p_{2}-p_{3}}-\mu)}{\displaystyle\int{{\rm d}\bar{\Omega}_{1}}\int{{\rm d}\bar{\Omega}_{2}}\int{{\rm d}\bar{\Omega}_{3}}W^{p_{1}p_{2}}_{p_{1}-p_{3}}v_{p_{1x}}({v}_{p_{1}x}-c\left({v}_{p_{2}x}+{v}_{p_{3}x}+{v}_{p_{1}+p_{2}-p_{3},x}\right))\delta(\varepsilon_{p_{1}+p_{2}-p_{3}}-\mu)}$ (9) $\displaystyle=$ $\displaystyle\frac{\displaystyle\sum_{1,2,3}W^{p_{1}p_{2}}_{p_{1}-p_{3}}\rho_{p_{1}}\rho_{p_{2}}\rho_{p_{3}}\rho_{p_{1}+p_{2}-p_{3}}v_{p_{1}x}\left({v}_{p_{1}x}+{v}_{p_{2}x}-{v}_{p_{3}x}-{v}_{p_{1}+p_{2}-p_{3},x}\right)}{\displaystyle\sum_{1,2,3}W^{p_{1}p_{2}}_{p_{1}-p_{3}}\rho_{p_{1}}\rho_{p_{2}}\rho_{p_{3}}\rho_{p_{1}+p_{2}-p_{3}}v_{p_{1}x}({v}_{p_{1}x}-c\left({v}_{p_{2}x}+{v}_{p_{3}x}+{v}_{p_{1}+p_{2}-p_{3},x}\right))}.$ (10) Note that one would find a trivial result ${\cal L}=1$, a constant Lorenz ratio, if one neglects both of the three vertex correction terms following ${v}_{p_{1}x}$ in the parentheses of the denominator as well as the numerator of Eq. (9). To have a nontrivial effect ${\cal L}\neq 1$, it is essential not to disregard detailed momentum structure of the relevant quasiparticle scatterings on the FS. In fact, ${\cal L}$ quantifies the effect of FS geometry on the availability of phase space for the quasiparticle scatterings to relax the transport currents. Essentially, the above expression represents the fact that thermal and electrical resistivities are determined mainly by different types of scattering processes, that is, normal processes are important for the thermal resistivity $B$, while the electrical $A$ is caused by Umklapp processes. In fact, for the numerator, or the resistivity coefficient $A$, to take a finite value, there must exist at least a set of four momenta ${\bf p}_{i}$ ($i=1,2,3,4$) on the FS satisfying the Umklapp condition ${\bf p}_{1}+{\bf p}_{2}={\bf p}_{3}+{\bf p}_{4}+{\bf G}$, where ${\bf G}(\neq 0)$ is a reciprocal lattice vector. This is not met if the FS is too small. Indeed, there is a limit Fermi radius estimated by the relation $|{\bf p}_{i}|=|{\bf G}|/4$ which holds at the threshold where all the vectors are parallel or antiparallel. In the vicinity of the threshold, one should expect a strong FS dependence of ${\cal L}$. This is a matter of our concern in the following. ## III results ### III.1 Simple Model The factor ${\cal L}$ may be sensitive to the momentum dependence of the scattering probability $W^{pp^{\prime}}_{k}=\frac{2\pi}{\hbar}\left(|A_{\uparrow\downarrow,k}^{pp^{\prime}}|^{2}+\frac{1}{2}|A_{\uparrow\uparrow,k}^{pp^{\prime}}|^{2}\right),$ where $A_{\uparrow\uparrow,k}^{pp^{\prime}}$ and $A_{\uparrow\downarrow,k}^{pp^{\prime}}$ are the scattering amplitudes for quasiparticles with parallel and antiparallel spins. In particular, the momentum dependence can give rise to a conspicuous effect in the vicinity of quantum critical points where there are quantum fluctuations localized in $k$ space. To take this into account, we may assume $A_{\uparrow\uparrow,k}^{pp^{\prime}}\propto A_{\uparrow\downarrow,k}^{pp^{\prime}}\propto A_{k}^{pp^{\prime}}=\frac{1}{{1+\xi^{2}\gamma({{\bf k}-{\bf Q}})}}$ (11) in order to represent the effect of fluctuations peaked around ${\bf k}={\bf Q}$. In addition to the correlation length $\xi$, one may adopt $\gamma({\bf k})\propto\sum_{{\bf d}}\left(1-{\rm e}^{{\rm i}{\bf k}\cdot{\bf d}}\right)$, where the sum is taken over the nearest neighboring lattice vectors ${\bf d}$, as it gives a simple lattice periodic function to give $\gamma({\bf k})\simeq k^{2}$ in the long wavelength limit $k\rightarrow 0$. To illustrate how ${\cal L}$ varies, let us first investigate a simple two- dimensional (2D) model. We should make a special remark, however. In a strictly 2D system, one will find that normal forward scatterings make the denominator $B$ of Eq. (9) logarithmically divergent, as so for the inverse lifetime of quasiparticle which corresponds to the denominator of Eq. (9) with $c=0$Hodges et al. (1971). On the other side, the numerator $A$ still remains finite as noted by Fujimoto $et$ $al.$Fujimoto et al. (1991). In effect, the divergence is suppressed by a small decay rate $\Gamma$ assigned to the quasiparticle states, $\rho_{p}=\delta(\varepsilon_{p}-\mu)$. In a real system, such a cutoff must be provided by an inevitable effect of three dimensionality of the system or by a finite density of impurities. As the dependence on $\Gamma$ is logarithmic and weak numerically, here we present a typical behaviour assuming $\Gamma$ as a given constant for simplicity. Figure 1: For a parabolic band on a square lattice, the dimensionless Lorenz factor ${\cal L}$ in Eq. (8) is shown as a function of the filling $n$ per orbital. For $\xi=0$ (solid line), for $\xi=1$ and ${\bf Q}=(\pi,\pi)$ (dashed line, AFM), and for $\xi=1$ and ${\bf Q}=(0,0)$ (dotted line, FM). In all cases, ${\cal L}$ must vanish at the threshold $n_{c,1}\simeq 0.2$. First we discuss the simplest case of a parabolic band on a square lattice. It is easy to show that the result is independent of the quasiparticle mass, so that we may simply use $\varepsilon_{p}=p^{2}$. We obtain Fig. 1 for $\Gamma=0.02$, which representatively shows ${\cal L}$ as a function of the electron number $n$ per orbital for three types of $W^{pp^{\prime}}_{k}$ according to Eq. (11). As expected, characteristic anomalies are clearly observed. The solid line for $\xi=0$ ($W^{pp^{\prime}}_{k}=$const.) typically shows the first onset of the Umklapp processes involving the smallest reciprocal lattice vector ${\bf G}=(2\pi,0)$ at $n_{c,1}=\pi/16\simeq 0.20$, as well as the second one at $n_{c,2}\simeq 0.39$ for ${\bf G}=(2\pi,2\pi)$. It is noted that the threshold fillings can be easily evaluated as they are geometrically determined by the given FS. The dashed line indicates that the latter structure for $n>n_{c,2}$ is particularly emphasized by the commensurate antiferromagnetic (AFM) fluctuations with ${\bf Q}=(\pi,\pi)$. On the other hand, the ferromagnetic (FM) fluctuations with ${\bf Q}=(0,0)$ generally suppress ${\cal L}$, as they strengthen the relative weight of the normal processes contributing to the thermal resistivity $B$. Figure 2: ${\cal L}(\equiv A/B)$ of a parabolic band on a sc lattice and a fcc lattice. In the inset, the resistivity coefficients $A\propto\sigma^{-1}/T^{2}$ and $B$ for the sc lattice are shown as a function of $n$. Similarly, one may obtain results for three dimensional systems, in which no cutoff is required. In Fig. 2, we observe the lattice-structure dependence of ${\cal L}$ for the parabolic band in a simple cubic (sc) lattice and a face centered cubic (fcc) lattice with $W_{k}^{pp^{\prime}}=$const. For the sc lattice the first threshold lies at $n_{c,1}\simeq 0.065$ due to ${\bf G}=(2\pi,0,0)$, while it is at $n_{c,1}\simeq 0.085$ with ${\bf G}=(2\pi,2\pi,-2\pi)$ for the fcc lattice. For the latter, the secondary kinks expected at $n_{c,2}\simeq 0.13$ and $n_{c,3}\simeq 0.37$ corresponding to ${\bf G}=(4\pi,4\pi,0)$ and $(4\pi,0,0)$, respectively, are not so conspicuous as that found clearly at $n_{c,2}\simeq 0.19$ due to ${\bf G}=(2\pi,2\pi,0)$ for the sc lattice. In the inset of Fig. 2, the resistivity coefficients $A$ and $B$ are shown respectively for the sc lattice. We remark that the presence of the threshold $n_{c,1}$ may be more easily anticipated from a relatively gradual $n$-dependence of ${\cal L}$ than from the electrical resistivity coefficient $A$, which drops abruptly at $n_{c,1}$. The results exemplify that the ideal Lorenz ratio is not a constant number but shows the lattice structure dependence interestingly. ### III.2 Transition metals Table 1: Calculated values for the ideal Lorenz ratio are compared with experiments. | ${\cal L}$ | $L_{0}$ ($10^{-8}$V2/K2) | $L_{\rm exp}$ ($10^{-8}$V2/K2) ---|---|---|--- Pd ($\xi=0$) | 1.0 | 1.3 | Pd ($\xi=5$Å) | 0.60 | 0.78 | 1.1 Schriempf (1968) Ni | 1.1 | 1.4 | 1.0 White and Tainsh (1967) Pt | 1.0 | 1.3 | 0.1 Anderson et al. (1968) Fe | 0.87 | 1.1 | 1.1 Beitchman et al. (1970) W | 0.61 | 0.79 | 0.2-0.4 Wagner et al. (1971) In principle, we can evaluate ${\cal L}$ for real materials though it would generally require a hard task numerically. Let us evaluate them for transition metals in the same manner as described elsewhereOkabe (2007), i.e., from the result of a first principle band calculation, we pick up a main band with the largest DOS, for which we apply Eq. (10). Table 1 shows calculated values along with experimental resultsSchriempf (1968); White and Tainsh (1967); Anderson et al. (1968); Beitchman et al. (1970); Wagner et al. (1971). For these to be evaluated, we regarded $W_{k}^{pp^{\prime}}$ as a constant, except Pd for which presented also is the result with a paramagnon ferromagnetic correlation effect with the correlation length $\xi=5$ÅHicks et al. (1968), which is taken into account as above in Eq. (11). Considering the approximations made to derive Eq. (10), for the scattering amplitudes and so on, we conclude that we could explain a small $L_{\rm exp}$ of tungsten, among others. This must be primarily due to a peculiar FSLoucks (1965). For platinum, however, the observed valueAnderson et al. (1968) is inexplicable by our single band result. Figure 3: For a single band tight-binding model of NaxCoO2, ${\cal L}$ as a function of $x$ as well as the Fermi surfaces at $x=0.6$ are shown to indicate the strong effect caused by a slight deformation due to the hopping integrals $t_{2}$ and $t_{3}$. ### III.3 NaxCoO2 Lastly, we discuss a tight-binding model of NaxCoO2. Though this material has attracted much interest by its large thermopower, it is of particular interest for us because of its simple cylindrical FS. As a nearly 2D system on a triangular lattice, it has an almost circular hole surface centered at the $\Gamma$ point with a slight hexagonal distortion bulging into the sides of the hexagonal Brillouin zoneSingh (2000); Qian et al. (2006); Geck et al. (2007). Therefore, it provides us with an unique case to realize the nontrivial effect as discussed above. Indeed the FS suggests that the Umklapp threshold must exist around $x_{c}\sim 0.6$, and one should expect that even a slight change in the shape of the FS around the threshold could have a striking effect on the doping dependence of the genuine quasiparticle transport properties and the ideal Lorenz factor. In fact, if we assume a undistorted parabolic band, it is straightforward to show that we should obtain $x_{c}=2n_{c,1}-1\simeq 0.55$. Nevertheless, experimentally, the $T^{2}$ dependence of the electrical resistivity has been observed up to $x\simeq 0.7$Li et al. (2004); Foo et al. (2004), around which, therefore, it must be necessary to take into account a three dimensional lattice distortionQian et al. (2006); Geck et al. (2007). For definiteness and simplicity, we restrict ourselves to the doping regime where the three dimensional effect is irrelevant. We do not consider incipient ferromagnetic in-plain correlationsBoothroyd et al. (2004), as they would not modify the result qualitatively. Below, we pay special attention to an expected nontrivial behavior that the factor ${\cal L}$ should tend to vanish as we approach a threshold $x\lesssim x_{c}$. We present results for a 2D tight- binding model on a triangular lattice, in which up to the third neighbor hopping integrals $t_{i}$ ($i=1,2,3$) are considered as in Ref. Kuroki and Arita (2007), where Kuroki and Arita have discussed the thermopower and the electrical conductivity by a relaxation time approximation with a single time scale $\tau$. Here the approximation must be abandoned from the outset, since otherwise one would only obtain a trivial result of a constant value regardless of the FS, as mentioned above. To see how ${\cal L}$ tends to vanish around $x\lesssim x_{c}$ qualitatively, we show Fig. 3 in which the $x$-dependence of ${\cal L}$ around $x\lesssim x_{c}$ is shown for $t_{1}>0$ and $\Gamma/t_{1}=0.02$, along with the Fermi surface for $x=0.6$. In comparison with the dashed line for $t_{2}=t_{3}=0$, the solid line shows the strong effect around $x\lesssim x_{c}$ due to a slight modification of the FS caused by $t_{2}$ and $t_{3}$. In effect, as the threshold value $x_{c}\sim 0.6$ itself depends sensitively on portions of the FS closest to the Brillouin zone boundary, the result cannot be regarded as a quantitative prediction. Furthermore, unfortunately, it is not easy to compare this nontrivial prediction directly with experimental results, because the intrinsic electron term of the thermal resistivity ($\propto BT$) for the perovskite oxides has been completely outweighed by contributions due to phonons and impuritiesLi et al. (2004); Foo et al. (2004). Those extrinsic terms have to be separated out properly to verify the nontrivial filling dependence. Figure 4: As in Fig. 3, shown are the filling $x$ dependences of the factor $F$, proportional to the KW ratio $A/\gamma^{2}$. Lastly, as the vanishment of the factor ${\cal L}$ essentially reflects that of the electrical resistivity, it would be interesting also to calculate the $x$ dependence of the resistivity coefficient $A$, or the Kadowaki Woods (KW) ratio $A/\gamma^{2}$. To evaluate the resistivity coefficients themselves, however, we need the absolute value of the scattering amplitude. Assuming the strong coupling $\rho^{2}W^{pp^{\prime}}_{p-p^{\prime\prime}}\simeq\pi$Okabe (2007), e.g., then we obtain $A/\gamma^{2}={9F}/{16\pi e^{2}}$ with $\displaystyle F$ $\displaystyle=$ $\displaystyle\frac{\displaystyle\sum_{1,2,3}\rho_{p_{1}}\rho_{p_{2}}\rho_{p_{3}}\rho_{p_{3}-p_{1}-p_{2}}v_{p_{1x}}({v}_{p_{1x}}+{v}_{p_{2x}}-{v}_{p_{3x}}-{v}_{p_{1}+p_{2}-p_{3},x})}{\displaystyle\rho^{4}\left(\sum_{p}\rho_{p}{v}_{px}^{2}\right)^{2}}.$ (12) As in Fig. 3 for ${\cal L}$, we show $F$ as a function of $x$ in Fig. 4. The ratio $A/\gamma^{2}$ also vanishes as $x$ approaches $x_{c}$, as expected. However, numerically, we find that it should remain of the order of a common value, $A/\gamma^{2}\simeq 1\times 10^{-5}\mu\Omega$ cm(mol K/mJ)2. Li $et$ $al.$Li et al. (2004) have observed a strongly enhanced deviation from this standard value for $x=0.7$, which should be well beyond the threshold $x_{c}$. It is difficult to regard the numerical deviation simply as a FS effect. Apart from taking into account a relevant three dimensional effect, one would have to assume an enhanced scattering $\rho^{2}W^{pp^{\prime}}_{p-p^{\prime\prime}}\gg 1,$ which may be due to strong scatterings caused by a proximity to some sort of instabilityLi et al. (2004). ## IV Summary In summary, on the basis of anisotropic Fermi liquid theory, we investigated the ideal Lorenz ratio for correlated metals by taking due care of the momentum dependence of transport relaxation processes due to mutual elastic scatterings between quasiparticles. It was shown explicitly that the ideal Lorenz ratio of a correlated electron system is not a constant, but may vary drastically in circumstances and even be made vanishingly small, not only by quantum fluctuations but by a filling control around thresholds of Umklapp scattering channels. Although it might not be easy to extract the ideal Lorenz value in practice, theoretically we pointed out and discussed that such a nontrivial effect should be expected in such a simple single-band system as NaxCoO2. The numerical calculations were carried out on Altix4700 at Shizuoka University Information Processing Center. ## References * Terasaki et al. (1997) I. Terasaki, Y. Sasago, and K. Uchinokura, Phys. Rev. B 56, R12685 (1997). * Mahan (1998) G. D. Mahan, Solid State Physics 51, 81 (1998). * Schriempf (1968) J. T. Schriempf, Phys. Rev. Lett. 20, 1034 (1968). * White and Tainsh (1967) G. K. White and R. J. Tainsh, Phys. Rev. Lett. 19, 165 (1967). * Anderson et al. (1968) A. C. Anderson, R. E. Peterson, and J. E. Robichaux, Phys. Rev. Lett. 20, 459 (1968). * Beitchman et al. (1970) J. G. Beitchman, C. W. Trussel, and R. V. Coleman, Phys. Rev. Lett. 25, 1291 (1970). * Wagner et al. (1971) D. K. Wagner, J. C. Garland, and R. Bowers, Phys. Rev. B 3, 3141 (1971). * Herring (1967) C. Herring, Phys. Rev. Lett. 19, 167 (1967). * Rice (1968) M. J. Rice, Phys. Rev. Lett. 20, 1439 (1968). * Schriempf et al. (1969) J. T. Schriempf, A. I. Schindler, and D. L. Mills, Phys. Rev. 187, 959 (1969). * Singh (2000) D. J. Singh, Phys. Rev. B 61, 13397 (2000). * Pines and Nozières (1966) D. Pines and P. Nozières, _The Theory of Quantum Liquids I_ (W. A. Benjamin, Inc., New York, 1966). * Okabe (1998) T. Okabe, J. Phys. Soc. Jpn. 67, 41 78 (1998). * Abrikosov and Khalatnikov (1959) A. A. Abrikosov and I. M. Khalatnikov, Rept. Progr. Phys. 22, 329 (1959). * Yamada and Yosida (1986) K. Yamada and K. Yosida, Prog. Theor. Phys. 76, 621 (1986). * Hodges et al. (1971) C. Hodges, H. Smith, and J. W. Wilkins, Phys. Rev. B 4, 302 (1971). * Fujimoto et al. (1991) S. Fujimoto, H. Kohno, and K. Yamada, J. Phys. Soc. Jpn. 60, 2724 (1991). * Okabe (2007) T. Okabe, Phys. Rev. B 76, 193109 (2007). * Hicks et al. (1968) T. J. Hicks, T. M. Holden, and G. G. Low, J. Phys. C 1, 528 (1968). * Loucks (1965) T. L. Loucks, Phys. Rev. 139, A1181 (1965). * Qian et al. (2006) D. Qian, D. Hsieh, L. Wray, Y.-D. Chuang, A. Fedorov, D. Wu, J. L. Luo, N. L. Wang, L. Viciu, R. J. Cava, et al., Phys. Rev. Lett. 96, 216405 (2006). * Geck et al. (2007) J. Geck, S. V. Borisenko, H. Berger, H. Eschrig, J. Fink, M. Knupfer, K. Koepernik, A. Koitzsch, A. A. Kordyuk, V. B. Zabolotnyy, et al., Phys. Rev. Lett. 99, 046403 (2007). * Li et al. (2004) S. Y. Li, L. Taillefer, D. G. Hawthorn, M. A. Tanatar, J. Paglione, M. Sutherland, R. W. Hill, C. H. Wang, and X. H. Chen, Phys. Rev. Lett. 93, 056401 (2004). * Foo et al. (2004) M. L. Foo, Y. Wang, S. Watauchi, H. W. Zandbergen, T. He, R. J. Cava, and N. P. Ong, Phys. Rev. Lett. 92, 247001 (2004). * Boothroyd et al. (2004) A. T. Boothroyd, R. Coldea, D. A. Tennant, D. Prabhakaran, L. M. Helme, and C. D. Frost, Phys. Rev. Lett. 92, 197201 (2004). * Kuroki and Arita (2007) K. Kuroki and R. Arita, J. Phys. Soc. Jpn. 76, 083707 (2007).

arxiv-papers

2024-09-04T02:48:53.625558

0802.1150

# Cellular neural networks for NP-hard optimization problems M. Ercsey-Ravasz1,2, T. Roska2 and Z. Néda1 1 Babeş-Bolyai University, Faculty of Physics, RO-400084, Cluj, Romania 2 Péter Pázmány Catholic University, Faculty of Information Technology, HU-1083, Budapest, Hungary ###### Abstract Nowadays, Cellular Neural Networks (CNN) are practically implemented in parallel, analog computers, showing a fast developing trend. Physicist must be aware that such computers are appropriate for solving in an elegant manner practically important problems, which are extremely slow on the classical digital architecture. Here, CNN is used for solving NP-hard optimization problems on lattices. It is proved, that a CNN in which the parameters of all cells can be separately controlled, is the analog correspondent of a two- dimensional Ising type (Edwards-Anderson) spin-glass system. Using the properties of CNN computers a fast optimization method can be built for such problems. Estimating the simulation time needed for solving such NP-hard optimization problems on CNN based computers, and comparing it with the time needed on normal digital computers using the simulated annealing algorithm, the results are astonishing: CNN computers would be faster than digital computers already at $10\times 10$ lattice sizes. Hardwares realized nowadays are of $176\times 144$ size. Also, there seems to be no technical difficulties adapting CNN chips for such problems and the needed local control is expected to be fully developed in the near future. ###### pacs: 07.05.Mh, 89.20.Ff, 05.10.-a, 87.85.Lf ††preprint: APS/123-QED ## I Introduction Solving NP-hard problems is a key task when testing novel computing paradigms. These complex problems frequently appear in physics, life sciences, biometrics, logistics, database search…etc., and in most cases they are associated with important practical applications as well NP . The deficiency of solving these problems in a reasonable amount of time is one of the most important limitation of digital computers, thus all novel paradigms are tested in this sense. Quantum computing for example is theoretically well-suited for solving NP-hard problems, but the technical realization of quantum computers seems to be quite hard. Here we present a computing paradigm based on Cellular Neural Networks (CNN) with good perspectives for fast NP-hard optimization. The advantage of this approach relative to quantum computing is that several practical realizations are already available. After the idea of CNN appeared in 1988 chua:cnn88 a detailed plan for a CNN computer was developed in 1993 roska:cnnum93 . Since than the chip had a fast developing trend and the latest - already commercialized - version is the EYE- RIS chip with lattice size $176\times 144$, mainly used as a visual microprocessor Eye-Ris . The physics community can also benefit from CNN based computers. It has been proven in several previous studies that this novel computational paradigm is usefull in solving partial differential equations pdeI ; pdeII , studying cellular automata models popdyn ; crounse:rand96 , doing image processing applications and making Monte Carlo simulations on lattice models cnnperc ; cnnising . Here it is shown that NP-hard optimization problems can be also effectively solved using such an approach. In the next section we will present briefly the structure and dynamics of cellular neural networks together with the most developed applications of CNN computing. In the third section we prove that a CNN computer on which the parameters of each cell can be separately controlled, is the analog correspondent of a locally coupled two-dimensional spin-glass system. Using the properties of CNN computers a fast optimization algorithm can be thus developed. The local control of the parameters of each cell is already partially realized on some hardwares (for details see section 2.) and it is expected to be fully functional in the near future. Beside the fundamental interest for solving NP hard problems, the importance of this study consists also in motivating the development of hardwares in such direction. ## II Cellular Neural Networks and CNN Computers The CNN Universal Machine (CNN-UM) roska:cnnum93 is one special case of cellular wave computers roska:cellwavecomp in which computation is achieved using the spatial-temporal dynamics of a cellular neural network chua:cnn88 . The CNN is composed by $L\times L$ cells placed on a square lattice and interconnected through their neighbors chua:cnn88 . Usually the $4$ nearest and the $4$ next-nearest neighbors (Moore neighborhood) are considered. Each cell is characterized by a state value: $x_{i,j}(t)$ representing a voltage in the circuit of the cell. The cell has also an input value (voltage) $u_{i,j}$, which is constant in time and can be defined at the beginning of an operation. The third characteristic quantity of the cell is the output value $y_{i,j}(t)$. This is equivalent with the $x_{i,j}$ state value in a given range. More specifically it is a piece-wise linear function, bounded between $-1$ (called as white) and $1$ (black): $y=f(x)\equiv\frac{1}{2}(\mid x+1\mid-\mid x-1\mid)$. The wiring between neighboring cells assures that the state value of each cell can be influenced by the input and output values of its neighbors. The equation governing the dynamics of the CNN cells results from the time- evolution of the equivalent circuits. Supposing the $8$ Moore neighbor interactions it has the following form chua:cnn88 : $\displaystyle\frac{dx_{i,j}(t)}{dt}=-x_{i,j}(t)+\sum_{k=i-1}^{i+1}\sum_{l=j-1}^{j+1}A_{i,j;k,l}y_{k,l}(t)+$ $\displaystyle+\sum_{k=i-1}^{i+1}\sum_{l=j-1}^{j+1}B_{i,j;k,l}u_{k,l}+z_{i,j}$ (1) where ${i,j}$ denotes the coordinates of the cell and the summation indeces ${k,l}$ are for its neighbors. Self-interaction ($k=i,l=j$) is also possible. The set of parameters $\\{A,B,z\\}$ is called a template and controls the whole system. An operation is performed by giving the initial states of the cells, the input image (the input values of all cells) and by defining a template. The states of all cells will vary in parallel and the result of the operation will be the final steady state of the CNN. If the state values ($x_{i,j}$) of all cells remain bounded in the $[-1,1]$ region (i.e. $y_{i,j}=x_{i,j}$ holds for each cell at any time $t$), than each operation is equivalent with solving a differential equation defined by the template itself chuaroskakonyv ; pdeI ; pdeII . When $x_{i,j}$ does not remain bounded, than the piece-wise linear function described at the definition of the output value $y_{i,j}$, takes an important role. The final steady state will not be simply the solution of the differential equation, and this case can be used for defining other useful operations as well chuaroskakonyv . The CNN-UM roska:cnnum93 is a programmable analogic (analog & logic) cellular wave computer. Beside the analog circuits described by Eq.1, each cell contains also a logic unit, local analog and logic memories and a local communication and control unit. The logic unit and logic memories are included to complement the analog computation. In this manner basic logic operations can be performed without defining complicated templates for it. In the local logic memories one can save a binary value ($1$ and $0$ respectively), and in the local analog memories it is possible to save real values between $-1$ and $1$. Since the CNN array is mainly used for image processing and acquisition, the binary values are often referred as black and white, and the real values bounded between $-1$ and $1$ are mapped in a gray-scale scheme. Beside these local units, the CNN-UM has also a global analog programming unit which controls the whole system, making it a programmable computer. It can be easily connected to PC type computers and programmed with special languages, for example the Analogic Macro Code (AMC). The physical implementations of these computers are numerous and widely different: mixed-mode CMOS, emulated digital CMOS, FPGA, and also optical. For practical purposes the most promising applications are for image processing, robotics or sensory computing purposes applications , so the main practical drive in the mixed-mode implementations was to build a visual microprocessor chuaroskakonyv . In the last decades the size of the engineered chips was constantly growing, the new cellular visual microprocessor EYE-RIS Eye-Ris for example has $176\times 144$ processors, each cell hosting also $4$ optical sensors. Parallel with increasing the lattice size of the chips, engineers are focusing on developing multi-layered, $3$ dimensional chips as well. For the first experimental versions the templates (coupling parameters) are defined identical for all cells. This means that for example $A(i,j;i+1,j)$ is the same for all $(i,j)$ coordinates. In such way, on the two-dimensional CNN chip, all the $A$ couplings are defined by a single $3\times 3$ matrix. Totally, $9+9+1=19$ parameters are needed to define the whole, globally valid, template ({A,B,z}). On the latest version of the CNN chips (ACE16K, EYE-RIS) the $z(i,j)$ parameter can already be locally varied. It is expected that on newer chips one will be able to separately control also the $A(i,j;k,l)$ and $B(i,j;k,l)$ connections as well. Many applications ideal for the analogic and parallel architecture of the CNN- UM were already developed and tested. Studies dealing with partial differential equations pdeI ; pdeII or cellular automata models popdyn ; crounse:rand96 prove this. In some of our latest publications we have shown that CNN computers are suitable also for stochastic simulations. The natural noise of the analog chip can be effectively used to generate random numbers approximately $4$ times faster than on digital computers cnnperc . We also presented experiments in which the site-percolation problem and the two- dimensional Ising model was properly solved on the ACE16K chip (with $128\times 128$ cells) cnnperc ; cnnising . ## III NP-hard optimization on CNN architectures The aim of the present study is to prove that CNN computing is also suitable for solving effectively complex optimization problems on spin-glass type lattice models. We consider a two-dimensional CNN where the templates (parameters of Eq. 1) can be locally controlled. Matrix $A$ is considered symmetric $A(i,j;k,l)=A(k,l;i,j)$, $A(i,j;i,j)=1$ for all $(i,j)$, and the elements are bounded $A(i,j;k,l)\in[-1,1]$ ($(i,j)$ and $(k,l)$ denote two neighboring cells). Matrix $B$, which controls the effect of the input image, will be taken simply as: $B(i,j;i,j)=b$ and $B(i,j;k,l)=0$, $\\{i,j\\}\neq\\{k,l\\}$. The parameter $z$ is chosen as $z=0$, so finally our template is defined by $\\{A,b\\}$ alone. In an earlier work Chua et al. chua:cnn88 defined a Lyapunov function for the CNN, which behaves like the ”energy” (Hamiltonian) of the system. For the CNN defined above it can be written simply as $E(t)=-\sum_{<i,j;k,l>}A_{i,j;k,l}y_{i,j}y_{k,l}-b\sum_{i,j}y_{i,j}u_{i,j},$ (2) where $<i,j;k,l>$ denotes pairs of Moore neighbors, each pair taken only once in the sum. $y_{i,j}$ denotes the output value of each cell and $u_{i,j}$ stands for an arbitrary input image. By choosing the parameter $b=0$, the energy of this special CNN is similar with the energy of an Ising type system on square lattice with locally varying coupling constants. The difference is that Ising spins are defined as $\pm 1$, while here we have continuous values between $[-1,1]$. Since the $A(i,j;k,l)$ coupling constants can be positive and negative as well, locally coupled spin-glasses can be mapped in such systems. In the following we will be especially interested in the case when the $A(i,j;k,l)$ couplings lead to a frustration and the quenched disorder in the system is similar with that of spin-glass systems (edwand ; sherkirk ). The Lyapunov function defined by Chua et al. has two important properties chua:cnn88 : 1.) it is always a monotone decreasing function in time, $dE/dt\leq 0$, so starting from an initial condition $E$ can only decrease during the dynamics of the CNN. 2.) the final steady state is a local minimum of the energy: $dE/dt=0$. In addition to these, our CNN has also another important property: due to the fact that all self-interaction parameters are $A(i,j;i,j)=1$, it can be shown that the output values of the cells in a final steady state will be always either $1$ or $-1$. The local minima achieved by the CNN is thus an Ising-like configuration. We can conclude thus that starting from any initial condition the final steady state of the template - meaning the result of an operation - will be always a local minimum of the spin-glass type Ising spin system with local connections defined by matrix $A$. The fact that one single operation is needed for finding a local minimum of the energy, gives us hope to develope fast optimization algorithms. As already emphasized, the complex frustrated case (locally coupled spin-glass type system), where the $A$ coupling parameters generates a non-trivial quenched disorder, will be considered here. The minimum energy configuration of such systems is searched by an algorithm which is similar with the well- known simulated annealing method simaneal . The noise is included with random input images ($u_{i,j}$ values in eq. 2) acting as an external locally variable magnetic field. The strength of this field is governed through parameter $b$. Whenever $b$ is different from zero, our CNN template minimizes the energy with form 2: the first part of it being the energy of the considered spin-glass type model and the second part an additional term, which gets minimal when the state of the system is equal to the input image (the external magnetic field). If $b$ is large, the result will be the input image itself, if $b=0$ the result is a local minimum of the pure Ising-type system. For values in between, our result is a ”compromise” between the two cases. Slowly decreasing the value of $b$ will result in a process similar with simulated annealing, where the temperature of the system is consecutively lowered. First big fluctuations of the energy are allowed, and by decreasing this we slowly drive the system to a low energy state. Since the method is a stochastic one, we can of course never be totally sure that the global minimum will be achieved. The steps of the algorithm are the following: 1\. One starts from a random initial condition $x$, and $b=5$ (with this value the result of the template is almost exactly the same as the input image). 2\. A binary random input image $u$ is generated with $1/2$ probability of black ($1$) pixels, 3\. Using the $x$ initial state and the $u$ input image the CNN template is applied, 4\. The value of $b$ is decreased with steps $\Delta b$, 5\. Steps 2-4 are repeated until $b=0$ is reached. The results of the previous step (minimization) is considered always as the initial state for the next step. 6\. When reaching $b=0$ the image (Ising spin configuration) is saved and the energy is calculated. In the classical simulated annealing algorithm several thousands of steps for a single temperature are needed. Here the CNN template working totally in parallel replaces all these steps. Similarly with choosing the cooling rate in simulated annealing, choosing the value of $\Delta b$ is also a delicate problem. A proper value providing an acceptable compromise between the quality of the results and speed of the algorithm has to be found. For each system size one can find an optimal value of $\Delta b$, but as one would expect this is rapidly decreasing by increasing the system size. It is much more effective, both for performance (meaning the probability of finding the real global optimum) and speed, to choose a constant $\Delta b=0.05$ step and repeat the whole cooling process several times. As a result, several different final states will be obtained, and we have a higher probability to get the right global minima between these. For testing the efficiency of the algorithm one needs to measure the number of steps necessary for finding the right global minima. To do this, one has to previously know the global minima. In case of small systems with $L=5,6$ this can be obtained by a quick exhaustive search in the phase-space. For bigger systems the classical simulated annealing algorithm was used. The temperature was decreased with a rate of $0.99$ ($T_{final}/T_{ini}$) and $1000$ Monte Carlo steps were performed for each temperature. In the present work spin-glass systems with $A(i,j;k,l)=\pm 1$ local connections were studied. The $p$ probability of the positive bonds was varied (influencing the amount of frustration in the system), and local interactions with the $8$ Moore neighbors were considered. For several $p$ densities of the positive links and various system sizes, we calculated the average number of steps needed for finding the energy minimum. As naturally is expected for the non-trivial frustrated cases, the needed simulation time exponentially increases with the system size. As an example, the $p=0.4$ case is shown on Fig. 1a. Circles represent the estimated time on CNN computers and stars illustrate the time measured on a Pentium 4, 3.4 GHz computer using the classical simulated annealing method. For the calculation time on CNN computers we have used a convention (arguments given in the next paragraph) that roughly $1000$ simulation steps are made in $1$ second. As observable in the figure, we could made estimates for relatively small system sizes only ($L\geq 12$). The reason for this is that we had to simulate also the operations on the CNN chip, and for large lattices a huge system of partial differential equations had to be solved. This process gets quite slow for bigger lattices. The needed average number of steps to reach the estimated energy minima depends also on the $p$ probability of the positive connections in the system. On Fig.1b we illustrate this for a system with size $L=7$. To obtain this data for each $p$ value $5000$ different systems were analyzed. As observable on Fig.1b the system is almost equally hard to solve for all $p$ values in the rage of $p\in(0,0.6)$. Figure 1: a) Time needed to reach the minimum energy as a function of the lattice size $L$. Circles are for estimates on CNN computers and stars are simulated annealing results on 3.4 GHz Pentium 4. Results are averaged on $10000$ different configurations with $p=0.4$ probability of positive bonds. For the CNN algorithm $\Delta b=0.05$ was chosen. For simulated annealing the initial temperature was $T_{0}=0.9$ , final temperature $T_{f}=0.2$ and the decreasing rate of the temperature was fixed as $0.99$. b) Number of steps needed for reaching the presumed global minima as a function of the probability $p$, of positive connections. Simulation results with the proposed algorithm on a CNN lattice with size $L=7$. Finally, let us have some thoughts about the estimated speed of such an optimization algorithm. As mentioned earlier, on the nowadays available CNN chips, only parameter $z$ can be locally varied, parameters $A$ and $B$ are $3\times 3$ matrices, uniformly applied for all cells. The reason for no local control of $A$ and $B$ seems to be simply the lack of motivations. In image processing applications no really useful algorithms were developed, which would require these locally variable connections. Realizing the local control of $A$ and $B$ is technically possible and is expected to be included in the newer versions of the CNN chips. This modification would not change the properties and the speed of the chip, only the control unit and template memories would become more complicated. Also, introducing the connection parameters in the local memories of the chip would take a slightly longer time. In the specific problem considered here the connection parameters have to be introduced only once for each problem, so this would not effect in a detectable manner the speed of calculations. Based on our previous experience with the ACE16K chip (with sizes $128\times 128$) cnnperc ; cnnising we can make an estimation of the speed for the presented optimization algorithm. This chip with its parallel architecture solves one template in a very short time - of the order of microseconds. For each step in the algorithm one also needs to generate a random binary image. This process is already $4$ times faster on the ACE16K chip than on a 3.4 GHz Pentium 4 computer and needs around $100\mu s$ (see cnnperc ). It is also helpful for the speed, that in the present algorithm it is not needed to save information at each step, only once at the end of each cooling process. Saving an image takes roughly $10$ milliseconds on the ACE16K, but this is done only once after several thousand of simulation steps. Making thus a rough estimate for our algorithm, a chip with similar properties like the ACE16K should be able to compute between $1000-5000$ steps in one second, _independently of the lattice size_. Using the lower estimation value ($1000$ steps /second) and following up the number of steps needed in case of $p=0.4$, the estimated average time for solving one problem is plotted as a function of the lattice size in Fig. 1a (empty circles). Comparing this with the speed of simulated annealing (SA) performed on a $3.4$ GHz Pentium 4 (stars on Fig. 1a), the results for larger lattice sizes are clearly in favour of the CNN chips. For testing the speed of simulated annealing we used the following parameters: initial temperature $T_{0}=0.9$ , final temperature $T_{f}=0.2$, decreasing rate of the temperature $0.99$. Results were avereged for $10000$ different bond distributions. From Fig. 1a it results that the estimated time needed for the presented algorthim on a CNN chip would be smaller than simulated annealing already at $10\times 10$ lattice sizes. Spin-glass like systems have many applications in which global minimum is not crucial to be exactly found, the minimization is needed only with a margin of error. In such cases the number of requested steps will decrease drastically. As an example in such sense, it has been shown that using spin-glass models as error-correcting codes, their cost-performance is excellent sourlas , and the systems are usually not even in the spin-glass phase. In this manner by using the CNN chip, finding acceptable results could be very fast, even on big lattices. ## IV Conclusion A cellular neural network with locally variable parameters was used for finding the optimal state of locally coupled, two-dimensional, Ising type spin-glass systems. By simulating the proposed optimization algorithm on a CNN chip, where all connections can be locally controlled, very good perspectives for solving such NP hard problems were predicted: CNN computers could be faster than digital computers already at a $10\times 10$ lattice size. Chips with $2$ and $3$ layers of cells were also produced (CACE1k, XENON) and increasing the number of layers is expected in the near future. This further extends the number of possible applications. On two layers is possible to map already a spin system with any connection matrix (even globally coupled spins) and also other important NP-hard problems (e.g. K-SAT) may become treatable. ###### Acknowledgements. Work supported from a Romanian CNCSIS No.1571 research grant (contract 84/2007) and a Hungarian ONR grant (N00014-07-1-0350). ## References * (1) H. Nishimori, Statistical Physics of Spin Glasses and Information Processing. An Introduction. (Clarendon Press, Oxford, 2001) * (2) L.O. Chua, L. Yang, IEEE Transactions on Circuits and Systems 35, 1257 (1988) * (3) T.Roska, L.O. Chua, IEEE Transactions on Circuits and Systems - II 40, 163 (1993) * (4) www.anafocus.com * (5) T. Roska, L.O. Chua, D. Wolf, T. Kozek, R. Tetzlaff, F. Puffer, IEEE Transactions on Circuits and Systems - I: Fundamental Theory and Applications 42, 807 (1995) * (6) T. Kozek, L.O. Chua, T. Roska, D. Wolf, R. Tetzlaff, F. Puffer, K. Lotz, IEEE Transactions on Circuits and Systems - I: Fundamental Theory and Applications 42, 816 (1995) * (7) J.M. Cruz, L.O. Chua, IEEE Transactions on Circuits and Systems I: Fundamental Theory and Applications 42, 715 (1995) * (8) K.R. Crounse, T. Yang, L.O. Chua, Fourth IEEE International Workshop on Cellular Neural Networks and their Applications, Seville, Spain (1996) * (9) K.R. Crounse, L.O. Chua, IEEE Trans. on Circuits and Systems 42, 583 (1995) * (10) M. Ercsey-Ravasz, T. Roska, Z. Néda, International Journal of Modern Physics C 17, No. 6, 909 (2006) * (11) M. Ercsey-Ravasz, T. Roska, Z. Néda, European Physical Journal B , 51, 407 (2006) * (12) T. Roska, Journal of Circuits, Systems and Computers 5, No.2, 539 (2003). * (13) L.O. Chua, T. Roska, Cellular Neural Networks and Visual Computing (Cambridge University Press, 2002) * (14) S. F. Edwards, P. W. Anderson, Journal of Physics F 5, 965 (1975) * (15) D. Sherrington, S. Kirkpatrick, Physical Review Letters 35, 1792 (1975) * (16) S. Kirckpatrick, C.D. Gelatt and M.P. Vecchi, Science 220, 671 (1983) * (17) N. Sourlas, Nature 339, 693 (1989)

arxiv-papers

2024-09-04T02:48:53.634044

0802.1213

# Cold atom confinement in an all-optical dark ring trap Spencer E. Olson Matthew L. Terraciano Mark Bashkansky Fredrik K. Fatemi Naval Research Laboratory, 4555 Overlook Ave. S.W., Washington, DC 20375 ###### Abstract We demonstrate confinement of 85Rb atoms in a dark, toroidal optical trap. We use a spatial light modulator to convert a single blue-detuned Gaussian laser beam to a superposition of Laguerre-Gaussian modes that forms a ring-shaped intensity null bounded harmonically in all directions. We measure a 1/$e$ spin-relaxation lifetime of $\approx$1.5 seconds for a trap detuning of 4.0 nm. For smaller detunings, a time-dependent relaxation rate is observed. We use these relaxation rate measurements and imaging diagnostics to optimize trap alignment in a programmable manner with the modulator. The results are compared with numerical simulations. ###### pacs: 32.80.Pj, 39.25.+k, 03.75.Be ††preprint: APS/123-QED Toroidal traps for cold atoms have recently been of interest for both fundamental and applied research. A toroidal geometry can enable studies of phenomena in non-simply connected or low dimensional topologies Helmerson et al. (2007); Gupta et al. (2005); Jain et al. (2007); Bludov and Konotop (2007); Jackson and Kavoulakis (2006); Lesanovsky and von Klitzing (2007); Fernholz et al. (2007); Morizot et al. (2006); Dutta et al. (2006); Arnold et al. (2006); Wu et al. (2004); Ruostekoski and Dutton (2005), e.g. superfluid persistent circulation states of Bose-Einstein condensates (BECs) Helmerson et al. (2007). A ring-shaped atom waveguide may also be suitable for inertial measurements Gustavson et al. (2000) and neutral atom storage Dutta et al. (2006); Arnold et al. (2006); Wu et al. (2004); Sauer et al. (2001). Several approaches for generating ring-shaped waveguides have been proposed and implemented. Magnetic fields have been used to create large ring traps for possible use as atom storage rings or Sagnac interferometry Gupta et al. (2005); Sauer et al. (2001); Arnold et al. (2006); Wu et al. (2004). Helmerson et al. Helmerson et al. (2007) used a combination of magnetic and optical fields to demonstrate persistent current flow of a BEC. Morizot et al. Morizot et al. (2006) proposed ring traps formed from the combination of an optical standing wave with rf-dressed atoms in a magnetic trap. All-optical approaches have also been considered for toroidal traps Wright et al. (2000); Courtade et al. (2006); Freegarde and Dholakia (2002). Wright et al. Wright et al. (2000) suggested the use of high-azimuthal-order Laguerre- Gaussian (LG) beams to confine atoms with red-detuning. Atoms in red-detuned optical traps seek high intensity, and with large detuning, spontaneous photon scattering can be negligible. Photon scattering can also be reduced by using blue-detuned optical traps. Such “dark” traps confine atoms to low intensity, allowing field-free measurements Ozeri et al. (1999); Grimm et al. (2000); Friedman et al. (2002); Kaplan et al. (2005), but are challenging to make because they require an intensity minimum bounded by higher intensity. This challenge is often overcome by crossing beams Kuga et al. (1997); Fatemi et al. (2007); Friedman et al. (2002) to plug a hollow optical potential, although dark point atom traps have been realized with a single laser beam containing a phase-engineered intensity null Ozeri et al. (1999). The single beam approach has the advantage of alignment simplicity over crossed-beam configurations. Lattices of dark rings have been proposed Freegarde and Dholakia (2002) and realized Courtade et al. (2006) using counterpropagating laser beams, but to the best of our knowledge, there have been no reports of atom confinement in a lone optical ring trap. In this paper, we report atom confinement within a different class of dark optical ring traps. We form a bounded, ring-shaped intensity null by converting a Gaussian laser beam to a dual-ringed beam with a programmable spatial light modulator (SLM). SLMs are of increasing value in cold atom manipulation experiments because of their ability to reconfigure trap parameters Fatemi et al. (2007); Bergamini et al. (2004); Chattrapiban et al. (2006); McGloin et al. (2003); Boyer et al. (2006). We measure the spin- relaxation lifetime, observe atom dynamics within the traps, and compare the experimental results with numerical simulations. Figure 1: (Color online) a) Phase profile for creating the dark optical ring. b) CCD image of a dual-ringed beam in the focal plane. c) Numerical simulation of $r$-$z$ cross section. Right: Transverse profile through the minimum, on a line through points 1 and 2. Bottom: Profile along the minimum-intensity path, indicated by the arrows, through points 3 and 4. Length scales are shown on the profile plots. The dashed curves in the profile plots are quadratic fits. The dark trap is formed in a ring through the centers of the dashed ellipses. We form the dual-ringed laser beam by modifying the spatial phase of a laser beam with an SLM, in a similar manner to that used for producing hollow laser beams Fatemi et al. (2007); Chattrapiban et al. (2006); Rhodes et al. (2006). The latter can be created by imparting an azimuthal phase $\Phi(r,\phi)=\ell\phi$, with integer $\ell$, to a Gaussian laser beam $E(r)=|E_{0}|exp\left(-r^{2}/w_{0}^{2}\right)$, where $w_{0}$ is the waist. The phase discontinuity at $r$ = 0 results in a hollow beam that, for low $\ell$, closely approximates a pure $LG_{p=0}^{\ell}$ mode, where $p$ and $\ell$ are radial and azimuthal indices. As shown in Figs. 1a-b, a dual ring is produced by introducing a $\pi$ phase discontinuity at $r=R_{c}>0$ such that the resulting beam has large overlap with the $LG_{p=1}^{\ell}$ mode, which has two radial nodes. The parameter $R_{c}/w_{0}$ controls the modal composition and thus the propagation characteristics. In Ref. Arlt et al. (1998), $R_{c}/w_{0}$ was set to generate high purity LG modes. Here, we adjust $R_{c}/w_{0}$ to create a superposition of $LG_{p}^{\ell}$ modes that produces a dark ring at the focus of a lens that is bounded in both the radial and longitudinal directions. Figure 1c shows the calculated $r$-$z$ cross-section of a toroidal beam with $\ell=1$ as it propagates along $z$ through the focus of an $f$=215 mm focal length lens ($w_{0}=1.7$mm). We have chosen values of $R_{c}$ such that the barrier heights in the longitudinal and transverse directions are equal. For $\ell$=0, 1, and 2, this condition is satisfied for $R_{c}/w_{0}\approx$ 0.71, 0.79, and 0.85. The small numerical aperture (NA=$w_{0}/f$=0.008) leads to a long aspect ratio of $\approx$1:300 for $\ell=1$, defined as the ratio of the longitudinal trap frequency $\omega_{\parallel}$ to the transverse trap frequency $\omega_{\perp}$. The mode composition is dominated by $p=0$ (single-ringed) and $p=1$ (dual-ringed) modes. For $\ell=0$, _e.g._ , the $p=0(1)$ fraction is 13%(78%). The potential is harmonic in all directions, as indicated in Fig. 1c. Under these conditions, the ratio of the inner radial barrier height to the outer radial barrier height is $\approx$25-35%. The radius of the trap depends linearly on $\ell$, as it does for hollow beams Curtis and Grier (2003); Fatemi and Bashkansky (2007). The trapping beam is derived from a 30 mW extended cavity diode laser tunable from 776-780 nm. The beam is amplified to $\approx$350 mW with a tapered amplifier of which $\approx$150 mW is coupled into polarization maintaining fiber. The linearly polarized fiber output is collimated with $w_{0}$=1.7 mm, and reshaped by a 512x512 reflective SLM (Boulder Nonlinear Systems) with 15 $\mu$m pixels and $\approx$90% absolute diffraction efficiency. A 4-$f$ imaging setup relays this modified Gaussian beam to a magneto-optical trap (MOT). The 4-$f$ relay roughly positions the focus of the ring trap over the MOT, but fine longitudinal adjustments are controlled entirely by the SLM by adding a lens phase profile $\Phi_{\rm{lens}}(r,\phi)=-{\pi}r^{2}/f\lambda$. We compensate for wavefront errors imposed by the SLM by calibrating the programmed phase on a pixel-by-pixel basis. The experiment begins with a MOT containing $10^{7}$ 85Rb atoms. After a 1 second loading time, the MOT coils are shut off, and the atoms are cooled to 5 ${\mu}K\approx\hbar\Gamma/60k_{B}$ during a 10 ms molasses cooling stage. All cooling and trapping beams are then extinguished, followed by a 100 $\mu$s pulse that optically pumps the atoms into the $F$=2 hyperfine level. The toroidal beam power is ramped to $\approx$150 mW over 5 ms during the molasses stage. This ramp adiabatically loads atoms into the trap and minimizes the energy gained in the loading process. The trap diameter is significantly smaller than the initial MOT size, so we typically load only a small fraction of atoms ($\approx$5$\times$104) into the traps. Collisions with background gas limit the trap 1/$e$ lifetimes to $\approx$1 s. After a variable delay, the trapped atoms are imaged onto an electron-multiplying CCD camera (Andor Luca) by a 500 $\mu$s pulse from the MOT and repump beams. Immediately prior to the imaging pulse, the trapping beam is switched off to avoid Stark shifting of the levels. For linear polarization, the optical potential is Grimm et al. (2000) $U(r)=\frac{\hbar{\Gamma}I(r)}{24I_{s}}\left(\frac{\Gamma}{\Delta+\Delta_{\rm{LS}}}+\frac{2\Gamma}{\Delta}\right)$ (1) where $I_{\rm s}$=1.6 mW/cm2 is the saturation intensity, $\Gamma$=2$\pi\times$6.1 MHz is the linewidth, and $\Delta_{\rm{LS}}$=2$\pi\times$ 7.1 THz (=15 nm) is the fine structure splitting. The resulting trap depths for $\ell=1$ and $\Delta=0.5$ nm, 1.0 nm, 2.0 nm, and 4 nm are 0.26$\hbar\Gamma$, 0.13$\hbar\Gamma$, 0.065$\hbar\Gamma$, and 0.033$\hbar\Gamma$ (at 780 nm, 1 nm$\leftrightarrow$493 GHz). At $\Delta=1$ nm, $\omega_{\perp}{\approx}2\pi\times$800 Hz and $\omega_{\parallel}{\approx}2\pi\times$3 Hz. Figure 2: (Color online) a) Image of atom cloud taken along $x$-axis for $\ell=1$. b) Images taken along $z$-axis (left) and profiles (right) for toroidal traps using $\ell=0-2$. The experimental beam profiles are shown for comparison (dotted line). Trap time for this figure is 600 ms. We record images of the trapped atoms with the camera axis along $x$ and along $z$. Images along $x$ show the longitudinal trap extent (Fig. 2a), while those along $z$ show the toroidal structure (Fig. 2b). The head-on views in Fig. 2b are taken after a trap time of 600 ms for $\ell=0{\rm-}2$. Also shown are the azimuthally-averaged beam intensity profiles in the focal plane and atom distributions in the $x$ and $y$ (gravity) directions. Because the trapping beam is propagating horizontally, the potential is not azimuthally symmetric. The gravitational potential energy difference between the intensity nulls for $\ell=2$ is $\Gamma$/30 $\approx 2\pi\times$200 kHz for 85Rb, which is larger than the atom cloud temperature of 2$\pi\times$100 kHz. Thus, most atoms are found in the bottom portion of the trap. For $\ell\geq 1$, atoms could initially be loaded on the axis of the beam, along which there is no barrier. This is seen for $\ell=2$ in Fig. 2. In our configuration it takes a few seconds for these atoms to drift away. Although there should be little interaction between axial atoms and the ring-trapped atoms under adiabatic loading, the axial atoms can be reduced by several means, such as orienting the trapping beam vertically, or loading from an atom distribution that has been dimpled by a blue-detuned Gaussian beam, as in Ref. Helmerson et al. (2007). A vertical propagation axis would permit a symmetric ring potential in a horizontal plane, but optical access in this direction was limited. Imaging constraints prevent high contrast images of the toroidal atom distributions. We use an 85 mm Nikon f/1.4 lens, the front element of which is $\approx$250 mm away from the trap location. This lens collects the maximum fluorescence and achieves a peak resolution of $\approx$5 $\mu$m but suffers from spherical aberration, which causes the observed loss of contrast. Figure 3: a) Measurement of F=3 fraction as a function of time for $\Delta$=0.5nm (triangles), 1.0 nm (diamonds), 2.0 nm (squares), and 4 nm (circles). Fits (solid lines) using the model described in the text. b) Comparison of 0.5 nm data with a single parameter exponential curve and simulations. One benefit of dark traps for coherent atom manipulation is the suppression of photon scattering events Ozeri et al. (1999); Grimm et al. (2000); Friedman et al. (2002); Kaplan et al. (2005). We measure the spin relaxation rate due to Raman scattering by measuring the fraction of atoms in the trap that transfer to $F$=3 as a function of trap Miller et al. (1993). The atoms are first pumped into the $F$=2 hyperfine level. After a variable trapping time, we image only the atoms that transfer to $F$=3 by using a 10 $\mu$s pulse of resonant cycling transition light. Within 2 ms, both the repump and the cycling transition beams are switched on to image the atoms in both the $F$=2 and $F$=3 states. For background subtraction, two images with the same pulse sequence are taken with no atoms present. This type of background subtraction is necessary to eliminate false counts due to CCD ghosting. By taking the images during a single loading cycle, the effect of atom number fluctuations is reduced. These images are recorded along $x$ (as in Fig. 2a). Between the first two imaging pulses, the atom distribution expands slightly beyond the few integrated rows of pixels. This leads to a slightly low estimate of the total atom count, but the resulting $F$=3 normalized signal is proportional to the actual $F$=3 fraction. We record the $F$=3 signal fraction as a function of trap time for four different detunings (Fig. 3a). In the simplest approximation that all atoms have an equal scattering rate, each curve can be modeled by a single exponential $N_{3}(t)=C(1$ \- exp(-${t/\tau}$)), as was used in Ref. Ozeri et al. (1999), where $\tau$ is the 1/$e$ decay time. For $\Delta\leq 1$ nm, however, a single relaxation rate was not observed (Fig. 3b). This difference between our results and those of Ref. Ozeri et al. (1999) is most likely due to differences in the trap loading technique, which we have found to affect the rate curves. We note that the $F$=3 fraction at long times should approach 7/12, but our measured values are higher due to the pixel integration described above. Instead of modeling the spin-relaxation with a single-parameter time constant, we phenomenologically “chirp” $\tau$ to be $\tau$(t) = $\tau_{0}+\beta{t^{1/2}}$ so that we can estimate the relaxation rate at different times. We choose a sublinear chirp rate so that the exponential will decay at long times, but the exact functional form will depend on trap geometry. A steadily increasing $\tau$ should be expected since atoms initially loaded into the trap in locations of high intensity scatter photons more quickly than those loaded into the dark portions of the trap. Thus, a rapid increase in the $F$=3 fraction is observed for small $t$, followed by longer relaxation times for the atoms with the least total energy. Using this form for the $F$=3 fraction, approximate spin-relaxation lifetimes at t = 0 for $\Delta=0.5$ nm, 1.0 nm, 2.0 nm, and $4.0$ nm are 35 ms, 115 ms, 460 ms, and 1440 ms; after 500 ms, these increase to 140 ms, 230 ms, 750 ms, and 1500 ms. The scattering time for atoms in a red-detuned trap of comparable depth at $\Delta$=0.5nm would be $\approx$2.5 ms, which is 50 times shorter than our recorded value. In Ref. Ozeri et al. (1999), the blue-detuned trap had a scattering lifetime 700 times longer than a comparable red-detuned trap at 0.5 nm. That work used significantly higher intensities, where the differences between red- and blue-detuning are more dramatic. Photon scattering may be reduced substantially by using commercially available lasers with higher power and larger detuning. For $\Delta>\Delta_{\rm{LS}}$, spin relaxation is further suppressed, asymptotically scaling as $\Delta^{-4}$ Miller et al. (1993). The time-dependent scattering rate is likely not limited to toroidal geometries, but to the best of our knowledge, it has been observed for the first time in this report. Also, we point out that we did not directly measure the recoil scattering rate, but for our $\Delta$ this is on the same order as the spin- relaxation rate. A recoil scattering rate of 1 s-1 corresponds to a heating rate of $\approx$400 nK/s. To demonstrate the time-dependent scattering rate numerically, we perform Monte Carlo simulations for $\Delta=0.5$ nm. The simulated trap is ramped on over 5 ms. The atom cloud is initially in the $F=2$ state and normally distributed in position and velocity to match the MOT size and temperature. Molasses effects are ignored. Within each time step, each atom’s hyperfine level is changed with a probability determined by the local scattering rate, as calculated from the Kramers-Heisenberg formula Miller et al. (1993). The simulation results, compared to data in Fig. 3b, confirm the time-dependent relaxation rate described above. For comparison, the data have been renormalized to have an asymptotic value of $7/12$. Figure 4: (Color online) a) Spin-relaxation curves for three different starting positions. b) Images of longitudinal oscillation of the atom cloud using trap displacement from left to right of 3 mm, 1.5 mm, and 0.0mm. To demonstrate axial confinement and to quantify the dependence of the scattering rates on the starting position of the atoms in the trap, we displace the trap minimum from the MOT by adjusting the lens function written to the SLM by a few MOT radii (MOT radius $\approx 250~{}\mu$m). Thus, most atoms are initially located in regions of high intensity, reducing the overall scattering lifetime. When the atom cloud is displaced 3 mm, 1.5 mm, and 0.0 mm away from the trap minimum, the single-parameter rate constants (for $\Delta=1$ nm) are 145 ms, 195 ms, and 230 ms (Fig. 4a). For each displacement, we show a composite image of the side views of the trap (Fig 4b), where each row in the image is a different slice in time. These images show the atom cloud oscillating in the longitudinal direction when the trap is not well overlapped with the atom cloud. By displacing the trap focus, we can also estimate the longitudinal trap frequency. For $\Delta=1$ nm, we measure $\omega_{\parallel}\approx 2\pi\times 2$ Hz. This agrees well with the estimate of $\omega_{\parallel}\approx 2\pi\times 3$ Hz from the calculated intensity profiles shown in Fig. 1. The scattering rate data and the composite images can be used to optimize the location of the trap focus, which is done to $\approx$100 $\mu$m with the SLM. As with all single-beam traps, the aspect ratio scales with the inverse of the trapping beam NA. For similar beam parameters, an aspect ratio of $\approx$10 could be realized by using a $f=10$ mm lens. A crossed beam geometry, in which additional beams cap the potential in the longitudinal direction, allows significantly tighter longitudinal confinement and larger diameter traps. In these cases, the ratio $R_{c}/w_{0}$ can be changed for optimal confinement. One possibility is to use values of $R_{c}/w_{0}$ such that the modified beam is primarily in a single $LG_{p=1}^{\ell}$ mode Arlt et al. (1998). Pure $LG_{p=1}^{\ell}$ modes have a radial intensity null that persists for all values of $z$. When $R_{c}/w_{0}$ is chosen such that the most pure $LG_{p=1}^{\ell}$ is formed, the inner radial barrier height is roughly 3$\times$ larger than the outer one, and the longitudinal barrier is minimized. Therefore, the crossing beam can be well outside the focal plane, where better beam quality is observed but the ring-shaped null remains dark. The reduction of aberration effects outside the focal plane was shown for hollow beams in Ref. Fatemi and Bashkansky (2007). We have used a spatial light modulator to generate superpositions of LG modes that form single-beam, dark ring traps for cold atoms. We have shown that the atoms can be held in these potentials with long state lifetimes. We have observed atom dynamics in the longitudinal direction and shown that by modifying the trap alignment with the SLM we can optimize the scattering lifetime. This work was funded by the Office of Naval Research and the Defense Advanced Research Projects Agency. ## References * Helmerson et al. (2007) K. Helmerson _et al._ , Nuclear Physics A 790, 705c (2007). * Gupta et al. (2005) S. Gupta _et al._ , Phys. Rev. Lett. 95, 143201 (2005). * Jain et al. (2007) P. Jain, A. S. Bradley, and C. W. Gardiner, Phys. Rev. A 76, 023617 (2007). * Bludov and Konotop (2007) Y. V. Bludov and V. V. Konotop, Phys. Rev. A 75, 053614 (2007). * Jackson and Kavoulakis (2006) A. D. Jackson and G. M. Kavoulakis, Phys. Rev. A 74, 065601 (2006). * Lesanovsky and von Klitzing (2007) I. Lesanovsky and W. von Klitzing, Phys. Rev. Lett. 99, 083001 (2007). * Fernholz et al. (2007) T. Fernholz _et al._ , Phys. Rev. A 75, 063406 (2007). * Morizot et al. (2006) O. Morizot _et al._ , Phys. Rev. A 74, 023617 (2006). * Dutta et al. (2006) O. Dutta, M. Jääskeläinen, and P. Meystre, Phys. Rev. A 74, 023609 (2006). * Arnold et al. (2006) A. S. Arnold, C. S. Garvie, and E. Riis, Phys. Rev. A 73, 041606(R) (2006). * Wu et al. (2004) S. Wu _et al._ , Phys. Rev. A 70, 013409 (2004). * Ruostekoski and Dutton (2005) J. Ruostekoski and Z. Dutton, Phys. Rev. A 72, 063626 (2005). * Gustavson et al. (2000) T. L. Gustavson, A. Landragin, and M. A. Kasevich, Class. Quantum Grav. 17, 2385 (2000). * Sauer et al. (2001) J. A. Sauer, M. D. Barrett, and M. S. Chapman, Phys. Rev. Lett. 87, 270401 (2001). * Wright et al. (2000) E. M. Wright, J. Arlt, and K. Dholakia, Phys. Rev. A 63, 013608 (2000). * Courtade et al. (2006) E. Courtade _et al._ , Phys. Rev. A 74, 031403(R) (2006). * Freegarde and Dholakia (2002) T. Freegarde and K. Dholakia, Opt. Commun. 201, 99 (2002). * Ozeri et al. (1999) R. Ozeri, L. Khaykovich, and N. Davidson, Phys. Rev. A 59, R1750 (1999). * Grimm et al. (2000) R. Grimm, M. Weidemuller, and Y. B. Ovchinnikov, Adv. Atom. Mol. Opt. Phys. 42, 95 (2000). * Friedman et al. (2002) N. Friedman, A. Kaplan, and N. Davidson, Adv. Atom. Mol. Opt. Phys. 48, 99 (2002). * Kaplan et al. (2005) A. Kaplan _et al._ , J. Opt. B: Quantum Semiclass. Opt. 7, R103 (2005). * Kuga et al. (1997) T. Kuga _et al._ , Phys. Rev. Lett. 78, 4713 (1997). * Fatemi et al. (2007) F. K. Fatemi, M. Bashkansky, and Z. Dutton, Opt. Express 15, 3589 (2007). * Bergamini et al. (2004) S. Bergamini _et al._ , J. Opt. Soc. Am. B 21, 1889 (2004). * Chattrapiban et al. (2006) N. Chattrapiban _et al._ , J. Opt. Soc. Am. B 23, 94 (2006). * McGloin et al. (2003) D. McGloin _et al._ , Opt. Express 11, 158 (2003). * Boyer et al. (2006) V. Boyer _et al._ , Phys. Rev. A 73, 031402(R) (2006). * Rhodes et al. (2006) D. P. Rhodes _et al._ , J. Mod. Optics 53, 547 (2006). * Arlt et al. (1998) J. Arlt _et al._ , J. Mod. Optics 45, 1231 (1998). * Curtis and Grier (2003) J. E. Curtis and D. G. Grier, Phys. Rev. Lett. 90, 133901 (2003). * Fatemi and Bashkansky (2007) F. K. Fatemi and M. Bashkansky, Appl. Opt. 46, 7573 (2007). * Miller et al. (1993) J. D. Miller, R. A. Cline, and D. J. Heinzen, Phys. Rev. A 47, R4567 (1993).

arxiv-papers

2024-09-04T02:48:53.642521

0802.1216

# High temperature superconductivity: from complexity to simplicity Sudip Chakravarty Department of Physics and Astronomy, University of California Los Angeles, Los Angeles, CA 90095, USA ###### Abstract I discuss the recent quantum oscillation experiments in the underdoped high temperature superconductors. > HAMLET: Do you see yonder cloud that’s almost in shape of a camel ? > POLONIUS: By th’ mass and ’tis, like a camel indeed. > HAMLET: Methinks it is like a weasel. > POLONIUS: It is backed like a weasel. > HAMLET: Or like a whale. > POLONIUS: Very like a whale. > —William Shakespeare More than 20 years ago, Bednorz and Müller discovered superconductivity in copper oxides at remarkably high temperatures Bednorz:1986 . Since then, physicists have struggled to understand the mechanisms at work. Recently, a set of experiments on cuprates in high magnetic fields Doiron-Leyraud et al. (2007); Bangura et al. (2007); LeBoeuf et al. (2007); Yelland et al. (2007); Jaudet et al. (2007) has completely changed the landscape of research in high- temperature superconductors (HTSs). In particular, the data suggest that the current carriers are both electrons and holes, when in fact the materials are hole doped —i.e., the current carriers should be positively charged. Moreover, the data cannot be reconciled with an important theorem about how electrons are organized in materials Luttinger (1960) unless one assumes that the signals arise from a combination of both holes and electrons. Until now, physicists have not been able to decide whether the cuprates, in Shakespeare’s terms, are camels or whales; in fact, these experiments foreshadow a remarkable degree of simplicity in these complex materials. The cuprates start out as insulators and become superconductors when doped with additional charge carriers. These so-called Mott insulators insulate by virtue of strong repulsive Coulomb interaction and need not break any symmetries in the lowest energy state, the ground state. A symmetry of a system is a transformation, such as a translation or a rotation, that keeps it unchanged. Such a symmetry is said to be broken, or spontaneously broken, if the system does not obey the symmetry of the underlying fundamental physical nature of the material; for example, a ferromagnet breaks the spin-rotational symmetry with its magnetization pointing in a definite direction. The notion of symmetry and broken symmetry finds many deep applications in physics. Soon after the discovery of the cuprate superconductors, Anderson proposed Anderson (1987) that their parent compounds begin as a featureless spin liquid that does not break any symmetries, called the resonating valence bond (RVB) state: “The preexisting magnetic singlet pairs of the insulating state become charged superconducting pairs when the insu lator is doped sufficiently strongly” Anderson (1987). Unfortunately, experiments show that the insulating phase is a simple antiferromagnet in which the spins are arranged in antiparallel manner, that is, with a broken symmetry. The materials remain antiferromagnets for a range of doping, and then, after a sequence of not well understood states as a function of doping, they become superconductors. How this plays out experimentally can be understood by looking at the Fermi surface, a fundamental concept in condensed matter physics. The Fermi surface differentiates the occupied electronic states from the unoccupied states (in coordinates of momentum rather than real space). Electrons fill the Fermi surface (FS) up to some highest occupied energy called the Fermi energy (see the figure). The excitations from the FS (e.g., when a current flows) are called Landau quasiparticles (quasi, because they are combinations or superpositions of real particles). The robustness of FS is due to its topological invariance, one of the most basic invariances in mathematical physics, which signifies stability with respect to “small deformations” Volovik:2003 . Even when the quasiparticles are absent due to electron- electron interactions, as in one-dimensional electronic systems, the FS is still defined by the same topological invariant. A reconstruction of this surface, such as a break up of a single surface into hole-like and electron- like pockets, requires a global deformation, most likely a broken symmetry. The new experimental work (2 6) yielded measurements of the oscillations that arise from energy levels created by imposing a magnetic field on the material (the Landau levels). As the magnetic field is increased, the highest fully occupied levels sweep past the Fermi energy, and the system periodically returns to itself, hence the oscillation in physical properties. The oscillations of the Hall resistance Doiron-Leyraud et al. (2007); LeBoeuf et al. (2007), capable of detecting the sign of the charge carriers, seem to show the presence of electron and hole pockets in the Fermi surface, suggesting that it undergoes some kind of reconstruction. One might complain that these high field measurements are still considerably below the upper critical field where superconductivity disappears (about 100 T or more) and are affected by the complex motion of vortices generated by the magnetic field. This may be true, but quantum oscillations in many superconductors are observed at fields as small as half the critical field, with the oscillation frequencies unchanged from the nonsuperconducting state (with an increased damping, however). It is also known that the quasiparticles of HTSs do not form Landau levels (9). Thus, it is very likely that the quantum oscillation experiments are accessing the normal state beyond the realm of superconductivity. But what kind of state? As the oscillations definitively point to both electron and hole pockets, it cannot be a conventional Fermi surface, rather one that has undergone a reconstruction due to a broken symmetry at variance with the RVB picture Chakravarty and Kee (2007). We may be finally beginning to understand these superconductors after two decades. The fly in the ointment is the lack of observation of electron and hole pockets in other measurements in hole-doped superconductors (in angle resolved photoemission spectroscopy, for instance) that are also capable of measuring Fermi surfaces [see, however, the work on electron-doped materials Armitage et al. (2001)]. Missing so far in experiments are also the higher frequency oscillations that must arise from the hole pockets, not just the electron pockets LeBoeuf et al. (2007). With further experimental work, we should be able to tell just what kind of animal we are dealing with. [Note added : A higher frequency is now seen and is believed to arise from an incommensurate order—S. Sebastian, N. Harrison, and G. Lonzarich, private communication.] This work was supported by NSF under grant: DMR-0705092. Special thanks are due to Hae-Young Kee for many discussions and collaborations. I would also like to thank E. Abrahams, N. P. Armitage, R. B. Laughlin, Z. Tesanovic, and J. Zaanen for important comments. ## References * (1) J. G. Bednorz and K. A. Müller, Z. Physik B 64, 189 (1986). * Doiron-Leyraud et al. (2007) N. Doiron-Leyraud, C. Proust, D. LeBoeuf, J. Levallois, J. B. Bonnemaison, R. X. Liang, D. A. Bonn, W. N. Hardy, and L. Taillefer, Nature 447, 565 (2007). * Bangura et al. (2007) A. F. Bangura, J. D. Fletcher, A. Carrington, J. Levallois, M. Nardone, B. Vignolle, P. J. Heard, N. Doiron-Leyraud, D. LeBoeuf, L. Taillefer, et al., arXiv:0707.4461 (2007). * LeBoeuf et al. (2007) D. LeBoeuf et al., Nature 450, 533 (2007). * Yelland et al. (2007) D. A. Yelland et al., arXiv:0707.0057 (2007). * Jaudet et al. (2007) C. Jaudet, D. Vignolles, A. Audouard, J. Levallois, D. LeBoeuf, N. Doiron-Leyraud, B. Vignolle, M. Nardone, A. Zitouni, R. Liang, et al., arXiv.org:0711.3559 (2007). * Luttinger (1960) J. M. Luttinger, Phys. Rev. 119, 1153 (1960). * Anderson (1987) P. W. Anderson, Science 235, 1196 (1987). * (9) G. E. Volovik, The universe in a Helium droplet (Cambridge University Press, Cambridge, 2003) * Franz and Tesanovic (2000) M. Franz and Z. Tesanovic, Phys. Rev. Lett. 84, 554 (2000). * Chakravarty and Kee (2007) S. Chakravarty and H.-Y. Kee, arXiv:0710.0608 (2007). * Armitage et al. (2001) N. P. Armitage, D. H. Lu, C. Kim, A. Damascelli, K. M. Shen, F. Ronning, D. L. Feng, P. Bogdanov, Z.-X. Shen, Y. Onose, et al., Phys. Rev. Lett. 87, 147003 (2001). Figure 1: Fermi surface reconstruction: In a crystalline lattice of periodicity $a$, the available quantum states are contained within the Brillouin zone (BZ). In (a) we have shown a two-dimensional example, $-\frac{\pi}{a}\leq k_{x}<\frac{\pi}{a}$, $-\frac{\pi}{a}\leq k_{y}<\frac{\pi}{a}$, which is pertinent to high temperature superconductors that consist of weakly coupled CuO-planes. For $N$-atoms, with unit cells containing 1 atom in a square $a\times a$, there are exactly $2N$ available states, as each state can be filled by electrons with two distinct spin orientations. The diamond in (a) is called the reduced Brillouin zone (RBZ) and contains exactly half the number of available states. In (a) the unoccupied states are colored red; these, holes, in a otherwise filled BZ, respond as charge carriers with positive sign in response to electric fields. The constant energy contours are shown as the set of black curves. The filled diamond corresponds to one electron per unit cell. The complementary sets within the BZ, when reassembled together, will form an identical diamond, which we may say is filled with holes. In (a) the red area corresponds to $(1+x)$ holes per unit cell of the crystal lattice. The excess, $x$, is called the doped holes. High temperature superconductors are schizophrenic. In some regimes they behave as though they consist of $(1+x)$ charge carriers, holes, while in the underdoped regime their properties are determined instead by $x$ doped holes. This is an important mystery. A class of theories posit that FS reconstructs in the uderdoped regime. Consider shifting the FS in (a) by vectors $(\pm\frac{\pi}{a},\pm\frac{\pi}{a})$, which will give rise to the Figure (b), ignoring the shading for clarity. Interesting quantum mechanical processes, about which there can be much debate, can result in reconnections shown in (c), as in a kaleidoscope. However, if we continued to consider the full BZ, we would double the number of states. All distinct states are contained in the RBZ, but there are now two distinct sets of energy levels, the upper band and the lower band. However, we continue to use the full BZ in (c), as a better aid for visualization. Because the RBZ is the fundamental unit in the wave vector space, the new unit cell of the crystal lattice is doubled, given by a square $\sqrt{2}a\times\sqrt{2}a$, and the full translational symmetry of the original lattice is broken. The FS now consists of disconnected sheets of blue and red areas. The remarkable fact is that the charge carriers in the blue region behave like electrons of fraction $n_{e}$ and in the red region like holes of fraction $n_{h}$. The doped holes is easily seen to be $x=2n_{h}-n_{e}$, as there are two hole pockets and one electron pocket in the RBZ. The broken symmetry invoked here is called commensurate, as the translational invariance of the crystal of integer multiples of the next nearest neighbor lattice vectors of the original lattice is still preserved. The broken symmetry can also be incommensurate with the original crystal lattice and can give rise to more complex FS reconstructions.

arxiv-papers

2024-09-04T02:48:53.646786

0802.1238

††thanks: Corresponding author, Electronic address: jljing@hunnu.edu.cn # Degradation of non-maximal entanglement of scalar and Dirac fields in non- inertial frames Qiyuan Pan Jiliang Jing Institute of Physics and Department of Physics, Hunan Normal University, Changsha, Hunan 410081, P. R. China and Key Laboratory of Low Dimensional Quantum Structures and Quantum Control of Ministry of Education, Hunan Normal University , Changsha, Hunan 410081, P.R. China ###### Abstract The entanglement between two modes of the free scalar and Dirac fields as seen by two relatively accelerated observers has been investigated. It is found that the same initial entanglement for an initial state parameter $\alpha$ and its “normalized partner” $\sqrt{1-\alpha^{2}}$ will be degraded by the Unruh effect along two different trajectories except for the maximally entangled state, which just shows the inequivalence of the quantization for a free field in the Minkowski and Rindler coordinates. In the infinite acceleration limit the state doesn’t have the distillable entanglement for any $\alpha$ for the scalar field but always remains entangled to a degree which is dependent of $\alpha$ for the Dirac field. It is also interesting to note that in this limit the mutual information equals to just half of the initially mutual information, which is independent of $\alpha$ and the type of field. ###### pacs: 03.65.Ud, 03.67.Mn, 04.70.Dy, 97.60.Lf The quantum information theory has made rapid progress in recent years and more and more efforts have been expended on the study of quantum information in the relativistic framework PT-B-P-E . Especially, the entanglement in a relativistic setting has received considerable attention because it is considered to be a major resource for quantum information tasks such as quantum teleportation, quantum computation and so on DAA . Despite the potential interest to quantum information, the study of entanglement can also help us get a deeper understanding of the black hole thermodynamics Bombelli- Callen and the black hole information paradox Hawking-Terashima . Thus, many authors have investigated the entanglement in the relativistic frames inertial or not for various fields PST ; Mann ; Alsing . More recently, Fuentes-Schuller _et al._ Mann and Alsing _et al._ Alsing explicitly demonstrated that the entanglement is a quantity depending on a relative acceleration of one of the observers who, before being accelerated, shared a maximally entangled bosonic or fermionic pair. Their results also showed that the different type of field will have a qualitatively different effect on the degradation of entanglement produced by the Unruh effect Davies- Unruh . Choosing a generic state as the initial entangled state in this Brief Report $\displaystyle|\Psi_{sk}\rangle=\sqrt{1-\alpha^{2}}|0_{s}\rangle^{M}|1_{k}\rangle^{M}+\alpha|1_{s}\rangle^{M}|0_{k}\rangle^{M},$ (1) where $\alpha$ is some real number which satisfies $|\alpha|\in(0,1)$, $\alpha$ and $\sqrt{1-\alpha^{2}}$ are the so-called “normalized partners”, we will try to see what effects this uncertainly initial entangled state will on the degradation of entanglement for two relatively accelerated observers due to the presence of an initial state parameter $\alpha$. Notice that the Schwarzschild space-time very close to the event horizon resembles the Rindler space in the infinite acceleration limit Wald ; Mann . Hence, as in Mann ; Alsing our results in this limit can be applied to discuss the entanglement between two free bosonic or fermionic modes seen by observers when one observer falls into a black hole and the other barely escapes through eternal uniform acceleration. Rindler coordinates are appropriate for describing the viewpoint of an observer moving with uniform acceleration. The world lines of uniformly accelerated observers in the Minkowski coordinates correspond to hyperbolae in the left (region I) and right (region II) of the origin which are bounded by light-like asymptotes constituting the Rindler horizon Mann ; Alsing , so two Rindler regions are causally disconnected from each other Birrell . An observer undergoing uniform acceleration remains constrained to either Rindler region I or II and has no access to the other sector. The system in Eq. (1) is bipartite from an inertial perspective, but in a non-inertial frame an extra set of modes in region II becomes relevant. Thus, we will study the mixed- state entanglement of the state as seen by an inertial observer Alice detecting the mode $s$ and a uniformly accelerated observer Bob with proper acceleration $a$ in region I detecting the second mode $k$. Bosonic entanglement For a free scalar field, the Minkowski vacuum state can be expressed as a two-mode squeezed state in the Rindler frame Davies-Unruh ; Birrell $\displaystyle|0_{k}\rangle^{M}=\frac{1}{\cosh r}\sum_{n=0}^{\infty}\tanh^{n}r|n_{k}\rangle_{I}|n_{k}\rangle_{II},$ (2) where $\cosh r=(1-e^{-2\pi|k|c/a})^{-1/2}$, $k$ is the wave vector and $r$ is the acceleration parameter, $|n\rangle_{I}$ and $|n\rangle_{II}$ indicate the Rindler-region-I-particle mode and -II-antiparticle mode respectively. Using Eq. (2) and the first excited state Mann ; Birrell $\displaystyle|1_{k}\rangle^{M}=\frac{1}{\cosh^{2}r}\sum_{n=0}^{\infty}\tanh^{n}r\sqrt{n+1}|(n+1)_{k}\rangle_{I}|n_{k}\rangle_{II},$ we can rewrite Eq. (1) in terms of Minkowski modes for Alice and Rindler modes for Bob. Since Bob is causally disconnected from region II, we will trace over the states in this region and obtain $\displaystyle\rho_{AB}=\frac{1}{\cosh^{2}r}\sum_{n=0}^{\infty}\tanh^{2n}r\rho_{n},$ $\displaystyle\rho_{n}=\alpha^{2}|1n\rangle\langle 1n|+\frac{\alpha\sqrt{(1-\alpha^{2})(n+1)}}{\cosh r}|1n\rangle\langle 0(n+1)|$ $\displaystyle\qquad+\frac{\alpha\sqrt{(1-\alpha^{2})(n+1)}}{\cosh r}|0(n+1)\rangle\langle 1n|$ $\displaystyle\qquad+\frac{(1-\alpha^{2})(n+1)}{\cosh^{2}r}|0(n+1)\rangle\langle 0(n+1)|,$ (3) where $|nm\rangle=|n_{s}\rangle^{M}|m_{k}\rangle_{I}$. The partial transpose criterion provides a sufficient condition for the existence of entanglement in this case peres : if at least one eigenvalue of the partial transpose is negative, the density matrix is entangled; but a state with positive partial transpose can still be entangled. It is well-known bound or nondistillable entanglement Vidal-Plenio . Interchanging Alice’s qubits, we get the eigenvalues of the partial transpose $\rho_{AB}^{T_{A}}$ in the ($n$,$n+1$) block $\displaystyle\lambda_{\pm}^{n}=\frac{\tanh^{2n}r}{2\cosh^{2}r}\left[\xi_{n}\pm\sqrt{\xi_{n}^{2}+\frac{4\alpha^{2}(1-\alpha^{2})}{\cosh^{2}r}}\right],$ where $\xi_{n}=\alpha^{2}\tanh^{2}r+(1-\alpha^{2})n/\sinh^{2}r$. Obviously the eigenvalue $\lambda_{-}^{n}$ is always negative for finite acceleration ($r<\infty$). Hence, this mixed state is always entangled for any finite acceleration of Bob. In the limit $r\rightarrow\infty$, the negative eigenvalue will go to zero. To discuss this further, we will use the logarithmic negativity which serves as an upper bound on the entanglement of distillation Vidal-Plenio . This entanglement monotone is defined as $N(\rho)=\log_{2}||\rho^{T}||$, where $||\rho^{T}||$ is the trace norm of the partial transpose $\rho^{T}$. We therefore find $\displaystyle N(\rho_{AB})=\log_{2}\left\\{\frac{\alpha^{2}}{\cosh^{2}{r}}+\sum_{n=0}^{\infty}\frac{\tanh^{2n}r}{\cosh^{2}r}\right.$ $\displaystyle\times\left.\sqrt{\left[\alpha^{2}\tanh^{2}r+\frac{(1-\alpha^{2})n}{\sinh^{2}r}\right]^{2}+\frac{4\alpha^{2}(1-\alpha^{2})}{\cosh^{2}r}}\right\\}.$ For vanishing acceleration ($r=0$), $N(\rho_{AB})=\log_{2}(1+2|\alpha|\sqrt{1-\alpha^{2}})$. In the range $0<|\alpha|\leq 1/\sqrt{2}$ the larger $\alpha$, the stronger the initial entanglement; but in the range $1/\sqrt{2}\leq|\alpha|<1$ the larger $\alpha$, the weaker the initial entanglement. For finite acceleration, the monotonous decrease of $N(\rho_{AB})$ with increasing $r$ for different $\alpha$ means that the entanglement of the initial state is lost to the thermal fields generated by the Unruh effect. From Fig. 1 it is surprisingly found that the same initial entanglement for $\alpha$ and its “normalized partner” $\sqrt{1-\alpha^{2}}$ will be degraded along two different trajectories except for the maximally entangled state, i.e., $|\alpha|=1/\sqrt{2}$. This phenomenon, due to the coupling of $\alpha$ and the hyperbolic functions related to $r$, just shows the inequivalence of the quantization for a scalar field in the Minkowski and Rindler coordinates. The logarithmic negativity is exactly zero for any $\alpha$ in the limit $r\rightarrow\infty$, which indicates that the state doesn’t have the distillable entanglement. Figure 1: Logarithmic negativity of the bosonic modes versus $r$ for different $\alpha$. The mutual information, which can be used to estimate the total amount of correlations between any two subsystem of the overall system, is defined as RAM $\displaystyle I(\rho_{AB})=S(\rho_{A})+S(\rho_{B})-S(\rho_{AB}),$ (4) where $S(\rho)=-\text{Tr}(\rho\log_{2}\rho)$ is the entropy of the density matrix $\rho$. From Eq. (Degradation of non-maximal entanglement of scalar and Dirac fields in non-inertial frames), we can obtain the entropy of this joint state $\displaystyle S(\rho_{AB})=-\sum_{n=0}^{\infty}\frac{\tanh^{2n}r}{\cosh^{2}r}\left[\alpha^{2}+\frac{(1-\alpha^{2})(n+1)}{\cosh^{2}r}\right]$ $\displaystyle\qquad\times\log_{2}\frac{{\tanh^{2n}r}}{\cosh^{2}r}\left[\alpha^{2}+\frac{(1-\alpha^{2})(n+1)}{\cosh^{2}r}\right].$ (5) Tracing over Alice’s states for $\rho_{AB}$, we get Bob’s density matrix in region I; its entropy is $\displaystyle S(\rho_{BI})=-\sum_{n=0}^{\infty}\frac{\tanh^{2n}r}{\cosh^{2}r}\left[\alpha^{2}+\frac{(1-\alpha^{2})n}{\sinh^{2}r}\right]$ $\displaystyle\qquad\times\log_{2}\frac{\tanh^{2n}r}{\cosh^{2}r}\left[\alpha^{2}+\frac{(1-\alpha^{2})n}{\sinh^{2}r}\right].$ (6) In the same way, we have Alice’s density matrix by tracing over Bob’s states; its entropy is given by $\displaystyle S(\rho_{A})=-[\alpha^{2}\log_{2}\alpha^{2}+(1-\alpha^{2})\log_{2}(1-\alpha^{2})].$ (7) We draw the behaviors of the mutual information $I(\rho_{AB})$ versus $r$ for different $\alpha$ in Fig. 2. For vanishing acceleration, the initially mutual information is $I_{bi}=-2[\alpha^{2}\log_{2}\alpha^{2}+(1-\alpha^{2})\log_{2}(1-\alpha^{2})]$. In the range $0<|\alpha|\leq 1/\sqrt{2}$ the larger $\alpha$, the stronger $I_{bi}$; but in the range $1/\sqrt{2}\leq|\alpha|<1$ the larger $\alpha$, the weaker $I_{bi}$. As the acceleration increases, the mutual information becomes smaller. It is interesting to note that except for the maximally entangled state, the same initially mutual information for $\alpha$ and $\sqrt{1-\alpha^{2}}$ will be degraded along two different trajectories. However, in the infinite acceleration limit, the mutual information converges to the same value again, i.e., $I_{bf}=-[\alpha^{2}\log_{2}\alpha^{2}+(1-\alpha^{2})\log_{2}(1-\alpha^{2})]$, which equals to just half of $I_{bi}$. Obviously if $I_{bi}$ is higher, it is degraded to a higher degree in this limit. Since the distillable entanglement in the infinite acceleration limit is zero, we are safe to say that the total correlations consist of classical correlations plus bound entanglement in this limit. Figure 2: Mutual information of the bosonic modes versus $r$ for different $\alpha$. Fermionic entanglement With the single-mode approximation used by Alsing _et al._ , the fermionic Minkowski vacuum can be written as Alsing $\displaystyle|0\rangle^{M}=\cos r|0\rangle_{I}|0\rangle_{II}+\sin r|1\rangle_{I}|1\rangle_{II},$ (8) and the only excited state is given by $\displaystyle|1\rangle^{M}=|1\rangle_{I}|0\rangle_{II},$ (9) where $\cos r=(1+e^{-2\pi\omega c/a})^{-1/2}$ and the acceleration parameter $r$ is in the range $0\leq r\leq\pi/4$ for $0\leq a\leq\infty$ in this case. Using Eq. (8) and (9) for the Minkowski particle states $|0_{k}\rangle^{M}$ and $|1_{k}\rangle^{M}$ and tracing over the modes in the region II, we get $\displaystyle\rho_{AB}=(1-\alpha^{2})|01\rangle\langle 01|$ $\displaystyle\qquad~{}+\alpha\sqrt{(1-\alpha^{2})}\cos r(|01\rangle\langle 10|+|10\rangle\langle 01|)$ $\displaystyle\qquad~{}+\alpha^{2}(\cos^{2}r|10\rangle\langle 10|+\sin^{2}r|11\rangle\langle 11|),$ (10) with $|mn\rangle=|m\rangle_{A}^{M}|n\rangle_{BI}$. The partial transpose criterion provides a necessary and sufficient condition for entanglement in a mixed state of two qubits peres : if at least one eigenvalue of the partial transpose is negative, the density matrix is entangled. Interchanging Alice’s qubits, we obtain an eigenvalue of the partial transpose $\rho_{AB}^{T_{A}}$ $\displaystyle\lambda_{-}=\frac{1}{2}\left[\alpha^{2}\sin^{2}{r}-\sqrt{\alpha^{4}\sin^{4}{r}+4\alpha^{2}(1-\alpha^{2})\cos^{2}{r}}\right],$ which is always negative for $0\leq r\leq\pi/4$. Thus, the state is always entangled for any uniform acceleration of Bob. The logarithmic negativity is expressed as $\displaystyle N(\rho_{AB})=\log_{2}[1-\alpha^{2}\sin^{2}{r}$ $\displaystyle\quad\quad+\sqrt{\alpha^{4}\sin^{4}{r}+4\alpha^{2}(1-\alpha^{2})\cos^{2}{r}}~{}].$ Figure 3: Logarithmic negativity of the fermionic modes versus $r$ for different $\alpha$ (notice that $\alpha_{N_{m}}=\sqrt{(4-\sqrt{2})/7}$ ). For vanishing acceleration ($r=0$), $N(\rho_{AB})=\log_{2}(1+2|\alpha|\sqrt{1-\alpha^{2}})$. For finite acceleration, the entanglement is degraded by the Unruh effect just as shown in Fig. 3. We find that in the range $0<|\alpha|\leq 1/\sqrt{2}$ for the larger $\alpha$, the initial entanglement is higher, but it isn’t always degraded to a higher degree. It should be noted that for $1/2<|\alpha|<1/\sqrt{2}$ the final entanglement of the initial state is higher than that of the maximally entangled state, i.e., $\log_{2}(3/2)\simeq 0.585$, and for $|\alpha|=\sqrt{(4-\sqrt{2})/7}$ the maximally final entanglement is $\log_{2}[(5+4\sqrt{2})/7]\simeq 0.606$. In the range $1/\sqrt{2}\leq|\alpha|<1$ the larger $\alpha$, the weaker the initial entanglement and the lower the final entanglement. Unlike the behaviors of the bosonic case, except for the maximally entangled state, the same initial entanglement of the fermionic modes for $\alpha$ and $\sqrt{1-\alpha^{2}}$ will be degraded along two different trajectories and asymptotically reach two differently nonvanishing minimum values in the infinite acceleration limit ($r=\pi/4$) due to the coupling of $\alpha$ and the trigonometric functions related to $r$. In the infinite acceleration limit $N(\rho_{AB})=\log_{2}(1-\alpha^{2}/2+|\alpha|\sqrt{2-7\alpha^{2}/4})\neq 0$, which means that the state is always entangled. This is in strong contrast to the bosonic case and shows that the fermionic system can be used as a resource for performing certain quantum information processing tasks. Similar to the bosonic case, we go through the same process again and get the mutual information for these fermionic modes $\displaystyle I(\rho_{AB})=(1-\alpha^{2}\sin^{2}{r})\log_{2}(1-\alpha^{2}\sin^{2}{r})$ $\displaystyle\quad\quad+\alpha^{2}\sin^{2}{r}\log_{2}\alpha^{2}\sin^{2}{r}$ $\displaystyle\quad\quad-(1-\alpha^{2}\cos^{2}{r})\log_{2}(1-\alpha^{2}\cos^{2}{r})$ $\displaystyle\quad\quad-\alpha^{2}\cos^{2}{r}\log_{2}\alpha^{2}\cos^{2}{r}$ $\displaystyle\quad\quad-\alpha^{2}\log_{2}\alpha^{2}-(1-\alpha^{2})\log_{2}(1-\alpha^{2}),$ (11) whose trajectories versus $r$ for different $\alpha$ are shown by Fig. 4. Figure 4: Mutual information of the fermionic modes versus $r$ for different $\alpha$. For vanishing acceleration, the initially mutual information is $I_{fi}=-2[\alpha^{2}\log_{2}\alpha^{2}+(1-\alpha^{2})\log_{2}(1-\alpha^{2})]$, whose behaviors are the same to $I_{bi}$ of the bosonic modes. The mutual information becomes smaller as the acceleration increases, and again we surprisingly find that the same initially mutual information for $\alpha$ and $\sqrt{1-\alpha^{2}}$ will be degraded along two different trajectories except for the maximally entangled state. In the infinite acceleration limit the mutual information converges to $I_{ff}=-[\alpha^{2}\log_{2}\alpha^{2}+(1-\alpha^{2})\log_{2}(1-\alpha^{2})]$, which is just half of $I_{fi}$. This behavior is reminiscent of that seen for the bosonic case, so we conclude that $\displaystyle I_{f}=\frac{1}{2}I_{i},$ (12) which is independent of $\alpha$ and the type of field. It should be noted that if we set the initial entangled state as $\displaystyle|\Psi_{sk}\rangle=\alpha|0_{s}\rangle^{M}|0_{k}\rangle^{M}+\sqrt{1-\alpha^{2}}|1_{s}\rangle^{M}|1_{k}\rangle^{M},$ (13) we will have the same behavior of the entanglement degradation for the same $\alpha$ just as shown in Figs. 1-4. Summarizing, the entanglement of the scalar and Dirac fields in non-inertial frames is degraded by the Unruh effect as the Bob’s rate of acceleration increases, but their behaviors of the degradation of entanglement are different for the same initial state parameter $\alpha$. It is surprisingly found that the same initial entanglement for $\alpha$ and $\sqrt{1-\alpha^{2}}$ will be degraded along two different trajectories except for the maximally entangled state, which just shows the inequivalence of the quantization for a free field in the Minkowski and Rindler coordinates. In the infinite acceleration limit, which can be applied to the case Alice falling into a black hole while Bob barely escapes, the state doesn’t have the distillable entanglement for any $\alpha$ for the scalar field but always remains entangled to a degree which is dependent of $\alpha$ for the Dirac field. It should be noted that for $|\alpha|=\sqrt{(4-\sqrt{2})/7}$, we will have the maximally final entanglement for the fermionic state in this limit. Further analysis shows that the mutual information is degraded to a nonvanishing minimum value which is dependent of $\alpha$ for these two fields with increasing acceleration parameter $r$. However, it is interesting to note that the mutual information in the infinite acceleration limit equals to just half of the initially mutual information, which is independent of $\alpha$ and the type of field. This work was supported by the National Natural Science Foundation of China under Grant No. 10675045; the FANEDD under Grant No. 200317; the Hunan Provincial Natural Science Foundation of China under Grant No. 07A0128; and the construct program of the key discipline in hunan province. ## References * (1) A. Peres and D. R. Terno, Rev. Mod. Phys. 76, 93 (2004); D. Boschi, S. Branca, F. De Martini, L. Hardy, and S. Popescu, Phys. Rev. Lett. 80, 1121 (1998); J. W. Pan, C. Simon, C. Brukner, and A. Zeilinger, Nature 410, 1067 (2001); S. J. van Enk and T. Rudolph, Quant. Inf. Comput. 3, 423 (2003). * (2) D. Bouwmeester, A. Ekert, and A. Zeilinger, The Physics of Quantum Information (Springer-Verlag, Berlin), 2000. * (3) L. Bombelli, R. K. Koul, J. Lee, and R. Sorkin, Phys. Rev. D 34, 373 (1986); C. Callen and F. Wilzcek, Phys. Lett. B 333, 55 (1994). * (4) S. W. Hawking, Commun. Math. Phys. 43, 199 (1975); Phys. Rev. D 14, 2460 (1976); H. Terashima, Phys. Rev. D 61, 104016 (2000). * (5) A. Peres, P. F. Scudo, and R. Terno, Phys. Rev. Lett. 88, 230402 (2002); R. M. Gingrich and C. Adami, Phys. Rev. Lett. 89, 270402 (2002); P. M. Alsing and G. J. Milburn, Phys. Rev. Lett. 91, 180404 (2003). * (6) I. Fuentes-Schuller and R. B. Mann, Phys. Rev. Lett. 95, 120404 (2005). * (7) P. M. Alsing, I. Fuentes-Schuller, R. B. Mann, and T. E. Tessier, Phys. Rev. A 74, 032326 (2006). * (8) P. C. W. Davies, J. Phys. A 8, 609 (1975); W. G. Unruh, Phys. Rev. D 14, 870 (1976). * (9) R. M. Wald, General Relativity (University of Chicago Press, Chicago), 1984. * (10) N. D. Birrell and P. C. W. Davies, Quantum fields in curved space (Cambridge University Press, New York), 1982. * (11) A. Peres, Phys. Rev. Lett. 77, 1413 (1996). * (12) G. Vidal and R. F. Werner, Phys. Rev. A 65, 032314 (2002); M. B. Plenio, Phys. Rev. Lett. 95, 090503 (2005). * (13) R. S. Ingarden, A. Kossakowski, and M. Ohya, Information Dynamics and Open Systems - Classical and Quantum Approach (Kluwer Academic Publishers, Dordrecht), 1997.

arxiv-papers

2024-09-04T02:48:53.650482

0802.1263

# Leibniz algebra deformations of a Lie algebra Alice Fialowski and Ashis Mandal The work was partially supported by grants from INSA India and HAS Hungary. ###### Abstract In this note we compute Leibniz algebra deformations of the $3$-dimensional nilpotent Lie algebra $\mathfrak{n}_{3}$ and compare it with its Lie deformations. It turns out that there are $3$ extra Leibniz deformations. We also describe the versal Leibniz deformation of $\mathfrak{n}_{3}$ with the versal base. Keywords: Nilpotent Lie algebra, Leibniz algebra, cohomology, infinitesimal, versal deformation. Mathematics Subject Classifications (2000): $13$D$10$, $14$D$15$, $13$D$03$. ## 1 Introduction Since a Lie algebra is also a Leibniz algebra, a natural question arises. If we consider a Lie algebra as a Leibniz algebra and compute its Leibniz algebra deformations, is it true that we can get more Leibniz algebra deformations, than just the Lie deformations of the original Lie algebra ? In this note we will demonstrate the problem on a three dimensional Lie algebra example for which we completely describe its versal Lie deformation and versal Leibniz deformation. It turns out that beside the Lie deformations we get three non-equivalent Leibniz deformations which are not Lie algebras. Our example is the following. Consider a three dimensional vector space $L$ spanned by $\\{e_{1},~{}e_{2},~{}e_{3}\\}$ over $\mathbb{C}$. Define the Heisenberg Lie algebra $\mathfrak{n}_{3}$ on it with the bracket matrix $A=\left(\begin{array}[]{lll}0&0&1\\\ 0&0&0\\\ 0&0&0\end{array}\right).$ Here the columns are the Lie brackets $[e_{1},e_{2}],[e_{1},e_{3}]$ and $[e_{2},e_{3}]$. That means $[e_{2},e_{3}]=e_{1}$ and all the other brackets are zero (except of course $[e_{3},e_{2}]=-e_{1}$). This is the only nilpotent three dimensional Lie algebra. We compute infinitesimal Lie and Leibniz deformations and show that there are three additional Leibniz cocycles beside the five Lie cocycles. We will show that all these infinitesimal deformations are extendable without any obstructions. We also describe the versal Leibniz deformation. The structure of the paper is as follows. In Section $1$ we recall the necessary preliminaries about Lie and Leibniz cohomology and deformations. In Section $2$ we recall the classification of three dimensional Lie algebras and describe all non-equivalent deformations of the nilpotent Lie algebra $\mathfrak{n}_{3}$ . In Section $3$ we give the classification of three dimensional nilpotent Leibniz algebras. Among those $\mathfrak{n}_{3}$ is the only nontrivial Lie algebra. Then we compute Leibniz cohomology and give explicitly all non-equivalent infinitesimal deformations. In Section $4$ we show that all infinitesimal deformations are extendable as all the Massey squares turn out to be zero. We identify our deformations with the classified objects. Finally we show that the versal Leibniz deformation is the universal infinitesimal one, and describe the base of the versal deformation. ## 2 Preliminaries Let us recall first the Lie algebra cohomology. ###### Definition 2.1. Suppose $\mathfrak{g}$ is a Lie algebra and $A$ is a module over $\mathfrak{g}$. Then a $q$-dimensional cochain of the Lie algebra $\mathfrak{g}$ with coefficients in $A$ is a skew -symmetric $q$-linear map on $\mathfrak{g}$ with values in $A$; the space of all such cochains is denoted by $C^{q}(\mathfrak{g};A)$. Thus, $C^{q}(\mathfrak{g};A)=Hom(\Lambda^{q}\mathfrak{g},A)$; this last representation transforms $C^{q}(\mathfrak{g};A)$ into a $\mathfrak{g}$-module. The differential $d=d_{q}:C^{q}(\mathfrak{g};A)\longrightarrow C^{q+1}(\mathfrak{g};A)$ is defined by the formula $\begin{split}dc(g_{1},\cdots,g_{q+1})=&\sum_{1\leq s<t\leq q+1}(-1)^{s+t-1}c([g_{s},g_{t}],g_{1},\cdots,\hat{g}_{s},\cdots,\hat{g}_{t},\cdots,g_{q+1})\\\ &\sum_{1\leq s\leq q+1}(-1)^{s}[g_{s},c(g_{1},\cdots,\hat{g}_{s},\cdots,g_{q+1})],\end{split}$ where $c\in C^{q}(\mathfrak{g};A)$ and $g_{1},\cdots,g_{q+1}\in\mathfrak{g}$. We complete the definition by putting $C^{q}(\mathfrak{g};A)=0$ for $q<0$, $d_{q}=0$ for $q<0$. As can be easily checked, $d_{q+1}\circ d_{q}=0$ for all $q$, so that $(C^{q}(\mathfrak{g},d_{q}))$ is an algebraic complex; this complex is denoted by $C^{*}((\mathfrak{g};A))$, while the corresponding cohomology is referred to as the cohomology of the Lie algebra $\mathfrak{g}$ with coefficients in $A$ and is denoted by $H^{q}(\mathfrak{g};A)$. Leibniz algebras were introduced by J.-L. Loday [7, 9]. Let $\mathbb{K}$ denote a field. ###### Definition 2.2. A Leibniz algebra is a $\mathbb{K}$-module $L$, equipped with a bracket operation that satisfies the Leibniz identity: $[x,[y,z]]=[[x,y],z]-[[x,z],y],~{}~{}\mbox{for}~{}x,~{}y,~{}z\in L.$ Any Lie algebra is automatically a Leibniz algebra, as in the presence of antisymmetry, the Jacobi identity is equivalent to the Leibniz identity. More examples of Leibniz algebras were given in [7, 8, 9], and recently for instance in [2, 1]. Let $L$ be a Leibniz algebra and $M$ a representation of $L$. By definition, $M$ is a $\mathbb{K}$-module equipped with two actions (left and right) of $L$, $[-,-]:L\times M\longrightarrow M~{}~{}\mbox{and}~{}[-,-]:M\times L\longrightarrow M~{}~{}\mbox{such that}~{}$ $[x,[y,z]]=[[x,y],z]-[[x,z],y]$ holds, whenever one of the variables is from $M$ and the two others from $L$. Define $CL^{n}({L};{M}):=\mbox{Hom}_{\mathbb{K}}({L}^{\otimes n},{M}),~{}n\geq 0.$ Let $\delta^{n}:CL^{n}({L};{M})\longrightarrow CL^{n+1}(L;M)$ be a $\mathbb{K}$-homomorphism defined by $\begin{split}&\delta^{n}f(x_{1},\cdots,x_{n+1})\\\ &:=[x_{1},f(x_{2},\cdots,x_{n+1})]+\sum_{i=2}^{n+1}(-1)^{i}[f(x_{1},\cdots,\hat{x}_{i},\cdots,x_{n+1}),x_{i}]\\\ &+\sum_{1\leq i<j\leq n+1}(-1)^{j+1}f(x_{1},\cdots,x_{i-1},[x_{i},x_{j}],x_{i+1},\cdots,\hat{x}_{j},\cdots,x_{n+1}).\end{split}$ Then $(CL^{*}(L;M),\delta)$ is a cochain complex, whose cohomology is called the cohomology of the Leibniz algebra $L$ with coefficients in the representation $M$. The $n$th cohomology is denoted by $HL^{n}(L;M)$. In particular, $L$ is a representation of itself with the obvious action given by the bracket in $L$. The $n$th cohomology of $L$ with coefficients in itself is denoted by $HL^{n}(L;L).$ Let $S_{n}$ be the symmetric group of $n$ symbols. Recall that a permutation $\sigma\in S_{p+q}$ is called a $(p,q)$-shuffle, if $\sigma(1)<\sigma(2)<\cdots<\sigma(p)$, and $\sigma(p+1)<\sigma(p+2)<\cdots<\sigma(p+q)$. We denote the set of all $(p,q)$-shuffles in $S_{p+q}$ by $Sh(p,q)$. For $\alpha\in CL^{p+1}(L;L)$ and $\beta\in CL^{q+1}(L;L)$, define $\alpha\circ\beta\in CL^{p+q+1}(L;L)$ by $\begin{split}&\alpha\circ\beta(x_{1},\ldots,x_{p+q+1})\\\ =&~{}\sum_{k=1}^{p+1}(-1)^{q(k-1)}\\{\sum_{\sigma\in Sh(q,p-k+1)}sgn(\sigma)\alpha(x_{1},\ldots,x_{k-1},\beta(x_{k},x_{\sigma(k+1)},\ldots,x_{\sigma(k+q)}),\\\ &~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}x_{\sigma(k+q+1)},\ldots,x_{\sigma(p+q+1)})\\}.\end{split}$ The graded cochain module $CL^{*}(L;L)=\bigoplus_{p}CL^{p}(L;L)$ equipped with the bracket defined by $[\alpha,\beta]=\alpha\circ\beta+(-1)^{pq+1}\beta\circ\alpha~{}~{}\mbox{for}~{}\alpha\in CL^{p+1}(L;L)~{}~{}\mbox{and}~{}\beta\in CL^{q+1}(L;L)$ and the differential map $d$ by $d\alpha=(-1)^{|\alpha|}\delta\alpha~{}\mbox{for}~{}\alpha\in CL^{*}(L;L)$ is a differential graded Lie algebra. Let now $\mathbb{K}$ a field of zero characteristic and the tensor product over $\mathbb{K}$ will be denoted by $\otimes$. We recall the notion of deformation of a Lie (Leibniz) algebra $\mathfrak{g}$ ($L$) over a commutative algebra base $A$ with a fixed augmentation $\varepsilon:{A}\rightarrow\mathbb{K}$ and maximal ideal $\mathfrak{M}$. Assume $dim(\mathfrak{M}^{k}/\mathfrak{M}^{k+1})<\infty$ for every $k$ (see [5]). ###### Definition 2.3. A deformation $\lambda$ of a Lie algebra $\mathfrak{g}$ (or a Leibniz algebra ${L}$) with base $({A},\mathfrak{M})$, or simply with base ${A}$ is an $A$-Lie algebra (or an ${A}$-Leibniz algebra) structure on the tensor product ${A}\otimes{\mathfrak{g}}$ (or $A\otimes L$) with the bracket $[,]_{\lambda}$ such that $\varepsilon\otimes id:{A}\otimes{\mathfrak{g}}\rightarrow\mathbb{K}\otimes{\mathfrak{g}}~{}~{}(\mbox{or}~{}\varepsilon\otimes id:{A}\otimes{L}\rightarrow\mathbb{K}\otimes{L})$ is an $A$-Lie algebra (${A}$-Leibniz algebra) homomorphism. A deformation of the Lie (Leibniz) algebra $\mathfrak{g}$ ($L$) with base $A$ is called infinitesimal, or first order, if in addition to this $\mathfrak{M}^{2}=0$. We call a deformation of order k, if $\mathfrak{M}^{k+1}=0$. A deformation with base is called local if $A$ is a local algebra over $\mathbb{K}$, which means $A$ has a unique maximal ideal. Suppose $A$ is a complete local algebra ( $A=\mathop{\varprojlim}\limits_{n\rightarrow\infty}({A}/{\mathfrak{M}^{n}})$), where $\mathfrak{M}$ is the maximal ideal in $A$. Then a deformation of $\mathfrak{g}$ ($L$) with base $A$ which is obtained as the projective limit of deformations of $\mathfrak{g}$ $(L)$ with base $A/\mathfrak{M}^{n}$ is called a formal deformation of $\mathfrak{g}$ $(L)$. ###### Definition 2.4. Suppose $\lambda$ is a given deformation of $L$ with base $(A,\mathfrak{M})$ and augmentation $\varepsilon:{A}\rightarrow\mathbb{K}$. Let $A^{\prime}$ be another commutative algebra with identity and a fixed augmentation $\varepsilon^{\prime}:{A^{\prime}}\rightarrow\mathbb{K}$. Suppose $\phi:A\rightarrow A^{\prime}$ is an algebra homomorphism with $\phi(1)=1$ and $\varepsilon^{\prime}\circ\phi=\varepsilon$. Let $ker(\varepsilon^{\prime})=\mathfrak{M}^{\prime}$. Then the push-out $\bf{\phi_{*}\lambda}$ is the deformation of $L$ with base $(A^{\prime},\mathfrak{M}^{\prime})$ and bracket $[{a_{1}}^{\prime}\otimes_{A}(a_{1}\otimes{l_{1}}),a_{2}^{\prime}\otimes_{A}(a_{2}\otimes l_{2})]_{\phi_{*}\lambda}=a_{1}^{\prime}a_{2}^{\prime}\otimes_{A}[a_{1}\otimes l_{1},a_{2}\otimes l_{2}]_{\lambda}$ where $a_{1}^{\prime},a_{2}^{\prime}\in{A}^{\prime},~{}a_{1},a_{2}\in A$ and $l_{1},l_{2}\in L$. Here $A^{\prime}$ is considered as an $A$-module by the map $a^{\prime}\cdot a=a^{\prime}\phi(a)$ so that $A^{\prime}\otimes L=(A^{\prime}{\otimes}_{A}A)\otimes L=A^{\prime}{\otimes}_{A}(A\otimes L).$ ###### Definition 2.5. (see [3]) Let $C$ be a complete local algebra. A formal deformation $\eta$ of a Lie algebra $\mathfrak{g}$ (Leibniz algebra $L$) with base $C$ is called versal, if (i) for any formal deformation $\lambda$ of $\mathfrak{g}$ ($L$) with base $A$ there exists a homomorphism $f:C\rightarrow A$ such that the deformation $\lambda$ is equivalent to $f_{*}\eta$; (ii) if $A$ satisfies the condition ${\mathfrak{M}}^{2}=0$, then $f$ is unique. ###### Theorem 2.6. If $H^{2}(\mathfrak{g};\mathfrak{g})$ is finite dimensional, then there exists a versal deformation of $\mathfrak{g}$ (similarly for $L$). ###### Proof. Follows from the general theorem of Schlessinger [11], like it was shown for Lie algebras in [3]. ∎ In [4] a construction for a versal deformation of a Lie algebra was given and it was generalized to Leibniz algebras in [5]. The computation of a specific example is given in [10]. Let us describe the universal infinitesimal deformation (see [4] and [5]). For simplicity we will only discuss the Leibniz algebra case. Assume that $dim(HL^{2}(L;L))<\infty$. Denote the space $HL^{2}(L;L)$ by $\mathbb{H}$. Consider the algebra $C_{1}=\mathbb{K}\oplus\mathbb{H}^{\prime}$ where $\mathbb{H}^{\prime}$ is the dual of $\mathbb{H}$ , by setting $(k_{1},h_{1})\cdot(k_{2},h_{2})=(k_{1}k_{2},k_{1}h_{2}+k_{2}h_{1})~{}\mbox{for}~{}(k_{1},h_{1}),(k_{2},h_{2})\in C_{1}.$ Observe that the second summand is an ideal of $C_{1}$ with zero multiplication. Fix a homomorphism $\mu:\mathbb{H}\longrightarrow CL^{2}(L;L)=Hom(L^{\otimes 2};L)$ which takes a cohomology class into a cocycle representing it. Notice that there is an isomorphism $\mathbb{H}^{\prime}\otimes L\cong Hom(\mathbb{H}~{};L)$, so we have $C_{1}\otimes L=(\mathbb{K}\oplus\mathbb{H}^{\prime})\otimes L\cong(\mathbb{K}\otimes L)\oplus(\mathbb{H}^{\prime}\otimes L)\cong L\oplus Hom(\mathbb{H}~{};L).$ Using the above identification, define a Leibniz bracket on $C_{1}\otimes L$ as follows. For $(l_{1},\phi_{1}),(l_{2},\phi_{2})\in L\oplus Hom(\mathbb{H}~{};L)$ let $[(l_{1},\phi_{1}),(l_{2},\phi_{2})]_{\eta_{1}}=([l_{1},l_{2}],\psi)$ where the map $\psi:\mathbb{H}\longrightarrow L$ is given by $\psi(\alpha)=\mu(\alpha)(l_{1},l_{2})+[\phi_{1}(\alpha),l_{2}]+[l_{1},\phi_{2}(\alpha)]~{}\mbox{for}~{}\alpha\in\mathbb{H}~{}.$ It is straightforward to check that $C_{1}\otimes L$ along with the above bracket $\eta_{1}$ is a Leibniz algebra over $C_{1}$. The Leibniz identity is a consequence of the fact that $\delta\mu(\alpha)=0~{}\mbox{for}~{}\alpha\in\mathbb{H}$ . Thus $\eta_{1}$ is an infinitesimal deformation of $L$ with base $C_{1}=\mathbb{K}\oplus\mathbb{H}^{\prime}$. It is proved in [5]: ###### Proposition 2.7. Up to an isomorphism, the deformation $\eta_{1}$ does not depend on the choice of $\mu$. ###### Remark 2.8. Suppose $\\{h_{i}\\}_{1\leq i\leq n}$ is a basis of $\mathbb{H}$ and $\\{g_{i}\\}_{1\leq i\leq n}$ is the dual basis. Let $\mu(h_{i})=\mu_{i}\in CL^{2}(L;L)$. Under the identification $C_{1}\otimes L=L\oplus Hom(\mathbb{H}~{};L)$, an element $(l,\phi)\in L\oplus Hom(\mathbb{H}~{};L)$ corresponds to $1\otimes l+\sum_{i=1}^{n}{g_{i}\otimes\phi(h_{i})}$. Then for $(l_{1},\phi_{1}),(l_{2},\phi_{2})\in L\oplus Hom(\mathbb{H};L)$ their bracket $([l_{1},l_{2}],\psi)$ corresponds to $1\otimes[l_{1},l_{2}]+\sum_{i=1}^{n}g_{i}\otimes(\mu_{i}(l_{1},l_{2})+[\phi_{1}(h_{i}),l_{2}]+[l_{1},\phi_{2}(h_{i})]).$ In particular, for $l_{1},l_{2}\in L$ we have $[1\otimes l_{1},1\otimes l_{2}]_{\eta_{1}}=1\otimes[l_{1},l_{2}]+\sum_{i=1}^{n}g_{i}\otimes\mu_{i}(l_{1},l_{2}).$ The main property of $\eta_{1}$ is the universality in the class of infinitesimal deformations with a finite dimensional base. ###### Proposition 2.9. For any infinitesimal deformation $\lambda$ of a Leibniz algebra $L$ with a finite dimensional base $A$ there exists a unique homomorphism $\phi:C_{1}=({\mathbb{K}}\oplus\mathbb{H}^{\prime})\longrightarrow A$ such that $\lambda$ is equivalent to the push-out $\phi_{*}\eta_{1}$. After obtaining the universal infinitesimal deformation we could like to extend it to higher order. If we have a universal infinitesimal deformation with basis cocycles $\\{\phi_{i}\\}_{i=i}^{r}$, the obstructions to extend it to a second order deformation are the Lie brackets $[\phi_{i},\phi_{j}]$ in the cochain complex (see [3, 5]). This are also called first order Massey operations. If these bracket cochains are coboundaries we can extend our infinitesimal to the second order deformation. The construction of a versal deformation for Leibniz algebras is given in [5]. ## 3 Deformations of the three dimensional Lie algebra $\mathfrak{n}_{3}$ Let us recall the classification of three dimensional complex Lie algebras. Fix a basis $\\{e_{1},e_{2},e_{3}\\}$. The nilpotent algebra $\mathfrak{n}_{3}$ with the commutator matrix $\left(\begin{array}[]{lll}0&0&1\\\ 0&0&0\\\ 0&0&0\end{array}\right).$ The solvable algebra $\mathfrak{r}_{3,1}$ with the matrix $\left(\begin{array}[]{lll}0&1&0\\\ 0&0&1\\\ 0&0&0\end{array}\right).$ The simple Lie algebra $\mathfrak{sl}_{2}$ with the matrix $\left(\begin{array}[]{lll}0&0&1\\\ 0&1&0\\\ 1&0&0\end{array}\right).$ and the projective family of pairwise non isomorphic algebras $d(r:s)$ with the matrix $\left(\begin{array}[]{lll}0&r&1\\\ 0&0&s\\\ 0&0&0\end{array}\right).$ In [6] the moduli space of these Lie algebras is described with the help of versal deformations. Let us recall the results for the nilpotent Lie algebra $\mathfrak{n}_{3}$. The cohomology spaces of the classified algebras are as follows. $\begin{split}&\mbox{Type}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}H^{1}~{}~{}H^{2}~{}~{}H^{3}\\\ &\mathfrak{n}_{3}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}4~{}~{}~{}~{}5~{}~{}~{}~{}2\\\ &d=\mathfrak{r}_{3,1}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}3~{}~{}~{}~{}3~{}~{}~{}~{}0\\\ &d(1:1)=\mathfrak{r}_{3}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}1~{}~{}~{}~{}1~{}~{}~{}~{}0\\\ &d(r:s)~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}1~{}~{}~{}~{}1~{}~{}~{}~{}0\\\ &d(1:0)=\mathfrak{r}_{2}\oplus\mathbb{C}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}2~{}~{}~{}~{}1~{}~{}~{}~{}0\\\ &d(1:-1)=\mathfrak{r}_{3,-1}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}1~{}~{}~{}~{}2~{}~{}~{}~{}1\\\ &\mathfrak{sl}_{2}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}0~{}~{}~{}~{}0~{}~{}~{}~{}0\end{split}$ We get $H^{2}(\mathfrak{n}_{3};\mathfrak{n}_{3})$ is five dimensional. Let us give explicit representative cocycles which form the basis of $H^{2}(\mathfrak{n}_{3};\mathfrak{n}_{3})$. We give the non zero values. $\begin{split}&(1)~{}f_{1}:f_{1}(e_{2},e_{3})=e_{3};\\\ &(2)~{}f_{2}:f_{2}(e_{1},e_{2})=e_{2},~{}f_{2}(e_{1},e_{3})=-e_{3};\\\ &(3)~{}f_{3}:f_{3}(e_{1},e_{2})=e_{3};\\\ &(4)~{}f_{4}:f_{4}(e_{1},e_{3})=e_{1};\\\ &(5)~{}f_{5}:f_{5}(e_{1},e_{3})=e_{2}.\end{split}$ It is easy to check that all the Massey brackets are zero. So the universal infinitesimal deformation is versal and is given by the matrix $\left(\begin{array}[]{lll}0&t_{4}&1\\\ t_{2}&t_{5}&0\\\ t_{3}&-t_{2}&t_{1}\end{array}\right).$ Let us check that how our infinitesimal deformation (which are real deformations) fit in the moduli space of three dimensional Lie algebras. The first deformation with cocycle $f_{1}$ has the matrix $\left(\begin{array}[]{lll}0&0&1\\\ 0&0&0\\\ 0&0&t\end{array}\right).$ which is equivalent to $\mathfrak{r}_{2}\oplus\mathbb{C}$. The second deformation with cocycle $f_{2}$ has the matrix $\left(\begin{array}[]{lll}0&0&1\\\ t&0&0\\\ 0&-t&0\end{array}\right).$ which is equivalent to $\mathfrak{sl}_{2}$. The third deformation with cocycle $f_{3}$ has the matrix $\left(\begin{array}[]{lll}0&0&1\\\ 0&0&0\\\ t&0&0\end{array}\right).$ which is equivalent to $\mathfrak{r}_{3,-1}$. The fourth deformation with cocycle $f_{4}$ has the matrix $\left(\begin{array}[]{lll}0&t&1\\\ 0&0&0\\\ 0&0&0\end{array}\right).$ which is equivalent to $\mathfrak{r}_{2}\oplus\mathbb{C}$. The fifth deformation with cocycle $f_{5}$ has the matrix $\left(\begin{array}[]{lll}0&0&1\\\ 0&t&0\\\ 0&0&0\end{array}\right).$ which is equivalent to $\mathfrak{r}_{3,-1}$. The first deformation is equivalent to $\mathfrak{sl}_{2}$, the second and fourth deformations give the Lie algebra $\mathfrak{r}_{3,-1}$ that means that the Lie algebra $\mathfrak{n}_{3}$ deforms to the family in two different ways. The third and fifth deformations are equivalent to $\mathfrak{r}_{2}\oplus\mathbb{C}$. which means that $\mathfrak{n}_{3}$ deforms to $\mathfrak{r}_{2}\oplus\mathbb{C}$ in two different ways. ## 4 Leibniz deformations of $\mathfrak{n}_{3}$ The classification of three dimensional nilpotent Leibniz algebras is known. Let us recall the definition. We take $L^{1}=L,L^{k+1}=[L^{k},L]~{}\mbox{for}~{}k\in\mathbb{N}$. ###### Definition 4.1. A Leibniz algebra $L$ is called nilpotent if there exists an integer $n\in\mathbb{N}$ such that $L^{1}\supset L^{2}\supset\ldots\supset L^{n}={0}.$ The smallest integer $n$ for which $L^{n}=0$ is called the nilindex of $L$. The classification of complex nilpotent Leibniz algebras up to isomorphism for dimension $2$ and $3$ is in [7] and [1]. In dimension three there are five non isomorphic algebras and one infinite family of pairwise not isomorphic algebras. The list of this classification is given below. $\begin{split}\lambda_{1}:&~{}~{}\mbox{abelian}\\\ \lambda_{2}:&~{}~{}[e_{1},e_{1}]=e_{2}\\\ \lambda_{3}:&~{}~{}[e_{2},e_{3}]=e_{1},[e_{3},e_{2}]=-e_{1}\\\ \lambda_{4}:&~{}~{}[e_{2},e_{2}]=e_{1},[e_{3},e_{3}]=\alpha e_{1},[e_{2},e_{3}]=e_{1};\alpha\in\mathbb{C}\\\ \lambda_{5}:&~{}~{}[e_{2},e_{2}]=e_{1},[e_{3},e_{2}]=e_{1},[e_{2},e_{3}]=e_{1}\\\ \lambda_{6}:&~{}~{}[e_{3},e_{3}]=e_{1},[e_{1},e_{3}]=e_{2}\end{split}$ Let us mention that in this list only $\lambda_{3}$ is a Lie algebra. This is the one which is denoted by $\mathfrak{n}_{3}$ and this is the case we are computing. Now we consider the Leibniz algebra $L=\lambda_{3}=\mathfrak{n}_{3}$ and compute its second Leibniz cohomology space. Our computation consists of the following steps: (i) To determine a basis of the space of cocycles $ZL^{2}(L;L)$, (ii) to find out a basis of the coboundary space $BL^{2}(L;L)$, (iii) to determine the quotient space $HL^{2}(L;L)$. (i) Let $\psi$ $\in$ $ZL^{2}(L;L)$. Then $\psi:L{\otimes 2}\longrightarrow L$ is a linear map and $\delta\psi=0$, where $\begin{split}\delta\psi(e_{i},e_{j},e_{k})&=[e_{i},\psi(e_{j},e_{k})]+[\psi(e_{i},e_{k}),e_{j}]-[\psi(e_{i},e_{j}),e_{k}]-\psi([e_{i},e_{j}],e_{k})\\\ &~{}+\psi(e_{i},[e_{j},e_{k}])+\psi([e_{i},e_{k}],e_{j})~{}\mbox{for}~{}0\leq i,j,k\leq 3.\end{split}$ Suppose $\psi(e_{i},e_{j})=\sum_{k=1}^{3}a_{i,j}^{k}e_{k}$ where $a_{i,j}^{k}\in\mathbb{C}$ ; for $1\leq i,j,k\leq 3$. Since $\delta\psi=0$ equating the coefficients of $e_{1},e_{2}~{}\mbox{and}~{}e_{3}$ in $\delta\psi(e_{i},e_{j},e_{k})$ we get the following relations: $\begin{split}&(i)~{}a_{1,1}^{1}=a_{1,1}^{2}=a_{1,1}^{3}=0;\\\ &(ii)~{}a_{1,2}^{1}=-a_{2,1}^{1};~{}a_{1,2}^{2}=-a_{2,1}^{2};~{}a_{1,2}^{3}=-a_{2,1}^{3};\\\ &(iii)~{}a_{1,3}^{1}=-a_{3,1}^{1};~{}a_{1,3}^{2}=-a_{3,1}^{2};~{}a_{1,3}^{3}=-a_{3,1}^{3};\\\ &(iv)~{}a_{2,2}^{2}=a_{2,2}^{3}=0;\\\ &(v)~{}a_{2,3}^{2}=-a_{3,2}^{2};~{}a_{2,3}^{3}=-a_{3,2}^{3};\\\ &(vi)~{}a_{3,3}^{2}=a_{3,3}^{3}=0;\\\ &(vii)~{}a_{1,2}^{2}=-a_{1,3}^{3}\end{split}$ Observe that there is no relation among $a_{2,2}^{1}$,$a_{2,3}^{1}$, $a_{3,2}^{1}$ and $a_{3,3}^{1}$. Therefore, in terms of the ordered basis $\\{e_{1}\otimes e_{1},e_{1}\otimes e_{2},e_{1}\otimes e_{3},e_{2}\otimes e_{1},e_{2}\otimes e_{2},e_{2}\otimes e_{3},e_{3}\otimes e_{1},e_{3}\otimes e_{2},e_{3}\otimes e_{3}\\}$ of $L^{\otimes 2}$ and $\\{e_{1},e_{2},e_{3}\\}$ of $L$,the matrix corresponding to $\psi$ is of the form $M=\left(\begin{array}[]{llrllllll}0&x_{9}&x_{4}&-x_{9}&x_{6}&x_{7}&-x_{4}&x_{10}&x_{11}\\\ 0&x_{2}&x_{5}&-x_{2}&0&x_{8}&-x_{5}&-x_{8}&0\\\ 0&x_{3}&-x_{2}&-x_{3}&0&x_{1}&~{}x_{2}&-x_{1}&0\end{array}\right).$ $\begin{split}&\mbox{where}~{}x_{1}=a_{2,3}^{3};~{}x_{2}=a_{1,2}^{2};~{}x_{3}=a_{1,2}^{3};~{}x_{4}=a_{1,3}^{1};~{}x_{5}=a_{1,3}^{2};~{}x_{6}=a_{2,2}^{1};\\\ &x_{7}=a_{2,3}^{1};~{}x_{8}=a_{2,3}^{2};~{}x_{9}=a_{1,2}^{1};~{}x_{10}=a_{3,2}^{1};~{}\mbox{and}~{}x_{11}=a_{3,3}^{1}\end{split}$ are in $\mathbb{C}$ . Let $\phi_{i}\in ZL^{2}(L;L)$ for $1\leq i\leq 11$, be the cocycle with $x_{i}=1$ and $x_{j}=0$ for $i\neq j$ in the above matrix of $\psi$. It is easy to check that $\\{\phi_{1},\cdots,\phi_{11}\\}$ forms a basis of $ZL^{2}(L;L)$. (ii) Let $\psi_{0}\in BL^{2}(L;L)$. We have $\psi_{0}=\delta g$ for some $1$-cochain $g\in CL^{1}(L;L)=Hom(L;L)$. Suppose the matrix associated to $\psi_{0}$ is same as the above matrix $M$. Let $g(e_{i})=g_{i}^{1}e_{1}+g_{i}^{2}e_{2}+g_{i}^{3}e_{3}$ for $i=1,2,3$. The matrix associated to $g$ is given by $\left(\begin{array}[]{lll}g_{1}^{1}&g_{2}^{1}&g_{3}^{1}\\\ g_{1}^{2}&g_{2}^{2}&g_{3}^{2}\\\ g_{1}^{3}&g_{2}^{3}&g_{3}^{3}\end{array}\right).$ From the definition of coboundary we get $\delta g(e_{i},e_{j})=[e_{i},g(e_{j})]+[g(e_{i}),e_{j}]-\psi([e_{i},e_{j}])$ for $1\leq i,j\leq 3$. The matrix $\delta g$ can be written as $\left(\begin{array}[]{rrrrrrrll}0&-g_{1}^{3}&g_{1}^{2}&g_{1}^{3}&0&-(g_{1}^{1}-g_{2}^{2}+g_{3}^{3})&-g_{1}^{2}&(g_{1}^{1}-g_{2}^{2}+g_{3}^{3})&0\\\ 0&0&0&0&0&-g_{1}^{2}&0&g_{1}^{2}&0\\\ 0&0&0&0&0&-g_{1}^{3}&0&g_{1}^{3}&0\end{array}\right).$ Since $\psi_{0}=\delta g$ is also a cocycle in $CL^{2}(L;L)$, comparing matrices $\delta g$ and $M$ we conclude that the matrix of $\psi_{0}$ is of the form $\left(\begin{array}[]{rrrrrrrll}0&x_{1}&x_{4}&-x_{1}&0&x_{7}&-x_{4}&-x_{7}&0\\\ 0&0&0&0&0&-x_{4}&0&x_{4}&0\\\ 0&0&0&0&0&x_{1}&0&-x_{1}&0\end{array}\right).$ Let ${\phi_{i}}^{\prime}\in BL^{2}(L;L)~{}\mbox{for}~{}i=1,4,7$ be the coboundary with $x_{i}=1$ and $x_{j}=0$ for $i\neq j$ in the above matrix of $\psi_{0}$. It follows that $\\{\phi_{7}^{\prime},\phi_{8}^{\prime},\phi_{9}^{\prime}\\}$ forms a basis of the coboundary space $BL^{2}(L;L)$. (iii) It is straightforward to check that $\\{[\phi_{1}],[\phi_{2}],[\phi_{3}],[\phi_{4}],[\phi_{5}],[\phi_{6}],[\phi_{10}],[\phi_{11}]\\}$ span $HL^{2}(L;L)$ where $[\phi_{i}]$ denotes the cohomology class represented by the cocycle $\phi_{i}$. Thus $dim(HL^{2}(L;L))=8$. The representative cocycles of the cohomology classes forming a basis of $HL^{2}(L;L)$ are given explicitly below. we give the non zero values. $\begin{split}&(1)~{}\phi_{1}:\phi_{1}(e_{2},e_{3})=e_{3},~{}\phi_{1}(e_{3},e_{2})=-e_{3}\\\ &(2)~{}\phi_{2}:\phi_{2}(e_{1},e_{2})=e_{2},~{}\phi_{2}(e_{2},e_{1})=-e_{2},\phi_{2}(e_{1},e_{3})=-e_{3},\phi_{2}(e_{3},e_{1})=e_{3};\\\ &(3)~{}\phi_{3}:\phi_{3}(e_{1},e_{2})=e_{3},~{}\phi_{3}(e_{2},e_{1})=-e_{3};\\\ &(4)~{}\phi_{4}:\phi_{4}(e_{1},e_{3})=e_{1},~{}\phi_{4}(e_{3},e_{1})=-e_{1};\\\ &(5)~{}\phi_{5}:\phi_{5}(e_{1},e_{3})=e_{2},~{}\phi_{5}(e_{3},e_{1})=-e_{2};\\\ &(6)~{}\phi_{6}:\phi_{6}(e_{2},e_{2})=e_{1};\\\ &(7)~{}\phi_{10}:\phi_{10}(e_{3},e_{2})=e_{1};\\\ &(8)~{}\phi_{11}:\phi_{11}(e_{3},e_{3})=e_{1};\\\ \end{split}$ Consider, $\mu_{i}=\mu_{0}+t\phi_{i}$ for $i=1,2,3,4,5,6,10,11$, where $\mu_{0}$ denotes the original bracket in $L$. This gives $8$ non-equivalent infinitesimal deformations of $L$. Here $\phi_{1},\phi_{2},\phi_{3},\phi_{4},\phi_{5}$ are skew-symmetric, so $\phi_{i}\in Hom(\Lambda^{2}L;L)\subset Hom(L^{\otimes 2};L)$ for $1\leq i\leq 5$. This are exactly the cocycles presented in the previous section. The last three cocycles define Leibniz deformations, more precisely their infinitesimal part. The Leibniz $2$-cocycle $\phi_{6}$ defines the infinitesimal deformation with matrix $\left(\begin{array}[]{lllllllll}0&0&0&0&t&1&0&-1&0\\\ 0&0&0&0&0&0&0&~{}~{}0&0\\\ 0&0&0&0&0&0&0&~{}~{}0&0\end{array}\right).$ $\phi_{10}$ defines the infinitesimal deformation with matrix $\left(\begin{array}[]{lllllllll}0&0&0&0&0&1&0&t-1&0\\\ 0&0&0&0&0&0&0&~{}~{}0&0\\\ 0&0&0&0&0&0&0&~{}~{}0&0\end{array}\right).$ $\phi_{11}$ defines the infinitesimal deformation with matrix $\left(\begin{array}[]{lllllllll}0&0&0&0&0&1&0&-1&t\\\ 0&0&0&0&0&0&0&~{}~{}0&0\\\ 0&0&0&0&0&0&0&~{}~{}0&0\end{array}\right).$ It is interesting to realize that all the three Leibniz deformations are nilpotent and they are real deformation. So they can be identified with the given list. Namely the $\mu_{6}$ is combination of $\lambda_{3}$ and $\lambda_{2}$. $\mu_{10}$ is combination of $\lambda_{2},\lambda_{3}$ and $\lambda_{6}$. $\mu_{11}$ is combination of $\lambda_{2}$ and $\lambda_{3}$. ###### Remark 4.2. Notice here that $f_{i}=\phi_{i}$ for $i=1,\cdots,5$. So, if we consider the infinitesimal deformations of the Leibniz algebra $L$, that automatically contain all infinitesimal Lie algebra deformations. It is interesting to note that we get few more deformations of the original bracket $\mu_{0}$ giving different Leibniz algebra structures, by considering the Leibniz algebra deformation. In order to get a simpler expression for the nontrivial cocycles let us denote $\phi_{1},\cdots,\phi_{6}$ by $\bar{\phi}_{1},\cdots,\bar{\phi}_{6}$, $\phi_{10}$ by $\bar{\phi}_{7}$ and $\phi_{11}$ by $\bar{\phi}_{8}$. First we describe the universal infinitesimal deformation for $L$. Let us denote a basis of $HL^{2}(L;L)^{\prime}$ by $\\{t_{i}\\}_{1\leq i\leq 8}$. By Remark 2.8 the universal infinitesimal deformation of $L$ can be written as $[1\otimes e_{i},1\otimes e_{j}]_{\eta_{1}}=1\otimes[e_{i},e_{j}]+\sum_{i=1}^{8}t_{i}\otimes\bar{\phi}_{i}(e_{i},e_{j})$ with base $C_{1}=\mathbb{C}~{}\oplus_{1\leq i\leq 8}\mathbb{C}~{}t_{i}$. ## 5 Extension of the infinitesimal deformation Let us try now to extend the universal infinitesimal deformation. All the $8$ infinitesimal deformations considered as one parameter deformations are real deformations as $[\bar{\phi}_{i},\bar{\phi}_{i}]=0$ for $i=1,\cdots,8$. Now let us consider the mixed brackets $[\bar{\phi}_{i},\bar{\phi}_{j}]$ for $i\neq j$. For the Lie part we get that $[\bar{\phi}_{1},\bar{\phi}_{2}],[\bar{\phi}_{1},\bar{\phi}_{5}],[\bar{\phi}_{3},\bar{\phi}_{4}]$ give nontrivial $3$-cochains from which two of them are linearly independent. We get two relations on the parameter space: $\begin{split}&t_{1}t_{2}+t_{3}t_{4}=0\\\ &t_{1}t_{5}=0.\end{split}$ This way the versal Lie deformation $[,]_{v_{1}}$ of the Lie algebra $\mathfrak{n}_{3}$ is defined by the infinitesimal part as follows. $\begin{split}&[e_{1},e_{2}]_{v_{1}}=t_{2}e_{2}+t_{3}e_{3}\\\ &[e_{1},e_{3}]_{v_{1}}=t_{4}e_{1}+t_{5}e_{2}-t_{2}e_{3}\\\ &[e_{2},e_{3}]_{v_{1}}=e_{1}+t_{1}e_{3}\\\ \end{split}$ The versal basis is the factor space $\mathbb{C}[[t_{1},t_{2},t_{3},t_{4},t_{5}]]/\\{t_{1}t_{5},t_{1}t_{2}+t_{3}t_{4}\\}.$ The cocycles corresponding to only Leibniz algebras result in trivial mixed brackets: $[\bar{\phi}_{k},\bar{\phi}_{l}]=0~{}\mbox{for}~{}k,l=6,7,8.$ If we take the bracket of a cocycle from the Lie set and take the bracket with a cocycle from the Leibniz set, it turns out that not all the Massey brackets are trivial. Namely, $[\bar{\phi}_{2},\bar{\phi}_{6}],[\bar{\phi}_{3},\bar{\phi}_{6}],[\bar{\phi}_{5},\bar{\phi}_{6}],[\bar{\phi}_{3},\bar{\phi}_{10}],[\bar{\phi}_{5},\bar{\phi}_{10}],[\bar{\phi}_{2},\bar{\phi}_{11}],[\bar{\phi}_{3},\bar{\phi}_{11}],~{}\mbox{and}~{}[\bar{\phi}_{5},\bar{\phi}_{11}]$ are nontrivial three cochains. They give us second order relation on the base of the versal deformation: $\begin{split}t_{5}t_{7}=0,~{}t_{3}t_{6}=0,~{}t_{5}t_{6}=0,~{}t_{5}t_{8}=0,~{}t_{3}t_{8}=0,t_{2}t_{8},~{}t_{3}t_{7}~{}\mbox{and}~{}t_{2}t_{6}=0.\end{split}$ together with the relations for the Lie part we get all the second order relations for the base of the Leibniz versal deformation. As no higher order brackets show up we get that the versal Leibniz deformation $[,]_{v}$ is defined as follows. $\begin{split}&[e_{1},e_{1}]_{v}=0\\\ &[e_{1},e_{2}]_{v}=t_{2}e_{2}+t_{3}e_{3}\\\ &[e_{1},e_{3}]_{v}=t_{4}e_{1}+t_{5}e_{2}-t_{2}e_{3}\\\ &[e_{2},e_{1}]_{v}=-t_{2}e_{2}-t_{3}e_{3}\\\ &[e_{2},e_{2}]_{v}=t_{6}e_{1}\\\ &[e_{2},e_{3}]_{v}=e_{1}+t_{1}e_{3}\\\ &[e_{3},e_{1}]_{v}=-t_{4}e_{1}-t_{5}e_{2}+t_{2}e_{3}\\\ &[e_{3},e_{2}]_{v}=(t_{7}-1)e_{1}-t_{1}e_{3}\\\ &[e_{3},e_{3}]_{v}=t_{8}e_{1}\end{split}$ The base of the versal Leibniz deformation is the factor space $\mathbb{C}[[t_{1},\cdots,t_{8}]]/<t_{1}t_{5},t_{1}t_{2}+t_{3}t_{4},t_{2}t_{6},t_{5}t_{6},t_{3}t_{6},t_{5}t_{7},t_{3}t_{8},t_{2}t_{8},t_{3}t_{7},t_{5}t_{8}>.$ ## References * [1] Albeverio, S. and Omirov, B. A. and Rakhimov, I. S. Varieties of nilpotent complex Leibniz algebras of dimension less than five, Comm. Alg., 33(2005), 1575-1585. * [2] Ayupov, Sh. A. and Omirov, B. A. On some classes of nilpotent Leibniz algebras, Siberian Math. Journal, 42(1) (2001) 18-29. * [3] Fialowski, A. An example of formal deformations of Lie algebras, “NATO Conference on deformation theory of algebras and applications, Proceedings”, Kluwer, Dordrecht, (1988), 375-401. * [4] Fialowski, A. and Fuchs, D. Construction of miniversal Deformation of Lie Algebras, Journal of Functional Analysis 161 (1999),76-110. * [5] Fialowski, A., Mandal, A. and Mukherjee, G. Versal deformations of Leibniz algebras, To appear in Journal of $K$\- Theory, arXiv:math.QA/0702476, 2007. * [6] Fialowski, A. and Penkava, M. Versal deformations of three dimensional Lie algebras as $L_{\infty}$ algebras. Commun. Contemp. Math. 7 (2), (2005), 145-165, math.RT/0512354. * [7] Loday J.-L, Une version con-commutative des algebres de Lie: Les algebres de Leibniz, Enseign. Math., 39, No.3-4 (1993), 269-293. * [8] Loday J.-L, Overview on Leibniz algebras, dialgebras and their homology, Fields Institute Communications, 17 (1997), 91-102. * [9] Loday J.-L and Pirashvili, T. Universal enveloping algebras of Leibniz algebras and (co)homology, Math.Ann., 296 (1993),139-158. * [10] Mandal, A. An Example of Constructing Versal Deformation for Leibniz Algebras, To appear in Comm. Alg., arXiv : math.QA/ 071212.2096v1 13 Dec 2007. * [11] Schlessinger, M. Functors of Artin rings, Trans. Amer. Math. Soc. 130 (1968), 208-222. Alice Fialowski E$\ddot{o}$tv$\ddot{o}$s Lor$\acute{a}$nd University, Budapest, Hungary. e-mail: fialowsk@cs.elte.hu Ashis Mandal Indian Statistical Institute, Kolkata, India. e-mail: ashis_r@isical.ac.in

arxiv-papers

2024-09-04T02:48:53.655026

0802.1341

# Topology of generalized complex quotients Thomas Baird, Yi Lin ###### Abstract. Consider the Hamiltonian action of a torus on a compact twisted generalized complex manifold $M$. We first observe that Kirwan injectivity and surjectivity hold for ordinary equivariant cohomology in this setting. Then we prove that these two results hold for the twisted equivariant cohomology as well. Dedicated to Prof. Victor Guillemin on the occasion of his seventieth birthday. ## 1\. Introduction Kirwan injectivity and surjectivity are two important results in equivariant symplectic geometry. Recall that for a symplectic manifold $(M,\omega)$, an action by a connected Lie group $G$ on $(M,\omega)$ is called Hamiltonian if it is regulated by a moment map $\mu:M\rightarrow\mathfrak{g}^{*}$ taking values in the dual of the Lie algebra of $G$. Contracting by $\xi\in\mathfrak{g}$ produces a real valued function $\mu^{\xi}:M\rightarrow\mathbb{R}$ called a component of the moment map. If $G$ is compact, then for any $\xi\in\mathfrak{g}$, $\mu^{\xi}$ is a Morse-Bott function and may be used to study the equivariant topology of $M$. We will mostly focus on the case that $G=T$ is a torus. In [Kir86], using ideas of Atiyah-Bott [AB82], Kirwan demonstrated that a Hamiltonian action on a compact symplectic manifold $M$ is equivariantly formal. In particular, the equivariant cohomology of $M$ with rational coefficients satisfies a noncanonical isomorphism $H_{G}^{*}(M)\cong H^{*}(M)\otimes H^{*}(BG)$ as graded $H^{*}(BG)$-modules, where $BG$ is the classifying space for $G$. Furthermore, if $G=T$ is a torus, and $i:M^{T}\hookrightarrow M$ denotes inclusion of the fixed point set, the localization map in equivariant cohomology $i^{*}:H_{T}^{*}(M)\rightarrow H_{T}^{*}(M^{T})$ is an injection, a result known as Kirwan injectivity. Her proof uses Morse theory of a component $\mu^{\xi}$ of the moment map. Kirwan also showed that the map $\kappa:H_{G}(M)\rightarrow H_{G}(\mu^{-1}(0))$ induced by inclusion is a surjection. This result is known as Kirwan surjectivity and the map $\kappa$ is known as the Kirwan map. If $0$ is a regular value of $\mu$, then $H_{G}(\mu^{-1}(0))\cong H(M//G)$, where $M//G=\mu^{-1}(0)/G$ is the symplectic quotient, so $H(M//G)$ is describable as a quotient ring $H_{G}(M)/\operatorname{\mathrm{ker}}(\kappa)$. Kirwan’s original proof of surjectivity involved studying the Morse theory of the norm square of the moment map $||\mu||^{2}$, which has minimum $\mu^{-1}(0)$. In fact $||\mu||^{2}$ is not Morse-Bott, but instead satisfies Kirwan’s minimal degeneracy condition, which allows the basic constructions of Morse theory to be carried out. Modern proofs of Kirwan surjectivity have avoided these technicalities. In [TW98], Tolman and Weitsman computed the kernel of $\kappa$ for torus actions using the honest Morse-Bott functions $\mu^{\xi}$, rather than $||\mu||^{2}$, the principle being that the kernel of $\kappa$ is built up of contributions from each circle in the torus. In [Gol02], Goldin used their ideas to produce a simplified proof of Kirwan surjectivity for torus actions, using circles actions and reduction in stages. Goldin’s proof contains a gap, which was resolved by Ginzburg-Guillemin- Karshon ([GGK02] appendix G). They introduce the notion of a nondegenerate abstract moment map, which abstracts the relevant Morse-theoretic properties from the symplectic case, and prove Kirwan’s theorems in this general setting. In this paper, we generalize Kirwan injectivity and surjectivity to Hamiltonian actions on compact generalized complex manifolds, in the sense of Lin-Tolman [LT05]. Generalized complex (GC) manifolds were introduced by Hitchin in [H02] and developed by Gualtieri in his thesis [Gua03]. They form a common generalization of both complex and symplectic manifolds and so are well suited to the study of Mirror Symmetry and conformal field theory. Generalized complex manifolds can also incorporate a twist by a closed 3-form $H\in\Omega^{3}(M)$. When $H$ is integral it may be interpreted as the curvature of a gerbe over $M$ and is known in the physics literature as the Neveu-Schwartz 3-form flux. In the presence of a twisting $H$, it becomes interesting to study the twisted de Rham cohomology of $M$, $H(M;H)$, which is defined to be the cohomology of the complex consisting of the usual de Rham forms $\Omega(M)$ with a twisted differential $d+H\wedge$. For example, Gualtieri [Gua04] showed that for an $H$-twisted generalized Kähler manifold $M$, $H(M;H)$ inherits a Hodge decomposition and in Kapustin-Li [KL04], $H(M;H)$ is identified as the BRST cohomology of states for the associated conformal field theory. In [LT05], Lin-Tolman extended the notion of Hamiltonian actions and reduction in symplectic geometry to the realm of generalized complex geometry. In the presence of a twisting 3-form $H\in\Omega^{3}(M)$, their construction involves a generalized moment map $\mu:M\rightarrow\mathfrak{g}^{*}$ and a moment 1-form $\alpha\in(\Omega^{1}(M)\otimes\mathfrak{g}^{*})^{G}$ for which $H+\alpha$ is an equivariantly closed 3-form in the Cartan model (c.f. [GS99] ). This construction turns out to be something very natural in physics. It has been shown by Kapustin and Tomasiello [KT06] that the mathematical notion of Hamiltonian actions on twisted generalized Kähler manifolds is in perfect agreement with the physical notion of general $(2,2)$ gauged sigma models with three-form fluxes. Inspired by Atiyah-Segal [AS05], the second author in [Lin07] used $H+\alpha$ to define the twisted equivariant cohomology, $H_{T}(M;\eta+\alpha)$. The basic properties of the twisted equivariant cohomologies were studied in [Lin07] using Hodge theory, especially in the case of Hamiltonian actions on compact generalized Kähler manifolds. In the current paper we study the twisted equivariant cohomology using Morse theory, in the more general case of Hamiltonian actions on compact generalized complex manifolds. The following proposition is crucial to our paper. ###### Proposition 1.1. Consider the Hamiltonian action of a (compact) torus $T$ on a compact twisted GC-manifold $M$ with a generalized moment map $\mu:M\rightarrow\mathfrak{t}^{*}$. Then $\mu$ is a nondegenerate abstract moment map in the sense of [GGK02] (see Definition 3.1). Proposition 1.1 paraphrases a result from Nitta’s very interesting recent work [NY07]. Nitta’s result was known by the authors to hold under additional hypotheses but his general result came out as a welcome surprise. Because Nitta’s theorem is central to our work, we provide a self-contained proof of Proposition 1.1 in Section 6. Our proof is a variation of Nitta’s which has the advantages of being somewhat simpler and of extending to some examples of noncompact manifolds $M$. One key ingredient in Nitta’s proof is the maximum principle for pseudo-holomorphic functions on almost complex manifolds, for which we provide a proof in Appendix A. In view of the above proposition, results from [GGK02] prove that Kirwan injectivity and surjectivity hold for ordinary equivariant cohomology. In our paper, we use parallel arguments to prove _twisted_ versions of equivariant formality, Kirwan injectivity, and Kirwan surjectivity: ###### Theorem 1.2 (Equivariant formality). Consider the Hamiltonian action of a compact connected group $G$ on a compact $H$-twisted generalized complex manifold $M$. Then we have a non-canonical isomorphism $H_{G}(M;H+\alpha)\cong H(M;H)\otimes H(BG),$ where $\alpha$ is the moment one form of the Hamiltonian action. ###### Theorem 1.3 (Kirwan injectivity). Let $T$ be a compact torus and let $M$ be a compact $H$-twisted generalized Hamiltonian $T$-space with induced equivariant 3-form $H+\alpha$, and let $i:M^{T}\rightarrow M$ denote the inclusion of the fixed point set. Then the induced map $i^{*}:H_{T}(M;H+\alpha)\rightarrow H_{T}(M^{T};H+\alpha)\cong H(M^{T};H)\otimes H(BT)$ is an injection. ###### Theorem 1.4 (Kirwan surjectivity). Let $M$ be a compact $H$-twisted generalized Hamiltonian $T$-space with induced equivariant 3-form $H+\alpha$ and moment map $\mu$, where $T$ is a compact torus. For $c\in\mathfrak{t}^{*}$ a regular value of $\mu$ we have: (1.1) $H_{T}(M;H+\alpha)\rightarrow H(\mu^{-1}(c)/T;\tilde{H})$ is a surjection, where $\tilde{H}$ is the twisting $3$-form inherited through reduction. These results are established more generally for compact nondegenerate abstract moment maps with _compatible_ equivariantly closed 3-form. We expect that Kirwan surjectivity remains true for the Hamiltonian action of a compact connected Lie group on a compact twisted generalized complex manifold, and we hope to return to this question in a later work. Non-symplectic examples of Hamiltonian torus actions on generalized complex manifolds to which our results may be applied have been constructed in [Lin07] and [Lin07b]. New examples constructed using surgery on toric varieties will be included in a forthcoming paper by the authors. We would also like to mention that a stronger version of Theorem 1.2 was previously proven for the case of Hamiltonian action of a compact Lie group on a generalized Kähler manifold in [Lin07], using the $\partial\bar{\partial}$ lemma and generalized Hodge theory. We discuss now one possible application of our results. Suppose $(M,\mathcal{J})$ is a twisted generalized complex manifold with a Hamiltonian $G$ action, and suppose $L$ is the $\sqrt{-1}$-eigenbundle of $\mathcal{J}$. Then $L$ has a natural Lie algebroid structure, c.f. [Gua03]. Moreover, the existence of a Hamiltonian $G$ action induces a Lie algebra map $\mathfrak{g}\rightarrow C^{\infty}(L)$. So there is an equivariant version of the Lie algebroid cohomology associated to the Hamiltonian $G$ action, in the sense of [BCRR05]. The twisted equivariant cohomology studied in the current paper is closely related to the equivariant Lie algebroid cohomology. Indeed, they are canonically isomorphic to each other if $M$ is a generalized Calabi- Yau manifold satisfying the $\overline{\partial}\partial$-lemma. It is well known that information on the deformation of generalized complex structures is contained in the Lie algebroid cohomology of $L$. Therefore, the results established in this paper may indicate a close relationship between the deformation theory of the generalized complex manifold $M$ and that of its generalized complex quotients. The layout of the paper is as follows. Section 2 reviews twisted equivariant cohomology and proves a few lemmas for later use. Section 3 uses Morse theory to prove twisted Kirwan injectivity and surjectivity for nondegenerate abstract moment maps. Section 4 gives a quick review of generalized complex geometry. Section 5 recalls the definition of generalized moment maps and proves Proposition 1.1. Section 6 establishes the main results of this paper, namely, Theorem 1.2, 1.3, 1.4. Appendix A proves the maximum principle for pseudoholomorphic functions on almost complex manifolds. Appendix B establishes a key Lemma about nondegenerate abstract moment maps postponed from §3. Appendix C compares several versions of twisted equivariant cohomology existed in the literature. Appendix D collects some commutative algebra results that we make frequent use of throughout, but particularly in Section 2. Acknowledgement: The authors would like to thank Marco Gualtieri, Yael Karshon, Eckhard Meinrenken, Paul Selick, and Ping Xu for useful discussions. Y.L. is grateful to Lisa Jeffrey for providing him a Postdoctoral Fellowship at the University of Toronto, where he started his work on the equivariant cohomology theory of GC manifolds. ## 2\. Review of twisted equivariant cohomology In this section we review twisted equivariant cohomology, as developed in Atiyah-Segal [AS05], Hu-Uribe [HuU06], Freed-Hopkins-Teleman [FHT02] and Lin [Lin07]. ### 2.1. Definitions Let $G$ be a compact connected Lie group with Lie algebra $\mathfrak{g}$ and dual $\mathfrak{g}^{*}$. For $M$ a smooth $G$-manifold, we denote by $\xi_{M}$ the vector field on $M$ generated by $\xi\in\mathfrak{g}$. The equivariant de Rham complex $(\Omega_{G}(M),d_{G})$ is a differential graded (super)commutative algebra associated to the $G$-manifold $M$. Here, $\Omega_{G}(M)=(\Omega(M)\otimes S\mathfrak{g^{*}})^{G}$ is the space of polynomial functions on $\mathfrak{g}$ taking values in the space of differential forms $\Omega(M)$, which are equivariant under the induced $G$-action on $\Omega(M)$ and the adjoint action on the symmetric algebra $S\mathfrak{g}^{*}$, and $d_{G}$ is defined by extending linearly the formula $(d_{G}(\sigma\otimes P))(\xi)=d\sigma\otimes P(\xi)-\iota_{\xi_{M}}\sigma\otimes P(\xi)$ where $\sigma\in\Omega(M),P\in S\mathfrak{g}^{*}$, and $\xi\in\mathfrak{g}$. It comes equipped with a grading $\Omega_{G}^{n}(M)=\bigoplus_{k}(\Omega^{n-2k}(M)\otimes S^{k}\mathfrak{g}^{*})^{G}.$ The equivariant de Rham complex computes the (Borel) equivariant cohomology of $M$ with real coefficients (we refer to [GS99] for more details). ###### Example 2.1. When $G$ is trivial $(\Omega_{G}(M),d_{G})$ is the usual de Rham complex. ###### Example 2.2. In the special case that $G=T$ is a compact torus with lie algebra $\mathfrak{t}$, $T$ acts trivially on $\mathfrak{t}$ we have: $\Omega_{T}(M)=\Omega(M)^{T}\otimes S\mathfrak{t^{*}}$ where $\Omega(M)^{T}$ is the space of $T$-invariant differential forms. Let $\hat{\Omega}_{G}(M)$ denote the direct product $\prod\Omega_{G}^{i}(M)$. The differential $d_{G}$ extends in a natural way to $\hat{\Omega}_{G}(M)$ and we adopt the convention that $H_{G}(M)$ is defined to be the cohomology of the complex $(\hat{\Omega}_{G}(M),d_{G})$ (where we have abusively reused $d_{G}$ to denote its extension to $\hat{\Omega}_{G}(M)$). It follows that (2.1) $H_{G}(M):=\prod_{i=0}^{\infty}H^{i}_{G}(M)$ as opposed to the more conventional direct sum. In the untwisted setting this is not a serious modification, but once twisting is introduced the direct product is much easier to work with. In this context, the equivariant cohomology of a point is $(\hat{S}\mathfrak{g}^{*})^{G}$, the ring of $G$-invariant formal power series on $\mathfrak{g}$. Given a $d_{G}$-closed 3-form, $\eta\in\Omega^{3}_{G}(M)$, we define a twisted differential $d_{G,\eta}=d_{G}+\eta\wedge$ on $\hat{\Omega}_{G}(M)$. Because $\eta$ is closed and of odd degree, it follows that $d^{2}_{G,\eta}=0$ and we define the $\eta$-twisted equivariant cohomology $H_{G}(M;\eta)=\text{ker}d_{G,\eta}/\text{im}d_{G,\eta}.$ Because $d_{G,\eta}$ is an odd operator, $H(M;\eta)$ inherits a ${\text{\bf Z}}_{2}$-grading from the Z-grading on $\Omega_{G}(M)$. Because $d_{G,\eta}$ is usually not a derivation, $H_{G}(M;\eta)$ is usually not a ring, but is instead a module for the untwisted equivariant cohomology ring $H_{G}(M)$ and hence also for $\hat{S}\mathfrak{g}^{*}$. ###### Remark 2.3. The cochain complex we are using to define twisted cohomology differs from those found in [FHT02] and [HuU06], but gives rise to a naturally isomorphic cohomology theory (see Appendix C). ###### Remark 2.4. We show in Appendix C that for a compact manifold $M$ the completion of twisted equivariant cohomology is obtained by extension of scalars from the uncompleted version, i.e. $H_{G}(M;\eta)\cong H(\Omega_{G}(M),d_{G,\eta})\otimes_{(S\mathfrak{g}^{*})^{G}}(\hat{S}\mathfrak{g}^{*})^{G}$ ###### Example 2.5. Suppose $G$ acts trivially on $M$. Then any $d$-closed 3-form $\eta\in\Omega^{3}(M)$ determines a $d_{G}$-closed 3-form $\eta\otimes 1\in\Omega_{G}^{3}(M)$. In this case it is easy to see that $H_{G}(M;\eta\otimes 1)\cong H(M;\eta)\otimes(\hat{S}\mathfrak{g}^{*})^{G}$ canonically as $(\hat{S}\mathfrak{g}^{*})^{G}$-modules. ###### Example 2.6. More generally, let $G$ act on $M$ such that a normal subgroup $H\subset G$ acts trivially. Then there is an induced $G/H$ action on $M$ and a chain isomorphism $\Omega_{G/H}(M)\otimes(S\mathfrak{h}^{*})^{H}\cong\Omega_{G}(M)$. If $\eta\in\Omega_{G/H}^{3}(M)$ is $d_{G/H}$-closed, then $H_{G}(M;\eta\otimes 1)\cong H_{G/H}(M;\eta)\otimes(\hat{S}\mathfrak{h}^{*})^{H}$ canonically. We may consider a more general class of twisted complexes using the notion of differential graded modules. Let $(C^{*},\delta)=(\oplus_{k\geq 0}C^{k},\delta)$ be a cochain complex graded by the nonnegative integers. We say that $(C^{*},\delta)$ is a (left) $(\Omega^{*}_{G}(M),d_{G})$-module, or simply a $\Omega_{G}(M)$-module, if $C^{*}$ is a graded module of the graded algebra $\Omega_{G}^{*}(M)$ and for all $\alpha\in\Omega_{G}(M)$ of pure degree and $x\in C^{*}$, the differential satisfies the identity: $\delta(\alpha\wedge x)=d_{G}(\alpha)\wedge x+(-1)^{\operatorname{\mathrm{deg}}\alpha}\alpha\wedge\delta(x).$ The differential $\delta$ extends naturally to a differential on $\hat{C}:=\prod_{i}C^{i}$, which by abuse of notation we also call $\delta$. A closed 3-form $\eta\in\Omega_{G}(M)$ determines a twisted differential $\delta_{\eta}:=\delta+\eta\wedge$ on $\hat{C}$ and we define $H_{G}(C^{*};\eta)=\text{ker}\delta_{\eta}/\text{im}\delta_{\eta}.$ The module structure descends to make $H_{G}(C^{*};\eta)$ a ${\text{\bf Z}}_{2}$-graded module for $H_{G}(M)$. ###### Example 2.7. Let $i:A\subset N$ a pair of embedded submanifolds of $M$ preserved by $G$. We use the algebraic mapping cone to define the differential graded complex $(\Omega_{G}^{*}(N,A),\delta)$ by $\Omega_{G}^{n}(N,A)=\Omega_{G}^{n+1}(N)\oplus\Omega_{G}^{n}(A)$ with differential $\delta(n,a)=(-d_{G}(n),d_{G}(a)+i^{*}(n))$ . Then $\Omega_{G}(N,A)$ is a $(\Omega_{G}(M),d_{G})$-module under the action $x\wedge(n,a)=(x\wedge n,x\wedge a)$. For $\eta\in\Omega_{G}^{3}(M)$ closed, we will use notation: $H_{G}(N,A;\eta)=H(\Omega_{G}(N,A);\eta)$ Notice that $H_{G}(N,A;\eta)=H_{G}(N,A;j^{*}\eta)$, where $j:N\rightarrow M$ is the inclusion map. ### 2.2. Basic Properties Twisted cohomology is invariant under quasiisomorphism. ###### Proposition 2.8. Let $\phi:(C^{*},\delta)\rightarrow(D^{*+n},\delta^{\prime})$ be a degree $n$ quasiisomorphism of $(\Omega_{G}(M),d_{G})$-modules. Then the induced map $H_{G}^{*}(C^{*};\eta)\rightarrow H_{G}^{*+n}(D^{*};\eta)$ is a degree $n~{}mod~{}2$ isomorphism for all $d_{G}$-closed $\eta\in\Omega_{G}^{3}(M)$. ###### Proof. Using the algebraic mapping cone construction, it suffices to prove that if $(C^{*},\delta)$ is acyclic, then so is $(\hat{C}^{*},\delta_{\eta})$. Let $c=c_{i}+c_{i+1}+c_{i+2}+...\in\hat{C}^{*}$ be $\delta_{\eta}$-closed, where $c_{k}\in C^{k}$. Then necessarily $\delta(c_{i})=0$. By acyclicity, there exists $b_{i-1}\in C^{i-1}$ such that $\delta(b_{i-1})=c_{i}$, so $c-\delta_{\eta}(b_{i-1})=c_{i+1}+(c_{i+2}-\eta\wedge b_{i-1})+...$ has lowest degree term lying in $C^{i+1}$. Iterating the process, we can construct $b=b_{i-1}+b_{i}+...$ satisfying $\delta_{\eta}(b)=c$. ∎ It follows that many important properties of untwisted equivariant cohomology, such as homotopy invariance and excision, extend to twisted cohomology. ###### Proposition 2.9. Let $G$ be compact connected with maximal torus $T$ and Weyl group $W=N(T)/T$. For any $G$-manifold $M$ and twisting $\eta\in\Omega_{G}^{3}(M)$, we have a natural isomorphism $H_{G}(M;\eta)\cong H_{T}(M;\eta^{\prime})^{W}$ where $\eta^{\prime}\in\Omega_{T}(M)^{W}$ is the image of $\eta$ under the map $\Omega_{G}(M)\rightarrow\Omega_{T}(M)$ induced by restricting the action. ###### Proof. The map $\Omega_{G}(M)\rightarrow\Omega_{T}(M)$ restricts to a quasiisomorphism $\Omega_{G}(M)\rightarrow\Omega_{T}(M)^{W}$ which is also a $\Omega_{G}(M)$-module homomorphism in the obvious way. Thus by Proposition 2.8, $H_{G}(M;\eta)\cong H_{T}(\Omega_{T}(M)^{W};\eta)\cong H_{T}(M;\eta^{\prime})^{W}$ ∎ This result helps justify our later focus on torus actions. Recall that in untwisted equivariant cohomology, we have the isomorphism $\phi:H_{G}(M)\cong H(M/G),$ provided that the action of $G$ on $M$ is free. We have the following generalization. ###### Proposition 2.10. ([Lin07], A.4.) Let $M$ be a smooth $G$-manifold upon which $G$ acts freely and suppose $\operatorname{\mathrm{dim}}H(M)<\infty$. For $d_{G}$-closed $\eta\in\Omega_{G}^{3}(M)$ we have isomorphisms $H_{G}(M;\eta)\cong H(M/G;\bar{\eta})$ where $\bar{\eta}\in\Omega^{3}(M/G)$ satisfies $\phi([\eta])=[\bar{\eta}]\in H(M/G)$. Given a short exact sequence $0\rightarrow C^{*}\rightarrow D^{*}\rightarrow E^{*}\rightarrow 0$ of $\Omega_{G}(M)$-modules, twisting by $\eta$ gives rise to a six term exact sequence in the twisted cohomology. ###### Example 2.11. Recall the notation of Example 2.7. The pair $i:A\hookrightarrow N$ gives rise to a short exact sequence of $\Omega_{G}(M)$-modules, $0\rightarrow\Omega_{G}(A)\rightarrow Cyl(i^{*})\rightarrow\Omega_{G}(N,A)\rightarrow 0$, where the algebraic mapping cylinder, $Cyl(i^{*})$, is naturally quasiisomorphic to $\Omega_{G}(N)$. We obtain a six term exact sequence: $\begin{CD}H^{0}_{G}(N,A;\eta)@>{}>{}>&H^{0}_{G}(N;\eta)@>{}>{}>H^{0}_{G}(A;\eta)\\\ @A{}A{}A&@V{}V{}V\\\ H^{1}_{G}(A;\eta)@<{}<{}<&H^{1}_{G}(N;\eta)@<{}<{}<H^{1}_{G}(N,A;\eta)\end{CD}$ The next lemma shows that up to (noncanonical) isomorphism, the $\eta$-twisted equivariant cohomology depends only on the cohomology class $[\eta]\in H_{G}^{3}(M)$. ###### Lemma 2.12. Let $b\in\Omega_{G}^{2}(M)$ be an equivariant 2-form and let $exp(b)=\sum_{i=0}^{\infty}b^{n}/n!$. Then for any $(\Omega_{G}(M),d_{G})$-module $(C^{*},\delta)$, wedging by $exp(b)$ determines an isomorphism of chain complexes, $exp(b)\wedge(\,\cdot\,):(\hat{C},\delta_{(\eta+d_{G}b)})\rightarrow(\hat{C},\delta_{\eta})$ which in particular determines an isomorphism $H_{G}(C^{*};\eta+d_{G}b)\cong H_{G}(C^{*};\eta)$. ###### Proof. The map $exp(b)\wedge(\,\cdot\,):\hat{C}\rightarrow\hat{C}$ is certainly even and linear. The equations $exp(b)\wedge exp(-b)=exp(-b)\wedge exp(b)=id_{\hat{C}}$ imply that $exp(b)\wedge(\,\cdot\,)$ is an isomorphism of vector spaces. Finally, for $\alpha\in\hat{C}$ we have $\begin{split}\delta_{\eta}(exp(b)\wedge\alpha)=d_{G}(exp(b))\wedge\alpha+exp(b)\wedge\delta_{\eta}\alpha\\\ =exp(b)\wedge d_{G}b\wedge\alpha+exp(b)\wedge\delta_{\eta}\alpha=exp(b)\delta_{(\eta+d_{G}b)}\alpha\end{split}$ so $exp(b)\wedge(\,\cdot\,)$ respects differentials. ∎ We may now state and prove the Thom isomorphism, which is due to Hu-Uribe [HuU06]. ###### Proposition 2.13. Let $\pi:E\rightarrow N$ be an orientable real vector bundle of rank $r$, let $i:N\rightarrow E$ denote inclusion as the zero section. Let $G$ be a compact torus acting on $E$ by bundle automorphisms, inducing an action on $N$. Let $\eta\in\Omega_{G}^{3}(E)$ be a $d_{G}$-closed form and let $\tau\in\Omega_{G}^{r}(E,E-N)$ be a $d_{G}$-closed form representing the usual equivariant Thom class (c.f. [GS99]). Then the composition: $H_{G}(N;\eta)\xrightarrow{\pi^{*}}H_{G}(E;\pi^{*}i^{*}\eta)\xrightarrow{\wedge\tau}H_{G}(E,E-N;\pi^{*}i^{*}\eta)\xrightarrow{exp(b)\wedge}H_{G}(E,E-N;\eta)$ is an isomorphism of degree $(r~{}mod~{}2)$, where $b\in\Omega^{2}_{G}(E)$ satisfies $d_{G}(b)=\pi^{*}i^{*}(\eta)-\eta$. ###### Proof. By homotopy invariance $\pi^{*}$ is an isomorphism. The map $\wedge\tau:(\Omega_{G}(E),d_{G})\rightarrow(\Omega_{G}(E,E-N),d_{G})$ is a $\Omega_{G}(E)$-module morphism and induces a degree $r$ isomorphism $H_{G}^{*}(E)\cong H_{G}^{*+r}(E,E-N)$, so by Proposition 2.8 $H_{G}(E;\eta)\xrightarrow{\wedge\tau}H_{G}(E,E-N;\pi^{*}i^{*}\eta)$ is also an isomorphism. Finally, $\operatorname{\mathrm{exp}}(b)\wedge$ is an isomorphism by Lemma 2.12. ∎ The equivariant Euler class plays the same role for twisted equivariant cohomology as it does for untwisted equivariant cohomology. ###### Lemma 2.14. Let $\pi:E\rightarrow N$ satisfy the hypotheses and notation of Proposition 2.13, and let $Eul_{G}(E)\in H_{G}^{r}(N)$ denote the equivariant Euler class of $E$. Then the following diagram is commutative: (2.2) $\begin{CD}H_{G}(E,E-N;\eta)@>{}>{j}>H_{G}(E;\eta)\\\ @A{}A{\phi}A@V{}V{i^{*}}V\\\ H_{G}(N;\eta)@>{}>{\cup Eul_{G}(E)}>H_{G}(N;\eta)\end{CD}$ where $\phi$ is the Thom isomorphism of Proposition 2.13, and $j$ is induced by the inclusion map of forms. ###### Proof. The map $i^{*}\circ j\circ\phi$ is induced by a map of forms $h:\Omega_{G}(N)\rightarrow\Omega_{G}(N)$ defined by $h(\alpha)=i^{*}(exp(b)\wedge\pi^{*}(\alpha)\wedge\tau)$ where $\tau$ is a form representing the equivariant Thom class and $b\in\Omega_{G}^{2}(N)$ satisfies $d_{G}(b)=\eta-\pi^{*}(i^{*}(\eta))$. We may choose $b$ so that $i^{*}(b)=0$, because if it doesn’t we can replace it by $b-\pi^{*}(i^{*}(b))$. Thus $h(\alpha)=i^{*}(exp(b)\wedge\pi^{*}(\alpha)\wedge\tau)=i^{*}(exp(b))\wedge\alpha\wedge i^{*}(\tau)=\alpha\wedge i^{*}(\tau)$. Because $i^{*}(\tau)$ represents the equivariant Euler class, this completes the proof. ∎ Hu and Uribe go on to prove the following twisted version of the localization theorem. ###### Theorem 2.15. ([HuU06]) Let $T$ be a compact torus acting on a smooth, closed manifold $M$ and let $i:M^{T}\hookrightarrow M$ denote the inclusion of the fixed point set. Then for any $d_{T}$-closed 3-form $\eta\in\Omega^{3}_{T}(M)$, the kernel and cokernel of the induced map $i^{*}:H_{T}(M;\eta)\hookrightarrow H_{T}(M^{T};\eta)$ are $\hat{S}\mathfrak{t}^{*}$-torsion. ### 2.3. Spectral Sequences We now consider two spectral sequences associated filtrations of the complex $(\hat{\Omega}_{G}(M),d_{\eta})$, both of which converge strongly to $H(M;\eta)$ (see Appendix C for an explanation of convergence properties). First consider the filtration of $(\hat{\Omega}_{G}(M),d_{G,\eta})$ (2.3) $F^{p}=F^{p}\hat{\Omega}_{G}(M):=\prod_{k\geq p}\Omega_{G}^{k}(M)$ which satisfies $d_{G,\eta}(F^{p})\subset F^{p+1}$. The resulting spectral sequence $(E_{r}^{*},d_{r})$ satisfies $E_{1}^{p}\cong E_{2}^{p}\cong H^{p}_{G}(M)$ the untwisted cohomology, while $d_{2}$ is the wedging map $\eta\wedge(.):H^{*}_{G}(M)\rightarrow H^{*}_{G}(M)$. Thus by Proposition 2.12, this spectral sequence collapses at $E_{1}$ if and only if $\eta$ is cohomologous to zero. In particular, if $\operatorname{\mathrm{dim}}H(M)<\infty$ then in the nonequivariant case: (2.4) $\operatorname{\mathrm{dim}}H(M;\eta)\leq\operatorname{\mathrm{dim}}H(M)$ with equality if and only if $\eta$ is $d$-exact. Now consider a different filtration: (2.5) $L^{p}=L^{p}\hat{\Omega}_{G}(M):=\prod_{k\geq p}(\Omega(M)\otimes S^{k}\mathfrak{g}^{*})^{G}$ This gives rise to a spectral sequence $(E_{r}^{*},d_{r})$ of $(\hat{S}\mathfrak{g}^{*})^{G}$-modules satisfying $E_{1}^{p}\cong H(M;\eta(0))\otimes(S^{p}\mathfrak{g}^{*})^{G}$, where $\eta(0)$ be the ordinary 3-form obtained by evaluating $\eta$ at $0\in\mathfrak{g}$. ###### Definition 2.16. We say that a $G$-manifold $M$ is $\eta$-equivariantly formal if the spectral sequence defined above collapses at $E_{1}$. In this case $H_{G}(M;\eta)$ is noncanonically isomorphic to $H(M;\eta(0))\otimes(\hat{S}\mathfrak{g}^{*})^{G}$ as a module over $(\hat{S}\mathfrak{g}^{*})^{G}$. Notice that the quotient complex $\hat{\Omega}_{G}(M)/L^{1}\hat{\Omega}_{G}(M)$ is canonically isomorphic to $\Omega(M)$. This gives rise to a natural map $H_{G}(M;\eta)\rightarrow H(M;\eta(0))$ for all twistings $\eta$. We have a version of the Leray-Hirsch theorem in this context. ###### Proposition 2.17. The $G$-manifold $M$ is $\eta$-equivariantly formal if and only if the natural map $H_{G}(M;\eta)\rightarrow H(M;\eta(0))$ is surjective. ###### Proof. The spectral sequence associated to the filtration collapses at page $E_{1}$ if and only if the injections $L^{p+1}\hookrightarrow L^{p}$ induce injections in cohomology $H(L^{p+1}\Omega_{G}(M);d_{G,\eta})\hookrightarrow H(L^{p}\Omega_{G}(M);d_{G,\eta})$ for all $p$. By the associated six term exact sequence, is true if and only if $H(L^{p};d_{G,\eta})\rightarrow H(L^{p}/L^{p+1};d_{G,\eta})$ is surjective. Of course $H(L^{p}/L^{p+1};d_{G,\eta})=E_{1}^{p}\cong H(M;\eta(0))\otimes(S^{p}\mathfrak{g}^{*})^{G}$. Collecting together, we see that $M$ is equivariantly formal if and only if the natural maps $\pi_{p}:H(L^{p};d_{G,\eta})\rightarrow H(M;\eta(0))\otimes(S^{p}\mathfrak{g}^{*})^{G}$ are surjective for all $p$. When $p=0$, $\pi_{0}$ is exactly $H_{G}(M;\eta)\rightarrow H(M;\eta(0))$ proving one direction of the equivalence. The opposite direction follows by noting that for $\sigma\in(S^{p}\mathfrak{g}^{*})^{G}$ and $d_{G,\eta}$-closed $\phi\in\Omega_{G}(M)$, we have $\pi_{p}(\phi\sigma)=\pi_{0}(\phi)\otimes\sigma$. ∎ ###### Proposition 2.18. A $G$-manifold $M$ is $\eta$-equivariantly formal if and only if $M$ it is $\eta^{\prime}$-equivariantly formal as a $T$-manifold under the restricted maximal torus action, where $\eta^{\prime}$ is the image of $\eta$ under the induced map $\Omega_{G}^{3}(M)\rightarrow\Omega_{T}^{3}(M)$. ###### Proof. We use the criterion of Proposition 2.17 The natural map $\phi:H_{G}(M;\eta)\rightarrow H(M;\eta(0))$ factors through the natural map $\phi^{\prime}$ via $H_{G}(M;\eta)\rightarrow H_{T}(M;\eta^{\prime})\stackrel{{\scriptstyle\phi^{\prime}}}{{\rightarrow}}H(M;\eta^{\prime}(0)).$ Thus if $\phi$ is surjective, so must $\phi^{\prime}$. On the other hand, the map $\phi^{\prime}$ is $W$ equivariant, where the action of the Weyl group $W$ on $H(M;\eta(0))$ is induced by $N(T)$ action restricted from $G$. Since this action is isotropically trivial, we find that if $\phi^{\prime}$ is invariant under the $W$ action, so is the restricted map $H_{G}(M;\eta)\cong H_{T}(M;\eta^{\prime})^{W}\rightarrow H(M;\eta(0))$ completing the proof. ∎ It is worth noting that a $G$-manifold that is equivariantly formal for $\eta=0$ may fail to be formal for $\eta\neq 0$. ###### Example 2.19. Let $U(1)$ act trivially on $S^{1}$. Then $H_{U(1)}(S^{1})\cong H(S^{1})\otimes\mathbb{R}[x]$, where $0\neq x\in\mathfrak{u(1)}^{*}$. Choose a twisting $\eta$ satisfying $0\neq[\eta]\in H^{1}(S^{1})\otimes x$. Then $H(S^{1};\eta(0))\cong H(S^{1})$ because $[\eta(0)]=0$, while $H_{U(1)}(M;\eta)=0$. The second assertion here follows from the $\\{L^{p}\\}$ spectral sequence where $E_{1}\cong H(S^{1})\otimes\mathbb{R}[x]$ with differential $d_{1}$ defined by wedging by $[\eta]$ so that $H(E_{1},d_{1})=E_{2}=0$. ## 3\. Morse theory Throughout this section, unless otherwise stated, $T$ is a compact torus with Lie algebra $\mathfrak{t}$ acting on a closed smooth manifold $M$. Recall the following definition from [GGK02] ###### Definition 3.1. An nondegenerate abstract moment map $\mu:M\rightarrow\mathfrak{t}^{*}$ is a smooth, equivariant map $\mu:M\rightarrow\mathfrak{t}^{*}$ such that for every vector $\xi\in\mathfrak{t}$, (1) $Crit(\mu^{\xi})=\\{\xi_{M}=0\\}$ , and (2) $\mu^{\xi}:M\rightarrow\mathbb{R}$ is a Morse-Bott function. Definition 3.1 is an abstraction of the Morse theoretic properties of symplectic moment maps which are responsible for results such as Kirwan injectivity and surjectivity, as well as convexity (c.f. [GGK02]). In [NY07] Nitta actually proved that the components of moment map for Hamiltonian torus actions on compact generalized complex manifolds are abstract nondegenerate moment maps (see also §5). Thus it follows that Kirwan injectivity and surjectivity for the usual equivariant cohomology must hold for GC-Hamiltonian actions. To prove twisted versions of these theorems, we must impose a compatibility condition on the twisting 3-form. ###### Definition 3.2. Let $\eta\in\Omega_{T}^{3}(M)=\Omega^{3}(M)^{T}\oplus(\Omega^{1}(M)^{T}\otimes\mathfrak{t}^{*})$ and let $\eta^{1}$ denote the component of $\eta$ lying in $\Omega^{1}(M)^{T}\otimes\mathfrak{t}^{*}$. We say that $\eta$ is compatible if for all $p\in M$, $ker(\eta^{1}_{p})\supseteq\mathfrak{t}_{p}$, where $\eta^{1}$ is regarded as a linear map $\eta^{1}:\mathfrak{t}\rightarrow\Omega^{1}(M)^{T}$ and $\mathfrak{t}_{p}$ is the Lie algebra of the isotropy subgroup of the point $p\in M$. Our goal in this section is to prove the following two results: ###### Theorem 3.3. [Kirwan Injectivity] Let $M$ be a smooth, compact $T$-manifold with abstract moment map $\mu:M\rightarrow\mathfrak{t}^{*}$, and compatible equivariantly closed 3-form $\eta\in\Omega_{T}^{3}(M)$. Then $M$ is $\eta$-equivariantly formal. In particular, the localization map $i^{*}:H_{T}(M;\eta)\rightarrow H_{T}(M^{T};\eta)$ is injective and $H_{T}(M;\eta)\cong H(M;\eta(0))\otimes\hat{S}(\mathfrak{t}^{*})$ noncanonically as $\hat{S}(\mathfrak{t}^{*})$-modules. Theorem 3.3 may be generalized to noncompact $M$ using the weaker hypothesis that the fixed point set $M^{T}$ is compact and that some nonzero component of the moment map $\mu^{\xi}:M\rightarrow\mathbb{R}$ is proper and bounded below. Working in such generality is cumbersome, so we stick with compact $M$. ###### Theorem 3.4. [Kirwan Surjectivity] Let $M$ be a smooth, compact $T$-manifold with abstract moment map $\mu:M\rightarrow\mathfrak{t}^{*}$, and compatible equivariantly closed 3-form $\eta\in\Omega_{T}(M)$. Suppose $M$ admits an invariant almost complex structure. Then the map in equivariant cohomology induced by inclusion of the zero level set: $H_{T}(M;\eta)\rightarrow H_{T}(\mu^{-1}(0);\eta)$ is a surjection. In the event that $T$ acts freely on $\mu^{-1}(0)$ then $H_{T}(\mu^{-1}(0);\eta)\cong H(\mu^{-1}(0)/T;\bar{\eta})$ as explained in Proposition 2.10. As before, Theorem 3.4 may be generalized to include some examples of noncompact manifolds but for the sake of simplicity we work with compact $M$. In our proof of Theorem 3.4 we found it necessary to require a invariant almost complex structure, though we suspect the theorem holds without this additional hypothesis. The presence of an invariant almost complex structure in the case of a GC Hamiltonian actions was proven by Nitta [NY07] and played an important part in his work. Let $f:M\rightarrow\mathbb{R}$ be a smooth function. We denote the critical set of $f$ by $Crit(f)=\\{x\in M|df_{x}=0\\}$. For $x\in Crit(f)$, the Hessian $Hess_{x}(f):T_{x}M\rightarrow T^{*}_{x}M$ is the symmetric linear map defined by the formula $<Hess_{x}(f)(v),w>=w\cdot L_{\tilde{v}}f,$ where $v,w\in T_{x}M$, $\tilde{v}$ is any vector field satisfying $\tilde{v}_{x}=v$, $L$ is the Lie derivative and $<,>$ is the pairing between $T_{x}M$ and $T_{x}^{*}M$. The Hessian is more often defined as the quadratic form $<Hess_{x}(.),.>$, but the definition as a linear map is more convenient for us. ###### Definition 3.5. Let $M$ be a smooth, closed manifold. A smooth function $f:M\rightarrow\mathbb{R}$ is called Morse-Bott if the connected components of $Crit(f)=\\{x\in M|df_{x}=0\\}$ are closed submanifolds of $M$ and for all $x\in Crit(f)$ the kernel of the Hessian satisfies $\operatorname{\mathrm{ker}}(Hess_{x}(f))=T_{x}Crit(f)$. Let $\\{C_{i}|i\in 0,1,2,...,n\\}$ be the set of connected components of $Crit(f)$. The function $f$ is constant on each component $C_{i}$ and we define $c_{i}:=f(C_{i})\in\mathbb{R}$. We will assume for simplicity of exposition that $c_{i}=c_{j}$ if and only if $i=j$, though all the proofs can be adapted to work without this assumption. We choose the indexing $i=0,1,...,n$ so that $c_{i}<c_{j}$ if and only if $i<j$. Choose a Riemannian metric $g$ on $M$. Using $g$ to identify $TM\cong T^{*}M$, we may regard $Hess_{x}(f)$ as an automorphism of $T_{x}M$ for $x\in Crit(f)$. Because it is symmetric, $Hess_{x}(f)$ is diagonalizable with real eigenvalues. We define the negative normal bundle $\nu_{i}$ of $C_{i}$ by setting $\nu_{i,x}$ to equal the sum of negative eigenspaces of $Hess_{x}(f)$. Up to isomorphism, $\nu_{i}$ is independent of the choice of $g$. We call the rank of $\nu_{i}$ the index of $C_{i}$ and denote it $\lambda(i)$. In the presence of a compact torus $T$-action on $M$ leaving $f$ and $g$ invariant, the $\nu_{i}$ become equivariant vector bundles over $C_{i}$. Let $M_{t}:=f^{-1}((-\infty,t))$. If the interval $[s,t]$ contains no critical values for $f$, the inclusion $M_{s}\hookrightarrow M_{t}$ is a homotopy equivalence. In particular, if a torus $T$ acts on $M$ leaving $f$ invariant and $\eta\in\Omega_{T}^{3}(M)$ is a closed equivariant 3-form, then $H_{T}(M_{t};\eta)\cong H_{T}(M_{s};\eta)$. Thus for some $\epsilon>0$ sufficiently small, we obtain for each critical value $c_{i}$ a six term exact sequences: (3.1) $\begin{CD}H_{T}^{0}(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon};\eta)@>{}>{}>H_{T}^{0}(M_{c_{i}+\epsilon};\eta)@>{}>{}>H_{T}^{0}(M_{c_{i}-\epsilon};\eta)\\\ @A{}A{}A@V{}V{}V\\\ H_{T}^{1}(M_{c_{i}-\epsilon};\eta)@<{}<{}<H_{T}^{1}(M_{c_{i}+\epsilon};\eta)@<{}<{}<H_{T}^{1}(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon};\eta)\end{CD}$ and canonical isomorphisms $H_{T}(M_{c_{i}+\epsilon})\cong H_{T}(M_{c_{i+1}-\epsilon})$. Using excision and the Thom isomorphism, we obtain isomorphisms: (3.2) $H_{T}^{*}(M_{c+\epsilon},M_{c-\epsilon};\eta)\cong H_{T}^{*}(\nu_{i},\nu_{i}-0;\eta)\cong H_{T}^{*+\lambda(i)}(C_{i};\eta)$ where the superscript grading is taken mod 2. ###### Definition 3.6. A $T$-invariant Morse-Bott function $f$ is called $\eta$-equivariantly perfect if the vertical arrows in (3.1) are zero for all critical values $c_{i}$. An important consequence is that $\oplus_{i}H_{T}(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon};\eta)$ is isomorphic to an associated graded object of $H_{T}(M;\eta)$. ###### Proposition 3.7. Suppose that $f$ is bounded below, $\eta$-equivariantly perfect Morse-Bott function on $M$ and that the negative normal bundles are all orientable. Then there is an isomorphism of $\hat{S}\mathfrak{t}^{*}$-modules $gr(H_{T}^{*}(M;\eta))\cong\oplus_{i}H^{*+\lambda(i)}_{T}(C_{i};\eta)$ where $gr(H_{T}(M;\eta))$ is the associated graded ring determined by the topological filtration $M_{s}$ of $M$ and $\lambda(i)\in\\{0,1\\}$ is the index of $C_{i}$ mod $2$. ###### Proof. Because $f$ is $\eta$-equivariantly perfect, the exact sequence (3.1) decomposes into exact sequences (3.3) $0\rightarrow H_{T}(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon})\rightarrow H_{T}(M_{c_{i}+\epsilon};\eta)\rightarrow H_{T}(M_{c_{i}-\epsilon};\eta)\rightarrow 0$ It follows that $gr(H_{T}(M;\eta))\cong\oplus_{i}H_{T}(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon})$. Applying (3.2) completes the proof. ∎ It was noticed by Atiyah and Bott that an invariant Morse-Bott function can sometimes be shown to be equivariantly perfect using only negative normal bundle data as follows. If the negative normal bundle is orientable, we construct a commutative diagram: (3.4) $\begin{CD}H_{T}(M_{c+\epsilon},M_{c-\epsilon};\eta)@>{j}>{}>H_{T}(M_{c+\epsilon};\eta)@>{}>{}>H_{T}(M_{c-\epsilon};\eta)\\\ @V{}V{\cong}V@V{}V{}V\\\ H_{T}(\nu_{i},\nu_{i}-0;\eta)@>{}>{}>H_{T}(\nu_{i};\eta)\\\ @A{}A{\cong}A@V{}V{\cong}V\\\ H_{T}(C_{i};\eta)@>{}>{\cup\text{Eul}_{T}(\nu_{i})}>H_{T}(C_{i};\eta)\end{CD}$ where the upper square is excision and the bottom square is from Lemma 2.14. If $\cup Eul_{T}(\nu_{i}):H_{T}(C_{i};\eta)\rightarrow H_{T}(C_{i};\eta)$ is injective then $j$ must also be injective. We obtain the self perfecting principle: ###### Lemma 3.8. Suppose that for all critical sets $C_{i}$, $Eul_{T}(\nu_{i})$ is not a zero divisor for $H_{T}(C_{i};\eta)$, i.e. for all $\alpha\in H_{T}(C_{i};\eta)$, we have $\alpha\cup Eul_{T}(\nu_{i})=0$ if and only if $\alpha=0$. Then $f$ is $\eta$-equivariantly perfect. Atiyah and Bott discovered a simple criterion implying that $Eul_{T}(\nu_{i})$ is not a zero divisor in the nontwisted setting. We adapt their proof to the twisted case. ###### Lemma 3.9. Let $\nu\rightarrow N$ be a $T$-equivariant oriented vector bundle over a compact manifold $N$ and suppose there exists a subtorus $S\subset T$ such that $\nu^{S}$ is exactly the zero section of $\nu$. Let $\eta\in\Omega_{T/S}(N)\hookrightarrow\Omega_{T}(N)$ under the natural inclusion (see Example 2.6). Then $Eul_{T}(\nu)$ is not a zero divisor for $H_{T}(N;\eta)$. ###### Proof. By Example 2.6 the untwisted equivariant cohomology satisfies $H_{T}(N)\cong H_{T/S}(N)\otimes\hat{S}(\mathfrak{s}^{*}).$ It was shown in [AB82] §13, that the equivariant Euler class $Eul_{T}(\nu)\in H_{T}(N)$ satisfies $Eul_{T}(\nu)=1\otimes\beta_{0}+\text{positive degree terms in }H^{*}_{T/S}(N)$ where $\beta_{0}\in S(\mathfrak{s}^{*})$ is nonzero. Also by Example 2.6, we have a natural isomorphism $H_{T}(N;\eta)\cong H_{T/S}(N;\eta)\otimes\hat{S}(\mathfrak{s}^{*})$ The ideal $I=\prod_{k>0}H_{T/S}^{k}(N)$ is the Jacobson ideal of $H_{T/S}(N)$, so by Nakayama’s Lemma the filtration $\\{F^{p}\\}$ of $H_{T}(N;\eta)$ defined by $F^{p}:=I^{p}\cup H_{T/S}(N;\eta)\otimes\hat{S}(\mathfrak{s}^{*})$ satisfies $\cap_{p}F^{p}=0$ (see Appendix D). For $\alpha\in H_{T}(N;\alpha)$ nonzero, define $p(\alpha)$ by $\alpha\in F^{p(\alpha)}-F^{p(\alpha)+1}$. It follows that $\alpha\cup Eul_{T}(\nu)=\alpha\cup\beta_{0}\text{ modulo }F^{p+1}$ which is nonzero. ∎ Notice that Definition 3.2 ensures that if $N$ is a component of $M^{S}$ and $\nu$ is a subbundle of the normal bundle of $N$ in $M$, then the hypotheses of Lemma 3.9 apply. ###### Lemma 3.10. Under the hypotheses of Theorem 3.3, for a generic choice of $\xi\in\mathfrak{t}$, the moment map component $\mu^{\xi}:M\rightarrow\mathbb{R}$ is $\eta$-equivariantly perfect. ###### Proof. For generic choice $\xi\in\mathfrak{t}$, the image of $exp:\text{span}\\{\xi\\}\rightarrow T$ is dense. Letting $f=\mu^{\xi}$, it follows that $Crit(f)=\\{p\in M|\xi_{M}=0\\}=M^{T}$ By Lemma 3.8, it suffices to show for each connected component $C_{i}$ of $M^{T}$ with negative normal bundle $\nu_{i}\rightarrow C_{i}$, that $H(C_{i};\eta)$ possesses no $Eul_{T}(\nu_{i})$-torsion. This is a consequence of Lemma 3.9 in the case $S=T$. ∎ ###### Proof of Theorem 3.3. By Proposition 2.17, equivariant formality is equivalent to surjectivity of the natural map $H_{T}(M;\eta)\rightarrow H(M;\eta)$. We prove this by induction on $H(M_{t};\eta)$ where $M_{t}:=f^{-1}((-\infty,t))$ and $f$ a generic component of the moment map as in Lemma 3.10. For the base case $M_{t}$ is empty for small $t$ because $M$ is compact. In the induction step, assume that $H_{T}(M_{c_{i}-\epsilon};\eta)\rightarrow H(M_{c_{i}-\epsilon};\eta)$ is surjective. Using long exact sequences for the pair and Lemma 3.10 we obtain a commutative diagram: (3.5) $\begin{CD}\lx@xy@svg{\hbox{\raise 0.0pt\hbox{\kern 5.5pt\hbox{\ignorespaces\ignorespaces\ignorespaces\hbox{\vtop{\kern 0.0pt\offinterlineskip\halign{\entry@#!@&&\entry@@#!@\cr&&&&\\\&&&&\crcr}}}\ignorespaces{\hbox{\kern-5.5pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{0\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 29.5pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 29.5pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{H_{T}(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon};\eta)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 69.60529pt\raise-29.68857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 133.71057pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 133.71057pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{H_{T}(M_{c_{i}+\epsilon};\eta)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 162.36502pt\raise-29.68857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 215.01947pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 215.01947pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{H_{T}(M_{c_{i}-\epsilon};\eta)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 293.83948pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 242.42947pt\raise-29.68857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\lower-1.99997pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 293.83948pt\raise 0.0pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{0}$}}}}}}}{\hbox{\kern-3.0pt\raise-40.18857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{}$}}}}}}}{\hbox{\kern 31.52515pt\raise-40.18857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{H(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon};\eta)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 135.7357pt\raise-40.18857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 135.7357pt\raise-40.18857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{H(M_{c_{i}+\epsilon};\eta)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$}}}}}}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces{}{\hbox{\lx@xy@droprule}}\ignorespaces{\hbox{\kern 217.04462pt\raise-40.18857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\lx@xy@tip{1}\lx@xy@tip{-1}}}}}}{\hbox{\lx@xy@droprule}}{\hbox{\lx@xy@droprule}}{\hbox{\kern 217.04462pt\raise-40.18857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{H(M_{c_{i}-\epsilon};\eta)}$}}}}}}}{\hbox{\kern 296.33948pt\raise-40.18857pt\hbox{\hbox{\kern 0.0pt\raise 0.0pt\hbox{\hbox{\kern 3.0pt\raise 0.0pt\hbox{$\textstyle{}$}}}}}}}\ignorespaces}}}}\ignorespaces\end{CD}$ By a diagram chase we are reduced to proving that $H_{T}(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon};\eta)\rightarrow H(M_{c_{i}+\epsilon},M_{c_{i}-\epsilon};\eta)$ surjects. By the Thom isomorphism this is equivalent to showing that the critical sets $C_{i}$ are equivariantly formal. By the compatibility of $\eta$ this follows from Example 2.5. The injectivity of $H_{T}(M;\eta)\rightarrow H_{T}(M^{T};\eta)$ follows from 2.15, because $H_{T}(M;\eta)$ is torsion free. ∎ ###### Corollary 3.11. Under the hypotheses of 3.3 we have an equality $\operatorname{\mathrm{dim}}H(M;\eta(0))=\operatorname{\mathrm{dim}}H(M^{T};\eta(0))$ ###### Proof. By equivariant formality $H_{T}(M;\eta)\cong H(M;\eta(0))\otimes\hat{S}\mathfrak{t}^{*}$ while by equivariant perfection of a generic component of the Morse map $gr(H_{T}(M;\eta))\cong\oplus H(C_{i};\eta(0))\otimes\hat{S}\mathfrak{t}^{*}$ if we ignore the ${\text{\bf Z}}_{2}$-grading. The summands $H(C_{i};\eta(0))\otimes\hat{S}\mathfrak{t}^{*}$ are free, hence projective over $\hat{S}\mathfrak{t}^{*}$ so $H_{T}(M;\eta)\cong H(\cup_{i}C_{i};\eta(0))\otimes\hat{S}\mathfrak{t}^{*}=H(M^{T};\eta(0))\otimes\hat{S}\mathfrak{t}^{*}.$ ∎ We now turn our attention to the proof of the Kirwan surjectivity Theorem 3.4. We will need the following proposition. ###### Proposition 3.12. Let $M$ be a compact $T$-manifold with nondegenerate abstract moment map $\mu:M\rightarrow\mathfrak{t}^{*}$ for which $0\in\mathfrak{t}^{*}$ is a regular value, and suppose that $M$ admits a $T$-invariant almost complex structure. Then we may choose a basis $\xi_{1},...,\xi_{n}$ of $\mathfrak{t}$ such that (1) each $\mathfrak{t}_{k}:=Span\\{\xi_{1},...,\xi_{k}\\}$ exponentiates to a rank $k$ torus $T_{k}$, (2) $0\in\mathfrak{t}_{k}^{*}$ is a regular value for the moment map $\mu_{k}=proj_{\mathfrak{t}_{k}^{*}}\circ\mu$ (3) The restriction of $\mu^{\xi_{k+1}}$ to the submanifold $M_{k}=\mu_{k}^{-1}(0)\subset M$ is Morse-Bott with critical set equal to the points where $T_{k+1}$ acts with positive dimensional stabilizer. The proof of Proposition 3.12 is postponed until Appendix B. Proposition 3.12 allows us to factor the Kirwan map $H_{T}(M;\eta)\rightarrow H_{T}(\mu^{-1}(0))$ through the sequence of submanifolds determined by Proposition 3.12 $H_{T}(M)\rightarrow H_{T}(\mu^{-1}_{1}(0))\rightarrow H_{T}(\mu^{-1}_{2}(0))...\rightarrow H_{T}(\mu^{-1}_{n}(0))=H_{T}(\mu^{-1}(0)).$ Our strategy to prove Theorem 3.4 is to show that each map in this composition is surjective. We do this by applying the following lemma to the $T$-manifold $\mu^{-1}_{k}(0)$ with function $(\mu^{\xi_{k+1}})^{2}$, which completes the proof of Theorem 3.4. ###### Lemma 3.13. Let $X$ be compact smooth $T$-manifold with no orbits of dimension smaller than $d$, and let $N$ be the union of dimension $d$ orbits. Let $f:X\rightarrow\mathbb{R}$ be an $T$-invariant function such that $Crit(f)=N\cup f^{-1}(c_{0})$, where $c_{0}$ is the minimum value of $f$. Suppose that $f$ is Morse-Bott except possibly at the minimum $f^{-1}(0)$. For $C=C_{i}$ a connected component of $N$, let $\mathfrak{t}_{C}$ denote the infinitesimal stabilizer of $C$ and let $T_{C}=exp(\mathfrak{t}_{C})$ its torus. If $\eta\in\Omega^{3}_{T}(X)$ is a $d_{T}$-closed form satisfying $\eta|_{C}\in\Omega_{T/T_{C}}(C)\otimes 1\subset\Omega_{T/T_{C}}(C)\otimes S\mathfrak{t}_{C}^{*}\cong\Omega_{T}(C),$ then the map induced by inclusion $H_{T}(X;\eta)\rightarrow H_{T}(f^{-1}(c_{0});\eta)$ is surjective. ###### Proof. The map $H_{T}(X_{c_{0}+\epsilon};\eta)\rightarrow H_{T}(f^{-1}(c_{0});\eta)$ is an isomorphism, hence surjective. Now suppose inductively that $H_{T}(X_{c_{i}-\epsilon};\eta)\cong H_{T}(X_{c_{(i-1)}+\epsilon};\eta)\rightarrow H_{T}(f^{-1}(c_{0});\eta)$ is surjective for some $i$. We must show that $H_{T}(X_{c_{i}+\epsilon};\eta)\rightarrow H_{T}(X_{c_{i}-\epsilon};\eta)$ is surjective. By Lemma 3.8, it will suffice to show that $Eul_{T}(\nu_{i})$ is not a zero divisor for $H_{T}(C_{i};\eta)$, where $C_{i}$ is a connected component of $N$. This follows from Lemma 3.9 using $S=T_{C_{i}}$. ∎ ###### Proof of Theorem 3.4. ∎ ## 4\. Generalized complex geometry Let $V$ be an $n$ dimensional vector space. There is a natural bi-linear pairing of signature $(n,n)$ on $V\oplus V^{*}$ which is defined by $\langle X+\alpha,Y+\beta\rangle=\dfrac{1}{2}(\beta(X)+\alpha(Y)).$ A generalized complex structure on a vector space $V$ is an orthogonal linear map $\mathcal{J}:V\oplus V^{*}\rightarrow V\oplus V^{*}$ such that $\mathcal{J}^{2}=-1$. Let $L\subset V_{\mathbb{C}}\oplus V^{*}_{\mathbb{C}}$ be the $\sqrt{-1}$ eigenspace of the generalized complex structure $\mathcal{J}$. Then $L$ is maximal isotropic and $L\cap\overline{L}=\\{0\\}$. Conversely, given a maximal isotropic $L\subset V_{\mathbb{C}}\oplus V^{*}_{\mathbb{C}}$ so that $L\cap\overline{L}=\\{0\\}$, there exists an unique generalized complex structure $\mathcal{J}$ whose $\sqrt{-1}$ eigenspace is exactly $L$. Let $M$ be a manifold. A generalized almost complex structure on a manifold $M$ is an orthogonal bundle map $\mathcal{J}:TM\oplus T^{*}M\rightarrow TM\oplus T^{*}M$ such that for any $x\in M$, $\mathcal{J}_{x}$ is a generalized complex structure on the vector space $T_{x}M$. Given a closed three form $H\in\Omega^{3}(M)$, an $H$-twisted generalized complex structure $\mathcal{J}$ is a generalized almost complex structure such that the sections of the $\sqrt{-1}$ eigenbundle of $\mathcal{J}$ are closed under the $\eta$-twisted Courant bracket, i.e., the bracket defined by the formula $[X+\xi,Y+\zeta]=[X,Y]+L_{X}\zeta- L_{Y}\xi-\dfrac{1}{2}d\left(\zeta(X)-\xi(Y)\right)+\iota_{Y}\iota_{X}H.$ A generalized almost Kähler structure is a pair of two commuting generalized almost complex structures $\mathcal{J}_{1}$,$\mathcal{J}_{2}$ such that $\langle-\mathcal{J}_{1}\mathcal{J}_{2}\xi,\xi\rangle>0$ for any $\xi\neq 0\in C^{\infty}(T_{\mathbb{C}}M\oplus T^{*}_{\mathbb{C}}M)$, where $\langle\cdot,\cdot\rangle$ is the canonical pairing on $T_{\mathbb{C}}M\oplus T^{*}_{\mathbb{C}}M$. A generalized almost Kähler structure $(\mathcal{J}_{1},\mathcal{J}_{2})$ is called an $H$-twisted generalized Kähler structure if both $\mathcal{J}_{1}$ and $\mathcal{J}_{2}$ are $H$-twisted generalized complex structures. Given a generalized almost Kähler structure $(\mathcal{J}_{1},\mathcal{J}_{2})$, define $\mathcal{G}(A,B):=\langle-\mathcal{J}_{1}\mathcal{J}_{2}A,B\rangle$, $A,B\in C^{\infty}(TM\oplus T^{*}M)$. Then $\mathcal{G}$ is a Riemannian metric on $TM\oplus T^{*}M$, and its restriction to $TM$ defines a Riemannian metric $g$ on $M$. Let $G=-\mathcal{J}_{1}\mathcal{J}_{2}$. Since $G^{2}=id$, $TM\oplus T^{*}M=C_{+}\oplus C_{-}$, where $C_{\pm}$ is the $\pm 1$-eigen-bundle of $G$. Let $\pi:TM\oplus T^{*}M\rightarrow TM$ be the projection map. Then $\pi\mid_{C_{\pm}}:C_{\pm}\rightarrow TM$ is an isomorphism. Since $\mathcal{J}_{1}$ commutes with $G$, $C_{\pm}$ is invariant under $\mathcal{J}_{1}$. By projecting from $C_{\pm}$, $\mathcal{J}_{1}$ induces two almost complex structure $I_{+}$ and $I_{-}$ on $TM$ which are compatible with the Riemannian metric $g$. Conversely, if there are two almost complex structures $I_{+}$ and $I_{-}$ which are compatible with a Riemannian metric $g$ on $M$, then (4.1) $\mathcal{J}_{1/2}=\dfrac{1}{2}\left(\begin{matrix}&1&0\\\ &b&1\end{matrix}\right)\left(\begin{matrix}&I_{+}\pm I_{-}&-(\omega_{+}^{-1}\mp\omega_{-}^{-1})\\\ &\omega_{+}\mp\omega_{-}&-(I_{+}^{*}\pm I_{-}^{*})\end{matrix}\right)\left(\begin{matrix}&1&0\\\ &-b&1\end{matrix}\right)$ is a generalized almost Kähler structure, where $\omega_{\pm}=gI_{\pm}$ are the fundamental $2$-forms of the Hermitian structures $(g,I_{\pm})$, and $b$ is a two form. The following remarkable result is due to Gualtieri. ###### Theorem 4.1. ([Gua03]) An $H$-twisted generalized Kähler structure is equivalent to a triple $(g,I_{+},I_{-})$ consisting of a Riemannian metric $g$ and two integrable almost complex structure compatible with $g$, satisfying the integrability conditions: $d_{+}^{c}\omega_{+}+d^{c}_{-}\omega_{-}=0,\,\,H+db=d_{+}^{c}\omega_{+},\,\,dd^{c}_{\pm}\omega_{\pm}=0,$ where $\omega_{\pm}=gI_{\pm}$, $d^{c}_{\pm}$ are the $i(\overline{\partial}-\partial)$ operator associated to the complex structure $I_{\pm}$, and $b$ is a two form. In particular, a triple $(g,I_{+},I_{-})$ satisfying the above assumption defines a generalized Kähler pair $\mathcal{J}_{1},\mathcal{J}_{2}$ by the formula (4.1). We close this section with a quick review of generalized complex submanifolds as introduced in [BB03] (See also [BS06]). Although [BB03] only defined generalized complex submanifolds for untwisted generalized complex structures, the definition given there extends naturally to the twisted case as well. Let $W$ be a submanifold of an $\eta$-twisted generalized complex manifold $(M,\mathcal{J})$, let $L\subset T_{\mathbb{C}}M\oplus T^{*}_{\mathbb{C}}M$ be the $\sqrt{-1}$-eigenbundle of $\mathcal{J}$, and let $i:W\rightarrow M$ be the inclusion map. At each point $x\in N$ set $L_{W,x}=\\{X+(\xi\mid_{T_{\mathbb{C}}W}):X+\xi\in L\cap(T_{\mathbb{C},x}W\oplus T^{*}_{\mathbb{C},x}M)\\}.$ This defines a maximally isotropic distribution of $T_{\mathbb{C}}W\oplus T_{\mathbb{C}}^{*}W$ whose sections are closed under the $i^{*}H$-twisted Courant bracket. If $L_{W}$ is a sub-bundle of $T_{\mathbb{C}}W\oplus T^{*}_{\mathbb{C}}W$ and if $L_{W}\cap\overline{L}_{W}=0$, then $W$ is said to be a generalized complex submanifold. 111 It is noteworthy that the sufficient and necessary conditions for $W$ to be a generalized complex submanifolds have been found in [BS06]. It is clear from the definition that if $W$ is a generalized complex submanifold then there exists a unique $i^{*}H$-twisted generalized complex structure $\mathcal{J}_{W}$ on $W$ whose $\sqrt{-1}$-eigenbundle is exactly $L_{W}$. It is well-known that the fixed point submanifold of a symplectic torus action on a symplectic manifold is a symplectic submanifold. [Lin06] extends this fact to generalized complex manifolds. ###### Proposition 4.2. Suppose the action of a torus $T$ on an $H$-twisted generalized complex manifold $(M,\mathcal{J})$ preserves the generalized complex structure $\mathcal{J}$. And suppose $Z$ is a connected component of the fixed point set. Then $Z$ is a generalized complex submanifold of $M$. Let $i:Z\rightarrow M$ be the inclusion map. Then $Z$ carries a $i^{*}H$-twisted generalized complex structure. ## 5\. Generalized moment maps First we recall the definition of Hamiltonian actions on $H$-twisted generalized complex manifolds given in [LT05]. ###### Definition 5.1. 222Indeed, Condition (b) was not imposed in [LT05, Definition A.2.]. However, in order to make the quotient construction work, Tolman and the author made it clear in [LT05, Prop. A.7, A.10] that $H+\alpha$ must be equivariantly closed in the usual equivariant Cartan model. [LT05]) Let a compact Lie group $G$ with Lie algebra $\mathfrak{g}$ act on a manifold $M$, preserving an $H$-twisted generalized complex structure $\mathcal{J}$, where $H\in\Omega^{3}(M)^{G}$ is closed. The action of $G$ is said to be Hamiltonian if there exists a smooth equivariant function $\mu:M\rightarrow\mathfrak{g}^{*}$, called the generalized moment map, and a 1-form $\alpha\in\Omega^{1}(M,\mathfrak{g}^{*})$, called the moment one form, so that * a) $\mathcal{J}d\mu^{\xi}=-\xi_{M}-\alpha^{\xi}$ for all $\xi\in\mathfrak{g}$, where $\xi_{M}$ denotes the induced vector field. * b) $H+\alpha$ is an equivariantly closed three form in the usual Cartan Model. ###### Remark 5.2. An $H$-twisted generalized complex structure $\mathcal{J}:TM\oplus T^{*}M\rightarrow TM\oplus T^{*}M$ induces by restriction and projection a map $\beta:T^{*}M\rightarrow TM$ which is a real Poisson bi-vector, see for instance [Gua03] and [BS06]. If the action of a compact Lie group $G$ on a generalized complex manifold $(M,\mathcal{J})$ is Hamiltonian with a generalized moment map $\mu\rightarrow\mathfrak{g}^{*}$, then a straightforward calculation shows $-\beta(d\mu^{\xi})=\xi_{M},$ where $\xi_{M}$ is the vector field on $M$ induced by $\xi\in\mathfrak{g}$. This shows clearly that the action of $G$ is Hamiltonian with respect to the Poisson bi-vector $\beta$. Let a compact Lie group $G$ act on a twisted generalized complex manifold $(M,\mathcal{J})$ with generalized moment map $\mu$. Let $\mathcal{O}_{a}$ be the co-adjoint orbit through $a\in\mathfrak{g}^{*}$. If $G$ acts freely on $\mu^{-1}(\mathcal{O}_{a})$, then $\mathcal{O}_{a}$ consists of regular values and $M_{a}=\mu^{-1}(\mathcal{O}_{a})/G$ is a manifold, which is called the generalized complex quotient. The following two results were proved in [LT05]. ###### Lemma 5.3. Let a compact Lie group $G$ act freely on a manifold $M$. Let $H$ be an invariant closed three form and let $\alpha$ be an equivariant mapping from $\mathfrak{g}$ to $\Omega^{1}(M)$. Fix a connection $\theta\in\Omega(M,\mathfrak{g}^{*})$. Then if $H+\alpha\in\Omega^{3}_{G}(M)$ is equivariantly closed, there exists a natural form $\Gamma\in\Omega^{2}(M)^{G}$ so that $\iota_{\xi_{M}}\Gamma=\alpha^{\xi}$. Thus $H+\alpha+d_{G}\Gamma\in\Omega^{3}(M)^{G}\subset\Omega_{G}^{3}(M)$ is closed and basic and so descends to a closed form $\widetilde{H}\in\Omega^{3}(M/G)$ so that $[\widetilde{H}]$ is the image of $[H+\alpha]$ under the Kirwan map. ###### Proposition 5.4. Assume there is a Hamiltonian action of a compact Lie group $G$ on an $H$-twisted generalized complex manifold $(M,\mathcal{J})$ with generalized moment map $\mu:M\to\mathfrak{g}^{*}$ and moment one-form $\alpha\in\Omega^{1}(M,\mathfrak{g}^{*})$. Let $\mathcal{O}_{a}$ be a co- adjoint orbit through $a\in\mathfrak{g}^{*}$ so that $G$ acts freely on $\mu^{-1}(\mathcal{O}_{a})$. Given a connection on $\mu^{-1}(\mathcal{O}_{a})$, the generalized complex quotient $M_{a}$ inherits an $\widetilde{H}$-twisted generalized complex structure $\widetilde{\mathcal{J}}$, where $\widetilde{H}$ is defined as in the Lemma 5.3. Up to $B$-transform, $\widetilde{J}$ is independent of the choice of connection. ### 5.1. Nitta’s theorem and compatibility The rest of this section is devoted to the proof of Proposition 1.1, which indeed has already been established by Nitta in [NY07]. However, since Proposition 1.1 is central to our paper and since we believe more details are needed in Nitta’s argument to make it more accessible, we will present a self- contained detailed proof in our paper. The essential step in the proof of Proposition 1.1 is the non-trivial observation that the restriction of $\alpha^{\xi}$ to the fixed point set $F^{\xi}$ vanishes. To prove it in full generality, we are going to use the maximum principle of pseudo-holomorphic functions on almost complex manifolds, as advocated in [NY07, Prop. 3.1]. Note that our proof differs slightly from the one given in [NY07]. For instance, our proof does not involve the use of a Levi-Civita connection, and we apply Proposition 4.2 in an essential way. We would also like to mention that when the generalized complex manifolds have constant types, one can construct more elementary proofs using the Darboux theorem of generalized complex structures [Gua03]. ###### Lemma 5.5. Suppose the trivial action of a torus $T$ on a compact $H$-twisted generalized complex manifold $(M,\mathcal{J})$ is Hamiltonian with a generalized moment map $\mu$ and a moment one form $\alpha$. Then $d\mu^{\xi}=\alpha^{\xi}=0$ for all $\xi\in\mathfrak{t}$. ###### Proof. It has been shown that there exists a generalized almost complex structure $\mathcal{J}_{2}$ such that $\mathcal{J}_{1}=\mathcal{J}$ and $\mathcal{J}_{2}$ form a generalized almost Kähler pair, see for instance, [Ca06, Sec. 3] and [NY07]. As we explained in Section 4, the generalized almost Kähler structure induces a triple $(g,I_{+},I_{-})$ consisting of a Riemannian metric $g$ and two almost complex structures $I_{+}$ and $I_{-}$ compatible with $g$; moreover, one can reconstruct $\mathcal{J}_{1}$ and $\mathcal{J}_{2}$ from the triple $(g,I_{+},I_{-})$ using Formula (4.1). Given $\xi\in\mathfrak{t}$, by assumption we have $\mathcal{J}_{1}d\mu^{\xi}=\alpha^{\xi}$, i.e., (5.1) $\dfrac{1}{2}\left(\begin{matrix}&1&0\\\ &b&1\end{matrix}\right)\left(\begin{matrix}&I_{+}+I_{-}&-(\omega_{+}^{-1}-\omega_{-}^{-1})\\\ &\omega_{+}-\omega_{-}&-(I_{+}^{*}+I_{-}^{*})\end{matrix}\right)\left(\begin{matrix}&1&0\\\ &-b&1\end{matrix}\right)\left(\begin{matrix}&0\\\ &d\mu^{\xi}\end{matrix}\right)=\left(\begin{matrix}&0\\\ &\alpha^{\xi}\end{matrix}\right),$ where $\omega_{\pm}=gI_{\pm}$. A straightforward calculation shows that $\omega_{+}^{-1}(d\mu^{\xi})=\omega_{-}^{-1}(d\mu^{\xi}),\,\,\,I_{+}^{*}d\mu+I_{-}^{*}d\mu^{\xi}=2\alpha^{\xi}.$ Since $I_{\pm}$ are compatible with $g$, we have $\omega_{\pm}=gI_{\pm}=-I_{\pm}^{*}g$ and so $\omega_{\pm}^{-1}=g^{-1}I_{\pm}^{*}$. It follows from $\omega_{+}^{-1}(d\mu^{\xi})=\omega_{-}^{-1}(d\mu^{\xi})$ that $I_{+}^{*}d\mu^{\xi}=I_{-}^{*}d\mu^{\xi}$. Thus $I^{*}_{\pm}d\mu^{\xi}=\alpha^{\xi}$. However, since by assumption the trivial action is Hamiltonian, condition b) in Definition 5.1 implies that $d\alpha^{\xi}=0$. Locally, we can always find a function $h$ such that $I^{*}_{\pm}d\mu^{\xi}=dh$. If the generalized almost complex structure $\mathcal{J}_{2}$ is integrable, i.e., $(\mathcal{J}_{1},\mathcal{J}_{2})$ forms a generalized Kähler pair, then $I_{\pm}$ must be integrable complex structures and $\mu^{\xi}$ is locally the real part of the $I_{\pm}$-holomorphic function $\mu^{\xi}+\sqrt{-1}h$. If $M$ is compact, it follows from the maximum principle of the real part of a holomorphic function that $\mu^{\xi}$ has to be a constant. In the general case, locally $\mu^{\xi}$ is the real part of a pseudo-holomorphic function with respect to almost complex structures $I_{\pm}$. By the maximum principle of the real part of a pseudo-holomorphic function as we explained in Appendix A, $\mu^{\xi}$ has to be a constant provided $M$ is compact. ∎ Now consider the Hamiltonian action of a compact connected torus $T$ on a compact $H$-twisted generalized complex manifold $M$ with a generalized moment map $\mu$. Suppose that $\xi\in\mathfrak{t}$ generates a compact connected sub-torus $T_{1}$ in $T$, i.e., $\xi$ is a generic element in the Lie algebra $\mathfrak{t}_{1}$ of $T_{1}\subset T$. ###### Lemma 5.6. Under the above assumptions, the critical set $\text{Crit}(\mu^{\xi})=\\{x\in M\mid(d\mu^{\xi})_{x}=0\\}$ coincides with the fixed point set $F$ of the $T_{1}$ action on $M$ for any $\xi\in\mathfrak{g}$. ###### Proof. The inclusion $\text{Crit}(\mu^{\xi})\subset F$ is obvious. It suffices to show that for any $x\in F$ we have $(d\mu^{\xi})_{x}=0$. By Proposition 4.2 the fixed point set $F$ of the $T_{1}$-action is a generalized complex submanifold. Moreover, it follows from [Lin07, Lemma 4.8] that the induced trivial action of $T_{1}$ on $F$ is Hamiltonian with the generalized moment map $\mu\mid_{F}:F\rightarrow\mathfrak{t}_{1}^{*}$. Since $M$ is compact, $F$ has to be compact itself. By Lemma 5.5, we have $(d\mu^{\xi})\mid_{F}=0$. Choose a $T_{1}$-invariant metric on $M$. Since $\mu^{\xi}$ is $T_{1}$-invariant, $\text{grad}(\mu^{\xi})$, the gradient flow of $\mu^{\xi}$, is also invariant under the linearized $T_{1}$ action. It follows that at each point $x\in F$, $\text{grad}(\mu^{\xi})$ is tangent to $F$. Thus $\langle\text{grad}\mu^{\xi},d\mu^{\xi}\rangle_{x}=0$. This implies that for any $x\in F$, $(d\mu^{\xi})_{x}=0$. ∎ ###### Remark 5.7. It is clear from the proof that in the statement of Lemma 5.6, we need only to assume that all fixed points submanifolds are compact. This compactness assumption here is essential. For instance, Hu [Hu05, Sec. 4.5] constructed an example of a Hamiltonian $S^{1}$ action on a generalized complex manifold with a proper moment map $f$ such that $\mathrm{crit}(f)\nsubseteqq M^{S^{1}}$. Lemma 5.6 fails because in Hu’s example the fixed point submanifold is a copy of complex plane $\mathbb{C}$ which is non-compact. We are ready to give a proof of Proposition 1.1. We are going to use the same notations as in Lemma 5.6. ###### Proof of Proposition 1.1. Let $f:=\mu^{\xi}$. In view of Lemma 5.6, it suffices to show that the Hessian of $f$ is nondegenerate in the normal direction to $M^{T_{1}}$. Because $\xi$ is generic, $M^{T_{1}}=ker(X)$. The vector field $X$ linearizes at $p\in M^{T_{1}}$ to $\mathcal{A}\in End(T_{p}M)$ defined by the formula $\mathcal{A}(w_{p})=[X,w]_{p}$ for $w\in C^{\infty}(TM)$ . Since $T_{1}$ is connected we have $T_{p}(M^{T_{1}})=\operatorname{\mathrm{ker}}\mathcal{A}$ as subsets of $T_{p}M$. The Hessian of $f$ at $p$ is a linear map $Hess_{p}(f):T_{p}M\rightarrow T_{p}^{*}M$, defined by $Hess_{p}(f)(w_{p})=(dL_{w}(f))_{p}$ where $w\in C^{\infty}(TM)$. We need to show that for all $p\in\mathrm{crit}(f)$, $\operatorname{\mathrm{ker}}(Hess_{p}(f))\subset T_{p}\mathrm{crit}(f)$. Now let $\beta$ be the canonical Poisson bivector associated to the generalized complex structure $\mathcal{J}$. As we explained in Remark 5.2, we have $X=-\beta df$. Thus $\displaystyle-[X,w]=L_{w}X=-L_{w}(\beta df)=-(L_{w}\beta)df-\beta(L_{w}df).$ Thus, $\displaystyle\mathcal{A}(w_{p})=(L_{w}\beta)_{p}df_{p}+\beta_{p}(L_{w}df)_{p}=\beta_{p}(L_{w}df)_{p}=\beta_{p}(Hess_{p}(f)(w_{p})),$ where we’ve used that $df_{p}=0$ and $(L_{w}df)_{p}=(dL_{w}f)_{p}=Hess_{p}(f)(w_{p})$. Thus $\operatorname{\mathrm{ker}}(Hess_{p}(f))\subset\operatorname{\mathrm{ker}}(\mathcal{A})=T_{p}(M^{T_{1}})=T_{p}\mathrm{crit}(f)$ as desired. ∎ ###### Remark 5.8. Choose an invariant generalized almost complex structure $\mathcal{J}_{2}$ such that $(\mathcal{J},\mathcal{J}_{2})$ form a generalized almost Kähler pair. It is easy to show that for any $p\in M$, $T_{p}M$ splits as the direct sum of $T_{p}M^{T_{1}}$ and $N$, where $N$ is the orthogonal complement of $T_{p}M^{T_{1}}$ in $T_{p}M$ with respect to the Riemannian metric induced by the generalized almost Kähler pair $(\mathcal{J},\mathcal{J}_{2})$; moreover, the vector space $N$ inherits a generalized Kähler structure which is invariant under the linearized action of $T_{1}$ on $N$. It follows that $N$ admits a complex structure which is invariant under the operator $\mathcal{A}$ which we defined in the proof of Proposition 1.1. As a direct consequence, we see that the $Hess_{p}(f)$ must have even index. So $f=\mu^{\xi}$ must be a Morse-Bott function of even index. It follows easily from Proposition 1.1 that the twisting form $H+\alpha$ is compatible with the torus action (Definition 3.2). ###### Corollary 5.9. Let $T\times M\rightarrow M$ be a Hamiltonian $T$-action for a compact, connected $H$-twisted generalized complex manifold $M$ with moment map $\mu:M\rightarrow\mathfrak{t}^{*}$ and moment 1-form $\alpha$. For $x\in M$, denote $\mathfrak{t}_{x}$ to be the infinitesimal stabilizer of $x$. Then $ker(\alpha_{x})\supseteq\mathfrak{t}_{x}$ where we regard $\alpha_{x}$ as an element of $Hom(\mathfrak{t},T_{x}^{*}M)$. ###### Proof. Condition a) of Definition 5.1 asserts $-\xi_{M}-\alpha^{\xi}=\mathcal{J}d\mu^{\xi}$ for all $\xi\in\mathfrak{t}$. Proposition 1.1 says that if $(\xi_{M})_{x}=0$ then $d\mu^{\xi}_{x}=0$ and consequently $\alpha_{x}^{\xi}=0$. ∎ ## 6\. Kirwan injectivity and Surjectivity and its application Proposition 1.1 establishes that the moment map $\mu$ of a compact $H$-twisted generalized Hamiltonian $T$-space is a nondegenerate abstract moment map and Corollary 5.9 establishes that the equivariant twisting 3-form $H+\alpha$ is compatible. Thus Theorems 1.3 and 1.4 follow from Theorem 3.3 and Theorem 3.4 respectively. Theorem 1.2 then follows from Proposition 2.18 because the restriction of a generalized Hamiltonian action is generalized Hamiltonian. One of the early motivations for this paper was to prove the following result. ###### Theorem 6.1. Let $(M,J,H)$ be a complex generalized complex manifold with nonexact twisting $H$. Then any generalized Hamiltonian torus action must have fixed point locus of dimension at least four. ###### Proof. Let $T$ be the torus and $\mu$ the moment map. By Proposition 1.1 and Corollary 5.9 $(M,T,\mu)$ is an abstract moment map with compatible equivariant twisting $H+\alpha$ where $\alpha$ is the moment 1-form. Thus by Corollary 3.11, $\operatorname{\mathrm{dim}}H(M;H)=\operatorname{\mathrm{dim}}H(M^{T};H).$ On the other hand, $0$ is always a compatible twisting so $\operatorname{\mathrm{dim}}H(M)=\operatorname{\mathrm{dim}}H(M^{T}).$ Since $H$ is not exact, we know by (2.4) that $\operatorname{\mathrm{dim}}H(M;H)<\operatorname{\mathrm{dim}}H(M)$. Consequently $H(M^{T};H)<H(M^{T})$ and we conclude that the restriction of $H$ to $M^{T}$ is _not_ exact. Since $H$ is a 3-form, it must be that $M^{T}$ has a component of dimension 3 or more, and because $M^{T}$ is even dimensional (see Prop. 4.2) this completes the proof. ∎ Theorem 6.1 stands in stark contrast with the symplectic world, where Hamiltonian actions with isolated fixed points abound (e.g. toric manifolds). In the course of writing this paper, we discovered a proof of Theorem 6.1 that avoids twisted cohomology and in fact leads to even stronger constraints. We will present these arguments in a future paper, along with new examples of compact GC Hamiltonian actions. ## Appendix A Maximum principle for pseudo-holomorphic functions on almost complex manifolds In this section, we give a self-contained proof of the maximum principle for pseudo-holomorphic functions on almost complex manifolds. We believe that the maximum principle in this setting should have been known to experts working in the related areas and we are not claiming any originality. We are presenting a proof here just because the central results of our paper are built upon it for which we can not find a good reference. Let $(M,J)$ be an almost complex manifold. Then the almost complex structure induces a splitting of the complexified cotangent bundle $T^{*}_{\mathbb{C}}(M)=T^{*}(M)^{1,0}\oplus T^{*}(M)^{0,1}$. In this context, a complex valued function $f+\mathfrak{i}g\in C^{\infty}(M)$ is defined to be a pseudo-holomorphic function on $M$ if $(df)_{x}+\mathfrak{i}(dg)_{x}\in T_{x}^{*}(M)^{1,0}$ for any $x\in M$. In this appendix, we prove the following result. ###### Theorem A.1. Suppose $(M,J)$ is an almost complex manifold. Suppose $f$ is the real part of a pseudo-holomorphic function $f+\mathfrak{i}g$ on $M$. Then for any $x\in M$ there exists an open neighborhood $B\ni x$ such that $\operatorname{\mathrm{sup}}_{B}f=\operatorname{\mathrm{sup}}_{\partial B}f,\,\,\,\operatorname{\mathrm{inf}}_{B}f=\operatorname{\mathrm{inf}}_{\partial B}f.$ Before beginning the proof, we first recall the maximum principle for the elliptic partial differential equations of second order as treated in [GT1977]. Let $L$ be a second order linear differential operator on a domain $\Omega$ of $R^{m}$ given by (A.1) $Lu=a^{ij}D_{ij}u+b^{i}D_{i}u+c_{i}u,$ where $a^{ij}=a^{ji}$, $D_{i}u=\frac{\partial u}{\partial x_{i}},D_{ij}u=\frac{\partial^{2}u}{\partial x_{i}\partial x_{j}}$. $L$ is said to be elliptic at $x\in\Omega$ if the coefficient matrix $[a^{ij}(x)]$ is positive definite. $L$ is said to be elliptic in $\Omega$ if it is elliptic at each point of $\Omega$. The following maximum principle is a fundamental result in the theory of elliptic operators. ###### Theorem A.2. ([GT1977, Thm. 3.1]) Let $L$ be an elliptic operator in the bounded domain $\Omega$. Suppose that (A.2) $Lu\geq 0(\leq 0)\text{ in }\Omega,\,\,\,c=0\,\text{ in }\Omega,$ with $u\in C^{2}(\Omega)\cap C^{0}(\overline{\Omega})$. Then $\operatorname{\mathrm{sup}}_{\Omega}u=\operatorname{\mathrm{sup}}_{\partial\Omega}u,\,\,\,\operatorname{\mathrm{inf}}_{\Omega}u=\operatorname{\mathrm{inf}}_{\partial\Omega}u.$ We are ready to present a proof of Theorem A.1. ###### Proof. Given an arbitrary point $p\in M$, we can choose a coordinate neighborhood $(U,x^{1},x^{2},\cdots,x^{2n})$ around $p$ such that under this coordinate system the almost complex $J(x)=[J^{i}_{j}(x)]$ coincides with the standard complex structure on $R^{2n}$ at the point $p$, i.e., $J(p)=\left[\begin{matrix}0_{n}&I_{n}\\\ -I_{n}&0_{n}\end{matrix}\right],$ where $0_{n}$ denotes the $n\times n$ zero matrix and $I_{n}$ the $n\times n$ identity matrix. Note that $\frac{\partial}{\partial x_{k}}+\mathfrak{i}J\frac{\partial}{\partial x_{k}}\in C^{\infty}(T(M)^{0,1})$ for any $1\leq k\leq 2n$. We have $<\frac{\partial}{\partial x_{k}}+\mathfrak{i}J\frac{\partial}{\partial x_{k}},df+\mathfrak{i}dg>=0$ since $df+\mathfrak{i}dg\in C^{\infty}(T^{*}(M)^{1,0})$. Observe $J\frac{\partial}{\partial x_{k}}=J_{k}^{p}\frac{\partial}{\partial x_{p}}$. We get the following generalized Riemann-Cauchy equations. (A.3) $\frac{\partial f}{\partial x_{k}}=J_{k}^{p}\frac{\partial g}{\partial x_{p}},\,\,\,\,\frac{\partial g}{\partial x_{k}}=-J_{k}^{p}\frac{\partial f}{\partial x_{p}}$ Therefore $\begin{split}\frac{\partial}{\partial x_{k}}\left(\frac{\partial f}{\partial x_{k}}\right)&=\frac{\partial}{\partial x_{k}}\left(J_{k}^{p}\frac{\partial g}{\partial x_{p}}\right)\\\ &=\frac{\partial J_{k}^{p}}{\partial x_{k}}\frac{\partial g}{\partial x_{p}}+J_{k}^{p}\frac{\partial}{\partial x_{p}}\left(\frac{\partial g}{\partial x_{k}}\right)\\\ &=\frac{\partial J_{k}^{p}}{\partial x_{k}}J_{p}^{q}\frac{\partial f}{\partial x_{q}}+J_{k}^{p}\frac{\partial}{\partial x_{p}}\left(-J_{k}^{q}\frac{\partial f}{\partial x_{q}}\right)\\\ &=\frac{\partial J_{k}^{p}}{\partial x_{k}}J_{p}^{q}\frac{\partial f}{\partial x_{q}}-J_{k}^{p}\frac{\partial J_{k}^{q}}{\partial x_{p}}\frac{\partial f}{\partial x_{q}}-J_{k}^{p}J_{k}^{q}\frac{\partial^{2}f}{\partial x_{p}\partial x_{q}}.\end{split}$ It follows $\frac{\partial^{2}f}{\partial x^{2}_{k}}+J_{k}^{p}J_{k}^{q}\frac{\partial^{2}f}{\partial x_{p}\partial x_{q}}+\left(J_{k}^{p}\frac{\partial J_{k}^{q}}{\partial x_{p}}-\frac{\partial J_{k}^{p}}{\partial x_{k}}J_{p}^{q}\right)\frac{\partial f}{\partial x_{q}}=0.$ Summing over the index $k$ we get a second order linear equation (A.4) $\sum_{k}\left(\frac{\partial^{2}f}{\partial x^{2}_{k}}+J_{k}^{p}J_{k}^{q}\frac{\partial^{2}f}{\partial x_{p}\partial x_{q}}\right)+\sum_{k}\left(J_{k}^{p}\frac{\partial J_{k}^{q}}{\partial x_{p}}-\frac{\partial J_{k}^{p}}{\partial x_{k}}J_{p}^{q}\right)\frac{\partial f}{\partial x_{q}}=0.$ Set $a^{pq}=\delta_{p}^{q}+\sum_{k}J_{k}^{p}J_{k}^{q},$ where $\delta_{p}^{q}$ is the Kronecker symbol. Since $J(p)$ coincides with the standard complex structure on $\mathbb{R}^{2n}$, a simple calculation shows that the matrix $[a^{pq}(p)]=2I_{2n}$, where $I_{2n}$ denotes the $2n\times 2n$ identity matrix. This shows clearly that the matrix $[a^{pq}]$ is a positive definite symmetric matrix at $p$ and so must be a positive definite symmetric matrix on an open ball $p\in B\subset U$. Thus Equation A.4 is a second order elliptic equation on a bounded ball $B$. Now Theorem A.1 is a simple consequence of Theorem A.2. ∎ ## Appendix B nondegenerate abstract moment maps In this section we will prove Lemma 3.12. This result was used without much explanation in appendix G of [GGK02] but was later recognized to be subtler than it appears. Here we provide a detailed proof using an extra hypothesis, the existence of an invariant almost (more generally stable) complex structure. Throughout, let $T$ be a compact torus with lie algebra $\mathfrak{t}$, $M$ be a smooth $T$-manifold, and $\phi:M\rightarrow\mathfrak{t}^{*}$ a nondegerate, abstract moment map ( Def. 3.1). As explained in [GGK02], condition 1 of Definition 3.1 is equivalent to the condition that for all $p\in M$, (B.1) $d\phi_{p}(T_{p}M)=\mathfrak{t_{p}}^{\perp}.$ Here given a subspace $\mathfrak{h}\subset\mathfrak{t}$, $\mathfrak{h}^{\perp}\subset\mathfrak{t}^{*}$ denotes the annihilator of $\mathfrak{h}\subset\mathfrak{t}$. That said, the next Lemma is not so surprising: ###### Lemma B.1. Let $p,q\in M$ satisfy $\mathfrak{t}_{p}=\mathfrak{t}_{q}=\mathfrak{g}$ where $\mathfrak{g}$ is the Lie algebra of a subtorus $G\subset T$. If $p$ and $q$ lie in the same connected component of $M^{G}$ then $\phi(p)+\mathfrak{t}_{p}^{\perp}=\phi(q)+\mathfrak{t}_{q}^{\perp}$ ###### Proof. The map $\phi_{\mathfrak{g}}:M\rightarrow\mathfrak{g}^{*}$ defined by composing $\phi$ with the projection $\pi_{\mathfrak{g}^{*}}:\mathfrak{t}^{*}\rightarrow\mathfrak{g}^{*}$ is a moment map for the restricted $G$ action. By definition, $\phi_{\mathfrak{g}}$ restricts to a locally constant function on $M^{G}$ so $\phi_{\mathfrak{g}}(p)=\phi_{\mathfrak{g}}(q)$ and so $\phi(p)-\phi(q)\in\operatorname{\mathrm{ker}}(\pi_{\mathfrak{g}^{*}})=\mathfrak{g}^{\perp}$. ∎ For compact $T$ manifold $M$, there can only be a finite number of distinct isotopy groups $T_{p}\subset T$, for each of which $M^{T_{p}}$ has a finite number of components. We deduce: ###### Corollary B.2. If $M$ is compact, then the set of vector spaces $\\{span(\phi(p))+\mathfrak{t}_{p}^{\perp}|p\in M\\}$ is finite. ###### Lemma B.3. Let $M$ be a compact $T$-manifold equipped with a nondegerate moment map $\phi$ and suppose that $0\in\mathfrak{t}^{*}$ is a regular value for $\phi$. There exists a codimension 1 subtorus $H\subset T$ with Lie algebra $\mathfrak{h}$ for which 0 is a regular value for $\phi_{\mathfrak{h}}=\pi_{\mathfrak{h}^{*}}\circ\phi$, where $\pi_{\mathfrak{h}^{*}}:\mathfrak{t}^{*}\rightarrow\mathfrak{h}^{*}$ is projection. ###### Proof. The hyperplane Grassmanian $Gr_{1}(\mathfrak{t})$ parametrizes the set of codimension one subspaces $\mathfrak{h}\subset\mathfrak{t}$. Those subspaces integrating to codimension one subtori form a dense subset of $Gr_{1}(\mathfrak{t})$, so to prove Lemma B.3 it will suffice to show that the set $U:=\\{\mathfrak{h}\in Gr_{1}(\mathfrak{t})|\text{ 0 is a regular value for }\phi_{\mathfrak{h}}=proj_{\mathfrak{h^{*}}}\circ\phi\\}$ contains a nonempty open set. It is somewhat clearer to work with the projective space $P(\mathfrak{t}^{*})$, which is canonically isomorphic to $Gr_{1}(\mathfrak{t})$ via the correspondence $\mathfrak{h}\leftrightarrow\mathfrak{h}^{\perp}$. Since $\mathfrak{h}^{\perp}=\operatorname{\mathrm{ker}}(\pi_{\mathfrak{h}^{*}})$ it follows easily that $0$ is a regular value for $\phi_{\mathfrak{h}}$ if and only if $\mathfrak{t}^{*}=im(d\phi_{p})+\mathfrak{h}^{\perp}=\mathfrak{t}_{p}^{\perp}+\mathfrak{h}^{\perp}$ for all $p\in M$ satisfying $\phi(p)\in\mathfrak{h}^{\perp}$. If $\phi(p)=0$ then $im(d\phi_{p})=\mathfrak{t}^{*}$ by hypothesis. If $\phi(p)\in\mathfrak{h}^{\perp}-0$, then $span(\phi(p))=\mathfrak{h}^{\perp}$. Thus if $\mathfrak{h}^{\perp}$ lies outside of the finite set of _proper_ vector subspaces described in Corollary B.2, the moment map $\phi_{\mathfrak{h}}$ is guaranteed to be regular at zero. This is an open and nonempty condition, completing the proof. ∎ Iterating Lemma B.3 enables us to prove (i) and (ii) of Proposition 3.12. In particular, we construct a sequence of subtori $T=T_{n}\supseteq T_{n-1}\supseteq T_{n-1}....\supseteq T_{1}$ such that the moment map $\phi_{k}$ for each $T_{k}$ is regular at $0$, and then choose arbitrarily $\xi_{k}\in\mathfrak{t}_{k}-\mathfrak{t}_{k-1}$. We prove (iii) in two steps. We denote the restricted function $f_{k}:=\phi^{\xi_{k+1}}|_{\phi_{k}^{-1}(0)}$. ###### Lemma B.4. The critical set $Crit(f_{k})=\\{p\in\phi^{-1}_{k}(0)|\text{ $(\mathfrak{t}_{k+1})_{p}\neq 0$ }\\}$. ###### Proof. For $p\in M$ we have $\operatorname{\mathrm{dim}}(d_{p}f_{k}(T_{p}\phi_{k}^{-1}(0)))=\operatorname{\mathrm{dim}}(d_{p}\phi_{k+1}(T_{p}\phi_{k}^{-1}(0)))=\operatorname{\mathrm{dim}}(d_{p}\phi_{k+1}(T_{p}M))-k=1-\operatorname{\mathrm{dim}}((\mathfrak{t}_{k})_{p}).$ ere we have used Equality B.1. ∎ It remains to prove that the critical points of $f_{k}:=\phi^{\xi_{k+1}}|_{\phi_{k}^{-1}(0)}$ are nondegenerate. According to [GGK02], if $M$ admits an invariant almost complex structure, then in the neighborhood of every orbit $M$ admits a symplectic structure for which the moment map is Hamiltonian. We can then use the following local canonical form [GS82] for symplectic Hamiltonian actions. ###### Lemma B.5. ([GS82]) Suppose that $M$ admits a $T$-invariant almost complex structure. For $p\in M$ choose a complimentary Lie subalgebra $\mathfrak{h}\subset\mathfrak{t}$ to $\mathfrak{t}_{p}$ so that $\mathfrak{t}^{*}=\mathfrak{h}^{*}\oplus\mathfrak{t}_{p}^{*}$. Then for some $T_{p}$ representation $V$, there is a $T$-equivariant diffeomorphism from an invariant neighborhood of $p$ to an invariant neighborhood of the zero section of the associated bundle $T\times_{T_{p}}(\mathfrak{h}^{*}\oplus V)$ sending the moment map $\phi$ to the map $\phi^{\prime}:T\times_{T_{p}}(\mathfrak{h}^{*}\oplus V)\rightarrow\mathfrak{h}^{*}\oplus\mathfrak{t}_{p}^{*}=\mathfrak{t}^{*}$ defined by $\phi^{\prime}(t,\eta,v)=(\eta,q(v))$, where $q:V\rightarrow\mathfrak{t}_{p}^{*}$ is a quadratic form. Applying this to the case $T=T_{k+1}$, $\phi=\phi_{k+1}$, $\mathfrak{h}=\mathfrak{t}_{k}$ at a critical point $p$ of $f_{k}$, we obtain a local model for $\phi_{k+1}$ near $p$ in $M$ $\phi^{\prime}:T_{k+1}\times_{(T_{k+1})_{p}}(\mathfrak{t}_{k}^{*}\oplus V)\rightarrow\mathfrak{t}_{k}^{*}\oplus\mathbb{R}=\mathfrak{t}_{k+1}^{*}$ where $\phi^{\prime}(t,\eta,v)=\eta+q(v)$. Here a neighborhood of $p$ in $\mu_{k}^{-1}(0)$ maps to $T_{k+1}\times_{(T_{k+1})_{p}}(\\{0\\}\oplus V)$ and $f_{k}$ corresponds (up to a nonzero scalar multiple) to the quadratic form $q$. Thus in some local coordinates, $f_{k}$ looks like a quadratic form near $p$ in $\phi^{-1}_{k}(0)$ and hence is nondegenerate, completing the proof of Lemma 3.12. ## Appendix C Further aspects of twisted equivariant cohomology We call a $G$-invariant open cover $\\{U_{\alpha}|\alpha\in I\\}$ of $M$ equivariantly good if all nonempty intersections of the $U_{\alpha}$ are tubular neighborhoods of some $G$-orbit in $M$. For example, every $G$-equivariant vector bundle over a compact manifold $M$ admits a _finite_ equivariantly good cover. ###### Lemma C.1. Let $M$ be a $G$-manifold admitting a finite equivariantly good open cover $\\{U_{\alpha}|\alpha\in I\\}$. Then for all $J\subset I$, and for any twisting $d_{G}$-closed 3-form $\eta\in\Omega_{G}^{3}(M)$ we have an isomorphism of $H_{G}(M)$-modules $H_{G}(\cap_{\alpha\in J}U_{\alpha};\eta)\cong H(BH)$ for some closed subgroup $H\subset G$. ###### Proof. By definition we know that $\cap_{\alpha\in J}U_{\alpha}$ is equivariantly homotopy equivalent to a homogeneous space $G/H$ for some subgroup $H\subset G$. Thus $\eta$, $H_{G}(\cap_{\alpha\in J}U_{\alpha})\cong H(G/H)$. But it is a standard result that $H_{G}(G/H)\cong H(BH)$ and $H^{3}(BH)=0$ (c.f. [AB84]). Thus $\eta$ is cohomologous to zero. It follows that $H_{G}(\cap_{\alpha\in J}U_{\alpha};\eta)\cong H_{G}(\cap_{\alpha\in J}U_{\alpha})\cong H(BH)$ ∎ Lemma C.1 can be used to prove twisted equivariant cohomology results using Mayer-Vietoris. ###### Proposition C.2. Let $M$ be a smooth $G$-manifold admitting a finite, equivariantly good open cover. Then for any twisting $\eta$, $H_{G}(M;\eta)$ is a finitely generated $(\hat{S}\mathfrak{g}^{*})^{G}$-module. ###### Proof. For any closed subgroup $H\subset G$, $H(BH)$ is finitely generated over $H(BG)\cong(\hat{S}\mathfrak{g}^{*})^{G}$ (the number of generators is bounded by the Weyl group). Then $H_{G}(M;\eta)$ is shown to be finitely generated by repeated application of Mayer-Vietoris and Lemma D.2. ∎ There a couple of different versions of twisted equivariant cohomology described in the literature and we take a moment to reconcile them. In [FHT02], it is defined as the cohomology of the complex of formal Laurent series $\Omega_{G}^{*}(M)((\beta))$ where $\beta$ has degree $-2$ with differential $d+\eta\beta$. This makes $\Omega_{G}^{*}(M)((\beta))$ into a graded complex, producing Z-graded cohomology $\tilde{H}_{G}^{*}(M;\eta)$. The reader can readily verify that $\sum_{k}\alpha^{n+2k}\otimes\beta^{k}\in\Omega^{n}_{G}(M)((\beta))$ is closed (exact) if and only if $\sum_{k}\alpha^{n+k}\in\hat{\Omega}_{G}(M)$ is closed (exact) for $d_{G}+\eta$. It follows that $\tilde{H}_{G}^{n}(M)\cong H_{G}^{[n]}(M)$ where on the $n$ is an integer and $[n]$ is its reduction mod $2$. Another version comes from [HuU06]. They define $\bar{H}_{G}(M;\eta)$ to be the cohomology of the complex $(\Omega(M)\otimes\hat{S}\mathfrak{g}^{*})$ with differential $d_{G}+\eta\wedge$. There is a natural injective chain map $(\Omega(M)\otimes\hat{S}\mathfrak{g}^{*})\hookrightarrow\hat{\Omega}_{G}(M)$ which induces an isomorphism in cohomology in most interesting cases: ###### Proposition C.3. Let $M$ be a smooth $G$-manifold admitting a finite, equivariantly good open cover. Then $\bar{H}_{G}(M)\cong H_{G}(M)$. ###### Proof. Using Equation 2.1, it is easy to prove the isomorphism holds on tubular neighborhoods of $G$-orbits. This extends to $M$ by repeated application of Mayer-Vietoris. ∎ This result allows us to compare $H_{G}(M;\eta)$ with the cohomology of the direct sum complex $(\Omega_{G}(M)=\bigoplus_{i}\Omega_{G}^{i}(M),d_{G}+\eta\wedge)$. ###### Proposition C.4. For $M$ admitting a finite equivariantly good cover, $H_{G}(M;\eta)$ is canonically isomorphic to $H(\Omega_{G}(M),d_{G}+\eta)\otimes_{(S\mathfrak{g}^{*})^{G}}(\hat{S}\mathfrak{g}^{*})^{G}$ (i.e. it is obtained by extension of scalars). ###### Proof. By Proposition C.3 $H_{G}(M;\eta)\cong\bar{H}_{G}(M;\eta)$. The formal power series ring $\hat{S}\mathfrak{g}^{*}$ is flat over $S\mathfrak{g}$ so it factors through taking cohomology (see Lemma D.5). Thus $\bar{H}_{G}(M;\eta)=H(\Omega_{G}(M)\otimes_{(S\mathfrak{g}^{*})^{G}}(\hat{S}\mathfrak{g}^{*})^{G};d+\eta)=H(\Omega_{G}(M);d+\eta)\otimes_{(S\mathfrak{g}^{*})^{G}}(\hat{S}\mathfrak{g}^{*})^{G}.$ ∎ Recall ([Mc01] Def. 3.8) that given a filtration $\\{F^{p}A\\}$ of a differential complex $(A,d)$, the associated spectral sequence $(E_{r}^{p},d_{r})$ is said to converge strongly to $H(A,d)$ if the induced maps $F^{p}H(A,d)/F^{p+1}H(A,d)\rightarrow E_{\infty}^{p}$ and $H(A,d)\rightarrow(\stackrel{{\scriptstyle lim}}{{\leftarrow}}H(A,d)/F^{p}H(A,d)$ are isomorphisms. The first of these ###### Proposition C.5. Let $M$ admit a finite $G$-equivariantly good open cover. Then the spectral sequences associated to the filtrations $\\{F^{p}\\}$ and $\\{L^{p}\\}$ of $(\hat{\Omega}_{G}(M),d_{G,\eta})$ described in §2.3 both converge strongly to $H_{G}(M;\eta)$. ###### Proof. First note that the filtrations are cofinal, satisfying $F^{2p-n}\subset L^{p}\subset F^{2p+n},$ so convergence of the spectral sequence determined by $\\{F^{p}\\}$ implies convergence of that for $\\{L^{p}\\}$. Clearly $\cup_{p}F^{p}=\hat{\Omega}_{G}(M)and\cap_{p}F^{p}=0$ so the filtration is exhaustive and weakly convergent. By ( [Mc01] Thm 3.2) it only remains to show that we have an isomorphism (C.1) $H_{G}(M;\eta)\cong\stackrel{{\scriptstyle lim}}{{\leftarrow}}H_{G}(M;\eta)/F^{p}H_{G}(M;\eta).$ First note that if $\eta=0$, then the filtration $F^{p}H_{G}(M)=\prod_{k\geq p}H_{G}^{k}(M)$ so (C.1) certainly holds. Next, note that wedging by $exp(b)$ for any $b\in\Omega_{G}^{2}(M)$ preserves the filtration $F^{p}$, so by Lemma 2.12, (C.1) must hold for exact $\eta$. Finally, because $\eta$ must become exact when restricted to tubular neighborhoods of orbits, we may use Mayer- Vietoris and the five lemma to prove the general case. ∎ Proposition C.5 was proven with more general hypotheses in [FHT02] using a Mittag-Lefler condition. We include our own proof because it is more elementary. ## Appendix D Commutative algebra The results in this section are standard. Our principal reference is Matsumura [Mat89] Chapter 8. Recall that a module is Noetherian is it satisfies the ascending chain condition. A ring is Noetherian if it is Noetherian over itself. Throughout, $R$ will denote a Noetherian, commutative integral domain. ###### Lemma D.1. Let $M$ be an $R$-module. The following are equivalent: (1) $M$ is Noetherian (2) $M$ is finitely generated (3) $M$ has a finite presentation ###### Lemma D.2. For a short exact sequence of $R$-modules $0\rightarrow K\rightarrow M\rightarrow N\rightarrow 0$ $M$ is Noetherian if and only if both $K$ and $N$ are Noetherian. An $R$-module $M$ is called flat if the functor $M\otimes_{R}:Mod_{R}\mapsto Ab$ from $R$-modules to abelian groups is exact. ###### Lemma D.3. For $R$ an Noetherian, integral domain, the following are flat over $R$: (1) the quotient field of $R$. (2) any $I$-adic completion of $R$, for $I\subset R$ an ideal. (3) any projective module over $R$. Recall that for $I\subset R$ an ideal and $M$ an $R$-module, the $I$-adic completion of $M$ is defined by the inverse limit: $\hat{M}=\stackrel{{\scriptstyle lim}}{{\leftarrow}}M/I^{k}M$ We have $\hat{R}$ is a ring and $\hat{M}$ is a $\hat{R}$-module. ###### Lemma D.4. If $M$ is finitely generated $R$-module, then there is a natural isomorphism of $\hat{R}$-modules: $\hat{M}\cong M\otimes_{R}\hat{R}$ Recall that a $R$-chain complex $(C,d)$ is an $R$-module $C$, equipped with a morphism $d:C\rightarrow C$ satisfying $d^{2}=0$. The homology is defined $H(C,d)=\operatorname{\mathrm{ker}}d/\operatorname{\mathrm{im}}d$ and is an $R$-module. Given a module $M$ over $R$, the tensor product $(C\otimes M,d\otimes 1)$ is a Z-chain complex. ###### Lemma D.5. (ex. 7.6 in [Mat89]) Let $(C,d)$ be an $R$-chain complex and let $M$ be flat over $R$. Then we have a natural isomorphism $H(C,d)\otimes_{R}M\cong H(C\otimes_{R}M,d\otimes_{R}1)$. The Jacobson radical is the ideal $J:=\\{r\in R|1-rs\text{ is a unit in R for all }s\in S\\}.$ ###### Lemma D.6. (Nakayama’s Lemma) Let $M$ be a nonzero finitely generated module over $R$ and $J$ the Jacobson radical of $R$. Then $JM\neq M$. We remark that Nakayama’s Lemma works for not necessarily commutative rings (see 4.3.10 [We94]) so we can apply it to super commutative rings like $H_{T}(M)$. We have some particular examples in mind. For instance the polynomial ring $A=\mathbb{R}[x_{1},...,x_{n}]$. If $I=(x_{1},...,x_{n})\subset A$ is the augmentation ideal, then the $I$-adic completion is $\hat{A}=\mathbb{R}[[x_{1},...,x_{n}]]$, the ring of formal power series, so $\hat{A}$ is flat over $A$. Both $A$ and $\hat{A}$ are Noetherian commutative integral domains so their quotient fields are flat over each of them. ## References * [AB82] M. Atiyah, R. Bott, Yang-Mills equations over Riemann surfaces, Phil. Trans. Soc. Lond. A, 308, 523 - 615, (1982). * [AB84] M. Atiyah, M. Bott, The moment map and equivariant cohomology, Topology 23 (1984), no. 1, 1 - 28. * [AS05] M. Atiyah, G. Segal, Twisted K-theory and cohomology, preprint, math.KT/0510674. * [AM99] M. Atiyah, I.G. Macdonald, Introduction to commutative algebra, Addison-Wesley Publishing Co.; Reading, Mass-London-Don Mills, Ont. 1969. * [BB03] O. Ben-Bassat, M. Boyarchenko, Submanifolds of generalized complex manifolds, J. Symplectic Geom. 2, no. 3, (2004) 309-355. * [BS06] J. Barton, M. Stienon, Generalized Complex Submanifolds, Preprint, math.DG/0603480 * [BCG05] H. Bursztyn, G. Cavalcanti, and M. Gualtieri, Reduction of Courant algebroids and generalized complex structures, math.DG/0509640, Adv. in Math., 211 (2), 726 - 765, 2007. * [BCRR05] U. Bruzzo, L. Cirio, P. Rossi, and V. Rubtsov, Equivariant cohomology and locolization for Lie algebroids, mah.DG/0506392 * [Ca05] Gil, Cavalcanti, New aspects of $dd^{c}$ lemma, Oxford D. Phil. thesis, math.DG/0501406. * [Ca06] Gil, Cavalcanti, The decomposition of forms and cohomology of generalized complex manifolds, J. Geom. Phyi. 57, 121-132, 2006 * [FHT02] D. Freed, M. Hopkins, C. Teleman, _T_ wisted equivariant K-theory with complex coefficients,Arxiv preprint math.AT/0206257, 2002. * [GHR84] Jr. S. Gates, C. Hull, and M. Rocek, Twisted multiplets and new supersymmetric non-linear $\sigma$-model, Nuclear Phys. B. 248(1) 157- 186, 1984. * [Gol02] R. F. Goldin, An effective algorithm for the cohomology ring of symplectic reductions, Geom. Funct. Anal., 12(3) 567-583, 2002. * [Gua03] M. Gualtieri, Generalized complex geometry, Oxford D. Phil. thesis, math.DG/0401221. * [Gua07] M. Gualtieri, Generalized complex geometry, math.DG/0703298. * [Gua04] Marco Gualtieri Generalized geometry and the Hodge decomposition, math.DG/0409093. * [GS82] Guillemin, Victor, Sternberg, Shlomo A normal form for the moment map, Differential geometric methods in mathematical physics (Jerusalem, 1982), 161–175, Math. Phys. Stud., 6, Reidel, Dordrecht, 1984. * [GS84] V. Guillemin, S.Sternberg, Symplectic Techniques in Physics, Cambridge University Press, Cambridge, 1984. * [GS99] V. Guillemin, S.Sternberg, Super-symmetry and equivariant de Rham theory, With an appendix containing two reprints by Henri Cartan. Mathematics Past and Present. Springer-Verlag, Berlin, 1999. * [GT1977] David Gilbarg ), Neil S. Trudinger, Elliptic Partial Differential Equations of Second Order, Springer-Verlag, 1998\. * [GGK02] gkk V. Ginzburg, V. Guillemin, Y. Karshon, Moment maps, Cobordisms, and Hamiltonian actions, American Mathematical Society, Mathematical surveys and Monographys, Vol. 98. * [H02] N. Hitchin, Generalized Calabi-Yau manifolds, Q.J. Math., 54 (3) 281-308, 2003 * [Hu05] S. Hu, Hamiltonian symmetries and reduction in generalized geometry, math.DG/0509060. * [HuU06] S. Hu, B. Uribe, Extended manifolds and extended equivariant cohomology, math.DG/0608319. * [Kir86] F., Kirwan, Cohomology of quotients in symplectic and algebraic geometry, Mathematical Notes, 31, Princeton, NJ, 1984. * [KL04] A. Kapustin, Y. Li, Topological sigma-models with H-flux and twisted generalized complex manifolds, hep-th/0407249. * [KT06] A. Kapustin, A. Tomasiello, The general $(2,2)$ gauged sigma model with three-form flux, preprint, hep-th/0610210. * [LT05] Yi Lin, Susan Tolman, Symmetries in generalized Kähler geometry, Comm. Math. Phys., 208 (2006), 199-222, math.DG/0509069. * [Lin06] Yi Lin, Generalized geometry, equivariant $\overline{\partial}\partial$-lemma and torus actions , the Journa of geometry and physics, 57 (2007) 1842-1860, also available on arxiv: math.DG/0607401. * [Lin07] Yi Lin, Equivariant cohomology theory of twisted generalized complex manifolds, to appear Commun. in Math. Physics, also available at math.DG/0704.2804. * [Lin07b] Yi Lin, Examples of Hamiltonian actions on twisted generalized complex manifolds, Preprint, 13 pages, 2007 * [Mar00] S. Martin, Symplectic quotients by a nonabelian group and by its maximal torus, Arxiv preprint math.SG/0001002, 2000. * [Mat89] H. Matsumura, Commutative Ring Theory, Cambridge University Press, 1989. * [Mc01] J. McCleary, A user’s Guide to Spectral Sequences, Second edition. Cambridge Studies in Advanced Mathematics, 58. Cambridge University press, Cambridge, 2001. * [NY06] Yasufumi Nitta, Reduction of generalized Calabi-Yau structures, to appear, J. Math. Soc. Japan, Vol 59, no. 4 (2007), also available on arxiv: math.DG/0611341. * [NY07] Yasufumi Nitta, Convexity properties for generalized moment maps I, math.DG/0710.3924. * [SX05] M. Stiénon and X. Ping, Reduction of generalized complex structure, math.DG/0509393 * [TW98] Susan Tolman and Jonathan Weitsman, The cohomology rings of abelian symplectic quotiens * [Va05] Vaisman, I. Reduction and submanifolds of generalized complex manifolds, math.DG/0511013. * [We94] C.A. Weibel An Introduction to Homological Algebra, Cambridge University Press, 1994.

arxiv-papers

2024-09-04T02:48:53.662255

0802.1350

# Equilibrium spin currents: Non-Abelian gauge invariance and color diamagnetism in condensed matter I. V. Tokatly Ilya.Tokatly@physik.uni-erlangen.de European Theoretical Spectroscopy Facility (ETSF), Dpto. Fisica Materiales, Universidad del Pais Vasco EHU-UPV, 20018 Donostia, Spain and Moscow Institute of Electronic Technology, Zelenograd, 124498 Russia ###### Abstract The spin-orbit (SO) interaction in condensed matter can be described in terms of a non-Abelian potential known in high-energy physics as a color field. I show that a magnetic component of this color field inevitably generates diamagnetic color currents which are just the equilibrium spin currents discussed in a condensed matter context. These dissipationless spin currents thus represent a universal property of systems with SO interaction. In semiconductors with linear SO coupling the spin currents are related to the effective non-Abelian field via Yang-Mills magnetostatics equation. ###### pacs: 72.25.-b, 72.25.Dc SO interaction is considered as an important ingredient of spintronics ZutFabSar2004 ; Rashba2007_Book as it allows to control spin degrees of freedom by electric means. Despite an increasing interest and growing number of publications on spin dynamics and spin currents in systems with SO interaction, a few basic questions remain unresolved up to now. In fact, even the very definition of spin currents is still debated Rashba2007_Book ; EngRasHal2007_Book ; Shi2006 . The reason for the controversies is that the spin is not conserved in a usual sense if SO interaction is present. The time derivative of the spin density ${\bf s}({\bf r},t)$ can not be represented in form of divergence of a current, but always contains an extra term – the spin torque. Hence it appears that any redefinition of the current can be compensated by correcting the torque in a way that preserves $\partial_{t}{\bf s}$. The problem of ambiguity of the spin current was sharpened by Rashba who noticed the presence of spin currents in a thermodynamically equilibrium 2D electron gas with Rashba SO interactions Rashba2003 . The physical reality of these dissipationless currents has been questioned as they do not accompanied by any spin accumulation, and their very appearance has been attributed to the ambiguity of the spin current concept. Recently Sonin2007a ; Sonin2007b Sonin proposed a way to detect the equilibrium spin flows in a ”Rashba medium”, and argued that there is nothing miraculous or unique in the presence of such equilibrium currents. They are known to exist in many different physical situations, e. g., spin supercurrents in a superfluid He3 or Meissner currents in superconductors (see. Ref. Sonin2007a, and references therein). I would like to add another simple example that, as we will see, is much more relevant – the diamagnetic currents responsible for the Landau diamagnetism in metals. This paper is aimed at completely removing a flavor of ambiguity and mystery in the questions of equilibrium spin flows and the definition of spin currents. The problem is resolved by making a link to non-Abelian gauge theories, and exploiting SU(2) gauge invariance of a many-body theory with SO interactions. The interpretation of SO interaction together with an external magnetic field as components $\bm{\mathcal{A}}_{\mu}$ of a non-Abelian four- potential is known for many years FroStu1993 , and it is becoming more and more popular nowadays SO-SU2 . However, the full power of non-Abelian gauge invariance in SO context is by far not explored note_SpinHall . In the present work I use this analogy to discover a simple physics behind equilibrium spin currents. We will see that these are nothing but diamagnetic color currents that appear as a response to an effective Yang-Mills magnetic field produced by SO interaction. Due to the gauge invariance they are nonzero only if the field strength $\bm{\mathcal{F}}_{ij}$ is nonvanishing, which is the case in most real situations. This simple physical picture clearly demonstrates the universality of equilibrium spin currents in matter. They should be generically present in almost any system, like molecules or solids, if the SO coupling is non-negligible. Interestingly, in semiconductors with linear SO coupling the spin current is related to the non-Abelian field by a Yang-Mills magnetostatics equation, which makes one more unexpected connection between the condensed matter and high energy physics. The present results also apply to ultracold atomic gases where a background non-Abelian field can be generated optically Osterloh2005 ; Ruseckas2005 . In general the many-body Hamiltonian with first spin-dependent relativistic corrections can be represented in a form FroStu1993 $\displaystyle H$ $\displaystyle=$ $\displaystyle\int d{\bf r}\Big{\\{}\frac{1}{2m}[(i\partial_{i}+\bm{\mathcal{A}}_{i})\Psi]^{\dagger}[(i\partial_{i}+\bm{\mathcal{A}}_{i})\Psi]-\Psi^{\dagger}\bm{\mathcal{A}}_{0}\Psi$ (1) $\displaystyle+$ $\displaystyle U\Psi^{\dagger}\Psi+\frac{1}{2}\int d{\bf r^{\prime}}V_{{\bf r}{\bf r^{\prime}}}\Psi^{\dagger}({\bf r})\Psi^{\dagger}({\bf r^{\prime}})\Psi({\bf r^{\prime}})\Psi({\bf r})\Big{\\}}$ where $\Psi^{\dagger}=(\psi^{\dagger}_{\uparrow},\psi^{\dagger}_{\downarrow})$ is a two-component fermionic field, $V_{{\bf r}{\bf r^{\prime}}}$ is the interparticle interaction, and $U$ is the external scalar potential note_A . The components $\bm{\mathcal{A}}_{\mu}$ ($\mu=0,x,y,z$) of the gauge field are 2$\times$2 matrices of the form $\bm{\mathcal{A}}_{\mu}={\mathcal{A}}_{\mu}^{a}\tau^{a}$, where $\tau^{a}=\sigma^{a}/2$ are the generators of SU(2) (spin-1/2 operators) with the following algebra $[\tau^{a},\tau^{b}]=i\varepsilon^{abc}\tau^{c}$, ${\rm tr}\\{\tau^{a},\tau^{b}\\}=\delta^{ab}$ note_notation . The non-Abelian potential $\bm{\mathcal{A}}_{\mu}$ captures all spin effects if one makes the following identifications $\bm{\mathcal{A}}_{0}=-\frac{e\hbar}{mc}B^{a}\tau^{a},\quad\bm{\mathcal{A}}_{i}=\frac{e\hbar}{mc^{2}}\varepsilon_{ija}E_{j}\tau^{a}$ (2) where $B^{a}$ are the components of the external magnetic field, and $E_{j}$ is the electric field produced, for example, by nuclei in molecules or solids. The Hamiltonian of Eq. (1) also covers popular 2D models of semiconductors with linear SO interaction of Rashba and/or Dresselhaus form Winkler . In this case the time component of SU(2) potential is still given by ${\mathcal{A}}_{0}^{a}=g\mu_{\rm B}B^{a}$, but with an appropriate $g$-factor, while the spatial components are defined as follows, $\bm{\mathcal{A}}_{z}=0$, $\bm{\mathcal{A}}_{x}=2m(\beta\tau^{x}-\alpha\tau^{y}),\quad\bm{\mathcal{A}}_{y}=2m(\alpha\tau^{x}-\beta\tau^{y}),$ (3) where $\alpha$ and $\beta$ are the Rashba and Dresselhaus SO coupling constants respectively. The beauty of the representation (1) is that the corresponding action $S[\Psi,\bm{\mathcal{A}}_{\mu}]=\int dt(d{\bf r}\Psi^{\dagger}i\partial_{t}\Psi-H)$ (4) is invariant with respect to local non-Abelian gauge transformations $\Psi\mapsto{\mathcal{U}}\Psi,\qquad\bm{\mathcal{A}}_{\mu}\mapsto{\mathcal{U}}\bm{\mathcal{A}}_{\mu}{\mathcal{U}}^{-1}-i(\partial_{\mu}{\mathcal{U}}){\mathcal{U}}^{-1}$ (5) where ${\mathcal{U}}=e^{i\theta^{a}({\bf r},t)\tau^{a}}$ is an arbitrary SU(2) matrix. The above gauge invariance immediately implies covariant conservation of a color current, ${\bm{J}}_{\mu}=J_{\mu}^{a}\tau^{a}$, with components $J_{\mu}^{a}=\delta S/\delta{\mathcal{A}}_{\mu}^{a}$: $D_{t}{\bm{J}}_{0}+D_{i}{\bm{J}}_{i}=0$ (6) where $D_{\mu}\cdot=\partial_{\mu}\cdot-i[\bm{\mathcal{A}}_{\mu},\cdot]$ is a covariant derivative, and $\displaystyle J_{0}^{a}$ $\displaystyle=$ $\displaystyle\frac{\delta S}{\delta{\mathcal{A}}_{0}^{a}}=\Psi^{\dagger}\tau^{a}\Psi\equiv s^{a}({\bf r},t)$ (7) $\displaystyle J_{i}^{a}=\frac{\delta S}{\delta{\mathcal{A}}_{i}^{a}}$ $\displaystyle=$ $\displaystyle\frac{-i}{2m}[\Psi^{\dagger}\tau^{a}\partial_{i}\Psi-(\partial_{i}\Psi^{\dagger})\tau^{a}\Psi]-\frac{{\mathcal{A}}_{i}^{a}}{4m}\hat{n}$ (8) Explicitly the covariant conservation law of Eq. (6) reads $\partial_{t}J_{0}^{a}+\varepsilon^{abc}{\mathcal{A}}_{0}^{b}J_{0}^{c}+\partial_{i}J_{i}^{a}+\varepsilon^{abc}{\mathcal{A}}_{i}^{b}J_{i}^{c}=0.$ Apparently the second and the fourth terms in this equation violate conservation of the spin $J_{0}^{a}=s^{a}$. The second term causes the spin precession in the U(1) magnetic field ${\mathcal{A}}_{0}$. The fourth term is the ”internal torque” due to SO interaction. However, the variational definition of the spin four-current, Eqs. (7) and (8), based on the gauge invariance, leaves no room for an ambiguity. As soon as we identify the zeroth component, $\delta S/\delta{\mathcal{A}}_{0}^{a}$, with the spin density, we are forced to accept that the spatial part, $\delta S/\delta{\mathcal{A}}_{j}^{a}$, is the spin current. $J_{i}^{a}$ is coupled to ${\mathcal{A}}_{i}^{a}$ in exactly the same fashion as $J_{0}^{a}$ is coupled to ${\mathcal{A}}_{0}^{a}$, which is absolutely analogous to the familiar case of the charge four-current coupled to U(1) gauge field. One can also show that $J_{i}^{a}$, Eq. (8), is a proper dissipative current conjugated to an effective SU(2) electric field ${\mathcal{F}}_{0i}^{a}$ note_dissipation . It is worth noting that Eq. (8) coincides with the ”natural” definition of the spin current Rashba2003 ; Rashba2007_Book ; EngRasHal2007_Book . Armed with the gauge invariant Hamiltonian and the variational definition of $J_{i}^{a}$ we are ready to approach the problem of equilibrium spin currents. SO interaction enters the Hamiltonian as an effective background non-Abelian field. If a magnetic part of this color field is nonzero one naturally expects an orbital response in a form of color diamagnetic currents. These currents, if exist, are given by the derivative of the energy (thermodynamic potential) $E[{\mathcal{A}}_{i}^{a}]=\langle H\rangle$ with respect to ${\mathcal{A}}_{i}^{a}$. Since the energy is gauge invariant it can depend on ${\mathcal{A}}_{i}^{a}$ only via invariants composed of the field strength $\bm{\mathcal{F}}_{ij}=\partial_{i}\bm{\mathcal{A}}_{j}-\partial_{j}\bm{\mathcal{A}}_{i}-i[\bm{\mathcal{A}}_{i},\bm{\mathcal{A}}_{j}]$ (9) A particular form of invariants is determined by the symmetry of a particular system. For the sake of clarity I consider explicitly the case of semiconductors with linear SO coupling of the Rashba-Dresselhaus type. I also assume that the external scalar potential and the usual magnetic field are zero, i. e. in Eq. (1) $U=0$ and $\bm{\mathcal{A}}_{0}=0$. Since in the absence of ${\mathcal{A}}_{i}^{a}$ the system is rotationally invariant, the first SO correction to the energy must be proportional to ${\rm tr}(\bm{\mathcal{F}}_{ij}\bm{\mathcal{F}}_{ij})$, i. e. $E_{\rm SO}=\frac{\lambda}{4}\int d{\bf r}{\mathcal{F}}_{ij}^{a}{\mathcal{F}}_{ij}^{a},$ (10) where $\lambda$ is a constant (on dimensional grounds $\lambda\sim p_{F}^{D-2}/m$, where $p_{F}$ is the Fermi momentum and $D>1$ is the dimension of space). Calculation of the current, $J_{i}^{a}=-\delta E_{\rm SO}/\delta{\mathcal{A}}_{i}^{a}$, yields $\bm{J}_{j}=\lambda D_{i}\bm{\mathcal{F}}_{ij}=\lambda\left(\partial_{i}\bm{\mathcal{F}}_{ij}-i[\bm{\mathcal{A}}_{i},\bm{\mathcal{F}}_{ij}]\right),$ (11) which is exactly of the form of Yang-Mills magnetostatics equation. Physically the result is very similar to the case of U(1) magnetic field: an external field produces diamagnetic currents aimed at compensating that field. There is, however, an essential difference in the spatial distribution of diamagnetic currents. In U(1) case the currents in the bulk vanish when the magnetic field approaches a constant. Only the integral defining the induced magnetic moment remains finite. In contrast, in the non-Abelian case bulk diamagnetic currents exist even for a constant in space field. The reason is the commutator in the right hand side of Eq. (11). A similar commutator in Eq. (9) gives a nonzero magnetic field even for a space-independent vector potential. Thus, in the case of a homogeneous field (space-independent SO coupling constants), the bulk spin current is given by $\bm{J}_{j}=-\lambda[\bm{\mathcal{A}}_{i},[\bm{\mathcal{A}}_{i},\bm{\mathcal{A}}_{j}]]$ (12) Equation (12) shows that in the homogeneous system the spin current is proportional to the third power of non-Abelian potential. This naturally explains why Rashba’s equilibrium spin current is proportional to $\alpha^{3}$ Rashba2003 . The above phenomenology can be confirmed by direct microscopic calculations of the spin current for an exactly solvable model of noninteracting particles. First I consider homogeneous $\bm{\mathcal{A}}_{i}$. In this case the spin current, Eq. (8), is given by the expression $J_{i}^{a}=T\sum_{w}\sum_{\bf p}{\rm tr}\left[\frac{p_{i}}{m}\tau^{a}\hat{G}(\omega,{\bf p})\right]-\frac{n}{4m}{\mathcal{A}}_{i}^{a}$ (13) where $\omega$ is the fermionic Matsubara frequency, and $n$ is the density of particles. The one particle Green’s function $\hat{G}(\omega,{\bf p})$ is defined as follows $\displaystyle\hat{G}(\omega,{\bf p})=\left[i\omega+\mu-\frac{1}{2m}(p_{j}-{\mathcal{A}}_{j}^{b}\tau^{b})^{2}\right]^{-1}$ (14) $\displaystyle=$ $\displaystyle\frac{\hat{Z}_{+}({\bf p})}{i\omega+\mu- E_{+}({\bf p})}+\frac{\hat{Z}_{-}({\bf p})}{i\omega+\mu-E_{-}({\bf p})},$ where $\displaystyle\hat{Z}_{\pm}({\bf p})$ $\displaystyle=$ $\displaystyle\frac{1}{2}\left(1\mp\frac{2p_{i}{\mathcal{A}}_{i}^{a}\tau^{a}}{\sqrt{(p_{j}{\mathcal{A}}_{j}^{b})(p_{k}{\mathcal{A}}_{k}^{b})}}\right),$ (15) $\displaystyle E_{\pm}({\bf p})$ $\displaystyle=$ $\displaystyle\frac{{\bf p}^{2}}{2m}+\frac{{\mathcal{A}}_{i}^{a}{\mathcal{A}}_{i}^{a}}{8m}\pm\frac{1}{2m}\sqrt{(p_{j}{\mathcal{A}}_{j}^{b})(p_{k}{\mathcal{A}}_{k}^{b})}$ (16) After summation over $\omega$ Eq. (13) takes the form $\displaystyle J_{i}^{a}$ $\displaystyle=$ $\displaystyle\frac{1}{2m}\sum_{\bf p}\frac{p_{i}(p_{j}{\mathcal{A}}_{j}^{a})}{\sqrt{(p_{l}{\mathcal{A}}_{l}^{b})(p_{k}{\mathcal{A}}_{k}^{b})}}[n_{F}(E_{-})-n_{F}(E_{+})]$ (17) $\displaystyle-$ $\displaystyle\frac{{\mathcal{A}}_{i}^{a}}{2m}\sum_{\bf p}[n_{F}(E_{-})+n_{F}(E_{+})],$ where $n_{F}(E)$ is the Fermi distribution function. In the second term in Eq. (17) the density of particles is represented as $n=\sum_{\bf p}[n_{F}(E_{-})+n_{F}(E_{+})]$. The rest of calculations is straightforward. Assuming as usual that SO coupling is weak, we expand the distribution functions $n_{F}(E{\pm})$ in terms of $\sqrt{(p_{j}{\mathcal{A}}_{j}^{b})(p_{k}{\mathcal{A}}_{k}^{b})}$, and keep the first nonvanishing term in Eq. (17). This term is, as expected, of the third order in ${\mathcal{A}}_{j}^{a}$. The final result at zero temperature is the following $J_{i}^{a}=\frac{N_{F}}{24m^{2}}({\mathcal{A}}_{j}^{a}{\mathcal{A}}_{i}^{b}{\mathcal{A}}_{j}^{b}-{\mathcal{A}}_{i}^{a}{\mathcal{A}}_{j}^{b}{\mathcal{A}}_{j}^{b}),$ (18) where $N_{F}$ is the density of states at the Fermi level. It is easy to see that the expression in the brackets in Eq. (18) is exactly the double commutator entering the right hand side of Eq. (12). Hence the phenomenological coefficient in the SO energy is $\lambda=N_{F}/24m^{2}$. Thus the direct calculations indeed confirm a diamagnetic nature of the equilibrium spin currents. The spin current in the 2D Rashba-Dresselhaus model is obtained by inserting ${\mathcal{A}}_{i}^{a}$ of Eq. (3) and $N_{F}=m/2\pi$ into Eq. (18) $\displaystyle J_{x}^{y}=-J_{y}^{x}$ $\displaystyle=$ $\displaystyle\frac{m^{2}}{6\pi}\alpha(\alpha^{2}-\beta^{2}),$ (19) $\displaystyle J_{x}^{x}=-J_{y}^{y}$ $\displaystyle=$ $\displaystyle\frac{m^{2}}{6\pi}\beta(\alpha^{2}-\beta^{2}).$ (20) Setting $\beta=0$ we exactly recover the result by Rashba Rashba2003 . It is very interesting to realize that the formula for the spin current obtained in Ref. Rashba2003, is a hidden form of a covariant curl of the non-Abelian magnetic field! The spin current of Eqs. (19), (20) is zero at $\alpha=\pm\beta$. The reason is that the color magnetic field ${\mathcal{F}}_{xy}^{z}=4m^{2}(\alpha^{2}-\beta^{2})$ vanishes at these special values of SO constants. In the absence of magnetic field there are no diamagnetic currents. Now I will show that the gauge invariance allows to significantly simplify practical calculations of the spin current using powerful techniques of the linear response theory, in spite of an obvious nonlinearity in ${\mathcal{A}}_{i}^{a}$. Let us first assume that the non-Abelian potential ${\mathcal{A}}_{i}^{a}({\bf r})$ is weak (here we allow for a general inhomogeneity of the potential). The standard linear response theory GiulianiVignale yields $J_{i}^{a}({\bf r})=\int d{\bf r}\chi_{ij}^{ab}({\bf r},{\bf r^{\prime}})\bm{\mathcal{A}}_{j}^{b}({\bf r^{\prime}}),$ (21) where the response function is defined by the Kubo formula $\chi_{ij}^{ab}({\bf r},{\bf r^{\prime}})=\langle\langle\hat{J}_{i}^{a}({\bf r});\hat{J}_{j}^{b}({\bf r^{\prime}})\rangle\rangle-\delta({\bf r}-{\bf r^{\prime}})\delta_{ij}\delta^{ab}\frac{n}{4m}.$ (22) The operator $\hat{J}_{i}^{a}$ in Eq. (22) is given by the first term in the right hand side of Eq. (8). Due to the gauge invariance the vector potential ${\mathcal{A}}_{i}^{a}$ can enter Eq. (21) only via the linearized field strength, $\partial_{i}{\mathcal{A}}_{j}^{a}-\partial_{j}{\mathcal{A}}_{i}^{a}$, and, possibly, its spatial derivatives, $\partial_{i}$. By the same token, to recover the formula valid to the first order in the full nonlinear field strength, ${\mathcal{F}}_{ij}^{a}$ of Eq. (11), we simply have to make the following replacements $(\partial_{i}{\mathcal{A}}_{j}^{a}-\partial_{j}{\mathcal{A}}_{i}^{a})\mapsto{\mathcal{F}}_{ij}^{a},\qquad\partial_{i}\mapsto D_{i}.$ (23) All other changes are forbidden by the gauge invariance. To demonstrate how this prescription works in practice I consider again the example of a homogeneous (in the absence of ${\mathcal{A}}_{i}^{a}({\bf r})$) and noninteracting electron gas with linear SO interaction. In this case Eq. (22) for the spin current response function in the momentum representation simplifies as follows $\chi_{ij}^{ab}({\bf q})=\delta^{ab}\chi_{ij}^{S}({\bf q})$, $\chi_{ij}^{S}({\bf q})=\sum_{\bf p}p_{i}p_{j}\frac{n_{F}(\xi_{\bf{p}-\frac{\bf q}{2}})-n_{F}(\xi_{\bf{p}+\frac{\bf q}{2}})}{2m^{2}(\xi_{\bf{p}+\frac{\bf q}{2}}-\xi_{\bf{p}-\frac{\bf q}{2}})}-\delta_{ij}\frac{n}{4m}$ (24) where $\xi_{\bf p}=\frac{{\bf p}^{2}}{2m}$. Equation (24) coincides, up to a numerical factor, with a static charge current response function $\chi_{ij}({\bf q})$ GiulianiVignale . It the standard theory the function $\chi_{ij}({\bf q})$ determines the Landau diamagnetic response to an external magnetic field, which provides us with another confirmation of the diamagnetic nature of the equilibrium spin currents. To calculate the spin current response function we follow the usual route GiulianiVignale . In the limit of $q\ll p_{F}$, which means that SO constants are slowly changing on the scale of $p_{F}^{-1}$, $\chi_{ij}^{ab}({\bf q})$ takes the form $\chi_{ij}^{ab}({\bf q})=\frac{N_{F}}{24m^{2}}\delta^{ab}(q_{i}q_{j}-q^{2}\delta_{ij}).$ Transforming this equation from ${\bf q}$\- to ${\bf r}$-space, and inserting the result into Eq. (21) we get for the spin current $J_{j}^{a}=\frac{N_{F}}{24m^{2}}\partial_{i}(\partial_{i}{\mathcal{A}}_{j}^{a}-\partial_{j}{\mathcal{A}}_{i}^{a})$ (25) The final step is a substitution of Eq. (23), which yields the final gauge covariant expression valid to the first order in the total non-Abelian magnetic field $\bm{\mathcal{F}}_{ij}$, Eq. (9), $J_{j}^{a}=\frac{N_{F}}{24m^{2}}D_{i}{\mathcal{F}}_{ij}^{a}.$ Thus the standard linear response theory supplemented with the substitution of Eq. (23) exactly recovers the Yang-Mills form of the spin current, Eq. (11), with the correct coefficient $\lambda=N_{F}/24m^{2}$. An obvious advantage of this way is that it allows to straightforwardly include the effects of interaction and external inhomogeneities (for example, impurity scattering). The formalism is also easily transferable to nonequilibrium situations. Apparently the general conclusion about the nature of equilibrium spin currents is not restricted to the simple Rashba-Dresselhaus model. Such diamagnetic currents necessarily present in any system with SO interaction, provided the effective color magnetic field of Eq. (9) is nonzero. As vanishing $\bm{\mathcal{F}}_{ij}$ is actually an exception (like in the case $\alpha=\pm\beta$) it should be possible to find the spin currents in many molecules and solids using available codes for ab initio electronic structure calculations. In conclusion I identified the equilibrium spin currents with diamagnetic color currents in the presence of a non-Abelian field generated by SO coupling. If the particles have a color charge coupled to physical Yang-Mills fields, like in quark-gluon plasma, the color currents would produce a back reaction field to compensate the external one. The absence of such back reaction does not make the equilibrium spin currents less ”real”. They are as real as pseudo-diamagnetic currents in a rotating reference frame. One of the main outcomes of this work is a conclusion about universality of equilibrium spin currents that should exist in most real systems. The universality makes it especially intriguing to observe and, possibly, to control such currents experimentally. In this respect a proposal by Sonin Sonin2007b , connecting spin currents to mechanical deformations, looks especially interesting. The present results show that it is not necessarily to do experiments with a ”Rashba medium”. Any technologically convenient material with strong SO interaction should demonstrate the same effect. Possibly the color diamagnetic currents can be also observed in trapped atomic gases subjected to a proper configuration of artificial non-Abelian fields Osterloh2005 ; Ruseckas2005 . I am grateful to G. Vignale for valuable comments. This work was supported by the Ikerbasque Foundation. ## References * (1) I. Žutić, J. Fabian, and S. Das Sarma, Rev. Mod. Phys. 76, 323 (2004). * (2) E. I. Rashba, in Future Trends in Microelectronics. Up the Nano Creek, edited by S. Luryi, J. M. Xu, and A. Zaslavsky (Wiley, Hoboken, 2007); arXiv:cond-mat/0611194. * (3) J. Shi, P. Zhang, D. Xiao, and Q. Niu, Phys. Rev. Lett. 96, 076604 (2006). * (4) H.-A. Engel, E. .I. Rashba, and B. I. Halperin, in Handbook of Magnetism and Advanced Magnetic Materials, edited by H. Kronmüller and S. Parkin (Wiley, New York, 2007), Vol. 5; arXiv:cond-mat/0603306. * (5) E. I. Rashba, Phys. Rev. B 68, 241315(R) (2003). * (6) E. B. Sonin, Phys. Rev. B 76, 033306 (2007a). * (7) E. B. Sonin, Phys. Rev. Lett. 99, 266602 (2007b). * (8) J. Fröhlich and U. M. Studer, Rev. Mod. Phys. 65, 733 (1993). * (9) A. Rebei and O. Heinonen, Phys. Rev. B 73, 153306 (2006); P.-Q. Jin, Y.-Q. Li, and F.-C. Zhang, J. Phys. A: Math. Gen. 39, 7115 (2006); P.-Q. Jin and Y.-Q. Li, Phys. Rev. B 74, 085315 (2006); B. A. Bernevig, J. Orenstein, and S.-C. Zhang, Phys. Rev. Lett. 97, 236601 (2006); N. Hatano, R. Shirasaki, and H. Nakamura, Phys. Rev. A 75, 032107 (2007); B. W. A. Leurs, Z. Nazario, D. I. Santiago, and J. Zaanen, arXiv:0705.2953 (2007). * (10) For example, it remains unnoticed that the vanishing static spin Hall conductivity in Rashba systems is a direct consequence of the SU(2) gauge invariance. * (11) K. Osterloh, M. Baig, L. Santos, P. Zoller, and M. Lewenstein, Phys. Rev. Lett. 95, 010403 (2005). * (12) J. Ruseckas, G. Juzeliunas, P. Ohberg, and M. Fleischhauer, Phys. Rev. Lett. 95, 010404 (2005). * (13) The usual U(1) external vector potential $A_{i}$ is not shown in Eq. (1). If needed, it is added additively to $\bm{\mathcal{A}}_{i}$. * (14) The Greek subscripts $\mu,\nu\dots$ label space-time coordinates, while the Latin subscripts $i,j,k,\dots$, and superscripts $a,b,c,\dots$ are used to label spatial coordinates and the directions in the internal spin space, respectively. The summation over repeated indexes is assumed throughout. * (15) R. Winkler, _Spin-Orbit Coupling Effects in Two-Dimensional Electron and Hole Systems_ (Springer, Berlin, 2003). * (16) Using Eqs. (1) and (6) we find $\frac{d}{dt}\langle H\rangle=\int d{\bf r}J_{i}^{a}{\mathcal{F}}_{i0}^{a}$, where $\bm{\mathcal{F}}_{i0}=\partial_{i}\bm{\mathcal{A}}_{0}-\partial_{t}\bm{\mathcal{A}}_{i}-i[\bm{\mathcal{A}}_{i},\bm{\mathcal{A}}_{0}]$. * (17) G. F. Giuliani and G. Vignale, _Quantum Theory of the Electron Liquid_ (Cambridge University Press, Cambrige, 2005).

arxiv-papers

2024-09-04T02:48:53.670218

0802.1411

# Zero-field and Larmor spinor precessions in a neutron polarimeter experiment S. Sponara J. Kleppa G. Badureka Y. Hasegawaa,b Hasegawa@ati.ac.at aAtominstitut der Österreichischen Universitäten, Stadionallee 2, A-1020 Vienna, Austria bPRESTO, Japan Science and Technology Agency (JST), Kawaguchi, Saitama 332-0012, Japan Corresponding author Phone +43-1-58801-14190; Fax +43-1-58801-14199. ###### Abstract We present a neutron polarimetric experiment where two kinds of spinor precessions are observed: one is induced by different total energy of neutrons (zero-field precession) and the other is induced by a stationary guide field (Larmor precession). A characteristic of the former is the dependence of the energy-difference, which is in practice tuned by the frequency of the interacting oscillating magnetic field $\omega_{\textrm{\scriptsize{R}}}$. In contrast the latter completely depends on the strength of the guide field, namely Larmor frequency $\omega_{\textrm{\scriptsize{L}}}$. Our neutron- polarimetric experiment exhibits individual tuning as well as specific properties of each spinor precession, which assures the use of both spin precessions for multi-entangled spinor manipulation. ###### keywords: neutron , spin precession , Larmor precession , zero-field precession , polarimeter , , , and In the last years great effort has been devoted to the development of quantum information and communication technology [1]. Among them creation and manipulation of entanglement is a key issue and experimental realizations with the use of photons, atoms, ions, nuclear-magnet, superconducting qubits etc. have been reported [2]. There, the entanglement between spatially separated systems are mainly used to exploit non-local property of quantum systems. In contrast, investigations of two-qubit entanglement with neutrons, one of the most useful quantum system to be utilized for studies of quantum mechanical phenomena [3], have been performed with the use of their degrees of freedom, namely spin and path, in a single particle: a violation of a Bell-like inequality [4], contextual nature of quantum theory [5], and a full tomographic analysis of Bell-states [6] were carried out. For further development towards multi-entanglement, it is inevitable to establish manipulation of other degrees of freedom. It is know that the total energy of neutrons can be manipulated with the use of interactions between a neutron’s magnetic moment and a time-dependent oscillating magnetic field [7, 8, 9]. Recently, we have accomplished an experiment to exhibit a coherent energy manipulation of neutrons with the use of radio-frequency (RF) oscillating magnetic field [10], enabling neutrons to be applied for investigations of multi-entanglement. Thus, we find it significant to demonstrate explicitly a phase manipulation for different energy eigenstates which will be utilized in forthcoming experiments such as a demonstration of triple-entanglment with a GHZ-state [11]. In the neutron scattering community, a spin precession, alternatively a phase shift, due to energy difference is known as zero-field precession [12, 13] and utilized for zero-field spin-echo spectroscopy [14, 15, 16, 17]. When a neutron is exposed to a stationary magnetic field, the motion of its spin vector is thoroughly described by the Bloch-equation exhibiting Larmor precession: $\frac{d\vec{S}}{dt}=\gamma\vec{S}\times\vec{B}$ (1) where $\gamma$ is the gyromagnetic ratio given by $\mu/\lvert S\lvert$, $\mu$ and ${\vec{S}}$ are magnetic moment and the neutron’s spin, respectively. This is the equation of motion of a classical magnetic dipole in a magnetic field, which shows the precession of the spin vector ${\vec{S}}$ about the magnetic field $\vec{B}$ with the Larmor frequency $\omega_{\textrm{\scriptsize{L}}}=\left|{{{2\mu B}\mathord{\left/{\vphantom{{2\mu B}\hbar}}\right.\kern-1.2pt}\hbar}}\right|$. It is worth noting here that the Larmor precession angle (rotation angle) is obtained with a frequency $\omega_{\textrm{\scriptsize{L}}}$ and the propagation time T as $\omega_{\textrm{\scriptsize{L}}}\rm{T}$, solely depends on the strength of the applied magnetic field. In practical experiments, the Larmor precession is utilized by a variety of DC spin rotators. In contrast, when a neutron interacts with a time-dependent, rather oscillating, magnetic field, photon exchange occurs and the total energy of neutrons is shifted. In particular when a so-called resonant spin-flipper is activated, the neutron emits (or absorbs) a photon of energy $\hbar\omega_{\textrm{\scriptsize{R}}}$, thus the total energy of the neutron decreases (or increases) by $\hbar\omega_{\textrm{\scriptsize{R}}}$ during the interaction [9]. As a consequence, an additional phase, a zero-field phase, $\omega_{\textrm{\scriptsize{R}}}\rm{T}$ is accumulated during the propagation afterwards. This phase shift, alternatively a spin precession, due to energy shift emerges even without a guide field, thus it is referred to a zero-field precession in literature [12, 13]. It should be emphasized that this zero- field phase is independent of the applied guide field, namely the Larmor phase $\omega_{\textrm{\scriptsize{L}}}\rm{T}$, and purely depends on the frequency of the spin-flipper. In the following neutron polarimeter experiments, we demonstrate how one can tune a zero-field phase $\omega_{\textrm{\scriptsize{R}}}\rm{T}$ and a Larmor phase $\omega_{\textrm{\scriptsize{L}}}\rm{T}$ independently. Figure 1: Experimental setup for demonstrations of zero-field and Larmor spin precessions. From a polarized incident beam the first $\pi/2$ spin-rotator generates a superposition of $\lvert\pm z\rangle$. The first RF flipper induces the energy shift $\hbar\omega_{R}$ for each flipped spin eigenstate $\lvert\mp z\rangle$, which leads to the zero-field spin precession until the second RF flipper where the energy difference is compensated. The guide field $B_{0}$, permeated along the beam path, leads to the Larmor spin precession. The second $\pi/2$ spin-rotator together with the analyzer enables to exhibit the intensity modulation due to the spin precessions. The translation of the DC and the RF flippers manipulate individually the two phases when the RF flipper is turned on and off. The experiments were carried out at the neutron polarimetry facility at the 250-kW TRIGA research reactor of the Atomic Institute of the Austrian Universities in Vienna [18, 19, 20, 21]. A schematic view of the experimental setup is shown in Fig.1. The incident neutron beam is monochromatized (with a mean wave length of $\lambda=1.99\mbox{ \AA}$ which corresponds to a velocity of 1990 m/s) by the use of a pyrolitic graphite crystal and polarized (average degree of polarization over $98\%$) by reflection from a bent Co-Ti supermirror array. The diameter of the beam is confined to about 4mm in diameter by a Cd diaphragm. The initial polarization vector $\vec{P}_{\textrm{i}}$ is perpendicular to the beam trajectory and defines the $+\hat{\mathbf{z}}$-direction. Another suppermirror array is used to analyze the final polarization $\vec{P}_{\textrm{f}}$. Depolarization of the neutron beam is minimized by applying a guide field along the beam trajectory wherever necessary: this guide field $B_{0}$ determines the Larmor frequency $\omega_{\textrm{\scriptsize{L}}}$. In a neutron polarimeter setup described above, after going through a $\pi/2$ spin-rotator, an incident polarization is interpreted as superposition of two orthogonal spin states $\lvert\pm z\rangle$ and different phase shifts of these two states under further spinor manipulations result in a polarization change of the emerging neutron beam, i.e., we take a neutron polarization- interference scheme. Between the two $\pi/2$ spin-rotators, three spin- flippers are placed in the beam: the first two are RF spin-flippers followed by a DC spin-flipper. In addition, one of the RF and the DC flippers are mounted together on a single translator, allowing to tune the propagation time of neutrons between the RF spin-flippers. The polarization vector after passing through all spin-flippers is expected to lie in the xy-plane, resulting from the interference between the states $\lvert\pm z\rangle$. A spin analyzer together with the second $\pi/2$ spin-rotator is used to resolve the different phase shifts accumulated through spin-flips. Figure 2: (a) Typical intensity oscillations with least square fits for the Larmor spin precession. Curves are shown with the guide field $B_{0}=$10.90, 10.79, 10.69, 10.58G.(b) Dependence of the period on the strength of the guide field depicted for seven values of the guide field. A clear linear dependence agrees well with the theoretical predictions (see, Eq. (2)). The first experiment shows a pure Larmor precession: both RF flippers are turned off and only the DC flipper is in operation. In this case, the superposed states $\lvert\pm z\rangle$ (and the flipped states $|\mp z\rangle$ later) simply propagate within the guide field $B_{0}$. These states $|+z\rangle$ and $|-z\rangle$ are the eigenstates of the guide field: no additional spinor rotation occurs. Then, each state merely obtains a Larmor phase $\alpha$ due to the guide field, which is given by $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle{\omega}_{\textrm{\scriptsize{L}}}(\rm{T_{1}}-\rm{T_{2}})$ (2) $\displaystyle=$ $\displaystyle{\omega}_{\textrm{\scriptsize{L}}}((\rm{T_{1}^{0}}+\Delta T)-(\rm{T_{2}^{0}}-\Delta T))$ $\displaystyle=$ $\displaystyle\alpha_{0}+2{\omega}_{\textrm{\scriptsize{L}}}\Delta\rm{T},$ with $\alpha_{0}=\omega_{\textrm{\scriptsize{L}}}(\rm{T_{1}^{0}}-\rm{T_{2}^{0}}+\rm{T_{3}^{0}})$. $\rm{T_{1}}$, $\rm{T_{2}}$, $\rm{T_{j}^{0}}$ and $\Delta\rm{T}$ are the propagation time before and after the DC flipper, the propagation time at the initial position, and the time shift for $\rm{T_{1}}$, $\rm{T_{2}}$ by shifting the position of the DC flipper. After the beam passes the $\pi/2$ spin-rotator and the analyzer, clear sinusoidal intensity oscillations of high contrast were obtained. Typical experimental data with least square fits are shown in Fig.2(a): the strength of the guide field was varied at $B_{0}=$10.90, 10.79, 10.69, 10.58G. The shift of the oscillations is due to $\omega_{\textrm{\scriptsize{L}}}\rm{T}$ (see, $\alpha_{0}$ in Eq.(2)). In addition, the dependence of the period of the oscillations on the strength of the guide field is plotted in Fig.2(b). A linear dependence, of which inclination is in good agreement with the theoretically predicted value, is seen, which confirms the precession angle given by Eq.(2). Figure 3: (a) Typical intensity oscillations with least square fits for the zero-field spin precession. Curves are shown at the RF flipper frequencies $\omega_{R}=$ 30.9, 30.6, 30.3, 30.0 kHz. (b) Dependence of the period on the frequency depicted for seven values of the frequency. A clear linear dependence agrees well with the theoretical predictions (see, Eq. (3)). Next, characteristics of the zero-field precession are considered: both RF flippers are turned on. In this case, the superposed states $\lvert\pm z\rangle$ is once flipped to $\lvert\mp z\rangle$, and finally again flipped back to $\lvert\pm z\rangle$. Then, the spin precession angle $\alpha^{\prime}$ is expected to be given by the propagation time, $\rm{T_{1}}$, $\rm{T_{2}}$, and $\rm{T_{3}}$ after each spin flipper by $\displaystyle\alpha^{\prime}$ $\displaystyle=$ $\displaystyle{\omega}_{\textrm{\scriptsize{L}}}(\rm{T_{1}}-\rm{T_{2}}+\rm{T_{3}})+{\omega}_{\textrm{\scriptsize{R}}}\rm{T_{1}}$ (3) $\displaystyle=$ $\displaystyle{\omega}_{\textrm{\scriptsize{L}}}\\{(\rm{T_{1}^{0}}+\Delta T)-\rm{T_{2}^{0}}+(\rm{T_{3}^{0}}-\Delta T)\\}+{\omega}_{\textrm{\scriptsize{R}}}(\rm{T_{1}^{0}}+\Delta T)$ $\displaystyle=$ $\displaystyle\alpha^{\prime}_{0}+{\omega}_{\textrm{\scriptsize{R}}}\Delta\rm{T},$ with $\alpha^{\prime}_{0}=\omega_{\textrm{\scriptsize{L}}}(\rm{T_{1}^{0}}-\rm{T_{2}^{0}}+\rm{T_{3}^{0}})+\omega_{\textrm{\scriptsize{R}}}\rm{T_{1}^{0}}$. In this configuration, no Larmor precession is expected to be induced by the change of $\Delta\rm{T}$, since a positive and a negative change of $\rm{T_{1}}$ and $\rm{T_{3}}$ completely compensate each other by shifting the position of the DC flipper and the second RF flipper. In order to prove the frequency dependence of the precession, the frequency of the RF flippers were varied with keeping the strength of the guide field by $B_{0}=$10.59G. Typical intensity modulations are shown in Fig.3(a) at the RF flipper frequencies of $\omega_{\textrm{\scriptsize{R}}}$=30.9, 30.6, 30.3, and 30.0kHz. Clear sinusoidal oscillations of high contrast were again obtained. A slight reduction of the amplitude of the obtained oscillations is solely due to the detuning of the frequency $\omega_{\textrm{\scriptsize{R}}}$ in the vicinity of the resonance for the flip-mode. And the shift of the oscillations arises from $\omega_{\textrm{\scriptsize{R}}}\rm{T}$ (see, $\alpha^{\prime}_{0}$ in eq.(3)). In addition, a dependence of the period on the frequency is plotted in Fig.3(b). A linear dependence is seen and its inclination is in good agreement with the theory, which confirms the precession angle given by Eq.(3). Figure 4: (a) Typical intensity oscillations with least square fits for the zero-field precession. Curves are shown with the guide field $B_{0}=$10.90, 10.79, 10.69, 10.58G. All curves are identical as expected. (b) The period is plotted versus the strength of the guide field for seven values. It is constant and independent of the strength of the guide field. Note that the setup is constructed in a way that no spin rotation due to the Larmor precession will occur, since the Larmor precessions before and after the second RF flipper, i.e., $\Delta\rm{T_{1}}$ and $\Delta\rm{T_{3}}$, are compensated each other with all flippers turned on (see Eq.(3)). This independence of the Larmor precession can also be accessed in our experimental setup. In particular, the frequency of the RF flippers were tuned at the resonance $\omega_{0}=$31.8kHz and the strength of the guide field was varied at $B_{0}=$10.90, 10.79, 10.69, 10.58G. Typical experimental data with least square fits are shown Fig.4(a): identical sinusoidal oscillations are obtained, confirming no spin rotation due to Larmor precessions. In addition, the period is plotted versus the strength of the guide field in Fig.4(b). Independent behavior of the period from the field strength is seen as expected by Eq.(3). The results are in good agreement with theoretically predicted behavior: the frequency of the oscillating fields affects only on the zero-field precession while the strength of the field does on the Larmor precession. The tactics, that the RF and the DC flippers are displaced by the same amount, allows individual control of both precessions. This is a great advantage for the separate (phase) manipulation of the two-spaces, i.e., spin and energy spaces, in future neutron optical experiments. We are now proceeding further neutron interferometric and polarimetric experiments, where multi-entanglement in a single neutron system will be investigated, in particular, with the use of entanglement between degrees of freedoms such as path, spin, and energy. In summary, we have presented experiments with a neutron polarimeter where two spin-precessions are manipulated to show their characteristics. The zero-field precession is independent of the strength of the guide field and purely depends on the frequency of the RF spin-flippers, where an energy shift occurs. In contrast, the Larmor precession solely depends on the strength of the applied magnetic field. In addition, we have exhibited a method to manipulate individually the two phases, resulting from each spin precession. The method used here will be utilized in further neutron optical experiments dealing with multi-entanglement in a single particle system. This work has been partly supported by the Austrian Science Foundation, FWF (P17803-N02). Y.H. would like to thank the Japan Science and Technology Agency (JST) for financial support. ## References * [1] M.A. Nielsen and I. Chuang, _Quantum Computation and Quantum Information_ , (Cambridge Unviversity Press, Cambridge, 2000). * [2] S. Haroche and J. Raimond, _Exploring the Quantum_ , (Oxford Unviversity Press, Oxford, 2006). * [3] H. Rauch and S.A. Werner, _Neutron Interferometry_ , (Clarendon Press, Oxford, 2000). * [4] Y. Hasegawa, R. Loidl, G. Badurek, M. Baron, and H. Rauch, Nature 425, 45 (2003). * [5] Y. Hasegawa, R. Loidl, G. Badurek, M. Baron, and H. Rauch, Phys. Rev. Lett. 97, 230401 (2006). * [6] Y. Hasegawa, R. Loidl, G. Badurek, S. Filipp, J. Klepp, and H. Rauch, Phys. Rev. A 76, 052108 (2007). * [7] B.Alefeld, G. Badurek, and H. Rauch, Z. Phys B41 231 (1981). * [8] G. Badurek, H. Rauch, and J. Summhammer, Phys. Rev. Lett. 51, 1015 (1983). * [9] J. Summhammer, Phys. Rev. A 47, 56 (1993). * [10] S. Sponar, J. Klepp, R. Loidl, S. Filipp, G. Badurek, Y. Hasegawa, and H. Rauch, submitted to Phys. Rev. Lett. * [11] D.M. Greenberger, M.A. Horne, and A. Zeilinger, in _Bell‘s Theorem, Quantum Theory, and Concepts of the Universe_ , edited by M.Kafatos (Kluwer Academics, Dordrecht, The Netherlands, 1989). * [12] R. Golub, R. Gähler, and T. Keller, Am. J. Phys. 62 779 (1994). * [13] W.H. Kraan, S.V. Grigoriev, R. kreuger, F.M. Mulder and M.Th. Rekveldt, Physica B297, 23 (2001). * [14] R. Gähler, and R. Golub, Z. Phys. B65 269 (1987). * [15] R. Gähler, and R. Golub, Phys. Lett. A 123 43 (1987). * [16] D. Yamazaki, Nucl. instr. meth. A 488, 623 (2002). * [17] S.V. Grigoriev, W.H. Kraan, and M.Th. Rekveldt, Phys. Rev. A 69, 043615 (2004). * [18] Y. Hasegawa, S. Menhart, R. Meixner, and G. Badurek, Phys. Lett. A 234, 322 (1997) * [19] Y. Hasegawa, and G. Badurek, Phys. Rev. A 59, 4614 (1999) * [20] A.G. Wagh, G. Badurek, V.C. Rakhecha, R.J. Buchelt, A. Schricker, Phys. Lett. A 268, 209 (1997) * [21] J. Klepp, S. Sponar, Y. Hasegawa, E. Jericha,and G. Badurek, Phys. Lett. A 342 48 (2005).

arxiv-papers

2024-09-04T02:48:53.676163

0802.1486

# Neural network predictions for $Z^{\prime}$ boson within LEP2 data set of Bhabha process A.N.Buryk, V.V.Skalozub Dnipropetrovs’k National University Dnipropetrovs’k, 49025 Ukraine (01 May, 2009) ###### Abstract The neural network approach is applied to search for the $Z^{\prime}$-boson within the LEP2 data set for $e^{+}e^{-}\to e^{+}e^{-}$ scattering process. In the course of the analysis, the data set is reduced by 20 percent. The axial- vector $a_{e}^{2}$ and vector $v_{e}^{2}$ couplings of the $Z^{\prime}$ are estimated at 95% CL within a two-parameter fit. The mass is determined to be $0.53\leq m_{Z^{\prime}}\leq 1.05$ TeV. Comparisons with other results are given. ## 1 Introduction A lot of extended models includes a so-called $Z^{\prime}$ gauge boson - massive neutral vector particle associated with an extra $U(1)$ subgroup of underlying group [3]. In particular, this particle is predicted by numerous grand unification models. In all the models, the Abelian $Z^{\prime}$-boson is described by a low-energy $\tilde{U}(1)$ gauge subgroup originated in some symmetry breaking pattern. Searching for $Z^{\prime}$ is one of the goals of future experiments at LHC [4], and current ones at Tevatron. It can manifest itself either as a real or a virtual state dependently on the value of its mass. With this goals keeping in mind, it is reasonable to take into consideration all accessible nowadays information about $Z^{\prime}$-boson, following from different experiments established at low energies. In particular, in LEP searching for $Z^{\prime}$ it was established a model dependent approach and low bounds on its mass $m_{Z^{\prime}}>400$ GeV and/or $m_{Z^{\prime}}>800$ GeV have been obtained [9], which are dependent on a specific model. No actual signals of the $Z^{\prime}$ have been detected. In recent experiments at Tevatron the derived low bounds are a little bit larger and correspond to the masses $m_{Z^{\prime}}>850$ GeV [4]. On the other hand, in series of papers [1][2] a model independent method of searching for the $Z^{\prime}$ was developed. This approach accounts for a renormalizability of an underlying unknown in other respect theory. The requirement of renormalizability results in a series of relations between low energy parameters describing interactions of the $Z^{\prime}$ with fermions of the standard model. This reduces the number of low-energy parameters which must be fitted in experiments. In this way the couplings of $Z^{\prime}$ to the fermions of the standard model and the mass $m_{Z^{\prime}}$ were estimated at $1-2~{}\sigma$ CL in the one-parameter [1] and many-parameter [2] fits. If these relations are not taken into consideration, no signals (hints, in fact) follow. It was also concluded that the statistics of the LEP experiments is not too large to detect the $Z^{\prime}$ as the virtual state with enough high precision. So, some further analysis is reasonable. One needs in the estimates which could be used in future experiments. To determine them in a maximally full way we address to the analysis based on the predictions of the neural networks (for applications in high energy physics see, for example, [7]). The main idea of this approach is to constrain a given data set in such a way that an amount of the data is considered as an inessential background and omitted. The remaining data are expected to give a more precise fit of the parameters of interest. This is the goal of the present investigation. We treat the full data set on the Bhabha scattering process obtained in LEP2 experiments by using the neural network method in order to determine the couplings to the fermions and the mass of the $Z^{\prime}$ boson. ## 2 Neural network(NN) predictions The lack of statistics in LEP experiments does not allow the determination of the $Z^{\prime}$-boson mass with CL more than $2\sigma$ in one-parameter fit [1] and more than $1\sigma$ in two-parameter fit [2]. We propose two points to overcome this restriction for the case of the Bhabha scattering process when a many parameter fit is applied. First, an increase in parametric space could be compensated by an increase in the data set, if all possible information is included in consideration. In this paper we take into consideration the differential cross-sections measured by the L3 Collaboration at 183-189 GeV, OPAL Collaboration at 130-207 GeV, ALEPH Collaboration at 130-183 GeV and the cross-sections obtained by the DELPHI Collaboration at energies 189-207 GeV. They form the set of differential cross-sections and their uncertainties for the center of mass energies. Second, we use NNs to predict the data set of the investigated process with increased statistics. In experiments, the $4\pi$ scattering angle is divided into bins where the detectors are placed. For each bin we have a differential cross-section and an uncertainty [8]. The $Z^{\prime}$ extends the SM and therefore contributes to the differential cross-section $\frac{d\sigma}{dz}=\frac{d\sigma^{SM}}{dz}+\frac{d\sigma^{Z^{\prime}}}{dz},$ (1) where $z=\cos\theta$, $\theta$ is a scattering angle, $\frac{d\sigma^{SM}}{dz}$ is the contribution coming due to the SM particles and $\frac{d\sigma^{Z^{\prime}}}{dz}$ is the contribution due to the $Z^{\prime}$ presence. In actual calculations, the first term was taken from the results reported by the LEP Collaborations and the second one has been calculated in the improved Born approximation with the relations due to renormalizability been taken into account. In this case it is possible to construct the observables which uniquely pick out the $Z^{\prime}$ virtual state [1],[2]. Hence, the signal of the $Z^{\prime}$ could be searched in the deviation of experimental differential cross-section from the SM differential cross- section. Taking into account all the noted above we prepare the needed experimental data set. It consists of a differential cross-section for the Bhabha process and an uncertainty. In order to use it in our analysis we subtract from it the SM differential cross-section for the Bhabha process. Going this way we obtain the data set: 1. 1. scattering sector(bin). 2. 2. $\Delta\frac{d\sigma}{dz}$ \- the difference between a measured differential cross-section and the corresponding SM differential cross-section. 3. 3. $\varepsilon$ \- uncertainty. Figure 1: Experimental data are segments resulting from the experimental uncertainties. Bins are enumerated. 1-5 bins correspond to scattering in the backward direction ($Cos\theta<0$), the others - forward direction ($Cos\theta>0$). An overall distribution of the data is shown in Fig.1. As it is seen, this forms the set of segments determining the value of the cross-section with its uncertainty. To obtain a more strict constraint on the axial-vector $\bar{a}^{2}$ and vector $\bar{v}^{2}$ couplings squared, we have to decrease the lengths of segments. For this purpose the NN analysis is applied. At first stage, the NNs were trained to recognize the signal and the background. The contribution of the $Z^{\prime}$ to a differential cross-section for the Bhabha process obtained within the SM extended by $Z^{\prime}$ was taken as the signal. Analytic expression for this contribution reads [1]: $\frac{d\sigma^{Z^{\prime}}}{dz}=F_{v}(\sqrt{s},z)\bar{v}^{2}+F_{a}(\sqrt{s},z)\bar{a}^{2}+F_{av}(\sqrt{s},z)\bar{a}\bar{v}$ (2) where $F_{a}$ , $F_{v},F_{av}$ are the functions depend on the SM couplings. In actual analysis they have been calculated with accounting for the relations due to renormalizability. This contribution was computed for the values of $0.0\leq\bar{v}^{2}\leq 4\times 10^{-4}$ and $0.0\leq\bar{a}^{2}\leq 4\times 10^{-4}$. Such the choice is based on the results, obtained in [1]. As the background we set the deviation from the signal equaled to the redoubled uncertainty of LEP2 experimental data. Thus, the network was trained to discard data, which correspond to large deviations from the signal, but include the ones corresponding to the probable signal of the $Z^{\prime}$. The NN processing makes the cutoff of the data. The cutoff eliminates the data which are assumed to be the background signal. Side by side with the background signals the NN could discard the signal data too. But total amount of these events is negligibly small as compared to the amount of discarded background points. To create and train NNs the MLPFit program [6] was used. Three-layer NNs were used, with 2 neurons in input, 10 neurons in hidden and 1 neuron in output layer. An input vector for the networks consists of the scattering sector and the differential cross-section for this sector. The training algorithm with back propagation of errors was used. The type of training - with tutor - was applied. We also worked out a necessary computer program to solve the problem. During processing, the NNs discarded all the data that have produced a less than 0.9 at NN output. After NN processing, the length of segments was decreased. The total decrease of them is 3-27 percent (these values vary for different data sets). The comparative general plot of data set before and after the application of the NN is shown in Fig.2. Figure 2: Plot of $e^{+}e^{-}\to e^{+}e^{-}$ process data before and after NN processing. After NNs processing, the data were analyzed by means of the $\chi^{2}$ method. Denoting an observable by $\sigma_{i}$, one can construct the $\chi^{2}$-function $\chi^{2}(\bar{a},\bar{v}_{e})=\sum\limits_{i}[\frac{\Delta\sigma_{i}^{exp}-\Delta\sigma_{i}^{th}(\bar{a},\bar{v}_{e})}{\delta\sigma_{i}}]^{2},$ (3) where $\Delta\sigma^{exp}$ and $\delta\sigma$ is the deviation of the measured cross-section from the SM one and the uncertainty for the Bhabha process, and $\Delta\sigma^{th}$ is the calculated $Z^{\prime}$ contribution. The sum in Eq.(3) refers to either the data in one specific process or the combined data for several processes. By minimizing the $\chi^{2}$-function, the maximal- likelihood estimate for the $Z^{\prime}$ couplings was derived. The confidence area in the parameter space $(\bar{v},\bar{a})$ corresponding to the probability $\beta$ is defined as $\chi^{2}\leq\chi^{2}_{min}+\chi^{2}_{CL,\beta}$ (4) where $\chi^{2}_{CL,\beta}$ is the $\beta$-level of the $\chi^{2}$-distribution with 2 degrees of freedom. The analysis of the $\chi^{2}$-function for the $\bar{v}^{2}$ observable gives $\chi^{2}_{min}=158.49$ at $\bar{v}^{2}=2.37\times 10^{-4}$. The 95% CL area ($\chi^{2}_{CL}=5.99$) is shown in Fig.3. From this area the value $\bar{v}^{2}=(2.4\pm 1.99)\times 10^{-4}$ is obtained. The 95% CL area for data before the NN processing is shown in Fig.4. Figure 3: The 95% CL in $\bar{v}-\bar{a}$ plane for $\bar{v}^{2}$ observable. The data set after the NN processing. Figure 4: The 95% CL in $\bar{v}-\bar{a}$ plane for $\bar{v}^{2}$ observable. The data set before the NN processing. The analysis of the $\chi^{2}$-function for the $\bar{a}^{2}$ observable results in $\chi^{2}_{min}=629.66$ at $\bar{a}^{2}=1.6\times 10^{-7}$. The 95% CL area ($\chi^{2}_{CL}=5.99$) is shown in Fig.5. To compare the results, the 95% CL area for data before NN processing is shown in Fig.6. As one can see, the zero point of $\bar{a}$ is inside the confidence area. So, we are able to obtain only an upper limit on its value $\bar{a}^{2}\leq 1.1\times 10^{-4}$ at the $2\sigma$ CL. Figure 5: The 95% CL in $\bar{v}-\bar{a}$ plane for $\bar{a}^{2}$ observable. The data set after the NN processing. As it is seen, the zero point of $\bar{a}$ is inside the confidence area. Figure 6: The 95% CL in $\bar{v}-\bar{a}$ plane for $\bar{a}^{2}$ observable. The data set before the NN processing. To constrain the value of $Z^{\prime}$ mass by the derived bounds on the four- fermion couplings $\bar{v}^{2}$ let us assume that the coupling $\bar{g}$ is of the order of the SM gauge couplings, $\bar{g}^{2}/(4\pi)\simeq 0.01-0.03$. Then the obtained value $\bar{v}^{2}$ corresponds to $m_{Z^{\prime}}=0.53-1.05$ TeV. In Ref. [1] the value $\bar{v}^{2}=2.18\pm 1.82\times 10^{-4}$ was obtained for the Bhabha process at the $2\sigma$ CL within one-parameter fit. Table 1: Constraints on the $Z^{\prime}$ four-fermion couplings $\bar{v}^{2}$ obtained in different investigations one-parameter fit | two-parameter fit | NN two-parameter fit ---|---|--- $1\sigma$ | $2\sigma$ | $2\sigma$ | $2\sigma$ $\times 10^{-4}$ $2.24\pm 0.92$ | $2.18\pm 1.82$ | $\leq 1.44$ | $2.4\pm 1.99$ ## 3 Discussion It is shown in Fig.2 how the NN selects data corresponding to the background. We note once again that the NN discards not only a background but also the signal. Nevertheless, the discarded signal data are negligibly small as compared to the discarded background ones. The obtained values of $\bar{v}^{2}=(2.4\pm 1.99)\times 10^{-4}$ and $\bar{a}^{2}\leq 1.1\times 10^{-4}$ are comparable to the results obtained in [1]. In this paper the results were obtained within the one-parameter fit. In paper [2] it was shown that there is only a $1\sigma$ CL hint in the two- parameter fit. The analysis carried out in the present investigation shows the $2\sigma$ CL hint of the $Z^{\prime}$. The estimate of the mass value gives $m_{Z^{\prime}}=0.53-1.05$ TeV. This is signaling a comparably not heavy $Z^{\prime}$, that is of interest for the experiments at the Tevatron and LHC. The estimated couplings are also important to analyze the current and future experimental data. To conclude, we note that the derived predictions of the NNs analysis of the two parameter fits are in good agreement with other one-parameter model- independent fits accounting for the $Z^{\prime}$ gauge boson existence [5], [2]. ## References * [1] A.V. Gulov and V.V. Skalozub, Phys. Rev. D 70, 115010 (2004). * [2] A.V. Gulov and V.V. Skalozub, Phys. Rev. D 76, 075008 (2007). * [3] A. Leike, Phys. Rep. 317, 143 (1999). * [4] T. G. Rizzo, Presented at 34th International Conference on High Energy Physics (ICHEP 2008), Philadelphia, Pennsylvania, 30 Jul - 5 Aug 2008. * [5] A. Ferroglia, A. Lorca, J.J. van der Bij, Annalen Phys.16:563-578,2007 * [6] MLPFit, http://home.cern.ch/ schwind/MLPfit.html * [7] V.M.Abazov et al. [D0 Collaboration], Phys. Lett. B 517, 282 (2001). * [8] ALEPH Collaboration (R. Barate et al.), Eur. Phys. J. C 12, 183-207(2000); DELPHI Collaboration (J. Abdallah et al.), Eur. Phys. J. C 45, 589-632(2006); L3 Collaboration (M. Acciarri et al.), Phys. Lett. B 479, 101-117(2000); OPAL Collaboration (K. Ackerstaff et al.), Eur. Phys. J. C 2, 441-472(1998); OPAL Collaboration (G. Abbiendi et al.), Eur. Phys. J. C 6, 1-18(1999); OPAL Collaboration (G. Abbiendi et al.), Eur. Phys. J. C 33, 173-212(2004). * [9] ALEPH Collaboration, DELPHI Collaboration, L3 Collaboration, OPAL Collaboration, the LEP Electroweak Working Group, the SLD Heavy Flavour, Electroweak Working Group, hep-ex/0511027.

arxiv-papers

2024-09-04T02:48:53.680607

0802.1563

# YITP-08-3 KUNS-2123 KEK-CP-206 Determination of $B^{*}B\pi$ coupling in unquenched lattice QCD Hiroshi Ohkia,b , Hideo Matsufuruc , and Tetsuya Onogia aYukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan bDepartment of Physics, Kyoto University, Kyoto 606-8501, Japan cHigh Energy Accelerator Research Organization (KEK), Tsukuba 305-0801, Japan e-mail: ohki@yukawa.kyoto-u.ac.jpe-mail: hideo.matsufuru@kek.jpe-mail: onogi@yukawa.kyoto-u.ac.jp (February 12, 2008) The $B^{*}B\pi$ coupling is a fundamental parameter of chiral effective Lagrangian with heavy-light mesons and can constrain the chiral behavior of $f_{B}$, $B_{B}$ and the $B\rightarrow\pi l\nu$ form factor in the soft pion limit. We compute the $B^{*}B\pi$ coupling with the static heavy quark and the $O(a)$-improved Wilson light quark. Simulations are carried out with $n_{f}=2$ unquenched $12^{3}\times 24$ lattices at $\beta=1.80$ and $16^{3}\times 32$ lattices at $\beta=1.95$ generated by CP-PACS collaboration. To improve the statistical accuracy, we employ the all-to-all propagator technique and the static quark action with smeared temporal link variables following the quenched study by Negishi et al.. These methods successfully work also on unquenched lattices, and determine the $B^{*}B\pi$ coupling with 1–2% statistical accuracy on each lattice spacing. ## 1 Introduction One of the major subjects in particle physics is to determine the CKM matrix elements in order to test the standard model and find a clue to the physics beyond. While the precision of the experimental data from $B$ factories having been improving significantly, there are still large uncertainties in the CKM matrix elements due to the theoretical errors, which includes those in the lattice determination of the weak matrix elements for the $B$ mesons. It is often the case that a symmetry helps to obtain nonperturbative results in field theories. For example, the chiral Lagrangian based on the approximate chiral symmetry can help to understand the quark mass dependence of the light mesons and also to derive nontrivial relations between different physical quantities related by the chiral symmetry. For the $B$ mesons, there is another symmetry called ‘heavy quark symmetry’ which appears in the limit of infinitely large quark mass. Based on this symmetry one can construct the heavy meson effective theory, which gives a systematic description of the heavy-light mesons including $1/M$ corrections. Using this effective theory one can understand the light quark mass dependence of various physical observables of the $B$ meson weak matrix elements and can also derive nontrivial relations between different quantities, provided the low energy constants being determined from some method. The heavy meson effective Lagrangian has single low energy constant at the leading order of the $1/M$ expansion. This constant, $\hat{g}_{b}$, is called the $B^{*}B\pi$ coupling. Once the $B^{*}B\pi$ coupling is determined, the heavy meson effective theory can predict various quantities which are important for CKM phenomenology [1]. For example the light quark mass dependence of the $B$ meson decay constant and the bag parameter can be determined as $\displaystyle f_{B_{d}}=F\left(1+\frac{3}{4}(1+3\hat{g}_{b}^{2})\frac{m_{\pi}^{2}}{(4\pi f_{\pi})^{2}}\log(m_{\pi}^{2}/{\Lambda^{2}})\right)+\mbox{ analytic terms},$ (1) $\displaystyle B_{B_{d}}=B\left(1+\frac{3}{4}(1-3\hat{g}_{b}^{2})\frac{m_{\pi}^{2}}{(4\pi f_{\pi})^{2}}\log(m_{\pi}^{2}/{\Lambda^{2}})\right)+\mbox{ analytic terms}.$ (2) $F$ and $B$ are the low energy constants associated with these operators, and correspond to those quantities in the chiral limit of the light quark. Also the form factor $f^{+}(q^{2})$ for the semileptonic decay $B\rightarrow\pi l\nu$ can be expressed in terms of the $B^{*}$ meson decay constant $f_{B^{*}}$ and $\hat{g}_{b}$ as $\displaystyle f^{+}(q^{2})=-\frac{f_{B^{*}}}{2f_{\pi}}\left[\hat{g}_{b}\left(\frac{m_{B^{*}}}{v\cdot k-\Delta}-\frac{m_{B^{*}}}{m_{B}}\right)+\frac{f_{B}}{f_{B^{*}}}\right],$ (3) where $v$ is the velocity of the $B$ meson, $k$ is the pion momentum, and $\Delta=m_{B^{*}}-m_{B}$. Therefore it is quite important to determine the $B^{*}B\pi$ coupling very precisely from lattice QCD simulations. For this purpose, one of the promising approaches is to use the heavy quark effective theory (HQET) with nonperturbative accuracy including $1/M$ corrections. HQET allows systematic treatment of the $b$ quark in the continuum theory where $1/M$ corrections can also be systematically included with nonperturbative accuracy. Despite its usefulness, it is very difficult in practice to calculate the matrix elements for heavy-light systems with HQET [2, 3, 4]. This is because in the heavy-light system the self-energy correction to the static quark gives a significant contribution to the energy, which results in an exponential growth in time of the noise to signal ratio of the heavy-light meson correlators. In fact, recent results of $\hat{g}_{\infty}$ are $\displaystyle\hat{g}_{\infty}=0.51\pm 0.03_{\text{stat}}\pm 0.11_{\text{sys}}$ $\displaystyle\text{for}\quad n_{f}=0~{}\cite[cite]{[\@@bibref{}{Abada:2003un}{}{}]},$ (4) $\displaystyle\hat{g}_{\infty}=0.51\pm 0.10_{\text{stat}}$ $\displaystyle\text{for}\quad n_{f}=2~{}\cite[cite]{[\@@bibref{}{Becirevic:2005zu}{}{}]},$ (5) which have about $5\%$ and $15\%$ statistical errors for quenched and unquenched cases, respectively. An alternative method which extracts $\hat{g}_{\infty}$ from the $B$ quark potential was proposed in Ref. [5], but such accuracies would not be sufficient to test new physics. Therefore significant improvements for statistical precision in HQET are needed. Fortunately the two techniques to reduce the statistical error are developed recently, which are the new HQET action [6, 7] with HYP smearing [8] and the all-to-all propagators [9] with the low mode averaging [10, 11]. Negishi et al. [12] tested applicability of these methods on a quenched lattice, and found that the statistical accuracy is drastically improved as $\hat{g}_{\infty}=0.517(16)_{\rm stat.}\ \ \ \text{for}\ \ \ n_{f}=0,$ (6) namely at 2% level, even with a modest number of configurations. Our final goal is to extend the above strategy to unquenched simulations and give a precise value of the $B^{*}B\pi$ coupling $\hat{g}_{b}$ with $2+1$ flavors in the continuum limit. In this paper, we study the static $B^{*}B\pi$ coupling in $n_{f}=2$ unquenched QCD combining two techniques of the HYP smeared link and the all-to-all propagators. Our purpose is two-fold. The primary purpose is to perform the first high precision study of $\hat{g}_{\infty}$ in $n_{f}=2$ unquenched QCD, which serves a reference point for future studies with better control over the systematic errors. The secondary goal is to understand in what conditions the above methods apply efficiently. We observe the dependence of the statistical errors on the time and the numbers of low-lying eigenmodes, as well as their behavior against variation of the quark mass and the lattice spacing. This will help us to understand in which region of parameters the method can give good control over the statistical errors, which will also be useful to precision calculations of other physics parameters for heavy-light systems. This paper is organized as follows. In section 2, we describe the method to obtain $B^{*}B\pi$ coupling from the $B$ meson matrix element. Section 3 explains our simulation details. In this section we first arrive at our final result for the $B^{*}B\pi$ coupling with our best parameter setting. Then in Section 4, the efficiency of the low mode averaging is examined in detail. Conclusion is given in Section 5. ## 2 Lattice observables The Lagrangian of heavy meson effective theory is given as $\displaystyle L=-\mbox{Tr}\left[\bar{H}iv\cdot DH\right]+\hat{g}_{b}\mbox{Tr}\left[\bar{H}HA_{\mu}\cdot\gamma_{\mu}\gamma_{5}\right]+O(1/M),$ (7) where the low energy constant $\hat{g}_{b}$ is the $B^{*}B\pi$ coupling, $v$ is the four-velocity of the heavy-light meson $B$ or $B^{*}$, and $H$, $D_{\mu}$, $A_{\mu}$ are described by the $B$, $B^{*}$ and $\pi$ fields as $\displaystyle H$ $\displaystyle=$ $\displaystyle\frac{1}{2}(1+\gamma_{\mu}v_{\mu})(iB\gamma_{5}+B_{\mu}^{*}\gamma_{\mu}),\hskip 19.91684pt\xi=\exp(i\pi/f),$ (8) $\displaystyle D_{\mu}$ $\displaystyle=$ $\displaystyle\partial_{\mu}+\frac{1}{2}(\xi^{\dagger}\partial_{\mu}\xi+\xi\partial_{\mu}\xi^{\dagger}),\hskip 19.91684ptA_{\mu}=\frac{i}{2}(\xi^{\dagger}\partial_{\mu}\xi-\xi\partial_{\mu}\xi^{\dagger}).$ (9) The $B^{*}B\pi$ coupling can be obtained from the form factor at zero recoil which corresponds to the matrix element $\displaystyle\langle B^{*}(p_{B^{*}},\lambda)|A_{i}|B(p_{B})\rangle|_{\vec{p}_{B^{*}}=\vec{p}_{B}=0}=(m_{B}+m_{B^{*}})A_{1}(q^{2}=0)\epsilon^{(\lambda)}_{i},$ (10) where $A_{1}(q^{2}=0)$ is the matrix element of the transition from $B$ to $B^{*}$ at zero recoil with axial current $A_{i}\equiv\bar{\psi}\gamma_{5}\gamma_{i}\psi$ and $\lambda$ stands for polarization [2]. In the static limit, $\displaystyle\hat{g}_{\infty}=A_{1}(q^{2}=0)$ (11) holds. The matrix element $\langle B^{*}|A_{\mu}|B\rangle$ at the zero recoil can be obtained from the ratio of 3-point and 2-point functions, $R(t)$. $\frac{\langle B^{*}(0)|A_{i}|B(0)\rangle}{2m_{B}}=\lim_{t,t_{A}\rightarrow\infty}R(t,t_{A}),\\\ $ (12) where $R(t,t_{A})=\frac{\langle{\cal O}_{B^{*}}^{i}(t+t_{A})A_{i}(t_{A}){\cal O}_{B}(0)\rangle}{\langle{\cal O}_{B}(t+t_{A}){\cal O}_{B}(0)\rangle}\equiv\frac{C_{3}(t+t_{A})}{C_{2}(t+t_{A})}$ (13) with ${\cal O}_{B}$ and ${\cal O}_{B^{*}}$ are some operator having quantum numbers of the $B$ and $B^{*}$ mesons, respectively. We apply the smearing technique to enhance the ground state contributions to the correlators as $\displaystyle{\cal O}_{B}(t,\vec{x})$ $\displaystyle=$ $\displaystyle\sum_{\vec{r}}\phi(\vec{r})\bar{q}(t,\vec{x}+\vec{r})\gamma^{5}h(t,\vec{x}),$ (14) $\displaystyle{\cal O}_{B^{*}}^{i}(t,\vec{x})$ $\displaystyle=$ $\displaystyle\sum_{\vec{r}}\phi(\vec{r})\bar{q}(t,\vec{x}+\vec{r})\gamma^{i}h(t,\vec{x}),$ (15) where $\phi(\vec{x})$ is the smearing function. The lattice HQET action in the static limit is defined as $S=\sum_{x}\bar{h}(x)\frac{1+\gamma_{0}}{2}\left[h(x)-U_{4}^{\dagger}(x-\hat{4})h(x-\hat{4}))\right],$ (16) where $h(x)$ is the heavy quark field. The static quark propagator is obtained by solving the time evolution equation. As is well known, the HQET propagator is very noisy, and it becomes increasingly serious as the continuum limit is approached. In order to reduce the noise, the Alpha collaboration [6, 7] studied the HQET action in which the link variables $U_{\mu}(x)$ are replaced by the smeared links $W_{\mu}(x)$ in order to suppress the power divergence. They found that the noise of the static heavy-light meson is significantly suppressed with so-called HYP smearing [8]. The statistical error is further suppressed by applying the all-to-all propagator technique developed by the TrinLat collaboration [9]. Defining the Hermitian lattice Dirac operator $Q\equiv\gamma_{5}D$, where $D$ is the lattice Dirac operator, the quark propagator $S_{q}(x,y)$ is expressed by the inverse of the Hermitian Dirac operator $\bar{Q}=Q^{-1}$ as $\displaystyle S_{q}(x,y)=\bar{Q}(x,y)\gamma_{5}.$ (17) We divide the light quark propagator into two parts: the low mode part and the high mode part. The low mode part can be obtained using low eigenmodes of Hermitian Dirac operator $Q$. The high mode part can be obtained by the standard random noise methods with time, color, and spin dilutions. With the projection operators into the low and high mode parts, $P_{0}=\sum_{i=1}^{N_{\rm ev}}v^{(i)}(x)\otimes v^{(i)\dagger}(y),\hskip 19.91684ptP_{1}=1-P_{0},$ (18) respectively, the propagator can be decomposed into two parts as $\bar{Q}=\bar{Q}_{0}+\bar{Q}P_{1},$ (19) $\displaystyle\bar{Q}_{0}(x,y)$ $\displaystyle=$ $\displaystyle\sum_{i=1}^{N_{\rm ev}}\frac{1}{\lambda_{i}}v^{(i)}(x)\otimes v^{(i)\dagger}(y),$ (20) $\displaystyle(\bar{Q}P_{1})(x,y)$ $\displaystyle=$ $\displaystyle\frac{1}{N_{r}}\sum_{r}^{N_{r}}\sum_{j}\psi_{[r]}^{(j)}(x)\otimes\eta^{(j)\dagger}_{[r]}(y),$ (21) where $N_{r}$ is the number of random noise and $j$ is the index for dilution to label the set of time, spin and color sources, $j=(t_{0},\alpha_{0},a_{0})$. The low mode part $Q_{0}$ is constructed from the eigenvectors $v^{(i)}$ with their eigenvalues $\lambda_{i}$, which are to be obtained at a preceding stage. As the random noise vector for the high mode part, we adopt the complex $Z_{2}$ noise. The random noise vector with dilution is given as $\displaystyle\eta^{(j)}_{[r]}(\vec{x},t)^{a}_{\alpha}=\eta_{[r]}(\vec{x})^{a}_{\alpha}\delta_{t,t_{0}}\delta_{a,a_{0}}\delta_{\alpha,\alpha_{0}}.$ (22) $\psi$ is given as $\displaystyle\psi_{[r]}(x)=\sum_{y}(\bar{Q}P_{1})(x,y)\eta_{[r]}(y),$ (23) which is obtained by solving a linear equation $Q\psi_{[r]}=P_{1}\eta_{[r]}$. Further details of the computation methods are given in Ref. [12]. Combining these propagators, we can obtain the 2-point functions for the heavy-light meson which are averaged all over the spacetime. Similarly, the 3-point functions can be divided into four parts: low-low, low-high, high-low and high-high parts. ## 3 Results ### 3.1 Simulation setup $\beta$ | lattice size | $c_{sw}$ | $a^{-1}$[GeV] | $\kappa$ | $m_{\pi}$[GeV] | $N_{\rm ev}$ | $N_{conf}$ ---|---|---|---|---|---|---|--- 1.80 | $12^{3}\times 24$ | 1.60 | 0.9177(92) | 0.1409 | 1.06 | 200 | 100 | | | | 0.1430 | 0.90 | 200 | 100 | | | | 0.1445 | 0.75 | 200 | 100 | | | | 0.1464 | 0.49 | 200 | 100 1.95 | $16^{3}\times 32$ | 1.53 | 1.269(14) | 0.1375 | 1.13 | 0 | 120 | | | | 0.1390 | 0.92 | 200 | 150 | | | | 0.1400 | 0.76 | 200 | 150 | | | | 0.1410 | 0.54 | 200 | 150 Table 1: The simulation parameters. The values of the lattice spacing and the pion mass are from Ref. [13]. Numerical simulations are carried out on $12^{3}\times 24$ lattices at $\beta=1.80$ and $16^{3}\times 32$ lattices at $\beta=1.95$ with two flavors of $O(a)$-improved Wilson quarks and Iwasaki gauge action. We make use of about 100 to 150 gauge configurations provided by CP-PACS collaboration [13] through JLDG (Japan Lattice DataGrid). We use the $O(a)$-improved Wilson fermion for the light valence quark with the masses set equal to the sea quark masses. We use the static quark action with the HYP smeared links with the smearing parameter values $(\alpha_{1},\alpha_{2},\alpha_{3})=(0.75,0.6,0.3)$ (HYP1) [6, 7]. The $B$ and $B^{*}$ meson operators are smeared with a function $\phi(r)=\exp{(-0.9\,\hat{r})}$, where $\hat{r}$ is the distance between the heavy quark and the light quark in lattice units. The configurations are fixed to the Coulomb gauge. We obtain the low-lying eigenmodes of the Hermitian Dirac operator using implicitly restarted Lanczos algorithm. The low mode parts of the correlation functions are computed with $N_{\rm ev}$ = 200 low-lying eigenmodes, except for the case of $\kappa=0.1375$ at $\beta=1.95$ which is obtained with $N_{\rm ev}=0$. The reason of this choice will be explained in Sec. 4. The high mode parts of the correlation functions are computed with complex $Z_{2}$ random noise vector with $N_{r}=1$. The number of time dilution for each configuration are set to $N_{t_{0}}=24$ at $\beta=1.80$ and $N_{t_{0}}=32$ at $\beta=1.95$, respectively. This setup is based on the experience from the work by Negishi et al. [12]. ### 3.2 Correlation function and effective mass Figure 1: The effective mass plot of the 2-point and 3-point functions at $\beta=1.80$. Top left, top right, bottom left, bottom right panels correspond to $\kappa=0.1409$, 0.1430, 0.1445, 0.1464, respectively. Figure 2: The effective mass plot of the 2-point and 3-point functions at $\beta=1.95$. Top left, top right, bottom left, bottom right panels correspond to $\kappa=0.1375$, 0.1390, 0.1400, 0.1410, respectively. Figures 1 and 2 show the effective mass plots for the 2-point and 3-point functions . We find that the 2-point functions exhibit nice plateaux at $t\geq 4$ for $\beta=1.80$ and at $t\geq 5$ for $\beta=1.95$. From this result we take $t_{A}=5$ for $\beta=1.8$ and $t_{A}=6$ for $\beta=1.95$ as a reasonable choice for the time separation between the current $A_{i}$ and the $B$ meson source. We also find that the effective masses of the 3-point functions give consistent values with those of the 2-point functions. We fit the 2-point and 3-point functions to exponential functions with single exponent as $\begin{split}C_{2}(t)=Z_{2}\exp{(-E_{stat}t)},\hskip 19.91684ptC_{3}(t)=Z_{3}\exp{(-E_{stat}t)},\end{split}$ (24) where $Z_{2}$ and $Z_{3}$ are constant parameters and $E_{stat}$ corresponds to the energy of the heavy-light meson. The fit ranges are chosen appropriately by observing the effective mass plots as listed in Table 2. The bare $B^{*}B\pi$ coupling can be obtained by the ratio of the fit parameters as $\hat{g}_{\infty}^{\rm bare}=Z_{3}/Z_{2}$ . Alternatively, the $B^{*}B\pi$ coupling is also extracted from the ratio of the 3-point and 2-point functions, $C_{3}(t)/C_{2}(t)$, as shown in Figures 3. We find that the fit of the ratio $C_{3}(t)/C_{2}(t)$ to a constant value gives consistent value with the value of $Z_{3}/Z_{2}$. Since the statistical accuracy is better, we employ $Z_{3}/Z_{2}$ to determine $\hat{g}_{\infty}$ in the following analyses. The results are summarized in Table 2. $\beta$ | $\kappa$ | $(t^{\rm 2pt}_{\rm min},t^{\rm 2pt}_{\rm max})$ | $(t^{\rm 3pt}_{\rm min},t^{\rm 3pt}_{\rm max})$ | $aE_{stat}$ | $Z_{3}/Z_{2}$ | $\hat{g}_{\infty}$ ---|---|---|---|---|---|--- 1.80 | 0.1409 | (5,10) | (8,10) | 0.9412(19) | 2.252(21) | 0.612(5) | 0.1430 | (5,10) | (8,10) | 0.8839(21) | 2.294(23) | 0.598(5) | 0.1445 | (5,10) | (8,10) | 0.8343(16) | 2.342(13) | 0.591(4) | 0.1464 | (5,10) | (8,10) | 0.7488(17) | 2.381(27) | 0.578(5) 1.95 | 0.1375 | (8,14) | (11,14) | 0.7669(27) | 2.435( 8) | 0.618(8) | 0.1390 | (8,14) | (11,14) | 0.7093(18) | 2.471(16) | 0.615(5) | 0.1400 | (8,14) | (11,14) | 0.6638(15) | 2.461(14) | 0.599(4) | 0.1410 | (8,14) | (11,14) | 0.6098(14) | 2.400(13) | 0.571(4) Table 2: The numerical results of the heavy-light meson energy $aE$, the ratio of the 3-point and 2-point functions $Z_{3}/Z_{2}$, and $\hat{g}_{\infty}$. $(t^{\rm 2pt}_{\rm min},t^{\rm 2pt}_{\rm max})$ and $(t^{\rm 3pt}_{\rm min},t^{\rm 3pt}_{\rm max})$ are the fit ranges for the 2-point and 3-point correlators, respectively. For the values of $\hat{g}_{\infty}$, only the statistical errors are quoted. Figure 3: The ratio of the 3-point and 2-point functions versus $t$ at $\beta=1.95$. Correspondence of the panels and the values of $\kappa$ is the same as in Fig. 2. ### 3.3 Physical value of the $B^{*}B\pi$ coupling and chiral extrapolation The physical value of the $B^{*}B\pi$ coupling is obtained by multiplying the bare value by the renormalization constant. We use the one-loop result of the renormalization factor for the axial vector current $\displaystyle A_{i}=2\kappa u_{0}Z_{A}\left(1+b_{A}\frac{m}{u_{0}}\right)A_{i}^{lat},$ (25) $\displaystyle u_{0}=\left(1-\frac{0.8412}{\beta}\right)^{\frac{1}{4}},\ \ \ \ b_{A}=1+0.0378g_{\overline{\rm MS}}^{2}(\mu),$ where the gauge coupling $g_{\overline{\rm MS}}^{2}(\mu)=3.155$, 2.816 and $Z_{A}=0.932$, 0.939 for $\beta=1.80$ and 1.95, respectively, as given in Ref. [13]. We arrive at the results of $\hat{g}_{\infty}$ for our $\kappa$ values in Table 2. We take the chiral extrapolation of the $B^{*}B\pi$ coupling employing the following fit functions: $\displaystyle{\rm(a)}$ $\displaystyle g^{(a)}(m_{\pi}^{2})=g(0)+A_{1}m_{\pi}^{2},$ $\displaystyle{\rm(b)}$ $\displaystyle g^{(b)}(m_{\pi}^{2})=g(0)+A_{1}m_{\pi}^{2}+A_{2}(m_{\pi}^{2})^{2},$ $\displaystyle{\rm(c)}$ $\displaystyle g^{(c)}(m_{\pi}^{2})=g(0)\left[1-g(0)^{2}\frac{1}{8\pi^{2}}\frac{m_{\pi}^{2}}{f_{\pi}^{2}}\log{(m_{\pi}^{2})}\right]+A_{1}m_{\pi}^{2}+A_{2}(m_{\pi}^{2})^{2},$ corresponding respectively to (a) the linear extrapolation, (b) the quadratic extrapolation, and (c) the quadratic plus chiral log extrapolation where the log coefficient is determined from ChPT [14, 15, 16]. We use three lightest data points for the fit (a), while all the four points for (b) and (c). We obtain physical values of the $B^{*}B\pi$ coupling in the chiral limit as $\hat{g}_{\infty}=0.57(1)$, 0.57(2), 0.52(1) at $\beta=1.80$ and $\hat{g}_{\infty}=0.548(6)$, 0.529(10), 0.480(8) at $\beta=1.95$ from the fit (a), (b), (c), respectively. Figure 4 shows these chiral extrapolations. We take the average of the results from the linear fit and the quadratic plus chiral log fit as our best value and take half the difference as the systematic error from the chiral extrapolation: $\displaystyle\hat{g}(m_{\pi}=0)$ $\displaystyle=$ $\displaystyle 0.543(5)_{\text{stat}}(26)_{\text{chiral}}\hskip 19.91684pt\text{at}\ \ \beta=1.80,$ (26) $\displaystyle=$ $\displaystyle 0.516(5)_{\text{stat}}(31)_{\text{chiral}}\hskip 19.91684pt\text{at}\ \ \beta=1.95.$ (27) Figure 4: The chiral extrapolation of the physical $B^{*}B\pi$ coupling at $\beta=1.80$ (left panel) and $\beta=1.95$ (right). Since we have only two lattice spacings, naive continuum extrapolation would not give a reliable result. However, the results at these two lattice spacings are consistent within quoted errors.111 We do not observe a large scaling violation for $\hat{g}_{\infty}$ as opposed to the case of $f_{\pi}$ by CP- PACS collaboration. A possible explanation is that the large scaling violation for $f_{\pi}$ might come from the perturbative error of $c_{A}$, which is the $O(a)$-improvement coefficient of the light-light axial vector current. If this is the case, $f_{\pi}$ receives significant systematic error from the $O(a)$-improvement term for the temporal axial vector current $ac_{A}\partial_{4}P$ due to the chiral enhancement, whereas $\hat{g}_{\infty}$ does not receive such a systematic error since the corresponding term $ac_{A}\partial_{i}P$ drops out from the matrix element by the sum over the space. Therefore, we take the result at $\beta=1.95$ as our best estimate for the physical value of $\hat{g}_{\infty}$, and estimate the discretization error of $O((a\Lambda)^{2})$ by order counting with $\Lambda\sim 0.3$ GeV. Including the perturbative error of $O(\alpha^{2})$ also by order counting, our results for $\hat{g}_{\infty}$ is $\hat{g}_{\infty}^{n_{f}=2}=0.516(5)_{\text{stat}}(31)_{\text{chiral}}(28)_{\text{pert}}(28)_{\text{disc}}.$ (28) In our study, each of the chiral extrapolation error, the perturbative error, and the discretization error is about 6% level. The perturbative error can be removed by employing the nonperturbative renormalization such as the RI-MOM scheme, which is successfully applied to the light-light axial vector current. The discretization error could be reduced by computing on finer lattices. For example, an order counting estimate suggests that the discretization error would be reduced to about 2% on the configurations of CP-PACS at $\beta=2.10$. In contrast, it is not straightforward to control the chiral extrapolation error. It is definitely necessary to use recent unquenched configurations with smallest pion mass $m_{\pi}\sim 0.3$ GeV. More predominant approach is to employ a fermion formulation possessing the chiral symmetry, such as the overlap fermions, which makes the extrapolation theoretically more transparent. ## 4 Applicability of the low mode averaging In this section, we examine under what condition the all-to-all propagator technique, in particular the low mode averaging, is efficient to reduce the statistical error. This would give us a guide to extend our computation to unquenched simulations with smaller quark masses and finer lattices. We mainly investigate the case of $\beta=1.80$ in the following. ### 4.1 Observation of the noise to signal ratio Figure 5: The low-lying eigenmode distribution for various $\kappa$ at $\beta=1.80$ with 40 configurations. Figure 5 shows the distribution of about 250 lowest eigenmodes for each $\kappa$. Since the contribution of each mode $v^{(i)}$ to the correlator is multiplied by $1/\lambda_{i}$, one naively expects that with a fixed number of modes the low mode averaging should be particularly effective for the smallest quark mass. This is indeed true for the 2-point and 3-point heavy-light meson correlators. Figure 6 represents the low and high mode contributions to the 2-point correlators. For the smaller quark mass, the low mode part is indeed more dominant the 2-point correlator in the earlier $t$. Figure 6: The low and high mode contributions to the 2-point correlators versus $t$ at $\beta=1.80$. The left and right panels show the results at $\kappa=0.1464$ and 0.1430, respectively. Figure 7: The time dependence of the noise to signal ratio for $\kappa=0.1464$ (left panel) and $\kappa=0.1430$ (right) at $\beta=1.80$ with 40 configurations. In Figure 7 we show the comparison of the noise to signal ratio of the 2-point functions with different number of low eigenmodes, $N_{\rm ev}=0$, 50, 100, 200 for $\kappa=0.1464$ and $\kappa=0.1430$ at $\beta=1.80$. For the smallest quark mass, the statistical error of the 2-point function is 1.5–2 times improved as $N_{\rm ev}$ is changed from 0 to 200. While this is not a drastic improvement, comparing the costs to determine the low-lying eigenmodes and to solve the quark propagator (the latter is 8–10 times larger than the former), there is still an advantage to adopt the low-mode averaging. Projecting out the low-lying modes also improves the cost of solving the quark propagator. These effects are amplified as going to smaller quark mass region. In the region of larger light quark mass, however, the situation is different. The right panel of Fig. 7 shows the noise to signal ratio of the 2-point function for $\kappa=0.1430$. The noise to signal ratio for $t<7$ achieves about factor 1.3 improvement in the statistical error as we change $N_{\rm ev}$ from 0 to 200. For $t>7$, on the contrary, the noise to signal ratio with $N_{\rm ev}\neq 0$ starts to grow more rapidly than that with $N_{\rm ev}=0$ which keeps to grow steadily. As the result, the low-mode averaging deteriorates the statistical accuracy at this and larger light quark masses. ### 4.2 High mode and low mode contributions to the noise Figure 8: The time dependence of the noise from the high mode part for $\kappa=0.1464$ (left panel) and $\kappa=0.1430$ (right) at $\beta=1.80$ with 40 configurations. To investigate the origin of this behavior, we examine the high mode and low mode contributions to the error of correlator separately. In general, by projecting out larger number of low modes, the high mode part decreases. Therefore one might naively expect that the error also decreases. This is indeed the case for $\kappa=0.1464$. However for $\kappa=0.1430$ such a naive expectation does not hold. Figs. 8 shows the time dependence of the error for the high mode part of the 2 point correlator with various values of $N_{\rm ev}$ at $\kappa=0.1464$ and $\kappa=0.1430$. As is displayed in the right panel of Fig. 8, although the error of the high mode contribution to the correlator at $\kappa=0.1430$ does decrease at $t<7$ with larger $N_{\rm ev}$, the errors of the high mode part at $t>7$ with $N_{\rm ev}=50,100,200$ exceed that with $N_{\rm ev}=0$. This clearly indicates that for $t>7$ both the high and low mode parts of the correlator individually have large errors, but when they are combined the error of the total correlator becomes small. In this situation if the low mode part is improved by the low mode averaging, which reduces the error by certain factor, the error of the high mode part of the same size remains unreduced and dominates the error of the correlator. Figure 9: The noise to signal ratio of the low and high mode parts for the 2-point correlators for $\kappa=0.1430$ at $\beta=1.80$ with 40 configurations. The signal part is always taken to be the total correlator using only the noisy estimator. The projection is made with 50 eigenmodes. Figure 10: The noise to signal ratio of the low and high mode parts for the 2-point correlators at $\beta=1.80$ with 40 configurations. The signal part is always taken to be the total correlator using only the noisy estimator. The left and right panels show the dependence of the error of ‘low(noise)’ on $N_{\rm ev}$ for $\kappa=0.1464$ and $\kappa=0.1430$, respectively. To see it more clearly, let us decompose the 2-point correlator computed by the noisy estimator (corresponding to $N_{\rm ev}$=0) into the high and low mode parts. The total correlators are computed only with the noisy estimator (denoted as ‘total(noise)’). The high and low mode parts (‘high(noise)’ and ’low(noise)’) are separately computed using the exactly same random source as for the total correlators but projected into the high and low mode spaces with the projection operators $P_{1}$ and $P_{0}$, respectively. Figure 9 displays the statistical errors from the low and high mode parts normalized with the total correlator, in the case of $N_{\rm ev}=50$ at $\kappa=0.1430$. For comparison we also show the error of the low mode part determined with low mode averaging (‘low(eigen)’). This figure confirms that the fluctuations of the low and high mode parts are almost the same size and compensate in the total correlator so as to give much smaller error. The errors of the low mode parts, $E_{noise}(t)$ and $E_{eigen}(t)$, exponentially grow with similar rates, while different from that of the total correlator. Such a behavior continues as we decrease $N_{\rm ev}$ even down to a few $N_{\rm ev}$. The right panel of Fig. 10 shows the case of $N_{\rm ev}=1$, 4, 16, 50, 200 for $\kappa=0.1430$, where the fluctuations of the low mode part, $E_{noise}(t)$, grows similar rates, while absolute values are shifted downward. The left panel of Fig. 10 shows the case for $\kappa=0.1464$. We observe that the low mode part (‘low(noise)’) does not exceed the signal by large amount, which explains the behavior observed in Fig. 7. The reason why projecting into the low or high mode part provides a drastic enhancement of the error is still unknown. Although the phenomena themselves are quite interesting and deserves for further studies, in this paper we restrict ourselves within their implication to applicability of the all-to-all propagator technique to the static heavy-light system. ### 4.3 When is the low mode averaging efficient? We have seen that the low mode averaging is efficient only if the error of the noisy estimator (not the correlator itself) is dominated by the low mode part. Once the errors from the high and low mode parts of the correlator start to exceed the error of the total correlator, the low mode averaging is no longer effective but it makes the situation even much worse. Our result implies that at $\beta=1.80$ and $\kappa<0.1430$, the rapid growth of the error at $t\geq 7$ in Fig. 7 signals the breakdown of the above condition. For $\kappa\leq 0.1430$, the noisy estimator without the low mode averaging works better. Thus the low mode averaging is only efficient in the small quark mass region. However, since we have already taken the data and they provided satisfactory statistical accuracy of 2% level, we adopted the result with the low mode averaging propagator at $\beta=1.80$. As for $\beta=1.95$, the low mode averaging has not provided sufficient statistical accuracy for $\kappa=0.1375$. Thus we adopted the noisy estimator without the low mode averaging at this $\kappa$ as was already noted in the previous section. Figure 11: The time dependence of the noise to signal ratio of the 2-point correlators at $\beta=1.95$. The results are determined with $N_{\rm ev}=200$, while for $\kappa=0.1375$ the $N_{\rm ev}=0$ result is also displayed. The results at $\beta=1.95$ are displayed in Figure 11. The figure shows the noise to signal ratio of the 2-point correlators at each $\kappa$ against $t$ in physical units. For $\kappa=0.1375$ both the results with $N_{\rm ev}=0$ and 200 are shown, and the former indeed exhibits smaller statistical error. For all the values of $\kappa$ with $N_{\rm ev}=200$, the slopes of the exponential growth rate of the noise to signal ratio change around $t\sim 16$, and beyond that $t$ the slopes become steeper as the quark mass increases. This behavior is clearly explained with the breakdown mechanism of the low mode averaging mentioned above. We can also extract a hint on the lattice spacing dependence of the statistical accuracy by comparing $\beta=1.80$ and $\beta=1.95$. Comparison of Figs. 7 and 11 implies that the noise to signal ratio is similar or even smaller for finer lattices. This is partly explained by the fact that as going the finer lattices one has the larger number of lattice points (if the volume is kept unchanged) which are used for the all-to-all propagator. As observed in Figs. 1 and 7, the statistical errors at $\beta=1.80$ rapidly increases beyond $t\sim 8$, which corresponds to $t\sim 1.6$ fm in physical units. Thus the low mode averaging breaks down almost at the same physical distances at these two lattice spacings. This implies that also on finer lattices of $a\sim 0.1$ fm, one can extract precise values of $B^{*}B\pi$ coupling from the region $t<1.6$ fm by applying the same methods as this work, while careful tuning of the smearing function would be indispensable. ## 5 Conclusion In this paper, we computed the $B^{*}B\pi$ coupling on unquenched lattices using the HYP smearing and the all-to-all propagators. Using the low mode averaging with 200 eigenmodes, the statistical errors are kept sufficiently small for smaller quark masses. On the other hand, as was investigated in Sec. 4 in detail, the low mode averaging is not efficient for larger light quark mass region, where the simple noisy estimator provides better precision. In either case, the statistical error is controlled below 2% level in the chiral limit. We obtained consistent results at two lattice spacings. Our best estimate of the $B^{*}B\pi$ coupling in the static limit is represented in Eq. (28). Figure 12 compares our results with other recent works on the $B^{*}B\pi$ coupling [2, 4, 12]. The improvement in statistical precision is drastic, which proves the power of the improvement techniques employed in this paper. Figure 12: Comparison of $\hat{g}_{\infty}$ with other calculations [3, 4, 12]. In our results, the small and large errors represent the statistical error and the perturbative error, respectively. For future prospects, better control over the systematic error from the chiral extrapolation is indispensable. For this purpose, the configurations with dynamical overlap fermions by JLQCD collaboration would be a good choice [18, 19, 20]. In order to obtain $\hat{g}_{b}$ at the physical bottom quark mass, one needs to understand the mass dependence of $\hat{g}$. Simulations with the charm quark mass region and interpolation with the static limit are desired. The methods adopted in this work are in principle also applicable to other weak matrix elements of the $B$ mesons, such as $f_{B}$, $B_{B}$, and the form factors, and expected to provide high precision results required in precision flavor physics. ## Acknowledgments We would like to thank S. Aoki, M. Della Morte, N. Ishizuka, C. Sachrajda, T. Umeda for fruitful discussions. We are also grateful to S. Fajfer and J. Kamenik for useful comments. We acknowledge JLDG for providing with unquenched configurations from CP-PACS collaboration. The numerical calculations were carried out on the vector supercomputer NEC SX-8 at Yukawa Institute for Theoretical Physics, Kyoto University ,Research Center for Nuclear Physics, Osaka University and also Blue Gene/L at High Energy Accelerator Organization (KEK). The simulation also owes to a gigabit network SINET3 supported by National Institute of Informatics, for efficient data transfer supported by JLDG. This work is supported in part by the Grant-in-Aid of the Ministry of Education (Nos. 19540286, 19740160). ## References * [1] C. G. Boyd and B. Grinstein, Nucl. Phys. B 442, 205 (1995) [arXiv:hep-ph/9402340]. * [2] G. M. de Divitiis, L. Del Debbio, M. Di Pierro, J. M. Flynn, C. Michael and J. Peisa [UKQCD Collaboration], JHEP 9810, 010 (1998) [arXiv:hep-lat/9807032]. * [3] A. Abada, D. Becirevic, Ph. Boucaud, G. Herdoiza, J. P. Leroy, A. Le Yaouanc and O. Pene, JHEP 0402, 016 (2004) [arXiv:hep-lat/0310050]. * [4] D. Becirevic, B. Blossier, Ph. Boucaud, J. P. Leroy, A. LeYaouanc and O. Pene, PoS LAT2005, 212 (2006) [arXiv:hep-lat/0510017]. * [5] W. Detmold, K. Orginos and M. J. Savage, Phys. Rev. D 76, 114503 (2007) [arXiv:hep-lat/0703009]. * [6] M. Della Morte, S. Durr, J. Heitger, H. Molke, J. Rolf, A. Shindler, and R. Sommer [ALPHA Collaboration], Phys. Lett. B 581, 93 (2004) [Erratum-ibid. B 612, 313 (2005)] [arXiv:hep-lat/0307021]. * [7] M. Della Morte, A. Shindler and R. Sommer, JHEP 0508, 051 (2005) [arXiv:hep-lat/0506008]. * [8] A. Hasenfratz and F. Knechtli, Phys. Rev. D 64, 034504 (2001) [arXiv:hep-lat/0103029]. * [9] J. Foley, K. Jimmy Juge, A. O’Cais, M. Peardon, S. M. Ryan and J. I. Skullerud, Comput. Phys. Commun. 172, 145 (2005) [arXiv:hep-lat/0505023]. * [10] T. A. DeGrand and U. M. Heller [MILC collaboration], Phys. Rev. D 65, 114501 (2002) [arXiv:hep-lat/0202001]. * [11] L. Giusti, P. Hernandez, M. Laine, P. Weisz and H. Wittig, JHEP 0404, 013 (2004) [arXiv:hep-lat/0402002]. * [12] S. Negishi, H. Matsufuru and T. Onogi, Prog. Theor. Phys. 117, 275 (2007) [arXiv:hep-lat/0612029]. * [13] A. Ali Khan et al. [CP-PACS Collaboration], Phys. Rev. D 65, 054505 (2002) [Erratum-ibid. D 67, 059901 (2003)] [arXiv:hep-lat/0105015]. * [14] H. Y. L. Cheng, C. Y. L. Cheung, G. L. L. Lin, Y. C. Lin, T. M. Yan and H. L. Yu, Phys. Rev. D 49, 5857 (1994) [Erratum-ibid. D 55, 5851 (1997)] [arXiv:hep-ph/9312304]. * [15] J. F. Kamenik, arXiv:0709.3494 [hep-ph]. * [16] S. Fajfer and J. Kamenik, Phys. Rev. D 74, 074023 (2006) [arXiv:hep-ph/0606278]. * [17] G. P. Lepage, Nucl. Phys. Proc. Suppl. 26, 45 (1992). * [18] For an overview, H. Matsufuru [JLQCD Collaboration], arXiv:0710.4225 [hep-lat]. * [19] S. Hashimoto et al. [JLQCD collaboration], arXiv:0710.2730 [hep-lat]. * [20] T. Kaneko et al. [JLQCD Collaboration], PoS LAT2006, 054 (2006) [arXiv:hep-lat/0610036].

arxiv-papers

2024-09-04T02:48:53.685762

0802.1666

# Ice emission and the redshifts of submillimeter sources C. C. Dudley Formerly: Naval Research Laboratory, Remote Sensing Division, 4555 Overlook Ave. SW, Washington, DC 20375-5351, USA C.Christopher.Dudley.85@alum.dartmouth.org M. Imanishi11affiliation: Department of Astronomy, School of Science, Graduate University for Advanced Studies, Mitaka, Tokyo 181-8588 National Astronomical Observatory, Mitaka, Tokyo 181-8588, Japan P. R. Maloney Center for Astrophysics and Space Astronomy, University of Colorado, Bolder, CO 80309-0839, USA ###### Abstract Observations at submillimeter wavelengths have revealed a population of sources thought to be at relatively large redshifts. The position of the 850 $\mu$m passband on the Rayleigh-Jeans portion of the Planck function leads to a maximum redshift estimate of $z\sim$4.5 since sources will not retain their redshift independent brightness close to the peak of the Planck function and thus drop out of surveys. Here we review evidence that ice absorption is present in the spectra of local ultraluminous infrared galaxies which are often taken as analogs for the 850 $\mu$m source population. We consider the implication of this absorption for ice induced spectral structure at far infrared wavelengths and present marginal astronomical evidence that amorphous ice may have a feature similar to crystalline ice near 150 $\mu$m. Recent corroborative laboratory evidence is supportive of this conclusion. It is argued that early metal enrichment by pair instability SN may lead to a high ice content relative to refractory dust at high redshift and a fairly robust detection of ice emission in a $z=6.42$ quasar is presented. It is further shown that ice emission is needed to understand the 450 $\mu$m sources observed in the GOODS-N field. We are thus encouraged to apply far infrared ice emission models to the available observations of HDF 850.1, the brightest submillimeter source in the Hubble Deep Field. We suggest that a redshift as large as 13 may need to be considered for this source, nearly a factor of three above the usual top estimate. Inclusion of the possibility of far infrared ice emission in the spectral energy distributions of model sources generally broadens the range of redshifts to be considered for submillimeter sources compared to models without ice emission. galaxies: high redshift, formation — infrared: galaxies — galaxies: individual: IRAS 00188-0856, IRAS 14348-1447, QSO J1148+5251, HDF 850.1 ## 1 Introduction Sources discovered in surveys undertaken using 450/850 $\mu$m SCUBA camera mounted on the JCMT have a high number density per unit solid angle ($\sim$ 104 deg-2; S${}_{850}>$ 1 mJy; Sanders 2000; Wang et al. 2004) and are thought to trace the dust emission from sources at a range of redshifts. Sources brighter than 1 mJy which have Arp 220-like far infrared (FIR) spectral energy distributions (SEDs) are ultraluminous in the $h_{100}=0.71$, $\Omega_{m}=0.27$, and $\Omega_{\Lambda}=0.73$ cosmology consistent with cosmic microwave background (CMB) and Type 1a SNe observations (Spergel et al. 2003; Nolta et al. 2004; Riess et al. 2004) unless they are at redshifts $<$ 1 so that, thus far, submillimeter surveys do not detect “normal” galaxies at high redshift. This should be the case even for gravitational lens amplification up to a factor of 4 for 2 mJy limit surveys. It is therefore a sensible approach to consider the properties of local ultraluminous infrared galaxies (ULIRGs; Sanders & Mirabel 1996) as a guide to understanding sources detected in submillimeter surveys with the proviso that the high redshift universe may not conserve these properties. It has already been noted that the number density of ULIRGs increases with redshift at a rate much larger than can be explained using geometric considerations alone (Kim & Sanders 1998) and deep field observations using the Hubble Space Telescope have found quite obvious morphological evolution at higher redshifts compared to local conditions (e.g. Brinchmann et al. 1998). While these particular evidences of evolution as a function of redshift do not directly speak to ULIRG SED evolution, they do suggest that this proviso cannot be lightly discounted. Ice is present in the dense ISM of the Galaxy. It is observed to occur above a threshold depth of A${}_{V}\sim$ 3 within molecular clouds but is absent on the cloud edges and in the diffuse (atomic/ionized) ISM (Whittet et al. 1988). Because ice is not exposed to star light, far infrared (FIR) emission from ice is not expected from quiescent clouds. The FIR emission from these ice bearing clouds arises from the first few ice-free AV. To observe ice emission in the Galaxy, centrally heated sources are needed. If mid-infrared (MIR) radiation can illuminate the ice bearing grains so that they may be heated, then the FIR emission will be dominated by ice if the ice mantles provide a moderate fraction of the total grain volume. This FIR dominance is owing to greater FIR opacity of ice relative to silicate or carbonaceous grains. In the models presented by Aannestad (1975), a mantle-to-core volume ratio of 0.33 doubles the 60 $\mu$m opacity: a 10% increase in grain radius doubles the opacity! Protostars in the Galaxy which are centrally heated turn out to be good sources for observing FIR ice emission (e.g. Dartois et al. 1998). And, there is mounting evidence that a significant fraction of local ULIRGs are also centrally heated (Dudley & Wynn-Williams 1997; Soifer et al. 2002; Imanishi & Dudley 2000; Imanishi & Maloney 2003; Imanishi et al. 2006a; Spoon et al. 2006) so that the FIR emission is a result of MIR heating. Thus, if ice is present in the FIR emitting regions, its effects on the FIR opacity may be important. This situation differs from what occurs in lower luminosity starburst galaxies where the FIR emission arises from the surfaces of molecular clouds which are heated by optical and far UV rather than MIR light. The interiors of the clouds, where ice presumably exists, are not strongly heated by the mainly FIR emission arising from their surfaces. If submillimeter galaxies are similar to ULIRGs in more than just their luminosity, then they may also sometimes have FIR emission that is powered by MIR light. For redshift estimation the question of the similarity between submillimeter galaxies and local ULIRGs is as important as the differences that distinguish ULIRGs from their lower luminosity starburst counterparts. Attempts to estimate redshifts for the submillimeter sources based on their radio fluxes could be guided by local ULIRG samples. For the 13 ULIRGs brighter than 5.24 Jy at 60 $\mu$m (Soifer et al. 1987; Sanders et al. 1995) available to the VLA, the mean logarithm of the ratio of FIR-to-20 cm radio flux is 0.22 dex larger than $q$=2.43$\pm$0.19 (Condon et al. 1991) for starburst galaxies when calculated using radio measurements reported by Condon et al. (1991) and Yun et al. (2001). While the mean differs at only the 1$\sigma$ level, the dispersion of $q$ is a factor 1.6 (in dex) larger than for the lower luminosity star forming galaxies. Using a different luminosity binning, Helou et al. (1985) report a similar trend in increased dispersion with increasing luminosity. The implication for redshift estimation is that the range of likely redshifts would increase when using ULIRG properties rather than starburst properties. And, a trend away from what is seen in local ULIRGs, say through more powerful starbursts at earlier times, could have complex implications for estimating redshifts of submillimeter galaxies using radio observations. However, broad limits on the redshift distribution derive from submillimeter observations alone where the relative infrequency of detecting 450 $\mu$m emission leads to (weak) lower limits on the redshifts while the detections themselves at 850 $\mu$m lead to upper limits since, baring extraordinary luminosities, the turnover from the Rayleigh-Jeans-like portion of the SED at high redshift leads to source fluxes dropping below survey detection limits. Given the large potential effect of ice on the FIR opacity of submillimeter galaxies, we first review the evidence for the presence of ice in centrally heated ULIRGs and estimate an expected FIR SED for a case where FIR ice opacity can be expected to be dominant based on a rough abundance analysis. We then search for FIR ice spectral features in the available data on ULIRGs. We discuss why a high relative abundance of ice during the epoch of reionization might be expected and present somewhat firmer evidence for FIR ice emission in the $z=6.42$ source QSO J1148+5251 than can yet be demonstrated for ULIRGs. We then take the submillimeter galaxy HDF 850.1 as an example to show how FIR ice emission allows a very large redshift estimate. ## 2 Ice in ULIRGs ### 2.1 Absorption Observations of strong absorption features from refractory dust in ULIRGs trace dust that is cooler than the temperature of a blackbody emitting at the wavelength of the feature: $\sim$ 300 K for the 10 $\mu$m silicate feature and $\sim$ 150 K for the 20 $\mu$m silicate feature for example. Thus, 20 $\mu$m silicate features are weak in ULIRGs compared to 10 $\mu$m features owing to the dust radial temperature gradient. Ice absorption, on the other hand, should only occur in the cooler outer regions of the dust envelope and so may show the same relative feature strengths found in the laboratory unless the absorbing covering factor is a strong function of the emitting source’s wavelength dependent size (buried source radius $r\approx 0.8({L\over 10^{12}L_{\odot}})^{0.5}({\lambda\over 3\mu m})^{2}$ pc where T has been eliminated between the Wein Displacement and Stefan-Boltzmann Laws, $L$ is the portion of IR luminosity corrected for extended star formation, $\lambda$ runs from 2 $\mu$m to the lesser of either the wavelength where the source is no longer optically thick or the peak FIR wavelength and scattering is neglected). MIR ice absorption is therefore consistent with a portion of the FIR emission being powered by the MIR rather than the optical and UV portion of the electromagnetic spectrum. However, it is necessary to be certain that the reported absorption is indeed owing to ice. As an example, Spoon et al. (2002) claim ice absorption in 18 galaxies based on 6 $\mu$m absorption including IRAS 17208-0014 the third brightest ULIRG. However OH bond bending needn’t occur in ice alone. The absence of 3 $\mu$m absorption in the spectra of the dust continuum dominated sources Mkn 231 and IRAS 05189-2524 presented by Spoon et al. (2002), where veiling by starlight is unimportant, suggests that an identification of water ice may be premature for these sources and refractory materials such as those studied by Greenberg et al. (1995) may provide a more convincing model. An apparent 6 $\mu$m dip in the spectrum of Mkn 231 plotted by Weedman et al. (2005) similar in amplitude to the 6.2 $\mu$m hump for which they report a measurement tends to confirm the Spoon et al. (2002) claim of 6 $\mu$m absorption in this source. The spectrum of Arp 220 may be consistent with a contribution of 12.5 $\mu$m libration mode absorption, supportive of the Spoon et al. (2002) identification of ice absorption in this source but, until recently (Imanishi et al. 2006a; Risaliti et al. 2006; Imanishi 2006), the only well confirmed case of ice absorption among ULIRGs in their sample was that of IRAS 00188-0856 where ice absorption at 3 $\mu$m confirms the identification (Imanishi & Maloney 2003). See Imanishi et al. (2006b) for further well confirmed ice identifications. To illustrate this we present Fig. 1 which employs a slightly detailed model to explain the absorption features observed between 2.7 and 12.4 $\mu$m. The upper panel reproduces the 3 $\mu$m spectrum reported by Imanishi & Maloney (2003) together with an absorption model composed of amorphous ice and refractory carbonaceous absorption applied to a power-law continuum, and veiled by some starburst-like emission. The model is then applied in the lower panel with some adjustment made for expected reduced refractory absorption at longer wavelengths owing to a presumed dust temperature gradient. It has not been possible to fit the entire 3 $\mu$m absorption without using ice. Thus, the absorption at 6 $\mu$m also is partly attributed to water ice. With these data, 12.5 $\mu$m water ice absorption, while included in the model, does not seem to be required so that one may conclude that two features, and particularly the degeneracy breaking short wavelength edge of the 3 $\mu$m absorption contribute to the fairly robust identification of ice in IRAS 00188-0856. We note that the need for strong silicate absorption in this source is confirmed by the clear presence of 18 $\mu$m silicate absorption (Imanishi et al. 2006b) Thus, ice absorption could play a significant role in ULIRGs, but the evidence is not yet as strong as that showing that the MIR emission is often absorbed by refractory dust. Evidence for or against FIR ice emission or absorption in ULIRGs is in worse shape still. This is primarily owing to a lack of observational data. In Fig. 2 we present archival ISO LWS spectra of Mkn 231 (filled diamonds) and IRAS 17208-0014 (filled circles) which show spectral structure in the shortest wavelength detector band SW 1. This structure is compared to the spectrum of the icy source HD 161796 (Hoogzaad et al. 2002) transformed to absorption. For Mkn 231, as just noted, the absence of 3 $\mu$m ice absorption would lead to somewhat strained geometric arguments to allow acceptance of the Spoon et al. (2002) identification of the 6 $\mu$m absorption, but once made, such arguments would likely allow absorption at even longer wavelengths. A complex geometry has already been invoked for this source on a smaller physical scale to explain its low X-ray flux (Maloney & Reynolds 2000). For IRAS 17208-0014, the present spectrum would tend to confirm the Spoon et al. (2002) identification of the 6 $\mu$m absorption as do 3 $\mu$m spectra (Risaliti et al. 2006; Imanishi 2006). For either source, the presence of crystalline ice implied by the possible 43 $\mu$m absorption would imply a low cosmic ray abundance at least in the region where the absorption is produced since ion bombardment erases crystalline structure in cold ices (Moore & Hudson 1992). The inset in Fig. 2 shows the relationship between the wavelength of the 43 $\mu$m peak and ice temperature measured by Smith et al. (1994) which we have invoked to shift the ice spectrum derived from HD 161796, an evolved ice forming star (Hoogzaad et al. 2002), beyond the corrections made for the source redshifts to better match the spectra. Inversion of these corrections would imply temperatures at which amorphous ice may exist and would be preferentially formed. Constraints on cosmic ray abundance would need account for the rate at which crystalline ice is introduced into the absorbing region either through (substantial) changes in luminosity of the source which may allow the annealing of amorphous ice to crystalline ice, processing of ice by local heat sources in the region, or the migration of crystalline ice into the region. The discovery of substantial radial molecular gas motion in the ULIRG IRAS 08572+3915 (Geballe et al. 2006) suggests the third possibility needs close scrutiny. Since the SW1 detector of LWS is problematic, these sources must be independently re-observed and the potential features confirmed before fully accepting that ice is present in these sources. ### 2.2 Millimeter morphology We have examined the LWS SW1 spectra of a two other ULIRGS, Arp 220 and Mkn 273, finding no feature consistent with crystalline ice absorption. In the case of Arp 220, the FIR spectrum appears to be optically thick (González- Alfonso et al. 2004). There is every reason to think ice should be present owing to the source temperature, the clear presence of water vapor, as well as supportive MIR observations already mentioned. Here, ice may help to resolve an issue raised by Soifer et al. (1999) with regard to FIR sources that are optically thick showing any detectable emission in the MIR owing to overwhelmingly large implied MIR opacity. Since ice strongly enhances the FIR opacity compared to refractory dust, a given FIR optical depth can be achieved with less dust, up to a factor of seven in the models computed by Aannestad (1975). This effect can make the issue raised by Soifer et al. (1999) less troubling, though other approaches such as examining the effects of the dust density distribution (e.g. Nenkova et al. 2002) may also need to be explored as well. Ice may also have the effect of making millimeter continuum size estimates smaller than what might be expected for refractory dust that is optically thick in the FIR since the opacity of ice mantled grains is lower than that of bare grains at millimeter wavelengths. This could yield a cuspy morphology in the millimeter where one would usually expect a relatively more flattened morphology owing to the cooler outer regions being relatively brighter in the millimeter under the assumption of higher refractory dust opacity. ### 2.3 FIR ice spectrum The infrared laboratory spectrum of crystalline water ice has been well known for some time (Bertie et. al 1969) and differences between the spectrum of crystalline and amorphous ice have been exploited to aid understanding of astronomical sources (e.g. Dartois et al. 1998). However owing to the low thermal conductivity of amorphous ice, preparation of laboratory samples, via vapor deposition, that are thick enough to conduct investigations at wavelengths longer than 100 $\mu$m has been too difficult until recently. In particular it has been unknown whether or not amorphous ice has a feature corresponding to the broad feature centered near 150 $\mu$m identified by Bertie et. al (1969) as a phonon mode in crystalline ice. Since nature provides sources rich in amorphous ice, it may be that such a comparison could be made through observations of these sources. The spectral coverage of the ISO LWS spectrometer extended to $\sim$ 200 $\mu$m, too short to fully cover the broad 150 $\mu$m crystalline ice feature. Comparison between the ISO LWS data presented by Dartois et al. (1998) for RAFGL 7009S, a deeply embedded ultracompact H II region, and the submillimeter observations provided by McCutcheon et al. (1995) yield a modified optically thin blackbody parameter set of T=27 K, $\beta$=2.1 (where the dust emissivity is proportional to $\lambda^{-\beta}$) when attempting to the match data near 190 $\mu$m. The large value of $\beta$ is suggestive of the presence of a feature similar to that of crystalline ice in this amorphous ice dominated source. However, full spectroscopic confirmation is needed to decide if the long wavelength departure of the data from the model of Dartois et al. (1998) indicates the existence of an amorphous ice feature analogous to the 150 $\mu$m crystalline ice feature. Laboratory work by Curtis et al. (2005) has explored the optical properties of amorphous ice between 15 and 200 $\mu$m. There is little question that most of the ice in the thick films used by Curtis et al. (2005) deposited at $T\leq 126$ K differs from ice deposited above this temperature since the wing of the libration band is clearly shifted so it seems safe to conclude that amorphous ice, clearly present in the thinner samples, is also largely present in the thicker samples despite issues related to the low thermal conductivity of amorphous ice which hampers the (rapid) growth of thick samples. Gerakines et al. (2005) have also had recent success in growing thick samples of amorphous ice using low deposition rates. Schober et al. (2000) take a different approach: annealing samples of high density amorphous ice into the low density amorphous ice found in space. As with the LWS spectroscopy, the new amorphous ice optical constants (Curtis et al. 2005) do not extend to the wavelength regime (300 $\mu$m) where crystalline ice shows absorption whose wavelength dependence is described by $\lambda^{-4}$ (Bertie et. al 1969). Unlike natural sources, however, where temperature and grain size effects make full spectral coverage of a broad feature nearly absolutely necessary to ascertain its reality, the controlled laboratory conditions allow a reasonable extrapolation to be considered reliable. The finding of Curtis et al. (2005) that the imaginary part of the refractive indices of amorphous and crystalline ice are very similar between 100 and 200 $\mu$m would appear to be consistent with the suggestion of Schober et al. (2000) that the order in amorphous ice indicated by sharp features in inelastic X-ray scattering data may be owing to a hydrogen network that is also present in crystalline ice. If this order is indeed present, then a shared spectral structure for the interaction of (FIR) photons with phonons between the two types would allow a first order substitution of crystalline ice optical constants for the unknown values for amorphous ice beyond 200 $\mu$m. ### 2.4 FIR ice emission Turning to photometry, and returning to ULIRGs, if we accept the Spoon et al. (2002) classification scheme then the spectrum of IRAS 14348-1447 as reported by Genzel et al. (1998) or Charmandaris et al. (2002) must be considered Class II, that is 6 $\mu$m “ice” absorption is present and partially filled in by 6.2 $\mu$m polycyclic aromatic hydrocarbon emission. This source occupies an extreme position in the dust emissivity ($\beta$) vs. 100 $\mu$m optical depth ($\tau_{100}$) diagram given by Klaas et al. (2001) with both high $\tau_{100}$ (5) and $\beta$ (2). Since the effects of optical depth and emissivity are intertwined, a lower optical depth can imply a larger value of $\beta$. For example, $\beta=2.4$ and $\tau_{100}=0.1$ does about as well as the Klaas et al. (2001) parameters for data longward of 100 $\mu$m. However, values of $\beta$ larger than 2 are not generally found except where ice is thought to be present (e.g. Lis et al. 1998) or dust temperatures are lower than found in ULIRGs (67 K for the Klaas et al. (2001) fit or 26 K at the lower optical depth and larger $\beta$ tested here). In Fig. 3 we compare the fit proposed for IRAS 14348-1447 by Klaas et al. (2001) with an alternative model. The model shown by the dashed line is that of Klaas et al. (2001), while that shown by the thick solid line is a core-mantle grain model taken from Aannestad (1975) but modified to have amorphous ice features in the following way: First, opacity data from Smith et al. (1994) for crystalline ice cooled to 70 K was matched to the Aannestad (1975) curve for Vm/Vc=0.95 by inducing a $\nu^{-1}$ dependence to approximate the Mie calculation results along with the addition of an underlying continuum opacity. Second, amorphous ice opacity data (Smith et al. 1994; deposited at 10 K and warmed to 70 K) was substituted between 30 and 70 $\mu$m based on this matching after adjustment for the differing sample thicknesses. This opacity curve was used to calculate a model with 16 and 71 K components. If we take temperature and optical depth as free parameters for the Klaas et al. (2001) fit (2 parameters; normalization fixed by IRAS) and two temperatures, two optical depths, and the relative contribution at 90 $\mu$m of these two components as 5 free parameters for the present model then we may estimate how well the two models represent the data if we know something about the errors in the data. The statistical errors for the data are thought to be small but the absolute calibration error could be large, $\sim$30%. What is important for comparing models is the relative accuracy from filter to filter. We have thus assumed 13% errors for the ISOPHOT data to give a filter-to-filter accuracy of 20% (Klaas et al. 2000). The fit of Klaas et al. (2001) for data longward of 20 $\mu$m is unacceptable (P($\chi^{2}$,$\nu$)=0.02) under our assumption about the errors. A marginally acceptable fit (P($\chi^{2}$,$\nu$)=0.4) is found for the amorphous ice model. Detailed inspection of the ice fit suggests that the 150 $\mu$m ice feature is not accounting for all of structure in the data. The thin solid line (Vm/Vc=0.33, T=28.8 K; $\tau_{90}=0.2$ and Vm/Vc=0.95, T=80.5 K, $\tau_{90}=0.08$ crystalline ice; onion-skin model with warm, T=180 K, refractory dust in the interior) shows that this might be possible. However, our method of substituting amorphous for crystalline ice is unlikely to work well for the thin mantle used here. Given the assumption regarding the data errors made here, it is not possible to make a clear claim that ice emission is detected in this source. Assuming larger errors could make both low free parameter models acceptable while reducing the errors could make the simpler ice model unacceptable while the more complex model cannot be evaluated via reduced $\chi^{2}$. Clearly, Spitzer MIPS SED mode observations could be definitive since the strong 70 $\mu$m curvature in the ice models cannot be produced with standard refractory dust. Thus, both astronomical and laboratory far infrared data leave questions about ice unanswered. Crystalline ice has a feature near 150 $\mu$m which has yet to be completely compared with amorphous ice. Comparison between crystalline and amorphous ice has been made near 40 and 60 $\mu$m and they are quite distinct whereas they are similar between 100 and 200 $\mu$m. Beyond 200 $\mu$m questions remain which are presently answered by extrapolation of the laboratory data and hints from the astronomical data. Substituting crystalline ice for more common (in space) amorphous ice to consider wavelengths longer than 100 $\mu$m could add to (systematic) uncertainties in photometric redshift estimation that can not be well addressed without further observations or measurements. Regardless of the remaining issues concerning the 150 $\mu$m feature, for the main purpose of this work it is prudent to consider the more extreme consequences of crystalline ice emission on redshift estimation which are examined after the next section. ## 3 Ice at high redshift ### 3.1 Pair instability SNe enrichment Just as interplanetary grains may preserve a record of the sources of ISM enrichment occurring at what is presently observed to be $z=0.43$, the solid phase of the ISM at high redshift may also record the conditions of the earliest enrichment. Particularly for sources that contribute to the reionization of the universe ($7<z<14$; Spergel et al. 2006), scenarios which attribute substantial enrichment to pair instability SNe early on may leave a mark on the solid phase of the ISM which could linger through some subsequent processing. A chief attribute of pair instability SN enrichment is an early high abundance of oxygen relative to other metals that usually deplete to the the solid phase of the ISM. For example, a 186 M⊙ progenitor yields number abundance ratios of O:Si:C:S:Mg:56Ni = 126:29:15:10:8:1 (Heger & Woosley 2002) which, after formation of CO (29 O removed), olivine (16 O and 4 Si removed), silica (50 O removed), and SO2 (20 O removed), leaves 9% of the oxygen available to form OH and H2O. For a 100 M⊙ progenitor, O:C:Mg=48:9:1 with little Si, S or 56Ni leaving 80% of the oxygen available to combine with hydrogen assuming the refractory solid state is MgO. In the absence of 56Ni, the formation of solids should be enhanced since it may commence at a higher density in the SNR. These estimates for the available oxygen are essentially lower limits, more complete treatments which consider the possibility that the ejecta are unmixed (Nozawa et al. 2003; Schneider et al. 2004; allowing the formation of graphite for example) would leave an even larger fraction of oxygen available. Thus, we might expect grains with large mantle-to-core volume ratios to result from a period of pair instability SN enrichment. The continued presence of these ice grains would depend on the rate of reprocessing of this initial solid phase in the ISM. ### 3.2 Ice emission in a high redshift quasar The Gunn-Peterson trough in QSO J1148+5251 ($z=6.42$; Bertoldi et al. 2003b) suggests that this source may be participating in the last phases of reionization (Fan et al. 2003). That the quasar exists indicates that the source has been contributing to reionization for some time and we may take the beginning of reionization as an estimate of it’s age ($\sim$500 Myr) which gives a high average accretion rate for a blackhole mass of $3\times 10^{9}$ M⊙ (Willot et al. 2003). Thus, it seems plausible that pair instability SN enriched material with a high ice content is being supplied toward the center of this activity as the blackhole and presumably its stellar system grow in mass. On the other hand, ice also forms in the more evolved ISM of the Galaxy so an indication of the presence of ice needn’t be fully attributed to ice rich regions that might be formed by pair instability SNe. Fig. 4 shows an indication that the rest frame FIR emission from QSO J1148+5251 is better explained by ice than refractory dust emission. Three models are compared to photometric observations: 1) a refractory model (dashed line T=55 K; $\tau_{60}=0.3$; $\beta=2$) is similar to that given by Beelen et al. (2006), 2) a simple ice model (dot-dashed line; T=43 K; $\tau_{60}=0.4$, Vm/Vc=5.6 amorphous), and 3) a more complex ice model described below (solid line). In Fig. 4, it is not possible to formally access the goodness of fit through a reduced $\chi_{\nu}^{2}$ calculation even for the refractory dust model (dashed line) since the number of free parameters (normalization, temperature, optical depth, and opacity index) is the same as the number of points to be fit. If one were to assume $\beta=2$ owing to insufficient time since the first enrichment to produce large grains then $\chi_{\nu}^{2}=10.7$ which strongly rules out (P($\chi,\nu$)=0.001) this model. That the fit is done by eye probably does not affect this assessment but a failure to account for systematic errors could since different groups give 850 $\mu$m photometric estimates using the same data that vary by as much as the refractory dust model differs from the present data. By selecting a particular mantle-to-core volume ratio (5.6) we may comparably provide a single degree of freedom and obtain $\chi_{\nu}^{2}=3.9$ (dot-dashed line) which would indicate that the model is not fully ruled out by the data (P($\chi,\nu$)=0.05) and is an improvement on the refractory dust model, again with the proviso that we have only considered statistical errors. More elaborate (multi-temperature) refractory dust or ice mantled grain models can fit the detections essentially perfectly and are useful to calculate as they can indicate what further observations would be decisive. The solid line in Fig. 4 is such a model where T=71 K; $\tau_{60}=0.4$ ice-mantled dust is obscured by T=24 K; $\tau_{60}=0.4$ ice-mantled dust. A similar model using $\beta=2$ refractory dust (not plotted; T=300 K; $\tau_{60}=0.1$ is obscured by T=18 K; $\tau_{60}=1$ dust) exceeds the plotted 400 $\mu$m upper limit (arrow in Fig. 4; 0.39 mJy 3$\sigma$; Bertoldi et al. 2003a) by a factor of 1.9 and is unphysical in a number of ways including that the cool component is cooler than the CMB at $z=6.42$. The ice models distinguish themselves from the plotted refractory dust model by being fainter by at least a factor of 4 at rest wavelength 400 $\mu$m (3 mm observed frame). The present upper limit does not distinguish these cases. Such a measurement should be possible using CARMA. Determining the composition of the emitting material can also affect estimates of the dust mass and thus estimates of the amount of past enrichment (and star formation) in the QSO J1148+5251 system. For the present ice model, 7 times as much refractory dust would be needed to produce the same FIR opacity so that a lower level of total enrichment might be accommodated. Finally, at high redshift, where theory suggests that we might most anticipate it owing to expectations about early ISM enrichment, we find evidence for FIR ice emission which, while still debatable, seems moderately persuasive and which can be relatively easily checked with further observations using existing telescopes. Now we may turn to our main theme. We have employed amorphous ice to consider IRAS 14348-1447 and QSO J1148+5251, but now we consider crystalline ice both because its effect on redshift estimation is the most dramatic and because ice may form at fairly warm temperatures in the SNRs associated with pair instability SNe. ### 3.3 Ice emission in 450 $\mu$m submillimeter sources A set of sources that could be at high redshift and which demonstrate the effects of ice are reported by Pope et al. (2005). They are detected at 450 $\mu$m but not at 850 $\mu$m. We consider the 7 sources tabulated by Pope et al. (2005) and two sources from the original list of Borys et al. (2003) which were recovered above 3.5 $\sigma$ by Pope et al. (2005). These have drawn our attention because the reported 850 $\mu$m upper limits indicate large values of $\beta$ (Borys et al. 2004; based on original list). Large values of $\beta$ can indicate the presence of ice emission. If we generate 3$\sigma$ lower limits to the 450-to-850 $\mu$m flux density ratios using the 450 $\mu$m measurements less 3/$\sqrt{(}2)\times\sigma_{450}$ together with 3/$\sqrt{(}2)\times\sigma_{850}$ 850 $\mu$m upper limits then the lowest lower limit is 8.7 and is only one of two that can be consistent with $\beta<2$ dust. This method of calculating limits can be more conservative than calculating the ratio of the 450 measurement to the 850 $\mu$m 3$\sigma$ upper limit (giving 13.4 in this case) but its extension to 2$\sigma$ should be viewed with caution since detections at the 2.9 $\sigma$ level are customarily reported as 3$\sigma$ limits. In such situations, when data with comparable signal-to-noise ratios become available, the “2$\sigma$ extension” may be violated much more frequently than the name implies if many of these near misses lurk in the original data. However, Pope et al. (2005) report no detection from applying the method of Caillault & Helfand (1985) (stacking) to the 850 $\mu$m data at the positions of the 450 $\mu$m sources they detect above 3.5$\sigma$. In five cases where upper limits are calculated below we substitute the ratio of the 450 $\mu$m 3$\sigma$ upper limit-to-the 850 measurement when it gives a more conservative value. In Fig. 5 we plot, as a function of redshift, the 450-to-850 $\mu$m flux density ratio calculated for a series of SEDs. The dashed line is for a model with T=60 K, $\beta$=2, and $\tau_{90}$=2. On this line are plotted at $z=0.9$ and $z=0.5$ the 3$\sigma$ lower limits (filled circle and diamond) for SMM J123603+620942 and J123631+620657. The arrows in the figure extend to the 2$\sigma$ lower limits. In these cases the argument given by Borys et al. (2003) for $z<1$ may be applied directly, though tenuously, at the 3$\sigma$ level with no assumptions about ice but this is not the case for the 7 remaining lower limits. There may be some difficulty however accommodating these two sources at these low redshifts unless sources with 0.2–2 Jy level flux densities are present at 70 $\mu$m while Frayer et al. (2006) find about four $\sim 30$ mJy sources at 70 $\mu$m in this region and we are unaware of any IRAS sources in this field. At these low redshifts, it is a losing proposition to attempt to increase the dust temperature to increase the redshift to avoid the 70 $\mu$m constraint in an effort to avoid requiring ice emission. When ice is invoked, then these and the remaining lower limits may all be accommodated but the range of consistent redshifts expands. The remaining curves in Fig. 5 correspond to increasing ice temperature. The dot- dashed curve uses an SED similar to that shown in Fig. 1 (T=30 K), the triple- dot-dashed curve has T=50 K as in Fig. 7., and the solid curve has T=150 K; $\tau_{90}=0.04$ and Vm/Vc=5.6 for all ice models. For the solid curve, ambiguity between amorphous and crystalline ice is physically resolved in favor of crystalline ice. The SMM J123603+620942 and J123631+620657 limits are replotted on the solid curve giving a redshift limit $z<10.2$. This limit is little changed if the maximum possible ice temperature (170 K) is used. All the 3$\sigma$ limits can be accommodated at $z=9.5$ but only SMM J123603+620942 and J123631+620657 can also be accommodated there at the 2$\sigma$ level. SMM J123657+622033 (filled square) shows this clearly where it is plotted at $z=9.6$. The 2$\sigma$ level for SMM J133747+621600 (filled clover leaf) also constrains the ice temperature to be T$\gtrapprox$40 K. It is plotted on the T=30 K curve at $z=1.9$. The remaining sources other than SMM J123727+621042 are plotted with redshift upper limits well within the epoch of reionization. However, as with sources plotted on the T = 30 K curve, the 2$\sigma$ limits cannot be accommodated for $z\gtrapprox 2.5$. That 7 of 9 of the 450 $\mu$m sources in the GOODS-N field require ice to explain the 850 $\mu$m non-detections without recourse to any other constraint suggests strongly that the possible effects of ice emission must be considered when interpreting submillimeter observations. Ice dramatically changes redshift limits: by a factor of 20 for SMM 123631+620657. From the other direction, for the sources with 850 $\mu$m detections and 450 $\mu$m upper limits that also have estimated redshifts (Borys et al. 2004) we may also begin to constrain the presence of ice. 14 upper limits are plotted in Fig. 5 as open circles for sources where a redshift estimate is given. The open diamonds are also upper limits plotted (but without arrows to avoid crowding) at the lower limits to their redshift given by Borys et al. (2004). For redshift estimates with $z<3$ the upper limits are constraining on the likelihood that optically thin ice contributes to the FIR opacity: at the 2$\sigma$ level (arrow heads) little T$>$30 K optically thin ice seems to be needed. At the 3$\sigma$ level 4 of 11 sources could be consistent with a contribution of optically thin ice emission to the source SEDs. One source with a redshift lower limit ($z>2.7$; SMM J123652+621225) needs to be optically thick and cool (T $\sim$40 K) or at a higher redshift as already noted by Borys et al. (2004). The remaining sources with with redshift estimates above 3 or redshift lower limits are not constraining on the presence of ice. The present considerations suggest that the 450 and ($z<3$) 850 $\mu$m detected sources differ physically in that the former have optically thin ice emission that dominates their FIR opacity while the latter may be optically thick at 850/(1+$z$) $\mu$m or that refractory dust dominates their FIR opacity or both. The average 450-to-850 $\mu$m ratio for the 850 $\mu$m source calculated from the stacked 450 $\mu$m and the average of the 850 $\mu$m flux densities reported by Pope et al. (2005) is 0.5 which we have plotted as a small open square at $z=4.25$. The plotted redshift would be consistent with T=30 optically thick material or T=20 optically thin ice or $\beta=2$ dust. If it is true that 850 $\mu$m sources are on average fainter at 450 $\mu$m than at 850 $\mu$m then it seems at though many of the 850 $\mu$m sources must lie at $z>4$ and ice may be needed to make them detectable at 850 $\mu$m with ULIRG luminosities (see end of Sec. 4.2). We do worry, however, that the stacked 450 $\mu$m flux is underestimated owing to the smaller 450 $\mu$m beam or that the data were normalized prior to stacking making the reported units ambiguous. QSO J1148-5251 is also plotted (small filled circle). As discussed above, it seems to require ice emission. Just as detected 850 $\mu$m sources must be at least ultraluminous, so also must the detected 450 $\mu$m sources be hyperluminous (L$>10^{13}$ L⊙). In Fig. 6 we plot the models shown in Fig. 5 giving their 450 $\mu$m brightness as a function of redshift under the following assumptions about luminosity: The models for refractory dust, and ice at T=30 and 50 K are ten times more luminous than ULIRG IRAS 00188-0856 (Fig. 1), the model for T=150 K ice is 10 times more luminous still, and we have also plotted the complex amorphous ice model for QSO J1148+5251 at it’s observed luminosity (short dashed line and filled circle; $\sim 10^{13}$ L⊙; Beelen et al. 2006). The brightnesses of the 450 $\mu$m sources given by Pope et al. (2005) lie in the range 77-to-291 mJy so that the plotted models would need to be scaled up by a factor of a few to cover this range fully unless the sources are at $z<0.2$. For example, one of the brighter sources (SMM J123631+620657; 263 mJy) would have a luminosity of 6$\times 10^{14}$ L⊙ for the T = 150 K ice model if located at $z=9.5$, a factor of three higher than the most luminous hyperluminous source listed by Verma et al. (2002). It’s luminosity would be 10 times lower for the T = 30 K ice model located at $z=2$, near the middle of the distribution of luminosities in their list. From Fig. 5 we estimate that the 450 $\mu$m sources would populate the redshift ranges 1–3 and 5–10, with the latter range having a larger comoving volume by a factor of 1.7. If ice is only important close to reionization so that the sources were all at 9.2$<z<10.2$ (see solid line in Fig. 6) then the luminosities require about a tenth of the stellar mass of M87 per source to be involved in star formation. The space density would be 2.5$\times 10^{-5}$ Mpc-3, comparable to the local density of clusters of galaxies (Bahcall et al. 2003) and the luminosity density would be $\sim 5\times 10^{9}$ L⊙/Mpc3 (comoving), about a factor of 100 higher than the local FIR value (Saunders et al. 1990). This would then imply a minimum of 103 recombinations per proton if the 450 $\mu$m luminosity traces a factor of ten smaller ionizing luminosity: about 102 more recombinations than for the lifetime of a typical H II region which seems too high for reionization unless dust competes efficiently for reionizing photons. On the other hand, there are six 850 $\mu$m sources with flux densities above 20 mJy listed by Pope et al. (2005) which could be close to hyperluminous (see Fig. 7) so that we need only postulate a similar population which is (thus far) invisible a 850 $\mu$m owing to the effects of ice and thus expect to find these sources at a range of redshifts. There is very clear evidence of ice rather than dust providing the FIR opacity in 7 of the 9 450 $\mu$m sources considered here. But, the redshifts may or may not be high and additional photometry is needed to constrain the redshifts. ## 4 HDF 850.1 and ice ### 4.1 $z=5$ The search for counterparts at other wavelengths to the brightest 850 $\mu$m source in the Hubble Deep Field has been arduous but Dunlop et al. (2004) make a convincing case for detections in the radio and near infrared (NIR) and summarize submillimeter observations. Modeling (Aretxaga et al. 2003) of these observations leads them to conclude that the redshift of HDF 850.1 lies in the range 4.1$\pm$0.5. Since the SED of a high $q$ ULIRG (IRAS 08572+3915) is included in the 20 templates in the model input, the effects of the change in the radio-FIR relation for ULIRGs compared to lower luminosity galaxies noted in the introduction should be partly represented in their estimate. Similarly, the suggestion of ice emission in the SED of IRAS 14348-1447 (Fig. 3) could also influence the estimate if the photometric data were used directly rather than fitted with a modified blackbody. In general, reliance on modified blackbodies will give lower estimated redshifts for cooler sources when more than one point on the Rayleigh-Jeans curve is available, as is the case for HDF 850.1, but the addition of an ice emission feature can increase the estimated redshift even for a cool source. As one example, the bow-tie shown in Fig. 1 on the FIR continuum corresponds to the range of redshift one would calculate for HDF 850.1 based on two estimates of its 850 $\mu$m flux density (Dunlop et al. 2004; Wang et al. 2004) and the 1.3 mm flux density reported by Downes et al. (1999). That is 5.4 $<z<$5.6. It should be noted, even before describing the model upon which this is based, that a detailed comparison between the SED shown in Fig. 1 and the optical and NIR limits and detections for HDF 850.1 is not possible owing to insufficient wavelength coverage for IRAS 00188-0856. See Dunlop et al. (2004) fig. 6 for a comparison with data from two ULIRGs and a luminous infrared galaxy calculated for a range of redshifts. However, the effects of far infrared ice emission can be demonstrated with this example. The model, the short-dashed line in the lower panel of Fig. 1, is based on curves given by Aannestad (1975) which are the results of calculations of the optical properties of ice mantled crystalline silicate grains. And, it is crystalline water ice in the mantles. We chose the mantle-to-core volume ratio = 5.6 curve to most closely match the ratio of 3 $\mu$m ice to silicate optical depth used in the model shown by the solid lines in Fig. 1. For a number of reasons this choice is probably not fully constrained by this ratio: the optical depth to which the silicate absorption is sensitive may be larger than for ice; an odd shift in the wavelength of the silicate feature may indicate a pyroxene composition with a different feature-to-continuum optical depth ratio than usual; the crystalline silicate in the Aannestad (1975) calculation is somewhat dissimilar to astrophysical silicate models. However the first two effects may be compensating so that as a rough abundance analysis, the expectation that ice would dominate the FIR opacity in this source seems well founded. This condition, and the need for the ice emission to be moderately optically thin are really the only two requirements for ice to strongly affect redshift estimation. The optical depth of the model of far infrared emission shown in the lower panel of Fig. 1 is set to correspond to the 3 $\mu$m ice optical depth used in models shown by the solid line in the upper panel (where ice and refractory material contribute equally to the absorption optical depth at 3.1 $\mu$m). The temperature of the far infrared emission model is set to roughly reproduce the IRAS 60 and 100 $\mu$m measurements. This temperature also accentuates the effects of the broad 150 $\mu$m feature over which the HDF 850.1 data are displayed. The range of redshifts needed to place HDF 850.1 on this model falls outside of the range estimated by Dunlop et al. (2004) and gives a first indication that the effects of ice on photometric redshift estimation (as opposed to just limits) can be significant. ### 4.2 $z=13$ As noted in the introduction, using local ULIRGs to understand submillimeter sources is a sensible approach if their properties are not strongly affected by cosmic evolution. However, at higher redshift, this proviso seems fairly likely to break down. Many local ULIRGs appear to be produced in the mergers of gas rich spiral galaxies and it has been argued (Mihos & Hernquist 1996) that the dynamical stability provided by bulges may preserve gas reservoirs for the final merger stages associated with some ULIRGs. Owing to the finite lifetime of the universe, there are earlier periods when the precursor systems may not resemble present day precursors in their dynamical state or star-to- gas mass ratio. It is not at all clear that seed SMBHs (Osterbrock 1993) would be present at sufficient mass to allow accretion to be an energetically important or dominant power source as suspected for local ULIRGs (e.g. Imanishi et al. 2006a). Thus, in what follows we assume star formation as the main power source at higher redshift. A sketch of how crystalline ice might exist in a primordial galaxy environment is also given to explore the plausibility of a very large effect of ice on redshift estimation. It is to be noted that giving up a central power source such as would be provided by an AGN would usually mean that ice mantled grains would contribute little to the FIR emission since most of the energy that powers FIR emission would be absorbed in reactive photodissociation regions (PDRs) rather than in molecular clouds where ice is thought to exist unless an improbable scenario involving an overabundance of protostars such as that proposed by Roussel et al. (2003) but previously discounted by Dudley & Wynn-Williams (1997) is invoked. Below we argue for such an overabundance. In Fig. 7 we show the observations of HDF 850.1 at $z=12.6$ This is 18 confidence intervals beyond the Dunlop et al. (2004) estimate. So far as we can tell, neither the new $Z$ band upper limit reported here, nor the released Spitzer IRAC and MIPS GOODS images (Dickinson et al. 2006) rule out either the model shown in Fig. 7 or the analysis of Dunlop et al. (2004) so long as a reddish ULIRG (see Dudley (1999) for a discussion of 12-to-60 $\mu$m flux density ratios in ULIRGs) is used as a basis for the Dunlop et al. (2004) analysis. Dunlop et al. (2004) argue that the lensing source (solid diamonds in Fig. 7) is evolved, in which case much of the tentative observed frame 24 $\mu$m flux density could be owing to HDF 850.1 but little of the 8 $\mu$m flux density. At $z=4.1$, Mkn 231 and 273 could be too blue to account for this level of emission whereas Arp 220 or IRAS 17208-0014 would not violate this datum if treated as an upper limit. On the other hand, at $z=1.1$ the chances of finding a younger elliptical galaxy to act as a lens are increased so that some or all of the observed 24 $\mu$m flux density could be owing to the lensing galaxy. This would still not rule out the Dunlop et al. (2004) redshift estimate for HDF 850.1 but would accommodate the present estimate under the assumption of the plotted starburst model. The shaded region in Fig. 7 shows a possible range for the spectral shape of the lens source emission. The source labeled 1b in fig. 8 of Wang et al. (2004) may also contribute to the 24 $\mu$m emission though only slightly to the IRAC data. If it is similar to NGC 253, and at a redshift of $z_{\rm ph}=1.76$ (Fernández-Soto et al. 1999) then up to $\sim$30% of the 24 $\mu$m flux density might be attributed to this source but again negligibly at 850 $\mu$m. The model for HDF 850.1 shown in Fig. 7 (thick gray line) is similar to that shown in Fig. 1 in that it has the same mantle-to-core volume ratio and optical depth however the temperature is higher as it must be owing to the higher CMB temperature at $z=12.6$. Whereas in Fig. 1, the 150 $\mu$m ice feature spectral structure influences the redshift placement of the HDF 850.1 data, in Fig. 7, shorter wavelength spectral structure also owing to ice is influential. Interestingly, assuming that there is a foreground neutral intergalactic medium as a result of incomplete reionization (Spergel et al. 2006), one would expect Gunn-Peterson saturation or a Ly$\alpha$ blackout to occur in the $H$ band for $z=12.6$ the shortest wavelength band for which HDF 850.1 is detected. This is represented by the change from black to gray in the thin line using a Heaviside function as multiplier. Here the thin line is a young starburst spectrum generated using Starburst99 (Leitherer et al. 1999). It is notable that taking the onset in redshift of the Gunn-Peterson trough to be $\sim$6 (Fan et al. 2004) absorption at the Lyman limit at $z<12.6$ easily overlaps Ly$\alpha$ absorption between $z=6$ and $z=9.2$. One expects pretty thorough opacity shortward of Ly$\alpha$. While the choices of parameters of the starburst model are meant only to be suggestive, for the model shown in Fig. 7, the free-free emission expected based on the rate of production of ionizing photons would explain about 20% of the observed radio emission (following Scoville et al. 1991) while retaining the starburst model as it is and correcting for sufficient extinction to account for the model infrared emission would over produce the radio emission so that a lower contribution of stars which produce ionizing photons to the observed frame $K$’ band flux might provide a more likely scenario. Alternatively, invoking a chemical trigger for star formation discussed below, it is conceivable that an object could be caught in a phase when a sufficient number of stars are in an ultracompact H II region configuration that, given a sufficiently high ionization parameter, higher than usually required if the gas-to-dust mass ratio is high owing to low enrichment, free-free emission would be suppressed (e.g. Bottorff et al. 1998). As noted in the introduction, sources at high redshift should eventually drop out of 850 $\mu$m surveys. In Fig. 8 we plot, for the same far infrared luminosity, the expected 850 $\mu$m brightness as a function of redshift for an Arp 220-like SED and the two distributions shown in Figs. 1 and 7\. The 30 K model is truncated at $z=11$ since the cosmic background radiation temperature exceeds the model temperature above this redshift. As can be seen, the models with ice select for high redshift sources assuming such sources exist. For the 30 K model the brightest (non-local) source would be found near a redshift of 5 and for the 50 K model near a redshift of 13. Should ice play the role proposed here in a moderate fraction of the detected 850 $\mu$m sources, it would be of little surprise that the brightest source in the HDF (850.1) would be among them. ### 4.3 Why star formation? Our choice of presenting a model that uses star formation as a power source may seem strange given that PDRs would be inefficient at producing ice emission which we require to allow, for example, HDF 850.1 to be detected at $z=13$. One phase of star formation that does allow MIR radiation to heat ice is the protostar phase. This phase is brief compared to O star life times and thus would not typically be expected to provide the dominant source of FIR radiation in a galaxy. Protogalaxies, however, are by definition required to be young, so that while the argument concerning the relative brevity of the protostar phase cannot be ignored, it may be overcome if the star formation rate is increasing rapidly with time. And, it is just such sources that one might expect to be observed as they populate the top of the infrared luminosity function since they would be the most efficient at converting radiation to FIR wavelengths. A rapidly increasing star formation rate might be expected in a situation where the gas cooling efficiency is increasing owing to ongoing initial enrichment. A positive feedback can lead to an exponentially increasing system response. The number at the right hand of Fig. 8 shows the 850 brightness (in mJy) of the three SED shapes considered here if we adopt the lensing amplification proposed by Dunlop et al. (2004) for HDF 850.1 for a source with the FIR luminosity of IRAS 00188-0856. The increase in luminosity required to match the Hughes et al. (1998) estimate for HDF 850.1 would be a factor of 2.3 for the dashed curve taken at a redshift of 13\. This would then correspond to $\sim 4\times 10^{12}$ L⊙ (e.g. Fig. 7) or 107 O stars and 1.5$\times 10^{11}$ G0 stars for a Salpeter initial mass function (m-2.35), or roughly a galaxy’s worth of star formation. This then would be the limit for the continued exponential increase in the star formation rate: the amount of available enrichable gas. Night et al. (2005) find 4 objects with $10^{11}$ M⊙ in stars at $z=6$ in their largest cosmological simulation (142 Mpc; large enough to contain $\sim$1 ULIRG at $z=0$). Thus, there is some suggestion that relatively massive objects (or perhaps regions that will become objects) are forming their stars a few dynamical timescales earlier than this. Notwithstanding the existence of QSO J1148+5251 and other high redshift quasars, it seems to us that there is a need to account for massive sources at intermediate redshift which last formed stars near $z=13$ (Stockton et al. 2004; Jimenez 2000) as well as the apparent monotonic increase in the comoving rate of gamma-ray bursts out to $z=7$ (Le & Dermer 2007). However, in the low-to-moderate redshift universe, we would usually take ice emission as an indication of a buried compact power source at least at ULIRG luminosities, and, should blackhole growth commonly reach $\sim 10^{7}$ M⊙ by $z=13$, as must happen at least occasionally given the existence of QSO J1148+5251, then our low-to-moderate redshift explanation could also apply during reionization. ## 5 Summary and prospects Far infrared ice emission may be crucially important to understanding the far infrared emission of sources at high redshift and may strongly affect the range of redshifts estimated for submillimeter sources, even placing them at reionization redshifts. Observations of infrared ice emission are definitive in Galactic protostars and the winds of evolved stars but are scant and not fully persuasive for the local analogs of the submillimeter sources, the ULIRGs. This, despite mounting evidence that ice is involved in absorbing much of the energy that is finally emitted in the far infrared in a substantial fraction of ULIRGs. We find stronger indications that ice emission is important at high redshift. The data for QSO J1148+5251 as well as the 450 $\mu$m sources cataloged by Pope et al. (2005) are both strongly suggestive. We feel that further investigation along these lines could be quite fruitful. We’ve considered two models where the presence of far-infrared emission features owing to ice in the SED of HDF 850.1 would place it at redshifts larger than the range estimated without ice by Dunlop et al. (2004). In the first, the use of crystalline ice at 150 $\mu$m as a substitute for amorphous ice is probably justified, subject to further laboratory investigations of amorphous ice. In the second, consideration of a first chemical enrichment mechanism suggests that crystalline ice could be present at early times while energetic and timescale considerations may be compatible with protostars in protogalaxies producing the sort of features we consider here. In the second model, a Lyman $\alpha$ blackout would be consistent with the NIR photometry for HDF 850.1. However, we do not prefer this model and its higher redshift over the redshift range estimated by Dunlop et al. (2004) or $z\sim 5.5$ from our first model. What we find to be important is that far infrared ice emission could have such a dramatic effect on the estimated redshift. It should be noted that redshifts estimated from the radio-FIR relation could also turn out to be lower when ice emission is invoked since a large submillimeter-to-radio flux density ratio is easier to produce with high $\beta$ at some redshifts. So, the main conclusion of this work is that ice broadens the range of redshifts to be considered for a given submillimeter source and that ice is quite likely to be important at high redshift. Some astronomical observations which could aid progress on open issues are: * • Continuum observations at 3 mm of QSO J1148+5251 may confirm ice emission in this source. * • The Atacama Large Millimeter Array may be able to isolate the transition from dust to free-free/synchrotron emission for sources of this type and allow more reliable redshift estimates. * • Investigating the role of ice emission locally could be taken up using SOFIA or Spitzer. * • Very low resolution H-band spectroscopy might test the possibility of a Lyman $\alpha$ blackout for HDF 850.1. * • Molecular studies which target H2O and OH maser emission may give definite redshifts using a new generation of radio telescope arrays even if the continuum is below detection limits. Colleagues have commented helpfully at various stages of this work. Some are: S. Chubb, E. Dwek, J. Fischer, P. Gerakines, A. Li, M. Moore, L. J. Rickard, E. Shettle, and H. Smith and an anonymous referee. Research in infrared astronomy and particularly the astrophysics of celestial backgrounds at the Naval Research Laboratory is supported by the Office of Naval Research (USA). This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation. Use is also made of the USNOFS Image and Catalog Archive operated by the United States Naval Observatory, the archive of results from the ESA sponsored Infrared Space Observatory mission, the NASA/IPAC Extragalactic Database operated by JPL and Caltech, and data from the GOODS project from NASA’s Spitzer and Hubble Space Telescopes. ## References * Aannestad (1975) Aannestad, P. A. 1975, ApJ, 200, 30 * Aretxaga et al. (2003) Aretxaga I., Hughes D. H., Chapin E. L., Gaztanaga E., Dunlop J. S., 2003, MNRAS, 342, 759 * Athey et al. (2002) Athey, A., Bregman, J., Bregman, J., Temi, P., 2002, ApJ, 571, 272 * Bahcall et al. (2003) Bahcall, N., et al., 2003, ApJS, 148, 243 * Beelen et al. (2006) Beelen, A., Cox, P., Benford, D. J., Dowell, C. D., Kovács, A., Bertoldi, F., Omont, A., Carilli, C. L., 2006, ApJ, 642, 694 * Bertie et. al (1969) Bertie, J. E., Labbé, H. J. & Whalley, E. 1969, J Chem Phys, 50, 4501 * Bertoldi et al. (2003a) Bertoldi, F. et al. 2003, A&A, 409, L47 * Bertoldi et al. (2003b) Bertoldi, F., Carilli, C. L., Cox, P., Fan, X., Strauss, M. A., Beelen, A., Omont, A., Zylka, R., 2003, A&A, 406, L55 * Borys et al. (2003) Borys, C., Chapman, S., Halpern, M., Scott, D., 2003, MNRAS, 344, 385 * Borys et al. (2004) Borys, C., Scott, D., Chapman, S., Halpern, M., Nandra, K., Pope, A., 2004, MNRAS, 355, 485 * Bottorff et al. (1998) Bottorff, M., Lamothe, J. Momjian, E. Verner, E. Vinković, D. & Ferland, G. 1998, PASP, 110, 1040 * Brinchmann et al. (1998) Brinchmann, J. et al. 1998, ApJ, 499, 112 * Caillault & Helfand (1985) Caillault, J.-P., Helfand, D. J., 1985, ApJ, 289, 279 * Charmandaris et al. (2002) Charmandaris, V. et al. 2002, A&A, 391, 429 * Condon et al. (1991) Condon J. J., Huang Z.-P., Yin Q. F., Thuan, T. X., 1991, ApJ, 378, 65 * Curtis et al. (2005) Curtis, D. B., Rajaram, B., Toon, O. B., Tolbert, M. A., 2005, Applied Optics, 44, 4102 * Cutri et al. (2003) Cutri R. M., Skrutskie M. F., Van Dyk S., et al. 2003, 2MASS All Sky Data Release * Dartois et al. (1998) Dartois, E., et al. 1998, A&A, 338, L21 * Dickinson et al. (2006) Dickinson, M. et al., in prep. * Downes et al. (1999) Downes D. et al., 1999, A&A, 347, 809 * Dudley & Wynn-Williams (1997) Dudley. C. C. & Wynn-Williams, C. G. 1997, ApJ, 488, 720 * Dudley (1999) Dudley C. C., 1999, MNRAS, 304, 549 * Dunlop et al. (2004) Dunlop, J. S. et al. 2004, MNRAS, 350, 769 * Dunne et. al (2000) Dunne, L., Eales, S., Edmunds, M., Ivison, R., Alexander, P., Clements, D. L. 2000, MNRAS. 315, 115 * Elvis et al. (1994) Elvis, M., Wilkes, B. J., McDowell, J. C., Green, R. F., Bechtold, J., Willner, S. P., 1994, ApJS, 95, 1 * Fan et al. (2003) Fan, X., et al. 2004, AJ, 125, 1649 * Fan et al. (2004) Fan, X., et al., 2004, AJ, 128, 515 * Fernández-Soto et al. (1999) Fernández-Soto, A., Lanzetta, K. M., Yahil, A., 1999, ApJ, 513, 34 * Frayer et al. (2006) Frayer, D. T., et al., 2006, ApJ, 647, L9 * Geballe et al. (2006) Geballe, T. R., Goto, M., Usuda, T., Oka, T., McCall, B. J. 2006, ApJ, 644, 907 * Genzel et al. (1998) Genzel, R. et al. 1998, ApJ, 498, 579 * Gerakines et al. (2005) Gerakines P. A., Bray J. J., Davis A., Richey C. R., 2005, ApJ, 620, 1140 * Giavalisco et al. (2004) Giavalisco M. et al., 2004, ApJ, 600, L93 * González-Alfonso et al. (2004) González-Alfonso, E., Smith, H. A., Fischer, J., Cernicharo, J., 2004, ApJ, 613, 247 * Greenberg et al. (1995) Greenberg, J. M., Li, A. Mendoza-Gomez, C. X., Schutte, W. A., Gerakines, P. A. & de Groot, M. 1995, ApJ, 455, L 177 * Hagen et al. (1983) Hagen, W., Greenberg, J. M. & Tielens, A. G. G. M. 1983, A&AS, 51, 389 * Heger & Woosley (2002) Heger, A. & Woosley, S. E. 2002, ApJ, 567, 532 * Helou et al. (1985) Helou, G., Soifer, B. T., Rowan-Robinson, M., 1985, ApJ, 298, L7 * Hoogzaad et al. (2002) Hoogzaad S. N., Molster E. F., Dominik C., Waters L. B. F. M., Barlow M. J., de Koter A., 2002, A&A, 389, 547 * Huchra (1977) Huchra, J. P., 1977, ApJS, 35, 171 * Hughes et al. (1998) Hughes D. H. et al., 1998, Nature, 394, 241 * Imanishi & Dudley (2000) Imanishi, M., Dudley, C. C., 2000, ApJ, 545, 701 * Imanishi et al. (2001) Imanishi, M., Dudley, C. C., Maloney, P. R., 2001, ApJ, 558, L93 * Imanishi & Maloney (2003) Imanishi, M., Maloney, P. R., 2003, ApJ, 588, 165 * Imanishi et al. (2006a) Imanishi, M., Dudley, C. C., Maloney, P. R., 2006, ApJ, 367, 114 * Imanishi (2006) Imanishi, M., 2006, AJ, 131, 2406 * Imanishi et al. (2006b) Imanishi, M., Dudley, C. C., Maiolino, R., Maloney, P. R., Nakagawa, T., Risaliti, G. 2006, ApJ, submitted * Jimenez (2000) Jimenez et al. 2000, ApJ, 532, 152 * Kim & Sanders (1998) Kim, D.-C. & Sanders, D. B. 1998, ApJS, 119, 41 * Kim, Veilleux & Sanders (2002) Kim D.-C., Veilleux S., Sanders D. B. 2002, ApJS, 143, 277 * Klaas et al. (2000) Klaas, U., Laureijs, R. J., Radovich, M., Schulz, B., Wilke, K., 2000, ISOPHOT Calibration Accuracies, SAI/1998-092/Dc, Ver. 4 * Klaas et al. (2001) Klaas, U. et al. 2001, A&A, 379, 823 * Le & Dermer (2007) Le, T., Dermer, C. D., 2006, ApJ, 661, 394L * Leitherer et al. (1999) Leitherer, C. et al. 1999, ApJS, 123, 3 * Lis et al. (1998) Lis, D. C., Serabyn, E., Keene, J., Dowell, C. D., Benford, D. J., Phillips, T. G., Hunter, T. R. & Wang, N. 1998, ApJ, 509, 299 * Lisenfeld et. al (2000) Lisenfeld, U., Isaak, K. G., Hills, R. 2000, MNRAS, 312, 433 * Maloney & Reynolds (2000) Maloney, P. R. & Reynolds, C. S. 2000, ApJ, 545, 23 * McCutcheon et al. (1995) McCutcheon, W. H., Sato, T. Purton, C. R. Matthews, H. E. & Dewdney, P. E. 1995, AJ, 110, 1762 * Mihos & Hernquist (1996) Mihos, J. C., Hernquist, L., 1996, ApJ, 464, 64 * Monet et al. (1998) Monet D., Canzian B., Harris H., Reid N., Rhodes A., Sell S. 1998, US Naval Obs. Flagstaff Station 1977, 1243 * Moore & Hudson (1992) Moore, M. H. & Hudson, R. L. 1992, ApJ, 401, 353 * Moshir et al. (1990) Moshir M. et al. RAS Faint Source Catalogue, version 2.0 * Nenkova et al. (2002) Nenkova,M., Ivezić, Z., Elitzur, M., 2002, ApJ, 570, L9 * Nolta et al. (2004) Nolta, M. R. et al. 2004, ApJ, 608, 10 * Nozawa et al. (2003) Nozawa, T., Kozasa, T., Umeda, H., Maeda, K., Nomoto, K., 2003, ApJ, 598, 785 * Osterbrock (1993) Osterbrock, D. E., 1993, ApJ, 404, 551 * Persson et al. (1979) Persson, S. E., Frogel, J. A., Aaronson, M., 1979, ApJS, 39, 61 * Pope et al. (2005) Pope, A., Borys, C., Scott, D. Conselice, C., Dickinson, M., Mobasher, B., 2005, MNRAS, 358, 149 * Riess et al. (2004) Riess, A. G. et al. 2004, ApJ, 607, 665 * Rigopoulou et al. (1996) Rigopoulou, D., Lawrence, A., Rowan-Robinson, M. 1996, MNRAS, 278, 1049 * Risaliti et al. (2006) Risaliti, G., et al., 2006, MNRAS, 365, 303 * Robson et al. (2004) Robson, I., Priddey, R. S., Isaak, K. G., McMahon, R. G., 2004, MNRAS, 351, L29 * Roche & Aitken (1984) Roche, P. F.,Aitken, D. K.., 1984, MNRAS, 208, 481 * Roussel et al. (2003) Roussel, H., Helou, G., Beck, R., Condon, J. J., Bosma, A., Matthews, K., Jarrett, T. H., 2003, ApJ, 593, 733 * Sanders (2000) Sanders, D. B. 2000, AdSpR, 25, 2251 * Sanders et al. (1995) Sander D. B., Egami E., Lipari S., Mirabel I. F., Soifer, B. T., 1995, AJ, 110, 1993 * Sanders & Mirabel (1996) Sanders, D. B. & Mirabel, I. F. 1996, ARAA, 37, 749 * Saunders et al. (1990) Saunders, W., Rowan-Robinson, M., Lawrence, A., Efstathiou, G., Kaiser, N., Ellis, R. S., Frenk, C. S., 1990, MNRAS, 242, 318 * Schober et al. (2000) Schober, H., Koza, M. M., Tölle, A., Masciovecchio, C., Sette, F., Fujara, F. 2000, PhRvL, 85, 4100 * Schneider et al. (2004) Schneider, R., Ferrara, A., Salvaterra, R., 2004, MNRAS, 351, 1379 * Scoville et al. (1991) Scoville, N. Z. et al. 1991, ApJ, 366, L8 * Soifer et al. (1987) Soifer B. T. et al., 1987, ApJ, 320, 238 * Soifer et al. (1999) Soifer, B. T., Neugebauer, G., Matthews, Becklin, E. E., Ressler, M., K., Werner, M. W., Weinberger, A. J., Egami, E., 1999, AJ, 118, 2065 * Soifer et al. (2002) Soifer, B. T., Neugebauer, G., Matthews, K., Egami, E., Weinberger, A. J., 2002, AJ, 124, 2980 * Smith et al. (1994) Smith, R. G. et al. 1994, MNRAS, 271, 481 * Spergel et al. (2006) Spergel, D. N. et al. 2006, ApJ, submitted * Spergel et al. (2003) Spergel, D. N. et al. 2003, ApJS, 148, 175 * Spoon et al. (2002) Spoon, H. W. W. et al. 2002, A&A, 414, 873 * Spoon et al. (2006) Spoon, H. W. W., et al., 2006, ApJ, 638, 759 * Stockton et al. (2004) Stockton, A., Canalizo, G., Maihara, T. 2004, ApJ, 605, 37 * Temi et al. (2005) Temi, P., Mathews, W. G., Fabrizio, B., 2005, ApJ, 622, 235 * Verma et al. (2002) Verma, A., Rowan-Robinson, M., McMahon, R., Efstathiou, A., 2002, MNRAS, 335, 574 * Wang et al. (2004) Wang, W.-H., Cowie, L. L., Barger, A. J. 2004, ApJ, 613, 655 * Weedman et al. (2005) Weedman, D. W., et al. 2005, ApJ, 633, 706 * Willot et al. (2003) Willott, C. J., McLure, R. J., Jarvis, M. J., 2003, ApJ, 587, L15 * Whittet et al. (1988) Whittet, D. C. B., Bode, M. F., Longmore, A. J., Adamson, A. J., McFadzean, A. D., Aitken, D. K., Roche, P. F., 1988, MNRAS, 233, 321 * Xilouris et al. (2004) Xilouris, E. M., Madden, S. C., Galliano, F., Vigroux, L., Sauvage, M., 2004, A&A, 416, 41 * Yun et al. (2001) Yun M. S., Reddy N. A., Condon J. J., 2001, ApJ, 554, 803 Figure 1: Upper Panel: Data from Imanishi & Maloney (2003) is reproduced for IRAS 00188-0856 together with a model (solid line) subsequently employed in the lower panel. The model consists of an $\alpha=2$ (flux density $f_{\nu}\propto\nu^{-\alpha}$) power-law continuum to which absorption is applied together with an unabsorbed component which is a combination of an archival spectrum of M 82 (ISO SWAA11600319) and a blackbody of temperature 2500 K to represent additional stellar photospheric emission (displayed in the lower panel long-dashed line). The two absorption components used are: (1) a volatile admixture (H2O:CO=2:1) of amorphous ice measured at 10 K (Hagen et al. 1983) and (2) the refractory EURECA-B curve presented by Greenberg et al. (1995). The effects of absorption by the refractory material are shown with the dot-dashed line and those of the combined volatile and refractory absorption with the dashed line. The volatile and refractory components have approximately equal contributions to the absorption at 3 $\mu$m but the refractory component contributes little at the short wavelength edge suggesting that a volatile component is required. Lower Panel: The same model (solid line), now adjusted to lower optical depth in the refractory component to better match the 6 $\mu$m absorption and with a smaller additive M 82 component is applied to longer wavelength data. Here, the effects of silicate absorption have also been included based on fig. 5 of Dudley & Wynn-Williams (1997) but oddly shifted to the red by about 0.6 $\mu$m at 10 $\mu$m, perhaps owing to differing composition. The inset shows the effects of adding the spectrum of M 82 as in the upper panel and without this component in the model at all (bounding dotted lines). The spectral resolution of the model matches that of the data in the inset. The short-dashed line is an emission model described more fully in the text (Sec. 4.1). Data sources for IRAS 00188-0856 in descending order in the legend are from Imanishi & Maloney (2003); Monet et al. (1998); Cutri et al. (2003); Moshir et al. (1990); and Charmandaris et al. (2002). R and K’ data are in rough agreement with the photometry given by Kim, Veilleux & Sanders (2002). Figure 2: Highly processed data products from the ISO Archive (hpdp_05100540_2.tar and hpdp_65000608_2.tar with minimal ISAP post processing) are compared with a 43 $\mu$m ice absorption feature derived from the spectrum of HD 161796 (also from the ISO archive: SWAA52100537). The diamonds and circles are data for Mkn 231 and IRAS 17208$-$0014 respectively. The solid lines are derived from the ISO SWS spectrum of HD 161796 using a single temperature (140 K) to estimate emissivity after the manner of Roche & Aitken (1984) and then converted to absorption. The inset is described in the text. Figure 3: Observational data for IRAS 14348-1447 (filled circles) are plotted at their rest wavelength in Jy as reported by Klaas et al. (2001). Horizontal bars indicate filter passbands. The solid line shows an optically thin (in the FIR) ice emission model whose $\beta$ parameter (100 and 300–900$\mu$m) is $\sim 2.3$ as compared to $\sim 1.9$ for a bare (refractory) core. Adjustment of the core to smaller $\beta$ would result in a smaller $\beta$ for the mantled grain as well. In fitting the model data at 95 and 450 $\mu$m were matched and various mantle-to-core volume ratios were compared with photometry between these wavelengths. The dashed line reproduces the fit proposed by Klaas et al. (2001) which is quite optically thick in the FIR and has $\beta$=2. Figure 4: Observational data for QSO J1154+5251 (filled circles and arrow) are plotted at their rest wavelength in mJy as reported by Bertoldi et al. (2003a) (3 mm); Bertoldi et al. (2003b) (1.2 mm); Robson et al. (2004) (850 and 450 $\mu$m); Beelen et al. (2006) (350 $\mu$m). The lines represent emission models. The dashed and dot-dashed are single temperature models using refractory and ice mantled dust respectively while the solid line two temperature ice model with colder ice obscuring warmer ice. Figure 5: The observed frame 450-to-850 flux density ratio calculated for a series of emission models is plotted as a function of redshift together with 9 observed lower limits (large filled symbols), 34 observed upper limits (medium open symbols), and two detections (smallest symbols). Individual 450 $\mu$m source and detection identifications are given in the legend. The models, also in the legend, are more fully described in the text. Figure 6: Brightness in mJy at the observed wavelength 450 $\mu$m is plotted as a function of redshift for a set of models. Four of the models are carried over from Fig. 5: The solid line represents 150 K ice, the triple dot dashed line, 50 K ice, the dot dashed line 30 K ice and the long dashed line 60 K refractory dust. A fifth model (short dashed line) is taken from the solid line in Fig. 4: 71 K ice is obscured by 24 K ice. Infrared luminosities for the models are constant as a function of redshift and are $2\times 10^{13}$ L⊙: long-dashed, dot-dashed, and triple-dot-dashed lines, $2\times 10^{14}$ L⊙: solid line, and $1\times 10^{13}$ L⊙: short-dashed line. We remark that just as standard dust shows roughly redshift independent brightness at 850 $\mu$m, so too does T=50 K ice (triple dot dashed line) at 450 $\mu$m. The filled circle is QSO J1148+5251. Figure 7: Observational data for HDF 850.1 (filled circles) are plotted for a rest wavelength and luminosity scale corresponding to a redshift of 12.6 and accounting for the effects of gravitational lensing by reducing the apparent luminosity by a factor of 3. The data, except for the $Z$ band limit, are reported or summarized in Dunlop et al. (2004). The $Z$ band limit is measured from GOODS ACS data (Giavalisco et al. 2004) by subtracting the image of the lensing galaxy from itself after a rotation of 180°. The limit is 2$\sigma$ drawn from the weight image for an 0$\farcs$5 aperture centered at the IRAM position (Downes et al. 1999) and corresponds to $Z-K^{\prime}>3.5$ (Vega). Data for the lensing source (filled diamonds) should be read relative to the HDF 850.1 data (filled circles) as the scalings in wavelength and luminosity do not apply. Here the $I-K^{\prime}$ photometry are from Dunlop et al. (2004) while the 3.6–24 $\mu$m points are measured from GOODS IRAC and MIPS images (Dickinson et al. 2006). The IRAC data are 6$\farcs$0 diameter aperture photometry subtracting an equal area outer annulus. The 24 $\mu$m datum is obtained through radial profile fitting after subtraction of a confusing source and is perhaps a 2.5$\sigma$ measurement of 20 $\mu$Jy. The dashed line is scaled ($z=1.1$ wrt observed frame; Fernández-Soto et al. 1999) data for the elliptical galaxy NGC 1399 (Persson et al. 1979;Athey et al. 2002) and the triple-dot-dashed line is for the galaxy NGC 3928 (Huchra 1977; 2MASS; Xilouris et al. 2004) with the shaded region between them showing the age dependent range of mid-infrared flux suggested by models for elliptical galaxies reported by Temi et al. (2005). The models shown as solid lines are described more fully in the text. Figure 8: The 850 $\mu$m brightness as a function of redshift is shown for three calculated SEDs. The solid line corresponds to the FIR SED shown in Fig. 1. The long-dashed line corresponds to an optically thin $T=45$, $\beta=1.2$ modified blackbody similar to the Scoville et al. (1991) fit for Arp 220. The short-dashed line corresponds to the FIR SED shown in Fig. 7 all normalized to the FIR luminosity of IRAS 00188-0856. The arrows for the solid, dot-dashed, and dashed lines indicate redshifts for HDF 850.1 estimated using Fig. 1, Dunlop et al. (2004), and Fig. 7 respectively. The numbers at the right are a conversion of the vertical scale for the gravitational lens amplification factor of three estimated for HDF 850.1. Figure 1: Figure 2: Figure 3: Figure 4: Figure 5: Figure 6: Figure 7: Figure 8:

arxiv-papers

2024-09-04T02:48:53.693720

0802.1703

# Lagrangian Floer theory on compact toric manifolds I Kenji Fukaya, Yong-Geun Oh, Hiroshi Ohta, Kaoru Ono (Date: Sep. 3, 2009) ###### Abstract. The present authors introduced the notion of _weakly unobstructed_ Lagrangian submanifolds and constructed their _potential function_ $\mathfrak{PO}$ purely in terms of $A$-model data in [FOOO3]. In this paper, we carry out explicit calculations involving $\mathfrak{PO}$ on toric manifolds and study the relationship between this class of Lagrangian submanifolds with the earlier work of Givental [Gi1] which advocates that quantum cohomology ring is isomorphic to the Jacobian ring of a certain function, called the Landau- Ginzburg superpotential. Combining this study with the results from [FOOO3], we also apply the study to various examples to illustrate its implications to symplectic topology of Lagrangian fibers of toric manifolds. In particular we relate it to Hamiltonian displacement property of Lagrangian fibers and to Entov-Polterovich’s symplectic quasi-states. ###### Key words and phrases: Floer cohomology, toric manifolds, weakly unobstructed Lagrangian submanifolds, potential function, Jacobian ring, balanced Lagrangian fibers, quantum cohomology KF is supported partially by JSPS Grant-in-Aid for Scientific Research No.18104001 and Global COE Program G08, YO by US NSF grant # 0503954, HO by JSPS Grant-in-Aid for Scientific Research No.19340017, and KO by JSPS Grant- in-Aid for Scientific Research, Nos. 17654009 and 18340014 ###### Contents 1. 1 Introduction 2. 2 Compact toric manifolds 1. 2.1 Complex structure 2. 2.2 Symplectic structure 3. 3 Deformation theory of filtered $A_{\infty}$-algebras 4. 4 Potential function 5. 5 Examples 6. 6 Quantum cohomology and Jacobian ring 7. 7 Localization of quantum cohomology ring at moment polytope 8. 8 Further examples and remarks 9. 9 Variational analysis of potential function 10. 10 Elimination of higher order term in nondegenerate cases 11. 11 Calculation of potential function 12. 12 Non-unitary flat connection on $L(u)$ 13. 13 Floer cohomology at a critical point of potential function 14. 14 Appendix 1 : Algebraically closedness of Novikov fields 15. 15 Appendix 2 : $T^{n}$-equivariant Kuranishi structure 16. 16 Appendix 3 : Smooth correspondence via the zero set of multisection ## 1\. Introduction Floer theory of Lagrangian submanifolds plays an important role in symplectic geometry since Floer’s invention [Fl] of the Floer cohomology and subsequent generalization to the class of _monotone_ Lagrangian submanifolds [Oh1]. After the introduction of $A_{\infty}$ structure in Floer theory [Fu1] and Kontsevich’s homological mirror symmetry proposal [Ko], it has also played an essential role in a formulation of mirror symmetry in string theory. In [FOOO1], we have analyzed the anomaly $\partial^{2}\neq 0$ and developed an obstruction theory for the definition of Floer cohomology and introduced the class of _unobstructed_ Lagrangian submanifolds for which one can deform Floer’s original definition of the ‘boundary’ map by a suitable bounding cochain denoted by $b$. Expanding the discussion in section 7 [FOOO1] and motivated by the work of Cho-Oh [CO], we also introduced the notion of _weakly unobstructed_ Lagrangian submanifolds in Chapter 3 [FOOO3] which turns out to be the right class of Lagrangian submanifolds to look at in relation to the mirror symmetry of Fano toric $A$-model and Landau-Ginzburg $B$-model proposed by physicists (see [Ho], [HV]). In the present paper, we study the relationship between this class of Lagrangian submanifolds with the earlier work of Givental [Gi1] which advocates that quantum cohomology ring is isomorphic to the Jacobian ring of a certain function, which is called the Landau-Ginzburg superpotential. Combining this study with the results from [FOOO3], we also apply this study to symplectic topology of Lagrangian fibers of toric manifolds. While appearance of bounding cochains is natural in the point of view of deformation theory, explicit computation thereof has not been carried out. One of the main purposes of the present paper is to perform this calculation in the case of fibers of toric manifolds and draw its various applications. Especially we show that each fiber $L(u)$ at $u\in\mathfrak{t}^{*}$ is weakly unobstructed for _any_ toric manifold $\pi:X\to\mathfrak{t}^{*}$ (see Proposition 4.3), and then show that the set of the pairs $(L(u),b)$ of a fiber $L(u)$ and a weak bounding cochain $b$ with nontrivial Floer cohomology can be calculated from the quantum cohomology of the ambient toric manifold, at least in the Fano case. Namely the set of such pairs $(L(u),b)$ is identified with the set of ring homomorphisms from quantum cohomology to the relevant Novikov ring. We also show by a variational analysis that for any compact toric manifold there exists at least one pair of $(u,b)$’s for which the Floer cohomology of $(L(u),b)$ is nontrivial. We call a Lagrangian fiber (that is, a $T^{n}$-orbit) balanced, roughly speaking, if its Floer cohomology is nontrivial. (See Definition 4.11 for its precise definition.) The main result of this paper is summarized as follows. 1. (1) When $X$ is a compact Fano toric manifold, we give a method to locate all the balanced fibers. 2. (2) Even when $X$ is not Fano, we can still apply the same method to obtain a finite set of Lagrangian fibers. We prove this set coincides with the set of balanced Lagrangian fibers under certain nondegeneracy condition. This condition can be easily checked, when a toric manifold is given. Now more precise statement of the main results are in order. Let $X$ be an $n$ dimensional smooth compact toric manifold. We fix a $T^{n}$-equivariant Kähler form on $X$ and let $\pi:X\to\mathfrak{t}^{*}\cong({\mathbb{R}}^{n})^{*}$ be the moment map. The image $P=\pi(X)\subset({\mathbb{R}}^{n})^{*}$ is called the moment polytope. For $u\in\text{\rm Int}\,P$, we denote $L(u)=\pi^{-1}(u)$. The fiber $L(u)$ is a Lagrangian torus which is an orbit of the $T^{n}$ action. (See section 2. We refer readers to, for example, [Au], [Ful] for the details on toric manifolds.) We study the Floer cohomology defined in [FOOO3]. According to [FOOO1, FOOO3], we need an extra data, the bounding cochain, to make the definition of Floer cohomology more flexible to allow more general class of Lagrangian submanifolds. In the current context of Lagrangian torus fibers in toric manifolds, we use _weak bounding cochains_. Denote by ${\mathcal{M}}_{\text{\rm weak}}(L(u);\Lambda_{0})$ the moduli space of (weak) bounding cochains for a weakly unobstructed Lagrangian submanifold $L(u)$. (See the end of section 4.) In this situation we first show that each element in $H^{1}(L(u);\Lambda_{0})$ gives rise to a weak bounding cochain, i.e., there is a natural embedding $H^{1}(L(u);\Lambda_{0})\hookrightarrow\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{0}).$ (1.1) (See Proposition 4.3.) Here we use the universal Novikov ring $\Lambda=\left\\{\sum_{i=1}^{\infty}a_{i}T^{\lambda_{i}}\,\Big{|}\,a_{i}\in{\mathbb{Q}},\lambda_{i}\in{\mathbb{R}},\lim_{i\to\infty}\lambda_{i}=\infty\right\\}$ (1.2) where $T$ is a formal parameter. (We do not use the grading parameter $e$ used in [FOOO3] since it will not play much role in this paper.) Then $\Lambda_{0}$ is a subring of $\Lambda$ defined by $\Lambda_{0}=\left\\{\sum_{i=1}^{\infty}a_{i}T^{\lambda_{i}}\in\Lambda\,\Big{|}\,\lambda_{i}\geq 0\right\\}.$ (1.3) We also use another subring $\Lambda_{+}=\left\\{\sum_{i=1}^{\infty}a_{i}T^{\lambda_{i}}\in\Lambda\,\Big{|}\,\lambda_{i}>0\right\\}.$ (1.4) We note $\Lambda$ is the field of fractions of $\Lambda_{0}$ and $\Lambda_{0}$ is a local ring with maximal ideal $\Lambda_{+}$. Here we take the universal Novikov ring over ${\mathbb{Q}}$ but we also use universal Novikov ring over ${\mathbb{C}}$ or other ring $R$ which we denote $\Lambda^{{\mathbb{C}}}$, $\Lambda^{R}$, respectively. (In case $R$ does not contain ${\mathbb{Q}}$, Floer cohomology over $\Lambda^{R}$ is defined only in Fano case.) ###### Remark 1.1. If we strictly follow the way taken in [FOOO3], we only get the embedding $H^{1}(L(u);\Lambda_{+})\hookrightarrow\mathcal{M}_{\text{\rm weak}}(L(u))$, not (1.1). Here $\mathcal{M}_{\text{\rm weak}}(L(u))=\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{+})$ is defined in [FOOO3]. (We note that $\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{+})\neq\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{0})$ where the right hand side is defined at the end of section 4.) However we can modify the definition of weak unobstructedness so that (1.1) follows, using the idea of Cho [Cho]. See section 12. Hereafter we use the symbol $b$ for an element of $\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{+})$ and $\mathfrak{x}$ for an element of $\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{0})$. We next consider the quantum cohomology ring $QH(X;\Lambda)$ with the universal Novikov ring $\Lambda$ as a coefficient ring. (See section 6.) It is a commutative ring for the toric case, since $QH(X;\Lambda)$ is generated by even degree cohomology classes. ###### Definition 1.2. 1. (1) We define the set $\text{\rm Spec}(QH(X;\Lambda))(\Lambda^{{\mathbb{C}}})$ to be the set of $\Lambda$ algebra homomorphisms $\varphi:QH(X;\Lambda)\to\Lambda^{{\mathbb{C}}}$. (In other words it is the set of all $\Lambda^{{\mathbb{C}}}$ valued points of the scheme $\text{\rm Spec}(QH(X;\Lambda))$. 2. (2) We next denote by $\mathfrak{M}(\mathfrak{Lag}(X))$ the set of all pairs $(\mathfrak{x},u)$, $u\in\operatorname{Int}P$, $\mathfrak{x}\in H^{1}(L(u);\Lambda_{0}^{{\mathbb{C}}})/H^{1}(L(u);2\pi\sqrt{-1}{\mathbb{Z}})$ such that $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x});\Lambda^{{\mathbb{C}}})\neq\\{0\\}.$ ###### Theorem 1.3. If $X$ is a Fano toric manifold then $\text{\rm Spec}(QH(X;\Lambda))(\Lambda^{{\mathbb{C}}})\cong\mathfrak{M}(\mathfrak{Lag}(X)).$ If $QH(X;\Lambda)$ is semi-simple in addition, we have $\sum_{d}\text{\rm rank}_{{\mathbb{Q}}}H_{d}(X;{\mathbb{Q}})=\\#\left(\mathfrak{M}(\mathfrak{Lag}(X))\right).$ (1.5) We remark that a commutative ring that is a finite dimensional vector space over a field (e.g., $\Lambda$ in our case) is semi-simple if and only if it does not contain any nilpotent element. We also remark that a compact toric manifold is Fano if and only if every nontrivial holomorphic sphere has positive Chern number. We believe that (1.5) still holds in the non-Fano case but are unable to prove it at the time of writing this paper. We however can prove that there exists a fiber $L(u)$ whose Floer cohomology is nontrivial, by a method different from the proof of Theorem 1.3. Due to some technical reason, we can only prove the following slightly weaker statement. ###### Theorem 1.4. Assume the Kähler form $\omega$ of $X$ is rational. Then, there exists $u\in\operatorname{Int}P$ such that for any ${{\mathcal{N}}}\in{\mathbb{R}}_{+}$ there exists $\mathfrak{x}\in H^{1}(L(u);\Lambda^{{\mathbb{R}}}_{0})$ with $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x});\Lambda^{{\mathbb{R}}}_{0}/(T^{{\mathcal{N}}}))\cong H(T^{n};{{\mathbb{R}}})\otimes_{{\mathbb{R}}}\Lambda_{0}^{{\mathbb{R}}}/(T^{{\mathcal{N}}}).$ We suspect that the rationality assumption in Theorem 1.4 can be removed. It is also likely that we can prove $\mathfrak{M}(\mathfrak{Lag}(X))$ is nonempty but its proof at the moment is a bit cumbersome to write down. We can however derive the following theorem from Theorem 1.4, without rationality assumption. ###### Theorem 1.5. Let $X$ be an $n$ dimensional compact toric manifold. There exists $u_{0}\in\text{\rm Int}P$ such that the following holds for any Hamiltonian diffeomorphism $\psi:X\to X$, $\psi(L(u_{0}))\cap L(u_{0})\neq\emptyset.$ (1.6) If $\psi(L(u_{0}))$ is transversal to $L(u_{0})$ in addition, then $\\#(\psi(L(u_{0}))\cap L(u_{0}))\geq 2^{n}.$ (1.7) Theorem 1.5 will be proved in section 13. We would like to point out that (1.6) can be derived from a more general intersection result, Theorem 2.1 [EP1], obtained by Entov-Polterovich with a different method using a very interesting notion of partial symplectic quasi- state constructed out of the spectral invariants defined in [Sc], [Oh3]. (See also [Vi], [Oh2] for similar constructions in the context of exact Lagrangian submanifolds.) ###### Remark 1.6. Strictly speaking, Theorem 2.1 [EP1] is stated under the assumption that $X$ is semi-positive and $\omega$ is rational because the theory of spectral invariant was developed in [Oh3] under these conditions. The rationality assumption has been removed in [Oh4], [Us] and the semi-positivity assumption of $\omega$ removed by Usher [Us]. Thus the spectral invariant satisfying all the properties listed in [EP1] section 5 is now established for an arbitrary compact symplectic manifold. By the argument of [EP1] section 7, this implies the existence of a partial symplectic quasi-state. Therefore the proof of Theorem 2.1 [EP1] goes through without these assumptions (semi-positivity and rationality) and hence it implies (1.6). (See the introduction of [Us].) But the result (1.7) is new. Our proof of Theorem 1.5 gives an explicit way of locating $u_{0}$, as we show in section 9. (The method of [EP1] is indirect and does not provide a way of finding such $u_{0}$. See [EP2]. Below, we will make some remarks concerning Entov-Polterovich’s approach in the perspective of homological mirror symmetry.) In various explicit examples we can find more than one element $u_{0}$ that have the properties stated in this theorem. Following terminology employed in [CO], we call any such torus fiber $L(u_{0})$ as in Theorem 1.4 a _balanced_ Lagrangian torus fiber. (See Definition 4.11 for its precise definition.) A criterion for $L(u_{0})$ to be balanced, for the case $\mathfrak{x}=0$, is provided by Cho-Oh [CO] and Cho [Cho] under the Fano condition. Our proofs of Theorems 1.4, 1.5 are much based on this criterion, and on the idea of Cho [Cho] of twisting _non-unitary_ complex line bundles in the construction of Floer boundary operator. This criterion in turn specializes to the one predicted by physicists [HV], [Ho], which relates the location of $u_{0}$ to the critical points of the Landau-Ginzburg superpotential. A precise description of balanced Lagrangian fibers including the data of bounding cochains involves the notion of a _potential function_ : In [FOOO3], the authors have introduced a function $\mathfrak{PO}^{L}:{\mathcal{M}}_{\text{\rm weak}}(L)\to\Lambda_{0}$ for an arbitrary weakly unobstructed Lagrangian submanifold $L\subset(X,\omega)$. By varying the function $\mathfrak{PO}^{L}$ over $L\in\\{\pi^{-1}(u)\mid u\in\text{Int}\,P\\}$, we obtain the potential function $\mathfrak{PO}:\bigcup_{L\in\\{\pi^{-1}(u)\mid u\in\text{Int}\,P\\}}{\mathcal{M}}_{\text{\rm weak}}(L)\to\Lambda_{0}.$ (1.8) This function is constructed purely in terms of $A$-model data of the general symplectic manifold $(X,\omega)$ _without_ using mirror symmetry. For a toric $(X,\omega)$, the restriction of $\mathfrak{PO}$ to $H^{1}(L(u);\Lambda_{+})$ (see (1.1)) can be made explicit when combined with the analysis of holomorphic discs attached to torus fibers of toric manifolds carried out in [CO], at least in the Fano case. (In the non-Fano case we can make it explicit modulo ‘higher order terms’.) This function extends to $H^{1}(L(u);\Lambda_{0})$. ###### Remark 1.7. In [EP3] some relationships between quantum cohomology, quasi-state, spectral invariant and displacement of Lagrangian submanifolds are discussed : Consider an idempotent i of quantum cohomology. The (asymptotic) spectral invariants associated to i gives rise to a partial symplectic quasi-state via the procedure concocted in [EP3], which in turn detects non-displaceability of certain Lagrangian submanifolds. (The assumption of [EP1] is weaker than ours.) In the current context of toric manifolds, we could also relate them to Floer cohomology and mirror symmetry in the following way : Quantum cohomology is decomposed into indecomposable factors. (See Proposition 7.7.) Let i be the idempotent corresponding to one of the indecomposable factors. Let $L=L(u(1,\text{\bf i}))$ be a Lagrangian torus fiber whose non-displaceability is detected by the partial symplectic quasi-state obtained from i. We conjecture that Floer cohomology $HF(L(u(1,\text{\bf i}),\mathfrak{x}),(L(u(1,\text{\bf i}),\mathfrak{x})))$ is nontrivial for some $\mathfrak{x}$. (See Remark 5.8.) This bounding cochain $\mathfrak{x}$ in turn is shown to be a critical point of the potential function $\mathfrak{PO}$ defined in [FOOO3]. On the other hand, i also determines a homomorphism $\varphi_{\text{\bf i}}:QH(X;\Lambda)\to\Lambda$. It corresponds to some Lagrangian fiber $L(u(2,\text{\bf i}))$ by Theorem 1.3. Then this will imply via Theorem 4.10 that the fiber $L(u(2,\text{\bf i}))$ is non-displaceable. We conjecture that $u(1,\text{\bf i})=u(2,\text{\bf i})$. We remark that $u(2,\text{\bf i})$ is explicitly calculable. Hence in view of the way $u(1,\text{\bf i})$ is found in [EP1], $u(1,\text{\bf i})=u(2,\text{\bf i})$ will give some information on the asymptotic behavior of the spectral invariant associated with i. We fix a basis of the Lie algebra $\mathfrak{t}$ of $T^{n}$ which induces a basis of $\mathfrak{t}^{*}$ and hence a coordinate of the moment polytope $P\subset\mathfrak{t}^{*}$. This in turn induces a basis of $H^{1}(L(u);\Lambda_{0})$ for each $u\in\operatorname{Int}P$ and so identification $H^{1}(L(u);\Lambda_{0})\cong(\Lambda_{0})^{n}$. We then regard the potential function as a function $\mathfrak{PO}(x_{1},\cdots,x_{n};u_{1},\cdots,u_{n}):(\Lambda_{0})^{n}\times\operatorname{Int}P\to\Lambda_{0}$ and prove in Theorem 4.10 that Floer cohomology $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x});\Lambda)$ with $\mathfrak{x}=(\mathfrak{x}_{1},\cdots,\mathfrak{x}_{n})$, $u=(u_{1},\cdots,u_{n})$ is nontrivial if and only if $(\mathfrak{x},u)$ satisfies $\frac{\partial\mathfrak{PO}}{\partial x_{i}}(\mathfrak{x};u)=0,\quad i=1,\cdots,n.$ (1.9) To study (1.9), it is useful to change the variables $x_{i}$ to $y_{i}=e^{x_{i}}.$ In these variables we can write potential function as a sum $\mathfrak{PO}(x_{1},\cdots,x_{n};u_{1},\cdots,u_{n})=\sum T^{c_{i}(u)}P_{i}(y_{1},\cdots,y_{n})$ where $P_{i}$ are Laurent polynomials which do not depend on $u$, and $c_{i}(u)$ are positive real valued functions. When $X$ is Fano, we can express the right hand side as a finite sum. (See Theorem 4.5.) We define a function $\mathfrak{PO}^{u}$ of $y_{i}$’s by $\mathfrak{PO}^{u}(y_{1},\cdots,y_{n})=\mathfrak{PO}(x_{1},\cdots,x_{n};u_{1},\cdots,u_{n})$ as a Laurent polynomial of $n$ variables with coefficient in $\Lambda$. We denote the set of Laurent polynomials by $\Lambda[y_{1},\cdots,y_{n},y_{1}^{-1},\cdots,y_{n}^{-1}]$ and consider its ideal generated by the partial derivatives of $\mathfrak{PO}^{u}$. Namely $\left(\frac{\partial\mathfrak{PO}^{u}}{\partial y_{i}};i=1,\cdots,n\right).$ ###### Definition 1.8. We call the quotient ring $Jac(\mathfrak{PO}^{u})=\frac{\Lambda[y_{1},\cdots,y_{n},y_{1}^{-1},\cdots,y_{n}^{-1}]}{\left(\frac{\partial\mathfrak{PO}^{u}}{\partial y_{i}};i=1,\cdots,n\right)}$ the Jacobian ring of $\mathfrak{PO}^{u}$. We will prove that the Jacobian ring is independent of the choice of $u$ up to isomorphism (see the end of section 6) and so we will just write $Jac(\mathfrak{PO})$ for $Jac(\mathfrak{PO^{u}})$ when there is no danger of confusion.. ###### Theorem 1.9. If $X$ is Fano, there exists a $\Lambda$ algebra isomorphism $\psi_{u}:QH(X;\Lambda)\to Jac(\mathfrak{PO})$ from quantum cohomology ring to the Jacobian ring such that $\psi_{u}(c_{1}(X))=\mathfrak{PO}^{u}.$ Theorem 1.9 (or Theorem 1.12 below) enables us to explicitly determine all the pairs $(\mathfrak{x},u)$ with $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x});\Lambda)\neq 0$ out of the quantum cohomology of $X$. More specifically Batyrev’s presentation of quantum cohomology in terms of the Jacobian ring plays an essential role for this purpose. We will explain how this is done in sections 7, 8. ###### Remark 1.10. 1. (1) The idea that quantum cohomology ring coincides with the Jacobian ring begins with a celebrated paper by Givental. (See Theorem 5 (1) [Gi1].) There it was claimed also that the $D$ module defined by an oscillatory integral with the superpotential as its kernel is isomorphic to $S^{1}$-equivariant Floer cohomology of periodic Hamiltonian system. When one takes its WKB limit, the former becomes the ring of functions on its characteristic variety, which is nothing but the Jacobian ring. The latter becomes the (small) quantum cohomology ring under the same limit. _Assuming_ the Ansatz that quantum cohomology can be calculated by fixed point localization, these claims are proved in a subsequent paper [Gi2] for, at least, toric Fano manifolds. Then the required fixed point localization is made rigorous later in [GrPa]. See also Iritani [Iri1]. In physics literature, it has been advocated that Landau-Ginzburg model of superpotential (that is, the potential function $\mathfrak{PO}$ in our situation) calculates quantum cohomology of $X$. A precise mathematical statement thereof is our Theorem 1.9. (See for example p. 473 [MIRROR].) Our main new idea entering in the proof of Theorem 1.3 other than those already in [FOOO3] is the way how we combine them to extract information on Lagrangian submanifolds. In fact Theorem 1.9 itself easily follows if we use the claim made by Batyrev that quantum cohomology of toric Fano manifold is a quotient of polynomial ring by the ideal of relations, called quantum Stanley- Reisner relation and linear relation. (This claim is now well established.) We include this simple derivation in section 6 for completeness’ sake, since it is essential to take the Novikov ring $\Lambda$ as the coefficient ring in our applications the version of which does not seem to be proven in the literature in the form that can be easily quoted. 2. (2) The proof of Theorem 1.9 given in this paper does not contain a serious study of pseudo-holomorphic spheres. The argument which we outline in Remark 6.15 is based on open-closed Gromov-Witten theory, and different from other various methods that have been used to calculate Gromov-Witten invariant in the literature. In particular this argument does not use the method of fixed point localization. We will present this conceptual proof of Theorem 1.9 in a sequel to this paper. 3. (3) The isomorphism in Theorem 1.9 may be regarded as a particular case of the conjectural relation between quantum cohomology and Hochschild cohomology of Fukaya category. See Remark 6.15. 4. (4) In this paper, we only involve small quantum cohomology ring but we can also include big quantum cohomology ring. Then we expect Theorem 1.9 can be enhanced to establish a relationship between the Frobenius structure of the deformation theory of quantum cohomology (see, for example, [Man]) and that of Landau-Ginzburg model (which is due to K. Saito [Sa]). This statement (and Theorems 1.3, 1.9) can be regarded as a version of mirror symmetry between the toric A-model and the Landau-Ginzburg $B$-model. In various literature on mirror symmetry, such as [Ab], [AKO], [Ue], the B-model is dealt for Fano or toric manifolds in which the derived category of coherent sheaves is studied while the A-model is dealt for Landau-Ginzburg $A$-models where the directed $A_{\infty}$ category of Seidel [Se2] is studied. 5. (5) In [Aur], Auroux discussed a mirror symmetry between the $A$-model side of toric manifolds and the $B$-model side of Landau-Ginzburg models. The discussion of [Aur] uses Floer cohomology with ${\mathbb{C}}$-coefficients. In this paper we use Floer cohomology over the Novikov ring which is more suitable for the applications to symplectic topology. 6. (6) Even when $X$ is not necessarily Fano we can still prove a similar isomorphism $\psi_{u}:QH^{\omega}(X;\Lambda)\cong Jac(\mathfrak{PO}_{0})$ (1.10) where the left hand side is the Batyrev quantum cohomology ring (see Definition 6.4) and the right hand side is the Jacobian ring of some function $\mathfrak{PO}_{0}$ : it coincides with the actual potential function $\mathfrak{PO}$ ‘up to higher order terms’. (See (4.9).) In the Fano case $\mathfrak{PO}_{0}=\mathfrak{PO}$. (1.10) is Proposition 6.8. 7. (7) During the final stage of writing this article, a paper [CL] by Chan and Leung was posted in the Archive which studies the above isomorphism via SYZ transformations. They give a proof of this isomorphism for the case where $X$ is a product of projective spaces and with the coefficient ring ${\mathbb{C}}$, not with Novikov ring. Leung presented their result [CL] in a conference held in Kyoto University in January 2008 where the first named author also presented the content of this paper. From our definition, it follows that the leading order potential function $\mathfrak{PO}_{0}$ (see (4.9)) can be extended to the whole product $(\Lambda^{\mathbb{C}}_{0})^{n}\times{\mathbb{R}}^{n}$ in a way that they are invariant under the translations by elements in $(2\pi\sqrt{-1}{\mathbb{Z}})^{n}\subseteq(\Lambda^{\mathbb{C}}_{0})^{n}$. Hence we may regard $\mathfrak{PO}_{0}$ as a function defined on $\left(\Lambda^{\mathbb{C}}_{0}/(2\pi\sqrt{-1}{\mathbb{Z}})\right)^{n}\times{\mathbb{R}}^{n}\cong\left(\Lambda^{\mathbb{C}}_{0}/(2\pi\sqrt{-1}{\mathbb{Z}})\right)^{n}\times{\mathbb{R}}^{n}.$ In the non-Fano case, the function $\mathfrak{PO}$ is invariant under the translations by elements in $(2\pi\sqrt{-1}{\mathbb{Z}})^{n}\subseteq(\Lambda^{\mathbb{C}}_{0})^{n}$ but may not extend to $\Lambda^{\mathbb{C}}_{0}/(2\pi\sqrt{-1}{\mathbb{Z}}))^{n}\times{\mathbb{R}}^{n}$. This is because the infinite sum appearing in the right hand side of (4.7) may not converge in non-Archimedean topology for $u\notin\mbox{Int}\,P$. ###### Definition 1.11. We denote by $\text{\rm Crit}(\mathfrak{PO}_{0}),\quad(\mbox{respectively }\,\text{\rm Crit}(\mathfrak{PO}))$ the subset of pairs $(\mathfrak{x},u)\in(\Lambda_{0}^{{\mathbb{C}}}/(2\pi\sqrt{-1}{\mathbb{Z}}))^{n}\times{\mathbb{R}}^{n},\quad(\mbox{respectively }\,(\mathfrak{x},u)\in(\Lambda_{0}^{{\mathbb{C}}}/(2\pi\sqrt{-1}{\mathbb{Z}}))^{n}\times\mbox{Int}\,P)$ satisfying the equation $\frac{\partial\mathfrak{PO}_{0}}{\partial x_{i}}(\mathfrak{x};u)=0,\quad\left(\mbox{respectively }\,\frac{\partial\mathfrak{PO}}{\partial x_{i}}(\mathfrak{x};u)=0\right)$ $i=1,\cdots,n$. We define $\mathfrak{M}(\mathfrak{Lag}(X))$ in Definition 1.2. (We use the same definition in non-Fano case.) In view of Theorem 1.12 (2) below, we also introduce the subset $\mathfrak{M}_{0}(\mathfrak{Lag}(X))=\\{(\mathfrak{x},u)\in\text{\rm Crit}(\mathfrak{PO}_{0})\mid u\in\text{\rm Int}\,P\\}.$ We also remark $\mathfrak{PO}_{0}=\mathfrak{PO}$ in case $X$ is Fano. The following is a more precise form of Theorem 1.3. ###### Theorem 1.12. 1. (1) There exists a bijection : $\text{\rm Spec}(QH^{\omega}(X;\Lambda))(\Lambda^{{\mathbb{C}}})\cong\text{\rm Crit}(\mathfrak{PO}_{0}).$ 2. (2) There exists a bijection : $\mathfrak{M}(\mathfrak{Lag}(X))\cong\text{\rm Crit}(\mathfrak{PO}).$ 3. (3) If $X$ is Fano and $\mathfrak{x}$ is a critical point of $\mathfrak{PO}_{0}^{u}$ then $u\in\mbox{\rm Int}\,P$. 4. (4) If $QH^{\omega}(X;\Lambda)$ is semi-simple, then $\sum_{d}\text{\rm rank}_{\Lambda}QH^{\omega}(X;\Lambda)=\\#\left(\text{\rm Crit}(\mathfrak{PO}_{0})\right).$ ###### Remark 1.13. (1) Theorem 1.12 (3) does not hold in non-Fano case. We give a counter example (Example 8.2) in section 8. In fact, in the case of Example 8.2 some of the critical points of $\mathfrak{PO}_{0}$ correspond to a point $u\in{\mathbb{R}}^{n}$ which lies outside the moment polytope. (2) In the Fano case Theorem 1.12 (3) implies $\mathfrak{M}(\mathfrak{Lag}(X))=\mbox{Crit}(\mathfrak{PO})=\mbox{Crit}(\mathfrak{PO}_{0}).$ We would like to point out that $\mathfrak{PO}_{0}$ is explicitly computable. But we do not know the explicit form of $\mathfrak{PO}$. However we can show that elements of $\text{\rm Crit}(\mathfrak{PO}_{0})$ and of $\text{\rm Crit}(\mathfrak{PO})$ can be naturally related to each other under a mild nondegeneracy condition. (Theorem 10.4.) So we can use $\mathfrak{PO}_{0}$ in place of $\mathfrak{PO}$ in most of the cases. For example we can use it to prove that the following : ###### Theorem 1.14. For any $k$, there exists a Kähler form on $X(k)$, the $k$ points blow up of ${\mathbb{C}}P^{2}$, that is toric and has exactly $k+1$ balanced fibers. See Definition 4.11 for the definition of balanced fibers. Balanced fiber satisfies the conclusions (1.6), (1.7) of Theorem 1.5. We prove Theorem 1.14 in section 10. ###### Remark 1.15. The cardinality of $\mathfrak{x}\in H^{1}(L(u);\Lambda^{{\mathbb{C}}}_{0})/H^{1}(L(u);2\pi\sqrt{-1}{\mathbb{Z}})$ with nonvanishing Floer cohomology is an invariant of Lagrangian submanifold $L(u)$. This is a consequence of [FOOO3] Theorem G (= [FOOO2] Theorem G). The organization of this paper is now in order. In section 2 we gather some basic facts on toric manifolds and fix our notations. Section 3 is a brief review of Lagrangian Floer theory of [FOOO1, FOOO3]. In section 4, we describe our main results on the potential function $\mathfrak{PO}$ and on its relation to the Floer cohomology. We illustrate these theorems by several examples and derive their consequences in sections 5 \- 10. We postpone their proofs until sections 11 \- 13. In section 5, we illustrate explicit calculations involving the potential functions in such examples as ${\mathbb{C}}P^{n}$, $S^{2}\times S^{2}$ and the two points blow up of ${\mathbb{C}}P^{2}$. We also discuss a relationship between displacement energy of Lagrangian submanifold (see Definition 5.9) and Floer cohomology. In sections 6 and 7, we prove the results that mainly apply to the Fano case. Especially we prove Theorems 1.3, 1.9 and 1.12 in these sections. Section 7 contains some applications of Theorem 1.9 especially to the case of monotone torus fibers and to the ${\mathbb{Q}}$-structure of quantum cohomology ring. In section 8 we first illustrate usage of (the proof of) Theorem 1.3 to locate balanced fibers by the example of one point blow up of ${\mathbb{C}}P^{2}$. We then turn to the study of non-Fano cases and discuss Hirzebruch surfaces. Section 8 also contains some discussion on the semi-simplicity of quantum cohomology. In sections 9 and 10, we prove the results that can be used in all toric cases, whether they are Fano or not. In section 9, using variational analysis, we prove existence of a critical point of the potential function, which is an important step towards the proof of Theorems 1.4 and 1.5. Using the arguments of this section, we can locate a balanced fiber in any compact toric manifold, explicitly solving simple linear equalities and inequalities finitely many times. In section 10, we prove that we can find the solution of (1.9) by studying its reduction to ${\mathbb{C}}=\Lambda_{0}^{{\mathbb{C}}}/\Lambda_{+}^{{\mathbb{C}}}$, which we call the leading term equation. This result is purely algebraic. It implies that our method of locating balanced fibers, which is used in the proof of Theorem 1.3, can be also used in the non-Fano case under certain nondegeneracy condition. We apply this method to prove Theorem 1.14. We discuss an example of blow up of ${\mathbb{C}}P^{n}$ along the high dimensional blow up center ${\mathbb{C}}P^{m}$ in that section. We also give several other examples and demonstrate various interesting phenomena which occur in Lagrangian Floer theory. For example we provide a sequence $((X,\omega_{i}),L_{i})$ of pairs that have nonzero Floer cohomology for some choice of bounding cochains, while its limit $((X,\omega),L)$ has vanishing Floer cohomology for any choice of bounding cochain (Example 10.17). We also provide an example of Lagrangian submanifold $L$ such that it has a nonzero Floer cohomology over $\Lambda^{{\mathbb{C}}}$ for some choice of bounding cochain, but vanishing Floer cohomology for any choice of bounding cochain over the field $\Lambda^{F}$ with a field $F$ of characteristic $3$ (Example 10.19). In section 11, we review the results on the moduli space of holomorphic discs from [CO] which are used in the calculation of the potential function. We rewrite them in the form that can be used for the purpose of this paper. We also discuss the non-Fano case in this section. (Our result is less explicit in the non-Fano case, but still can be used to explicitly locate balanced fibers in most of the cases.) In section 12, we use the idea of Cho [Cho] to deform Floer cohomology by an element from $H^{1}(L(u);\Lambda_{0})$ rather than from $H^{1}(L(u);\Lambda_{+})$. This enhancement is crucial to obtain an optimal result about the non-displacement of Lagrangian fibers. In section 13 we use those results to calculate Floer cohomology and complete the proof of Theorems 1.4 and 1.5 etc. We attempted to make this paper largely independent of our book [FOOO1, FOOO3] as much as possible and also to make the relationship of the contents of the paper with the general story transparent. Here are a few examples : 1. (1) Our definition of potential function for the fibers of toric manifolds in this paper is given in a way independent of that of [FOOO3] except the statement on the existence of compatible Kuranishi structures and multi-sections on the moduli space of pseudo-holomorphic discs which provides a rigorous definition of Floer cohomology of single Lagrangian fiber. Such details are provided in section 3.1 [FOOO3] (= section 29 [FOOO2]). 2. (2) Similarly the definition of $A_{\infty}$ algebra in this paper on the Lagrangian fiber of toric manifolds is also independent of the book except the process going from $A_{n,K}$ structure to $A_{\infty}$ structure, which we refer to section 7.2 [FOOO3] (= section 30 [FOOO2]). However, for all the applications in this paper, only existence of $A_{n,K}$ structures is needed. 3. (3) The property of the Floer cohomology $HF(L,L)$ detecting Lagrangian intersection of $L$ with its Hamiltonian deformation relies on the fact that Floer cohomology of the pair is independent under the Hamiltonian isotopy. This independence is established in [FOOO3]. In the toric case, its alternative proof based on the de Rham version is given in [FOOO4] in a more general form than we need in this paper. The authors would like to thank H. Iritani and D. McDuff for helpful discussions. They would also like to thank the referee for various helpful comments. ## 2\. Compact toric manifolds In this section, we summarize basic facts on the toric manifolds and set-up our notations to be consistent with those in [CO], which in turn closely follow those in Batyrev [B1] and M. Audin [Au]. ### 2.1. Complex structure In order to obtain an $n$-dimensional compact toric manifold $X$, we need a combinatorial object $\Sigma$, a complete fan of regular cones, in an $n$-dimensional vector space over ${\mathbb{R}}$. Let $N$ be the lattice ${\mathbb{Z}}^{n}$, and let $M=Hom_{\mathbb{Z}}(N,{\mathbb{Z}})$ be the dual lattice of rank $n$. Let $N_{{\mathbb{R}}}=N\otimes{\mathbb{R}}$ and $M_{{\mathbb{R}}}=M\otimes{\mathbb{R}}$. ###### Definition 2.1. A convex subset $\sigma\subset N_{{\mathbb{R}}}$ is called a regular $k$-dimensional cone $(k\geq 1)$ if there exists $k$ linearly independent elements $v_{1},\cdots,v_{k}\in N$ such that $\sigma=\\{a_{1}v_{1}+\cdots+a_{k}v_{k}\mid a_{i}\in{\mathbb{R}},a_{i}\geq 0\\},$ and the set $\\{v_{1},\cdots,v_{k}\\}$ is a subset of some ${\mathbb{Z}}$-basis of $N$. In this case, we call $v_{1},\cdots,v_{k}\in N$ the integral generators of $\sigma$. ###### Definition 2.2. A regular cone $\sigma^{\prime}$ is called a face of a regular cone $\sigma$ (we write $\sigma^{\prime}\prec\sigma$) if the set of integral generators of $\sigma^{\prime}$ is a subset of the set of integral generators of $\sigma$. ###### Definition 2.3. A finite system $\Sigma={\sigma_{1},\cdots,\sigma_{s}}$ of regular cones in $N_{{\mathbb{R}}}$ is called a complete $n$-dimensional fan of regular cones, if the following conditions are satisfied. 1. (1) if $\sigma\in\Sigma$ and $\sigma^{\prime}\prec\sigma$, then $\sigma^{\prime}\in\Sigma$; 2. (2) if $\sigma,\sigma^{\prime}$ are in $\Sigma$, then $\sigma^{\prime}\cap\sigma\prec\sigma$ and $\sigma^{\prime}\cap\sigma\prec\sigma^{\prime}$; 3. (3) $N_{{\mathbb{R}}}=\sigma_{1}\cup\cdots\cup\sigma_{s}$. The set of all $k$-dimensional cones in $\Sigma$ will be denoted by $\Sigma^{(k)}.$ ###### Definition 2.4. Let $\Sigma$ be a complete $n$-dimensional fan of regular cones. Denote by $G(\Sigma)=\\{v_{1},\cdots,v_{m}\\}$ the set of all generators of 1-dimensional cones in $\Sigma$ ($m=$ Card $\Sigma^{(1)}$). We call a subset ${\mathcal{P}}=\\{v_{i_{1}},\cdots,v_{i_{p}}\\}\subset G(\Sigma)$ a primitive collection if $\\{v_{i_{1}},\cdots,v_{i_{p}}\\}$ does not generate $p$-dimensional cone in $\Sigma$, while for all $k\,(0\leq k<p)$ each $k$-element subset of ${\mathcal{P}}$ generates a $k$-dimensional cone in $\Sigma$. ###### Definition 2.5. Let ${\mathbb{C}}^{m}$ be an $m$-dimensional affine space over ${\mathbb{C}}$ with the set of coordinates $z_{1},\cdots,z_{m}$ which are in the one-to-one correspondence $z_{i}\leftrightarrow v_{i}$ with elements of $G(\Sigma)$. Let ${\mathcal{P}}=\\{v_{i_{1}},\cdots,v_{i_{p}}\\}$ be a primitive collection in $G(\Sigma)$. Denote by $\mathbb{A}({\mathcal{P}})$ the $(m-p)$-dimensional affine subspace in ${\mathbb{C}}^{n}$ defined by the equations $z_{i_{1}}=\cdots=z_{i_{p}}=0.$ Since every primitive collection ${\mathcal{P}}$ has at least two elements, the codimension of $\mathbb{A}({\mathcal{P}})$ is at least 2. ###### Definition 2.6. Define the closed algebraic subset $Z(\Sigma)$ in ${\mathbb{C}}^{m}$ as follows $Z(\Sigma)=\bigcup_{{\mathcal{P}}}\mathbb{A}({\mathcal{P}}),$ where ${\mathcal{P}}$ runs over all primitive collections in $G(\Sigma)$. Put $U(\Sigma)={\mathbb{C}}^{m}\setminus Z(\Sigma).$ ###### Definition 2.7. Let ${\mathbb{K}}$ be the subgroup in ${\mathbb{Z}}^{m}$ consisting of all lattice vectors $\lambda=(\lambda_{1},\cdots,\lambda_{m})$ such that $\lambda_{1}v_{1}+\cdots+\lambda_{m}v_{m}=0.$ Obviously ${\mathbb{K}}$ is isomorphic to ${\mathbb{Z}}^{m-n}$ and we have the exact sequence: $0\to{\mathbb{K}}\to{\mathbb{Z}}^{m}\stackrel{{\scriptstyle\pi}}{{\to}}{\mathbb{Z}}^{n}\to 0,$ (2.1) where the map $\pi$ sends the basis vectors $e_{i}$ to $v_{i}$ for $i=1,\cdots,m$. ###### Definition 2.8. Let $\Sigma$ be a complete $n$-dimensional fan of regular cones. Define $D(\Sigma)$ to be the connected commutative subgroup in $({\mathbb{C}}^{*})^{m}$ generated by all one-parameter subgroups $a_{\lambda}:{\mathbb{C}}^{*}\to({\mathbb{C}}^{*})^{m},$ $t\mapsto(t^{\lambda_{1}},\cdots,t^{\lambda_{m}})$ where $\lambda=(\lambda_{1},\cdots,\lambda_{m})\in{\mathbb{K}}$. It is easy to see from the definition that $D(\Sigma)$ acts freely on $U(\Sigma)$. Now we are ready to give a definition of the compact toric manifold $X_{\Sigma}$ associated with a complete $n$-dimensional fan of regular cones $\Sigma$. ###### Definition 2.9. Let $\Sigma$ be a complete $n$-dimensional fan of regular cones. Then the quotient $X_{\Sigma}=U(\Sigma)/D(\Sigma)$ is called the compact toric manifold associated with $\Sigma$. There exists a simple open coverings of $U(\Sigma)$ by affine algebraic varieties. ###### Proposition 2.10. Let $\sigma$ be a $k$-dimensional cone in $\Sigma$ generated by $\\{v_{i_{1}},\cdots,v_{i_{k}}\\}.$ Define the open subset $U(\sigma)\subset{\mathbb{C}}^{m}$ as $U(\sigma)=\\{(z_{1},\cdots,z_{m})\in{\mathbb{C}}^{m}\mid z_{j}\neq 0\;\;\textrm{for all}\;j\notin\\{i_{1},\cdots,i_{k}\\}\\}.$ Then the open sets $U(\sigma)$ have the following properties: 1. (1) $U(\Sigma)=\bigcup_{\sigma\in\Sigma}U(\sigma);$ 2. (2) if $\sigma\prec\sigma^{\prime}$, then $U(\sigma)\subset U(\sigma^{\prime})$; 3. (3) for any two cone $\sigma_{1},\sigma_{2}\in\Sigma$, one has $U(\sigma_{1})\cap U(\sigma_{2})=U(\sigma_{1}\cap\sigma_{2})$; in particular, $U(\Sigma)=\sum_{\sigma\in\Sigma^{(n)}}U(\sigma).$ ###### Proposition 2.11. Let $\sigma$ be an $n$-dimensional cone in $\Sigma^{(n)}$ generated by $\\{v_{i_{1}},\cdots,v_{i_{n}}\\}$, which spans the lattice $M$. We denote the dual ${\mathbb{Z}}$-basis of the lattice $N$ by $\\{u_{i_{1}},\cdots,u_{i_{n}}\\}$. i.e. $\langle v_{i_{k}},u_{i_{l}}\rangle=\delta_{k,l}$ (2.2) where $\langle\cdot,\cdot\rangle$ is the canonical pairing between lattices $N$ and $M$. Then the affine open subset $U(\sigma)$ is isomorphic to ${\mathbb{C}}^{n}\times({\mathbb{C}}^{*})^{m-n}$, the action of $D(\Sigma)$ on $U(\sigma)$ is free, and the space of $D(\Sigma)$-orbits is isomorphic to the affine space $U_{\sigma}={\mathbb{C}}^{n}$ whose coordinate functions $y_{1}^{\sigma},\cdots,y_{n}^{\sigma}$ are $n$ Laurent monomials in $z_{1},\cdots,z_{m}$: $\begin{cases}y_{1}^{\sigma}=z_{1}^{\langle v_{1},u_{i_{1}}\rangle}\cdots z_{m}^{\langle v_{m},u_{i_{1}}\rangle}\\\ \qquad\vdots\\\ y_{n}^{\sigma}=z_{1}^{\langle v_{1},u_{i_{n}}\rangle}\cdots z_{m}^{\langle v_{m},u_{i_{n}}\rangle}\end{cases}$ (2.3) The last statement yields a general formula for the local affine coordinates $y_{1}^{\sigma},\cdots,y_{n}^{\sigma}$ of a point $p\in U_{\sigma}$ as functions of its “homogeneous coordinates” $z_{1},\cdots,z_{m}$. ### 2.2. Symplectic structure In the last subsection, we associated a compact manifold $X_{\Sigma}$ to a fan $\Sigma$. In this subsection, we review the construction of symplectic (Kähler) manifold associated to a convex polytope $P$. Let $M$ be a dual lattice, we consider a convex polytope $P$ in $M_{{\mathbb{R}}}$ defined by $\\{u\in M_{{\mathbb{R}}}\mid\langle u,v_{j}\rangle\geq\lambda_{j}\;\textrm{for}\;j=1,\cdots,m\\}$ (2.4) where $\langle\cdot,\cdot\rangle$ is a dot product of $M_{{\mathbb{R}}}\cong{\mathbb{R}}^{n}$. Namely, $v_{j}$’s are inward normal vectors to the codimension 1 faces of the polytope $P$. We associate to it a fan in the lattice $N$ as follows: With any face $\Gamma$ of $P$, fix a point $u_{0}$ in the (relative) interior of $\Gamma$ and define $\sigma_{\Gamma}=\bigcup_{r\geq 0}r\cdot(P-u_{0}).$ The associated fan is the family $\Sigma(P)$ of dual convex cones $\displaystyle\check{\sigma}_{\Gamma}$ $\displaystyle=$ $\displaystyle\\{x\in N_{\mathbb{R}}\mid\langle y,x\rangle\geq 0\;\;\forall y\in\sigma_{\Gamma}\\}$ (2.5) $\displaystyle=$ $\displaystyle\\{x\in N_{\mathbb{R}}\mid\langle u,x\rangle\leq\langle p,x\rangle\;\;\forall p\in P,u\in\Gamma\\}$ (2.6) where $\langle\cdot,\cdot\rangle$ is dual pairing $M_{\mathbb{R}}$ and $N_{\mathbb{R}}$. Hence we obtain a compact toric manifold $X_{\Sigma(P)}$ associated to a fan $\Sigma(P)$. Now we define a symplectic (Kähler) form on $X_{\Sigma(P)}$ as follows. Recall the exact sequence : $0\to{\mathbb{K}}\stackrel{{\scriptstyle i}}{{\to}}{\mathbb{Z}}^{m}\stackrel{{\scriptstyle\pi}}{{\to}}{\mathbb{Z}}^{n}\to 0.$ It induces another exact sequence : $0\to K\to{\mathbb{R}}^{m}/{\mathbb{Z}}^{m}\to{\mathbb{R}}^{n}/{\mathbb{Z}}^{n}\to 0.$ Denote by $k$ the Lie algebra of the real torus $K$. Then we have the exact sequence of Lie algebras: $0\to k\to{\mathbb{R}}^{m}\stackrel{{\scriptstyle\pi}}{{\to}}{\mathbb{R}}^{n}\to 0.$ And we have the dual of above exact sequence: $0\to({\mathbb{R}}^{n})^{*}\to({\mathbb{R}}^{m})^{*}\stackrel{{\scriptstyle i^{*}}}{{\to}}k^{*}\to 0.$ Now, consider ${\mathbb{C}}^{m}$ with symplectic form $\frac{i}{2}\sum dz_{k}\wedge d\overline{z}_{k}$. The standard action $T^{n}$ on ${\mathbb{C}}^{n}$ is hamiltonian with moment map $\mu(z_{1},\cdots,z_{m})=\frac{1}{2}(|z_{1}|^{2},\cdots,|z_{m}|^{2}).$ (2.7) For the moment map $\mu_{K}$ of the $K$ action is then given by $\mu_{K}=i^{*}\circ\mu:{\mathbb{C}}^{m}\to k^{*}.$ If we choose a ${\mathbb{Z}}$-basis of ${\mathbb{K}}\subset{\mathbb{Z}}^{m}$ as $Q_{1}=(Q_{11},\cdots,Q_{m1}),\cdots,Q_{k}=(Q_{1k},\cdots,Q_{mk})$ and $\\{q^{1},\cdots,q^{k}\\}$ be its dual basis of ${\mathbb{K}}^{*}$. Then the map $i^{*}$ is given by the matrix $Q^{t}$ and so we have $\mu_{K}(z_{1},\cdots,z_{m})=\frac{1}{2}\left(\sum_{j=1}^{m}Q_{j1}|z_{j}|^{2},\cdots,\sum_{j=1}^{m}Q_{jk}|z_{j}|^{2}\right)\in{\mathbb{R}}^{k}\cong k^{*}$ (2.8) in the coordinates associated to the basis $\\{q^{1},\cdots,q^{k}\\}$. We denote again by $\mu_{K}$ the restriction of $\mu_{K}$ on $U(\Sigma)\subset{\mathbb{C}}^{m}$. ###### Proposition 2.12 (See Audin [Au]). Then for any $r=(r_{1},\cdots,r_{m-n})\in\mu_{K}(U(\Sigma))\subset k^{*}$, we have a diffeomorphism $\mu_{K}^{-1}(r)/K\cong U(\Sigma)/D(\Sigma)=X_{\Sigma}$ (2.9) And for each (regular) value of $r\in k^{*}$, we can associate a symplectic form $\omega_{P}$ on the manifold $X_{\Sigma}$ by symplectic reduction [MW]. To obtain the original polytope $P$ that we started with, we need to choose $r$ as follows: Consider $\lambda_{j}$ for $j=1,\cdots,m$ which we used to define our polytope $P$ by the set of inequalities $\langle u,v_{j}\rangle\geq\lambda_{j}$. Then, for each $a=1,\cdots,m-n$, let $r_{a}=-\sum_{j=1}^{m}Q_{ja}\lambda_{j}.$ Then we have $\mu_{K}^{-1}(r_{1},\cdots,r_{m-n})/K\cong X_{\Sigma(P)}$ and for the residual $T^{n}\cong T^{m}/K$ action on $X_{\Sigma(P)}$, and for its moment map $\pi$, we have $\pi(X_{\Sigma(P)})=P.$ Using Delzant’s theorem [De], one can reconstruct the symplectic form out of the polytope $P$ (up to $T^{n}$-equivariant symplectic diffeomorphisms). In fact, Guillemin [Gu] proved the following explicit closed formula for the $T^{n}$-invariant Kähler form associated to the canonical complex structure on $X=X_{\Sigma}(P)$ ###### Theorem 2.13 (Guillemin). Let $P$, $X_{\Sigma(P)}$, $\omega_{P}$ be as above and $\pi:X_{\Sigma(P)}\to({\mathbb{R}}^{m}/k)^{*}\cong({\mathbb{R}}^{n})^{*}$ be the associated moment map. Define the functions on $({\mathbb{R}}^{n})^{*}$ $\displaystyle\ell_{i}(u)$ $\displaystyle=$ $\displaystyle\langle u,v_{i}\rangle-\lambda_{i}\,\mbox{ for }\,i=1,\cdots,m$ (2.10) $\displaystyle\ell_{\infty}(u)$ $\displaystyle=$ $\displaystyle\sum_{i=1}^{m}\langle u,v_{i}\rangle=\left\langle u,\sum_{i=1}^{m}v_{i}\right\rangle.$ Then we have $\omega_{P}=\sqrt{-1}\partial\overline{\partial}\left(\pi^{*}\Big{(}\sum_{i=1}^{m}\lambda_{i}(\log\ell_{i})+\ell_{\infty}\Big{)}\right)$ (2.11) on $\text{\rm Int}\,P$. The affine functions $\ell_{i}$ will play an important role in our description of potential function as in [CO] since they also measure symplectic areas $\omega(\beta_{i})$ of the canonical generators $\beta_{i}$ of $H_{2}(X,L(u);{\mathbb{Z}})$. More precisely we have $\omega(\beta_{i})=2\pi\ell_{i}(u)$ (2.12) (see Theorem 8.1 [CO]). We also recall $P=\\{u\in M_{\mathbb{R}}\mid\ell_{i}(u)\geq 0,\,i=1,\cdots,m\\}$ (2.13) by definition (2.4). ## 3\. Deformation theory of filtered $A_{\infty}$-algebras In this section, we provide a quick summary of the deformation and obstruction theory of Lagrangian Floer cohomology developed in [FOOO1, FOOO3] for readers’ convenience. We also refer readers to the third named author’s survey paper [Ohta] for a more detailed review, and refer to [FOOO3] for complete details of the proofs of the results described in this section. We start our discussion with the classical unfiltered $A_{\infty}$ algebra. Let $C$ be a graded $R$-module where $R$ is the coefficient ring. We denote by $C[1]$ its suspension defined by $C[1]^{k}=C^{k+1}$. Define the bar complex $B(C[1])$ by $B_{k}(C[1])=\underbrace{C[1]\otimes\cdots\otimes C[1]}_{k},\quad B(C[1])=\bigoplus_{k=0}^{\infty}B_{k}(C[1]).$ Here $B_{0}(C[1])=R$ by definition. $B(C[1])$ has the structure of graded coalgebra. ###### Definition 3.1. The structure of $A_{\infty}$ algebra is a sequence of $R$ module homomorphisms $\mathfrak{m}_{k}:B_{k}(C[1])\to C[1],\quad k=1,2,\cdots,$ of degree +1 such that the coderivation $\widehat{d}=\sum_{k=1}^{\infty}\widehat{\mathfrak{m}}_{k}$ satisfies $\widehat{d}\,\widehat{d}=0$, which is called the _$A_{\infty}$ -relation_. Here we denote by $\widehat{\mathfrak{m}}_{k}:B(C[1])\to B(C[1])$ the unique extension of $\mathfrak{m}_{k}$ as a coderivation on $B(C[1])$, that is $\widehat{\mathfrak{m}}_{k}(x_{1}\otimes\cdots\otimes x_{n})=\sum_{i=1}^{k-i+1}(-1)^{*}x_{1}\otimes\cdots\otimes\mathfrak{m}_{k}(x_{i},\cdots,x_{i+k-1})\otimes\cdots\otimes x_{n}$ (3.1) where $*=\deg x_{1}+\cdots+\deg x_{i-1}+i-1$. The relation $\widehat{d}\,\widehat{d}=0$ can be written as $\sum_{k=1}^{n-1}\sum_{i=1}^{k-i+1}(-1)^{*}\mathfrak{m}_{n-k+1}(x_{1}\otimes\cdots\otimes\mathfrak{m}_{k}(x_{i},\cdots,x_{i+k-1})\otimes\cdots\otimes x_{n})=0,$ where $*$ is the same as above. In particular, we have $\mathfrak{m}_{1}\mathfrak{m}_{1}=0$ and so it defines a complex $(C,\mathfrak{m}_{1})$. A weak (or curved) $A_{\infty}$-algebra is defined in the same way, except that it also includes the $\mathfrak{m}_{0}$-term $\mathfrak{m}_{0}:R\to B(C[1]).$ The first two terms of the $A_{\infty}$ relation for a weak $L_{\infty}$ algebra are given as $\mathfrak{m}_{1}(\mathfrak{m}_{0}(1))=0,\quad\mathfrak{m}_{1}\mathfrak{m}_{1}(x)+(-1)^{\deg x+1}\mathfrak{m}_{2}(x,\mathfrak{m}_{0}(1))+\mathfrak{m}_{2}(\mathfrak{m}_{0}(1),x)=0.$ In particular, for the case of weak $A_{\infty}$ algebras, $\mathfrak{m}_{1}$ will not satisfy boundary property, i.e., $\mathfrak{m}_{1}\mathfrak{m}_{1}\neq 0$ in general. We now recall the notion of _unit_ in $A_{\infty}$ algebra. ###### Definition 3.2. An element ${\bf e}\in C^{0}=C[1]^{-1}$ is called a unit if it satisfies 1. (1) $\mathfrak{m}_{k+1}(x_{1},\cdots,{\bf e},\cdots,x_{k})=0$ for $k\geq 2$ or $k=0$. 2. (2) $\mathfrak{m}_{2}({\bf e},x)=(-1)^{\deg x}\mathfrak{m}_{2}(x,{\bf e})=x$ for all $x$. Combining this definition of unit and (3.1), we have the following immediate lemma. ###### Lemma 3.3. Consider an $A_{\infty}$ algebra $(C[1],\mathfrak{m})$ over a ground ring $R$ for which $\mathfrak{m}_{0}(1)=\lambda{\bf e}$ for some $\lambda\in R$. Then $\mathfrak{m}_{1}\mathfrak{m}_{1}=0$. Now we explain the notion of the _filtered $A_{\infty}$ algebra_. We define the universal Novikov ring $\Lambda_{0,nov}$ by $\Lambda_{0,nov}=\left.\left\\{\sum_{i=1}^{\infty}a_{i}T^{\lambda_{i}}e^{n_{i}}\,\,\right|\,\,a_{i}\in R,n_{i}\in{\mathbb{Z}},\lambda_{i}\in{\mathbb{R}}_{\geq 0},\lim_{i\to\infty}\lambda_{i}=\infty\right\\}.$ This is a graded ring by defining $\deg T=0$, $\deg e=2$. Let $\Lambda_{0,nov}^{+}$ be its maximal ideal which consists of the elements $\sum_{i=1}^{\infty}a_{i}T^{\lambda_{i}}e^{n_{i}}$ with $\lambda_{i}>0$. Let $\bigoplus_{m\in{\mathbb{Z}}}C^{m}$ be the free graded $\Lambda_{0,nov}$ module over the basis $\\{\mathbb{v}_{i}\\}$. We define a filtration $\bigoplus_{m\in{\mathbb{Z}}}F^{\lambda}C^{m}$ on it such that $\\{T^{\lambda}\mathbb{v}_{i}\\}$ is a free basis of $\bigoplus_{m\in{\mathbb{Z}}}F^{\lambda}C^{m}$. Here $\lambda\in{\mathbb{R}}$, $\lambda\geq 0$. We call this filtration the energy filtration. (Our algebra $\bigoplus_{m\in{\mathbb{Z}}}C^{m}$ may not be finitely generated. So we need to take completion.) We denote by $C$ the completion of $\bigoplus_{m\in{\mathbb{Z}}}C^{m}$ with respect to the energy filtration. The filtration induces a natural non-Archimedean topology on $C$. A filtered $A_{\infty}$ algebra $(C,\mathfrak{m})$ is a weak $A_{\infty}$ algebra such that $A_{\infty}$ operators $\mathfrak{m}$ have the following properties : 1. (1) $\mathfrak{m}_{k}$ respect the energy filtration, 2. (2) $\mathfrak{m}_{0}(1)\in F^{\lambda}C^{1}$ with $\lambda>0$, 3. (3) The reduction $\mathfrak{m}_{k}\mod\Lambda_{0,nov}^{+}:B_{k}\overline{C}[1]\otimes R[e,e^{-1}]\to\overline{C}\otimes R[e,e^{-1}]$ does not contain $e$. More precisely speaking, it has the form $\overline{\mathfrak{m}}_{k}\otimes\,_{R}R[e,e^{-1}]$ where $\overline{\mathfrak{m}}_{k}:B_{k}\overline{C}[1]\to\overline{C}$ is an $R$ module homomorphism. (Here $\overline{C}$ is the free $R$ module over the basis $\mathbb{v}_{i}$.) (See Definition 3.2.20 [FOOO3] = Definition 7.20 [FOOO2].) ###### Remark 3.4. 1. (1) In [FOOO3] we assume $\mathfrak{m}_{0}=0$ for (unfiltered) $A_{\infty}$ algebra. On the other hand, $\mathfrak{m}_{0}=0$ is not assumed for filtered $A_{\infty}$ algebra. Filtered $A_{\infty}$ algebra satisfying $\mathfrak{m}_{0}=0$ is called to be strict. 2. (2) In this section, to be consistent with the exposition given in [FOOO3], we use the Novikov ring $\Lambda_{0,nov}$ which includes the variable $e$. In [FOOO3], the variable $e$ is used so that the operations $\mathfrak{m}_{k}$ become to have degree one for all $k$ (with respect to the shifted degree.) But for the applications of this paper, it is enough to use the ${\mathbb{Z}}_{2}$-grading and so encoding the degree with a formal parameter is not necessary. Therefore we will use the ring $\Lambda_{0}$ in other sections, which does not contain $e$. An advantage of using the ring $\Lambda_{0}$ is that it is a local ring while $\Lambda_{0,nov}$ is not. This makes it easier to use some standard results from commutative algebra in later sections. We would like to remark that as a ${\mathbb{Z}}_{2}$ graded complex a ${\mathbb{Z}}$ graded complex over $\Lambda_{0,nov}$ is equivalent to the complex over $\Lambda_{0}$. Next we explain how one can deform the given filtered $A_{\infty}$ algebra $(C,\mathfrak{m})$ by an element $b\in F^{\lambda}C[1]^{0}$ with $\lambda>0$, by re-defining the $A_{\infty}$ operators as $\mathfrak{m}_{k}^{b}(x_{1},\cdots,x_{k})=\mathfrak{m}(e^{b},x_{1},e^{b},x_{2},e^{b},x_{3},\cdots,x_{k},e^{b}).$ This defines a new weak $A_{\infty}$-algebra for arbitrary $b$. Here we simplify notations by writing $e^{b}=1+b+b\otimes b+\cdots+b\otimes\cdots\otimes b+\cdots.$ Note that each summand in this infinite sum has degree 0 in $C[1]$. When the ground ring is $\Lambda_{0,nov}$, the infinite sum will converge in the non-Archimedean topology since $b\in F^{\lambda}C[1]^{0}$ with $\lambda>0$ ###### Proposition 3.5. For $A_{\infty}$ algebra $(C,\mathfrak{m}_{k}^{b})$, $\mathfrak{m}_{0}^{b}\equiv 0\mod\Lambda_{0,nov}\\{\bf e\\}$ if and only if $b$ satisfies $\sum_{k=0}^{\infty}\mathfrak{m}_{k}(b,\cdots,b)\equiv 0\mod\Lambda_{0,nov}\\{\bf e\\}.$ (3.2) We call the equation (3.2) the _$A_{\infty}$ Maurer-Cartan equation_. ###### Definition 3.6. Let $(C,\mathfrak{m})$ be a filtered $A_{\infty}$ algebra in general and $BC[1]$ be its bar-complex. An element $b\in F^{\lambda}C[1]^{0}$ ($\lambda>0$) is called a _weak bounding cochain_ if it satisfies the equation (3.2). If the $b$ satisfies the strict equation $\sum_{k=0}^{\infty}\mathfrak{m}_{k}(b,\cdots,b)=0$ we call it a (strict) _bounding cochain_. From now on, we will also call a weak bounding cochain just a bounding cochain since we will mainly concern weak bounding cochains. In general a given $A_{\infty}$ algebra may or may not have a solution to (3.2). ###### Definition 3.7. A filtered $A_{\infty}$-algebra is called _weakly unobstructed_ if the equation (3.2) has a solution $b\in F^{\lambda}C[1]^{0}$ with $\lambda>0$. One can define a notion of gauge equivalence between two bounding cochains as described in section 4.3 [FOOO3] (=section 16 [FOOO2]). The way how a filtered $A_{\infty}$ algebra is attached to a Lagrangian submanifold $L\subset(M,\omega)$ arises as an $A_{\infty}$ deformation of the classical singular cochain complex including the instanton contributions. In particular, when there is no instanton contribution as in the case $\pi_{2}(M,L)=0$, it will reduce to an $A_{\infty}$ deformation of the singular cohomology in the chain level including all possible higher Massey products. We now describe the basic $A_{\infty}$ operators $\mathfrak{m}_{k}$ in the context of $A_{\infty}$ algebra of Lagrangian submanifolds. For a given compatible almost complex structure $J$, consider the moduli space ${\mathcal{M}}_{k+1}(\beta;L)$ of stable maps of genus zero. It is a compactification of $\\{(w,(z_{0},z_{1},\cdots,z_{k}))\mid\overline{\partial}_{J}w=0,\,z_{i}\in\partial D^{2},\,[w]=\beta\,\,\,\mbox{in}\,\pi_{2}(M,L)\\}/\sim$ where $\sim$ is the conformal reparameterization of the disc $D^{2}$. The expected dimension of this space is given by $n+\mu(\beta)-3+(k+1)=n+\mu(\beta)+k-2$. Now given $k$ singular chains $[P_{1},f_{1}],\cdots,[P_{k},f_{k}]\in C_{*}(L)$ of $L$, we put the cohomological grading $\mbox{deg}P_{i}=n-\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{i}$ and regard the chain complex $C_{*}(L)$ as the cochain complex $C^{\mathop{\kern 0.0pt{\rm dim}}\nolimits L-*}(L)$. We consider the fiber product $ev_{0}:{\mathcal{M}}_{k+1}(\beta;L)\times_{(ev_{1},\cdots,ev_{k})}(P_{1}\times\cdots\times P_{k})\to L,$ where $ev_{i}([w,(z_{0},z_{1},\cdots,z_{k})])=w(z_{i})$. A simple calculation shows that we have the expected degree $\mbox{deg}\left[{\mathcal{M}}_{k+1}(\beta;L)\times_{(ev_{1},\cdots,ev_{k})}(P_{1}\times\cdots\times P_{k}),ev_{0}\right]=\sum_{j=1}^{n}(\mbox{deg}P_{j}-1)+2-\mu(\beta).$ For each given $\beta\in\pi_{2}(M,L)$ and $k=0,\cdots$, we define $\mathfrak{m}_{k,\beta}(P_{1},\cdots,P_{k})=\left[{\mathcal{M}}_{k+1}(\beta;L)\times_{(ev_{1},\cdots,ev_{k})}(P_{1}\times\cdots\times P_{k}),ev_{0}\right]$ (3.3) and $\mathfrak{m}_{k}=\sum_{\beta\in\pi_{2}(M,L)}\mathfrak{m}_{k,\beta}\cdot T^{\omega(\beta)}e^{\mu(\beta)/2}$. Now we denote by $C[1]$ the completion of a _suitably chosen countably generated_ (singular) chain complex with $\Lambda_{0,nov}$ as its coefficients with respect to the non-Archimedean topology. (We regard $C[1]$ as a cochain complex.) Then by choosing a system of multivalued perturbations of the right hand side of (3.3) and a triangulation of its zero sets, the map $\mathfrak{m}_{k}:B_{k}(C[1])\to C[1]$ is defined, has degree 1 and continuous with respect to non-Archimedean topology. We extend $\mathfrak{m}_{k}$ as a coderivation $\widehat{\mathfrak{m}}_{k}:BC[1]\to BC[1]$ by (3.1). Finally we take the sum $\widehat{d}=\sum_{k=0}^{\infty}\widehat{\mathfrak{m}}_{k}:BC[1]\to BC[1].$ (3.4) A main theorem proven in [FOOO1, FOOO3] then is the following coboundary property ###### Theorem 3.8. (Theorem 3.5.11 [FOOO3] = Theorem 10.11 [FOOO2]) Let $L$ be an arbitrary compact relatively spin Lagrangian submanifold of an arbitrary tame symplectic manifold $(M,\omega)$. The coderivation $\widehat{d}$ is a continuous map that satisfies the $A_{\infty}$ relation $\widehat{d}\widehat{d}=0$. The $A_{\infty}$ algebra associated to $L$ in this way has the _homotopy unit_ , not a _unit_. In general a filtered $A_{\infty}$ algebra with homotopy unit canonically induces another filtered _unital_ $A_{\infty}$-algebra called a _canonical model_ of the given filtered $A_{\infty}$ algebra. In the geometric context of $A_{\infty}$ algebra associated to a Lagrangian submanifold $L\subset M$ of a general symplectic manifold $(M,\omega)$, the canonical model is defined on the cohomology group $H^{*}(L;\Lambda_{0,nov})$. We refer to [FOOO5] for a quick explanation of construction by summing over trees of this canonical model. Once the $A_{\infty}$ algebra is attached to each Lagrangian submanifold $L$, we then construct a filtered _$A_{\infty}$ bimodule_ $C(L,L^{\prime})$ for the transversal pair of Lagrangian submanifolds $L$ and $L^{\prime}$. Here $C(L,L^{\prime})$ is the free $\Lambda_{0,nov}$ module such that the set of its basis is identified with $L\cap L^{\prime}$. The filtered $A_{\infty}$ bimodule structure is by definition is a family of operators $\mathfrak{n}_{k_{1},k_{2}}:B_{k_{1}}(C(L)[1])\,\,\widehat{\otimes}_{\Lambda_{0,nov}}\,\,C(L,L^{\prime})\,\,\widehat{\otimes}_{\Lambda_{0,nov}}\,\,B_{k_{1}}(C(L^{\prime})[1])\to C(L,L^{\prime})$ for $k_{1},k_{2}\geq 0$. (Here $\widehat{\otimes}_{\Lambda_{0,nov}}$ is the completion of algebraic tensor product.) Let us briefly describe the definition of $\mathfrak{n}_{k_{1},k_{2}}$. A typical element of the tensor product $B_{k_{1}}(C(L)[1])\,\,\widehat{\otimes}_{\Lambda_{0,nov}}\,\,C(L,L^{\prime})\,\,\widehat{\otimes}_{\Lambda_{0,nov}}\,\,B_{k_{2}}(C(L^{\prime})[1])$ has the form $P_{1,1}\otimes\cdots,\otimes P_{1,k_{1}}\otimes\langle p\rangle\otimes P_{2,1}\otimes\cdots\otimes P_{2,k_{2}}$ with $p\in L\cap L^{\prime}$. Then the image $\mathfrak{n}_{k_{1},k_{2}}$ thereof is given by $\sum_{q,B}T^{\omega(B)}e^{\mu(B)/2}\\#\left({\mathcal{M}}(p,q;B;P_{1,1},\cdots,P_{1,k_{1}};P_{2,1},\cdots,P_{2,k_{2}})\right)\langle q\rangle.$ Here $B$ denotes homotopy class of Floer trajectories connecting $p$ and $q$, the summation is taken over all $(q,B)$ with $\text{vir.dim }{\mathcal{M}}(p,q;B;P_{1,1},\cdots,P_{1,k_{1}};P_{2,1},\cdots,P_{2,k_{2}})=0,$ and $\\#\left({\mathcal{M}}(p,q;B;P_{1,1},\cdots,P_{1,k_{1}};P_{2,1},\cdots,P_{2,k_{2}})\right)$ is the ‘number’ of elements in the ‘zero’ dimensional moduli space ${\mathcal{M}}(p,q;B;P_{1,1},\cdots,P_{1,k_{1}};P_{2,1},\cdots,P_{2,k_{2}})$. Here the moduli space ${\mathcal{M}}(p,q;B;P_{1,1},\cdots,P_{1,k_{1}};P_{2,1},\cdots,P_{2,k_{2}})$ is the Floer moduli space ${\mathcal{M}}(p,q;B)$ cut-down by intersecting with the given chains $P_{1,i}\subset L$ and $P_{2,j}\subset L^{\prime}$. ###### Theorem 3.9. (Theorem 3.7.21 [FOOO3] = Theorem 12.21 [FOOO2]) Let $(L,L^{\prime})$ be an arbitrary relatively spin pair of compact Lagrangian submanifolds. Then the family $\\{\mathfrak{n}_{k_{1},k_{2}}\\}$ define a left $(C(L),\mathfrak{m})$ and right $(C(L^{\prime}),\mathfrak{m}^{\prime})$ filtered $A_{\infty}$-bimodule structure on $C(L,L^{\prime})$. What this theorem means is explained below as Proposition 3.10. Let $B(C(L)[1])\,\widehat{\otimes}_{\Lambda_{0,nov}}\,C(L,L^{\prime})\,\widehat{\otimes}_{\Lambda_{0,nov}}\,B(C(L^{\prime})[1])$ be the completion of the direct sum of $B_{k_{1}}(C(L)[1])\,\widehat{\otimes}_{\Lambda_{0,nov}}\,C(L,L^{\prime})\,\widehat{\otimes}_{\Lambda_{0,nov}}\,B_{k_{2}}(C(L^{\prime})[1])$ over $k_{1}\geq 0$, $k_{2}\geq 0$. We define the boundary operator $\widehat{d}$ on it by using the maps $\mathfrak{n}_{k_{1},k_{2}}$ and $\mathfrak{m}_{k}$, $\mathfrak{m}^{\prime}_{k}$ as follows : $\displaystyle\widehat{d}((x_{1}\otimes\cdots\otimes x_{n})\otimes x\otimes(x^{\prime}_{1}\otimes\cdots\otimes x^{\prime}_{m}))$ $\displaystyle=\widehat{d}(x_{1}\otimes\cdots\otimes x_{n})\otimes x\otimes(x^{\prime}_{1}\otimes\cdots\otimes x^{\prime}_{m})$ $\displaystyle\quad+(-1)^{\deg x_{1}+\cdots+\deg x_{n}+\deg x+n+1}(x_{1}\otimes\cdots\otimes x_{n})\otimes x\otimes\widehat{d}(x^{\prime}_{1}\otimes\cdots\otimes x^{\prime}_{m})$ $\displaystyle\quad+\sum_{k_{1}\leq n}\sum_{k_{2}\leq m}(-1)^{\deg x_{1}+\cdots+\deg x_{n-k_{1}}+n-k_{1}}(x_{1}\otimes\cdots\otimes x_{n-k_{1}})$ $\displaystyle\hskip 65.44142pt\otimes\mathfrak{n}_{k_{1},k_{2}}((x_{n-k_{1}+1}\otimes\cdots\otimes x_{n})\otimes x\otimes(x^{\prime}_{1}\otimes\cdots\otimes x^{\prime}_{k_{2}}))$ $\displaystyle\hskip 65.44142pt\otimes(x^{\prime}_{k_{2}+1}\otimes\cdots\otimes x^{\prime}_{m}).$ Here $\widehat{d}$ in the second and the third lines are induced from $\mathfrak{m}$ and $\mathfrak{m}^{\prime}$ by Formula (3.4) respectively. ###### Proposition 3.10. The map $\widehat{d}$ satisfies $\widehat{d}\,\widehat{d}=0$. The $A_{\infty}$ bimodule structure, which define a boundary operator on the bar complex, induces an operator $\delta=\mathfrak{n}_{0,0}$ on a much smaller, ordinary free $\Lambda_{0,nov}$ module $C(L,L^{\prime})$ generated by the intersections $L\cap L^{\prime}$. However the boundary property of this Floer’s ‘boundary’ map $\delta$ again meets obstruction coming from the obstructions cycles of either $L$, $L^{\prime}$ or of both. We need to deform $\delta$ using suitable bounding cochains of $L,\,L^{\prime}$. In the case where both $L,\,L^{\prime}$ are weakly unobstructed, we can carry out this deformation of $\mathfrak{n}$ using weak bounding chains $b$ and $b^{\prime}$ of fibered $A_{\infty}$ algebras associated to $L$ and $L^{\prime}$ respectively, in a way similar to $\mathfrak{m}^{b}$. Namely we define $\delta_{b,b^{\prime}}:C(L,L^{\prime})\to C(L,L^{\prime})$ by $\delta_{b,b^{\prime}}(x)=\sum_{k_{1},k_{2}}\mathfrak{n}_{k_{1},k_{2}}(b^{\otimes k_{1}}\otimes x\otimes b^{\prime\otimes k_{2}})=\mathfrak{\widehat{n}}(e^{b},x,e^{b^{\prime}}).$ We can generalize the story to the case where $L$ has clean intersection with $L^{\prime}$, especially to the case $L=L^{\prime}$. In the case $L=L^{\prime}$ we have $\mathfrak{n}_{k_{1},k_{2}}=\mathfrak{m}_{k_{1}+k_{2}+1}$. So in this case, we have $\delta_{b,b^{\prime}}(x)=\mathfrak{m}(e^{b},x,e^{b^{\prime}}).$ In general $\delta_{b,b^{\prime}}$ does not satisfy the equation $\delta_{b,b^{\prime}}\delta_{b,b^{\prime}}=0$. It turns out that there is an elegant condition for $\delta_{b,b^{\prime}}\delta_{b,b^{\prime}}=0$ to hold in terms of the _potential function_ introduced in [FOOO3], which we explain in the next section. In the case $\delta_{b,b^{\prime}}\delta_{b,b^{\prime}}=0$ we define Floer cohomology by $HF((L,b),(L^{\prime},b^{\prime});\Lambda_{0,nov})=\mbox{Ker}\,\delta_{b,b^{\prime}}/\mbox{Im}\,\delta_{b,b^{\prime}}.$ Let $\Lambda_{nov}$ be the field of fraction of $\Lambda_{0,nov}$. We define $HF((L,b),(L^{\prime},b^{\prime});\Lambda_{nov})$ by extending the coefficient ring $\Lambda_{0,nov}$ to $\Lambda_{nov}$. Then $HF((L,b),(L^{\prime},b^{\prime});\Lambda_{nov})$ is invariant under the Hamiltonian isotopies of $L$ and $L^{\prime}$. Therefore we can use it to obtain the following result about non-displacement of Lagrangian submanifolds. ###### Theorem 3.11. (Theorem G [FOOO3] = Theorem G [FOOO2]) Assume $\delta_{b,b^{\prime}}\delta_{b,b^{\prime}}=0$. Let $\psi:X\to X$ be a Hamiltonian diffeomorphism such that $\psi(L)$ is transversal to $L^{\prime}$. Then we have $\\#(\psi(L)\cap L^{\prime})\geq\text{\rm rank}_{\Lambda_{nov}}HF((L,b),(L^{\prime},b^{\prime});\Lambda_{nov}).$ The Floer cohomology $HF((L,b),(L^{\prime},b^{\prime});\Lambda_{0,nov})$ with coefficient $\Lambda_{0,nov}$ is not invariant under the Hamiltonian isotopy. We however can prove the following Theorem 3.12. There exists an integer $a$ and positive numbers $\lambda_{i}$ ($i=1,\cdots,b$) such that $HF((L,b),(L^{\prime},b^{\prime});\Lambda_{0,nov})=\Lambda_{0,nov}^{\oplus a}\oplus\bigoplus_{i=1}^{b}(\Lambda_{0,nov}/T^{\lambda_{i}}\Lambda_{0,nov}).$ (See Theorem 6.1.20 [FOOO3] (= Theorem 24.20 [FOOO2]).) Let $\psi:X\to X$ be a Hamiltonian diffeomorphism and $\|\psi\|$ is its Hofer distance (see [H]) from the identity map. ###### Theorem 3.12. (Theorem J [FOOO3] = Theorem J [FOOO2]) If $\psi(L)$ is transversal to $L^{\prime}$, we have an inequality $\\#(\psi(L)\cap L^{\prime})\geq a+2\,\\#\\{i\mid\lambda_{i}\geq\|\psi\|\\}.$ In later sections, we will apply Theorems 3.11,3.12 to study non-displacement of Lagrangian fibers of toric manifolds. ## 4\. Potential function The $A_{\infty}$ structure defined on a countably generated chain complex $C(L,\Lambda_{0})$ itself explained in the previous section is not suitable for explicit calculations as in our study of toric manifolds. For this computational purpose, we work with the filtered $A_{\infty}$ structure on the canonical model defined on $H(L;\Lambda_{0})$ which has a _finite_ rank over $\Lambda_{0}$. Furthermore this has a strict unit ${\bf e}$ given by the dual of the fundamental class $PD([L])$. Recall $C(L,\Lambda_{0})$ itself has only a homotopy-unit. An element $b\in H^{1}(L;\Lambda_{+})$ is called a _weak bounding cochain_ if it satisfies the $A_{\infty}$ Maurer-Cartan equation $\sum_{k=0}^{\infty}\mathfrak{m}_{k}(b,\cdots,b)\equiv 0\mod PD([L])$ (4.1) where $\\{\mathfrak{m}_{k}\\}_{k=0}^{\infty}$ is the $A_{\infty}$ operators associated to $L$. $[L]\in H_{n}(L)$ is the fundamental class, and $PD([L])\in H^{0}(L)$ is its Poincaré dual. We denote by $\widehat{{\mathcal{M}}}_{\text{weak}}(L)$ the set of weak bounding cochains of $L$. We say $L$ is weakly unobstructed if $\widehat{{\mathcal{M}}}_{\text{weak}}(L)\neq\emptyset$. The moduli space ${\mathcal{M}}_{\text{weak}}(L)$ is then defined to be the quotient space of $\widehat{{\mathcal{M}}}_{\text{weak}}(L)$ by suitable gauge equivalence. (See chapter 3 and 4 [FOOO3] for more explanations.) ###### Lemma 4.1 (Lemma 3.6.32 [FOOO3] = Lemma 11.32 [FOOO2]). If $b\in\widehat{{\mathcal{M}}}_{\text{\rm weak}}(L)$ then $\delta_{b,b}\circ\delta_{b,b}=0$, where $\delta_{b,b}$ is the deformed Floer operator defined by $\delta_{b,b}(x)=\mathfrak{m}_{1}^{b}(x)=:\sum_{k,\ell\geq 0}\mathfrak{m}_{k+\ell+1}(b^{\otimes k},x,b^{\otimes\ell}).$ For $b\in\widehat{{\mathcal{M}}}_{\text{weak}}(L)$, we define $HF((L;b),(L;b))=\frac{\operatorname{Ker}(\delta_{b,b}:C\to C)}{\operatorname{Im}(\delta_{b,b}:C\to C)},$ where $C$ is an appropriate subcomplex of the singular chain complex of $L$. When $L$ is weakly unobstructed i.e., $\widehat{{\mathcal{M}}}_{\text{weak}}(C)\neq\emptyset$, we define a function $\mathfrak{PO}:\widehat{{\mathcal{M}}}_{\text{weak}}(C)\to\Lambda_{+}$ by the equation $\mathfrak{m}(e^{b})=\mathfrak{PO}(b)\cdot PD([L]).$ This is the _potential function_ introduced in [FOOO3]. ###### Theorem 4.2 (Proposition 3.7.17 [FOOO3] = Proposition 12.17 [FOOO2]). For each $b\in\widehat{{\mathcal{M}}}_{\text{\rm weak}}(L)$ and $b^{\prime}\in\widehat{{\mathcal{M}}}_{weak}(L^{\prime})$, the map $\delta_{b,b^{\prime}}$ defines a continuous map $\delta_{b,b^{\prime}}:CF(L,L^{\prime})\to CF(L,L^{\prime})$ that satisfies $\delta_{b,b^{\prime}}\delta_{b,b^{\prime}}=0$, provided $\mathfrak{PO}(b)=\mathfrak{PO}(b^{\prime}).$ (4.2) Therefore for each pair $(b,b^{\prime})$ of $b\in\widehat{{\mathcal{M}}}_{\text{\rm weak}}(L)$ and $b^{\prime}\in\widehat{{\mathcal{M}}}_{\text{\rm weak}}(L^{\prime})$ that satisfy (4.2), we define the _$(b,b^{\prime})$ -Floer cohomology_ of the pair $(L,L^{\prime})$ by $HF((L,b),(L^{\prime},b^{\prime});\Lambda_{nov})=\frac{\mbox{Ker}\delta_{b,b^{\prime}}}{\mbox{\rm Im }\delta_{b,b^{\prime}}}.$ Now in the rest of this section we state the main results concerning the detailed structure of the potential function for the case of Lagrangian fibers of toric manifolds. For the later analysis of examples, we recall from [FOOO1, FOOO3] that $\mathfrak{m}_{k}$ is further decomposed into $\mathfrak{m}_{k}=\sum_{\beta\in\pi_{2}(M,L)}\mathfrak{m}_{k,\beta}\otimes T^{\omega(\beta)}e^{\mu(\beta)/2}.$ Here $\mu$ is the Maslov index. Firstly we remove the grading parameter $e$ from the ground ring. Secondly to eliminate many appearance of $2\pi$ in front of the affine function $\ell_{i}$ in the exponents of the parameter $T$ later in this paper, we redefine $T$ as $T^{2\pi}$. Under this arrangement, we get the formal power series expansion $\mathfrak{m}_{k}=\sum_{\beta\in\pi_{2}(M,L)}\mathfrak{m}_{k,\beta}\otimes T^{\omega(\beta)/2\pi}$ (4.3) which we will use throughout the paper. Now we restrict to the case of toric manifold. Let $X=X_{\Sigma}$ be associated a complete regular fan $\Sigma$ (in other words $\Sigma$ is the normal fan of $X$), and $\pi:X\to\mathfrak{t}^{*}$ be the moment map of the action of the torus $T^{n}\cong T^{m}/K$. We make the identifications $\mathfrak{t}=Lie(T^{n})\cong N_{\mathbb{R}}^{n}\cong{\mathbb{R}}^{n},\,\mathfrak{t}^{*}\cong M_{\mathbb{R}}\cong({\mathbb{R}}^{n})^{*}.$ We will exclusively use $N_{\mathbb{R}}$ and $M_{\mathbb{R}}$ to be consistent with the standard notations in toric geometry instead of $\mathfrak{t}$ (or ${\mathbb{R}}^{n}$) and $\mathfrak{t}^{*}$ (or $({\mathbb{R}}^{n})^{*}$) as much as possible. Denote the image of $\pi:X\to M_{\mathbb{R}}$ by $P\subset M_{\mathbb{R}}$ which is the moment polytope of the $T^{n}$ action on $X$. We will prove the following in section 11. ###### Proposition 4.3. For any $u\in\text{\rm Int}P$, the fiber $L(u)$ is weakly unobstructed. Moreover we have the canonical inclusion $H^{1}(L(u);\Lambda_{+})\hookrightarrow{\mathcal{M}}_{\text{\rm weak}}(L(u)).$ Choose an integral basis $\text{\bf e}^{*}_{i}$ of $N$ and $\text{\bf e}_{i}$ be its dual basis on $M$. With this choice made, we identify $M_{\mathbb{R}}$ with ${\mathbb{R}}^{n}$ as long as its meaning is obvious from the context. Identifying $H_{1}(T^{n};{\mathbb{Z}})$ with $N\cong{\mathbb{Z}}^{n}$ via $T^{n}={\mathbb{R}}^{n}/{\mathbb{Z}}^{n}$, we regard $\text{\bf e}_{i}$ as a basis of $H^{1}(L(u);{\mathbb{Z}})$. The following immediately follows from definition. ###### Lemma 4.4. We write $\pi=(\pi_{1},\cdots,\pi_{n}):X\to M_{\mathbb{R}}$ using the coordinate of $M_{\mathbb{R}}$ associated to the basis $\text{\bf e}_{i}$. Let $S^{1}_{i}\subset T^{n}$ be the subgroup whose orbit represents $\text{\bf e}^{*}_{i}\in H_{1}(T^{n};{\mathbb{Z}})$. Then $\pi_{i}$ is proportional to the moment map of $S^{1}_{i}$ action on $X$. Let $b=\sum x_{i}\text{\bf e}_{i}\in H^{1}(L(u);\Lambda_{+})\subset{{\mathcal{M}}}_{\text{\rm weak}}(L(u)).$ We study the potential function $\mathfrak{PO}:H^{1}(L(u);\Lambda_{+})\to\Lambda_{+}.$ Once a choice of the family of bases $\\{\text{\bf e}_{i}\\}$ on $H^{1}(L(u);{\mathbb{Z}})$ for $u\in\text{\rm Int}\,P$ is made as above starting from a basis on $N$ , then we can regard this function as a function of $(x_{1},\cdots,x_{n})\in(\Lambda_{+})^{n}$ and $(u_{1},\cdots,u_{n})\in P\subset M_{\mathbb{R}}$. We denote its value by $\mathfrak{PO}(x;u)=\mathfrak{PO}(x_{1},\cdots,x_{n};u_{1},\cdots,u_{n})$. We put $y_{i}=e^{x_{i}}=\sum_{k=0}^{\infty}\frac{x_{i}^{k}}{k!}\in\Lambda_{0}.$ Let $\partial P=\bigcup_{i=1}^{m}\partial_{i}P$ be the decomposition of the boundary of the moment polytope into its faces of codimension one. ($\partial_{i}P$ is a polygon in an $n-1$ dimensional affine subspace of $M_{\mathbb{R}}$.) Let $\ell_{i}$ be the affine functions $\ell_{i}(u)=\langle u,v_{i}\rangle-\lambda_{i}\,\mbox{ for }\,i=1,\cdots,m$ appearing in Theorem 2.13. Then the followings hold from construction : 1. (1) $\ell_{i}\equiv 0$ on $\partial_{i}P$. 2. (2) $P=\\{u\in M_{\mathbb{R}}\mid\ell_{i}(u)\geq 0,i=1,\cdots,m\\}.$ 3. (3) The coordinates of the vectors $v_{i}=(v_{i,1},\cdots,v_{i,n})$ satisfy $v_{i,j}=\frac{\partial\ell_{i}}{\partial u_{j}}$ (4.4) and are all integers. ###### Theorem 4.5. Let $L(u)\subset X$ be as in Theorem 1.5 and $\ell_{i}$ be as above. Suppose $X$ is Fano. Then we can take the canonical model of $A_{\infty}$ structure of $L(u)$ over $u\in\text{\rm Int}\,P$ so that the potential function restricted to $\bigcup_{u\in\text{\rm Int }P}H^{1}(L(u);\Lambda_{+})\cong(\Lambda_{+})^{n}\times\text{\rm Int}\,P$ has the form $\displaystyle\mathfrak{PO}(x;u)$ $\displaystyle=$ $\displaystyle\sum_{i=1}^{m}y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}T^{\ell_{i}(u)}$ (4.5) $\displaystyle=$ $\displaystyle\sum_{i=1}^{m}e^{\langle v_{i},x\rangle}T^{\ell_{i}(u)}$ (4.6) where $(x;u)=(x_{1},\cdots,x_{n};u_{1},\cdots,u_{n})$ and $v_{i,j}$ is as in $(\ref{eq:vij})$. Theorem 4.5 is a minor improvement of a result from [CO] (see (15.1) of [CO]) and [Cho] : The case considered in [CO] corresponds to the case where $y_{i}\in U(1)\subset\\{z\in{\mathbb{C}}\mid|z|=1\\}$ and the case in [Cho] corresponds to the one where $y_{i}\in{\mathbb{C}}\setminus\\{0\\}$. The difference of Theorem 4.5 from the ones thereof is that $y_{i}$ is allowed to contain $T$, the formal parameter of the universal Novikov ring encoding the energy. For the non-Fano case, we prove the following slightly weaker statement. The proof will be given in section 11. ###### Theorem 4.6. Let $X$ be an arbitrary toric manifold and $L(u)$ be as above. Then there exist $c_{j}\in{\mathbb{Q}}$, $e_{j}^{i}\in{\mathbb{Z}}_{\geq 0}$ and $\rho_{j}>0$, such that $\sum_{i=1}^{m}e_{j}^{i}>0$ and $\displaystyle\mathfrak{PO}(x_{1},\cdots,x_{n};u_{1},\cdots,u_{n})$ $\displaystyle-$ $\displaystyle\sum_{i=1}^{m}y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}T^{\ell_{i}(u)}$ (4.7) $\displaystyle=$ $\displaystyle\sum_{j=1}c_{j}y_{1}^{v^{\prime}_{j,1}}\cdots y_{n}^{v^{\prime}_{j,n}}T^{\ell^{\prime}_{j}(u)+\rho_{j}},$ where $v^{\prime}_{j,k}=\sum_{i=1}^{m}e_{j}^{i}v_{i,k},\quad\ell^{\prime}_{j}=\sum_{i=1}^{m}e_{j}^{i}\ell_{i}.$ If there are infinitely many non-zero $c_{j}$’s, we have $\lim_{j\to\infty}\ell^{\prime}_{j}(u)+\rho_{j}=\infty.$ Moreover $\rho_{j}=[\omega]\cap\alpha_{j}$ for some $\alpha_{j}\in\pi_{2}(X)$ with nonpositive first Chern number $c_{1}(X)\cap[\alpha_{j}]$. We note that although $\mathfrak{PO}$ is defined originally on $(\Lambda_{+}^{{\mathbb{C}}})^{n}\times P$, Theorems 4.5 and 4.6 imply that $\mathfrak{PO}$ extends to a function on $(\Lambda_{0}^{{\mathbb{C}}})^{n}\times P$ in general, and to one on $(\Lambda_{0}^{\mathbb{C}})^{n}\times M_{\mathbb{R}}$ for the Fano case. Furthermore these theorems also imply the periodicity of $\mathfrak{PO}$ in $x_{i}$’s, $\mathfrak{PO}(x_{1},\cdots,x_{i}+2\pi\sqrt{-1},\cdots,x_{n};u)=\mathfrak{PO}(x_{1},\cdots,x_{n};u).$ (4.8) We write $\mathfrak{PO}_{0}=\sum_{i=1}^{m}y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}T^{\ell_{i}(u)}$ (4.9) to distinguish it from $\mathfrak{PO}$. We call $\mathfrak{PO}_{0}$ the leading order potential function. We will concern the existence of the bounding cochain $\mathfrak{x}$ for which the Floer cohomology $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x}))$ is not zero, and prove that critical points of the $\mathfrak{PO}^{u}$ (as a function of $y_{1},\cdots,y_{n}$) have this property. (Theorem 4.10.) This leads us to study the equation $\frac{\partial\mathfrak{PO}^{u}}{\partial y_{k}}(\mathfrak{y}_{1},\cdots,\mathfrak{y}_{n})=0,\qquad k=1,\cdots,n,$ (4.10) where $\mathfrak{y}_{i}\in\Lambda_{0}\setminus\Lambda_{+}$. We regard $\mathfrak{PO}^{u}$ as either a function of $x_{i}$ or of $y_{i}$. Since the variable ($x_{i}$ or $y_{i}$) is clear from situation, we do not mention it occasionally. ###### Proposition 4.7. We assume that the coordinates of the vertices of $P$ are rational. Then there exists $u_{0}\in\text{\rm Int}P\cap{\mathbb{Q}}^{n}$ such that for each ${\mathcal{N}}$ there exists $\mathfrak{y}_{1},\cdots,\mathfrak{y}_{n}\in\Lambda_{0}\setminus\Lambda_{+}$ satisfying : $\frac{\partial\mathfrak{PO}^{u_{0}}}{\partial y_{k}}(\mathfrak{y}_{1},\cdots,\mathfrak{y}_{n})\equiv 0,\mod T^{{\mathcal{N}}}\qquad k=1,\cdots,n.$ (4.11) Moreover there exists $\mathfrak{y}^{\prime}_{1},\cdots,\mathfrak{y}^{\prime}_{n}\in\Lambda_{0}\setminus\Lambda_{+}$ such that $\frac{\partial\mathfrak{PO}_{0}^{u_{0}}}{\partial y_{k}}(\mathfrak{y}^{\prime}_{1},\cdots,\mathfrak{y}^{\prime}_{n})=0,\qquad k=1,\cdots,n.$ (4.12) We will prove Proposition 4.7 in section 9 . ###### Remark 4.8. 1. (1) $u_{0}$ is independent of ${\mathcal{N}}$. But $\mathfrak{y}_{i}$ may depend on ${\mathcal{N}}$. (We believe it does not depend on ${\mathcal{N}}$, but are unable to prove it at the time of writing this paper.) 2. (2) If $[\omega]\in H^{2}(X;{\mathbb{R}})$ is contained in $H^{2}(X;{\mathbb{Q}})$ then we may choose $P$ so that its vertices are rational. 3. (3) We believe that rationality of the vertices of $P$ is superfluous. We also believe there exists not only a solution of (4.11) or of (4.12) but also of (4.10). However then the proof seems to become more cumbersome. Since we can reduce the general case to the rational case by approximation in most of the applications, we will be content to prove the above weaker statement in this paper. We put $\mathfrak{x}_{i}=\log\mathfrak{y}_{i}\in\Lambda_{0}$ and write $\mathfrak{x}=\sum_{i}\mathfrak{x}_{i}\text{\bf e}_{i}\in H^{1}(L(u_{0});\Lambda_{0}).$ (4.13) Since $\mathfrak{y}_{i}\in\Lambda_{0}\setminus\Lambda_{+}$, $\log\mathfrak{y}_{i}$ is well-defined (by using non-Archimedean topology on $\Lambda_{0}$) and is contained in $\Lambda_{0}$. We remark that $\mathfrak{x}_{i}$ is determined from $\mathfrak{y}_{i}$ up to addtion by an element of $2\pi\sqrt{-1}{\mathbb{Z}}$. It will follow from (4.8) that changing $\mathfrak{x}_{i}$ by an element of $2\pi\sqrt{-1}{\mathbb{Z}}$ does not change corresponding Floer cohomology. So we take for example $\text{\rm Im}\,\mathfrak{x}_{i}\in[0,2\pi)$. (See also Definition 1.2 (2).) Let $\mathfrak{y}_{i,0}\in{\mathbb{C}}\setminus\\{0\\}$ be the zero-order term of $\mathfrak{y}_{i}$ i.e., the complex number such that $\mathfrak{y}_{i}-\mathfrak{y}_{i,0}\equiv 0\mod\Lambda_{+}^{{\mathbb{C}}}.$ If we put an additional assumption that $\mathfrak{y}_{i,0}=1$ for $i=1,\cdots,n$, then $\mathfrak{x}$ lies in $H^{1}(L(u_{0});\Lambda_{+})\subset H^{1}(L(u_{0});\Lambda_{0}).$ Therefore Proposition 4.3 implies the Floer cohomology $HF((L(u_{0}),\mathfrak{x}),(L(u_{0}),\mathfrak{x});\Lambda_{0})$ is defined. Then (4.11) combined with the argument from [CO] (see section 13) imply $HF((L(u_{0}),\mathfrak{x}),(L(u_{0}),\mathfrak{x});\Lambda_{0}^{{\mathbb{C}}}/(T^{{\mathcal{N}}}))\cong H(T^{n};\Lambda_{0}^{{\mathbb{C}}}/(T^{{\mathcal{N}}})).$ (4.14) We now consider the case when $\mathfrak{y}_{i,0}\neq 1$ for some $i$. In this case, we follow the idea of Cho [Cho] of twisting the Floer cohomology of $L(u)$ by non-unitary flat line bundle and proceed as follows : We define $\rho:H_{1}(L(u);{\mathbb{Z}})\to{\mathbb{C}}\setminus\\{0\\}$ by $\rho(\text{\bf e}_{i}^{*})=\mathfrak{y}_{i,0}.$ (4.15) Let $\mathfrak{L}_{\rho}$ be the flat complex line bundle on $L(u)$ whose holonomy representation is $\rho$. We use $\rho$ to twist the operator $\mathfrak{m}_{k}$ in the same way as [Fu3], [Cho] to obtain a filtered $A_{\infty}$ algebra, which we write $(H(L(u);\Lambda_{0}),\mathfrak{m}^{{\rho}})$. It is weakly unobstructed and ${\mathcal{M}}_{\text{\rm weak}}(H(L(u);\Lambda_{0}),\mathfrak{m}^{{\rho}})\supseteq H^{1}(L(u);\Lambda_{+})$. (See section 12.) We deform the filtered $A_{\infty}$ structure $\mathfrak{m}^{{\rho}}$ to ${\mathfrak{m}^{{\rho},b}}$ using $b\in H^{1}(L(u);\Lambda_{+})$ for which $\mathfrak{m}^{\rho,b}_{1}\mathfrak{m}^{\rho,b}_{1}=0$ holds. Denote by $HF((L(u_{0}),{\rho},b),(L(u_{0}),{\rho},b);\Lambda_{0}^{{\mathbb{C}}})$ the cohomology of $\mathfrak{m}^{{\rho},b}_{1}$. We denote the potential function of $(H(L(u);\Lambda_{0}),\mathfrak{m}^{{\rho}})$ by $\mathfrak{PO}^{u}_{\rho}:H^{1}(L(u);\Lambda_{+})\to\Lambda_{+}$ which is defined in the same way as $\mathfrak{PO}^{u}$ by using $\mathfrak{m}^{\rho}$ instead of $\mathfrak{m}$. Let $\mathfrak{x}$ be as in (4.13) and put $\mathfrak{x}_{i,0}=\log\mathfrak{y}_{i,0},\qquad b=\sum(\mathfrak{x}_{i}-\mathfrak{x}_{i,0})\text{\bf e}_{i}\in H^{1}(L(u);\Lambda_{+}).$ (4.16) From the way how the definition goes, we can easily prove ###### Lemma 4.9. $\mathfrak{PO}^{u}_{\rho}(b)=\mathfrak{PO}^{u}(\mathfrak{x}).$ We note from the remark right after Theorem 4.6 that $\mathfrak{PO}^{u}$ has been extended to a function on $(\Lambda_{0}^{\mathbb{C}})^{n}$ and hence the right hand side of the identity in this lemma has a well-defined meaning. Lemma 4.9 will be proved in section 13. Now we have : ###### Theorem 4.10. Let $\mathfrak{x}_{i}$, $\mathfrak{y}_{i}=e^{\mathfrak{x}_{i}}$, and $\rho$ satisfy $(\ref{formula:critical})$, $(\ref{frakxdef})$ and $(\ref{repdef})$. Let $\mathfrak{x}_{i,0}$ and $b$ be as in $(\ref{bdef})$. Then we have $HF((L(u_{0}),{\rho},b),(L(u_{0}),{\rho},b),\Lambda_{0}^{{\mathbb{C}}})\cong H(T^{n};\Lambda_{0}^{{\mathbb{C}}}).$ (4.17) If $(\ref{formula:criticalweak})$, $(\ref{frakxdef})$, $(\ref{repdef})$ and $(\ref{bdef})$ are satisfied instead then we have $HF((L(u_{0}),{\rho},b),(L(u_{0}),{\rho},b),\Lambda_{0}^{{\mathbb{C}}}/(T^{{\mathcal{N}}}))\cong H(T^{n};\Lambda_{0}^{{\mathbb{C}}}/(T^{{\mathcal{N}}})).$ (4.18) Theorem 4.10 is proved in section 13. Using this we prove Theorem 1.5 in section 13. More precisely, we will also discuss the following two points in that section : 1. (1) We need to study the case where $\omega$ is not necessarily rational. 2. (2) We only have (4.18) instead of (4.17). Next we give ###### Definition 4.11. Let $(X,\omega)$ be a smooth compact toric manifold, $P$ be its moment polytope. We say that a fiber $L(u_{0})$ at $u_{0}\in P$ is balanced if there exists a sequence $\omega_{i}$, $u_{i}$ such that 1. (1) $\omega_{i}$ is a $T^{n}$ invariant Kähler structure on $X$ such that $\lim_{i\to\infty}\omega_{i}=\omega$. 2. (2) $u_{i}$ is in the interior of the moment polytope $P_{i}$ of $(X,\omega_{i})$. We make an appropriate choice of moment polytope $P_{i}$ so that they converge to $P$. Then $\lim_{i\to\infty}u_{i}=u_{0}$. 3. (3) For each ${{\mathcal{N}}}$, there exist a sufficiently large $i$ and $\mathfrak{x}_{i,{{\mathcal{N}}}}\in H^{1}(L(u_{i});\Lambda^{{\mathbb{C}}}_{0})$ such that $HF((L(u_{i}),\mathfrak{x}_{i,{{\mathcal{N}}}}),(L(u_{i}),\mathfrak{x}_{i,{{\mathcal{N}}}});\Lambda^{{\mathbb{C}}}/(T^{{{\mathcal{N}}}}))\cong H(T^{n};{\mathbb{C}})\otimes\Lambda^{{\mathbb{C}}}/(T^{{{\mathcal{N}}}}).$ We say that $L(u_{0})$ is strongly balanced if there exists $\mathfrak{x}\in H^{1}(L;\Lambda^{{\mathbb{C}}}_{0})$ such that $HF((L(u_{0}),\mathfrak{x}),(L(u_{0}),\mathfrak{x});\Lambda^{{\mathbb{C}}}_{0})\cong H(T^{n};{\mathbb{Q}})\otimes\Lambda^{{\mathbb{C}}}_{0}$. Obviously ‘strongly balanced’ implies ‘balanced’. The converse is not true in general. See Example 10.17. We also refer readers to Remark 13.9 for other characterizations of being balanced (or strongly balanced). Theorem 4.10 implies that $L(u_{0})$ in Proposition 4.7 is balanced. (Proposition 13.2.) We will prove the next following intersection result in section 13. Theorem 1.5 will then be a consequence of Propositions 4.7 and 4.12. ###### Proposition 4.12. If $L(u_{0})$ is a balanced Lagrangian fiber then the following holds for any Hamiltonian diffeomorphism $\psi:X\to X$. $\psi(L(u_{0}))\cap L(u_{0})\neq\emptyset.$ (4.19) If $\psi(L(u_{0}))$ is transversal to $L(u_{0})$ in addition, then $\\#(\psi(L(u_{0}))\cap L(u_{0}))\geq 2^{n}.$ (4.20) Denoting $\mathfrak{x}=b+\sum\mathfrak{x}_{i,0}\text{\bf e}_{i}$, we sometime write $HF((L(u_{0}),\mathfrak{x}),(L(u_{0}),\mathfrak{x}),\Lambda_{0})$ for $HF((L(u_{0}),{\rho},b),(L(u_{0}),{\rho},b),\Lambda_{0})$ from now on. We also define $\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{0}):=\\{(\rho,b)\mid\rho:\pi_{1}L(u)\to{\mathbb{C}}\setminus\\{0\\},b\in\mathcal{M}_{\text{\rm weak}}(H(L(u)),\mathfrak{m}^{\rho})\\}.$ Namely it is the set of pairs $(\rho,b)$ where $\rho$ is a holonomy of a flat ${\mathbb{C}}$ bundle over $L(u)$ and $b\in H(L(u);\Lambda_{+})$ is a weak bounding cochain of the filtered $A_{\infty}$ algebra associated to $L(u)$ and twisted by $\rho$. With this definition of $\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{0})$, we have : $H^{1}(L(u);\Lambda_{0})\subseteq\mathcal{M}_{\text{\rm weak}}(L(u);\Lambda_{0}).$ ## 5\. Examples In this section, we discuss various examples of toric manifolds which illustrate the results of section 4. ###### Example 5.1. Consider $X=S^{2}$ with standard symplectic form with area $2\pi$. The moment polytope of the standard $S^{1}$-action by rotations along an axis becomes $P=[0,1]$ after a suitable translation. We have $\ell_{1}(u)=u$, $\ell_{2}(u)=1-u$ and $\mathfrak{PO}(x;u)=e^{x}T^{u}+e^{-x}T^{1-u}=yT^{u}+y^{-1}T^{1-u}.$ The zero of $\frac{\partial\mathfrak{PO}^{u}}{\partial y}=T^{u}-y^{-2}T^{1-u}$ is $\mathfrak{y}=\pm T^{(1-2u)/2}.$ If $u\neq 1/2$ then $\log\mathfrak{y}=\frac{1-2u}{2}\log(\pm T)$ is not an element of universal Novikov ring. In particular, there is no critical point in $\Lambda^{{\mathbb{C}}}_{0}\setminus\Lambda^{{\mathbb{C}}}_{+}$. If $u=1/2$ then $\mathfrak{y}=\pm 1$. The case $\mathfrak{y}=1$ corresponds to $\mathfrak{x}=0$. We have $HF((L(1/2),0),(L(1/2),0);\Lambda_{0})\cong H(S^{1};\Lambda^{{\mathbb{C}}}_{0}).$ The other case $\mathfrak{y}=-1$, corresponds to a nontrivial flat bundle on $S^{1}$. ###### Example 5.2. We consider $X={\mathbb{C}}P^{n}$. Then $P=\\{(u_{1},\cdots,u_{n})\mid 0\leq u_{i},u_{1}+\cdots+u_{n}\leq 1\\},$ is a simplex. We have $\mathfrak{PO}(x_{1},\cdots,x_{n};u_{1},\cdots,u_{n})=\sum_{i=1}^{n}e^{x_{i}}T^{u_{i}}+e^{-\sum x_{i}}T^{1-\sum u_{i}}.$ We put $u=u_{0}=\left(\frac{1}{n+1},\cdots,\frac{1}{n+1}\right).$ Then $\mathfrak{PO}^{u_{0}}=(y_{1}+\cdots+y_{n}+y_{1}^{-1}y_{2}^{-1}\cdots y_{n}^{-1})T^{1/(n+1)}.$ Solutions of the equation (4.10) are given by $\mathfrak{y}_{1}=\cdots=\mathfrak{y}_{n}=e^{2\pi k\sqrt{-1}/{(n+1)}},\quad k=0,\cdots,n.$ Hence the conclusion of Theorem 1.5 holds for our torus. The case $k=0$ corresponds to $b=0$. The other cases correspond to appropriate flat bundles on $T^{n}$. ###### Remark 5.3. The critical values of the potential function is $(n+1)e^{2\pi\sqrt{-1}k/(n+1)}$, $k=0,\cdots,n$. We consider the quantum cohomology ring $QH({\mathbb{C}}P^{n};\Lambda_{0})\cong\Lambda_{0}[z,T]/(z^{n+1}-T).$ The first Chern class $c_{1}$ is $(n+1)z$. The eigenvalues of the operator $c:QH({\mathbb{C}}P^{n})\to QH({\mathbb{C}}P^{n}),\alpha\mapsto c_{1}\cup_{Q}\alpha$ are $(n+1)e^{2\pi\sqrt{-1}k/(n+1)}$, $k=0,\cdots,n$. It coincides with the set of critical values. Kontsevich announced this statement at the homological mirror symmetry conference at Vienna 2006. (According to some physicists, this statement is known to them before.) See [Aur]. In our situation of Lagrangian fiber of compact toric manifolds, we can prove it by using Theorem 1.9. In the rest of this subsection, we discuss 2 dimensional examples. Let $\text{\bf e}_{1}$, $\text{\bf e}_{2}$ be the basis of $H^{1}(T^{2};{\mathbb{Z}})$ as in Lemma 4.4. We put $\text{\bf e}_{12}=\text{\bf e}_{1}\cup\text{\bf e}_{2}\in H^{2}(T^{2};{\mathbb{Z}})$. Let $\text{\bf e}_{\emptyset}$ be the standard basis of $H^{0}(T^{2};{\mathbb{Z}})\cong{\mathbb{Z}}$. The proof of the following proposition will be postponed until section 13. ###### Proposition 5.4. Let $\mathfrak{x}=\sum\mathfrak{x}_{i}\text{\bf e}_{i}\in H^{1}(L(u);\Lambda_{0})$, $\mathfrak{y}_{i}=e^{\mathfrak{x}_{i}}$. Then the boundary operator $\mathfrak{m}_{1}^{\mathfrak{x}}$ is given as follows : $\left\\{\begin{aligned} \mathfrak{m}_{1}^{\mathfrak{x}}(\text{\bf e}_{i})&=\frac{\partial\mathfrak{PO}^{u}}{\partial y_{i}}(\mathfrak{y})\text{\bf e}_{\emptyset},\\\ \mathfrak{m}_{1}^{\mathfrak{x}}(\text{\bf e}_{12})&=\frac{\partial\mathfrak{PO}^{u}}{\partial y_{1}}(\mathfrak{y})\text{\bf e}_{2}-\frac{\partial\mathfrak{PO}^{u}}{\partial y_{2}}(\mathfrak{y})\text{\bf e}_{1},\\\ \mathfrak{m}_{1}^{\mathfrak{x}}(\text{\bf e}_{\emptyset})&=0.\end{aligned}\right.$ (5.1) We remark that we do not use the grading parameter $e$, which was introduced in [FOOO3]. So the boundary operator $\mathfrak{m}_{1}^{\mathfrak{x}}$ is of degree $-1$ rather than $+1$. (Note we are using cohomology notation.) In other words, our Floer cohomology is only of ${\mathbb{Z}}_{2}$ graded. With (5.1) in our disposal, we examine various examples. ###### Example 5.5. We consider $M={\mathbb{C}}P^{2}$ again. We put $u_{1}=\epsilon+1/3$, $u_{2}=1/3$. ($\epsilon>0$.) Using (5.1) we can easily find the following isomorphism for the Floer cohomology with $\Lambda_{0}$ coefficients : $HF^{odd}((L(u),0),(L(u),0))\cong HF^{even}((L(u),0),(L(u),0))\cong\Lambda_{0}/(T^{1/3-\epsilon}).$ Let us apply Theorem J [FOOO3] (= Theorem 3.12) in this situation. (See also Theorem 5.11 below.) We consider a Hamiltonian diffeomorphism $\psi:{\mathbb{C}}P^{2}\to{\mathbb{C}}P^{2}$. We denote by $\|\psi\|$ the Hofer distance of $\psi$ from identity. Then we have $\\#(\psi(L(u))\cap L(u))\geq 4$ if $\|\psi\|<2\pi(\frac{1}{3}-\epsilon)$ and $\psi(L(u))$ is transversal to $L(u)$. We remark $\omega\cap[{\mathbb{C}}P^{1}]=2\pi$ by (2.12). We remark that this fact was already proved by Chekanov [Che]. (Actually the basic geometric idea behind our proof is the same as Chekanov’s.) ###### Example 5.6. Let $M=S^{2}(\frac{a}{2})\times S^{2}(\frac{b}{2})$, where $S^{2}(r)$ denotes the 2-sphere with radius $r$. We assume $a<b$. Then $B=[0,a]\times[0,b]$ and we have : $\mathfrak{PO}(x_{1},x_{2};u_{1},u_{2})=y_{1}T^{u_{1}}+y_{2}T^{u_{2}}+y_{1}^{-1}T^{a-u_{1}}+y_{2}^{-1}T^{b-u_{2}}.$ Let us take $u_{1}=a/2$, $u_{2}=b/2$. Then $\frac{\partial\mathfrak{PO}^{u}}{\partial y_{1}}=(1-y_{1}^{-2})T^{a/2},\quad\frac{\partial\mathfrak{PO}^{u}}{\partial y_{2}}=(1-y_{2}^{-2})T^{b/2}.$ Therefore $y_{1}=\pm 1$ $y_{2}=\pm 1$ are solutions of (4.10). Hence we can apply Theorem 4.10 to our torus. We next put $u_{1}=a/2$, $a<2u_{2}<b$. Then $\frac{\partial\mathfrak{PO}^{u}}{\partial y_{1}}=(1-y_{1}^{-2})T^{a/2},\quad\frac{\partial\mathfrak{PO}^{u}}{\partial y_{2}}=T^{u_{2}}-y_{2}^{-2}T^{b-u_{2}}.$ We put $y_{1}=y_{2}=1$. Then $\frac{\partial\mathfrak{PO}^{u}}{\partial y_{1}}=0$, $\frac{\partial\mathfrak{PO}^{u}}{\partial y_{2}}\neq 0$. We find that $HF^{odd}((L(u),0),(L(u),0))\cong HF^{even}((L(u),0),(L(u),0))\cong\Lambda_{0}/(T^{u_{2}}).$ Let $\psi:{\mathbb{C}}P^{2}\to{\mathbb{C}}P^{2}$ be a Hamiltonian diffeomorphism. Then, Theorem J [FOOO3] (= Theorem 3.12) implies that $\\#(\psi(L(u))\cap L(u))\geq 4$ if $\|\psi\|<2\pi u_{2}$ and $\psi(L(u))$ is transversal to $L(u)$. Note there exists a pseudo-holomorphic disc with symplectic area $\pi a$ ($<2\pi u_{2}$). Hence our result improves a result from [Che]. ###### Example 5.7. Let $X$ be the two-point blow up of ${\mathbb{C}}P^{2}$. We may take its Kähler form so that the moment polytope is given by $P=\\{(u_{1},u_{2})\mid-1\leq u_{1}\leq 1,-1\leq u_{2}\leq 1,u_{1}+u_{2}\leq 1+\alpha\\},$ where $-1<\alpha<1$ depends on the choice of Kähler form. We have $\displaystyle\mathfrak{PO}(x_{1},x_{2};u_{1},u_{2})=y_{1}T^{1+u_{1}}$ $\displaystyle+y_{2}T^{1+u_{2}}+y_{1}^{-1}T^{1-u_{1}}$ (5.2) $\displaystyle+y_{2}^{-1}T^{1-u_{2}}+y_{1}^{-1}y_{2}^{-1}T^{1+\alpha- u_{1}-u_{2}}.$ Note $X$ is Fano in our case. (Case 1: $\alpha=0$). In this case $X$ is monotone. We put $u_{0}=(0,0)$. $L(u_{0})$ is a monotone Lagrangian submanifold. We have $\frac{\partial\mathfrak{PO}^{u_{0}}}{\partial y_{1}}=(1-y_{1}^{-2}-y_{1}^{-2}y_{2}^{-1})T,\quad\frac{\partial\mathfrak{PO}^{u_{0}}}{\partial y_{2}}=(1-y_{2}^{-2}-y_{1}^{-1}y_{2}^{-2})T.$ The solutions of (4.10) are given by $y_{2}=\frac{1}{y_{1}^{2}-1}$, $y_{1}^{5}+y_{1}^{4}-2y_{1}^{3}-2y_{1}^{2}+1=0$ in ${\mathbb{C}}$. (There are 5 solutions.) (Case 2: $\alpha>0$). We put $u_{0}=(0,0)$. Then $\frac{\partial\mathfrak{PO}^{u_{0}}}{\partial y_{1}}=(1-y_{1}^{-2})T-y_{1}^{-2}y_{2}^{-1}T^{1+\alpha},\quad\frac{\partial\mathfrak{PO}^{u_{0}}}{\partial y_{2}}=(1-y_{2}^{-2})T-y_{1}^{-1}y_{2}^{-2}T^{1+\alpha}.$ We consider, for example, the case $y_{1}=y_{2}=\tau$. Then (4.10) becomes $\tau^{3}-\tau-T^{\alpha}=0.$ (5.3) The solution of (5.3) with $\tau\equiv 1\mod\Lambda_{+}$ is given by $\tau=1+\frac{1}{2}T^{\alpha}-\frac{3}{8}T^{2\alpha}+\frac{1}{2}T^{3\alpha}+\sum_{k=4}^{\infty}c_{k}T^{k\alpha}.$ Let us put ${\mathfrak{x}}=x_{1}\text{\bf e}_{1}+x_{2}\text{\bf e}_{2}$ with $x_{1}=x_{2}=\log\left(1+\frac{1}{2}T^{\alpha}-\frac{3}{8}T^{2\alpha}+\frac{1}{2}T^{3\alpha}+\cdots\right)\in\Lambda_{+}.$ Then by Theorem 4.10 we have $HF((L(u_{0}),{\mathfrak{x}}),(L(u_{0}),{\mathfrak{x}});\Lambda_{0})\cong H(T^{2};\Lambda_{0}).$ We like to point out that in this example it is essential to deform Floer cohomology using an element ${\mathfrak{x}}$ of $H^{1}(L(u_{0});\Lambda_{+})$ containing the formal parameter $T$ to obtain nonzero Floer cohomology. At $u_{0}$, there are actually 4 solutions such that $(y_{1},y_{2})\equiv(1,1),(1,-1),(-1,1),(-1,-1)\mod\Lambda_{+},$ respectively. In the current case there is another point $u^{\prime}_{0}=(\alpha,\alpha)\in P$ at which $L(u^{\prime}_{0})$ is balanced 111Using the method of spectral invariants and symplectic quasi-states, Entov and Polterovich discovered some non-displaceable Lagrangian fiber which was not covered by the criterion given in [CO] (see section 9 [EP1]). Recently this example, among others, was explained by Cho [Cho] via Lagrangian Floer cohomology twisted by non-unitary line bundles.. In fact at $u^{\prime}_{0}=(\alpha,\alpha)$ the equation (5.3) becomes $0=-(y_{1}^{-2}y_{2}^{-1}+y_{1}^{-2})T^{1-\alpha}+T^{1+\alpha},\quad 0=-(y_{1}^{-1}y_{2}^{-2}+y_{2}^{-2})T^{1-\alpha}+T^{1+\alpha}.$ we put $\tau=y_{1}=y_{2}$ to obtain $\tau^{3}T^{2\alpha}-\tau-1=0$ This equation has a unique solution with $\tau\equiv-1\mod\Lambda_{+}$. (The other solution is $T^{2\alpha}\tau^{3}\equiv 1\mod\Lambda_{+}$, for which Theorem 4.10 is not applicable.) The total number of the solutions $({\mathfrak{x}},u)$ is $5$. (Case 3: $\alpha<0$). We first consider $u_{0}=(0,0)$. Then $\frac{\partial\mathfrak{PO}^{u_{0}}}{\partial y_{1}}=-y_{1}^{-2}y_{2}^{-1}T^{1+\alpha}+(1-y_{1}^{-2})T,\quad\frac{\partial\mathfrak{PO}^{u_{0}}}{\partial y_{2}}=-y_{1}^{-1}y_{2}^{-2}T^{1+\alpha}+(1-y_{2}^{-2})T.$ We assume $y_{i}$ satisfies (4.10). It is then easy to see that $y_{1}^{-1}\equiv 0$, or $y_{2}^{-1}\equiv 0\mod\Lambda_{+}$. In other words, there is no $(y_{1},y_{2})$ to which we can apply Theorem 4.10. Actually it is easy to find a Hamiltonian diffeomorphism $\psi:X\to X$ such that $\psi(L(u_{0}))\cap L(u_{0})=\emptyset$. We next take $u^{\prime}_{0}=(\alpha/3,\alpha/3)$. Then $\displaystyle\frac{\partial\mathfrak{PO}^{u^{\prime}_{0}}}{\partial y_{1}}$ $\displaystyle=(1-y_{1}^{-2}y_{2}^{-1})T^{1+\alpha/3}-y_{1}^{-2}T^{1-\alpha/3},$ $\displaystyle\frac{\partial\mathfrak{PO}^{u^{\prime}_{0}}}{\partial y_{2}}$ $\displaystyle=(1-y_{1}^{-1}y_{2}^{-2})T^{1+\alpha/3}-y_{2}^{-2}T^{1-\alpha/3}.$ By putting $y_{1}=y_{2}=\tau$ for example, (4.10) becomes $\tau^{3}-T^{-2\alpha/3}\tau-1=0.$ (5.4) Let us put ${\mathfrak{x}}=x_{1}\text{\bf e}_{1}+x_{2}\text{\bf e}_{2}$ with $x_{1}=x_{2}=\log\tau=\log\left(1+\frac{1}{3}T^{-2\alpha/3}-\frac{1}{81}T^{-6\alpha/3}+\cdots\right)\in\Lambda_{+},$ where $\tau$ solves (5.4). Theorem 4.10 is applicable. (There are actually 3 solutions of (4.10) corresponding to the 3 solutions of (5.4).) There are two more points $u=(\alpha+1,\alpha),(\alpha,\alpha+1)$ where (4.10) has a solution in $(\Lambda_{0}\setminus\Lambda_{+})$. Each $u$ has one solution $b$. Thus the total number of the pair $({\mathfrak{x}},u)$ is again 5. We remark $5=\sum\text{\rm rank}\,H^{k}(X;{\mathbb{Q}}).$ This is not just a coincidence but an example of general phenomenon stated as in Theorem 1.3. We remark that as $\alpha\to 1$ our $X$ blows down to $S^{2}(1)\times S^{2}(1)$. On the other hand, as $\alpha\to-1$ our $X$ blows down to ${\mathbb{C}}P^{2}$. The situation of the case $\alpha>0$ can be regarded as a perturbation of the situation of $S^{2}(1)\times S^{2}(1)$, by the effect of exceptional curve corresponding to the segment $u_{1}+u_{2}=1+\alpha$. The situation of the case $\alpha<0$ can be regarded as a perturbation of the situation of ${\mathbb{C}}P^{2}$ by the effect of the two exceptional curves corresponding to the segments $u_{1}=1$ and $u_{2}=1$. An interesting phase change occurs at $\alpha=0$. We remark that $H^{2}(X;{\mathbb{R}})$ is two dimensional. So there are actually two parameter family of symplectic structures. We study two points blow up of ${\mathbb{C}}P^{2}$ more in Example 10.17. The discussion of this section strongly suggests that Lagrangian Floer theory (Theorems G, J [FOOO3] = Theorems 3.11, 3.12) gives the optimal result for the study of non-displacement of Lagrangian fibers in toric manifolds. ###### Remark 5.8. Let $X$ be a compact toric manifold and $L(u)=\pi^{-1}(u)$, $u\in\text{\rm Int}P$. We consider the following two conditions : 1. (1) There exists no Hamiltonian diffeomorphism $\psi:X\to X$ such that $\psi(L(u))\cap L(u)=\emptyset.$ 2. (2) $L(u)$ is balanced. $(2)\Rightarrow(1)$ follows from Proposition 4.12. In many cases $(1)\Rightarrow(2)$ can be proved by the method of [Mc]. However there is a case $(1)\Rightarrow(2)$ does not follow, as we will mention in Remark 10.18 and will prove in a sequel of this series of papers, using the bulk deformation of Lagrangian Floer cohomology. We conjecture that after including this wider class of Floer cohomology, we can detect all the non-displacable Lagrangian fibers in toric manifolds, by Floer cohomology. Using the argument employed in Example 5.6 we can discuss the relationship between the Hofer distance and displacement. First we introduce some notations for this purpose. We denote by $Ham(X,\omega)$ the group of Hamiltonian diffeomorphisms of $(X,\omega)$. For a time-dependent Hamiltonian $H:[0,1]\times X\to{\mathbb{R}}$, we denote by $\phi_{H}^{t}$ the time $t$-map of Hamilton’s equation $\dot{x}=X_{H}(t,x)$. The Hofer norm of $\psi\in Ham(X,\omega)$ is defined to be $\|\psi\|=\inf_{H;\phi_{H}^{1}=\psi}\int_{0}^{1}(\max H_{t}-\min H_{t})\,dt.$ (See [H].) ###### Definition 5.9. Let $Y\subset X$. We define the displacement energy $e(Y)\in[0,\infty]$ by $e(Y):=\inf\\{\|\psi\|\mid\psi\in Ham(X,\omega),\,\psi(Y)\cap\overline{Y}=\emptyset\\}.$ We put $e(Y)=\infty$ if there exists no $\psi\in Ham(X,\omega)$ with $\psi(Y)\cap\overline{Y}=\emptyset$. Let us consider $\mathfrak{PO}(y_{1},\cdots,y_{n};u_{1},\cdots,u_{n}):\Lambda_{0}^{n}\times P\to\Lambda_{+}$ as in Theorem 4.5. ###### Definition 5.10. We define the number $\mathfrak{E}(u)\in(0,\infty]$ as the supremum of all $\lambda$ such that there exists $\mathfrak{y}_{1},\cdots,\mathfrak{y}_{n}\in(\Lambda_{0}\setminus\Lambda_{0}^{+})^{n}$ satisfying $\frac{\partial\mathfrak{PO}}{\partial y_{i}}(\mathfrak{y}_{1},\cdots,\mathfrak{y}_{n};u)\equiv 0\mod T^{\lambda}$ (5.5) for $i=1,\cdots,n$. (Here we consider universal Novikov ring with ${\mathbb{C}}$-coefficients.) We call $\mathfrak{E}(u)$ the _$\mathfrak{PO}$ -threshold_ of the fiber $L(u)$. We put $\overline{\mathfrak{E}}(u)=\limsup_{\omega_{i}\to\omega,u_{i}\to u}{\mathfrak{E}}(u_{i}).$ Here limsup is taken over all sequences $\omega_{i}$ and $u_{i}$ such that $\omega_{i}$ is a sequence of $T^{n}$ invariant symplectic structures on $X$ with $\lim_{i\to\infty}\omega_{i}=\omega$ and $u_{i}$ is a sequence of points of moment polytopes $P_{i}$ of $(X,\omega_{i})$ such that $P_{i}$ converges to $P$ and $u_{i}$ converges to $u$. Clearly $\overline{\mathfrak{E}}(u)\geq{\mathfrak{E}}(u)$. We will give an example where $\overline{\mathfrak{E}}(u)\neq{\mathfrak{E}}(u)$ in Example 10.17 (10.16). ###### Theorem 5.11. For any compact toric manifold $X$ and $L(u)=\pi^{-1}(u)$, $u\in\text{\rm Int}P$, we have $e(L(u))\geq 2\pi\overline{\mathfrak{E}}(u).$ (5.6) We will prove Theorem 5.11 in section 13. ###### Remark 5.12. The equality in $(\ref{displaceineq})$ holds in various examples. However there are cases that the equality in $(\ref{displaceineq})$ does not hold. The situation is the same as Remark 5.8. We like to remark $\mathfrak{E}(u)$, $\overline{\mathfrak{E}}(u)$ can be calculated in most of the cases once the toric manifold $X$ is given explicitly. In fact the leading order potential function $\mathfrak{PO}_{0}$ is explicitly calculated by Theorem 4.5. We can then find the maximal value $\lambda$ for which the polynomial equations $\frac{\partial{\mathfrak{PO}}_{0}}{\partial y_{i}}(u;\mathfrak{y}_{1},\cdots,\mathfrak{y}_{n})\equiv 0\mod T^{\lambda}$ has a solution $\mathfrak{y}_{i}\in\Lambda_{0}\setminus\Lambda_{+}$. In a weakly degenerate case this value of $\lambda$ for $\mathfrak{PO}(u;\cdots)$ is the same as the value for $\mathfrak{PO}_{0}(u;\cdots)$. (See section 10.) ###### Remark 5.13. Appearance of a new family of pseudo-holomorphic discs with Maslov index 2 after blow up, which we observed in Examples 5.7 can be related to the operator $\mathfrak{q}$ that we introduced in section 3.8 [FOOO3] (= section 13 [FOOO2]) in the following way. We denote by $\mathcal{M}^{\text{\rm main}}_{k+1,l}(\beta)$ the moduli space of stable maps $f:(\Sigma,\partial\Sigma)\to(X,L)$ from bordered Riemann surface $\Sigma$ of genus zero with $l$ interior and $k+1$ boundary marked points and in homology class $\beta$. (See section 3 [FOOO1] (= subsection 2.1.2 [FOOO3]). The symbol main means that we require the boundary marked points to respect the cyclic order of $\partial\Sigma$.) Let us consider the case when Maslov index of $\beta$ is $2n$. More precisely we take the following class $\beta$. We use notation introduced at the beginning of section 11. We put $\beta=\beta_{i_{1}}+\cdots+\beta_{i_{n}}$, where $\partial_{i_{1}}P\cap\cdots\cap\partial_{i_{n}}P=\overline{p}$ is an vertex of $P$. We assume $[f]\in\mathcal{M}^{\text{\rm main}}_{0+1,1}(\beta)$ and $\mathcal{M}^{\text{\rm main}}_{0+1,1}(\beta)$ is Fredholm regular at $f$. The virtual dimension of $\mathcal{M}^{\text{\rm main}}_{0+1,1}(\beta)$ is $3n$. Let us take the unique point $p\in X$ such that $\pi(p)=\overline{p}$. $p$ is the $T^{n}$ fixed point. We assume moreover $f(0)=p$. We blow up $X$ at a point $p=f(0)\in X$ and obtain $\widehat{X}$. Let $[E]\in H_{2n-2}(\widehat{X})$ be the homology class of the exceptional divisor $E=\pi^{-1}(p)$. Now $f$ induces a map $\widehat{f}:(\Sigma;\partial\Sigma)\to(\widehat{X},L)$. The Maslov index of the homology class $[\widehat{f}]\in H_{2}(\widehat{X},L)$ becomes $2$. We put $\widehat{\beta}=[\widehat{f}]$. Since $p$ is a fixed point of $T^{n}$ action, a $T^{n}$-invariant perturbation lifts to a perturbation of the moduli space $\mathcal{M}^{\text{\rm main}}_{0+1,0}(\widehat{\beta})$. Then any $T^{n}$-orbit of the moduli space $\mathcal{M}^{\text{\rm main}}_{0+1,0}(X;\beta)$ of holomorphic discs passing through $p$ corresponds to the $T^{n}$-orbit of $\mathcal{M}^{\text{\rm main}}_{0+1,0}(\widehat{X};\widehat{\beta})$ and vice versa. Namely we have an isomorphism $\mathcal{M}^{\text{\rm main}}_{0+1,1}(\beta)\,\,{}_{ev}\times_{X}\\{p\\}\cong\mathcal{M}^{\text{\rm main}}_{0+1,0}(\widehat{\beta}).$ (5.7) Here $ev$ in the left hand side is the evaluation map at the interior marked point. (Actually we need to work out analytic detail of gluing construction etc.. It seems very likely that we can do it in the same way as the argument of Chapter 10 [FOOO2]. See also [LiRu].) Using (5.7) we may prove : $\mathfrak{q}_{1,k;\beta}(PD([p]);b,\cdots,b)=\mathfrak{m}_{k,\widehat{\beta}}(b,\cdots,b),$ where $\mathfrak{q}_{1,k;\beta}(Q;P_{1},\cdots,P_{k})=ev_{0*}\left(\mathcal{M}^{\text{\rm main}}_{k+1,1}(\beta)\times_{(X\times L^{k})}(Q\times P_{1}\times\cdots\times P_{k})\right)$ is defined in section 3.8 [FOOO3] (=section 13 [FOOO2]). (Here $Q$ is a chain in $X$ and $P_{i}$ are chains in $L(u)$, and $ev_{0}:\mathcal{M}^{\text{\rm main}}_{k+1,1}(\beta)\to X$ is the evaluation map at the $0$-th boundary marked point. In the right hand side, we take fiber product over $X\times L^{k}$.) This is an example of a blow-up formula in Lagrangian Floer theory. ## 6\. Quantum cohomology and Jacobian ring In this section, we prove Theorem 1.9. Let $\mathfrak{PO}_{0}$ be the leading order potential function. (Recall if $X$ is Fano, we have $\mathfrak{PO}_{0}=\mathfrak{PO}$.) We define the monomial $\overline{z}_{i}(u)=y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}T^{\ell_{i}(u)}\in\Lambda_{0}[y_{1},\cdots,y_{n},y^{-1}_{1},\cdots,y^{-1}_{n}].$ (6.1) Compare this with (2.3). It is also suggestive to write $\overline{z}_{i}$ also as $\overline{z}_{i}(u)=e^{\langle x,v_{i}\rangle}T^{\ell_{i}(u)},\,x=(x_{1},\cdots,x_{n}),\,y_{i}=e^{x_{i}}.$ (6.2) By definition we have $\displaystyle\mathfrak{PO}_{0}^{u}$ $\displaystyle=$ $\displaystyle\sum_{i=1}^{m}\overline{z}_{i}(u)$ (6.3) $\displaystyle y_{j}\frac{\partial\overline{z}_{i}}{\partial y_{j}}$ $\displaystyle=$ $\displaystyle v_{i,j}\overline{z}_{i}(u).$ (6.4) The following is a restatement of Theorem 1.9. Let $z_{i}\in H^{2}(X;{\mathbb{Z}})$ be the Poincaré dual of the divisor $\pi^{-1}(\partial_{i}P)$. ###### Theorem 6.1. If $X$ is Fano, there exists an isomorphism $\psi_{u}:QH(X;\Lambda)\cong Jac(\mathfrak{PO})$ such that $\psi_{u}(z_{i})=\overline{z}_{i}(u)$. Since $c_{1}(X)=\sum_{i=1}^{m}z_{i}$ (see [Ful]) and $\mathfrak{PO}_{0}^{u}=\sum_{i=1}^{m}\overline{z}_{i}(u)$ by definition, Theorem 1.9 follows from Theorem 6.1. In the rest of this section, we prove Theorem 6.1. We remark that $z_{i}$ ($i=1,\cdots,m$) generates the quantum cohomology ring $QH(X;\Lambda)$ as a $\Lambda$-algebra (see Theorem 6.6 below). Therefore it is enough to prove that the assignment $\tilde{\psi}_{u}(z_{i})=\overline{z}_{i}(u)$ extends to a homomorphism $\tilde{\psi}_{u}:\Lambda[z_{1},\cdots,z_{m}]\to Jac(\mathfrak{PO}_{0}^{u})$ that induces an isomorphism in $QH(X;\Lambda)$. In other words, it suffices to show that the relations among the generators in $\Lambda[z_{1},\cdots,z_{m}]$ and in $Jac(\mathfrak{PO}_{0}^{u})$ are mapped to each other under the assignment $\tilde{\psi}_{u}(z_{i})=\overline{z}_{i}(u)$. To establish this correspondence, we now review Batyrev’s description of the relations among $z_{i}$’s. We first clarify the definition of quantum cohomology ring over the universal Novikov rings $\Lambda_{0}$ and $\Lambda$. Let $(X,\omega)$ be a symplectic manifold and $\alpha\in\pi_{2}(X)$. Let $\mathcal{M}_{3}(\alpha)$ be the moduli space of stable map with homology class $\alpha$ of genus $0$ with $3$ marked points. Let $ev:\mathcal{M}_{3}(\alpha)\to X^{3}$ be the evaluation map. We can define the virtual fundamental class $ev_{*}[\mathcal{M}_{3}(\alpha)]\in H_{d}(X^{3};{\mathbb{Q}})$ where $d=2(\mathop{\kern 0.0pt{\rm dim}}\nolimits_{{\mathbb{C}}}X+c_{1}(X)\cap\alpha).$ Let $a_{i}\in H^{*}(X;{\mathbb{Q}})$. We define $a_{1}\cup_{Q}a_{2}\in H^{*}(X;\Lambda_{0})$ by the following formula. $\langle a_{1}\cup_{Q}a_{2},a_{3}\rangle=\sum_{\alpha}T^{\omega\cap\alpha/2\pi}ev_{*}[\mathcal{M}_{3}(\alpha)]\cap(a_{1}\times a_{2}\times a_{3}).$ (6.5) Here $\langle\cdot,\cdot\rangle$ is the Poincaré duality. Extending this linearly we obtain the quantum product $\cup_{Q}:H(X;\Lambda_{0})\otimes H(X;\Lambda_{0})\to H(X;\Lambda_{0}).$ Extending the coefficient ring further to $\Lambda$, we obtain the (small) quantum cohomology ring $QH(X;\Lambda)$. Now we specialize to the case of compact toric manifolds and review Batyrev’s presentation of quantum cohomology ring. We consider the exact sequence $0\longrightarrow\pi_{2}(X)\longrightarrow\pi_{2}(X;L(u))\longrightarrow\pi_{1}(L(u))\longrightarrow 0.$ (6.6) We note $\pi_{2}(X;L(u))\cong{\mathbb{Z}}^{m}$ and choose its basis adapted to this exact sequence as follows : Consider the divisor $\pi^{-1}(\partial_{i}P)$ and take a small disc transversal to it. Each such disc gives rise to an element $[\beta_{i}]\in H_{2}(X;\pi^{-1}(\text{\rm Int}P))\cong H_{2}(X;L(u))\cong\pi_{2}(X,L(u)).$ (6.7) The set of $[\beta_{i}]$ with $i=1,\cdots,m$ forms a basis of $\pi_{2}(X;L(u))\cong{\mathbb{Z}}^{m}$. The boundary map $[\beta]\mapsto[\partial\beta]:\pi_{2}(X;L(u))\to\pi_{1}(L(u))$ is identified with the corresponding map $H_{2}(X;L(u))\to H_{1}(L(u))$. Using the basis chosen in Lemma 4.4 on $H_{1}(L(u))$ we identify $H_{1}(L(u))\cong{\mathbb{Z}}^{n}$. Then this homomorphism maps $[\beta_{i}]$ to $[\partial\beta_{i}]\cong v_{i}=(v_{i,1},\cdots,v_{i,n}),$ (6.8) where $v_{i,j}$ is as in (4.4). By the exactness of (6.6), we have an isomorphism $H_{2}(X)\cong\\{\beta\in H_{2}(X;L(u))\mid[\partial\beta]=0\\}.$ (6.9) ###### Lemma 6.2. We have $\omega\cap\left[\sum k_{i}\beta_{i}\right]=2\pi\sum k_{i}\ell_{i}(u).$ (6.10) If $\left[\sum k_{i}\partial\beta_{i}\right]=0$ then $\sum k_{i}\frac{d\ell_{i}}{du_{j}}=0.$ (6.11) In particular, the right hand side of $(\ref{form:symparea})$ is independent of $u$. ###### Proof. (6.10) follows from the area formula (2.12), $\omega(\beta_{i})=2\pi\ell_{i}(u).$ On the other hand if $\left[\sum k_{i}\partial\beta_{i}\right]=0$, we have $\sum_{i=1}^{m}k_{i}v_{i}=0.$ By the definition of $\ell_{i}$, $\ell_{i}(u)=\langle u,v_{i}\rangle-\lambda_{i}$, from Theorem 2.13, this equation is precisely (6.11) and hence the proof. ∎ Let $\mathcal{P}\subset\\{1,\cdots,m\\}$ be a primitive collection (see Definition 2.4). There exists a unique subset $\mathcal{P}^{\prime}\subset\\{1,\cdots,m\\}$ such that $\sum_{i\in\mathcal{P}}v_{i}$ lies in the interior of the cone spanned by $\\{v_{i^{\prime}}\mid i^{\prime}\in\mathcal{P}^{\prime}\\}$, which is a member of the fan $\Sigma$. (Since $X$ is compact, we can choose such $\mathcal{P}^{\prime}$. See section 2.4 [Ful].) We write $\sum_{i\in\mathcal{P}}v_{i}=\sum_{i^{\prime}\in\mathcal{P}^{\prime}}k_{i^{\prime}}v_{i^{\prime}}.$ (6.12) Since $X$ is assumed to be nonsingular $k_{i^{\prime}}$ are all positive integers. (See p.29 of [Ful].) We put $\omega(\mathcal{P})=\sum_{i\in\mathcal{P}}\ell_{i}(u)-\sum_{i^{\prime}\in\mathcal{P}^{\prime}}k_{i^{\prime}}\ell_{i^{\prime}}(u).$ (6.13) It follows from (6.10) that $2\pi\omega(\mathcal{P})$ is the symplectic area of the homotopy class $\beta(\mathcal{P})=\sum_{i\in\mathcal{P}}\beta_{i}-\sum_{i^{\prime}\in\mathcal{P}^{\prime}}k_{i^{\prime}}\beta_{i^{\prime}}\in\pi_{2}(X).$ (6.14) ###### Lemma 6.3. Let $\mathcal{P}$ be any primitive collection. Then $\omega(\mathcal{P})>0$. ###### Proof. Since the cone spanned by $\\{v_{i^{\prime}}|i^{\prime}\in{\mathcal{P}}^{\prime}\\}$ belongs to the fan $\Sigma$, we have $\bigcap_{i^{\prime}\in\mathcal{P}^{\prime}}\pi^{-1}(\partial_{i^{\prime}}P)\neq\emptyset.$ Then for any $u^{\prime}\in\bigcap_{i^{\prime}\in\mathcal{P}^{\prime}}\partial_{i^{\prime}}P$, we have $\ell_{i^{\prime}}(u^{\prime})=0$. By the choice of $k_{i^{\prime}}$ in (6.12), we have $\partial\beta(\mathcal{P})=0$. Therefore by Lemma 6.2 and by the continuity of the right hand side of (6.13) we can evaluate $\omega(\mathcal{P})$ at a point $u^{\prime}\in\bigcap_{i^{\prime}\in\mathcal{P}^{\prime}}\pi^{-1}(\partial_{i^{\prime}}P)$. Then we obtain $\omega(\mathcal{P})=\sum_{i\in\mathcal{P}}\ell_{i}(u^{\prime}).$ Since $\mathcal{P}$ is a primitive collection, and in particular does not form a member of the fan, there must be an element $v_{i}\in\mathcal{P}$ such that $\ell_{i}(u^{\prime})>0$ and so $\omega(\mathcal{P})>0$. This finishes the proof. ∎ Now we associate formal variables, $z_{1},\cdots,z_{m}$, to $v_{1},\cdots,v_{m}$, respectively. ###### Definition 6.4 (Batyrev [B1]). 1. (1) The quantum Stanley-Reisner ideal $SR_{\omega}(X)$ is the ideal generated by $z(\mathcal{P})=\prod_{i\in\mathcal{P}}z_{i}-T^{\omega(\mathcal{P})}\prod_{i^{\prime}\in\mathcal{P}^{\prime}}z_{i^{\prime}}^{k_{i^{\prime}}}$ (6.15) in the polynomial ring $\Lambda\left[z_{1},\cdots,z_{m}\right]$. Here $\mathcal{P}$ runs over all primitive collections. 2. (2) We denote by $P(X)$ the ideal generated by $\sum_{i=1}^{m}v_{i,j}z_{i}$ (6.16) for $j=1,\cdots,n$. In this paper we call $P(X)$ the linear relation ideal. 3. (3) We call the quotient $QH^{\omega}(X;\Lambda)=\frac{\Lambda\left[z_{1},\cdots,z_{m}\right]}{(P(X)+SR_{\omega}(X))}$ (6.17) the Batyrev quantum cohomology ring. ###### Remark 6.5. We do not take closure of our ideal $P(X)+SR_{\omega}(X)$ here. See Proposition 8.6. ###### Theorem 6.6 (Batyrev [B1], Givental [Gi2]). If $X$ is Fano there exists a ring isomorphism from $QH^{\omega}(X;\Lambda)$ to the quantum cohomology ring $QH(X;\Lambda)$ of $X$ such that $z_{i}$ is sent to the Poincaré dual to $\pi^{-1}(\partial_{i}P)$. The main geometric part of the proof of Theorem 6.6 is the following. ###### Proposition 6.7. The Poincaré dual to $\pi^{-1}(\partial_{i}P)$ satisfy the quantum Stanley- Reisner relation. We do not prove Proposition 6.7 in this paper. See Remarks 6.15 and 6.16. However since our choice of the coefficient ring is different from other literature, we explain here for reader’s convenience how Theorem 6.6 follows from Proposition 6.7. Proposition 6.7 implies that we can define a ring homomorphism $h:QH^{\omega}(X;\Lambda)\to QH(X;\Lambda)$ by sending $z_{i}$ to $PD(\pi^{-1}(\partial_{i}P))$. Let $F^{k}QH(X;\Lambda)$ be the direct sum of elements of degree $\leq k$. Let $F^{k}QH^{\omega}(X;\Lambda)$ be the submodule generated by the polynomial of degree at most $k/2$ on $z_{i}$. Clearly $h(F^{k}QH^{\omega}(X;\Lambda))\subset F^{k}QH(X;\Lambda)$. Since $X$ is Fano, it follows that, $x\cup_{Q}y-x\cup y\in F^{\deg x+\deg y-2}QH(X;\Lambda).$ We also recall the cohomology ring $H(X;{\mathbb{Q}})$ is obtained by putting $T=0$ in quantum Stanley-Reisner relation. Moreover we find that the second product of the right hand side of (6.15) has degree strictly smaller than the first since $X$ is Fano. Therefore the graded ring $gr(QH(X;\Lambda))=\bigoplus_{k}F^{k}(QH(X;\Lambda))/F^{k-1}(QH(X;\Lambda)),$ is isomorphic to the (usual) cohomology ring (with $\Lambda$ coefficient) as a ring. The same holds for $QH^{\omega}(X;\Lambda)$. It follows that $h$ is an isomorphism. ∎ In the rest of this section, we will prove the following Proposition 6.8. Theorem 6.1 follows immediately from Proposition 6.8 and Theorem 6.6. ###### Proposition 6.8. There exists an isomorphism : $\psi_{u}:QH^{\omega}(X;\Lambda)\cong Jac(\mathfrak{PO}_{0}^{u})$ such that $\psi_{u}(z_{i})=\overline{z}_{i}(u)$. We remark that we do not assume that $X$ is Fano in Proposition 6.8. We also remark that for our main purpose to calculate $\mathfrak{M}_{0}(\mathfrak{Lag}(X))$, Proposition 6.8 suffices. Proposition 6.8 is a rather simple algebraic result whose proof does not require study of pseudo-holomorphic discs or spheres. ###### Proof of Proposition 6.8. We start with the following proposition. ###### Proposition 6.9. The assignment $\widehat{\psi}_{u}(z_{i})=\overline{z}_{i}(u)$ (6.18) induces a well-defined ring isomorphism $\widehat{\psi}_{u}:\frac{\Lambda\left[z_{1},\cdots,z_{m}\right]}{SR_{\omega}(X)}\to\Lambda[y_{1},\cdots,y_{n},y_{1}^{-1},\cdots,y_{n}^{-1}].$ (6.19) ###### Proof. Let $\mathcal{P}$ be a primitive collection and $\mathcal{P}^{\prime}$, $k_{i^{\prime}}$ be as in (6.12). We calculate $\prod_{i\in\mathcal{P}}\overline{z}_{i}(u)=\prod_{i\in\mathcal{P}}y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}T^{\ell_{i}(u)}$ (6.20) by (6.1). On the other hand, $\displaystyle\prod_{i^{\prime}\in\mathcal{P}^{\prime}}\overline{z}_{i^{\prime}}^{k_{i^{\prime}}}(u)$ $\displaystyle=\prod_{i^{\prime}\in\mathcal{P}^{\prime}}y_{1}^{k_{i^{\prime}}v_{i^{\prime},1}}\cdots y_{n}^{k_{i^{\prime}}v_{i^{\prime},n}}T^{k_{i^{\prime}}\ell_{i^{\prime}}(u)}$ $\displaystyle=\prod_{i\in\mathcal{P}}y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}\prod_{i^{\prime}\in\mathcal{P}^{\prime}}T^{k_{i^{\prime}}\ell_{i^{\prime}}(u)}$ by (6.12). Moreover $\sum_{i\in\mathcal{P}}\ell_{i}(u)-\sum_{i\in\mathcal{P}^{\prime}}k_{i^{\prime}}\ell_{i^{\prime}}(u)=\omega(\mathcal{P})$ by (6.13). Therefore $\prod_{i\in\mathcal{P}}\overline{z}_{i}(u)=T^{\omega(\mathcal{P})}\prod_{i^{\prime}\in\mathcal{P}^{\prime}}\overline{z}_{i^{\prime}}^{k_{i^{\prime}}}(u)$ in $\Lambda[y_{1},\cdots,y_{n},y_{1}^{-1},\cdots,y_{n}^{-1}]$. In other words, (6.18) defines a well-defined ring homomorphism (6.19). We now prove that $\widehat{\psi}_{u}$ is an isomorphism. Let $pr:{\mathbb{Z}}^{m}\cong\pi_{2}(X;L(u))\longrightarrow{\mathbb{Z}}^{n}\cong\pi_{1}(L(u))$ be the homomorphism induced by the boundary map $pr([\beta])=[\partial\beta].$ (See (2.1).) We remark $pr(c_{1},\dots,c_{m})=(d_{1},\dots,d_{n})$ with $d_{j}=\sum_{i}c_{i}v_{i,j}$. Let $A=\sum_{i}c_{i}\beta_{i}$ be an element in the kernel of $pr$. We write it as $\sum_{i\in I}a_{i}\beta_{i}-\sum_{j\in J}b_{j}\beta_{j}$ where $a_{i}$ $b_{j}$ are positive and $I\cap J=\emptyset$. We define $r(A)=\prod_{i\in I}z_{i}^{a_{i}}-T^{\sum_{i}a_{i}\ell_{i}(u)-\sum_{j}b_{j}\ell_{j}(u)}\prod_{j\in J}z_{j}^{b_{j}}.$ (6.21) We remark that a generator of quantum Stanley-Reisner ideal corresponds to $r(A)$ for which $I$ is a primitive collection $\mathcal{P}$ and $J=\mathcal{P}^{\prime}$. We also remark that the case $I=\emptyset$ or $J=\emptyset$ is included. ###### Lemma 6.10. $r(A)\in SR_{\omega}(X).$ ###### Proof. This lemma is proved in [B1]. We include its proof (which is different from one in [B1]) here for reader’s convenience. We prove the lemma by an induction over the values $E(A)=\sum_{i\in I}a_{i}\ell_{i}(u_{0})+\sum_{j\in J}b_{j}\ell_{j}(u_{0}).$ Here we fix a point $u_{0}\in\text{Int}P$ during the proof of Lemma 6.10. Since $I\cap J=\emptyset$, at least one of $\\{v_{i}\mid i\in I\\}$, $\\{v_{j}\mid j\in J\\}$ can not span a cone that is a member of the fan $\Sigma$. Without loss of generality, we assume that $\\{v_{i}\mid i\in I\\}$ does not span such a cone. Then it contains a subset $\mathcal{P}\subset I$ that is a primitive collection. We take $\mathcal{P}^{\prime}$, $k_{i^{\prime}}$ as in (6.12) and define $Z=\prod_{i\in I}z_{i}^{a_{i}}-T^{\omega(\mathcal{P})}\prod_{i\in I\setminus\mathcal{P}}z_{i}^{a_{i}}\prod_{i\in\mathcal{P}}z_{i}^{a_{i}-1}\prod_{i^{\prime\prime}\in\mathcal{P}^{\prime}}z_{i^{\prime\prime}}^{k_{i^{\prime\prime}}}.$ (6.22) Then $Z$ lies in $SR_{\omega}(X)$ by construction. We recall from Lemma 6.2 that the values $\sum_{i\in\mathcal{P}}\ell_{i}(u)-\sum_{i\in\mathcal{P}^{\prime}}k_{i}\ell_{i}(u)=\omega(\mathcal{P})$ are independent of $u$ and positive. By the definitions (6.21), (6.22) of $r(A)$ and $Z$, we can express $r(A)-Z=T^{\omega(\mathcal{P})+c}\left(\prod_{h\in K}z_{h}^{n_{h}}\right)r(B)$ for an appropriate $B$ in the kernel of $pr$ and a constant $c$. Moreover we have $E(B)+2\sum_{h\in K}n_{h}\ell_{h}(u_{0})+\omega(\mathcal{P})=E(A).$ Since $u_{0}\in\text{Int}\,P$ it follows that $\ell_{h}(u_{0})>0$ which in turn gives rise to $E(B)<E(A)$. The induction hypothesis then implies $r(B)\in SR_{\omega}(X)$. The proof of the lemma is now complete. ∎ ###### Corollary 6.11. $z_{i}$ is invertible in $\frac{\Lambda\left[z_{1},\cdots,z_{m}\right]}{SR_{\omega}(X)}.$ ###### Proof. Since $X$ is compact, the vector $-v_{i}$ is in some cone spanned by $v_{j}$ ($j\in I$). Namely $-v_{i}=\sum_{j\in I}k_{j}v_{j}$ where $k_{j}$ are nonnegative integers. Then $T^{\ell_{i}(u)+\sum_{j}k_{j}\ell_{j}(u)}=z_{i}\prod_{j\in I}z_{j}^{k_{j}}\mod SR_{\omega}(X)$ by Lemma 6.10. Since $T^{\ell_{i}(u)+\sum_{j}k_{j}\ell_{j}(u)}$ is invertible in the field $\Lambda$, it follows that $\prod_{j\in I}z_{j}^{k_{j}}$ defines the inverse of $z_{i}$ in the quotient ring. ∎ We recall from Lemma 6.2 that $\ell_{i}(u)+\sum_{j}k_{j}\ell_{j}(u)$ is independent of $u$. We define $z_{i}^{-1}=T^{-\ell_{i}(u)-\sum_{j}k_{j}\ell_{j}(u)}\prod_{j\in I}z_{j}^{k_{j}}.$ (6.23) (Note we have not yet proved that ${\Lambda\left[z_{1},\cdots,z_{m}\right]}/{SR_{\omega}(X)}$ is an integral domain. This will follow later when we prove Proposition 6.9.) Since $v_{1},\cdots,v_{m}$ generates the lattice ${\mathbb{Z}}^{n}$, we can always assume the following by changing the order of $v_{i}$, if necessary. ###### Condition 6.12. The determinant of the $n\times n$ matrix $(v_{i,j})_{i,j=1,\cdots,n}$ is $\pm 1$. Let $(v^{i,j})$ be the inverse matrix of $(v_{i,j})$. Namely $\sum_{j}v^{i,j}v_{j,k}=\delta_{i,k}.$ Condition 6.12 implies that each $v^{i,j}$ is an integer. Inverting the matrix $(v_{i,j})$, we obtain $y_{i}=T^{-c_{i}(u)}\prod_{j=1}^{n}\overline{z}_{i}^{v^{i,j}}$ (6.24) from (6.20) where $c_{i}(u)=\sum v^{i,j}\ell_{j}(u).$ We define using Corollary 6.11 $\widehat{\phi}_{u}(y_{i}^{\pm 1})=T^{-\pm c_{i}(u)}\prod_{j=1}^{n}z_{j}^{\pm v^{i,j}}\in\frac{\Lambda\left[z_{1},\cdots,z_{m}\right]}{SR_{\omega}(X)}.$ More precisely, we plug (6.23) here if $\pm v^{i,j}$ is negative. The identity $\widehat{\psi}_{u}\circ\widehat{\phi}_{u}=id$ is a consequence of (6.24). We next calculate $(\widehat{\phi}_{u}\circ\widehat{\psi}_{u})(z_{h})=\widehat{\phi}_{u}(\overline{z}_{h}(u))$ and prove $(\widehat{\phi}_{u}\circ\widehat{\psi}_{u})(z_{h})=T^{\ell_{h}(u)}\widehat{\phi}_{u}(y_{1}^{v_{h,1}}\cdots y_{n}^{v_{h,n}})=T^{e(h;u)}\prod_{j=1}^{n}z_{j}^{m_{j}},$ where $m_{j}\geq 0$ and $v_{h}=\sum m_{j}v_{j},\qquad\ell_{h}(u)=e(h;u)+\sum m_{j}\ell_{j}(u):$ (6.25) To see (6.25), we consider any monomial $Z$ of $y_{i},z_{i},\overline{z}_{i},T^{\alpha}$. We define its multiplicative valuation $\mathfrak{v}_{u}(Z)\in{\mathbb{R}}$ by putting $\mathfrak{v}_{u}(y_{i})=0,\quad\mathfrak{v}_{u}(z_{i})=\mathfrak{v}_{u}(\overline{z}_{i})=\ell_{i}(u),\quad\mathfrak{v}_{u}(T^{\alpha})=\alpha.$ We also define a (multiplicative) grading $\rho(Z)\in{\mathbb{Z}}^{n}$ by $\rho(y_{i})=\text{\bf e}_{i},\quad\rho(z_{i})=\rho(\overline{z}_{i})=v_{i},\quad\rho(T^{\alpha})=0.$ and by $\rho(ZZ^{\prime})=\rho(Z)+\rho(Z^{\prime})$. We remark that $\mathfrak{v}_{u}$ and $\rho$ are consistent with (6.1). We next observe that both $\mathfrak{v}_{u}$ and $\rho$ are preserved by $\widehat{\psi}_{u}$, $\widehat{\phi}_{u}$ and by (6.23). This implies (6.25). Now we use Lemma 6.10 and (6.25) to conclude $z_{h}-T^{e(h;u)}\prod_{j=1}^{n}z_{j}^{m_{j}}\in SR_{\omega}(X).$ The proof of Proposition 6.9 is now complete. ∎ Next we prove ###### Lemma 6.13. Let $P(X)$ be the linear relation ideal defined in Definition 6.4. Then $\widehat{\psi}_{u}(P(X))=\left(\frac{\partial\mathfrak{PO}_{0}^{u}}{\partial y_{i}};i=1,\cdots,n\right).$ ###### Proof. Let $\sum_{i=1}^{m}v_{i,j}z_{i}$ be in $P(X)$. Then we have $\widehat{\psi}_{u}\left(\sum_{i=1}^{m}v_{i,j}z_{i}\right)=\sum_{i=1}^{m}v_{i,j}\overline{z}_{i}=\sum_{i=1}^{m}y_{j}\frac{\partial\overline{z}_{i}}{\partial y_{j}}=y_{j}\frac{\partial\mathfrak{PO}_{0}^{u}}{\partial y_{j}}$ by (6.1) and (6.4). Since $y_{j}$’s are invertible in $\Lambda[y_{1},\cdots,y_{n},y_{1}^{-1},\cdots,y^{-1}_{n}]$, this identity implies the lemma. ∎ The proof of Theorem 6.1 and of Proposition 6.8 is now complete. ∎ ###### Remark 6.14. Proposition 6.8 holds over $\Lambda^{R}$ coefficient for arbitrary commutative ring $R$ with unit. The proof is the same. We define $\psi_{u^{\prime},u}:Jac(\mathfrak{PO}_{0}^{u})\to Jac(\mathfrak{PO}_{0}^{u^{\prime}})$ by $\psi_{u^{\prime},u}(\overline{z}_{i}(u))=\overline{z}_{i}(u^{\prime})=T^{\ell_{i}(u^{\prime})-\ell_{i}(u)}\overline{z}_{i}(u).$ (6.26) It is an isomorphism. We have $\psi_{u^{\prime},u}\circ{\psi}_{u}={\psi}_{u^{\prime}}.$ The well-definedness $\psi_{u^{\prime},u}$ is proved from this formula or by checking directly. In case no confusion can occur, we identify $Jac(\mathfrak{PO}_{0}^{u})$, $Jac(\mathfrak{PO}_{0}^{u^{\prime}})$ by $\psi_{u^{\prime},u}$ and denote them by $Jac(\mathfrak{PO}_{0})$. Since $\psi_{u^{\prime},u}(\overline{z}_{i}(u))=\overline{z}_{i}(u^{\prime})$ we write them $\overline{z}_{i}$ when we regard it as an element of $Jac(\mathfrak{PO}_{0})$. Note $\psi_{u^{\prime},u}(y_{i})\neq y_{i}$. In case we regard $y_{i}\in Jac(\mathfrak{PO}_{0}^{u})$ as an element of $Jac(\mathfrak{PO}_{0})$ we write it as $y_{i}(u):=\psi_{u}(y_{i})$. ###### Remark 6.15. The above proof of Theorem 6.1 uses Batyrev’s presentation of quantum cohomology ring and is not likely generalized beyond the case of compact toric manifolds. (In fact the proof is purely algebraic and do not contain serious study of pseudo-holomorphic curve, except Proposition 6.7, which we quote without proof and Theorem 4.5, which is a minor improvement of an earlier result of [CO].) There is an alternative way of constructing the ring homomorphism $\psi_{u}$ which is less computational. (This will give a new proof of Proposition 6.7.) We will give this conceptual proof in a sequel to this paper. We use the operations $\mathfrak{q}_{1,k;\beta}:H(X;{\mathbb{Q}})[2]\otimes B_{k}H(L(u);{\mathbb{Q}})[1]\to H(L(u);{\mathbb{Q}})[1]$ which was introduced by the authors in section 3.8 [FOOO3] (= section 13 [FOOO2]). Using the class $z_{i}\in H^{2}(X;{\mathbb{Z}})$ the Poincaré dual to $\pi^{-1}(\partial_{i}P)$ we put $\psi_{u}(z_{i})=\sum_{k}\sum_{i=1}^{m}T^{\beta_{i}\cap\omega/2\pi}\int_{L(u)}\mathfrak{q}_{1,k;\beta_{i}}(z_{i}\otimes x^{\otimes k}).$ (6.27) Here we put $x=\sum x_{i}\text{\bf e}_{i}$ and the right hand side is a formal power series of $x_{i}$ with coefficients in $\Lambda$. Using the description of the moduli space defining the operators $\mathfrak{q}_{1,k;\beta}$ (See section 11.) it is easy to see that the right hand side of (6.27) coincides with the definition of $\overline{z}_{i}$ in the current case, when $X$ is Fano toric. Extending the expression (6.27) to an arbitrary homology class $z$ of arbitrary degree we obtain $\psi_{u}(z)=\sum_{k}\sum_{\beta;\mu(\beta)=\deg z}T^{\beta\cap\omega/2\pi}\int_{L(u)}\mathfrak{q}_{1,k;\beta}(z\otimes x^{\otimes k}).$ (6.28) Since $\mu(\beta)=\deg z$, $\mathfrak{q}_{1,k;\beta}(z\otimes x^{\otimes k})\in H^{n}(L(u);{\mathbb{Q}})$. One can prove that (6.28) defines a ring homomorphism from quantum cohomology to the Jacobian ring $Jac(\mathfrak{PO}^{u})$. We may regard $Jac(\mathfrak{PO}^{u})$ as the moduli space of deformations of Floer theories of Lagrangian fibers of $X$. (Note the Jacobian ring parameterizes deformations of a holomorphic function up to an appropriate equivalence. In our case the equivalence is the right equivalence, that is, the coordinate change of the domain.) Thus (6.28) is a particular case of the ring homomorphism $QH(X)\to HH(\mathfrak{Lag}(X))$ where $HH(\mathfrak{Lag}(X))$ is the Hochschild cohomology of Fukaya category of $X$. (We remark that Hochschild cohomology parameterizes deformations of $A_{\infty}$ category.) Existence of such a homomorphism is a folk theorem (See [Ko]) which is verified by various people in various favorable situations. (See for example [Aur].) It is conjectured to be an isomorphism under certain conditions by various people including P. Seidel and M. Kontsevich. This point of view is suitable for generalizing our story to more general $X$ (to non-Fano toric manifolds, for example) and also for including big quantum cohomology group into our story. (We will then also need to use the operators $\mathfrak{q}_{\ell,k}$ mentioned above for $\ell\geq 2$.) These points will be discussed in subsequent papers in this series of papers. In this paper we follow more elementary approach exploiting the known calculation of quantum cohomology of toric manifolds, although it is less conceptual. ###### Remark 6.16. There are two other approaches towards a proof of Proposition 6.7 besides the fixed point localization. One is written by Cieliebak and Salamon [CS] which uses vortex equations (gauged sigma model) and the other is written by McDuff and Tolman [MT] which uses Seidel’s result [Se1]. ## 7\. Localization of quantum cohomology ring at moment polytope In this section, we discuss applications of Theorem 1.9. In particular, we prove Theorem 1.12. (Note Theorem 1.3 is a consequence of Theorem 1.12.) The next theorem and Theorem 1.9 immediately imply (1) of Theorem 1.12. ###### Theorem 7.1. There exists a bijection $\text{\rm Crit}(\mathfrak{PO}_{0})\cong Hom(Jac(\mathfrak{PO}_{0});\Lambda^{{\mathbb{C}}}).$ Here the right hand side is the set of unital $\Lambda^{{\mathbb{C}}}$-algebra homomorphisms. We start with the following definition ###### Definition 7.2. For an element $x\in\Lambda\setminus\\{0\\}$, we define its valuation $\mathfrak{v}_{T}(x)$ as the unique number $\lambda\in{\mathbb{R}}$ such that $T^{-\lambda}x\in\Lambda_{0}\setminus\Lambda_{+}.$ We note that $\mathfrak{v}_{T}$ is multiplicative non-Archimedean valuation, i.e., satisfies $\displaystyle\mathfrak{v}_{T}(x+y)$ $\displaystyle\geq$ $\displaystyle\min(\mathfrak{v}_{T}(x),\mathfrak{v}_{T}(y)),$ $\displaystyle\mathfrak{v}_{T}(xy)$ $\displaystyle=$ $\displaystyle\mathfrak{v}_{T}(x)+\mathfrak{v}_{T}(y).$ ###### Lemma 7.3. For any $\varphi\in Hom(Jac(\mathfrak{PO}_{0});\Lambda^{{\mathbb{C}}})$ there exists a unique $u\in M_{\mathbb{R}}$ such that $\mathfrak{v}_{T}(\varphi(y_{j}(u)))=0$ (7.1) for all $j=1,\cdots,n$. ###### Proof. We still assume Condition 6.12. By definition (6.1) of $\overline{z}_{i}$, homomorphism property of $\varphi$ and multiplicative property of valuation, we obtain $\mathfrak{v}_{T}(\varphi(\overline{z}_{i}))=\ell_{i}(u)+\sum_{j=1}^{n}v_{i,j}\mathfrak{v}_{T}(\varphi(y_{j}(u))),$ (7.2) for $i=1,\cdots,m$. On the other hand, since $\ell_{i}(u)=\langle u,v_{i}\rangle-\lambda_{i}$ and $(v_{i,j})_{i,j=1,\cdots,n}$ is invertible, there is a unique $u$ that satisfies $\mathfrak{v}_{T}(\varphi(\overline{z}_{i}))=\ell_{i}(u)$ (7.3) for $i=1,\cdots,n$. But by the invertibility of $(v_{i,j})_{i,j=1,\cdots,n}$ and (7.2), this is equivalent to (7.1) and hence the proof. ∎ We remark that obviously by the above proof the formula (7.3) automatically holds for $i=n+1,\cdots,m$ and $u$ in Lemma 7.3 as well. ###### Proof of Theorem 7.1. Consider the maps $\Psi_{1}(\varphi)=\sum_{i=1}^{n}(\log\varphi(y_{i}(u)))\text{\bf e}_{i}\in H^{1}(L(u);\Lambda_{0}),\qquad\Psi_{2}(\varphi)=u\in M_{\mathbb{R}}$ where $u$ is obtained as in Lemma 7.3. Since $y_{i}(u)\in\Lambda_{0}\setminus\Lambda_{+}$ it follows that we can define its logarithm on $\Lambda_{0}$ as a convergent power series with respect to the non-Archimedean norm. Set $(\mathfrak{x},u)=(\Psi_{1}(\varphi),\Psi_{2}(\varphi))$. Since $\varphi$ is a ring homomorphism from $Jac(\mathfrak{PO}_{0})\cong Jac(\mathfrak{PO}_{0}^{u})$ it follows from the definition of the Jacobian ring that $\frac{\partial\mathfrak{PO}_{0}^{u}}{\partial y_{i}}(\mathfrak{x})=0.$ Therefore by Theorem 4.10, $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x});\Lambda)\neq 0$. We have thus defined $\Psi:Hom(Jac(\mathfrak{PO}_{0});\Lambda^{{\mathbb{C}}})\to\text{\rm Crit}(\mathfrak{PO}_{0}).$ Let $(\mathfrak{x},u)\in\text{\rm Crit}(\mathfrak{PO}_{0})$. We put $\mathfrak{x}=\sum\mathfrak{x}_{i}\text{\bf e}_{i}$. We define a homomorphism $\varphi:Jac(\mathfrak{PO}_{0})\to\Lambda$ by assigning $\varphi(y_{i}(u))=e^{\mathfrak{x}_{i}}.$ It is straightforward to check that $\varphi$ is well-defined. Then we define $\Phi(\mathfrak{x},u):=\varphi$. It easily follows from definition that $\Phi$ is an inverse to $\Psi$. The proof of Theorem 7.1 is complete. ∎ We next work with the (Batyrev) quantum cohomology side. ###### Definition 7.4. For each $z_{i}$, we define a $\Lambda$-linear map $\widehat{z}_{i}:QH^{\omega}(X;\Lambda^{{\mathbb{C}}})\to QH^{\omega}(X;\Lambda^{{\mathbb{C}}})$ by $\widehat{z}_{i}(z)=z_{i}\cup_{Q}z,$ where $\cup_{Q}$ is the product in $QH^{\omega}(X;\Lambda^{{\mathbb{C}}})$. Since $QH^{\omega}(X;\Lambda)$ is generated by even degree elements it follows that it is commutative. Therefore we have $\widehat{z}_{i}\circ\widehat{z}_{j}=\widehat{z}_{j}\circ\widehat{z}_{i}.$ (7.4) ###### Definition 7.5. For $\mathfrak{w}=(\mathfrak{w}_{1},\cdots,\mathfrak{w}_{n})\in(\Lambda^{{\mathbb{C}}})^{n}$ we put $QH^{\omega}(X;\mathfrak{w})=\\{x\in QH^{\omega}(X;\Lambda^{{\mathbb{C}}})\mid(\widehat{z}_{i}-\mathfrak{w}_{i})^{N}x=0\quad\text{for $i=1,\cdots,n$ and large $N$.}\\}$ We say that $\mathfrak{w}$ is a weight of $QH^{\omega}(X)$ if $QH^{\omega}(X;\mathfrak{w})$ is nonzero. We denote by $W(X;\omega)$ the set of weights of $QH^{\omega}(X)$ . We remark that $\mathfrak{w}_{i}\neq 0$ since $z_{i}$ is invertible. (Corollary 6.11.) ###### Remark 7.6. Since $z_{i}$, $i=1,\cdots,n$ generates $QH^{\omega}(X;\Lambda)$ by Condition 6.12, we have the following. For each $\mathfrak{w}=(\mathfrak{w}_{1},\cdots,\mathfrak{w}_{n})$, there exists $\mathfrak{w}_{n+1},\cdots,\mathfrak{w}_{m}$ depending only on $\mathfrak{w}$ such that $(\widehat{z}_{i}-\mathfrak{w}_{i})^{N}x=0$ also holds for $i=n+1,\cdots,m$, if $N$ is sufficiently large and $x\in QH^{\omega}(X;\mathfrak{w})$. ###### Proposition 7.7. 1. (1) There exists a factorization of the ring $QH^{\omega}(X;\Lambda^{{\mathbb{C}}})\cong\prod_{\mathfrak{w}\in W(X;\omega)}QH^{\omega}(X;\mathfrak{w}).$ 2. (2) There exists a bijection $W(X;\omega)\cong Hom(QH^{\omega}(X;\Lambda);\Lambda^{{\mathbb{C}}}).$ 3. (3) $QH^{\omega}(X;\mathfrak{w})$ is a local ring and $(1)$ is the factorization to indecomposables. ###### Proof. Existence of decomposition (1) as a $\Lambda^{{\mathbb{C}}}$-vector space is a standard linear algebra, using the fact that $\Lambda^{{\mathbb{C}}}$ is an algebraically closed field. (We will prove this fact in appendix as Lemma 14.1.) If $z\in QH^{\omega}(X;\mathfrak{w})$ and $z^{\prime}\in QH^{\omega}(X;\mathfrak{w}^{\prime})$ then $\displaystyle(z_{i}-\mathfrak{w}_{i})^{N}\cup_{Q}(z\cup_{Q}z^{\prime})$ $\displaystyle=((z_{i}-\mathfrak{w}_{i})^{N}\cup_{Q}z)\cup_{Q}z^{\prime}=0,$ $\displaystyle(z_{i}-\mathfrak{w}^{\prime}_{i})^{N}\cup_{Q}(z\cup_{Q}z^{\prime})$ $\displaystyle=((z_{i}-\mathfrak{w}^{\prime}_{i})^{N}\cup_{Q}z^{\prime})\cup_{Q}z=0.$ Therefore $z\cup_{Q}z^{\prime}\in QH^{\omega}(X;\mathfrak{w})\cap QH^{\omega}(X;\mathfrak{w}^{\prime})$. This implies that the decomposition (1) is a ring factorization. Let $\varphi:QH^{\omega}(X;\mathfrak{w})\to\Lambda^{{\mathbb{C}}}$ be a unital $\Lambda^{{\mathbb{C}}}$ algebra homomorphism. It induces a homomorphism $QH^{\omega}(X;\Lambda)\to\Lambda^{{\mathbb{C}}}$ by (1). We denote this ring homomorphism by the same letter $\varphi$. Let $z\in QH^{\omega}(X;\mathfrak{w})$ be an element such that $\varphi(z)\neq 0$. Then we have $(\varphi(z_{i})-\mathfrak{w}_{i})^{N}\varphi(z)=\varphi((z_{i}-\mathfrak{w}_{i})^{N}\cup_{Q}z)=0.$ Therefore $\mathfrak{w}_{i}=\varphi(z_{i}).$ (7.5) Since $z_{i}$ generates $QH^{\omega}(X;\Lambda)$, it follows from (7.5) that there is a unique $\Lambda^{{\mathbb{C}}}$ algebra homomorphism $:QH^{\omega}(X;\mathfrak{w})\to\Lambda^{{\mathbb{C}}}$. (2) follows. Since $QH^{\omega}(X;\mathfrak{w})$ is a finite dimensional $\Lambda^{{\mathbb{C}}}$ algebra and $\Lambda^{{\mathbb{C}}}$ is algebraically closed, we have an isomorphism $\frac{QH^{\omega}(X;\mathfrak{w})}{\text{rad}}\cong(\Lambda^{{\mathbb{C}}})^{k}$ (7.6) for some $k$. (Here $\text{rad}=\\{z\in QH^{\omega}(X;\mathfrak{w})\mid z^{N}=0\quad\text{for some $N$.}\\}$) Since there is a unique unital $\Lambda^{{\mathbb{C}}}$-algebra homomorphism $:QH^{\omega}(X;\mathfrak{w})\to\Lambda^{{\mathbb{C}}}$, it follows that $k=1$. Namely ${QH^{\omega}(X;\mathfrak{w})}$ is a local ring. It also implies that ${QH^{\omega}(X;\mathfrak{w})}$ is indecomposable. ∎ The result up to here also works for the non-Fano case. But the next theorem will require the fact that $X$ is Fano since we use the equality $QH^{\omega}(X;\Lambda)\cong QH(X;\Lambda)$. ###### Theorem 7.8. If $X$ is Fano then $\mbox{\rm Crit}(\mathfrak{PO}_{0})=\mathfrak{M}(\mathfrak{Lag}(X))$. ###### Proof. Let $\mathfrak{w}$ be a weight. We take $z\in QH^{\omega}(X;\mathfrak{w})\subset H(X;\Lambda^{{\mathbb{C}}})\cong H(X;{\mathbb{C}})\otimes\Lambda^{{\mathbb{C}}}$. We may take $z$ so that $z\in(H(X;{\mathbb{C}})\otimes\Lambda_{0}^{{\mathbb{C}}})\,\,\setminus(H(X;{\mathbb{C}})\otimes\Lambda_{+}^{{\mathbb{C}}}).$ Since $z_{i}\cup_{Q}z\equiv z_{i}\cup z\mod\Lambda_{+}^{{\mathbb{C}}},$ where $\cup$ is the classical cup product. (We use $QH^{\omega}(X;\Lambda)=QH(X;\Lambda)$ here.) It follows that $\mathfrak{w}_{i}^{n}z=(\widehat{z}_{i})^{n}(z)=(z_{i})^{n}\cup_{Q}z\equiv(z_{i})^{n}\cup z\mod\Lambda_{+}^{{\mathbb{C}}}.$ Therefore $\mathfrak{w}_{i}\in\Lambda_{+}^{{\mathbb{C}}}$ as $(z_{i})^{n}\cup z=0$. (7.3) and (7.5) then imply $\ell_{i}(u)=\mathfrak{v}_{T}(\mathfrak{w}_{i})>0.$ Namely $u\in\text{Int}P$. ∎ We are now ready to complete the proof of Theorem 1.12. (1) is Theorem 7.1. (2) is a consequence of Theorem 4.10. (3) is Theorem 7.8. If $QH^{\omega}(X;\Lambda^{{\mathbb{C}}})$ is semi-simple, then (7.6) and $k=1$ there implies $QH^{\omega}(X;\Lambda^{{\mathbb{C}}})\cong(\Lambda^{{\mathbb{C}}})^{\\#W(X;\omega)}$ (7.7) as a $\Lambda^{{\mathbb{C}}}$ algebra. (4) follows from (7.7), Proposition 7.7 (2), and Theorem 7.1. The proof of Theorem 1.12 is complete. ∎ We next explain the factorization in Proposition 7.7 (1) from the point of view of Jacobian ring. Let $(\mathfrak{x},u)\in\text{\rm Crit}(\mathfrak{PO}_{0})$. ###### Definition 7.9. We consider the ideal generated by $\frac{\partial}{\partial w_{i}}\mathfrak{PO}_{0}^{u}(\mathfrak{y}_{1}+w_{1},\cdots,\mathfrak{y}_{n}+w_{n})$ (7.8) $i=1,\cdots,n$, in the ring $\Lambda[[w_{1},\cdots,w_{n}]]$ of formal power series where $\mathfrak{x}=\sum\mathfrak{x}_{i}\text{\bf e}_{i}$ and $\mathfrak{y}_{i}=e^{\mathfrak{x}_{i}}$. We denote its quotient ring by $Jac(\mathfrak{PO}_{0};\mathfrak{x},u)$. ###### Proposition 7.10. 1. (1) There is a direct product decomposition : $Jac(\mathfrak{PO}_{0})\cong\prod_{(\mathfrak{x},u)\in\text{\rm Crit}(\mathfrak{PO}_{0})}Jac(\mathfrak{PO}_{0};\mathfrak{x},u),$ as a ring. 2. (2) If $(\mathfrak{x},u)\in\text{\rm Crit}(\mathfrak{PO}_{0})$ corresponds to $\mathfrak{w}\in W(X;\omega)$ via the isomorphism given in Proposition 7.7 $(2)$ and Theorem 7.1, then $\psi_{u}$ induces an isomorphism $\psi_{u}:QH^{\omega}(X;\mathfrak{w})\cong Jac(\mathfrak{PO}_{0};\mathfrak{x},u).$ 3. (3) $Jac(\mathfrak{PO}_{0};\mathfrak{x},u)$ is one dimensional (over $\Lambda$) if and only if the Hessian $\left(\frac{\partial^{2}\mathfrak{PO}_{0}^{u}}{\partial y_{i}\partial y_{j}}\right)_{i,j=1,\cdots,n}$ is invertible over $\Lambda$ at $\mathfrak{x}$. ###### Proof. We put $\mathfrak{x}=\sum\mathfrak{x}_{i}\text{\bf e}_{i}$ and $\mathfrak{y}_{i}=e^{\mathfrak{x}_{i}}$. Let $\mathfrak{m}(\mathfrak{x},u)$ be the ideal generated by $y_{i}-\mathfrak{y}_{i}$, in the ring : $Jac(\mathfrak{PO}_{0})=\frac{\Lambda[y_{1}^{\pm},\cdots,y_{n}^{\pm}]}{(y_{i}\frac{\partial\mathfrak{PO}_{0}^{u}}{\partial y_{i}})}.$ Since $Jac(\mathfrak{PO}_{0})$ is finite dimensional over $\Lambda$ it follows that $Jac(\mathfrak{PO}_{0})\cong\prod_{(\mathfrak{x},u)\in\text{\rm Crit}(\mathfrak{PO}_{0})}Jac(\mathfrak{PO}_{0})_{\mathfrak{m}(\mathfrak{x},u)},$ where $Jac(\mathfrak{PO}_{0})_{\mathfrak{m}(\mathfrak{x},u)}$ is the localization of the ring $Jac(\mathfrak{PO}_{0})$ at $\mathfrak{m}(\mathfrak{x},u)$. Using finite dimensionality of $Jac(\mathfrak{PO}_{0})$ again we have an isomorphism $Jac(\mathfrak{PO}_{0})_{\mathfrak{m}(\mathfrak{x},u)}\cong Jac(\mathfrak{PO}_{0};\mathfrak{x},u)$ which sends $y_{i}-\mathfrak{y}_{i}$ to $w_{i}$. Here $\mathfrak{x}=\sum_{i}\mathfrak{x}_{i}\text{\bf e}_{i}$ and $\mathfrak{y}_{i}=e^{\mathfrak{x}_{i}}$ (1) follows. Now we prove (2). If $z\in QH^{\omega}(X;\mathfrak{w})$ then $(\widehat{z}_{i}-\mathfrak{w}_{i})^{N}z=0$. Let $\pi_{\mathfrak{x},u}:Jac(\mathfrak{PO}_{0})\to Jac(\mathfrak{PO}_{0};\mathfrak{x},u)$ be the projection. We then have $(T^{\ell_{i}(u)}\mathfrak{y}_{1}^{v_{i,1}}\cdots\mathfrak{y}_{n}^{v_{i,n}}-\mathfrak{w}_{i})^{N}\pi_{\mathfrak{x},u}(\psi_{u}(z))=0.$ (7.9) We remark that $\mathfrak{w}_{i}=T^{\ell_{i}(u^{\prime})}\mathfrak{y}_{1}^{\prime v_{i,1}}\cdots\mathfrak{y}_{n}^{\prime v_{i,n}}$ (7.10) if $\mathfrak{w}_{i}$ corresponds $(\mathfrak{x}^{\prime},u^{\prime})$ and $\mathfrak{y}^{\prime}_{i}$ are exponential of the coordinates of $\mathfrak{x}^{\prime}$. We define the operator $\widehat{\mathfrak{y}}_{i}:Jac(\mathfrak{PO}_{0};\mathfrak{x},u)\to Jac(\mathfrak{PO}_{0};\mathfrak{x},u)$ by $\widehat{\mathfrak{y}}_{i}(c)={\mathfrak{y}}_{i}c.$ By definition of $Jac(\mathfrak{PO}_{0};\mathfrak{x},u)$ the eigenvalue of $\widehat{\mathfrak{y}}_{i}$ is $\mathfrak{y}_{i}$. Therefore (7.9) and (7.10) imply that $\pi_{\mathfrak{x},u}(\psi_{u}(z))=0$ unless $(\mathfrak{x},u)=(\mathfrak{x}^{\prime},u^{\prime})$. (2) follows. Let us prove (3). We first remark $\Lambda=\Lambda^{{\mathbb{C}}}$ is an algebraically closed field (Lemma 14.1). Therefore $\mathop{\kern 0.0pt{\rm dim}}\nolimits_{\Lambda}Jac(\mathfrak{PO}_{0};\mathfrak{x},u)=1$ if and only if the ideal generated by (7.8) (for $i=1,\cdots,n$) is the maximal ideal $\mathfrak{m}=(w_{1},\cdots,w_{n})$. We remark that $\mathfrak{m}/\mathfrak{m}^{2}=\Lambda^{n}$ and elements (7.8) reduces to $\left(\frac{\partial^{2}\mathfrak{PO}_{0}^{u}}{\partial y_{i}\partial y_{j}}\right)_{j=1,\cdots,n}\in\Lambda^{n},$ modulo $\mathfrak{m}^{2}$. (3) follows easily. ∎ We recall that a symplectic manifold $(X,\omega)$ is said to be (spherically) monotone if there exists $\lambda>0$ such that $c_{1}(X)\cap\alpha=\lambda\,[\omega]\cap\alpha$ for all $\alpha\in\pi_{2}(X)$. Lagrangian submanifold $L$ of $(X,\omega)$ is said to be monotone if there exists $\lambda>0$ such that $\mu(\beta)=\lambda\omega(\beta)$ for any $\beta\in\pi_{2}(X,L)$. (Here $\mu$ is the Maslov index.) In the monotone case we have the following : ###### Theorem 7.11. If $X$ is a monotone compact toric manifold then there exists a unique $u_{0}$ such that $\mathfrak{M}(\mathfrak{Lag}(X))\subset\Lambda\times\\{u_{0}\\}$ i.e., whenever $(\mathfrak{x},u)\in\mathfrak{M}(\mathfrak{Lag}(X))$, $u=u_{0}$. Moreover $L(u_{0})$ is monotone. ###### Remark 7.12. Related results are discussed in [EP1]. ###### Proof. Since $X$ is Fano, we have $QH^{\omega}(X;\Lambda)=QH(X;\Lambda)$. We assume $c_{1}(X)\cap\alpha=\lambda\,[\omega]\cap\alpha$ with $\lambda>0$. Let $\cup_{\alpha}$ be the contribution to the moduli space of pseudo-holomorphic curve of homology class $\alpha\in H_{2}(X;{\mathbb{Z}})$ in the quantum cup product. (See (6.5).) We have a decomposition : $x\cup_{Q}y=x\cup y+\sum_{\alpha\in\pi_{2}(X)\setminus\\{0\\}}T^{\alpha\cap[\omega]/2\pi}x\cup_{\alpha}y.$ Then $\deg(x\cup_{\alpha}y)=\deg x+\deg y-2c_{1}(X)\cap\alpha=\deg x+\deg y-2\lambda\alpha\cap[\omega].$ (7.11) We define $\mathfrak{v}_{\deg}(T^{1/2\pi})=2\lambda,\quad\mathfrak{v}_{\deg}(x)=\deg x\quad(\text{for $x\in H(X;{\mathbb{Q}})$}).$ $\mathfrak{v}_{\deg}$ is a multiplicative non-Archimedean valuation on $QH(X;\Lambda)$ such that $\mathfrak{v}_{\deg}(a\cup_{Q}b)=\mathfrak{v}_{\deg}(a)+\mathfrak{v}_{\deg}(b),$ by virtue of (7.11). Moreover for $c\in\Lambda$ and $a\in QH(X;\Lambda)$ we have $\mathfrak{v}_{\deg}(ca)=2\lambda\mathfrak{v}_{T}(c)+\mathfrak{v}_{\deg}(a).$ Now let $\mathfrak{w}$ be a weight and $x\in QH^{\omega}(X;\mathfrak{w})$. Since $\mathfrak{v}_{\deg}(z_{i})=2$ it follows that $2\lambda\mathfrak{v}_{T}(\mathfrak{w}_{i})+\mathfrak{v}_{\deg}(x)=\mathfrak{v}_{\deg}(z_{i}x)=2+\mathfrak{v}_{\deg}(x).$ Therefore if $(\mathfrak{x},u)$ corresponds to $\mathfrak{w}$ then $\ell_{i}(u)=\mathfrak{v}_{T}(\mathfrak{w}_{i})=1/\lambda$. Namely $u$ is independent of $\mathfrak{w}$. We denote it by $u_{0}$. For $\beta_{i}\in H_{2}(X,L(u_{0}))$ ($i=1,\cdots,m$) given by (6.7), we have $\omega(\beta_{i})=2\pi\ell_{i}(u_{0})=2\pi/\lambda$. Hence $\mu(\beta_{i})=\lambda\omega(\beta_{i})/\pi$. Since $\beta_{i}$ generates $H_{2}(X,L(u_{0}))$, it follows that $L(u_{0})$ is monotone, as required. ∎ So far we have studied Floer cohomology with $\Lambda^{{\mathbb{C}}}$-coefficients. We next consider the case of $\Lambda^{F}$ coefficient where $F$ is a finite Galois extension of ${\mathbb{Q}}$. We choose $F$ so that each of the weight $\mathfrak{w}$ lies in $(\Lambda^{F}_{0})^{n}$. (Since every finite extension of $\Lambda^{{\mathbb{Q}}}$ is contained in such $\Lambda^{F}$ we can always find such an $F$. See appendix.) Then we have a decomposition $QH^{\omega}(X;\Lambda^{F})\cong\prod_{\mathfrak{w}\in W(X;\omega)}QH^{\omega}(X;\mathfrak{w};F).$ (7.12) It follows that the Galois group $Gal(F/{\mathbb{Q}})$ acts on $W(X;\omega)$. It induces a $Gal(F/{\mathbb{Q}})$ action on $\text{\rm Crit}(\mathfrak{PO}_{0})$. (We use Remark 6.14 here.) We write it as $(\mathfrak{x},u)\mapsto(\sigma(\mathfrak{x}),\sigma(u))$. We remark the following: ###### Proposition 7.13. 1. (1) $\sigma(u)=u$. 2. (2) We write by $y_{i}(\mathfrak{x})$ the exponential of the coordinates of $\mathfrak{x}$. Then $y_{i}(\mathfrak{x})\in\Lambda^{F}$ and $y_{i}(\sigma(\mathfrak{x}))=\sigma(y_{i}(\mathfrak{x}))$. 3. (3) If $QH^{\omega}(X;\Lambda^{{\mathbb{Q}}})$ is indecomposable, there exists $u_{0}$ such that whenever $(\mathfrak{x},u)\in\text{\rm Crit}(\mathfrak{PO}_{0}))$, $u=u_{0}$. ###### Proof. Let $\mathfrak{w}_{i}(\mathfrak{x})$ corresponds to $(\mathfrak{x},u)$. Then $\ell_{i}(\sigma(u))=\mathfrak{v}_{T}(\mathfrak{w}_{i}(\sigma(\mathfrak{x},u)))=\mathfrak{v}_{T}(\sigma\mathfrak{w}_{i}(\mathfrak{x},u))=\mathfrak{v}_{T}(\mathfrak{w}_{i}(\mathfrak{x},u))=\ell_{i}(u).$ (1) follows. (2) follows from the definition and (1). (3) is a consequence of (1). ∎ A monotone blow up of ${\mathbb{C}}P^{2}$ (at one point) gives an example where the assumption of Proposition 7.13 (3) is satisfied. It seems interesting to observe that the ring $QH(X;\Lambda^{{\mathbb{Q}}})$ jumps sometimes when we deform symplectic structure of $X$. The point where this jump occurs is closely related to the point where the number of balanced Lagrangian fibers jumps. In the case of Example 5.7 we have $QH(X;\Lambda^{{\mathbb{Q}}})\cong\begin{cases}(\Lambda^{{\mathbb{Q}}})^{5}&\alpha>0,\\\ (\Lambda^{{\mathbb{Q}}})^{3}\times\Lambda^{{\mathbb{Q}}(\sqrt{-3})}&\alpha<0,\\\ \Lambda^{{\mathbb{Q}}(\sqrt{5})}\times\Lambda^{F}&\alpha=0,\end{cases}$ where $F={\mathbb{Q}}[x]/(x^{3}-x-1).$ We remark that $x^{5}+x^{4}-2x^{3}-2x^{2}+1=(x^{2}+x-1)(x^{3}-x-1)$. We also refer readers to Example 10.10 for further example. ###### Remark 7.14. In sections 11 \- 13, we will use de Rham cohomology of Lagrangian submanifold to define and study Floer cohomology. As a consequence, our results on Floer cohomology is proved over $\Lambda^{{\mathbb{R}}}_{0}$ or $\Lambda^{\mathbb{C}}_{0}$ but not over $\Lambda^{{\mathbb{Q}}}_{0}$ or $\Lambda^{F}_{0}$. (The authors believe that those results can be also proved over $\Lambda^{{\mathbb{Q}}}_{0}$ by using the singular cohomology version developed in [FOOO3], although the detail of their proofs could be more complicated.) On the other hand, Proposition 6.8 and Theorem 6.1 are proved over $\Lambda^{{\mathbb{Q}}}_{0}$. Therefore the discussion on quantum cohomology here works over $\Lambda^{F}_{0}$. We also remark that, though Proposition 7.13 (3) is related to Floer cohomology, its proof given above does not use Floer cohomology over $\Lambda^{{\mathbb{Q}}}_{0}$ but uses only Floer cohomology over $\Lambda^{{\mathbb{C}}}_{0}$ and quantum cohomology over $\Lambda^{{\mathbb{Q}}}_{0}$. In fact, the proof above implies the following : If $u\in\text{\rm Int}P$ and $\mathfrak{x}\in H^{1}(L(u);\Lambda^{{\mathbb{C}}}_{0})$ satisfy $\frac{\partial\mathfrak{PO}^{u}_{0}}{\partial x_{i}}(\mathfrak{x})=0$ then $u=u_{0}$. This is because $Jac(\mathfrak{PO}^{u}_{0};\Lambda^{{\mathbb{C}}}_{0})=Jac(\mathfrak{PO}^{u}_{0};\Lambda^{{\mathbb{Q}}}_{0})\otimes_{\Lambda^{{\mathbb{Q}}}_{0}}\Lambda^{{\mathbb{C}}}_{0}$ and $Jac(\mathfrak{PO}^{u}_{0};\Lambda^{{\mathbb{Q}}}_{0})\cong QH(X;\Lambda^{{\mathbb{Q}}}).$ ## 8\. Further examples and remarks In this section we show how we can use the argument of the last 2 sections to illustrate calculations of $\mathfrak{M}(\mathfrak{Lag}(X))$ in examples. ###### Example 8.1. We consider one point blow up $X$ of ${\mathbb{C}}P^{2}$. We choose its Kähler form so that the moment polytope is $P=\left\\{(u_{1},u_{2})\mid 0\leq u_{1},u_{2},\,\,u_{1}+u_{2}\leq 1,\,\,u_{2}\leq 1-\alpha\right\\},$ $0<\alpha<1$. The potential function is $\mathfrak{PO}=y_{1}T^{u_{1}}+y_{2}T^{u_{2}}+(y_{1}y_{2})^{-1}T^{1-u_{1}-u_{2}}+y_{2}^{-1}T^{1-\alpha- u_{2}}.$ We put $\overline{z}_{1}=y_{1}T^{u_{1}}$, $\overline{z}_{2}=y_{2}T^{u_{2}}$, $\overline{z}_{3}=(y_{1}y_{2})^{-1}T^{1-u_{1}-u_{2}}$, $\overline{z}_{4}=y_{2}^{-1}T^{1-\alpha-u_{2}}$. The quantum Stanley-Reisner relation is $\overline{z}_{1}\overline{z}_{3}=\overline{z}_{4}T^{\alpha},\quad\overline{z}_{2}\overline{z}_{4}=T^{1-\alpha},$ (8.1) and linear relation is $\overline{z}_{1}-\overline{z}_{3}=0,\quad\overline{z}_{2}-\overline{z}_{3}-\overline{z}_{4}=0.$ (8.2) We put $X=\overline{z}_{1}$ and $Y=\overline{z}_{2}$ and solve (8.1), (8.2). We obtain $X^{3}(T^{\alpha}+X)=T^{1+\alpha},$ (8.3) with $Y=X+T^{-\alpha}X^{2}$. We consider valuations of both sides of (8.3). There are three different cases to consider. Case 1; $\mathfrak{v}_{T}(X)>\alpha$ : (8.3) implies $3\mathfrak{v}_{T}(X)+\alpha=1+\alpha$. Namely $\mathfrak{v}_{T}(X)=1/3$. So $\alpha<1/3$. Moreover $\mathfrak{v}_{T}(Y)=1/3$. We have $u_{1}=\mathfrak{v}_{T}(X)=1/3$, $u_{2}=\mathfrak{v}_{T}(Y)=1/3$. (See Lemma 7.3.) Writing $X=a_{1}T^{1/3}+a_{2}T^{\lambda}+$ higher order terms with $\lambda>\frac{1}{3}$ and substituting this into (8.3), we obtain $a_{1}^{3}=1$ which has 3 simple roots. Each of them corresponds to the solution for $\mathfrak{x}$ by Hensel’s lemma (see Proposition 3 in p 144 [BGR], for example). (It also follows from Theorem 10.4 in section 10.) Case 2; $\mathfrak{v}_{T}(X)<\alpha$ : By taking the valuation of (8.3) we obtain $u_{1}=\mathfrak{v}_{T}(X)=(1+\alpha)/4$. Hence $\alpha>1/3$. Moreover $u_{2}=\mathfrak{v}_{T}(Y)=(1-\alpha)/2$. In the same way as Case 1, we can check that there are four solutions. Case 3; $\mathfrak{v}_{T}(X)=\alpha$ : We put $X=a_{1}T^{\alpha}+a_{2}T^{\lambda}+$ higher order terms where $\lambda>\alpha$. (Case 3-1: $a_{1}\neq-1$) : By taking valuation of (8.3), we obtain $u_{1}=\mathfrak{v}_{T}(X)=1/3$. Then $\alpha=1/3$ and $u_{2}=\mathfrak{v}_{T}(Y)=1/3$. (8.3) becomes $a_{1}^{4}+a_{1}^{3}-1=0.$ (8.4) (In this case $X=a_{1}T^{\alpha}$ has no higher term.) There are four solutions. We remark that (8.4) is irreducible over ${\mathbb{Q}}$, since it is so over ${\mathbb{Z}}_{2}$. Namely the assumption of Proposition 7.13 (3) is satisfied. Actually $X$ is monotone in the case $\alpha=1/3$. Hence the same conclusion (uniqueness of $u$) follows from Theorem 7.11 also. (Case 3-2: $a_{1}=-1$) : By taking valuation of (8.3), we obtain $\lambda=1-2\alpha$. $\lambda>\alpha$ implies $\alpha<1/3$. $u_{2}=\mathfrak{v}_{T}(Y)=1-2\alpha$. ($u_{1}=\mathfrak{v}_{T}(X)=\alpha$.) There is one solution. In summary, if $\alpha<1/3$ there are two choices of $u=(\alpha,1-2\alpha),(1/3,1/3)$. On the other hand the numbers of choices of $\mathfrak{x}$ are $1$ and $3$ respectively. If $\alpha\geq 1/3$ there is the unique choice $u=((1+\alpha)/4,(1-\alpha)/2)$. The number of choices of $\mathfrak{x}$ is $4$. We next study a non-Fano case. We will study Hirzebruch surface $F_{n}$. Note $F_{1}$ is one point blow up of ${\mathbb{C}}P^{2}$ which we have already studied. We leave the case $F_{2}$ to the reader. ###### Example 8.2. We consider Hirzebruch surface $F_{n}$, $n\geq 3$. We take its Kähler form so that the moment polytope is $P=\left\\{(u_{1},u_{2})\mid 0\leq u_{1},u_{2},\,\,u_{1}+nu_{2}\leq n,\,\,u_{2}\leq 1-\alpha\right\\},$ $0<\alpha<1$. The leading order potential function is $\mathfrak{PO}_{0}=y_{1}T^{u_{1}}+y_{2}T^{u_{2}}+y_{1}^{-1}y_{2}^{-n}T^{n-u_{1}-nu_{2}}+y_{2}^{-1}T^{1-\alpha- u_{2}}.$ We put $\overline{z}_{1}=y_{1}T^{u_{1}}$, $\overline{z}_{2}=y_{2}T^{u_{2}}$, $\overline{z}_{3}=y_{1}^{-1}y_{2}^{-n}T^{n-u_{1}-nu_{2}}$, $\overline{z}_{4}=y_{2}^{-1}T^{1-\alpha-u_{2}}$. The quantum Stanley-Reisner relation and linear relation gives $\displaystyle\overline{z}_{1}\overline{z}_{3}=\overline{z}_{4}^{n}T^{n\alpha},$ $\displaystyle\quad\overline{z}_{2}\overline{z}_{4}=T^{1-\alpha},$ (8.5) $\displaystyle\overline{z}_{1}-\overline{z}_{3}=0,$ $\displaystyle\quad\overline{z}_{2}-n\overline{z}_{3}-\overline{z}_{4}=0.$ (8.6) Let us assume $n$ is odd. We put $\overline{z}_{1}=Z^{n},\qquad\overline{z}_{4}=Z^{2}T^{-\alpha}.$ (In case $n=2n^{\prime}$ is even we put $\overline{z}_{1}=Z^{n^{\prime}}$, $\overline{z}_{4}=\pm ZT^{-\alpha}.$ The rest of the argument are similar and is omitted.) Then $\overline{z}_{2}=T^{-\alpha}Z^{2}+nZ^{n}$ and $Z^{4}(nZ^{n-2}+T^{-\alpha})=T.$ (8.7) Case 1 ; $(n-2)\mathfrak{v}_{T}(Z)>-\alpha$ : In the first case, we have $\mathfrak{v}_{T}(Z)=(\alpha+1)/4$. (Then $(n-2)\mathfrak{v}_{T}(Z)>-\alpha$ is automatically satisfied.) Therefore $u_{1}=\mathfrak{v}_{T}(z_{1})=n(\alpha+1)/4$, $u_{2}=\mathfrak{v}_{T}(z_{2})=(1-\alpha)/2$. We also can check that there are $4$ solutions. We remark that we are using $\mathfrak{PO}_{0}$ in place of $\mathfrak{PO}$. However we can use Corollary 10.6 to prove the following lemma. This lemma in particular implies that $L(n(\alpha+1)/4,(1-\alpha)/2)$ is balanced which was already shown above in Example 8.1 for the case $n=1$. ###### Lemma 8.3. Let $y^{(i)}\in\Lambda_{0}\times\Lambda_{0}$ ($i=1,\cdots,4$) be critical points $\mathfrak{PO}_{0}^{u}$ for $u=(n(\alpha+1)/4,(1-\alpha)/2)$. Then there exists $y^{(i)\prime}\in\Lambda_{0}\times\Lambda_{0}$ which is a critical point of $\mathfrak{PO}^{u}$ and $y^{(i)}\equiv y^{(i)\prime}\mod\Lambda_{+}$. We will prove Lemma 8.3 in section 10. Case 2 ; $(n-2)\mathfrak{v}_{T}(Z)<-\alpha$ : We have $v_{T}(Z)=1/(n+2)$. This can never occur since $1/(n+2)>0>-\alpha/(n-2)$. Case 3 ; $(n-2)\mathfrak{v}_{T}(Z)=-\alpha$ : We put $Z=a_{1}T^{-\alpha/(n-2)}+a_{2}T^{\lambda}+$ higher order term. (Case 3-1 : $na_{1}^{n-2}\neq-1$) : Then $\mathfrak{v}_{T}(Z)=(\alpha+1)/4$. Since $(\alpha+1)/4\neq-\alpha/(n-2)$, this case never occur. (Case 3-2 : $na_{1}^{n-2}=-1$) : We have $4v_{T}(Z)+(n-3)v_{T}(Z)+\lambda=1.$ Therefore $\lambda=\frac{n-2+(n+1)\alpha}{n-2}.$ We have $u_{1}=v_{T}(z_{1})=-\frac{n\alpha}{n-2},\qquad u_{2}=v_{T}(z_{2})=1-\alpha- v_{T}(z_{4})=\frac{n-2+2\alpha}{n-2}.$ Thus $(u_{1},u_{2})$ is not in the moment polytope. In Example 8.2, we have $\mathfrak{M}(\mathfrak{Lag}(X))=\mathfrak{M}_{0}(\mathfrak{Lag}(X))\neq\text{\rm Crit}(\mathfrak{PO}_{0}).$ On the other hand, the order of $\mathfrak{M}(\mathfrak{Lag}(X))$ is $4$ and is equal to the sum of Betti numbers. ###### Remark 8.4. In a sequel of this series of papers, we will prove the equality $\sum_{d}\text{\rm rank}\,H_{d}(X;{\mathbb{Q}})=\\#(\mathfrak{M}(\mathfrak{Lag}(X)))$ for any compact toric manifold $X$ (which is not necessarily Fano) such that $QH(X;\Lambda)$ is semi-simple. If we count the right hand side with multiplicity, the same equality holds without assuming semi-simplicity. We next discuss the version of the above story where we substitute some explicit number into the formal variable $T$. Let $u\in\mbox{Int}P$. We define a Laurent polynomial $\mathfrak{PO}^{u}_{0,T=t}\in{\mathbb{C}}[y_{1},\cdots,y_{n},y_{1}^{-1},\cdots,y_{n}^{-1}]$ by substituting a complex number $t\in{\mathbb{C}}\setminus\\{0\\}$. In the same way we define the algebra $QH^{\omega}(X;T=t;{\mathbb{C}})$ over ${\mathbb{C}}$ by substituting $T=t$ in the quantum Stanley-Reisner relation. The argument of section 6 goes through to show $QH^{\omega}(X;T=t;{\mathbb{C}})\cong Jac(\mathfrak{PO}^{u}_{0,T=t}).$ (8.8) In particular the right hand side is independent of $u$ up to an isomorphism. Here the ${\mathbb{C}}$-algebra in the right hand side of (8.8) is the quotient of the polynomial ring ${\mathbb{C}}[y_{1},\cdots,y_{n},y_{1}^{-1},\cdots,y_{n}^{-1}]$ by the ideal generated by $\partial\mathfrak{PO}_{0,T=t}/\partial y_{i}$. ($i=1,\cdots,n$.) We remark that right hand side of (8.8) is always nonzero, for small $t$, by Proposition 4.7. It follows that the equation $\frac{\partial\mathfrak{PO}^{u}_{0,T=t}}{\partial y_{i}}=0$ (8.9) has a solution $y_{i}\neq 0$ for any $u$. Namely, as far as the Floer cohomology after $T=t$ substituted, there always exists $b\in H^{1}(X;{\mathbb{C}})$ with nonvanishing Floer cohomology $HF((L(u),b),(L(u),b);{\mathbb{C}})$ for any $u\in\mbox{Int}P$. Since the version of Floer cohomology after substituting $T=t$ is not invariant under the Hamiltonian isotopy, this is not useful for the application to symplectic topology. (Compare this with section 14.2 [CO].) The relation between the set of solutions of (8.9) and that of (4.10) is stated as follows : Let $(y^{(c)}_{1}(t;u),\cdots,y^{(c)}_{n}(t;u))$ be a branch of the solutions of (8.9) for $t\neq 0$ where $c$ is an integer with $1\leq c\leq l$ for some $l\in{\mathbb{N}}$. We can easily show that it is a holomorphic function of $t$ on ${\mathbb{C}}\setminus{\mathbb{R}}_{-}$. We consider its behavior as $t\to 0$. For generic $u$ the limit either diverges or converges to $0$. However if $(\mathfrak{x},u)\in\mathfrak{M}_{0}(\mathfrak{Lag}(X))$ and $\mathfrak{x}=\sum x_{i}\text{\bf e}_{i}$ then there is some $c$ such that $\lim_{t\to 0}y^{(c)}_{i}(t;u)\in{\mathbb{C}}\setminus\\{0\\}\quad\text{and that}\quad y^{(c)}_{i}(t;u)=e^{x_{i}(t)}.$ To prove this claim it suffices to show that if $(\mathfrak{x},u)\in\text{\rm Crit}(\mathfrak{PO}_{0})$ and $\mathfrak{x}=\sum x_{i}\text{\bf e}_{i}$, $y_{i}=e^{x_{i}}$, $y_{i}=\sum_{j}y_{ij}T^{\lambda_{ij}}$ then $\sum_{j}y_{ij}t^{\lambda_{ij}}$ converges for $0<|t|<\epsilon$. (Here $\epsilon$ is sufficiently small positive number.) This follows from the following Lemma 8.5. Let $\Lambda_{0}^{conv}$ be the ring $\left\\{\sum_{i}a_{i}T^{\lambda_{i}}\in\Lambda_{0}^{{\mathbb{C}}}\,\left|\,\exists\epsilon>0\,\,\text{such that}\,\,\sum_{i}|a_{i}||t|^{\lambda_{i}}\,\,\text{\rm converges for $|t|<\epsilon$.}\right\\}\right.$ and $\Lambda^{conv}$ be its field of fractions. We put $\Lambda_{+}^{conv}=\Lambda_{0}^{conv}\cap\Lambda_{+}$. ###### Lemma 8.5. The field $\Lambda^{conv}$ is algebraically closed. We will prove Lemma 8.5 in section 14. We go back to the discussion on the difference between two sets $\mathfrak{M}_{0}(\mathfrak{Lag}(X))$ and $\text{\rm Crit}(\mathfrak{PO}_{0})$. (See Definition 1.11 for its definition.) The rest of this section owes much to the discussion with H. Iritani and also to his papers [Iri1], [Iri2]. The results we describe below will not be used in the other part of this paper. We recall that we did not take closure of the ideal $(P(X)+SR_{\omega}(X))$ in section 6. This is actually the reason why we have $\mathfrak{M}_{0}(\mathfrak{Lag}(X))\neq\text{\rm Crit}(\mathfrak{PO}_{0})$. More precisely we have the following Proposition 8.6. We consider the polynomial ring $\Lambda[z_{1},\cdots,z_{m}]$. We define its norm $\|\cdot\|$ so that $\left\|\sum_{\vec{i}}a_{\vec{i}}z_{1}^{i_{1}}\cdots z_{m}^{i_{m}}\right\|=\exp\left(-\inf_{\vec{i}}\mathfrak{v}_{T}(a_{\vec{i}})\right).$ We take the closure of the ideal $(P(X)+SR_{\omega}(X))$ with respect to this norm and denote it by $\text{\rm Clos}(P(X)+SR_{\omega}(X))$. We put $\overline{QH}^{\omega}(X;\Lambda)=\frac{\Lambda[z_{1},\cdots,z_{m}]}{\text{\rm Clos}(P(X)+SR_{\omega}(X))}.$ (8.10) Let $W^{\text{\rm geo}}(X;\omega)$ be the set of all weight such that the corresponding $(\mathfrak{x},u)$ satisfies $u\in\text{\rm Int}\,P$. We remark that $\mathfrak{w}\in W^{\text{\rm geo}}(X;\omega)$ if and only if $\mathfrak{v}_{T}(\mathfrak{w}_{i})>0$ for all $i$. ###### Proposition 8.6 (Iritani). There exists an isomorphism $\overline{QH}^{\omega}(X;\Lambda^{{\mathbb{C}}})\cong\prod_{\mathfrak{w}\in W^{\text{\rm geo}}(X;\omega)}QH^{\omega}(X;\mathfrak{w}).$ ###### Proof. Let $\mathfrak{w}\in W(X;\omega)\setminus W^{\text{\rm geo}}(X;\omega)$. We first assume $\mathfrak{v}_{T}(\mathfrak{w}_{i})=-\lambda<0$. (The case $\mathfrak{v}_{T}(\mathfrak{w}_{i})=0$ will be discussed at the end of the proof.) Then, there exists $f\in\Lambda_{0}\setminus\Lambda_{+}$ such that $T^{\lambda}f\mathfrak{w}_{i}=1$. Let $x\in QH^{\omega}(X;\mathfrak{w})$. We assume $x\neq 0$. We take $k$ such that $(\widehat{z}_{i}-\mathfrak{w}_{i})^{k}x\neq 0$, $(\widehat{z}_{i}-\mathfrak{w}_{i})^{k+1}x=0$ and replace $x$ by $(\widehat{z}_{i}-\mathfrak{w}_{i})^{k}x$. We then have $T^{\lambda}f\widehat{z}_{i}x=x$. Since $\lim_{N\to\infty}\|(fz_{i}T^{\lambda})^{N}\|=0$, it follows that $x=0$ in $\overline{QH}^{\omega}(X;\Lambda^{{\mathbb{C}}})$. This is a contradiction. We next assume $\mathfrak{v}_{T}(\mathfrak{w}_{i})>0$ for all $i$. We consider the homomorphism $\varphi:\Lambda[z_{1},\cdots,z_{m}]\to Hom_{\Lambda}(QH^{\omega}(X;\mathfrak{w}),QH^{\omega}(X;\mathfrak{w})),$ defined by $\varphi(z_{i})(x)=z_{i}\cup_{Q}x.$ We have $\varphi(P(X)+SR_{\omega}(X))=0$. We may choose the basis of $QH^{\omega}(X;\mathfrak{w})$ so that $\varphi(z_{i})$ is upper triangular matrix whose diagonal entries are all $\mathfrak{w}_{i}$ and whose off diagonal entries are all $0$ or $1$. We use it and $\mathfrak{v}_{T}(\mathfrak{w}_{i})>0$ to show that $\varphi(\text{\rm Clos}(P(X)+SR_{\omega}(X)))=0$. Namely $\varphi$ induces a homomorphism from $\overline{QH}^{\omega}(X;\Lambda)$. It follows easily that the restriction of the projection ${QH}^{\omega}(X;\Lambda^{{\mathbb{C}}})\to\overline{QH}^{\omega}(X;\Lambda^{{\mathbb{C}}})$ to $QH^{\omega}(X;\mathfrak{w})$ is an isomorphism to its image. We finally show that for $u\in\partial P$, there is no critical point of $\mathfrak{PO}_{0}$ on $(\Lambda_{0}\setminus\Lambda_{+})^{n}$. Let $u\in\bigcup_{i\in I}\partial_{i}P\setminus\bigcup_{i\notin I}\partial_{i}P_{i}.$ Then $\mathfrak{PO}_{0}^{u}\equiv\sum_{i\in I}y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}\mod\Lambda_{+}.$ We remark that $v_{i}$ ($i\in I$) is a part of the ${\mathbb{Z}}$ basis of ${\mathbb{Z}}^{n}$, since $X$ is nonsingular toric. Hence by changing the variables to appropriate $y^{\prime}_{i}$ it is easy to see that there is no nonzero critical point of $\sum_{i\in I}y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}=\sum_{i\in I^{\prime}}y^{\prime}_{i}$. The proof of Proposition 8.6 is now complete. ∎ To further discuss the relationship between the contents of sections 6 and 7 and those in [Iri2], we compare the coefficient rings used here and in [Iri2]. In [Iri2] (like many of the literatures on quantum cohomology such as [Gi1]) the formal power series ring ${\mathbb{Q}}[[q_{1},\cdots,q_{m-n}]]$ is taken as the coefficient ring. ($m-n$ is the rank of $H^{2}(X;{\mathbb{Q}})$ and we choose a basis of it.) The superpotential in [Iri2] (which is the same as the one used in [Gi1]) is given as 222We change the notation so that it is consistent to ours. $m,n,v_{i,j}$ here corresponds to $r+N$, $r$, $x_{i,b}$ in [Iri2], respectively. $F_{q}=\sum_{i=1}^{m}\left(\prod_{a=1}^{m-n}q_{a}^{l_{a,i}}\prod_{j=1}^{n}s_{j}^{v_{i,j}}\right).$ (8.11) Here $l_{a,i}$ is a matrix element of a splitting of $H_{2}(X;{\mathbb{Z}})\to H_{2}(X,T^{n};{\mathbb{Z}})$. We will show that (8.11) pulls back to our potential function $\mathfrak{PO}^{u}_{0}$ after a simple change of variables. Let $\alpha_{a}\in H_{2}(X;{\mathbb{Z}})$ be the basis we have chosen. (We choose it so that $[\omega]\cap\alpha_{a}$ is positive.) ###### Lemma 8.7. There exists $f_{j}(u)\in{\mathbb{R}}$ ($j=1,\cdots,n$) such that $\frac{1}{2\pi}\sum_{a}l_{a,i}[\omega]\cap\alpha_{a}=\ell_{i}(u)-\sum_{j=1}^{n}v_{i,j}f_{j}(u).$ ###### Proof. We consider the exact sequence $0\longrightarrow H_{2}(X;{\mathbb{Z}})\overset{i_{*}}{\longrightarrow}H_{2}(X,L(u);{\mathbb{Z}})\longrightarrow H_{1}(L(u);{\mathbb{Z}})\longrightarrow 0.$ $(c_{1},\cdots,c_{m})\in H_{2}(X,L(u);{\mathbb{Z}})$ is in the image of $H_{2}(X;{\mathbb{Z}})$ if and only if $\sum_{i}c_{i}v_{i}=0$. (Here $v_{i}=(v_{i,1},\cdots,v_{i,n})\in{\mathbb{Z}}^{n}$.) For given $\alpha\in H_{2}(X,{\mathbb{Z}})$ denote $i_{*}(\alpha)=(c_{1},\cdots,c_{m})$. Then we have $\sum_{a}[\omega]\cap c_{i}l_{a,i}\alpha_{a}=[\omega]\cap\alpha=2\pi\sum c_{i}\ell_{i}(u).$ This implies the lemma. ∎ We now put $q_{a}=T^{[\omega]\cap\alpha_{a}/2\pi},\qquad s_{j}(u)=T^{f_{j}(u)}y_{j}.$ (8.12) We obtain the identity $F_{q}(s_{1}(u),\cdots,s_{n}(u))=\mathfrak{PO}_{0}^{u}(y_{1},\cdots,y_{n}).$ (8.13) We remark that if we change the choice of Kähler form then the identification (8.12) changes. In other words, the story over ${\mathbb{Q}}[[q_{1},\cdots,q_{m-n}]]$ corresponds to studying all the symplectic structures simultaneously, while the story over $\Lambda$ focuses on one particular symplectic structure. In [Iri2] Corollary 5.12, Iritani proved semi-simplicity of quantum cohomology ring of toric manifold with coefficient ring ${\mathbb{Q}}[[q_{1},\cdots,q_{m-n}]]$. It does not imply the semi-simplicity of our $QH^{\omega}(X;\Lambda)$ since the semi-simplicity in general is not preserved by the pull-back. (On the other way round, semi-simplicity follows from semi-simplicity of the pull-back.) However it is preserved by the pull- back at a generic point. Namely we have: ###### Proposition 8.8. The set of $T^{n}$-invariant symplectic structures on $X$ for which $Jac(\mathfrak{PO}_{0}^{u})$ is semi-simple is open and dense. ###### Proof. We give a proof for completeness, following the argument in the proof of Proposition 5.11 [Iri2]. Consider the polynomial $F_{w_{1},\cdots,w_{m}}=\sum_{i=1}^{m}w_{i}y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}$ where $w_{i}\in{\mathbb{C}}\setminus\\{0\\}$. By Kushnirenko’s theorem [Ku] the Jacobian ring of $F_{w_{1},\cdots,w_{m}}$ is semi-simple for a generic $w_{1},\cdots,w_{m}$. We put $w_{i}=\exp\left(\frac{1}{2\pi}\sum_{a}l_{a,i}[\omega]\cap\alpha+\sum_{j}v_{i,j}f_{j}(u)\right).$ It is easy to see that when we move $[\omega]\cap\alpha_{a}$ and $u$ (there are $m-n$, $n$ parameters respectively) then $w_{i}$ moves in an arbitrary way. Therefore for generic choice of $\omega$ and $u$, the Jacobian ring $Jac(\mathfrak{PO}_{0}^{u})$ is semi-simple. Since $Jac(\mathfrak{PO}_{0}^{u})$ is independent of $u$ up to isomorphism, the proposition follows. ∎ ###### Remark 8.9. Combined with Theorem 1.9, this proposition gives a partial answer to Question in section 3 [EP2]. ## 9\. Variational analysis of potential function In this section, we prove Proposition 4.7. Let $\mathfrak{PO}$ be defined as in (4.7). We define $s_{1}(u)=\inf\\{\ell_{i}(u)\mid i=1,\cdots,m\\}.$ $s_{1}$ is a continuous, piecewise affine and convex function and $s_{1}\equiv 0$ on $\partial P$. Recall if $u\in\partial_{i}P$ then $\ell_{i}(u)=0$ by definition. We put $\displaystyle S_{1}$ $\displaystyle=\sup\\{s_{1}(u)\mid u\in P\\},$ $\displaystyle P_{1}$ $\displaystyle=\\{u\in P\mid s_{1}(u)=S_{1}\\}.$ ###### Proposition 9.1. There exist $s_{k}$, $S_{k}$, and $P_{k}$ with the following properties. 1. (1) $P_{k+1}$ is a convex polyhedron in $M_{\mathbb{R}}$. $\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{k+1}\leq\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{k}$. 2. (2) $s_{k+1}:P_{k}\to{\mathbb{R}}$ is a continuous, convex piecewise affine function. 3. (3) $s_{k+1}(u)=\inf\\{\ell_{i}(u)\mid\ell_{i}(u)>S_{k}\\}$ for $u\in\text{\rm Int}P_{k}$. 4. (4) $s_{k+1}(u)=S_{k}$ for $u\in\partial P_{k}$. 5. (5) $S_{k+1}=\sup\\{s_{k+1}(u)\mid u\in P_{k}\\}.$ 6. (6) $P_{k+1}=\\{u\in P_{k}\mid s_{k+1}(u)=S_{k+1}\\}.$ 7. (7) $P_{k+1}\subset\text{\rm Int}P_{k}$. 8. (8) $s_{k}$, $S_{k}$, $P_{k}$ are defined for $k=1,2,\cdots,K$ for some $K\in{\mathbb{Z}}_{+}$ and $P_{K}$ consists of a single point. ###### Example 9.2. Let $P=[0,a]\times[0,b]$ ($a<b$.) Then $s_{1}(u_{1},u_{2})=\inf\\{u_{1},u_{2},a-u_{1},b-u_{2}\\}.$ $S_{1}=a/2$, $P_{1}=\\{(a/2,u_{2})\mid a/2\leq u_{2}\leq b-a/2\\}$, $s_{2}(1/2,u_{2})=\inf\\{u_{2},b-u_{2}\\}$, $S_{2}=b/2$, $P_{2}=\\{(a/2,b/2)\\}$. ###### Proof. We define $s_{k}$, $S_{k}$, $P_{k}$ inductively over $k$. We assume that $s_{k}$, $S_{k}$, $P_{k}$ are defined for $k=1,\cdots,k_{0}$ so that (1) - (7) of Proposition 9.1 are satisfied for $k=1,\cdots,k_{0}-1$. We define $s_{k_{0}+1}$ by (3) and (4). We will prove that it satisfies (2). We use the following lemma for this purpose. ###### Lemma 9.3. Let $u_{j}\in\text{\rm Int}P_{k_{0}}$ and $\lim_{j\to\infty}u_{j}=u_{\infty}\in\partial P_{k_{0}}$. Then $\lim_{j\to\infty}s_{k_{0}+1}(u_{j})=S_{k_{0}}.$ ###### Proof. We put $I_{k_{0}}^{\prime}=\\{\ell_{i}|\ell_{i}(u_{\infty})=S_{k_{0}}\\}.$ (9.1) By (6) for $k=k_{0}-1$, we find that $s_{k_{0}}(u_{\infty})=S_{k_{0}}$. Then (3) for $k=k_{0}-1$ implies that there is $\ell_{i}$ such that $\ell_{i}(u_{\infty})=S_{k_{0}}$. Thus $I_{k_{0}}^{\prime}$ is non-empty. We take the affine space $A_{k_{0}}\subset M_{\mathbb{R}}$ such that $\text{\rm Int}P_{k_{0}}$ is relatively open in $A_{k_{0}}$. Now since $u_{\infty}\in\partial P_{k_{0}}$, we can take a $\vec{u}\in T_{u_{\infty}}A_{k_{0}}$ such that $u_{\infty}+\epsilon\vec{u}\notin P_{k_{0}}$ for any sufficiently small $\epsilon>0$. It follows from (7) for $k=k_{0}-1$ that $u_{\infty}+\epsilon\vec{u}\in\operatorname{Int}P_{k_{0}-1}$, and hence $u+\epsilon\vec{u}\in\operatorname{Int}P_{k_{0}-1}\setminus P_{k_{0}}$. By definition, we also have $s_{k_{0}}(u)\leq S_{k_{0}}$ for all $u\in P_{k_{0}-1}$. Therefore we have $s_{k_{0}}(u_{\infty}+\epsilon\vec{u})<S_{k_{0}}.$ It follows that there exists $\ell_{i}\in I^{\prime}_{k_{0}}$ such that $\ell_{i}(u_{\infty}+\epsilon\vec{u})<\ell_{i}(u_{\infty})=S_{k_{0}}<\ell_{i}(u_{\infty}-\epsilon\vec{u}).$ (9.2) Since (9.2) holds for any $\vec{u}\in T_{u_{\infty}}A_{k_{0}}$ with $u_{\infty}+\epsilon\vec{u}\notin P_{k_{0}}$, it holds for $\epsilon\vec{u}:=u_{j}-u_{\infty}$ for any sufficiently large $j$. We note that since $u_{j}\in\operatorname{Int}P_{k_{0}}\subset A_{k_{0}}$, $\vec{u}_{j}=u_{\infty}-u_{j}$ is an ‘outward’ vector as a tangent vector in $T_{u_{\infty}}A_{k_{0}}$ at $u_{\infty}\in\partial P_{k_{0}}$. Therefore we have $u_{\infty}+\vec{u}_{j}\notin P_{k_{0}}$. Because $u_{j}=u_{\infty}-\vec{u}_{j}$, it follows from (9.2) that $\ell_{i}(u_{j})>S_{k_{0}}$ (9.3) for any sufficiently large $j$. Therefore we have $s_{k_{0}+1}(u_{j})=\inf\\{\ell_{i}(u_{j})\mid\ell_{i}\in I^{\prime}_{k_{0}},\,\,\ell_{i}(u_{j})>S_{k_{0}}\\}$ (9.4) and $\lim_{j\to\infty}s_{k_{0}+1}(u_{j})=\lim_{j\to\infty}\ell_{i}(u_{j})=\ell_{i}(u_{\infty})=S_{k_{0}}$. This finishes the proof of the lemma. ∎ Lemma 9.3 implies that $s_{k_{0}+1}$ is continuous and piecewise linear in a neighborhood of $\partial P_{k_{0}}$. We can then check (2) easily. We define $S_{k_{0}+1}$ by (5). Then we can define $P_{k_{0}+1}$ by (6). (In other words the right hand side of (6) is nonempty.) (7) is a consequence of Lemma 9.3. We can easily check that $P_{k_{0}+1}$ satisfies (1). We finally prove that $P_{K}$ becomes a point for some $K$. Let $\text{\bf u}_{k}\in\text{\rm Int}\,P_{k}$ and put $I_{k}=\\{\ell_{i}\mid\ell_{i}(\text{\bf u}_{k})=S_{k}\\}.$ (9.5) Here $k=1,\cdots,K$. We remark that $I_{k}$ is independent of the choice of $\text{\bf u}_{k}\in\text{\rm Int}\,P_{k}$. Note we defined $I^{\prime}_{k_{0}}$ by the formula (9.1). We have $I_{k_{0}}\subseteq I^{\prime}_{k_{0}}$. But the equality may not hold in general. In fact $u_{\infty}$ in the boundary of $P_{k_{0}}$ but $\text{\bf u}_{k_{0}}$ is an interior point of $P_{k_{0}}$. Therefore if $\ell_{i}\in I_{k_{0}}$ then $\ell_{i}$ is constant on $P_{k_{0}}$. But element of $I^{\prime}_{k_{0}}$ may not have this property. In case some $\ell_{i}\in I_{k_{0}+1}$ is not constant on $P_{k_{0}}$ it is easy to see that $\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{k_{0}+1}<\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{k_{0}}$. We remark that there exists some $\ell_{j}\notin\bigcup_{k\leq k_{0}}I_{k_{0}}$ which is not constant on $S_{k_{0}}$ unless $S_{k_{0}}$ is a point. Therefore if $\mathop{\kern 0.0pt{\rm dim}}\nolimits S_{k_{0}}\neq 0$, there exists $k^{\prime}>k_{0}$ such that $\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{k^{\prime}}<\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{k_{0}}$. Therefore there exists $K$ such that $P_{K}$ becomes $0$-dimensional (namely a point). Hence we have achieved (8). The proof of Proposition 9.1 is now complete. ∎ ###### Remark 9.4. In a recent preprint [Mc], McDuff pointed out an error in the statement (1) of Proposition 9.1 in the previous version of this paper. We have corrected the statement and modified the last paragraph of its proof, following the corresponding argument in section 2.2 of [Mc]. We thank her for pointing out this error. The next lemma easily follows from construction. ###### Lemma 9.5. If all the vertices of $P$ lie in ${\mathbb{Q}}^{n}$ then $u_{0}\in{\mathbb{Q}}^{n}$. Here $\\{u_{0}\\}=P_{K}$. By parallelly translating the polytope, we may assume, without loss of generality, that $u_{0}=\text{\bf 0}$, the origin. In the rest of this subsection, we will prove that $\mathfrak{PO}^{\text{\bf 0}}$ has a critical point on $(\Lambda_{0}\setminus\Lambda_{+})^{n}$. More precisely we prove Proposition 4.7 for $u_{0}=\text{\bf 0}$. (We remark that if $P$ and $\ell_{i}$ are given we can easily locate $u_{0}$.) ###### Example 9.6. Let us consider Example 8.1 in the case $\alpha>1/3$. At $u_{0}=((1+\alpha)/4,(1-\alpha)/2)$ we have $\mathfrak{PO}^{u_{0}}=(y_{2}+y_{2}^{-1})T^{(1-\alpha)/2}+(y_{1}+(y_{1}y_{2})^{-1})T^{(1+\alpha)/4}.$ Therefore the constant term $\mathfrak{y}_{i;0}$ of the coordinate $y_{i}$ of the critical point is given by $1-\mathfrak{y}_{2;0}^{-2}=0,\quad 1-\mathfrak{y}_{1;0}^{-2}\mathfrak{y}_{2;0}^{-1}=0.$ (9.6) Note the first equation comes from the term of the smallest exponent and contains only $\mathfrak{y}_{2;0}$. The second equation comes from the term which has second smallest exponent and contains both $\mathfrak{y}_{1;0}$ and $\mathfrak{y}_{2;0}$. So we need to solve the equation inductively according to the order of the exponent. This is the situation we want to work out in general. We remark that the affine space $A_{i}$ defined above in the proof of Lemma 9.3 $M_{\mathbb{R}}=A_{0}\supseteq A_{1}\supseteq\cdots\supseteq A_{K-1}\supseteq A_{K}=\\{\text{\bf 0}\\}$ is a nonincreasing sequence of linear subspaces such that $\text{\rm Int }P_{k}$ is an open subset of $A_{k}$. Let $A_{l}^{\perp}\subset(M_{\mathbb{R}})^{*}\cong N_{\mathbb{R}}$ be the annihilator of $A_{l}\subset M_{\mathbb{R}}$. Then we have $\\{\text{\bf 0}\\}=A_{0}^{\perp}\subseteq A_{1}^{\perp}\subseteq\cdots\subseteq A_{K-1}^{\perp}\subseteq A_{K}^{\perp}=N_{\mathbb{R}}.$ We recall: $I_{k}=\\{\ell_{i}\mid\ell_{i}(\text{\bf 0})=S_{k}\\},$ (9.7) for $k=1,\cdots,K$. In fact $\text{\bf 0}\in P_{k+1}\subseteq\text{\rm Int}\,P_{k}$ for $k<K$. We renumber each of $I_{k}$ in (9.7) so that $\\{\ell_{k,j}\mid j=1,\cdots,a(k)\\}=I_{k}.$ (9.8) By construction $s_{k}(u)=\inf_{j}\ell_{k,j}(u)$ (9.9) in a neighborhood of 0 in $P_{k-1}$. In fact $s_{k-1}(\text{\bf 0})=S_{k-1}<S_{k}=s_{k}(\text{\bf 0})$ and $\\{\ell_{i}(\text{\bf 0})\mid i=1,\cdots,m\\}\cap(S_{k-1},S_{k})=\emptyset.$ ###### Lemma 9.7. If $u\in A_{k}$ then $\ell_{k,j}(u)=S_{k}$. ###### Proof. We may assume $k<K$. Hence $\text{\bf 0}\in\text{\rm Int}P_{k}$. We regard $u\in A_{k}=T_{\text{\bf 0}}A_{k}$. By (9.9), we have $s_{k}(\varepsilon u)=\inf\\{\ell_{k,j}(\epsilon u)\mid j=1,\cdots,a(k)\\}.$ Since $s_{k}(\varepsilon u)=S_{k}$ for $\varepsilon u\in P_{k}$ it follows that $\ell_{k,j}(u)=S_{k}$. ∎ Lemma 9.7 implies that the linear part $d\ell_{k,j}$ of $\ell_{k,j}$ is an element of $A_{k}^{\perp}\subset\mathfrak{t}=N_{\mathbb{R}}$. In fact if $\ell_{k,j}=\ell_{i}$, we have $d\ell_{k,j}=v_{i}$ from the definition of $\ell_{i}$, $\ell_{i}(u)=\langle u,v_{i}\rangle-\lambda_{i}$ given in Theorem 2.13. ###### Lemma 9.8. For any $v\in A_{k}^{\perp}$, there exist nonnegative real numbers $c_{j}\geq 0$, $j=1,\cdots,a(k)$ such that $v-\sum_{j=1}^{a(k)}c_{j}d\ell_{k,j}\in A_{k-1}^{\perp}.$ ###### Proof. Suppose to the contrary that $\left\\{v-\sum_{j=1}^{a(k)}c_{j}d\ell_{k,j}\,\Big{|}\,c_{j}\geq 0,\,\,j=1,\cdots,a(k)\right\\}\cap A_{k-1}^{\perp}=\emptyset.$ Then we can find $u\in A_{k-1}\setminus A_{k}$ such that $d\ell_{k,j}(u)\geq 0$ (9.10) for all $j=1,\cdots,a(k)$. Since $\varepsilon u\in A_{k-1}\setminus A_{k}$ it follows that $s_{k}(\varepsilon u)<S_{k}$ for a sufficiently small $\varepsilon$. On the other hand, (9.10) implies $d\ell_{k,j}(\varepsilon u)\geq 0$ for all $\varepsilon>0$ and so $\ell_{k,j}(\varepsilon u)\geq\ell_{k,j}(\text{\bf 0})=S_{k}$. Therefore by definition of $s_{k}$ in Proposition 9.1 we have $\displaystyle s_{k}(\varepsilon u)$ $\displaystyle\geq$ $\displaystyle\inf\\{\ell_{k,j}(\varepsilon u)\mid j=1,\cdots,a(k)\\}$ $\displaystyle\geq$ $\displaystyle\inf\\{\ell_{k,j}(\text{\bf 0})\mid j=1,\cdots,a(k)\\}=S_{k}.$ This is a contradiction. ∎ Applying Lemma 9.8 inductively downwards starting from $\ell=k$ ending at $\ell=1$, we immediately obtain the following ###### Corollary 9.9. For any $v\in A_{k}^{\perp}$, there exist $c_{l,j}\geq 0$ for $l=1,\cdots,k$, $j=1,\cdots,a(l)$ such that $v=\sum_{l=1}^{k}\sum_{j=1}^{a(l)}c_{l,j}d\ell_{l,j}.$ We denote $\mathfrak{I}=\\{\ell_{i}\mid i=1,\cdots,m\\}\setminus\bigcup_{k=1}^{K}I_{k}.$ (9.11) It is easy to see that $\ell\in\mathfrak{I}\Rightarrow\ell(\text{\bf 0})>S_{K}.$ (9.12) Now we go back to the situation of (4.7). We use the notation of (4.7). In this case, for each $k=1,\cdots,K$, we also associate, in Definition 9.10, a set $\mathfrak{I}_{k}$ consisting of pairs $(\ell,\rho)$ with an affine map $\ell:M_{\mathbb{R}}\to{\mathbb{R}}$ and $\rho\in{\mathbb{R}}_{+}$. ###### Definition 9.10. We say that a pair $(\ell,\rho)=(\ell^{\prime}_{j},\rho_{j})$ is an element of $\mathfrak{I}_{k}$ if the following holds : 1. (1) If $e_{j}^{i}\neq 0$ then $\ell_{i}\in\bigcup_{l=1}^{k}I_{l}$. (Note $\ell^{\prime}_{j}=\sum_{i}e_{j}^{i}\ell_{i}$.) 2. (2) (1) does not hold for some $i,j$ if we replace $k$ by $k-1$. A pair $(\ell,\rho)=(\ell^{\prime}_{j},\rho_{j})$ as in (4.7) is, by definition, an element of $\mathfrak{I}_{K+1}$ if it is not contained in any of $\mathfrak{I}_{k}$, $k=1,\cdots,K$. ###### Lemma 9.11. 1. (1) If $(\ell,\rho)\in\mathfrak{I}_{k}$ then $d\ell\in A_{k}^{\perp}$. 2. (2) If $(\ell,\rho)\in\mathfrak{I}_{k}$ then $\ell(\text{\bf 0})+\rho>S_{k}$. 3. (3) If $(\ell,\rho)\in\mathfrak{I}_{K+1}$ then $\ell(\text{\bf 0})+\rho>S_{K}$. ###### Proof. (1) follows from Definition 9.10 (1) and Lemma 9.7. If $(\ell,\rho)=(\ell_{j}^{\prime},\rho_{j})\in\mathfrak{I}_{k}$ then there exists $e_{j}^{i}\neq 0$, $\ell_{i}=\ell_{k,j^{\prime}}$. Then $\ell(\text{\bf 0})+\rho\geq e^{i}_{j}\ell_{i}(\text{\bf 0})+\rho_{j}>\ell_{i}(\text{\bf 0})=S_{k}.$ (2) follows. The proof of (3) is the same. ∎ ###### Lemma 9.12. The vector space $A_{k}$ is defined over ${\mathbb{Q}}$. ###### Proof. $A_{k}$ is defined by equalities of the type $\ell_{i}=S_{k}$ on $A_{k-1}$. Since the linear part of $\ell_{i}$ has integer coefficients, the lemma follows by induction on $k$. ∎ We put $d(k)=\mathop{\kern 0.0pt{\rm dim}}\nolimits A_{k-1}-\mathop{\kern 0.0pt{\rm dim}}\nolimits A_{k}=\mathop{\kern 0.0pt{\rm dim}}\nolimits A^{\perp}_{k}-\mathop{\kern 0.0pt{\rm dim}}\nolimits A^{\perp}_{k-1}$. We choose $\text{\bf e}_{i,j}^{*}\in Hom(M_{\mathbb{Q}},{\mathbb{Q}})\cong N_{\mathbb{Q}}$ ($i=1,\cdots,K$, $j=1,\cdots,d(k)$) such that the following condition holds. Here $M_{\mathbb{Q}}=M\otimes{\mathbb{Q}}$ and $N_{\mathbb{Q}}=N\otimes{\mathbb{Q}}$ ###### Condition 9.13. 1. (1) $\text{\bf e}_{1,1}^{*},\cdots,\text{\bf e}_{k,d(k)}^{*}$ is a ${\mathbb{Q}}$ basis of $A^{\perp}_{k}\cap N_{{\mathbb{Q}}}$. 2. (2) $d\ell_{k,j}=\sum_{k^{\prime},j^{\prime}}v_{(k,j),(k^{\prime},j^{\prime})}\text{\bf e}_{k^{\prime},j^{\prime}}^{*}$ with $v_{(k,j),(k^{\prime},j^{\prime})}\in{\mathbb{Z}}$. 3. (3) If $(\ell,\rho)\in\mathfrak{I}_{k}$ or $\ell\in\mathfrak{I}$, then $d\ell=\sum_{k^{\prime},j^{\prime}}v_{\ell,(k^{\prime},j^{\prime})}\text{\bf e}_{k^{\prime},j^{\prime}}^{*}$ with $v_{\ell,(k^{\prime},j^{\prime})}\in{\mathbb{Z}}$. Note $d(k)=0$ if $A_{k}=A_{k-1}$. We identify ${\mathbb{R}}^{n}$ with $H^{1}(L(u);{\mathbb{R}}^{n})$ in the same way as Lemma 4.4 and let $x_{k,j}\in Hom(H^{1}(L(u);{\mathbb{R}}),{\mathbb{R}})$ be the element corresponding to $\text{\bf e}_{k,j}^{*}$ by this identification. In other words, if $\text{\bf e}_{k,j}^{*}=\sum_{i}a_{(k,j);i}\text{\bf e}_{i}^{*},$ where $\text{\bf e}_{i}^{*}$ is as in Lemma 4.4, then we have $x_{k,j}=\sum_{i}a_{(k,j);i}x_{i}.$ We put $y_{k,j}=e^{x_{k,j}}.$ We define $Y(k,j)=\prod_{k^{\prime}=1}^{K}\prod_{j^{\prime}=1}^{d(k^{\prime})}y_{k^{\prime},j^{\prime}}^{v_{(k,j),(k^{\prime},j^{\prime})}}.$ (9.13) And for $(\ell,\rho)\in\mathfrak{I}_{k}$ or $\ell\in\mathfrak{I}$, we define $Y(\ell)=\prod_{k=1}^{K}\prod_{j=1}^{d(k)}y_{k,j}^{v_{\ell,(k,j)}}.$ (9.14) By Theorem 4.6 there exists $c_{(\ell,\rho)}\in{\mathbb{Q}}$ such that : $\displaystyle\mathfrak{PO}^{\text{\bf 0}}=$ $\displaystyle\sum_{k=1}^{K}\left(\sum_{j=1}^{a(k)}Y(k,j)\right)T^{S_{k}}+\sum_{\ell\in\mathfrak{I}}Y(\ell)T^{\ell(\text{\bf 0})}$ (9.15) $\displaystyle+\sum_{k=1}^{K+1}\sum_{(\ell,\rho)\in\mathfrak{I}_{k}}c_{(\ell,\rho)}Y(\ell)T^{\ell(\text{\bf 0})+\rho}$ where $\mathfrak{PO}^{\text{\bf 0}}$ is $\mathfrak{PO}^{u}$ with $u=\text{\bf 0}$. ###### Lemma 9.14. 1. (1) If $k^{\prime}<k$ then $\frac{\partial Y(k^{\prime},j^{\prime})}{\partial y_{k,j}}=0.$ (9.16) 2. (2) If $(\ell,\rho)\in\mathfrak{I}_{k^{\prime}}$, $k^{\prime}<k$ then $\frac{\partial Y(\ell)}{\partial y_{k,j}}=0.$ (9.17) 3. (3) If $(\ell,\rho)\in\mathfrak{I}_{k}$ then $\ell(\text{\bf 0})+\rho>S_{k}$. 4. (4) If $(\ell,\rho)\in\mathfrak{I}_{K+1}$ then $\ell(\text{\bf 0})+\rho>S_{K}$. 5. (5) If $\ell\in\mathfrak{I}$ then $\ell(\text{\bf 0})>S_{K}$. ###### Proof. Since $d\ell_{k^{\prime},j^{\prime}}\in A^{\perp}_{k^{\prime}}$ by Lemma 9.7 it follows that $v_{(k^{\prime},j^{\prime}),(k,j)}=0$ for $k>k^{\prime}$. (1) follows. (2) follows from Lemma 9.11 (1) in the same way. (3) follows from Lemma 9.11 (2). (4) follows from Lemma 9.11 (3). (5) follows from (9.12). ∎ Now equation (4.10) becomes $0=\frac{\partial\mathfrak{PO}^{\text{\bf 0}}}{\partial y_{k,j}}.$ We calculate this equation using Lemma 9.14 to find that it is equivalent to : $\displaystyle 0=$ $\displaystyle\sum_{j^{\prime}=1}^{a(k)}\frac{\partial Y(k,j^{\prime})}{\partial y_{k,j}}+\sum_{k^{\prime}>k}\sum_{j^{\prime}=1}^{a(k^{\prime})}\frac{\partial Y(k^{\prime},j^{\prime})}{\partial y_{k,j}}T^{S_{k^{\prime}}-S_{k}}$ (9.18) $\displaystyle+\sum_{k^{\prime}=k}^{K+1}\sum_{(\ell,\rho)\in\mathfrak{I}_{k^{\prime}}}c_{(\ell,\rho)}\frac{\partial Y(\ell)}{\partial y_{k,j}}T^{\ell(\text{\bf 0})+\rho- S_{k}}+\sum_{\ell\in\mathfrak{I}}\frac{\partial Y(\ell)}{\partial y_{k,j}}T^{\ell(\text{\bf 0})-S_{k}}.$ Note the exponents of $T$ in the second, third, and fourth terms of (9.18) are all strictly positive. So after putting $T=0$ we have $0=\sum_{j^{\prime}=1}^{a(k)}\frac{\partial Y(k,j^{\prime})}{\partial y_{k,j}}.$ (9.19) Note that the equation (9.19) does not involve $T$ but becomes a numerical equation. We call (9.19) the leading term equation. ###### Lemma 9.15. There exist positive real numbers $\mathfrak{y}_{k,j;0}$, $k=1,\cdots,K$, $j=1,\cdots,d(k)$, solving the leading term equations for $k=1,\cdots,K$. ###### Proof. We remark the leading term equation for $k,j$ contain the monomials involving only $y_{k^{\prime},j}$ for $k^{\prime}\leq k$. We first solve the leading term equation for $k=1$. Denote $f_{1}(x_{1,1},\cdots,x_{1,d(1)})=\sum_{j=1}^{a(1)}Y(1,j).$ It follows from Corollary 9.9 that for any $(x_{1,1},\cdots,x_{1,d(1)})\neq 0$, there exists $j$ such that $d\ell_{1,j}(x_{1,1},\cdots,x_{1,d(1)})>0.$ Therefore, we have $\lim_{t\to\infty}f_{1}(tx_{1,1},\cdots,tx_{1,d(1)})\geq\lim_{t\to\infty}C\exp(td\ell_{1,j}(x_{1,1},\cdots,x_{1,d(1)}))=+\infty.$ Hence $f_{1}(x_{1,1},\cdots,x_{1,d(1)})$ attains its minimum at some point of ${\mathbb{R}}^{d(1)}$. Taking its exponential, We obtain $\mathfrak{y}_{1,j;0}\in{\mathbb{R}}_{+}$. Suppose we have already found $\mathfrak{y}_{k^{\prime},j;0}$ for $k^{\prime}<k$. Then we put $F_{k}(x_{1,1},\cdots,x_{k,1},\cdots,x_{k,d(k)})=\sum_{j=1}^{a(k)}Y(k,j)$ and $f_{k}(x_{k,1},\cdots,x_{k,d(k)})=F_{k}(\mathfrak{x}_{1,1;0},\cdots,\mathfrak{x}_{k-1,d(k-1);0},x_{k,1},\cdots,x_{k,d(k)})$ where $\mathfrak{x}_{k^{\prime},j;0}=\log\mathfrak{y}_{k^{\prime},j;0}$. Again using Corollary 9.9, we find $\lim_{t\to\infty}f_{k}(tx_{k,1},\cdots,tx_{k,d(k)})=+\infty.$ for any $(x_{k,1},\cdots,x_{k,d(k)})\neq 0$. Hence $f_{k}(x_{k,1},\cdots,x_{k,d(k)})$ attains a minimum and we obtain $\mathfrak{y}_{k,j;0}$. Lemma 9.15 now follows by induction. ∎ We next find the solution of our equation (4.11) or (4.12). We take a sufficiently large ${\mathcal{N}}$ and put $\displaystyle\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}}}=$ $\displaystyle\sum_{j=1}^{a(k)}Y(k,j)+\sum_{k^{\prime}>k}\sum_{j^{\prime}=1}^{a(k^{\prime})}\frac{\partial Y(k^{\prime},j^{\prime})}{\partial y_{k,j}}T^{S_{k^{\prime}}-S_{k}}$ (9.20) $\displaystyle+\sum_{\ell\in\mathfrak{I},\,\,\ell(\text{\bf 0})\leq{\mathcal{N}}}Y(\ell)T^{\ell(\text{\bf 0})-S_{k}}$ $\displaystyle+\sum_{k^{\prime}=k+1}^{K+1}\sum_{(\ell,\rho)\in\mathfrak{I}_{k^{\prime}},\ell(\text{\bf 0})+\rho\leq{\mathcal{N}}}c_{(\ell,\rho)}Y(\ell)T^{\ell(\text{\bf 0})+\rho- S_{k}}.$ We remark that (4.11) is equivalent to $\frac{\partial\mathfrak{PO}_{k,{\mathcal{N}}}^{\text{\bf 0}}}{\partial y_{k,j}}(\mathfrak{y}_{1},\cdots,\mathfrak{y}_{n})\equiv 0\mod T^{{\mathcal{N}}-S_{k}}\qquad k=1,\cdots,K,\,j=1,\cdots,a(k).$ (9.21) We also put $\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k}=\sum_{j=1}^{a(k)}Y(k,j).$ It satisfies $\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k}\equiv\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}}}\mod\Lambda_{+}.$ (9.22) For given positive numbers $R(1),\cdots,R(K)$ we define the discs $D(R(k))=\\{(x_{k,1}\cdots,x_{k,d(k)})\mid x_{k,1}^{2}+\cdots+x_{k,d(k)}^{2}\leq R(k)\\}\subset{\mathbb{R}}^{d(k)}$ and the poly-discs $\displaystyle D(R(\cdot))$ $\displaystyle=$ $\displaystyle\prod_{k=1}^{K}D(R(k))$ $\displaystyle=$ $\displaystyle\\{(x_{1,1}\cdots,x_{K,d(K)})\mid x_{k,1}^{2}+\cdots+x_{k,d(k)}^{2}\leq R(k),\,k=1,\cdots,K\\}.$ We factorize ${\mathbb{R}}^{n}=\prod_{k=1}^{K}{\mathbb{R}}^{d(k)}.$ Then we consider the Jacobian of $\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k}$ $\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k}:{\mathbb{R}}^{n}\to{\mathbb{R}}^{d(k)}$ i.e., the map $(\mathfrak{x}_{1,1},\cdots,\mathfrak{x}_{K,d(K)})\mapsto\left(\frac{\partial\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k}}{\partial x_{k,j}}(\mathfrak{x}_{1,1},\cdots,\mathfrak{x}_{K,d(K)})\right)_{j=1,\cdots,d(k)}.$ (9.23) We remark that $\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k}$ depends only on ${\mathbb{R}}^{d(1)}\times\cdots\times{\mathbb{R}}^{d(k)}$ components. Combining all $\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k}$, $k=1,\cdots,K$ (9.23) induces a map $\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}:{\mathbb{R}}^{n}\to{\mathbb{R}}^{n}$ defined by $\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}=(\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}_{1},\cdots,\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}_{K}).$ The next lemma is closely related to Lemma 9.15. ###### Lemma 9.16. We may choose the positive numbers $R(k)$ for $k=1,\cdots,K$ such that the following holds : 1. (1) $\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}$ is nonzero on $\partial(D(R(\cdot)))$. 2. (2) The map $:\partial(D(R(\cdot)))\to S^{n-1}$ $\mathfrak{x}\mapsto\frac{\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}}{\|\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}\|}$ has degree $1$. ###### Proof. We first prove the following sublemma by an upward induction on $k_{0}$. ###### Sublemma 9.17. There exist $R(k)$’s for $1\leq k\leq K$ such that for any given $1\leq k_{0}\leq K$ we have $\sum_{j=1}^{d(k_{0})}x_{k_{0},j}\frac{\partial\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k_{0}}}{\partial x_{k_{0},j}}(x_{1,1},\cdots,x_{k_{0},d(k_{0})})>0$ (9.24) if $(x_{k,1},\cdots,x_{k,d(k)})\in D(R(k))$ for all $1\leq k\leq k_{0}-1$ and $(x_{k_{0},1},\cdots,x_{k_{0},d(k_{0})})\in\partial D(R(k_{0}))$. ###### Proof. In case $k_{0}=1$ the existence of $R(1)$ satisfying (9.24) is a consequence of Corollary 9.9. We assume that the sublemma is proved for $1,\cdots,k_{0}-1$. For each fixed $\text{\bf x}=(x_{1,1},\cdots,x_{k_{0}-1,d(k_{0}-1)})$ we can find $R(k_{0})_{\text{\bf x}}$ such that (9.24) holds for $(x_{k_{0},1},\cdots,x_{k_{0},d(k_{0})})\in{\mathbb{R}}^{d(k_{0})}\setminus D(R(k_{0})_{\text{\bf x}}/2)$. This is also a consequence of Corollary 9.9. We take supremum of $R(k_{0})_{\text{\bf x}}$ over the compact set $\text{\bf x}\in\prod_{k=1}^{k_{0}-1}D(R(k))$ and obtain $R(k_{0})$. The proof of Sublemma 9.17 is complete. ∎ It is easy to see that Lemma 9.16 follows from Sublemma 9.17. ∎ We now use our assumption that the vertices of $P$ lies in $M_{\mathbb{Q}}={\mathbb{Q}}^{n}$ and that $\rho_{j}\in{\mathbb{Q}}$. Replacing $T$ by $T^{1/\mathcal{C}}$ if necessary, we may assume that all the exponents of $y_{k,j}$ and $T$ appearing in (9.20) are integers. Then $\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}}}=\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}}}(y_{1,1},\cdots,y_{K,d(K)};T)$ are polynomials of $y_{k,j}$, $y_{k,j}^{-1}$ and $T$. Define the set $\mathfrak{X}$ by the set consisting of $(\mathfrak{y}_{1,1},\cdots,\mathfrak{y}_{K,d(K)};q)\in({\mathbb{R}}_{+})^{n}\times{\mathbb{R}}$ that satisfy $\frac{\partial\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}}}}{\partial y_{k,j}}(\mathfrak{y}_{1,1},\cdots,\mathfrak{y}_{K,d(K)};q)=0,$ (9.25) for $k=1,\cdots,K$, $j=1,\cdots,d(k)$. Clearly $\mathfrak{X}$ is a real affine algebraic variety. (Note the equation for $y_{i}$ are polynomials. So we need to regard $y_{i}$ (not $x_{i}$) as variables to regard $\mathfrak{X}$ as a real affine algebraic variety. ) Consider the projection $\pi:\mathfrak{X}\to{\mathbb{R}},\quad\pi(\mathfrak{y}_{1,1},\cdots,\mathfrak{y}_{K,d(K)};q)=q$ which is a morphism of algebraic varieties. ###### Lemma 9.18. There exists a sufficiently small $\epsilon>0$ such that if $|q|<\epsilon$ then $\pi^{-1}(q)\cap\\{(e^{x_{1}},\cdots,e^{x_{n}})\mid(x_{1},\cdots,x_{n})\in D(R(\cdot))\\}\neq\emptyset.$ ###### Proof. We consider the real analytic $q$-family of polynomials $\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}},q}(y_{1,1},\cdots,y_{K,d(K)})=\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}}}(y_{1,1},\cdots,y_{K,d(K)};q).$ Then $\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}},0}=\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k,{\mathcal{N}}}.$ (9.26) Replacing $\overline{\mathfrak{PO}}^{\text{\bf 0}}_{k,{\mathcal{N}}}$ by $\mathfrak{PO}^{\text{\bf 0}}_{k,{\mathcal{N}},q}$, we can repeat construction of the map $\nabla\mathfrak{PO}^{\text{\bf 0}}_{{\mathcal{N}},q}:{\mathbb{R}}^{n}\to{\mathbb{R}}^{n}$ for each fixed $q\in{\mathbb{R}}$ in the same way as we defined $\nabla\overline{\mathfrak{PO}}^{\text{\bf 0}}$. Then the conclusion of Lemma 9.16 holds for $\nabla\mathfrak{PO}^{\text{\bf 0}}_{{\mathcal{N}},q}$ if $|q|$ is sufficiently small. (This is a consequence of Lemma 9.16 and (9.26).) Lemma 9.18 follows from elementary algebraic topology. ∎ Lemma 9.18 implies that we can find $\mathfrak{y}_{0}=(\mathfrak{y}_{1,1;0},\cdots,\mathfrak{y}_{K,d(K);0})\in{\mathbb{R}}_{+}^{n}$ and a sequence $(\mathfrak{y}_{h},q_{h})=(\mathfrak{y}^{h}_{1,1;0},\cdots,\mathfrak{y}^{h}_{K,d(K);0};q_{h})\in\mathfrak{X}\subset{\mathbb{R}}^{n+1}$ $h=1,2,\cdots$ such that $q_{h}>0$ and $\lim_{h\to\infty}(\mathfrak{y}_{h},q_{h})=(\mathfrak{y}_{0},0).$ Therefore by the curve selection lemma (Lemma 3.1 [Mi]) there exists a real analytic map $\gamma:[0,\epsilon)\to\mathfrak{X}$ such that $\gamma(0)=(\mathfrak{y}_{0},0)$ and $\pi(\gamma(t))>0$ for $t>0$. We reparameterize $\gamma(t)$, so that its $q$-component is $t^{a/b}$, where $a$ and $b$ are relatively prime integers. We put $T=t^{a/b}$ i.e., $t=T^{b/a}$ and denote the $y_{k,j}$-components of $\gamma(t)$ by $\mathfrak{y}_{k,j}=\mathfrak{y}_{k,j;0}+\sum_{\ell=1}^{\infty}\mathfrak{y}_{k,j;\ell}T^{b\ell/a}.$ Since $\gamma(t)\in\mathfrak{X}$, the element $(\mathfrak{y}_{k,j})_{k,j}\in(\Lambda^{{\mathbb{R}}}_{0}\setminus\Lambda^{{\mathbb{R}}}_{+})^{n}$ is the required solution of (4.11). Since $\mathfrak{PO}_{0}$ contains only a finite number of summands, we can take $\mathfrak{PO}_{0,{\mathcal{N}}}=\mathfrak{PO}_{0}$. Therefore we can find a solution of (4.12) for $\mathfrak{PO}_{0}$. The proof of Proposition 4.7 is now complete. ∎ ## 10\. Elimination of higher order term in nondegenerate cases In this section, we prove a rather technical (but useful) result, which shows that solutions of the leading term equation (9.19) correspond to actual critical points under certain non-degeneracy condition. For this purpose, we slightly modify the argument of the last part of section 9. This result will be useful to determine $u\in\text{\rm Int}\,P$ such that $HF((L(u);\mathfrak{x}),(L(u);\mathfrak{x});\Lambda_{0})\neq 0$ for some $\mathfrak{x}$ in non-Fano cases. (In other words we study the image of $\mathfrak{M}(\mathfrak{Lag}(X))\to\text{\rm Int}\,P$ by the map $(\mathfrak{x},u)\mapsto u$.) In fact it shows that we can use $\mathfrak{PO}_{0}$ in place of $\mathfrak{PO}$ in most practical cases. We remark that we explicitly calculate $\mathfrak{PO}_{0}$ but do not know the precise form of $\mathfrak{PO}$ in non-Fano cases. In order to state the result in a general form, we prepare some notations. Let $u_{0}\in\text{\rm Int}\,P$. (In section 9, $u_{0}$ is determined as the unique element of $P_{K}$ defined in Proposition 9.1. The present situation is more general.) We define positive real numbers $S_{1}<S_{2}<\cdots$ by $\\{\ell_{i}(u_{0})\mid i=1,\cdots,m\\}=\\{S_{1},S_{2},\cdots,S_{m^{\prime}}\\}$ (10.1) and the sets $I_{k}=\\{\ell_{i}\mid\ell_{i}(u_{0})=S_{k}\\},$ (10.2) for $k=1,\cdots$. We renumber each of $I_{k}$ so that $\\{\ell_{k,j}\mid j=1,\cdots,a(k)\\}=I_{k}.$ (10.3) ###### Definition 10.1. Let $A^{\perp}_{l}$ be the linear subspace of $N_{\mathbb{R}}$ spanned by $d\ell_{k,j}$ $k\leq l$, $j\leq a(k)$. We define $K$ to be the smallest number such that $A^{\perp}_{K}=N_{\mathbb{R}}$. Note our notations here are consistent with one in section 9 in case $\\{u_{0}\\}=P_{K}$. We define $\mathfrak{I}$ and $\mathfrak{I}_{k}$ by (9.11) and Definition 9.10. Then Lemma 9.11 and (9.12) hold. We choose $\text{\bf e}_{i,j}^{*}\in Hom(M_{\mathbb{Q}},{\mathbb{Q}})$ such that Condition 9.13 is satisfied. (Note $A^{\perp}_{l}$ is defined over ${\mathbb{Q}}$.) $x_{i,j}$ and $y_{i,j}$ then are defined in the same way as section 9. We define $Y(k,j)$ by (9.13) and $Y(\ell)$ by (9.14). Then (9.15) and Lemma 9.14 hold. We remark that Corollary 9.9 does not hold in general in the current situation. In fact we can write $v=\sum_{l=1}^{k}\sum_{j=1}^{a(l)}c_{l,j}d\ell_{l,j}$ under the assumption of Corollary 9.9 but we may not be able to ensure $c_{l,j}\geq 0$. ###### Definition 10.2. 1. (1) We call $0=\sum_{j^{\prime}=1}^{a(k)}\frac{\partial Y(k,j^{\prime})}{\partial y_{k,j}},\quad k=1,\cdots,K,\,\,j=1,\cdots,d(k)$ the leading term equation at $u_{0}$. We regard it as a polynomial equation for $\mathfrak{y}_{k,j}\in{\mathbb{C}}\setminus\\{0\\}$, $k=1,\cdots,K$, $j=1,\cdots,d(k)$. 2. (2) A solution $\mathfrak{y}^{0}=(\mathfrak{y}_{k,j;0})_{k=1,\cdots,K,\,j=1,\cdots,d(k)}$ of leading term equation is said to be weakly nondegenerate if it is isolated in the set of solutions. 3. (3) A solution $\mathfrak{y}^{0}=(\mathfrak{y}_{k,j;0})_{k=1,\cdots,K,\,j=1,\cdots,d(k)}$ of leading term equation is said to be strongly nondegenerate if the matrices $\left(\sum_{j^{\prime}=1}^{a(k)}\frac{\partial^{2}Y(k,j^{\prime})}{\partial y_{k,j_{1}}\partial y_{k,j_{2}}}\right)_{j_{1},j_{2}=1,\cdots,a(k)}$ are invertible for $k=1,\cdots,K$, at $\mathfrak{y}^{0}$. 4. (4) We define the multiplicity of leading term equation in the standard way of algebraic geometry, in the weakly nondegenerate case. ###### Example 10.3. In Example 9.6, the equation (9.6) is the leading term equation. Let $\mathfrak{PO}^{u_{0}}_{*}$ be either $\mathfrak{PO}^{u_{0}}_{0}$ or $\mathfrak{PO}^{u_{0}}$. ###### Theorem 10.4. For any strongly nondegenerate solution $\mathfrak{y}^{0}=(\mathfrak{y}_{k,j;0})$ of leading term equation, there exists a solution $\mathfrak{y}=(\mathfrak{y}_{k,j})\in(\Lambda^{{\mathbb{C}}}_{0}\setminus\Lambda^{{\mathbb{C}}}_{+})^{n}$ of $\frac{\partial\mathfrak{PO}^{u_{0}}_{*}}{\partial y_{k,j}}(\mathfrak{y})=0$ (10.4) such that $\mathfrak{y}_{k,j}\equiv\mathfrak{y}_{k,j;0}\mod\Lambda^{{\mathbb{C}}}_{+}.$ If all the vertices of $P$ and $u_{0}$ are rational, the same conclusion holds for weakly nondegenerate $\mathfrak{y}^{0}$. We prove the following at the end of section 13. ###### Lemma 10.5. We assume $[\omega]\in H^{2}(X;{\mathbb{Q}})$ and choose the moment polytope $P$ such that its vertices are all rational. Let $u_{0}\in\text{\rm Int}P$ such that $\mathfrak{PO}_{0}^{u_{0}}$ has weakly nondegenerate critical point in $(\Lambda_{0}\setminus\Lambda_{+})^{n}$. Then $u_{0}$ is rational. The following corollary is an immediate consequence of Theorem 10.4 and Lemma 10.5. ###### Corollary 10.6. Let $(\mathfrak{x},u)\in\mathfrak{M}_{0}(\mathfrak{Lag}(X))$ and $u\in\text{\rm Int}\,P$. Assume one of the following conditions : 1. (1) The corresponding solution of the leading term equation is strongly nondegenerate. 2. (2) $[\omega]\in H^{2}(X;{\mathbb{R}})$ is rational and the corresponding solution of leading term equation is weakly nondegenerate. Then there exists $\mathfrak{x}^{\prime}$ such that $(\mathfrak{x}^{\prime},u)\in\mathfrak{M}(\mathfrak{Lag}(X))$ and $\mathfrak{x}^{\prime}\equiv\mathfrak{x}\mod\Lambda^{{\mathbb{C}}}_{+}$. ###### Remark 10.7. 1. (1) Using Proposition 10.8 below, we can also apply Theorem 10.4 and Corollary 10.6 to study non-displacement of Lagrangian fibers for the weakly nondegenerate case, without assuming rationality. See the last step of the proof of Theorem 1.5 given at the end of section 13. 2. (2) The conclusion of Theorem 10.4 does not hold in general without weakly nondegenerate assumption. We give an example (Example 10.17) where both the assumption of weak nondegeneracy and the conclusion of Theorem 10.4 fail to hold. 3. (3) In this section we work with $\Lambda^{{\mathbb{C}}}$ coefficients, while in the last section we work with $\Lambda^{{\mathbb{R}}}$ coefficients. We also remark that in the last section, we did not assume the weak nondegeneracy condition. 4. (4) If we define the multiplicity of the element of $\mathfrak{M}_{0}(\mathfrak{Lag}(X))$ as the dimension of the Jacobian ring $Jac(\mathfrak{PO}_{0};\mathfrak{x},u_{0})$ in Definition 7.9 (namely as the Milnor number) then the sum of the multiplicities of the solutions of (10.4) converging to $\mathfrak{y}^{0}$ as $T\to 0$, is equal to the multiplicity of $\mathfrak{y}^{0}$. 5. (5) In the strongly nondegenerate case, the solution of (10.4) with the given leading term is unique. ###### Proposition 10.8. Let $(X,\omega)$ be a compact toric manifold with moment polytope $P$ and $u_{0}\in\text{\rm Int}\,P$. Then there exist $(X,\omega^{h})$ with moment polytope $P^{h}$ and $u_{0}^{h}\in\text{\rm Int}\,P^{h}$ such that the following holds: 1. (1) $\lim_{h\to\infty}\omega^{h}=\omega$. $\lim_{h\to\infty}u^{h}_{0}=u_{0}$. 2. (2) The vertices of $P^{h}$ and $u_{0}^{h}$ are rational. 3. (3) The leading term equation at $u_{0}^{h}$ is the same as the leading term equation at $u_{0}$. We prove Proposition 10.8 at the end of section 13. We first derive Theorem 1.14 and Lemma 8.3 from Theorem 10.4 before proving Theorem 10.4. ###### Proof of Theorem 1.14. We start with ${\mathbb{C}}P^{2}$ and blow up a $T^{2}$ fixed point to obtain ${\mathbb{C}}P^{2}\\#(-{\mathbb{C}}P^{2})$. We take a Kähler form so that the volume of the exceptional ${\mathbb{C}}P^{1}$ is $2\pi\epsilon_{1}$ which is small. We next blow up again at one of the fixed points so that the volume of the exceptional ${\mathbb{C}}P^{1}$ is $\epsilon_{2}$ and is small compared with $\epsilon_{1}$. We continue $k$ times to obtain $X(k)$, whose Kähler structure depends on $\epsilon_{1},\cdots,\epsilon_{k}$. Note $X(k)$ is non- Fano for $k>3$. Let $P(k)$ be the moment polytope of $X(k)$ and $\mathfrak{PO}_{0,k}$ be the leading order potential function of $X(k)$. We remark that $P(k)$ is obtained by cutting out a vertex of $P(k-1)$. (See [Ful].) ###### Lemma 10.9. We may choose $\epsilon_{i}$ ($i=1,\cdots,k$) so that the following holds for $l\leq k$. 1. (1) The number of balanced fibers of $P(l)$ is $l+1$. We write them as $L(u^{(l,i)})$ $i=0,\cdots,l$. 2. (2) $u^{(l-1,i)}=u^{(l,i)}$ for $i\leq l-1$. $u^{(l,0)}=(1/3,1/3)$. 3. (3) $u^{(l,l)}$ is in an $o(\epsilon_{l})$ neighborhood of the vertices corresponding to the point we blow up. 4. (4) The leading term equation of $\mathfrak{PO}_{0,l-1}$ at $u^{(l-1,i)}$ is the same as the leading term equation of $\mathfrak{PO}_{0,l}$ at $u^{(l,i)}$ for $i\leq l-1$. 5. (5) The leading term equations are all strongly nondegenerate. ###### Proof. The proof is by induction on $k$. There is nothing to show for $k=0$. Suppose that we have proved Lemma 10.9 up to $k-1$. Let $w$ be the vertex of the polytope we cut out which corresponds to the blow up of $X(k-1)$. Let $\ell_{i}$, $\ell_{i^{\prime}}$ be the affine functions associated to the two edges containing $w$. It is easy to see that $P(k)=\\{u\in P(k-1)\mid\ell_{i}(u)+\ell_{i^{\prime}}(u)\geq\epsilon_{k}\\}.$ We also have : $\mathfrak{PO}_{0,k}=\mathfrak{PO}_{0,k-1}+T^{\ell_{i}(u)+\ell_{i^{\prime}}(u)-\epsilon_{k}}y_{1}^{v_{i,1}+v_{i^{\prime},1}}y_{2}^{v_{i,2}+v_{i^{\prime},2}}.$ Therefore if we choose $\epsilon_{k}$ sufficiently small, the leading term equation at $u^{(k-1,i)}$ does not change. We take $u^{(k,k)}$ such that $\ell_{i}(u^{(k,k)})=\ell_{i^{\prime}}(u^{(k,k)})=\epsilon_{k}.$ It is easy to see that there exists such $u^{(k,k)}$ uniquely if $\epsilon_{k}$ is sufficiently small. We put $y^{\prime}_{1}=y_{1}^{v_{i,1}}y_{2}^{v_{i,2}},\quad y^{\prime}_{2}=y_{1}^{v_{i^{\prime},1}}y_{2}^{v_{i^{\prime},2}}.$ (We remark that $v_{i}$ and $v_{i^{\prime}}$ are ${\mathbb{Z}}$ basis of ${\mathbb{Z}}^{2}$, since $X(k-1)$ is smooth toric.) Then we have $\mathfrak{PO}^{u^{(k,k)}}_{0,k}\equiv(y_{1}^{\prime}+y^{\prime}_{2}+y^{\prime}_{1}y^{\prime}_{2})T^{\epsilon_{k}}\mod T^{\epsilon_{k}}\Lambda_{+}.$ Therefore the leading term equation is $1+y^{\prime}_{1}=1+y^{\prime}_{2}=0$ and hence has a unique solution $(-1,-1)$. In particular it is strongly nondegenerate. We can also easily check that there is no other solution of leading term equation. The proof of Lemma 10.9 now follows by Theorem 10.4. ∎ Theorem 1.14 immediately follows from Lemma 10.9. ∎ Note that Theorem 1.14 can be generalized to ${\mathbb{C}}P^{n}$ by the same proof. ###### Proof of Lemma 8.3. Let $u_{0}=(n(\alpha+1)/4,(1-\alpha)/2)$. We calculate $\mathfrak{PO}_{0}^{u_{0}}=(y_{2}+y_{2}^{-1})T^{(1-\alpha)/2}+(y_{1}+y_{1}^{-1}y_{2}^{-1})T^{n(\alpha+1)/4}.$ The leading equation is $1-y^{-2}_{2}=0,\quad 1-y_{1}^{-2}y_{2}^{-1}=0.$ Its solution are $(1,1),(-1,1),(\sqrt{-1},-1),(-\sqrt{-1},-1)$, all of which are strongly nondegenerate. The lemma them follows from Corollary 10.6. ∎ We give another example which demonstrates the way how one can use the leading term equation and Theorem 10.4 to locate balanced fibers. ###### Example 10.10. Let us consider ${\mathbb{C}}P^{n}$ with moment polytope $P=\\{(u_{1},\cdots,u_{n})\mid u_{i}\geq 0,\sum u_{i}\leq 1\\}$. We take ${\mathbb{C}}P^{n-\ell}\subset{\mathbb{C}}P^{n}$ corresponding to $u_{1}=\cdots=u_{\ell}=0$. ($\ell\geq 2$.) We blow up ${\mathbb{C}}P^{n}$ along the center ${\mathbb{C}}P^{n-\ell}$ and denote the blow-up by $X$. (The case $\ell=n=2$ is Example 8.1.) We take $\alpha\in(0,1)$ so that the moment polytope of $X$ is $P_{\alpha}=\left\\{(u_{1},\cdots,u_{n})\in P~{}\Big{|}~{}\sum_{i=1}^{\ell}u_{i}\geq\alpha\right\\}.$ Below we use $\mathfrak{PO}_{0}$ in place of the potential function $\mathfrak{PO}$. Since the all the critical points of $\mathfrak{PO}_{0}$ are weakly nondegenerate, they correspond to the critical points of $\mathfrak{PO}$. (We thank D. McDuff for pointing out that this example is not Fano.) The function $\mathfrak{PO}_{0}$ is given by $\mathfrak{PO}_{0}=\sum_{i=1}^{n}T^{u_{i}}y_{i}+T^{1-\sum_{i=1}^{n}u_{i}}(y_{1}\cdots y_{n})^{-1}+T^{\sum_{i=1}^{\ell}u_{i}-\alpha}y_{1}\cdots y_{\ell}.$ We denote $z_{i}=T^{u_{i}}y_{i},\quad z_{0}=T^{1-\sum_{i=1}^{n}u_{i}}(y_{1}\cdots y_{n})^{-1},\quad z=T^{\sum_{i=1}^{\ell}u_{i}-\alpha}y_{1}\cdots y_{\ell}.$ Then the quantum Stanley-Reisner relations are : $z_{1}\cdots z_{n}z_{0}=T,\,\,\,z_{1}\cdots z_{\ell}=zT^{\alpha}.$ By computing the derivatives $y_{i}\frac{\partial\mathfrak{PO}_{0}}{\partial y_{i}}$, we obtain the linear relations which can be written as $\displaystyle z_{i}-z_{0}+z$ $\displaystyle=$ $\displaystyle 0\,\quad\mbox{for }\,i\leq\ell$ $\displaystyle z_{i}-z_{0}$ $\displaystyle=$ $\displaystyle 0\,\quad\mbox{for }\,i>\ell.$ Putting $X=z_{0}-z$, $Y=z$, we obtain $z_{i}=\begin{cases}X&\quad\mbox{for $1\leq i\leq\ell$}\\\ X+Y&\quad\mbox{for $\,i>\ell$ or $i=0$}.\end{cases}$ We also have $X^{\ell}=YT^{\alpha}$ and $X^{n+1}(X^{\ell-1}T^{-\alpha}+1)^{n-\ell+1}=T.$ (10.5) Case 1; $(\ell-1)\mathfrak{v}_{T}(X)<\alpha$ : In this case, we obtain $\mathfrak{v}_{T}(X)=\frac{1+\alpha(n-\ell+1)}{n+1+(n-\ell+1)(\ell-1)}.$ The condition $(\ell-1)\mathfrak{v}_{T}(X)<\alpha$ then is equivalent to $\alpha>\frac{\ell-1}{n+1}.$ And we have $\mathfrak{v}_{T}(X+Y)=\mathfrak{v}_{T}(Y)=\ell\mathfrak{v}_{T}(X)-\alpha<\mathfrak{v}_{T}(X)$ for $i>\ell$. At the point $\text{\bf u}=(u_{1},\cdots,u_{n})$ with $u_{i}=\mathfrak{v}_{T}(X)$ for $1\leq i\leq\ell$ and $u_{i}=\mathfrak{v}_{T}(X+Y)$ for $i\geq\ell$, we have $\mathfrak{PO}_{0}^{\text{\bf u}}=T^{\mathfrak{v}_{T}(Y)}(y_{\ell+1}+\cdots+y_{n}+(y_{1}\cdots y_{n})^{-1}+y_{1}\cdots y_{\ell})+T^{\mathfrak{v}_{T}(X)}(y_{1}+\cdots+y_{\ell}).$ To obtain the leading term equation, it is useful to make a change of variables from $y_{1},\cdots,y_{n}$ to $y_{2},\cdots,y_{n},y$ with $y=y_{1}\cdots y_{\ell}$. In fact, using the notation which we introduced at the beginning of this section, we have $S_{1}=\mathfrak{v}_{T}(Y),\quad S_{2}=\mathfrak{v}_{T}(X),$ and $A^{\perp}_{1}$ is the vector space generated by $\partial/\partial u_{i}$, $(i=\ell+1,\cdots,n)$ and $\partial/\partial u_{1}+\cdots+\partial/\partial u_{\ell}$. We also have $A^{\perp}_{2}={\mathbb{R}}^{n}$. Therefore a basis satisfying Condition 9.13 is $\frac{\partial}{\partial u_{2}},\quad\cdots,\quad\frac{\partial}{\partial u_{n}},\quad\frac{\partial}{\partial u_{1}}+\cdots+\frac{\partial}{\partial u_{\ell}}.$ The variables corresponding to this basis is $y_{2},\cdots,y_{n},y$. In these variables, $\mathfrak{PO}_{0}^{\text{\bf u}}$ has the form $\displaystyle\mathfrak{PO}_{0}^{\text{\bf u}}$ $\displaystyle=$ $\displaystyle T^{\mathfrak{v}_{T}(Y)}(y_{\ell+1}+\cdots+y_{n}+(yy_{\ell+1}\cdots y_{n})^{-1}+y)$ $\displaystyle\quad+T^{\mathfrak{v}_{T}(X)}(y(y_{2}\cdots y_{\ell})^{-1}+y_{2}+\cdots+y_{\ell}).$ Then the leading term equation becomes : $\begin{cases}0=1-y^{-1}(yy_{\ell+1}\cdots y_{n})^{-1}\\\ 0=1-y_{i}^{-1}(yy_{\ell+1}\cdots y_{n})^{-1}&\mbox{for }i>\ell\\\ 0=1-y_{i}^{-1}y(y_{2}\cdots y_{\ell})^{-1}\,&\mbox{for }\,2\leq i\leq\ell.\end{cases}$ Its solutions are $\displaystyle y_{\ell+1}=\cdots=y_{n}=y=\theta,$ $\displaystyle\qquad\theta^{n-\ell+2}=1$ $\displaystyle y_{2}=\cdots=y_{\ell}=\rho,$ $\displaystyle\qquad\rho^{\ell}=\theta.$ It follows that this leading term equation has $\ell(n-\ell+2)$ solutions, all of which are strongly nondegenerate. Case 2; $(\ell-1)\mathfrak{v}_{T}(X)>\alpha$ : We have $\mathfrak{v}_{T}(X)=\frac{1}{n+1}.$ We also have $\mathfrak{v}_{T}(Y)=\ell\mathfrak{v}_{T}(X)-\alpha>\mathfrak{v}_{T}(X)$ and hence $\mathfrak{v}_{T}(X+Y)=\mathfrak{v}_{T}(X)=\frac{1}{n+1}.$ The condition $(\ell-1)\mathfrak{v}_{T}(X)>\alpha$ becomes $\alpha<\frac{\ell-1}{n+1}.$ If we consider the point $\text{\bf u}=(u_{1},\cdots,u_{n})$ with $u_{i}=\mathfrak{v}_{T}(X)=\frac{1}{n+1}$ for $i=1,\cdots n$, $\mathfrak{PO}_{0}^{\text{\bf u}}$ has the form $\mathfrak{PO}_{0}^{\text{\bf u}}=T^{1/n+1}(y_{1}+\cdots+y_{n}+(y_{1}\cdots y_{n})^{-1})+T^{\frac{\ell}{n+1}-\alpha}y_{1}\cdots y_{\ell}$ and so the leading term equation is obtained by differentiating $y_{1}+\cdots+y_{n}+(y_{1}\cdots y_{n})^{-1}.$ Its solutions are $y_{1}=\cdots=y_{n}=\theta,\qquad\theta^{n+1}=1.$ There are $n+1$ solutions all of which are strongly nondegenerate. Case 3; $(\ell-1)\mathfrak{v}_{T}(X)=\alpha$ : (Case 3-1; $-X^{\ell-1}\not\equiv T^{\alpha}\mod T^{\alpha}\Lambda_{+}$.) We have $u_{i}=\mathfrak{v}_{T}(X)=\mathfrak{v}_{T}(Y)=1/(n+1)$ ($i=1,\cdots,n$), $\alpha=(\ell-1)/(n+1)$. In this case $\mathfrak{PO}_{0}^{\text{\bf u}}=T^{1/(n+1)}(y_{1}+\cdots+y_{n}+(y_{1}\cdots y_{n})^{-1}+y_{1}\cdots y_{\ell}).$ This formula implies that the symplectic area of all discs with Maslov index $2$ are $2\pi/(n+1)$. The leading term equation is $\begin{cases}1-\frac{1}{y_{i}}((y_{1}\cdots y_{n})^{-1}-y_{1}\cdots y_{\ell})=0,&i=1,\cdots,\ell\\\ 1-\frac{1}{y_{i}}((y_{1}\cdots y_{n})^{-1})=0,&i=\ell+1,\cdots,n.\end{cases}$ Its solutions are $y_{1}=\cdots=y_{\ell}=\rho$, $y_{\ell+1}=\cdots=y_{n}=\theta$ with $\rho^{\ell}(\rho+\rho^{\ell})^{n-\ell+1}=1,$ (10.6) and $\theta=\rho+\rho^{\ell}$. It looks rather cumbersome to check whether (10.6) has multiple root. Certainly all the solutions are weakly nondegenerate. The number of solutions counted with multiplicity is $\ell(n-\ell+2)$. (Case 3-2; $-X^{\ell-1}\equiv T^{\alpha}\mod T^{\alpha}\Lambda_{+}$.) We have $z_{0}=X+Y=T^{-\alpha}X(T^{\alpha}+X^{\ell-1}).$ We put $\mathfrak{v}_{T}(z_{0})=\lambda>\mathfrak{v}_{T}(X)=\frac{\alpha}{\ell-1}.$ Using (10.5) we obtain $\lambda=\frac{\ell-1-\ell\alpha}{(\ell-1)(n-\ell+1)}.$ The condition $\lambda>\frac{\alpha}{\ell-1}$ becomes $\alpha<\frac{\ell-1}{n+1}$. We have $u_{i}=\mathfrak{v}_{T}(X)=\alpha/(\ell-1)$, $i\leq\ell$ and $u_{i}=\mathfrak{v}_{T}(Y)=\lambda$, $i>\ell$. We have $\mathfrak{PO}_{0}^{\text{\bf u}}=T^{\mathfrak{v}_{T}(X)}(y_{1}+\cdots+y_{\ell}+y_{1}\cdots y_{\ell})+T^{\mathfrak{v}_{T}(Y)}(y_{\ell+1}+\cdots+y_{n}+(y_{1}\cdots y_{n})^{-1}).$ The first term gives the leading term equation : $1+y_{1}\cdots\widehat{y}_{i}\cdots y_{\ell}=0,\qquad i=1,\cdots,\ell.$ It solution is $y_{1}=\cdots=y_{\ell}=\rho$ and $\rho^{\ell-1}=-1$. The second term of $\mathfrak{PO}_{0}^{\text{\bf u}}$ gives the leading term equation $1-y_{i}^{-1}\rho^{-\ell}(y_{\ell+1}\cdots y_{n})^{-1}=0,\qquad i=\ell+1,\cdots,n.$ Its solutions are $y_{\ell+1}=\cdots y_{n}=\theta,$ with $\rho^{\ell}\theta^{n-\ell+1}=1.$ Thus the leading term equation has $(\ell-1)(n+1-\ell)$ solutions all of which are strongly nondegenerate. We remark that $(\ell-1)(n+1-\ell)+(n+1)=\ell(n-\ell+2)$. Hence the number of solutions are always $\ell(n-\ell+2)$ which coincides with the Betti number of $X$. There are $2$ balanced fibers for $\alpha<(\ell-1)/(n+1)$ and $1$ balanced fiber for $\alpha\geq(\ell-1)/(n+1)$. By the above discussion and Theorem 6.1 (see also Remark 6.14), we can calculate $QH(X;\Lambda^{{\mathbb{Q}}})$ as follows : $QH(X;\Lambda^{{\mathbb{Q}}})=\begin{cases}\Lambda^{R_{1}}&\alpha>(\ell-1)/(n+1),\\\ \Lambda^{R_{2}}&\alpha=(\ell-1)/(n+1),\\\ \Lambda^{R_{3}}\times\Lambda^{R_{4}}&\alpha<(\ell-1)/(n+1),\end{cases}$ (10.7) where $\displaystyle R_{1}={\mathbb{Q}}[Z]/(Z^{\ell(n-\ell+2)}-1),$ $\displaystyle R_{2}={\mathbb{Q}}[Z]/(Z^{\ell}(Z+Z^{\ell})^{n-\ell+1}-1),$ $\displaystyle R_{3}={\mathbb{Q}}[Z]/(Z^{n+1}-1),$ $\displaystyle R_{4}={\mathbb{Q}}[Z_{1},Z_{2}]/(Z_{1}^{\ell-1}+1,Z_{1}^{\ell}Z_{2}^{n-\ell+1}-1).$ Here we assume (10.6) has only a simple root in case $\alpha=(\ell-1)/(n+1)$. We use the next lemma to show (10.7). ###### Lemma 10.11. Let $\mathfrak{x}=\sum\mathfrak{x}_{i}\text{\bf e}_{i}$ be a critical point of $\mathfrak{PO}^{\text{\bf u}}_{*}$. We assume $\mathfrak{y}_{i;0}=e^{\mathfrak{x}_{i;0}}$ (where $\mathfrak{x}_{i}=\mathfrak{x}_{i;0}\mod\Lambda_{+}$) is a strongly nondegenerate solution of leading term equation. We also assume that $\mathfrak{y}_{i;0}\in F$ where $F\subset{\mathbb{C}}$ is a field. Then $\mathfrak{y}_{i}=e^{\mathfrak{x}_{i}}$ is an element of $\Lambda_{0}^{F}$. We prove Lemma 10.11 later this section. We are now ready to give the proof of Theorem 10.4. ###### Proof of Theorem 10.4. We first consider the weakly nondegenerate case. Let $\mathfrak{m}$ be the multiplicity of $\mathfrak{y}^{0}$. We choose $\delta$ such that the ball $B_{\delta}(\mathfrak{y}^{0})$ centered at $\mathfrak{y}^{0}$ and of radius $\delta$ does not contain a solution of the leading term equation other than $\mathfrak{y}^{0}$. For $y\in\partial B_{\delta}(\mathfrak{y}^{0})$ we define $\nabla\overline{\mathfrak{PO}}(y)=\left(\sum_{j^{\prime}=1}^{a(k)}\frac{\partial Y(k,j^{\prime})}{\partial y_{k,j}}(y)\right)_{k=1,\cdots,K,\,j=1,\cdots,d(k)}\in{\mathbb{C}}^{n}.$ The map $y\mapsto\frac{\nabla\overline{\mathfrak{PO}}(y)}{\|\nabla\overline{\mathfrak{PO}}(y)\|}\in S^{2n-1}$ is well defined and of degree $\mathfrak{m}\neq 0$ by the definition of multiplicity. We define $\mathfrak{PO}^{u_{0}}_{*,k,{\mathcal{N}}}$ in the same way as (9.20). For $q\in{\mathbb{C}}$, we define $\mathfrak{PO}^{u_{0}}_{*,k,{\mathcal{N}}}(\cdots;q)$ by substituting $q$ to $T$. Then in the same way as the proof of Lemma 9.18 we can prove the following. ###### Lemma 10.12. There exists $\epsilon>0$ such that if $|q|<\epsilon$, the equation $0=\frac{\partial}{\partial y_{k,j}}\mathfrak{PO}^{u_{0}}_{*,k,{\mathcal{N}}}(\cdots;q)$ (10.8) has a solution in $B_{\delta}(\mathfrak{y}^{0})$. The sum of multiplicities of the solutions of $(\ref{eq:orderN})$ converging to $\mathfrak{y}_{k,j;0}$ is $\mathfrak{m}$. (10.8) is a polynomial equation. Hence multiplicity of its solution is well- defined in the standard sense of algebraic geometry. Now we assume that all the vertices of $P$ and $u_{0}$ are contained in ${\mathbb{Q}}^{n}$. (10.8) also depends polynomially on $q^{\prime}=T^{\prime}$, where $T^{\prime}=T^{1/\mathcal{C}!}$ for a sufficiently large integer $\mathcal{C}$. (We remark that $\mathcal{C}$ is determined by the denominators of the coordinates of the vertices of $P$ and of $u_{0}$. In particular it can be taken to be independent of ${\mathcal{N}}$.) We denote $y=(y_{1},\cdots,y_{n})$ and put $\mathfrak{X}=\\{(y,q^{\prime})\mid y\in B_{\delta}(\mathfrak{y}^{0}),\,q^{\prime}\,\text{with $|q^{\prime}|<\epsilon$ and $q=(q^{\prime})^{\mathcal{C}!}$ satisfying (\ref{eq:orderN})}\\}.$ We consider the projection $\pi_{q^{\prime}}:\mathfrak{X}\to\\{q^{\prime}\in{\mathbb{C}}\mid|q^{\prime}|<\epsilon\\}.$ (10.9) By choosing a sufficiently small $\epsilon>0$, we may assume that (10.9) is a local isomorphism on the punctured disc $\\{q^{\prime}\mid 0<|q^{\prime}|<\epsilon\\}$. Namely, $\pi_{q^{\prime}}$ is an étale covering over the punctured disc. We remark that for each $q^{\prime}$ the fiber consists of at most $\mathfrak{m}$ points, since the multiplicity of the leading term equation is $\mathfrak{m}$. We put $q^{\prime\prime}=(q^{\prime})^{1/\mathfrak{m}!}$. Then the pull-back $\pi_{q^{\prime\prime}}:\mathfrak{X}^{\prime}\to\\{q^{\prime\prime}\in{\mathbb{C}}\mid 0<|q^{\prime\prime}|<\epsilon\\}$ (10.10) of (10.9) becomes a trivial covering space. Namely there exists a single valued section of $\pi_{q^{\prime\prime}}$ on $\\{q^{\prime\prime}\mid 0<|q^{\prime\prime}|<\epsilon\\}$. It extends to a holomorphic section of $\\{q^{\prime\prime}\mid|q^{\prime\prime}|<\epsilon\\}$. In other words there exists a holomorphic family of solutions of (10.8) which is parameterized by $q^{\prime\prime}\in\\{q^{\prime\prime}\mid|q^{\prime\prime}|<\epsilon\\}$. We put $T^{\prime\prime}=(T^{\prime})^{1/\mathfrak{m}!}$. Then by taking the Taylor series of the $q^{\prime\prime}$-parameterized family of solutions at $0$, we obtain the following : ###### Lemma 10.13. If all the vertices of $P$ and $u_{0}$ are rational, then for each ${\mathcal{N}}$ there exists $\mathfrak{y}^{({\mathcal{N}})}=(\mathfrak{y}^{({\mathcal{N}})}_{k,j})$ $\mathfrak{y}^{({\mathcal{N}})}_{k,j}=\sum_{l=0}^{{\mathcal{N}}}\mathfrak{y}^{({\mathcal{N}})}_{k,j;l}(T^{\prime\prime})^{l}$ $(\mathfrak{y}^{({\mathcal{N}})}_{k,j;l}\in{\mathbb{C}})$ such that $\frac{\partial\mathfrak{PO}^{u_{0}}_{*}}{\partial y_{k,j}}(\mathfrak{y}^{({\mathcal{N}})}_{k,j})\equiv 0\mod(T^{\prime\prime})^{{\mathcal{N}}+1}$ (10.11) and that $\mathfrak{y}^{({\mathcal{N}})}_{k,j;0}\equiv\mathfrak{y}_{k,j;0}.$ We remark that Lemma 10.13 is sufficient for most of the applications. In fact it implies that $L(u_{0})$ is balanced if there exists a weakly nondegenerate solution of leading term equation at $u_{0}$. Hence we can apply Lemma 4.12. For completeness we prove the slightly stronger statement made for the weakly nondegenerate case in Theorem 10.4. The argument is similar to one in subsection 7.2.11 [FOOO3] (= subsection 30.11 [FOOO2]). For each ${\mathcal{N}}$, we denote by $\widetilde{\mathfrak{M}}((\mathfrak{y}_{k,j;0});{\mathcal{N}})$ the set of all $(\mathfrak{y}_{k,j;l}^{({\mathcal{N}})})_{k,j;l}\in{\mathbb{C}}^{n{\mathcal{N}}}$, where $k=1,\cdots,K$, $j=1,\cdots,a(k)$, $l=1,\cdots,{\mathcal{N}}$, such that $\mathfrak{y}^{({\mathcal{N}})}_{k,j}=\mathfrak{y}_{k,j;0}+\sum_{l=1}^{{\mathcal{N}}}\mathfrak{y}^{({\mathcal{N}})}_{k,j;l}(T^{\prime\prime})^{l}$ satisfies (10.11). By definition, $\widetilde{\mathfrak{M}}((\mathfrak{y}_{k,j;0});{\mathcal{N}})$ is the set of ${\mathbb{C}}$-valued points of certain complex affine algebraic variety (of finite dimension). Lemma 10.13 implies that $\widetilde{\mathfrak{M}}((\mathfrak{y}_{k,j;0});{\mathcal{N}})$ is nonempty. For ${\mathcal{N}}_{1}>{\mathcal{N}}_{2}$ there exists an obvious morphism $I_{{\mathcal{N}}_{1},{\mathcal{N}}_{2}}:\widetilde{\mathfrak{M}}((\mathfrak{y}_{k,j;0});{\mathcal{N}}_{1})\to\widetilde{\mathfrak{M}}((\mathfrak{y}_{k,j;0});{\mathcal{N}}_{2})$ of complex algebraic variety. To complete the proof of Theorem 10.4 in the weakly nondegenerate case, it suffices to show that the projective limit $\lim_{\longleftarrow}(\widetilde{\mathfrak{M}}(\mathfrak{y}_{k,j;0};{\mathcal{N}}))$ (10.12) is nonempty. ###### Lemma 10.14. $\bigcap_{{\mathcal{N}}>1}\text{\rm Im}I_{{\mathcal{N}},1}\neq\emptyset.$ ###### Proof. By a theorem of Chevalley (see Chapter 6 [Mat]), each $\text{\rm Im}I_{{\mathcal{N}},1}$ is a constructible set. It is nonempty and its dimension $\mathop{\kern 0.0pt{\rm dim}}\nolimits\text{\rm Im}I_{{\mathcal{N}},1}$ is nonincreasing as ${\mathcal{N}}\to\infty$. Therefore we may assume $\mathop{\kern 0.0pt{\rm dim}}\nolimits\text{\rm Im}I_{{\mathcal{N}},1}=d$ for ${\mathcal{N}}\geq{\mathcal{N}}_{1}$. We consider the number of $d$ dimensional irreducible components of $\text{\rm Im}I_{{\mathcal{N}},1}$. This number is nonincreasing for ${\mathcal{N}}\geq{\mathcal{N}}_{1}$. Therefore, there exists ${\mathcal{N}}_{2}$ such that for ${\mathcal{N}}\geq{\mathcal{N}}_{2}$ the number of $d$ dimensional irreducible components of $\text{\rm Im}I_{{\mathcal{N}},1}$ is independent of ${\mathcal{N}}$. It follows that there exists $X_{{\mathcal{N}}}$ a sequence of $d$ dimensional irreducible components of $\text{\rm Im}I_{{\mathcal{N}},1}$ such that $X_{{\mathcal{N}}+1}\subset X_{{\mathcal{N}}}$. Since $\mathop{\kern 0.0pt{\rm dim}}\nolimits(X_{{\mathcal{N}}}\setminus X_{{\mathcal{N}}+1})<d$, it follows from Baire’s category theorem that $\cap_{{\mathcal{N}}}X_{{\mathcal{N}}}\neq\emptyset$. Hence the lemma. ∎ ###### Lemma 10.15. There exists a sequence $(\mathfrak{y}_{k,j;l}^{(n)})_{k,j;l}$ $n=1,2,3,\cdots\,m$ such that $I_{n,n-1}((\mathfrak{y}_{k,j;l}^{(n)})_{k,j;l}=(\mathfrak{y}_{k,j;l}^{(n-1)})_{k,j;l})$ for $n=2,\cdots,m$ and that $(\mathfrak{y}_{k,j;l}^{(m)})_{k,j;l}\in\bigcap_{{\mathcal{N}}>m}\text{\rm Im}I_{{\mathcal{N}},m}.$ ###### Proof. The proof is by induction on $m$. The case $m=1$ is Lemma 10.14. Each of the inductive step is similar to the proof of Lemma 10.14 and so it omitted. ∎ Lemma 10.15 implies that the projective limit (10.12) is nonempty. The proof of weakly nondegenerate case of Theorem 10.4 is complete. We next consider the strongly nondegenerate case. We prove the following lemma by induction on ${\mathcal{N}}$. Let $G$ be a submonoid of $({\mathbb{R}}_{\geq 0},+)$ generated by the numbers appearing in the exponent of (9.18). Namely it is generated by $\displaystyle S_{k^{\prime}}-S_{k}\,\,(k^{\prime}>k),$ $\displaystyle\ell(u_{0})+\rho- S_{k}\,\,((\ell,\rho)\in\mathfrak{I}_{k^{\prime}},\,\,k^{\prime}\geq k),$ (10.13) $\displaystyle\ell(u_{0})-S_{k}\,\,(\ell\in\mathfrak{I}).$ We define $0<\lambda_{1}<\lambda_{2}<\cdots$ by $\\{\lambda_{i}\mid i=1,2,\dots\\}=G.$ ###### Lemma 10.16. We assume that $\mathfrak{y}^{0}=(\mathfrak{y}_{k,j;0})_{k=1,\cdots,K,\,j=1,\cdots,d(k)}$ is a strongly nondegenerate solution of the leading term equation. Then, there exists $\mathfrak{y}^{({\mathcal{N}})}_{k,j}=\mathfrak{y}_{k,j;0}+\sum_{l=1}^{{\mathcal{N}}}\mathfrak{y}_{k,j;l}T^{\lambda_{l}}$ such that $\sum_{j^{\prime}=1}^{a(k)}\frac{\partial Y(k,j^{\prime})}{\partial y_{k,j}}(\mathfrak{y}^{({\mathcal{N}})}_{k,1},\cdots,\mathfrak{y}^{({\mathcal{N}})}_{K,d(K)})\equiv 0\mod T^{\lambda_{{\mathcal{N}}+1}}.$ (10.14) Moreover we may choose $\mathfrak{y}^{({\mathcal{N}})}_{k,j}$ so that $\mathfrak{y}^{({\mathcal{N}})}_{k,j}\equiv\mathfrak{y}^{({\mathcal{N}}+1)}_{k,j}\mod T^{\lambda_{{\mathcal{N}}+1}}.$ ###### Proof. The proof is by induction on ${\mathcal{N}}$. There is nothing to show in the case ${\mathcal{N}}=0$. Assume we have proved the lemma up to ${\mathcal{N}}-1$. Then we have $\sum_{j^{\prime}=1}^{a(k)}\frac{\partial Y(k,j^{\prime})}{\partial y_{k,j}}(\mathfrak{y}^{({\mathcal{N}}-1)}_{k,1},\cdots,\mathfrak{y}^{({\mathcal{N}}-1)}_{K,d(K)})\equiv c_{k,j,M}T^{{\mathcal{N}}}\mod T^{\lambda_{{\mathcal{N}}+1}}.$ Consider $\mathfrak{y}^{({\mathcal{N}})}_{k,j}$ of the form $\mathfrak{y}^{({\mathcal{N}})}_{k,j}=\mathfrak{y}^{({\mathcal{N}}-1)}_{k,j}+\Delta_{k,j,{\mathcal{N}}}T^{\lambda_{{\mathcal{N}}}}$ for some $\Delta_{k,j,{\mathcal{N}}}$. Then we can write $\displaystyle\sum_{j^{\prime}=1}^{a(k)}$ $\displaystyle\frac{\partial Y(k,j^{\prime})}{\partial y_{k,j}}(\mathfrak{y}^{(M-1)}_{k,1}+\Delta_{k,1,{\mathcal{N}}}T^{\lambda_{{\mathcal{N}}}},\cdots,\mathfrak{y}^{({\mathcal{N}}-1)}_{K,d(K)}+\Delta_{K,d(K),{\mathcal{N}}}T^{\lambda_{{\mathcal{N}}}})$ (10.15) $\displaystyle\equiv\left(c_{k,j,{\mathcal{N}}}+\sum_{j^{\prime},j^{\prime\prime}=1}^{a(k)}\frac{\partial^{2}Y(k,j^{\prime})}{\partial y_{k,j}\partial y_{k,j^{\prime\prime}}}\Delta_{k,j^{\prime\prime},{\mathcal{N}}}\right)T^{\lambda_{\mathcal{N}}}\mod T^{\lambda_{{\mathcal{N}}+1}}.$ Since $\mathfrak{y}^{0}=(\mathfrak{y}_{k,j;0})_{k=1,\cdots,K,\,j=1,\cdots,d(k)}$ is strongly nondegenerate, we can find $\Delta_{k,j^{\prime\prime},{\mathcal{N}}}\in{\mathbb{C}}$ so that the right hand side become zero module $T^{\lambda_{{\mathcal{N}}+1}}$. The proof of Lemma 10.16 is complete. ∎ By Lemma 10.16, the limit $\lim_{{\mathcal{N}}\to\infty}\mathfrak{y}^{({\mathcal{N}})}_{k,j}$ exists. We set $\mathfrak{y}_{k,j}:=\lim_{{\mathcal{N}}\to\infty}\mathfrak{y}^{({\mathcal{N}})}_{k,j}.$ This is the required solution of (10.4). The proof of Theorem 10.4 is complete ∎ ###### Proof of Lemma 10.11. We put $\mathfrak{y}_{k,j}=\mathfrak{y}_{k,j;0}+\sum_{l=1}^{\infty}\mathfrak{y}_{k,j;l}T^{\lambda_{l}}.$ By assumption $\mathfrak{y}_{k,j;0}\in F$. We remark that (10.15) gives a linear equation which determines $\mathfrak{y}_{k,j;l}$ inductively on $l$. We use it to show $\mathfrak{y}_{k,j;l}\in F$ inductively on $l$. ∎ We next give an example where weakly nondegeneracy condition is not satisfied. ###### Example 10.17. Consider the 2-point blow up $X(\alpha,\beta)$ of ${\mathbb{C}}P^{2}$ with its moment polytope given by $P=\\{(u_{1},u_{2})\mid 0\leq u_{1}\leq 1,\,\,0\leq u_{2}\leq 1-\alpha,\,\,\beta\leq u_{1}+u_{2}\leq 1\\}.$ We consider the case when $1-\alpha$ is sufficiently small. The potential function is $\mathfrak{PO}=T^{u_{1}}y_{1}+T^{u_{2}}y_{2}+T^{1-\alpha- u_{2}}y_{2}^{-1}+T^{1-u_{1}-u_{2}}y_{1}^{-1}y_{2}^{-1}+T^{u_{1}+u_{2}-\beta}y_{1}y_{2}.$ (We remark that $X$ is Fano.) We fix $\alpha$ and move $\beta$ starting from zero. When $\beta$ is small compared to $1-\alpha$, there are two balanced fibers. One is located at $((1+\alpha)/4,(1-\alpha)/2)$. This corresponds to the location of the balanced fiber of the one point blow up, which is nothing but the case $\beta=0$. The other appears near the origin and is $(\beta,\beta)$. The leading term equation at the first point is $1-y_{2}^{-2}=0,\quad 1-y_{1}^{-2}y_{2}^{-1}=0.$ The solutions are $(y_{1},y_{2})=(\pm 1,1),(\pm\sqrt{-1},-1)$, all of which are strongly nondegenerate. The leading term equation at the second point is $1+y_{1}=1+y_{2}=0.$ $(-1,-1)$ is the nondegenerate solution. Thus we have $5$ solutions. The situation jumps when $\beta=(1-\alpha)/2$. Denote $\beta_{0}=(1-\alpha)/2$ for the simplicity of notation. In that case the potential function at $(\beta_{0},\beta_{0})$ becomes $T^{\beta_{0}}(y_{1}+y_{2}+y_{1}y_{2}+y_{2}^{-1})+T^{1-2\beta_{0}}y_{1}^{-1}y_{2}^{-1}.$ The leading term equation is $1+y_{2}=0,\quad 1+y_{1}-y_{2}^{-2}=0.$ Its solution is $(0,-1)$. Since $y_{1}=0$ it follows that there is no solution in $(\Lambda_{0}\setminus\Lambda_{+})^{2}$. Hence there is no weak bounding cochain $\mathfrak{x}$ for which the Floer cohomology $HF((L(\beta_{0},\beta_{0}),\mathfrak{x}),(L(\beta_{0},\beta_{0}),\mathfrak{x});\Lambda)$ is nontrivial. In other words, the fiber $L(\beta_{0},\beta_{0})$ in $X(\alpha,\beta_{0})$ is not strongly balanced. On the other hand, this fiber $L(\beta_{0},\beta_{0})$ in $X(\alpha,\beta_{0})$ is balanced because by choosing $\beta$ arbitrarily close to $\beta_{0}$ and $\beta<\beta_{0}$, we can approximate it by the fibers $L(\beta,\beta)\subset X(\alpha,\beta)$ for which the Floer cohomology $HF((L(\beta,\beta),\mathfrak{x}),(L(\beta,\beta),\mathfrak{x}))$ is non- trivial. In particular $L(\beta_{0},\beta_{0})$ in $X(\alpha,\beta_{0})$ is not displaceable. We can also verify that $\overline{\mathfrak{E}}(\beta_{0},\beta_{0})=\infty\quad\mbox{in }\,X(\alpha,\beta_{0})$ (10.16) while $\mathfrak{E}(\beta_{0},\beta_{0})=\beta_{0}$ in $X(\alpha_{0},\beta_{0})$. Now we examine where the missing solutions at $\beta=\beta_{0}$ have gone. We consider $(u_{1},(1-\alpha)/2)$ where $\beta_{0}=(1-\alpha)/2<u_{1}<(1+\alpha)/4$. The potential function is $T^{\beta_{0}}(y_{2}+y_{2}^{-1})+T^{\beta_{0}+\lambda_{1}}(y_{1}+y_{1}y_{2})+T^{\beta_{0}+\lambda_{2}}y_{1}^{-1}y_{2}^{-1}.$ (10.17) Here $\lambda_{1}=u_{1}-\beta_{0}<\lambda_{2}=(1+\alpha)/2-u_{1}-\beta_{0}.$ The leading term equation is $1-y_{2}^{-2}=0,\quad 1+y_{2}=0.$ (10.18) The solution is $y_{2}=-1$ and $y_{1}$ is arbitrary. Thus there are infinitely many solutions of the leading term equation. Therefore these solutions of (10.18) are not weakly nondegenerate. So we need to study the critical point of (10.17) more carefully. The condition that $(y_{1},y_{2})$ is a critical point of (10.17) is written as $\left\\{\begin{aligned} &1-y_{2}^{-2}+T^{\lambda_{1}}y_{1}-T^{\lambda_{2}}y_{1}^{-1}y_{2}^{-2}=0\\\ &1+y_{2}-T^{\lambda_{2}-\lambda_{1}}y_{1}^{-2}y_{2}^{-1}=0.\end{aligned}\right.$ (10.19) The leading order term of $y_{2}$ should be $-1$. We need to study also the second order term. We can write $y_{2}=-1+cT^{\mu},\quad y_{1}=d,$ where $c,d\in\Lambda_{0}\setminus\Lambda_{+}$. Then we have $\displaystyle-2cT^{\mu}+dT^{\lambda_{1}}\equiv 0\mod T^{\min\\{\mu,\lambda_{1}\\}}\Lambda_{+},$ (10.20) $\displaystyle cT^{\mu}+d^{-2}T^{\lambda_{2}-\lambda_{1}}\equiv 0\mod T^{\min\\{\mu,\lambda_{2}-\lambda_{1}\\}}\Lambda_{+}.$ (10.21) (10.20) implies $\mu=\lambda_{1}$. (10.21) then implies $\lambda_{2}-\lambda_{1}=\lambda_{1}$. It implies $u_{1}=1/3$. Furthermore $c^{3}\equiv-1/4\mod\Lambda_{+},\quad d\equiv 2c\mod\Lambda_{+}.$ (10.22) Since the three solutions of the ${\mathbb{C}}$-reduction of (10.22) are all simple, we can show, by the same way as that of the proof of Theorem 10.4, that all solutions correspond to solutions of the equation (10.19). Therefore, $L(1/3,\beta_{0})$ is a strongly balanced fiber. We remark that solutions of the leading term equation (10.18) do not lift to solutions of (10.19) unless $u_{1}=1/3$ and $y_{1}=-1$. This shows that the weakly nondegeneracy assumption in Theorem 10.4 is essential. We remark that at $((1+\alpha)/4,(1-\alpha)/2)$ the leading term equation becomes $1-y_{2}^{-2}=0,\quad 1+y_{2}-y_{1}^{-2}y_{2}^{-1}=0.$ Its solutions in $({\mathbb{C}}\setminus\\{0\\})^{2}$ are $(\pm 1/\sqrt{2},1)$. The number of solutions jumps from 4 to 2 here. $2+3=5$. So this is consistent with Theorem 1.3. In summary for the case of $(\alpha,\beta_{0})$ with $\beta_{0}=(1-\alpha)/2$, there are 3 balanced fibers $(1/3,\beta_{0})$, $((1+\alpha)/4,\beta_{0})$ and $(\beta_{0},\beta_{0})$. The first two of them are strongly balanced and the last is not strongly balanced. The balanced fiber $L(1/3,\beta_{0})\subset X(\alpha,\beta_{0})$ disappears as we deform $X(\alpha,\beta_{0})$ to $X(\alpha,\beta)$ as $\beta_{0}$ moves to nearby $\beta$. To see this let us take $\beta$ which is slightly bigger than $\beta_{0}=(1-\alpha)/2$. Then $((\alpha+\beta)/2,(1-\alpha)/2)$, and $(1-\alpha-\beta,\beta)$ are the balanced fibers. The leading term equation at the first point is $1-y_{2}^{-2}=0,\quad y_{2}-y_{1}^{-2}y_{2}^{-1}=0.$ Hence there are 4 solutions $(\pm 1,\pm 1)$. The leading term equation at the second point is $1+y_{2}=0,\quad y_{1}-y_{2}^{-2}=0.$ Hence the solution is $(1,-1)$. Total number is again 5. We remark that the ${\mathbb{Q}}$-structure of quantum cohomology also jumps at $\beta=(1-\alpha)/2$. Namely $QH(X(\alpha,\beta);\Lambda^{{\mathbb{Q}}})=\begin{cases}\Lambda^{{\mathbb{Q}}(\sqrt{-1})}\times(\Lambda^{{\mathbb{Q}}})^{3}&\text{$\beta$ is slightly smaller than $(1-\alpha)/2$},\\\ \Lambda^{{\mathbb{Q}}(\sqrt{2})}\times\Lambda^{{\mathbb{Q}}((-2)^{1/3})}&\text{$\beta=(1-\alpha)/2$},\\\ (\Lambda^{{\mathbb{Q}}})^{5}&\text{$\beta$ is slightly larger than $(1-\alpha)/2$}.\end{cases}$ ###### Remark 10.18. In a sequel of this series of papers, we will prove that $L(u_{1},(1-\alpha)/2))$ is not displacable for any $u_{1}\in((1-\alpha)/2,1/3)\cup(1/3,(1+\alpha)/4)$, in the case $\beta=\beta_{0}$. We will use the bulk deformation introduced by [FOOO3] section 3.8 (= [FOOO2] section 13) to prove it. The next example shows that Theorems 1.3, 1.4 and 10.4 can not be generalized to the case of a positive characteristic. ###### Example 10.19. Consider the $2$-point blow up $X$ of ${\mathbb{C}}P^{2}$ with moment polytope $P=\\{(u_{1},u_{2})\mid 0\leq u_{i}\leq 1-\epsilon,\sum u_{i}\leq 1\\}.$ Since $X$ is monotone for $\epsilon=1/3$, it follows that $X$ is Fano. We will assume $\epsilon>0$ is sufficiently small. Then the fiber at $u_{0}=(1/3,1/3)$ is balanced. Now we consider the Novikov ring $\Lambda^{F}$ with $F=\mathbb{F}_{3}$ a field of characteristic $3$. We will prove that there exists no element $\mathfrak{x}\in H(L(u);\Lambda_{0}^{F})$ such that $HF((L(u_{0}),\mathfrak{x}),(L(u_{0}),\mathfrak{x});\Lambda^{F})\neq 0$. The potential function at $u_{0}$ is $\mathfrak{PO}^{u_{0}}=T^{1/3}(y_{1}+y_{2}+1/(y_{1}y_{2}))+T^{2/3-\epsilon}(y_{1}^{-1}+y_{2}^{-1}).$ Therefore the critical point equation is given by $1-1/(y_{i}y_{1}y_{2})-ty_{i}^{-2}=0\quad i=1,2$ (10.23) where $t=T^{1/3-\epsilon}$. From this it follows that $y_{i}\equiv 1\mod\Lambda_{+}$. In fact the leading term equation is $a_{i}a_{1}a_{2}=1$ for $i=1,2$ which is reduced to $a_{1}=a_{2}=a,\quad a^{3}=1.$ Obviously this equation has the unique solution $a_{1}=a_{2}=1$ in $\mathbb{F}_{3}$. Going back to the study of solutions of the critical point equation (10.23), we first prove $y_{1}=y_{2}$. We put $z_{i}=y_{i}^{-1}$. We assume $z_{i}-z_{j}\neq 0$ and put $z_{i}-z_{j}\equiv t^{\lambda}c\mod t^{\lambda}\Lambda_{+}$ with $c\in\mathbb{F}_{3}\setminus\\{0\\}$. Then by (10.23) we have $(z_{i}-z_{j})z_{1}z_{2}+t(z_{i}-z_{j})(z_{i}+z_{j})=0.$ This is a contradiction since the left hand side is congruent to $ct^{\lambda}$ modulo $t^{\lambda}\Lambda_{+}^{\mathbb{F}_{3}}$. We now put $x=y_{i}$ and obtain $x^{3}-tx-1=0.$ (10.24) We prove: ###### Lemma 10.20. $(\ref{eq:x})$ has no solution in $\Lambda_{0}^{\mathbb{F}_{3}}$. ###### Proof. We put $x=1+t^{1/3}x^{\prime}$ and obtain $(x^{\prime})^{3}-t^{1/3}x^{\prime}-1=0.$ This equation resembles (10.24) except that $t$ is replaced by $t^{1/3}$. We now put $x_{N}\equiv 1+\sum_{k=1}^{N}t^{\sum_{i=1}^{k}3^{-i}}.$ Then $x_{N}^{3}=1+tx_{N-1}.$ Therefore $(x_{N})^{3}-tx_{N}-1=-t^{1+\sum_{i=1}^{N}3^{-i}}.$ Namely $x_{N}$ is a solution of (10.24) modulo $t^{1+\sum_{i=1}^{N}3^{-i}}$. It is easy to see that there are no other solution of (10.24) modulo $t^{1+\sum_{i=1}^{N}3^{-i}}$. However since $1+\sum_{i=1}^{\infty}3^{-i}=\frac{3}{2}<\infty,$ it follows that $\lim_{N\to\infty}x_{N}$ does not converge in $\Lambda_{0}^{\mathbb{F}_{3}}$. Thus there is no solution of (10.24) over a field of characteristic $3$. ∎ Lemma 10.20 implies that the field of fraction of the Puiseux series ring with coefficients in an algebraically closed field of _positive_ characteristic is not algebraically closed. It is well known that this phenomenon does not occur in the case of characteristic zero. See, for example, Corollary 13.15 [Ei]. Since we could not find a proof of a similar result for universal Novikov ring in the literature, we will provide its proof in the appendix for completeness. (We used it in the proof of Theorem 1.12 in section 7.) ## 11\. Calculation of potential function In this section, we prove Theorems 4.5 and 4.6. We begin with a review of [CO]. Let $\pi:X\to P$ be the moment map and $\partial P=\bigcup_{i=1}^{m}\partial_{i}P$ be the decomposition of the boundary of $P$ into $n-1$ dimensional faces. Let $\beta_{i}\in H_{2}(X,L(u);{\mathbb{Z}})$ be the elements such that $\beta_{i}\cap[\pi^{-1}(\partial P_{j})]=\begin{cases}1&\text{if $i=j$,}\\\ 0&\text{if $i\neq j$.}\end{cases}$ The Maslov index $\mu(\beta_{i})$ is 2. (Theorem 5.1 [CO].) Let $\beta\in\pi_{2}(X,L(u))$ and $\mathcal{M}_{k+1}^{\text{\rm main}}(L(u),\beta)$ be the moduli space of stable maps from bordered Riemann surfaces of genus zero with $k+1$ boundary marked points in homology class $\beta$. (See [FOOO1] section 3 = [FOOO3] subsection 2.1.2. We require the boundary marked points to respect the cyclic order of $S^{1}=\partial D^{2}$. (In other words we consider the main component in the sense of [FOOO1] section 3.)) Let $\mathcal{M}^{\text{\rm main},\text{\rm reg}}_{k+1}(L(u),\beta)$ be its subset consisting of all maps from a disc. (Namely the stable map without disc or sphere bubble.) The next theorem easily follows from the results of [CO]. In (3) Theorem 11.1 we use the spin structure of $L(u)$ which is induced by the diffeomorphism of $L(u)\cong T^{n}$ by the $T^{n}$ action and the standard trivialization of the tangent bundle of $T^{n}$. ###### Theorem 11.1. 1. (1) If $\mu(\beta)<0$, or $\mu(\beta)=0$, $\beta\neq 0$, then $\mathcal{M}^{\text{\rm main},\text{\rm reg}}_{k+1}(L(u),\beta)$ is empty. 2. (2) If $\mu(\beta)=2$, $\beta\neq\beta_{1},\cdots,\beta_{m}$, then $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{k+1}(L(u),\beta)$ is empty. 3. (3) For $i=1,\cdots,m$, we have $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{1}(L(u),\beta_{i})=\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i}).$ (11.1) Moreover $\mathcal{M}_{1}^{\text{\rm main}}(L(u),\beta_{i})$ is Fredholm regular. Furthermore the evaluation map $ev:\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i})\to L(u)$ is an orientation preserving diffeomorphism. 4. (4) For any $\beta$, the moduli space $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{1}(L(u),\beta)$ is Fredholm regular. Moreover $ev:\mathcal{M}^{\text{\rm main,reg}}_{1}(L(u),\beta)\to L(u)$ is a submersion. 5. (5) If $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)$ is not empty then there exists $k_{i}\in{\mathbb{Z}}_{\geq 0}$ and $\alpha_{j}\in H_{2}(X;{\mathbb{Z}})$ such that $\beta=\sum_{i}k_{i}\beta_{i}+\sum_{j}\alpha_{j}$ and $\alpha_{j}$ is realized by holomorphic sphere. There is at least one nonzero $k_{i}$. ###### Proof. For reader’s convenience and completeness, we explain how to deduce Theorem 11.1 from the results in [CO]. By Theorems 5.5 and 6.1 [CO], $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{k+1}(L(u),\beta)$ is Fredholm regular for any $\beta$. Since the complex structure is invariant under the $T^{n}$ action and $L(u)$ is $T^{n}$ invariant, it follows that $T^{n}$ acts on $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{k+1}(L(u),\beta)$ and $ev:\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{k+1}(L(u),\beta)\to L(u)$ is $T^{n}$ equivariant. Since the $T^{n}$ action on $L(u)$ is free and transitive, it follows that $ev$ is a submersion if $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{k+1}(L(u),\beta)$ is nonempty. (4) follows. We assume $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{k+1}(L(u),\beta)$ is nonempty. Since $ev$ is a submersion it follows that $n=\mathop{\kern 0.0pt{\rm dim}}\nolimits L(u)\leq\mathop{\kern 0.0pt{\rm dim}}\nolimits\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{k+1}(L(u),\beta)=n+\mu(\beta)-2$ if $\beta\neq 0$. Therefore $\mu(\beta)\geq 2$. (1) follows. We next assume $\mu(\beta)=2$, and $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{k+1}(L(u),\beta)$ is nonempty. Then by Theorem 5.3 [CO], we find $\beta=\beta_{i}$ for some $i$. (2) follows. We next prove (5). It suffices to consider $[f]\in\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)\setminus\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{1}(L(u),\beta).$ We decompose the domain of $u$ into irreducible components and restrict $f$ there. Let $f_{j}:D^{2}\to M$ and $g_{k}:S^{2}\to M$ be the restriction of $f$ to disc or sphere components respectively. We have $\beta=\sum[f_{j}]+\sum[g_{k}].$ Theorem 5.3 [CO] implies that each of $f_{j}$ is homologous to the sum of the element of $\beta_{i}$. It implies (5). We finally prove (3). The fact that $ev$ is a diffeomorphism for $\beta=\beta_{i}$, follows dirctly from Theorem 5.3 [CO]. We next prove that $ev$ is orientation preserving. Since $L(u),u\in\text{\rm Int}\,P,$ is a principal homogeneous space of $T^{n}$, the tangent bundle $TL(u)$ is trivialized once we fix an isomorphism $T^{n}\equiv S^{1}\times\dots\times S^{1}$. Using the orientation and the spin structure on $L(u)$ induced by such a trivialization, we orient the moduli space $\mathcal{M}_{1}(\beta)$ of holomorprhic discs. If we change the identification $T^{n}\equiv S^{1}\times\dots\times S^{1}$ by an orientation preserving, (resp. reversing), isomorphism, the corresponding orientations on $L(u)$ and $\mathcal{M}_{1}(\beta)$ is preserved, (resp. reversed). Therefore whether $ev:\mathcal{M}(\beta_{i})\to L(u)$ is orientation preserving or not does not depend on the choice of the identification $T^{n}$ and $S^{1}\times\dots\times S^{1}$. For each $i=1,\dots,m$, we can find an automorphism $\phi$ of $({\mathbb{C}}^{*})^{n}$ and a biholomorphic map $f:X\setminus\cup_{j\neq i}\pi^{-1}(\partial_{j}P)\to{\mathbb{C}}\times({\mathbb{C}}^{*})^{n-1}$ such that 1. (1) $f$ is $\phi$-equivariant, 2. (2) $f(L(u))=L_{\text{\rm std}}$, where $L_{\text{\rm std}}=\\{(w_{1},\dots,w_{n})\in({\mathbb{C}})^{n}||w_{1}|=\dots=|w_{n}|=1\\}.$ Under this identification, $\mathcal{M}_{1}(\beta_{i})$ is identified with the space of holomorphic discs $z\in D^{2}\mapsto(\zeta\cdot z,w_{2},\dots,w_{n})\in{\mathbb{C}}\times({\mathbb{C}}^{*})^{n-1},\quad\zeta\in S^{1}\subset{\mathbb{C}}^{*},$ where $w_{k}\in{\mathbb{C}},k=2,\dots,n$ with $|w_{k}|=1$. Therefore it is enough to check the statement that $ev$ is orientation preserving in a single example. Cho [Cho] proved it in the case of the Clifford torus in ${\mathbb{C}}P^{n}$, hence the proof. To prove (11.1) and complete the proof of Theorem 11.1, it remains to prove $\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{1}(L(u),\beta_{i_{0}})=\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i_{0}})$. (Here $i_{0}\in\\{1,\cdots,m\\}$.) Let $[f]\in\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i_{0}})$. We take $k_{i}$ and $\alpha_{j}$ as in (5). (Here $\beta=\beta_{i_{0}}$). We have $\partial\beta_{i_{0}}=\sum_{i}k_{i}\partial\beta_{i}.$ Using the convexity of $P$, (5) and $k_{i}\geq 0$, we show the inequality $\beta_{i_{0}}\cap\omega\leq\sum_{i}k_{i}\beta_{i}\cap\omega$ (11.2) holds and that the equality holds only if $k_{i}=0$ ($i\neq i_{0}$), $k_{i_{0}}=1$, as follows : By (5) we have $\ell_{i_{0}}=\sum_{i=1}^{m}k_{i}\ell_{i}+c$ where $c$ is a constant. Since $k_{i}\geq 0$ and $\ell_{i_{0}}(u^{\prime})=0$ for $u^{\prime}\in\partial_{i_{0}}P$, it follows that $c\leq 0$. (Note $\ell_{i}\geq 0$ on $P$.) Since $\beta_{i}\cap\omega=\ell_{i}(u)$, we have the inequality (11.2). Let us assume that the equality holds. If there exists $i\neq j$ with $k_{i},k_{j}>0$ then $\partial_{i_{0}}\,P=\\{u^{\prime}\in P\mid\ell_{i_{0}}(u^{\prime})=0\\}\subseteq\\{u^{\prime}\in P\mid\ell_{i}(u^{\prime})=\ell_{j}(u^{\prime})=0\\}\subseteq\partial_{i}P\cap\partial_{j}P.$ This is a contradiction since $\partial_{i_{0}}P$ is codimension 1. Therefore there is only one nonzero $k_{i}$. It is easy to see that $i=i_{0}$ and $k_{i_{0}}=1$. On the other hand since $\alpha_{j}\cap\omega>0$ it follows that $\beta_{i_{0}}\cap\omega\geq\sum_{i}k_{i}\beta_{i}\cap\omega.$ Therefore there is no sphere bubble (that is $\alpha_{j}$). Moreover the equality holds in (11.2). Hence the domain of our stable map is irreducible. Namely $\mathcal{M}^{\text{\rm main},\text{\rm reg}}_{1}(L(u),\beta_{i_{0}})=\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i_{0}}).$ The proof of Theorem 11.1 is now complete. ∎ Next we discuss one delicate point to apply Theorem 11.1 to the proofs of Theorems 4.5 and 4.6. (This point was already mentioned in section 16 [CO].) Let us consider the case where there exists a holomorphic sphere $g:S^{2}\to X$ with $c_{1}(X)\cap g_{*}[S^{2}]<0.$ We assume moreover that there exists a holomorphic disc $f:(D^{2},\partial D^{2})\to(X,L(u))$ such that $f(0)=g(1).$ We glue $D^{2}$ and $S^{2}$ at $0\in D^{2}$ and $1\in S^{2}$ to obtain $\Sigma$. $f$ and $g$ induce a stable map $h:(\Sigma,\partial\Sigma)\to(X,L(u))$. In general $h$ will not be Fredholm regular since $g$ may not be Fredholm regular or the evaluation is not transversal at the interior nodes. In other words, elements of $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)\setminus\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{1}(L(u),\beta)$ may not be Fredholm regular in general. Moreover replacing $g$ by its multiple cover, we obtain an element of $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)\setminus\mathcal{M}^{{\text{\rm main}},\text{\rm reg}}_{1}(L(u),\beta)$ such that $\mu(\beta)$ is negative. Theorem 11.1 says that all the holomorphic disc without any bubble are Fredholm regular. However we can not expect that all stable maps in $\mathcal{M}^{{\text{\rm main}}}_{1}(L(u),\beta)$ are Fredholm regular. In order to prove Theorem 4.6, we need to find appropriate perturbations of those stable maps. For this purpose we use the $T^{n}$ action and proceed as follows. (We remark that many of the arguments below are much simplified in the Fano case, where there exists no holomorphic sphere $g$ with $c_{1}(M)\cap g_{*}[S^{2}]\leq 0.$ ) We equip each of $\mathcal{M}_{1}(L(u),\beta)$ with Kuranishi structure. (See [FO] for the general theory of Kuranishi structure and section 17-18 [FOOO1] (or section 7.1 [FOOO3] = section 29 [FOOO2]) for its construction in the context we currently deal with.) We may construct Kuranishi neighborhoods and obstruction bundles that carry $T^{n}$ actions induced by the $T^{n}$ action on $X$, and choose $T^{n}$-equivariant Kuranishi maps. (See Definition 15.4.) We note that the evaluation map $ev:\mathcal{M}_{1}(L(u),\beta)\to L(u)$ is $T^{n}$-equivariant. We use the fact that the complex structure of $X$ is $T^{n}$-invariant and $L(u)$ is a free $T^{n}$-orbit to find such a Kuranishi structure. (See Proposition 15.7 for the detail.) We remark that the $T^{n}$ action on the Kuranishi neighborhood is free since the $T^{n}$ action on $L(u)$ is free and $ev$ is $T^{n}$ equivariant. We take a perturbation (that is, a multisection) of the Kuranishi map that is $T^{n}$ equivariant. We can find such a multisection which is also transversal to $0$ as follows : Since the $T^{n}$ action is free, we can take the quotient of Kuranishi neighborhood, obstruction bundle etc. to obtain a space with Kuranishi structure. Then we take a transversal multisection of the quotient Kuranishi structure and lift it to a multisection of the Kuranishi neighborhood of $\mathcal{M}_{1}(L(u),\beta)$. (See Corollary 15.15 for detail.) Let $\mathfrak{s}_{\beta}$ be such a multisection and let $\mathcal{M}_{1}(L(u),\beta)^{\mathfrak{s}_{\beta}}$ be its zero set. We remark that the evaluation map $ev:\mathcal{M}_{1}(L(u),\beta)^{\mathfrak{s}_{\beta}}\to L(u)$ (11.3) is a submersion. This follows from the $T^{n}$ equivariance. This makes our construction of system of multisections much simpler than the general one in section 7.2 [FOOO3] (= section 30 [FOOO2]) since the fiber product appearing in the inductive construction is automatically transversal. (See subsection 7.2.2 [FOOO3] = section 30.2 [FOOO2]) for the reason why this is crucial.) More precisely we prove the following Lemma 11.2. Let $\mathfrak{forget}_{0}:\mathcal{M}^{\text{\rm main}}_{k+1}(L(u),\beta)\to\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)$ (11.4) be the forgetful map which forgets the first, …, $k$-th marked points. (In other words, only the $0$-th marked point remains.) We can construct our Kuranishi structure so that it is compatible with $\mathfrak{forget}_{0}$ in the same sense as Lemma 7.3.8 [FOOO3] (= Lemma 31.8 [FOOO2]). ###### Lemma 11.2. For each given $E>0$, there exists a system of multisections $\mathfrak{s}_{\beta,k+1}$ on $\mathcal{M}^{\text{\rm main}}_{k+1}(L(u),\beta)$ for $\beta\cap\omega<E$ with the following properties : 1. (1) They are transversal to $0$. 2. (2) They are invariant under the $T^{n}$ action. 3. (3) The multisection $\mathfrak{s}_{\beta,k+1}$ is the pull-back of the multisection $\mathfrak{s}_{\beta,1}$ by the forgetful map $(\ref{37.183})$. 4. (4) The restriction of $\mathfrak{s}_{\beta,1}$ to the boundary of $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)$ is the fiber product of the multisections $\mathfrak{s}_{\beta^{\prime},k^{\prime}}$ with respect to the identification of the boundary $\partial{\mathcal{M}}_{1}^{\text{\rm main}}(L(u),\beta)=\bigcup_{\beta_{1}+\beta_{2}=\beta}{\mathcal{M}}_{1}^{\text{\rm main}}(L(u),\beta_{1}){}_{ev_{0}}\times_{ev_{1}}{\mathcal{M}}_{2}^{\text{\rm main}}(L(u),\beta_{2}).$ 5. (5) We do not perturb $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i})$ for $i=1,\cdots,m$. ###### Proof. We construct multisections inductively over $\omega\cap\beta$. Since (2) implies that fiber products of the perturbed moduli spaces which we have already constructed in the earlier stage of induction are automatically transversal, we can extend them so that (1), (2), (3), (4) are satisfied by the method we already explained above. We recall from Theorem 11.1 (3) that $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i})=\mathcal{M}_{1}^{{\text{\rm main}},\text{\rm reg}}(L(u),\beta_{i})$ and it is Fredholm regular and its evaluation map is surjective to $L(u)$. Therefore when we perturb the multisection we do not need to worry about compatibility of it with other multisections we have already constructed in the earlier stage of induction. This enable us to leave the moduli space $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i})$ unperturbed for all $\beta_{i}$. The proof of Lemma 11.2 is complete. ∎ ###### Remark 11.3. We need to fix $E$ and stop the inductive construction of multisections at some finite stage. Namely we define $s_{\beta,k+1}$ only for $\beta$ with $\beta\cap\omega<E$. The reason is explained in subsection 7.2.3 [FOOO3] (= section 30.3 [FOOO2]). We can go around this trouble in the same way as explained in section 7.2 [FOOO3] (= section 30 [FOOO2]). See Remark 11.11. ###### Remark 11.4. We explain one delicate point of the proof of Lemma 11.2. Let $\alpha\in\pi_{2}(X)$ be represented by a holomorphic sphere with $c_{1}(X)\cap\alpha<0$. We consider the moduli space $\mathcal{M}_{1}(\alpha)$ of holomorphic sphere with one marked point and in homology class $\alpha$. Let us consider $\beta\in\pi_{2}(X;L(u))$ and the moduli space $\mathcal{M}_{k+1,1}^{\text{\rm main}}(\beta)$ of holomorphic discs with one interior and $k+1$ boundary marked points and of homotopy class $\beta$. The fiber product $\mathcal{M}_{1}(\alpha)\times_{X}\mathcal{M}_{k+1,1}^{\text{\rm main}}(\beta)$ taken by the evaluation maps at interior marked points are contained in $\mathcal{M}_{k+1,1}^{\text{\rm main}}(\beta+\alpha)$. If we want to define a multisection compatible with the embedding $\mathcal{M}_{1}(\alpha)\times_{X}\mathcal{M}_{k+1,1}^{\text{\rm main}}(\beta)\subset\mathcal{M}_{k+1}^{\text{\rm main}}(\beta+\alpha)$ (11.5) then it is impossible to make it both transversal and $T^{n}$ equivariant in general : This is because the nodal point of such a singular curve could be contained in the part of $X$ with non-trivial isotropy group. Our perturbation constructed above satisfies (1) and (2) of Lemma 11.2 and so may not be compatible with the embedding (11.5). Our construction of the perturbation given in Lemma 11.2 exploits the fact that the $T^{n}$ action acts freely on the Lagrangian fiber $L(u)$ and carried out by induction on the number of disc components (and of energy) only, regardless of the number of sphere components. The following corollary is an immediate consequence of Lemma 11.2. ###### Corollary 11.5. If $\mu(\beta)<0$ or $\mu(\beta)=0$, $\beta\neq 0$, then $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)^{s_{\beta}}$ is empty. Now we consider $\beta\in\pi_{2}(X;L)$ with $\mu(\beta)=2$ and $\beta\cap\omega<E$, where $E$ is as in Lemma 11.2. One immediate consequence of Corollary 11.5 is that the virtual fundamental chain of $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)$ becomes a _cycle_. More precisely, we introduce ###### Definition 11.6. Let $\beta\in\pi_{2}(X;L)$ with $\mu(\beta)=2$ and $\beta\cap\omega<E$, where $E$ is as in Lemma 11.2. We define a homology class $c_{\beta}\in H_{n}(L(u);{\mathbb{Q}})\cong{\mathbb{Q}}$ by the pushforward $c_{\beta}=ev_{*}([\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)^{s_{\beta}}]).$ ###### Lemma 11.7. The number $c_{\beta}$ is independent of the choice of the system of multisections $\mathfrak{s}_{\beta,k+1}$ satisfying $(1)$ \- $(5)$ of Proposition 11.2. ###### Proof. If there are two such systems, we can find a $T^{n}$ invariant homotopy between them which is also transversal to $0$. By a dimension counting argument applied to the parameterized version of $\mathcal{M}_{1}^{\text{\rm main}}(L(u),\beta)$ and its perturbation, we will have the parameterized version of Corollary 11.5. This in turn implies that the perturbed (parameterized) moduli space defines a compact cobordism between the perturbed moduli spaces of $\mathcal{M}_{1}^{\text{\rm main}}(\beta)$ associated to the two such systems. The lemma follows. ∎ We remark $c_{\beta_{i}}=1$ where $\beta_{i}$ ($i=1,\cdots,m$) are the classes corresponding to each of the irreducible components of the divisor $\pi^{-1}(\partial P)$. If $X$ is Fano, then $c_{\beta}=0$, for $\beta\neq\beta_{i}$. But this may not be the case if $X$ is not Fano. We now will use our perturbed moduli space to define a structure of filtered $A_{\infty}$ algebra on the de Rham cohomology $H(L(u);\Lambda_{0}^{{\mathbb{R}}})\cong(H(L(u),{\mathbb{R}}))\otimes\Lambda_{0}$. We will write it $\mathfrak{m}^{can}$. We take a $T^{n}$ equivariant Riemannian metric on $L(u)$. We observe that a differential form $\rho$ on $L(u)$ is harmonic if and only if $\rho$ is $T^{n}$ equivariant. So we identify $H(L(u),{\mathbb{R}})$ with the set of $T^{n}$ equivariant forms from now on. We consider the evaluation map $ev=(ev_{1},\cdots,ev_{k},ev_{0}):\mathcal{M}^{\text{\rm main}}_{k+1}(L(u),\beta)^{\mathfrak{s}_{\beta}}\to L(u)^{k+1}.$ Let $\rho_{1},\cdots,\rho_{k}$ be $T^{n}$ equivariant differential forms on $L(u)$. We define $\mathfrak{m}^{can}_{k,\beta}(\rho_{1},\cdots,\rho_{k})=(ev_{0})_{!}(ev_{1},\cdots,ev_{k})^{*}(\rho_{1}\wedge\cdots\wedge\rho_{k}).$ (11.6) We remark that integration along fiber $(ev_{0})_{!}$ is well defined and gives a smooth form, since $ev_{0}$ is a submersion. (It is a consequence of $T^{n}$ equivariance.) More precisely, we apply Definition 16.7 in section 16 as follows. We put $\mathcal{M}=\mathcal{M}^{\text{\rm main}}_{k+1}(L(u),\beta)$, $L_{s}=L^{k}$, $L_{t}=L$. $ev_{s}=(ev_{1},\cdots,ev_{k}):\mathcal{M}\to L_{s}$, $ev_{t}=ev_{0}:\mathcal{M}\to L_{t}$. Thus we are in the situation we formulate at the beginning of section 16. Then using Lemma 16.9 and Remark 16.8 (1), we put $\mathfrak{m}^{can}_{k,\beta}(\rho_{1},\cdots,\rho_{k})=(\mathcal{M};ev_{s},ev_{t},\mathfrak{s}_{\beta})_{*}(\rho_{1}\times\cdots\times\rho_{k}).$ (11.7) We remark that the right hand side of (11.7) is again $T^{n}$ equivariant since $\mathfrak{s}_{\beta}$ etc. are $T^{n}$ equivariant. Lemma 11.2 (4) implies $\displaystyle\partial\mathcal{M}^{\text{\rm main}}_{k+1}(L(u),\beta)$ $\displaystyle=\bigcup_{k_{1}+k_{2}=k+1}\bigcup_{\beta_{1}+\beta_{2}=\beta}\bigcup_{l=1}^{k_{2}}\mathcal{M}^{\text{\rm main}}_{k_{1}+1}(L(u),\beta_{1}){}_{ev_{0}}\times_{ev_{l}}\mathcal{M}^{\text{\rm main}}_{k_{2}+1}(L(u),\beta_{2}).$ Therefore using Lemmata 16.9, 16.10, we have the following formula. $\sum_{\beta_{1}+\beta_{2}=\beta}\sum_{k_{1}+k_{2}=k+1}\sum_{l=1}^{k_{1}}(-1)^{*}\mathfrak{m}^{can}_{k_{1},\beta_{1}}(\rho_{1},\cdots,\mathfrak{m}^{can}_{k_{2},\beta_{2}}(\rho_{l},\cdots),\cdots,\rho_{k}).$ (11.8) Here $*=\sum_{i=1}^{l}(\deg\rho_{i}+1)$. (See the end of section 16 for sign.) We now put $\mathfrak{m}_{k}^{can}(\rho_{1},\cdots,\rho_{k})=\sum_{\beta}T^{\beta\cap\omega/(2\pi)}\mathfrak{m}_{k,\beta}^{can}(\rho_{1},\cdots,\rho_{k}).$ (11.9) We extend (11.9) to $\rho\in H(L(u);\Lambda^{{\mathbb{R}}}_{0})$ such that it is $\Lambda_{0}$ multilinear. Then (11.8) implies that it defines a structure of a filtered $A_{\infty}$ structure on $H(L(u);\Lambda_{0}^{{\mathbb{R}}})$ in the sense of section 3. We also remark that our filtered $A_{\infty}$ algebra is unital and the constant $0$ form $1\in H^{0}(L;{\mathbb{R}})$ is a unit. This is a consequence of Lemma 11.2 (3). We next calculate our filtered $A_{\infty}$ structure in the case when $\rho_{i}$ are degree $1$ forms. ###### Lemma 11.8. For $\mathfrak{x}\in H^{1}(L(u),\Lambda_{0})$ and $\beta\in\pi_{2}(X,L)$ with $\mu(\beta)=2$, we have $\mathfrak{m}_{k,\beta}^{can}(\mathfrak{x},\cdots,\mathfrak{x})=\frac{c_{\beta}}{k!}(\partial\beta\cap\mathfrak{x})^{k}\cdot PD([L(u)]).$ Here $PD([L(u)])$ is the Poincaré dual to the fundamental class. In other words it is the $n$ form with $\int_{L(u)}PD([L(u)])=1$. ###### Proof. It suffices to consider the case $\mathfrak{x}=\rho\in H^{1}(L(u);{\mathbb{R}})$, and show $\int_{L(u)}\mathfrak{m}^{can}_{k,\beta}(\rho,\cdots,\rho)=\frac{c_{\beta}}{k!}(\partial\beta\cap\mathfrak{x})^{k}.$ (11.10) Let $C_{k}=\\{(t_{1},\cdots,t_{k})\mid 0\leq t_{1}\leq\cdots\leq t_{k}\leq 1\\}.$ (11.11) We define an iterated blow-up, denoted by $\widehat{C}_{k}$, of $C_{k}$ in the following way. Let $S=\partial D$ be the boundary of the unit disc $D=D^{2}\subset{\mathbb{C}}$ and $\beta_{D}\in H_{2}({\mathbb{C}},S)$ be the homology class of the unit disc. We consider the moduli space $\mathcal{M}_{k+1}({\mathbb{C}},S;\beta_{D})$ and the evaluation map $\vec{ev}=(ev_{0},\cdots,ev_{k}):\mathcal{M}_{k+1}({\mathbb{C}},S;\beta_{D})\to(S^{1})^{k+1}.$ We fix a point $p_{0}\in S\subset{\mathbb{C}}$ and put $\widehat{C}_{k}:=ev_{0}^{-1}(p_{0})\subset\mathcal{M}_{k+1}({\mathbb{C}},S;\beta_{D}).$ We make the identification $S^{1}\setminus\\{p_{0}\\}\cong(0,1)$. Then $\vec{ev}$ induces a diffeomorphism $\widehat{C}_{k}\cap\mathcal{M}_{k+1}^{\rm reg}({\mathbb{C}},S;\beta_{D})\to\text{\rm Int }C_{k}$ given by $[w,z_{0},\cdots,z_{k}]\mapsto(w(z_{1})-w(z_{0}),\cdots,w(z_{k})-w(z_{0}))$ where $\text{\rm Int }C_{k}=\\{(t_{1},\cdots,t_{k})\mid 0<t_{1}<\cdots<t_{k}<1\\}\subset C_{k}.$ In this sense $\widehat{C}_{k}$ is regarded as an iterated blow up of $C_{k}$ along the diagonal (that is the set of points where $t_{i}=t_{i+1}$ for some $i$). We identify $\partial D=S\cong{\mathbb{R}}/{\mathbb{Z}}\cong S^{1}$. We have $\mathcal{M}^{\text{\rm main}}_{k+1}(L(u),\beta)^{\mathfrak{s}}\cong\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)^{\mathfrak{s}}\times\widehat{C}_{k}.$ (11.12) In fact, Corollary 11.5 implies $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)^{\mathfrak{s}}$ consists of finitely many free $T^{n}$ orbits (with multiplicity $\in{\mathbb{Q}}$) and $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)^{\mathfrak{s}}=\mathcal{M}^{\text{\rm main,reg}}_{1}(L(u),\beta)^{\mathfrak{s}}$. By Lemma 11.2 (3) we have a map $\mathcal{M}^{\text{\rm main,reg}}_{k+1}(L(u),\beta)^{\mathfrak{s}}\to\mathcal{M}^{\text{\rm main,reg}}_{1}(L(u),\beta)^{\mathfrak{s}}$. It is easy to see that the fiber can be identified with $\widehat{C}_{k}$. Under this identification, the evaluation map $\vec{ev}$ is induced by $ev_{i}(\mathfrak{u};t_{1},\cdots,t_{k})=[t_{i}\partial\beta]\cdot ev(\mathfrak{u})$ (11.13) for $(\mathfrak{u};t_{1},\cdots,t_{k})\in\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)\times\text{\rm Int }C_{k}\subset\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)\times\widehat{C}_{k}$. Here $\partial\beta\in H_{1}(L(u);{\mathbb{Z}})$ is identified to an element of the universal cover $\widetilde{L}(u)\cong{\mathbb{R}}^{n}$ of $L(u)$ and $[t_{i}\partial\beta]\in L(u)$ acts as a multiplication on the torus. $ev(\mathfrak{u})$ is defined by the evaluation map $ev:\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)\to L(u)$. We also have : $ev_{0}(\mathfrak{u};t_{1},\cdots,t_{k})=ev(\mathfrak{u}).$ (11.14) We remark that $ev:\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta_{i})\to L(u)$ is a diffeomorphism (See Theorem 11.1 (3)). Now we have $\int_{L(u)}\mathfrak{m}^{dR}_{k,\beta}(\rho,\cdots,\rho)=c_{\beta}\text{\rm Vol}(C_{k})\left(\int_{\partial\beta}\rho\right)^{k}=\frac{c_{\beta}}{k!}(\partial\beta\cap\mathfrak{x})^{k}.$ The proof of Lemma 11.8 is now complete. ∎ ###### Remark 11.9. We can prove that our filtered $A_{\infty}$ algebra $(H(L(u);\Lambda_{0}^{{\mathbb{R}}}),\mathfrak{m}_{*}^{can})$ is homotopy equivalent to the one in [FOOO3] Theorem A (= [FOOO2] Theorem A). The proof is a straight forward generalization of section 7.5 [FOOO3] (= section 33 of [FOOO2]) and is omitted here. In fact we do not need to use this fact to prove Theorem 1.5 if we use de Rham version in all the steps of the proof of Theorem 1.5 without involving the singular homology version. ###### Remark 11.10. We constructed our filtered $A_{\infty}$ structure directly on de Rham cohomology group $H(L(u);\Lambda_{0}^{{\mathbb{R}}})$. The above construction uses the fact that wedge product of harmonic forms are again harmonic. This is a special feature of our situation where our Lagrangian submanifold $L$ is a torus. (In other words we use the fact that rational homotopy type of $L$ is formal.) Alternatively we can construct filtered $A_{\infty}$ structure on the de Rham complex $\Omega(L(u))\widehat{\otimes}_{{\mathbb{R}}}\Lambda_{0}^{{\mathbb{R}}}$ and reduce it to the de Rham cohomology by homological algebra. Namely we consider smooth forms $\rho_{i}$ which are not necessarily harmonic and use (11.6) and (11.9) to define $\mathfrak{m}_{k}^{dR}(\rho_{1},\cdots,\rho_{k})$. (The proof of $A_{\infty}$ formula is the same.) Using the formality of $T^{n}$, we can show that the canonical model of $(\Omega(L(u))\widehat{\otimes}_{{\mathbb{R}}}\Lambda_{0}^{{\mathbb{R}}},\mathfrak{m}^{dR}_{*})$ is the same as $(H(L(u);\Lambda_{0}^{{\mathbb{R}}}),\mathfrak{m}^{can})$. We omit its proof since we do not use it. Using the continuous family of perturbations, this construction can be generalized to the case of arbitrary relatively spin Lagrangian submanifold in a symplectic manifold. See [Fu5]. ###### Proof of Proposition 4.3. Proposition 4.3 immediately follows from Corollary 11.5, Lemma 11.8 and Lemma 11.9 : We just take the sum $\displaystyle\sum_{k=0}^{\infty}\mathfrak{m}^{can}_{k}(b,\cdots,b)$ $\displaystyle=$ $\displaystyle\sum_{k=0}^{\infty}\sum_{\beta\in\pi_{2}(X,L(u))}T^{\omega\cap\beta/2\pi}\mathfrak{m}^{can}_{k,\beta}(b,\cdots,b)$ (11.15) $\displaystyle=$ $\displaystyle\sum_{k=0}^{\infty}\sum_{\beta}T^{\omega\cap\beta/2\pi}\mathfrak{m}^{can}_{k,\beta}(b,\cdots,b)$ $\displaystyle=$ $\displaystyle\sum_{\beta}\sum_{k=0}^{\infty}\frac{c_{\beta}}{k!}(\partial\beta\cap b)^{k}T^{\beta\cap\omega/2\pi}\cdot PD([L(u)]).$ Note, by the degree reason, we need to take sum over $\beta$ with $\mu(\beta)=2$. Since $b$ is assumed to lie in $H^{1}(L(u),\Lambda_{+})$ not just in $H^{1}(L(u),\Lambda_{0})$, the series appearing as the scalar factor in (11.15) converges in non-Archimedean topology of $\Lambda_{0}$ and so the sum $\sum_{k=0}^{\infty}\mathfrak{m}^{can}_{k}(b,\cdots,b)$ is a multiple of $PD([L(u)])$. Hence $b\in\widehat{\mathcal{M}}_{\text{\rm weak}}(L(u))$ by definition (4.1). We remark that the gauge equivalence relation in Chapter 4 [FOOO3] is trivial on $H^{1}(L(u);\Lambda_{0})$ and so $H^{1}(L(u);\Lambda_{+})\hookrightarrow{\mathcal{M}}_{\text{\rm weak}}(L(u))$. We omit the proof of this fact since we do not use it in this paper. ∎ ###### Proof of Theorem 4.5. Suppose that there is no nontrivial holomorphic sphere whose Maslov index is nonpositive. Then Theorem 11.1 (5) implies that if $\mu(\beta)\leq 2$, $\beta\neq\beta_{i}$, $\beta\neq 0$ then $\mathcal{M}^{\text{\rm main}}_{1}(L(u),\beta)$ is empty. Therefore again by dimension counting as in Corollary 11.5, we obtain $\sum_{k=0}^{\infty}\mathfrak{m}_{k}^{dR,can}(x,\cdots,x)=\sum_{i=1}^{m}\sum_{k=0}^{\infty}T^{\omega\cap\beta_{i}/2\pi}\mathfrak{m}_{k,\beta_{i}}^{dR,can}(x,\cdots,x)$ for $x\in H^{1}(L(u),\Lambda_{+})$. On the other hand, we obtain $\displaystyle\mathfrak{PO}(x;u)$ $\displaystyle=$ $\displaystyle\sum_{i=1}^{m}\sum_{k=0}^{\infty}\frac{1}{k!}(\partial\beta_{i}\cap x)^{k}T^{\ell_{i}(u)}$ $\displaystyle=$ $\displaystyle\sum_{i=1}^{m}\sum_{k=0}^{\infty}\frac{1}{k!}\langle v_{i},x\rangle^{k}T^{\ell_{i}(u)}=\sum_{i=1}^{m}e^{\langle v_{i},x\rangle}T^{\ell_{i}(u)}$ from (6.8), (11.15) and the definition of $\mathfrak{PO}$. Writing $x=\sum_{i=1}^{n}x_{i}\text{\bf e}_{i}$ and recalling $y_{i}=e^{x_{i}}$, we obtain $e^{\langle v_{i},x\rangle}=y_{1}^{v_{i,1}}\cdots y_{n}^{v_{i,n}}$ and hence the proof of Theorem 4.5. ∎ ###### Proof of Theorem 4.6. Let $\beta\in\pi_{2}(X)$, $\mu(\beta)=2$ and $\mathcal{M}_{\text{\rm weak}}(\beta)\neq\emptyset$. Theorem 11.1 (5) implies that $\partial\beta=\sum k_{i}\partial\beta_{i},\quad\beta=\sum_{i}k_{i}\beta_{i}+\sum_{j}\alpha_{j}.$ Hence $\sum_{k}T^{\beta\cap\omega/2\pi}\mathfrak{m}_{k,\beta}^{can}(b,\cdots,b)$ becomes one of the terms of the right hand side of (4.7). We remark that class $\beta$ with $\mu(\beta)\geq 4$ does not contribute to $\mathfrak{m}_{k}^{can}(b,\cdots,b)$ by the degree reason. When all the vertices of $P$ lie in ${\mathbb{Q}}^{n}$, then the symplectic volume of all $\alpha_{j}$ are in $2\pi{\mathbb{Q}}$. Moreover $\omega\cap\beta_{i}\in 2\pi{\mathbb{Q}}$. Therefore the exponents $\beta\cap\omega/2\pi$ are rational. The proof of Theorem 4.6 is complete. ∎ ###### Remark 11.11. We remark that in Lemma 11.2 we constructed a system of multisections only for $\mathcal{M}_{k+1}^{\text{main}}(L(u),\beta)$ with $\beta\cap\omega<E$. So we obtain only an $A_{n,K}$ structure instead of a filtered $A_{\infty}$ structure. Here $(n,K)=(n(E),K(E))$ depends on $E$ and $\lim_{E\to\infty}(n(E),K(E))=(\infty,\infty)$. It induces an $A_{n,K}$ structure $\mathfrak{m}^{(E)}$ on $H(L;\Lambda_{0})$. (See subsection 7.2.7 [FOOO3] = subsection 30.7 [FOOO2].) In the same way as section 7.2 [FOOO3] (= section 30 [FOOO2]), we can find $(n^{\prime}(E),K^{\prime}(E))$ such that $(n^{\prime}(E),K^{\prime}(n))\to(\infty,\infty)$ as $E\to\infty$ and the following holds. If $E_{1}<E_{2}$ then $A_{n(E_{1}),K(E_{1})}$ structure $\mathfrak{m}^{(E_{1})}$ is $(n^{\prime}(E),K^{\prime}(E))$-homotopy equivalent to $\mathfrak{m}^{(E_{2})}$. This implies that we can extend $\mathfrak{m}^{(E_{1})}$ (regarded as $A_{n^{\prime}(E_{1}),K^{\prime}(E_{1})}$ structure) to a filtered $A_{\infty}$ structure by Theorem 7.2.72 [FOOO3] (= Theorem 30.72 [FOOO2]). (We also remark that for all the applications in this paper we can use filtered $A_{n,K}$ structure for sufficiently large $n,K$, in place of filtered $A_{\infty}$ structure.) Moreover we can use Lemma 11.7 to show the following. If $\mathfrak{x}_{i}\in H^{1}(L(u);{\mathbb{Q}})$ then $\mathfrak{m}_{k,\beta}^{(E)}(\mathfrak{x}_{1},\cdots,\mathfrak{x}_{k})$ is independent of $E$. So in particular it coincides to one of filtered $A_{\infty}$ structure we define as above. In other words, since the number $c_{\beta}$ is independent of the choice of the system of $T^{n}$ invariant multisections it follows that the potential function in Theorem 4.6 is independent of it. However we do not know how to calculate it. ###### Remark 11.12. We used de Rham cohomology to go around the problem of transversality among chains in the classical cup product. One drawback of this approach is that we lose control of the rational homotopy type. Namely we do not prove here that the filtered $A_{\infty}$ algebra (partially) calculated above is homotopy equivalent to the one in Theorem A [FOOO3] (= Theorem A [FOOO2]) over ${\mathbb{Q}}$. (Note all the operations we obtain is defined over ${\mathbb{Q}}$, however.) We however believe that they are indeed homotopy equivalent over ${\mathbb{Q}}$. There may be several possible ways to prove this statement, one of which is to use the rational de Rham forms used by Sullivan. Moreover since the number $c_{\beta}$ is independent of the choices we made, the structure of filtered $A_{\infty}$ algebra on $H(L(u),\Lambda^{{\mathbb{Q}}})$ is well defined, that is independent of the choices involved. The ${\mathbb{Q}}$-structure is actually interesting in our situation. See for example Proposition 7.13. However homotopy equivalence of the ${\mathbb{Q}}$-version of Lemma 11.9 is not used in the statement of Proposition 7.13 or in its proof. ## 12\. Non-unitary flat connection on $L(u)$ In this section we explain how we can include (not necessarily unitary) flat bundles on Lagrangian submanifolds in Lagrangian Floer theory following [Fu3], [Cho]. ###### Remark 12.1. We need to use flat complex line bundle for our purpose by the following reason. In [FOOO3] we assumed that our bounding cochain $b$ is an element of $H(L;\Lambda_{+})$ since we want the series $\mathfrak{m}_{1}^{b}(x)=\sum_{k,\ell}\mathfrak{m}_{k+\ell+1}(b^{\otimes k},x,b^{\otimes\ell})$ to converge. There we used convergence with respect to the non-Archimedean norm. For the case of Lagrangian fibers in toric manifold, the above series converges for $b\in H^{1}(L;\Lambda_{0})$. The convergence is the usual (classical Archimedean) topology on ${\mathbb{C}}$ on each coefficient of $T^{\lambda}$. This is not an accident and was expected to happen in general. (See Conjecture 3.6.46 [FOOO3] = Conjecture 11.46 [FOOO2].) However for this convergence to occur, we need to choose the perturbations on $\mathcal{M}_{k+1}^{\text{\rm main}}(L,\beta)$ so that it is consistent with $\mathcal{M}_{k^{\prime}+1}^{\text{\rm main}}(L,\beta)$ ($k^{\prime}\neq k$) via the forgetful map. We can make this choice for the current toric situation by Lemma 11.2 (3). In a more general situation, we need to regard $\mathcal{M}_{1}^{\text{\rm main}}(L,\beta)$ as a chain in the free loop space. (See [Fu4].) On the other hand, if we use the complex structure other than the standard one, we do not know whether Lemma 11.2 (3) holds or not. So in the proof of independence of Floer cohomology under the various choices made, there is a trouble to use a bounding cochain $b$ lying in $H^{1}(L;\Lambda_{0})$. The idea, which is originally due to Cho [Cho], is to change the leading order term of $\mathfrak{x}$ by twisting the construction using _non-unitary_ flat bundles on $L$. Let $X$ be a symplectic manifold and $L$ be its relatively spin Lagrangian submanifold. Let $\rho:H_{1}(L;{\mathbb{Z}})\to{\mathbb{C}}\setminus\\{0\\}$ be a representation and $\mathfrak{L}_{\rho}$ be the flat ${\mathbb{C}}$ bundle induced by $\rho$. We replace the formula (11.9) by $\mathfrak{m}_{k}^{\rho,can}=\sum_{\beta\in H_{2}(M,L)}\rho(\partial\beta)\,\,\mathfrak{m}^{can}_{k,\beta}\otimes T^{\omega(\beta)/2\pi}.$ (Compare this with (4.3) in section 4.) ###### Proposition 12.2. $(H(L(u);\Lambda_{0}^{{\mathbb{R}}}),\mathfrak{m}_{k}^{\rho,can})$ is a filtered $A_{\infty}$ algebra. ###### Proof. Suppose that $[f]\in\mathcal{M}_{k+1}^{\text{\rm main}}(L,\beta)$ is a fiber product of $[f_{1}]\in\mathcal{M}_{\ell+1}^{\text{\rm main}}(L,\beta_{1})$ and $[f_{2}]\in\mathcal{M}_{k-\ell}^{\text{\rm main}}(L,\beta_{2})$. Namely $\beta_{1}+\beta_{2}=\beta$ and $ev_{0}(f_{2})=ev_{i}(f_{1})$ for some $i$. Then it is easy to see that $\rho(\partial\beta)=\rho(\partial\beta_{1})\rho(\partial\beta_{2}).$ (12.1) (12.1) and (11.8) imply the filtered $A_{\infty}$ relation. ∎ The unitality can also be proved in the same way. The well definedness (that is, independence of various choices up to homotopy equivalence) can also be proved in the same way. ###### Remark 12.3. We have obtained our twisted filtered $A_{\infty}$ structure on the (untwisted) cohomology group $H^{*}(L;\Lambda_{0})$. This is because the flat bundle $Hom(\mathfrak{L}_{\rho},\mathfrak{L}_{\rho})$ is trivial. In more general situation where we consider a flat bundle $\mathfrak{L}$ of higher rank, we obtain a filtered $A_{\infty}$ structure on cohomology group with local coefficients with values in $Hom(\mathfrak{L},\mathfrak{L})$. The filtered $A_{\infty}$ structure $\mathfrak{m}_{k}^{\rho,can}$ is different from $\mathfrak{m}_{k}^{can}$ in general as we can see from the expression of the potential function given in Lemma 4.9. In the rest of this section, we explain the way how the Floer cohomology detects the Lagrangian intersection. Namely we sketch the proof of Theorem 3.11 in our case and its generalization to the case where we include the nonunital flat connection $\rho$. Let $\psi_{t}:X\to X$ be a Hamiltonian isotopy with $\psi_{0}=$ identity. We put $\psi_{1}=\psi$. We consider the pair $L^{(0)}=L(u),\quad L^{(1)}=\psi(L(u))$ such that $L^{(1)}$ is transversal to $L^{(0)}$. We then take a one-parameter family $\\{J_{t}\\}_{t\in[0,1]}$ of compatible almost complex structures such that $J_{0}=J$ which is the standard complex structure of $X$ and $J_{1}=\psi_{*}(J)$. Let $p,q\in L^{(0)}\cap L^{(1)}$. We consider the homotopy class of a maps $\varphi:{\mathbb{R}}\times[0,1]\to X$ (12.2) such that 1. (1) $\lim_{\tau\to-\infty}\varphi(\tau,t)=p$, $\lim_{\tau\to+\infty}\varphi(\tau,t)=q$. 2. (2) $\varphi(\tau,0)\in L^{(0)}$, $\varphi(\tau,1)\in L^{(1)}$. We denote by $\pi_{2}(L^{(1)},L^{(0)};p,q)$ the set of all such homotopy classes. There are obvious maps $\displaystyle\pi_{2}(L^{(1)},L^{(0)};p,r)\times\pi_{2}(L^{(1)},L^{(0)};r,q)\to\pi_{2}(L^{(1)},L^{(0)};p,q),$ (12.3) $\displaystyle\pi_{2}(X;L^{(1)})\times\pi_{2}(L^{(1)},L^{(0)};p,q)\to\pi_{2}(L^{(1)},L^{(0)};p,q),$ $\displaystyle\pi_{2}(L^{(1)},L^{(0)};p,q)\times\pi_{2}(X;L^{(0)})\to\pi_{2}(L^{(1)},L^{(0)};p,q).$ We denote them by $\\#$. ###### Definition 12.4. We consider the moduli space of maps (12.2) satisfying (1), (2) above, in homotopy class $B\in\pi_{2}(L^{(1)},L^{(0)};p,q)$, and satisfies the following equation : $\frac{\partial\varphi}{\partial\tau}+J_{t}\left(\frac{\partial\varphi}{\partial t}\right)=0.$ (12.4) We denote it by $\widehat{\mathcal{M}}^{\text{reg}}(L^{(1)},L^{(0)};p,q;B).$ We take its stable map compactification and denote it by $\widehat{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B).$ We divide this space by the ${\mathbb{R}}$ action induced by the translation of $\tau$ direction to obtain ${\mathcal{M}}(L^{(1)},L^{(0)};p,q;B).$ We define evaluation maps $ev_{L^{(i)}}:\widehat{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)\to L^{(i)}$ by $ev_{L^{(i)}}(\varphi)=\varphi(0,i),$ for $i=0,1$. ###### Lemma 12.5. ${\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)$ has an oriented Kuranishi structure with corners. Its boundary is isomorphic to the union of the following three kinds of fiber products as spaces with Kuranishi structure. 1. (1) ${\mathcal{M}}(L^{(1)},L^{(0)};p,r;B^{\prime})\times{\mathcal{M}}(L^{(1)},L^{(0)};r,q;B^{\prime\prime}).$ Here $B^{\prime}\in\pi_{2}(L^{(1)},L^{(0)};p,r)$, $B^{\prime\prime}\in\pi_{2}(L^{(1)},L^{(0)};r,q)$. 2. (2) ${\mathcal{M}}_{1}(L(u);\beta^{\prime})\,\,{}_{ev_{0}}\times_{ev_{L^{(1)}}}\,\,\widehat{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B^{\prime\prime}).$ Here $\beta^{\prime}\in\pi_{2}(X;L^{(1)})\cong\pi_{2}(X;L(u))$, $\beta^{\prime}\\#B^{\prime\prime}=B$. The fiber product is taken over $L^{(1)}\cong L(u)$ by using $ev_{0}:{\mathcal{M}}(L(u);\beta^{\prime})\to L(u)$ and $ev_{L^{(1)}}:\widehat{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B^{\prime\prime})\to L^{(1)}$. 3. (3) $\widehat{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B^{\prime})\,\,{}_{ev_{L^{(0)}}}\times_{ev_{0}}\,\,{\mathcal{M}}(L(u);\beta^{\prime\prime}).$ Here $\beta^{\prime}\in\pi_{2}(X;L^{(1)})\cong\pi_{2}(X;L(u))$, $B^{\prime}\\#\beta^{\prime\prime}=B$. The fiber product is taken over $L^{(0)}\cong L(u)$ by using $ev_{0}:{\mathcal{M}}(L(u);\beta^{\prime\prime})\to L(u)$ and $ev_{L^{(0)}}:\widehat{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B^{\prime})\to L^{(0)}$. We have $\mathop{\kern 0.0pt{\rm dim}}\nolimits{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)=\mu(B)-1$ where $\displaystyle\mu(B_{1}\\#B_{2})$ $\displaystyle=\mu(B_{1})+\mu(B_{2}),$ $\displaystyle\mu(\beta^{\prime}\\#B^{\prime\prime})$ $\displaystyle=\mu(\beta^{\prime})+\mu(B^{\prime\prime}),\quad\mu(B^{\prime}\\#\beta^{\prime\prime})=\mu(B^{\prime})+\mu(\beta^{\prime\prime}).$ Here the notations are as in $(1)$, $(2)$, $(3)$ above. Lemma 12.5 is proved in subsection 7.1.4 [FOOO3] (= section 29.4 [FOOO2]). ###### Lemma 12.6. There exists a system of multisections on ${\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)$ such that 1. (1) It is transversal to $0$. 2. (2) It is compatible at the boundaries described in Lemma 12.5. Here we pull back multisection of ${\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)$ to one on $\widehat{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)$ and use the multisection in Lemma 11.2 on ${\mathcal{M}}(L(u);\beta)$. ###### Proof. We can find such system of multisections inductively over $\int_{B}\omega$ by using the fact that $ev_{0}:{\mathcal{M}}(L(u);\beta)^{\mathfrak{s}}\to L(u)$ is a submersion and Lemma 12.5 (1) is a direct product. ∎ In the case when $\mu(B)=1$, we define $n(B)=\\#{\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)^{\mathfrak{s}}\in{\mathbb{Q}}$ that is the number of zeros of our multisection counted with sign and multiplicity. We use it to define Floer cohomology. Let $CF(L^{(1)},L^{(0)})$ be the free $\Lambda_{0,nov}$ module generated by $L^{(0)}\cap L^{(1)}$. We will define boundary operator on it. Let $\rho_{i}:\pi_{1}(L^{(i)})\to{\mathbb{C}}^{*}$ be the representation. We take harmonic $1$ form $h_{i}\in H^{1}(L^{(i)};{\mathbb{C}})$ such that $\rho_{i}(\gamma)=\exp\left(\int_{\gamma}h_{i}\right).$ Let $b_{i,+}\in H^{1}(L^{(0)};\Lambda_{+})\subset\mathcal{M}(L^{(0)})$. An element $B\in\pi_{2}(L^{(1)},L^{(0)};p,q)$ induces a path $\partial_{i}:[0,1]\to L^{(i)}$ joining $p$ to $q$ in $L^{(i)}$ for $i=0,1$. We define $C(B;(h_{0},h_{1}),(b_{0,+},b_{1,+}))=\exp\left(\int_{\partial_{0}B}(h_{0}+b_{0,+})\right)\exp\left(-\int_{\partial_{1}B}(h_{1}+b_{1,+})\right).$ This is an element of $\Lambda_{0}\setminus\Lambda_{+}$. It is easy to see that $\displaystyle C(B_{1}\\#B_{2};(h_{0},h_{1}),(b_{0,+},b_{1,+}))$ (12.5) $\displaystyle=C(B_{1};(h_{0},h_{1}),(b_{0,+},b_{1,+}))\,C(B_{2};(h_{0},h_{1}),(b_{0,+},b_{1,+})),$ and $\displaystyle C(\beta^{\prime}\\#B^{\prime\prime};(h_{0},h_{1}),(b_{0,+},b_{1,+}))$ (12.6) $\displaystyle=C(B^{\prime\prime};(h_{0},h_{1}),(b_{0,+},b_{1,+}))\exp\left(\int_{\partial\beta^{\prime}}h_{1}+b_{0,+}\right).$ (Here $B^{\prime}$ and $\beta^{\prime\prime}$ are as in Lemma 12.5 (2).) Now we define $\displaystyle\langle\delta_{(h_{0},h_{1}),(b_{0,+},b_{1,+})}(p),q\rangle$ (12.7) $\displaystyle=\sum_{B\in\pi_{2}(L^{(1)},L^{(0)};p,q);\mu(B)=1}n(B)C(B;(h_{0},h_{1}),(b_{0,+},b_{1,+})).$ For the case $h_{1}=\psi_{*}(h_{0})$, $b_{1,+}=\psi_{*}(b_{+})$, $b_{0,+}=b_{+}$, we write $C(B;h_{0},b_{+})$ and $\delta^{h_{0},b_{+}}$, in place of $C(B;(h_{0},h_{1}),(b_{0,+},b_{1,+}))$ and $\delta_{(h_{0},h_{1}),(b_{0,+},b_{1,+})}$, respectively. ###### Lemma 12.7. $\delta_{(h_{0},h_{1}),(b_{0,+},b_{1,+})}\circ\delta_{(h_{0},h_{1}),(b_{0,+},b_{1,+})}=(\mathfrak{PO}(b(1))-\mathfrak{PO}(b(0)))\cdot id$ where $b(i)=h_{i}+b_{i,+}\in H^{1}(L(u);\Lambda_{0})$. ###### Proof. Let $p,q\in L^{(1)}\cap L^{(0)}$. We consider $B\in\pi_{2}(L^{(1)},L^{(0)};p,q)$ with $\mu(B)=2$. We consider the boundary of ${\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)$. We put $\delta_{B}=\begin{cases}1&\text{$B\in\pi_{2}(L^{(1)},L^{(0)};p,q)$, $p=q$, and $B$ is the class of constant map,}\\\ 0&\text{otherwise.}\end{cases}$ Then using the classification of the boundary of ${\mathcal{M}}(L^{(1)},L^{(0)};p,q;B)$ in Lemma 12.5 we have the following equality : $0=\sum_{r,B,B^{\prime\prime}}n(B^{\prime})n(B^{\prime\prime})+\sum_{\beta^{\prime},B^{\prime\prime}}c_{\beta^{\prime}}\delta_{B^{\prime\prime}}-\sum_{\beta^{\prime\prime},B^{\prime}}c_{\beta^{\prime\prime}}\delta_{B^{\prime}},$ (12.8) where the summention in the first, second, third terms of right hand side is taken over the set described in Lemma 12.5 (1), (2), (3), respectively, and $c_{\beta}$ is defined in Definition 11.6. We multiply (12.8) by $C(B;(h_{0},h_{1}),(b_{0,+},b_{1,+}))$ and calculate the right hand side by using Formulae (12.5), (12.6) and Lemma 11.8. It is easy to see that the first term gives $\langle\delta_{(h_{0},h_{1}),(h_{0,+},b_{1,+})}\circ\delta_{(h_{0},h_{1}),(b_{0,+},b_{1,+})}(p),q\rangle.$ The second term is $0$ if $p\neq q$, and is $\sum_{\beta}c_{\beta}\exp\left(\int_{\partial\beta}h_{0}+b_{0,+}\right)=\mathfrak{PO}(b(0))$ if $p=q$. The third term gives $\mathfrak{PO}(b(1))\cdot id$ in a similar way. ∎ ###### Definition 12.8. Let $b(0)=h_{0}+b_{+}$ and $b(1)=\psi_{*}(b(0))$. We define $HF((L^{(0)},b(0)),(L^{(1)},b(1));\Lambda_{0})\cong\frac{\text{Ker}\,\,\delta_{h_{0},h_{+}}}{\text{Im}\,\,\delta_{h_{0},h_{+}}}.$ This is well defined by Lemma 12.7 and by the identity $\mathfrak{PO}(\psi_{*}(b(0)))=\mathfrak{PO}((b(0)))$. Now we have ###### Lemma 12.9. $HF((L^{(0)},b(0)),(L^{(1)},b(1));\Lambda_{0})\otimes_{\Lambda_{0}}\Lambda\cong\frac{\text{\rm Ker}\,\,\mathfrak{m}_{1}^{\rho,can,b_{+}}}{\text{\rm Im}\,\,\mathfrak{m}_{1}^{\rho,can,b_{+}}}\otimes_{\Lambda_{0}}\Lambda.$ Here $\rho(\gamma)=\exp\left(\int_{\gamma}h_{0}\right)$ and $\mathfrak{m}_{1}^{\rho,can,b_{+}}(x)=\sum_{k_{1},k_{2}=0}^{\infty}\mathfrak{m}^{\rho,can}_{k_{1}+k_{2}+1}(b_{+}^{\otimes k_{1}},x,b_{+}^{\otimes k_{2}}).$ Lemma 12.9 implies (the $\rho$ twisted version of) Theorem 3.11 in our case. We omit the proof of Lemma 12.9 and refer section 5.3 [FOOO3] (= section 22 [FOOO2]) or section 8 [FOOO4]. ## 13\. Floer cohomology at a critical point of potential function In this section we prove Theorem 4.10 etc. and complete the proof of Theorem 1.5. ###### Proof of Lemma 4.9. Let $\beta\in H_{2}(X,L(u_{0}))$ with $\mu(\beta)=2$ and $\mathcal{M}_{1}^{\text{\rm main}}(L(u_{0}),\beta)$ be nonempty. We have $\beta=\sum_{i=1}^{m}c_{i}\beta_{i}+\sum_{j}\alpha_{j}$ by Theorem 11.1 (5). Let $\rho$ be as in (4.15). We have $\rho(\partial\beta)=\prod\rho(\partial\beta_{i})^{c_{i}}$. Note $\partial\beta_{i}=\sum_{j}v_{i,j}\text{\bf e}_{j}^{*}$. Thus we have $\left\\{\begin{aligned} \rho(\partial\beta_{i})&=\mathfrak{y}_{1,0}^{v_{i,1}}\cdots\mathfrak{y}_{n,0}^{v_{i,n}},\\\ \rho(\partial\beta)&=\prod_{i}\prod_{j}\mathfrak{y}_{j,0}^{c_{i}v_{i,j}}.\end{aligned}\right.$ (13.1) Therefore for $b\in H^{1}(L(u_{0});\Lambda_{+}^{{\mathbb{C}}})$ we have $\displaystyle\sum_{k=0}^{\infty}\mathfrak{m}_{k,\beta}^{\rho,can}(b,\cdots,b)$ $\displaystyle=\sum_{k=0}^{\infty}\mathfrak{y}_{1,0}^{v_{i,1}}\cdots\mathfrak{y}_{n,0}^{v_{i,n}}\mathfrak{m}_{k,\beta}^{can}(b,\cdots,b)$ $\displaystyle=\sum_{k=0}^{\infty}e^{\mathfrak{x}_{1,0}v_{i,1}}\cdots e^{\mathfrak{x}_{n,0}v_{i,n}}\frac{c_{\beta}}{k!}(b\cap\partial\beta)^{k}\cdot[PD(L)]$ $\displaystyle=\sum_{k=0}^{\infty}\mathfrak{m}^{can}_{k,\beta}\left(b+\sum_{j=1}^{n}\mathfrak{x}_{j,0}\text{\bf e}_{j},\cdots,b+\sum_{j=1}^{n}\mathfrak{x}_{j,0}\text{\bf e}_{j}\right).$ On the other hand, it follows from Theorems 4.5 and 4.6 that the left and the right sides of this identity correspond to those in Lemma 4.9 respectively. This finishes the proof of Lemma 4.9. ∎ ###### Proof of Theorem 4.10. Let $x^{+}=(x^{+}_{1},\cdots,x^{+}_{n})$, $x^{+}_{1},\cdots,x^{+}_{n}\in\Lambda_{+}$. We put $x(x^{+})=\sum_{i}(\mathfrak{x}_{i,0}+x^{+}_{i})\text{\bf e}_{i},\quad b(x^{+})=\sum_{i}x^{+}_{i}\text{\bf e}_{i}.$ From Lemma 4.9 we derive $\displaystyle\mathfrak{PO}^{u_{0}}_{\rho}(b(x^{+}))$ $\displaystyle=\sum\mathfrak{m}_{k}^{\rho,can}(b(x^{+}),\cdots,b(x^{+}))\cap[L(u_{0})]$ $\displaystyle=\sum\mathfrak{m}_{k}^{can}(x(x^{+}),\cdots,x(x^{+}))\cap[L(u_{0})]=\mathfrak{PO}^{u_{0}}(x(x^{+})).$ Let $\mathfrak{x}$ be as in (4.13). Then we have $\frac{\partial}{\partial x^{+}_{i}}\mathfrak{PO}^{u_{0}}_{\rho}(b(x^{+}))\Big{|}_{x(x^{+})=\mathfrak{x}}=\frac{\partial}{\partial x^{+}_{i}}\mathfrak{PO}^{u_{0}}(x(x^{+}))\Big{|}_{x(x^{+})=\mathfrak{x}}=\frac{\partial\mathfrak{PO}^{u_{0}}}{\partial x_{i}}(\mathfrak{x})=0,$ where the last equality follows from the assumption (4.12). (In case when (4.11) is assumed we have ‘$\equiv 0\mod T^{\mathcal{N}}$’ in place of ‘$=0$’.) On the other hand, we have $\displaystyle\frac{\partial}{\partial x^{+}_{i}}$ $\displaystyle\mathfrak{PO}^{u_{0}}_{\rho}(b(x^{+}))\Big{|}_{x(x^{+})=\mathfrak{x}}$ (13.2) $\displaystyle=\sum_{k}\sum_{\ell}\mathfrak{m}^{\rho,can}_{k}(b^{\otimes\ell},\text{\bf e}_{i},b^{\otimes(k-\ell-1)})\cap[L(u_{0})]$ $\displaystyle=\mathfrak{m}_{1}^{\rho,can,b}(\text{\bf e}_{i})\cap[L(u_{0})].$ Note here and hereafter we write $b$ in place of $b(x^{+})$ with $x(x^{+})=\mathfrak{x}$. Namely $b=\mathfrak{x}-\sum\mathfrak{x}_{i,0}\text{\bf e}_{i}$. Hence we obtain $\mathfrak{m}_{1}^{\rho,can,b}(\text{\bf e}_{i})\begin{cases}=0&\text{if (\ref{formula:critical0}) is satisfied}\\\ \equiv 0\mod T^{{\mathcal{N}}}&\text{if (\ref{formula:criticalweak}) is satisfied}.\end{cases}$ (13.3) We remark that by the degree reason $\mathfrak{m}^{\rho,can,b}_{1}(\text{\bf e}_{i})$ is proportional to $PD[L(u_{0})]$. We next prove the vanishing of $\mathfrak{m}_{1}^{\rho,can,b}(\text{\bf f})$ for the classes f of higher degree. Namely we prove ###### Lemma 13.1. For $\text{\bf f}\in H^{*}(L(u_{0});\Lambda_{0}^{{\mathbb{C}}})$ we have : $\mathfrak{m}_{1,\beta}^{\rho,can,b}(\text{\bf f})\begin{cases}=0&\text{if $(\ref{formula:critical0})$ is satisfied}\\\ \equiv 0\mod T^{{\mathcal{N}}}&\text{if $(\ref{formula:criticalweak})$ is satisfied}.\end{cases}$ ###### Proof. Let $d=\deg\text{\bf f}$ and $2\ell=\mu(\beta)$. We say $(d,\ell)<(d^{\prime},\ell^{\prime})$ if $\ell<\ell^{\prime}$ or $\ell=\ell^{\prime}$, $d<d^{\prime}$. We prove the lemma by upward induction on $(d,\ell)$. The case $d=1$ is (13.3). We remark that $\mathfrak{m}_{k,\beta}=0$ if $\mu(\beta)\leq 0$. We assume that the lemma is proved for $(d^{\prime},\ell^{\prime})$ smaller than $(d,\ell)$ and will prove the case of $(d,\ell)$. Since the case $d=1$ is already proved, we may assume that $d\geq 2$. Let $\text{\bf f}=\text{\bf f}_{1}\cup\text{\bf f}_{2}$ where $\deg\text{\bf f}_{i}\geq 1$. By the $A_{\infty}$-relation we have $\displaystyle\mathfrak{m}_{1,\beta}^{\rho,can,b}(\text{\bf f}_{1}\cup\text{\bf f}_{2})=$ $\displaystyle\sum_{\beta_{1}+\beta_{2}=\beta}\pm\mathfrak{m}^{\rho,can,b}_{2,\beta_{1}}(\mathfrak{m}^{\rho,can,b}_{1,\beta_{2}}(\text{\bf f}_{1}),\text{\bf f}_{2})$ $\displaystyle+\sum_{\beta_{1}+\beta_{2}=\beta}\pm\mathfrak{m}^{\rho,can,b}_{2,\beta_{1}}(\text{\bf f}_{1},\mathfrak{m}_{1,\beta_{2}}(\text{\bf f}_{2}))$ $\displaystyle+\sum_{\beta_{1}+\beta_{2}=\beta,\beta_{2}\neq 0}\pm\mathfrak{m}^{\rho,can,b}_{1,\beta_{1}}(\mathfrak{m}_{2,\beta_{2}}(\text{\bf f}_{1},\text{\bf f}_{2})).$ We remark that $\mathfrak{m}^{\rho,can,b}_{1,\beta_{0}}=0$ since we are working on a canonical model. The first two terms of the right hand side vanishes by the induction hypothesis since $\deg\text{\bf f}_{i}<\deg\text{\bf f}$ and $\mu(\beta_{i})\leq\mu(\beta)$. The third term also vanishes since $\mu(\beta_{1})<\mu(\beta)$. The proof of Lemma 13.1 is complete. ∎ Lemma 13.1 immediately implies Theorem 4.10. ∎ ###### Proof of Proposition 5.4. Let us specialize to the case of 2 dimension. In case $\mathop{\kern 0.0pt{\rm dim}}\nolimits L(u_{0})=2$, we can prove $\mathfrak{m}_{1,\beta}^{\rho,can,b}=0$ for $\mu(\beta)\geq 4$ also by dimension counting. We can use that to prove Proposition 5.4 in the same way as above. ∎ ###### Proof of Proposition 4.12. Let $\omega_{i}$, $u_{i}$, $\mathfrak{x}_{i,\mathcal{N}}$ be as in Definition 4.11. We assume $\psi:X\to X$ does not satisfy (4.19) or (4.20) and will deduce a contradiction. We use the same (time dependent) Hamiltonian as $\psi$ to obtain $\psi_{i}:(X,\omega_{i})\to(X,\omega_{i})$. Take an integer ${\mathcal{N}}$ such that $\|\psi_{i}\|<2\pi{\mathcal{N}}$ for large $i$. Then for sufficiently large $i$, $L(u_{0}^{i})$ and $\psi_{i}$ does not satisfy (4.19) or (4.20). In fact if $\psi(L(u_{0}))\cap L(u_{0})=\emptyset$ then for sufficiently large $i$, we have $\psi_{i}(L(u^{i}_{0}))\cap L(u^{i}_{0})=\emptyset$. If $\psi(L(u_{0}))$ is transversal to $L(u_{0})$ and if (4.20) is not satisfied, then $\\#(\psi(L(u_{0}))\cap L(u_{0}))\geq\\#(\psi_{\varepsilon_{i}}(L(u^{i}_{0}))\cap L(u^{i}_{0})).$ On the other hand, by Theorem 4.10 we have $HF((L(u_{i}),\mathfrak{x}_{i,k}),(L(u_{i}),\mathfrak{x}_{i,k});\Lambda_{0}^{{\mathbb{C}}}/(T^{{\mathcal{N}}}))\cong H(T^{n};\Lambda_{0}^{{\mathbb{C}}}/(T^{{\mathcal{N}}})).$ It follows from Universal Coefficient Theorem that $HF((L(u_{i}),\mathfrak{x}_{i,k}),(L(u_{i}),\mathfrak{x}_{i,k});\Lambda_{0}^{{\mathbb{C}}})\cong\Lambda_{0}^{\oplus a}\oplus\bigoplus_{i=1}^{b}\Lambda_{0}/(T^{c(i)})$ (13.4) such that $c(i)\geq{\mathcal{N}}$ and $a+2b\geq 2^{n}$. This contradicts to Theorem J [FOOO3]. (In fact Theorem J [FOOO3] (= Theorem 3.12) and (13.4) imply that (4.19) and (4.20) hold for $L(u_{i})$ and $\psi_{i}$ with $\|\psi_{i}\|<2\pi{\mathcal{N}}$.) Proposition 4.12 is proved. ∎ ###### Proof of Theorem 5.11. The proof is the same as the proof of Proposition 4.12 above. ∎ Now we are ready to complete the proof of Theorem 1.5. ###### Proof of Theorem 1.5. In the case where the vertices of $P$ are contained in ${\mathbb{Q}}^{n}$, Proposition 4.7 and Theorem 4.10 imply that $L(u_{0})$ is balanced in the sense of Definition 4.11. Therefore Proposition 4.12 implies Theorem 1.5 in this case. If the leading term equation is strongly nondegenerate, Theorem 1.5 also follows from Theorem 4.10, Theorem 10.4 and Proposition 4.12. We finally present an argument to remove the rationality assumption. In view of Lemma 4.12, it suffices to prove the following Proposition 13.2. ###### Proposition 13.2. In the situation of Theorem 1.5, there exists $u_{0}$ such that $L(u_{0})$ is a balanced Lagrangian fiber. ###### Proof. Let $\pi:X\to P$ be as in Theorem 1.5. Let us consider $s_{k}$, $S_{k}$, $P_{k}$ as in section 9. We obtain $u_{0}\in P$ such that $\\{u_{0}\\}=P_{K}$. We will prove that $L(u_{0})$ is balanced. We perturb the Kähler form $\omega$ of $X$ a bit so that we obtain $\omega^{\prime}$. Let $P^{\prime}$ be the corresponding moment polytope and $s_{k}^{\omega^{\prime}}$, $S_{k}^{\omega^{\prime}}$, $P_{k}^{\omega^{\prime}}$, be the corresponding piecewise affine function, number, subset of $P_{\omega^{\prime}}$ obtained for $\omega^{\prime}$, $P_{\omega^{\prime}}$ as in section 9. ###### Proposition 13.3. We can choose $\omega_{h}$ so that $\omega_{h}$ is rational and $\lim_{h\to\infty}s_{k}^{\omega_{h}}=s_{k}$, $\lim_{h\to\infty}S_{k}^{\omega_{h}}=S_{k}$, $\lim_{h\to\infty}P_{k}^{\omega_{h}}=P_{k}$, $\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{k}^{\omega_{h}}=\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{k}$ for all $k$. ###### Proof. We write $I_{k}^{\omega^{\prime}}$ the set $I_{k}$ defined in (9.7) section 9 for $\omega^{\prime}$, $P^{\prime}$. We will prove the following lemma. We remark that the set $\mathfrak{K}$ of $T^{n}$ invariant Kähler structure $\omega^{\prime}$ is regarded as an open set of an affine space defined on ${\mathbb{Q}}$ (that is the Kähler cone). We may regard $\mathfrak{K}$ as a moduli space of moment polytope as follows : We consider a polyhedron $P^{\prime}$ each of whose edges is parallel to a corresponding edge of $P$. Translation defines an ${\mathbb{R}}^{n}$ action on the set of such $P^{\prime}$. The quotient space can be identified with $\mathfrak{K}$. ###### Lemma 13.4. There exists a subset $\mathfrak{P}_{k}$ of $\mathfrak{K}$ which is a nonempty open subset of an affine subspace defined over ${\mathbb{Q}}$ such that any element $\omega^{\prime}\in\mathfrak{P}_{k}$ has the following properties : 1. (1) $\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{l}^{\omega^{\prime}}=\mathop{\kern 0.0pt{\rm dim}}\nolimits P_{l}^{\omega}$ for $l\leq k$. 2. (2) $I_{l}^{\omega^{\prime}}=I_{l}^{\omega}$ for $l\leq k$. ###### Remark 13.5. In the case of Example 8.1, the set $P^{\omega^{\prime}}_{k}$ etc. jumps at the point $\alpha=1/3$ in the Kähler cone. Hence the set $\mathfrak{P}_{k}$ may have strictly smaller dimension than $\mathfrak{K}$. ###### Proof. Let $A_{l}^{\omega^{\prime}}$ be the affine space defined in section 9. (We put ${\omega^{\prime}}$ to specify the symplectic form.) We write $\ell_{i}^{\omega}$, $\ell_{i}^{\omega^{\prime}}$ in place of $\ell_{i}$ to specify symplectic form and moment polytope. We remark that the linear part of $\ell_{i}^{\omega}$ is equal to the linear part of $\ell_{i}^{\omega^{\prime}}$. The proof of Lemma 13.4 is by induction on $k$. Let us first consider the case $k=1$. We put $\widehat{A}_{1}^{\omega^{\prime}}=\\{u\in M_{\mathbb{R}}\mid\ell_{1,1}^{\omega^{\prime}}(u)=\cdots=\ell_{1,a_{1}}^{\omega^{\prime}}(u)\\}.$ We remark that $\\{\ell^{\omega}_{1,1},\cdots,\ell^{\omega}_{1,a_{1}}\\}=I_{1}^{\omega}$ and so $\widehat{A}_{1}^{\omega}=A_{1}^{\omega}$. We put $\mathfrak{P}^{\prime}_{1}=\\{\omega^{\prime}\mid\mathop{\kern 0.0pt{\rm dim}}\nolimits\widehat{A}_{1}^{\omega^{\prime}}=\mathop{\kern 0.0pt{\rm dim}}\nolimits A_{1}^{\omega}\\}.$ It is easy to see that $\mathfrak{P}^{\prime}_{1}$ is a nonempty affine subset of $\mathfrak{K}$ and is defined over ${\mathbb{Q}}$. ###### Sublemma 13.6. If $\omega^{\prime}\in\mathfrak{P}^{\prime}_{1}$ and is sufficiently close to $\omega$, then $P_{1}^{\omega^{\prime}}$ is an equi-dimensional polyhedron in $\widehat{A}_{1}^{\omega^{\prime}}$. In particular $\widehat{A}_{1}^{\omega^{\prime}}=A_{1}^{\omega^{\prime}}$. ###### Proof. The tangent space of $\widehat{A}_{1}^{\omega^{\prime}}$ is parallel to the tangent space of $A_{1}^{\omega}$. Therefore $\ell^{\omega^{\prime}}_{1,j}$ is constant on $\widehat{A}_{1}^{\omega^{\prime}}$. We put $\widehat{S}_{1}^{\omega^{\prime}}=\ell^{\omega^{\prime}}_{1,1}(u)$ for some $u\in\widehat{A}_{1}^{\omega^{\prime}}$. On the other hand, if $\ell_{i}^{\omega}\notin I_{1}^{\omega}$ then $\ell_{i}^{\omega}(u)>S_{1}^{\omega}$ on $\text{Int}\,P_{1}^{\omega}$. Therefore if $\omega^{\prime}$ is sufficiently close to $\omega$ we have $\ell_{i}^{\omega^{\prime}}(u)>\widehat{S}_{1}^{\omega^{\prime}}$ on a neighborhood of a compact subset of $\text{Int}\,\,P_{1}^{\omega}$, which we identify with a subset of $P^{\prime}$. This implies the sublemma. ∎ The Condition (1), (2) of Lemma 13.3 in the case $k=1$ follows from Sublemma 13.6 easily. Let us assume that Lemma 13.4 is proved up to $k-1$. We remark $\\{\ell^{\omega}_{k,1},\cdots,\ell^{\omega}_{k,a_{k}}\\}=I_{k}^{\omega}$. We put $\widehat{A}_{k}^{\omega^{\prime}}=\\{u\in A_{k-1}^{\omega^{\prime}}\mid\ell_{k,1}^{\omega^{\prime}}(u)=\cdots=\ell_{k,a_{k}}^{\omega^{\prime}}(u)\\}$ and $\mathfrak{P}^{\prime}_{k}=\\{\omega^{\prime}\in\mathfrak{P}^{\prime}_{k-1}\mid\mathop{\kern 0.0pt{\rm dim}}\nolimits\widehat{A}_{k}^{\omega^{\prime}}=\mathop{\kern 0.0pt{\rm dim}}\nolimits A_{k}^{\omega}\\}.$ $\mathfrak{P}^{\prime}_{k}$ is a nonempty affine subset of $\mathfrak{K}$ and is defined over ${\mathbb{Q}}$. We can show that a sufficiently small open neighborhood $\mathfrak{P}_{k}$ of $\omega$ in $\mathfrak{P}^{\prime}_{k}$ has the required properties in the same way as the first step of the induction. The proof of Lemma 13.4 is complete. ∎ Proposition 13.3 follows immediately from Lemma 13.4. In fact the set of rational points are dense in $\mathfrak{P}_{K}$. ∎ Proposition 13.2 follows from Proposition 13.3, Proposition 4.7 and Theorem 4.10. ∎ The proof of Theorem 1.5 is now complete. ∎ ###### Proof of Proposition 10.8. The proof is similar to the proof of Proposition 13.3. Let $I_{k}$ be as in (10.2). We write it as $I_{k}(P,u_{0})$, where $P$ is the moment polytope of $(X,\omega)$. We define $I_{k}(P^{\prime},u^{\prime}_{0})$ as follows. Let $P^{\prime}$ be a polytope which is a small perturbation of $P$ and such that each of the faces are parallel to the corresponding face of $P$. Let $u^{\prime}_{0}\in\text{\rm Int}\,\,P^{\prime}$. Let us consider the set $\mathfrak{K}^{+}$ of all such pairs $(P^{\prime},u^{\prime}_{0})$. It is an open set of an affine space defined over ${\mathbb{Q}}$. Each of such $P^{\prime}$ is a moment polytope of certain Kähler form on $X$. (We remark that Kähler form on $X$ determine $P^{\prime}$ only up to translation.) For each $P^{\prime}$, we take corresponding Kähler form on $X$ and it determines a potential function. Therefore $I_{k}(P^{\prime},u^{\prime}_{0})$ is determined by (10.2). We define $A^{\perp}_{l}(P^{\prime},u^{\prime}_{0})$ in the same way as Definition 10.1. ###### Lemma 13.7. There exists a subset $\mathfrak{Q}_{k}$ of $\mathfrak{K}^{+}$ which is a nonempty open set of an affine subspace defined over ${\mathbb{Q}}$. All the elements $(P^{\prime},u^{\prime}_{0})$ of $\mathfrak{Q}_{k}$ have the following properties. 1. (1) $\mathop{\kern 0.0pt{\rm dim}}\nolimits A^{\perp}_{l}(P^{\prime},u^{\prime}_{0})=\mathop{\kern 0.0pt{\rm dim}}\nolimits A^{\perp}_{l}(P,u_{0})$ for $l\leq k$. 2. (2) $I_{l}(P^{\prime},u^{\prime}_{0})=I_{l}(P,u_{0})$ for $l\leq k$. The proof is the same as the proof of Lemma 13.4 and is omitted. Now we take a sequence of rational points $(P_{h},u^{h}_{0})\in\mathfrak{Q}_{k}$ converging to $(P,u_{0})$. Lemma 13.7 (2) implies that the leading term equation at $u^{h}_{0}$ is the same as the leading term equation at $u_{0}$. The proof of Proposition 10.8 is complete. ∎ ###### Proof of Lemma 10.5. Let $[\omega]\in H^{2}(X;{\mathbb{Q}})$. We may take the moment polytope $P$ such that its vertices are rational. (This time we do not change $P$ throughout the proof.) Let $u_{0}\in\text{\rm Int}\,P$ and we assume that $\mathfrak{PO}_{0}^{u_{0}}$ has a nondegenerate critical point in $(\Lambda_{0}\setminus\Lambda_{+})^{n}$. We define $I_{k}(P,u)$ as above. In the same way as the proof of Lemma 13.7 we can prove the following. ###### Sublemma 13.8. The set $\mathcal{P}_{k}$ of all $u\in\text{\rm Int}\,P$ satisfying the following conditions $(1)$,$(2)$ contains an open neighborhoods $u_{0}$ in certain affine subspace $A$ of ${\mathbb{R}}^{n}$. $A$ is defined on ${\mathbb{Q}}$. 1. (1) $\mathop{\kern 0.0pt{\rm dim}}\nolimits A^{\perp}_{l}(P,u)=\mathop{\kern 0.0pt{\rm dim}}\nolimits A^{\perp}_{l}(P,u_{0})$ for all $l<k$. 2. (2) $I_{l}(P,u)=I_{l}(P,u_{0})$ for all $l<k$. We omit the proof. We take $K$ such that $\\{d\ell_{i}\mid\ell_{i}\in I_{1}(P,u_{0})\cup\cdots\cup I_{K}(P,u_{0})\\}$ generates $N_{{\mathbb{R}}}$. (Note $P\subset N_{{\mathbb{R}}}=M_{{\mathbb{R}}}^{*}$.) By Sublemma 13.8 there exists a sequence $\\{u_{i}\\}_{i=1,2,\cdots}$ of rational points $u_{i}$ in $\mathcal{P}_{K}$ which converges to $u_{0}$. (1), (2) above implies $\mathfrak{PO}_{0}^{u_{0}}$ and $\mathfrak{PO}_{0}^{u_{i}}$ has the same leading term equation. Therefore by assumption $\mathfrak{PO}_{0}^{u_{i}}$ has a critical point on $(\Lambda_{0}\setminus\Lambda_{+})^{n}$. Since $Jac(\mathfrak{PO}_{0};\Lambda)$ is finite dimensional, it follows that we may take a subsequence $u_{k_{i}}$ such that $u_{k_{1}}=u_{k_{2}}=\cdots$. Hence $u_{0}=u_{k_{i}}$, is rational as required. ∎ ###### Remark 13.9. We can replace Definition 4.11 (3) by $HF((L(u_{i}),\mathfrak{x}_{i,{\mathcal{N}}}),(L(u_{i}),\mathfrak{x}_{i,{\mathcal{N}}});\Lambda^{{\mathbb{C}}}/(T^{{\mathcal{N}}}))\supseteq\Lambda^{{\mathbb{C}}}/(T^{{\mathcal{N}}}).$ In fact the following three conditions are equivalent to one another : 1. (1) $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x});\Lambda^{{\mathbb{C}}}/(T^{{\mathcal{N}}}))\cong H(T^{n};{\mathbb{C}})\otimes\Lambda^{{\mathbb{C}}}/(T^{{\mathcal{N}}}).$ 2. (2) $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x});\Lambda^{{\mathbb{C}}}/(T^{{\mathcal{N}}}))\supseteq\Lambda^{{\mathbb{C}}}/(T^{{\mathcal{N}}}).$ 3. (3) $\frac{\partial\mathfrak{PO}^{u}}{\partial y_{k}}\equiv 0,\mod T^{{\mathcal{N}}}\qquad k=1,\cdots,n,$ at $\mathfrak{x}$. (1) $\Rightarrow$ (2) is obvious. (3) $\Rightarrow$ (1) is Theorem 4.10. Let us prove (2) $\Rightarrow$ (3). Suppose (3) does not hold. We put $\frac{\partial\mathfrak{PO}^{u}}{\partial y_{k}}\equiv cT^{\lambda}\mod T^{\lambda}\Lambda_{+}$, where $c\in{\mathbb{C}}\setminus\\{0\\}$ and $0\leq\lambda<N$. Then (13.2) implies $T^{{\mathcal{N}}-\lambda}PD[L(u)]=0$ in $HF((L(u),\mathfrak{x}),(L(u),\mathfrak{x});\Lambda^{{\mathbb{C}}}/(T^{{\mathcal{N}}}))$. Since $PD[L(u)]$ is a unit, (2) does not hold. ###### Remark 13.10. The proof of Theorem 4.10 (or of Lemma 13.1) does not imply $\mathfrak{m}_{k,\beta}(\rho_{1},\cdots,\rho_{k})=0$ (13.5) for $\mu(\beta)\geq 4$. So it does not imply that the numbers $c_{\beta}$ (Definition 11.6) determine the filtered $A_{\infty}$ algebra $(H(L(u);\Lambda_{0}),\mathfrak{m})$ up to homotopy equivalence. We however believe that this is indeed the case. In fact the homology group $H(\mathcal{L}(T^{n});{\mathbb{Q}})$ of the free loop space $\mathcal{L}(T^{n})$ is trivial of degree $>n$. On the other hand, $\mathop{\kern 0.0pt{\rm dim}}\nolimits\mathcal{M}_{1}^{\text{\rm main}}(L(u_{0});\beta)=n+\mu(\beta)-2$. Hence if $\mu(\beta)\geq 4$ there is no nonzero homology class on the corresponding degree in the free loop space. Using the argument of [Fu4] it may imply that the contribution of those classes to the homotopy type of filtered $A_{\infty}$ structure is automatically determined from the contribution of the classes of Maslov index $2$. On the other hand, pseudo-holomorphic disc with Maslov index $\geq 4$ certainly contributes to the operator $\mathfrak{q}_{\ell,k,\beta}$ introduced in section 3.8 [FOOO3] (= section 13 [FOOO2]) : Namely $\mathfrak{q}_{\ell,k,\beta}$ is the operator that involves a cohomology class of the ambient symplectic manifold $X$. (See Remark 6.15) It seems that tropical geometry will play a role in this calculation. ## 14\. Appendix 1 : Algebraically closedness of Novikov fields ###### Lemma 14.1. If $F$ is an algebraically closed field of characteristic $0$, then $\Lambda^{F}$ is algebraically closed. ###### Proof. Let $\sum_{k=0}^{n}a_{k}x^{k}=0$ be a polynomial equation with $\Lambda^{F}$ coefficients. We will prove that it has a solution in $\Lambda^{F}$ by induction on $n$. We may assume that $a_{n}=1$. Replacing $x$ by $x-\frac{a_{n-1}}{n}$ we may assume $a_{n-1}=0$. (Here we use the fact that the characteristic of $F$ is $0$.) We may assume furthermore $a_{0}\neq 0$, since otherwise $0$ is a solution. We put $c=\inf_{k=0,\cdots,n-2}\frac{\mathfrak{v}_{T}(a_{k})}{n-k}.$ We put $x=T^{c}y$, $b_{k}=T^{c(k-n)}a_{k}$. Then our equation is equivalent to $P(y)=\sum_{k=0}^{n}b_{k}y^{k}=0$. We remark that $b_{n}=1$, $b_{n-1}=0$, $b_{0}\neq 0$. Moreover $\mathfrak{v}_{T}(b_{k})=c(k-n)+\mathfrak{v}_{T}(a_{k})\geq 0.$ Namely $b_{k}\in\Lambda_{0}$. We define $\overline{b}_{k}\in F$ to be the zero order term of $b_{k}$ i.e., to satisfy $b_{k}\equiv\overline{b}_{k}\mod\Lambda_{+}$. We consider the equation $\overline{P}(\overline{y})=\sum_{k=0}^{n}\overline{b}_{k}\overline{y}^{k}=0.$ By our choice of $c$ there exists $k<n-1$ such that $\overline{b}_{k}\neq 0$ and $\overline{b}_{n}=1$, $\overline{b}_{n-1}=0$. Therefore $\overline{P}$ has at least two roots. (We use the fact that the characteristic of $F$ is 0 here.) Since $F$ is algebraically closed we can decompose $\overline{P}=\overline{Q}\overline{R}$ where $\overline{Q}$ and $\overline{R}$ are monic, of nonzero degree and coprime. Therefore by Hensel’s lemma, there exists $Q,R\in\Lambda_{0}[y]$ such that $P=QR$ and $\deg Q=\deg\overline{Q}$, $Q\equiv\overline{Q}\mod\Lambda_{+}$, $R\equiv\overline{R}\mod\Lambda_{+}$333A proof of Hensel’s lemma, in the case when valuation is not necessarily discrete, is given, for example, in p 144 [BGR]. See also the proof of Lemma 8.5 given below.. Since the degree of $Q$ is smaller than the degree of $P$, it follows from induction hypothesis that $Q$ has a root in $\Lambda^{F}$. The proof of Lemma 14.1 is now complete. ∎ By a similar argument we can show that if $F$ is characteristic $0$ then a finite algebraic extension of $\Lambda^{F}$ is contained in $\Lambda^{K}$ for some finite extension $K$ of $F$. (We used this fact in section 7.) ###### Proof of Lemma 8.5. In view of the proof of Lemma 14.1, it suffices to show that $\Lambda_{0}^{conv}$ is Henselian. (Namely Hensel’s lemma holds for it.) Let $P(X)=\sum_{i=0}^{n-1}a_{i}X^{i}+X^{n}\in\Lambda_{0}^{conv}[X].$ We assume that its ${\mathbb{C}}$-reduction $\overline{P}\in{\mathbb{C}}[X]$ is decomposed as $\overline{P}=\overline{Q}\overline{R}$, where $\overline{Q}$ and $\overline{R}$ are monic and coprime. We put $a_{i}=a_{i,0}+a_{i,+},\qquad a_{i,0}\in{\mathbb{C}},\quad a_{i,+}\in\Lambda_{+}^{conv},$ and $\widetilde{P}(X)=\sum_{i=0}^{n-1}(a_{i,0}+Z_{i})X^{i}+X^{n}\in{\mathbb{C}}[Z_{0},\cdots,Z_{n-1}][X],$ where $Z_{i}$ are formal variables. The convergent power series ring ${\mathbb{C}}\\{Z_{0},\cdots,Z_{n-1}\\}$ is Henselian (see section 45 [Na]). Therefore there exist monic polynomials $\widetilde{Q},\widetilde{R}\in{\mathbb{C}}\\{Z_{0},\cdots,Z_{n-1}\\}[X]$ such that $\widetilde{Q}\widetilde{R}=\widetilde{P}$ and the ${\mathbb{C}}$-reduction of $\widetilde{Q}$, $\widetilde{R}$ are $\overline{Q}$, $\overline{R}$ respectively. On the other hand, it is easy to see that $Z_{i}\mapsto a_{i,+}$ induces a continuous ring homomorphism ${\mathbb{C}}\\{Z_{0},\cdots,Z_{n-1}\\}\to\Lambda_{0}^{conv}$. Thus $\widetilde{Q}$, $\widetilde{R}$ induce $Q,R\in\Lambda_{0}^{conv}[X]$ such that $QR=P$. Hence $\Lambda_{0}^{conv}$ is Henselian, as required. ∎ ## 15\. Appendix 2 : $T^{n}$-equivariant Kuranishi structure In this section we define the notion of $T^{n}$ equivariant Kuranishi structure and prove that our moduli space $\mathcal{M}_{k+1}^{\text{\rm main}}(\beta)$ has one. We also show the existence of $T^{n}$ equivariant perturbation of the Kuranishi map. Let $\mathcal{M}$ be a compact space with Kuranishi structure. The space ${\mathcal{M}}$ is covered by a finite number of Kuranishi charts $(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})$, $\alpha\in\mathfrak{A}$ which satisfy the following : ###### Condition 15.1. 1. (1) $V_{\alpha}$ is a smooth manifold (with boundaries or corners) and $\Gamma_{\alpha}$ is a finite group acting effectively on $V_{\alpha}$. 2. (2) $\text{pr}_{\alpha}:E_{\alpha}\to V_{\alpha}$ is a finite dimensional vector bundle on which $\Gamma_{\alpha}$ acts so that $\text{pr}_{\alpha}$ is $\Gamma_{\alpha}$\- equivariant. 3. (3) $s_{\alpha}$ is a $\Gamma_{\alpha}$ equivariant section of $E_{\alpha}$. 4. (4) $\psi_{\alpha}:s_{\alpha}^{-1}(0)/\Gamma_{\alpha}\to\mathcal{M}$ is a homeomorphism to its image. 5. (5) The union of $\psi_{\alpha}(s_{\alpha}^{-1}(0)/\Gamma_{\alpha})$ for various $\alpha$ is $\mathcal{M}$. We assume that $\\{(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})\mid\alpha\in\mathfrak{A}\\}$ is a good coordinate system, in the sense of Definition 6.1 [FO] or Lemma A1.11 [FOOO3] (= Lemma A1.11 [FOOO2]). This means the following : The set $\mathfrak{A}$ has a partial order $<$, where either $\alpha_{1}\leq\alpha_{2}$ or $\alpha_{2}\leq\alpha_{1}$ holds for $\alpha_{1},\alpha_{2}\in\mathfrak{A}$ if $\psi_{\alpha_{1}}(s_{\alpha_{1}}^{-1}(0)/\Gamma_{\alpha_{1}})\cap\psi_{\alpha_{2}}(s_{\alpha_{2}}^{-1}(0)/\Gamma_{\alpha_{2}})\neq\emptyset.$ Let $\alpha_{1},\alpha_{2}\in\mathfrak{A}$ and $\alpha_{1}\leq\alpha_{2}$. Then, there exists a $\Gamma_{\alpha_{1}}$-invariant open subset $V_{\alpha_{2},\alpha_{1}}\subset V_{\alpha_{1}}$, a smooth embedding $\varphi_{\alpha_{2},\alpha_{1}}:V_{\alpha_{2},\alpha_{1}}\to V_{\alpha_{2}}$ and a bundle map $\widehat{\varphi}_{\alpha_{2},\alpha_{1}}:E_{\alpha_{1}}|_{V_{\alpha_{2},\alpha_{1}}}\to E_{\alpha_{2}}.$ which covers $\varphi_{\alpha_{2},\alpha_{1}}$. Moreover there exists an injective homomorphism $\widehat{\widehat{\varphi}}_{\alpha_{2},\alpha_{1}}:\Gamma_{\alpha_{1}}\to\Gamma_{\alpha_{2}}.$ We require that they satisfy the following ###### Condition 15.2. 1. (1) The maps $\varphi_{\alpha_{2},\alpha_{1}}$, ${\widehat{\varphi}}_{\alpha_{2},\alpha_{1}}$ are $\widehat{\widehat{\varphi}}_{\alpha_{2},\alpha_{1}}$-equivariant. 2. (2) $\varphi_{\alpha_{2},\alpha_{1}}$ and $\widehat{\widehat{\varphi}}_{\alpha_{2},\alpha_{1}}$ induce an embedding of orbifold $\overline{\varphi}_{\alpha_{2},\alpha_{1}}:\frac{V_{\alpha_{2},\alpha_{1}}}{\Gamma_{\alpha_{1}}}\to\frac{V_{\alpha_{2}}}{\Gamma_{\alpha_{2}}}.$ (15.1) 3. (3) We have $s_{\alpha_{2}}\circ\varphi_{\alpha_{2},\alpha_{1}}={\widehat{\varphi}}_{\alpha_{2},\alpha_{1}}\circ s_{\alpha_{1}}.$ 4. (4) We have $\psi_{\alpha_{2}}\circ\overline{\varphi}_{\alpha_{2},\alpha_{1}}=\psi_{\alpha_{1}}$ on $\frac{V_{\alpha_{2},\alpha_{1}}\cap s_{\alpha_{1}}^{-1}(0)}{\Gamma_{\alpha_{1}}}.$ 5. (5) If $\alpha_{1}<\alpha_{2}<\alpha_{3}$ then $\varphi_{\alpha_{3},\alpha_{2}}\circ\varphi_{\alpha_{2},\alpha_{1}}=\varphi_{\alpha_{3},\alpha_{1}},$ on $\varphi_{\alpha_{2},\alpha_{1}}^{-1}(V_{\alpha_{3},\alpha_{2}})$. ${\widehat{\varphi}}_{\alpha_{3},\alpha_{2}}\circ{\widehat{\varphi}}_{\alpha_{2},\alpha_{1}}={\widehat{\varphi}}_{\alpha_{3},\alpha_{1}},$ and $\widehat{\widehat{\varphi}}_{\alpha_{3},\alpha_{2}}\circ\widehat{\widehat{\varphi}}_{\alpha_{2},\alpha_{1}}=\widehat{\widehat{\varphi}}_{\alpha_{3},\alpha_{1}},$ hold in the similar sense. 6. (6) $V_{\alpha_{2},\alpha_{1}}/\Gamma_{\alpha_{1}}$ contains $\psi_{\alpha_{1}}^{-1}(\psi_{\alpha_{1}}(s_{\alpha_{1}}^{-1}(0)/\Gamma_{\alpha_{1}})\cap\psi_{\alpha_{2}}(s_{\alpha_{2}}^{-1}(0)/\Gamma_{\alpha_{2}}))$. ###### Condition 15.3. $\mathcal{M}$ has a tangent bundle : i.e., the differential of $s_{\alpha_{2}}$ in the direction of the normal bundle induces a bundle isomorphism $ds_{\alpha_{2}}:\frac{{\varphi}_{\alpha_{2},\alpha_{1}}^{*}TV_{\alpha_{2}}}{TV_{\alpha_{2},\alpha_{1}}}\to\frac{\widehat{\varphi}_{\alpha_{2},\alpha_{1}}^{*}E_{\alpha_{2}}}{E_{\alpha_{1}}}.$ We say $\mathcal{M}$ is oriented if $V_{\alpha}$, $E_{\alpha}$ is oriented, the $\Gamma_{\alpha}$ action is orientation preserving, and $ds_{\alpha}$ is orientation preserving. ###### Definition 15.4. Suppose that $\mathcal{M}$ has a $T^{n}$ action. We say our Kuranishi structure on $\mathcal{M}$ is $T^{n}$ equivariant in the strong sense if the following holds : 1. (1) $V_{\alpha}$ has a $T^{n}$ action which commutes with the given $\Gamma_{\alpha}$ action. 2. (2) $E_{\alpha}$ is a $T^{n}$ equivariant bundle. 3. (3) The Kuranishi map $s_{\alpha}$ is $T^{n}$ equivariant and $\psi_{\alpha}$ is a $T^{n}$ equivariant map. 4. (4) The coordinate changes $\varphi_{\alpha_{2},\alpha_{1}}$ and $\varphi_{\alpha_{2},\alpha_{1}}$ are $T^{n}$ equivariant. ###### Remark 15.5. We remark that Condition (1) above is more restrictive than the condition that the orbifold $V_{\alpha}/\Gamma_{\alpha}$ has a $T^{n}$ action. This is the reason why we use the phrase in the strong sense in the above definition. Hereafter we will say $T^{n}$ equivariant instead for simplicity. Let $L$ be a smooth manifold. A strongly continuous smooth map $ev:\mathcal{M}\to L$ is a family of $\Gamma_{\alpha}$ invariant smooth maps $ev_{\alpha}:V_{\alpha}\to L$ (15.2) which induce $\overline{ev}_{\alpha}:V_{\alpha}/\Gamma_{\alpha}\to L$ such that $\overline{ev}_{\alpha_{2}}\circ\overline{\varphi}_{\alpha_{2},\alpha_{1}}=\overline{ev}_{\alpha_{1}}$ on $V_{\alpha_{2},\alpha_{1}}/\Gamma_{\alpha}$. (Note $\Gamma_{\alpha}$ action on $L$ is trivial.) We say that $ev$ is weakly submersive if each of ${ev}_{\alpha}$ in (15.2) is a submersion. ###### Definition 15.6. Assume that there exists $T^{n}$ actions on $L$ and $\mathcal{M}$. We say that $ev:\mathcal{M}\to L$ is $T^{n}$ equivariant if ${ev}_{\alpha}$ in (15.2) is $T^{n}$ equivariant. Now we show ###### Proposition 15.7. The moduli space $\mathcal{M}_{k+1}(\beta)$ has a $T^{n}$ equivariant Kuranishi structure such that $ev_{0}:\mathcal{M}_{k+1}(\beta)\to L$ is $T^{n}$ equivariant strongly continuous weakly submersive map. ###### Proof. Except the $T^{n}$ equivariance this is proved in section 7.1 [FOOO3] (= section 29 [FOOO2]). Below we explain how we choose our Kuranishi structure so that it is $T^{n}$ equivariant. We also include the case of the moduli space $\mathcal{M}_{k+1,\ell}(\beta)$ with interior marked points. We first review the construction of the Kuranishi neighborhood from section 7.1 [FOOO3] (= section 29 [FOOO2]). Let $\text{\bf x}=((\Sigma,\vec{z}),w)\in\mathcal{M}_{k+1,\ell}(\beta)$. Let $\Sigma_{a}$ be an irreducible component of $\Sigma$. We consider the operator $D_{w,a}\overline{\partial}:W^{1,p}(\Sigma_{a};w^{*}(TX);L,\vec{z}_{a})\to W^{0,p}(\Sigma_{a};w^{*}(TX)\otimes\Lambda^{0,1}).$ (15.3) Here $W^{1,p}(\Sigma_{a};w^{*}(TX);L,\vec{z}_{a})$ is the space of section $v$ of $w^{*}(TX)$ of $W^{1,p}$ class with the following properties. 1. (1) The restriction of $v$ to $\partial\Sigma_{a}$ is in $w^{*}(TL)$. 2. (2) $\vec{z}_{a}$ is the set of $\Sigma$ which is either singular or marked. We assume that $v$ is $0$ at those points. $\Lambda^{0,1}$ is the bundle of $(0,1)$ forms on $\Sigma_{a}$ and $W^{0,p}(\Sigma_{a};w^{*}(TX)\otimes\Lambda^{0,1})$ is the set of sections of $W^{0,p}$ class of $w^{*}(TX)\otimes\Lambda^{0,1}$. $D_{w,a}\overline{\partial}$ is the linearization of (nonlinear) Cauchy- Riemann equation. (See [Fl].) The operator (15.3) is Fredholm by ellipticity thereof. We choose open subsets $W_{a}$ of $\Sigma_{a}$ whose closure is disjoint from the boundary of each of the irreducible component $\Sigma_{a}$ of $\Sigma$ and from the singular points and marked points. By the unique continuation theorem, we can choose a finite dimensional subset $E_{0,a}$ of $C_{0}^{\infty}(W_{a};w^{*}TX)$ (the set of smooth sections of compact support in $W_{a}$) such that $\text{\rm Im}D_{w,a}\overline{\partial}+E_{0,a}=W^{0,p}(\Sigma_{a};w^{*}(TX)\otimes\Lambda^{0,1}).$ When x has nontrivial automorphisms, we choose $\bigoplus_{a}E_{0,a}$ to be invariant under the automorphisms. We next associate a finite dimensional subspace $E_{0,a}((\Sigma,\vec{z}),w^{\prime})$ for $w^{\prime}$ which is ‘$C^{0}$ close to $w$.’ We need some care to handle the case where some component $(\Sigma_{a},\vec{z}_{a})$ is not stable, that is the case for which the automorphism group $G_{a}$ of $(\Sigma_{a},\vec{z}_{a})$ is of positive dimension. (Note $G_{a}$ is different from the automorphism group of $((\Sigma_{a},\vec{z}_{a}),w|_{\Sigma_{a}})$. The latter group is finite.) We explain this choice of $E_{0,1}$ below following Appendix [FO]. For each unstable components $\Sigma_{a}$, pick points $z_{a,i}\in\Sigma$, $(i=1,\cdots,d_{a})$ with the following properties. ###### Condition 15.8. 1. (1) $w$ is immersed at $z_{a,i}$. 2. (2) $z_{a,i}$ in the interior of $\Sigma_{a}$, $z_{a,i}\neq z_{a,j}$ for $i\neq j$ and $z_{a,i}\notin\vec{z}$. 3. (3) $(\Sigma_{a};(\vec{z}\cap\Sigma_{a})\cup(z_{a,1},\cdots,z_{a,d_{a}}))$ is stable. 4. (4) Let $\Gamma=\Gamma(\text{\bf x})$ be the group of automorphisms of $\text{\bf x}=((\Sigma,\vec{z}),w)$. If $\gamma\in\Gamma$, $\gamma(\Sigma_{a})=\Sigma_{a^{\prime}}$ then $d_{a}=d_{a^{\prime}}$ and $\\{\gamma(z_{a,i})\mid i=1,\cdots,d_{a}\\}=\\{z_{a^{\prime},i^{\prime}}\mid i^{\prime}=1,\cdots,d_{a}\\}$. For each $a,i$ we choose a submanifold $N_{a,i}$ of codimension $2$ in $X$ that transversely intersects with $(\Sigma_{a},w)$ at $w_{a}(z_{a,i})$. In relation to Condition 15.8 (4) above we choose $N_{a,i}=N_{a^{\prime},i^{\prime}}$ if $\gamma(z_{a,i})=z_{a^{\prime},i^{\prime}}$. We add extra interior marked points $\\{z_{a,i}\mid a,i\\}$ in addition to $\vec{z}$ on $(\Sigma,\vec{z})$, and obtain a stable curve $(\Sigma,\vec{z}^{+})$. (Namely $\vec{z}^{+}=\vec{z}\sqcup\\{z_{a,i}\mid a,i\\}$.) (We choose an order of the added marked points and fix it.) We consider a neighborhood $\mathfrak{U}_{0}$ of $[\Sigma,\vec{z}^{+}]$ in $\mathcal{M}_{k+1,\ell^{\prime}}^{\text{\rm main}}$, that is the moduli space of bordered stable curve of genus $0$ with $k+1$ boundary and $\ell^{\prime}$ interior marked points. Let $\Gamma_{0}$ be the group of automorphisms of $((\Sigma,\vec{z}^{+}),w)$. Now both $\Gamma$ and $\Gamma_{0}$ are finite groups and $\Gamma\supseteq\Gamma_{0}$. An element $\gamma\in\Gamma$ induces an automorphism $\gamma:\Sigma\to\Sigma$ which fixes marked points in $\vec{z}$ and permutes $\ell-\ell^{\prime}$ marked points $\\{z_{a,i}\mid a,i\\}$, by Condition 15.8 (4). Therefore we obtain an element of $[\gamma_{*}(\Sigma,\vec{z}^{+})]$ that is different from $[\Sigma,\vec{z}^{+}]$ only by reordering of marked points. We take the union of neighborhoods of $[\gamma_{*}(\Sigma,\vec{z}^{+})]$ for various $\gamma\in\Gamma$ in $\mathcal{M}_{k+1,\ell^{\prime}}^{\text{\rm main}}$ and denote it by $\mathfrak{U}$. $\mathfrak{U}_{0}$ is written as $\mathfrak{V}_{0}/\Gamma_{0}$ where $\mathfrak{V}_{0}$ is a manifold. Moreover there exists a manifold $\mathfrak{V}$ on which $\Gamma$ acts such that $\mathfrak{V}/\Gamma_{0}=\mathfrak{U}$, $\mathfrak{V}/\Gamma\cong\mathfrak{U}_{0}$. Furthermore there is a ‘universal family’ $\mathfrak{M}\to\mathfrak{V}$ where the fiber $\Sigma(\text{\bf v})$ of $\text{\bf v}\in\mathfrak{V}$ is identified with the bordered marked stable curve that represents the element $[\text{\bf v}]\in\mathfrak{V}/\Gamma\subset\mathcal{M}_{k+1,\ell^{\prime}}^{\text{\rm main}}$. There is a $\Gamma$ action on $\mathfrak{M}$ such that $\mathfrak{M}\to\mathfrak{V}$ is $\Gamma$ equivariant. By the construction, each member $\Sigma(\text{\bf v})$ of our universal family is diffeomorphic to $\Sigma$ away from singularity. More precisely we have the following : Let $\Sigma_{0}=\Sigma\setminus S$ where $S$ is a small neighborhood of the union of the singular point sets and the marked point sets. Then for each v there exists a smooth embedding $i_{\text{\bf v}}:\Sigma_{0}\to\Sigma(\text{\bf v})$. The error of this embedding for becoming a biholomorphic map goes to $0$ as v goes to $0$. We may assume $\text{\bf v}\mapsto i_{\text{\bf v}}$ is $\Gamma$ invariant in an obvious sense and $i_{\text{\bf v}}$ depends smoothly on v. We may choose $W_{a}$ so that $W_{a}\subset\Sigma_{0}$ for each $a$. We moreover assume that $\bigoplus_{a}E_{0,a}$ is invariant under the $\Gamma$ action in the following sense : if $\gamma\in\Gamma$ and $\Sigma_{a^{\prime}}=\gamma(\Sigma_{a})$ then the isomorphism induced by $\gamma$ sends $E_{0,a}$ to $E_{0,a^{\prime}}$. Now we consider a pair $(w^{\prime},\text{\bf v})$ where $w^{\prime}:(\Sigma(\text{\bf v}),\partial\Sigma(\text{\bf v}))\to(X,L).$ We assume : ###### Condition 15.9. There exists $\epsilon>0$ depending only on x such that the following holds. 1. (1) $\sup_{x\in\Sigma_{0}}\text{\rm dist}(w^{\prime}(i_{\text{\bf v}}(x)),w(x))\leq\epsilon$. 2. (2) For any connected component $D_{c}$ of $\Sigma(\text{\bf v})\setminus\text{\rm Im}(i_{\text{\bf v}})$, the diameter of $w^{\prime}(D_{c})$ in $X$ (with a fixed Riemannian metric on it) is smaller than $\epsilon$. For each point $x\in W_{a}$ we use the parallel transport to make the identification $T_{w(x)}X\otimes\Lambda^{0,1}_{x}(\Sigma)\equiv T_{w^{\prime}(i_{\text{\bf v}}(x))}X\otimes\Lambda^{0,1}_{i_{\text{\bf v}}(x)}(\Sigma(\text{\bf v})).$ Using this identification we obtain an embedding $I_{(\text{\bf v},w^{\prime})}:\bigoplus_{a}E_{0,a}\longrightarrow w^{\prime*}(TX)\otimes\Lambda^{0,1}(\Sigma(\text{\bf v})).$ Via this embedding $I_{(\text{\bf v},w^{\prime})}$, we consider the equation $\overline{\partial}w^{\prime}\equiv 0\mod\bigoplus_{a}I_{(\text{\bf v},w^{\prime})}(E_{0,a}),$ (15.4) together with the additional conditions $w^{\prime}(z_{a,i})\in N_{a,i}.$ (15.5) Let $V(\text{\bf x})$ be the set of solution of (15.4), (15.5). This is a smooth manifold (with boundary and corners) by the implicit function theorem and a gluing argument. (See [FOOO3] section A1.4 = [FOOO2] section A1.4 for the smoothness at boundary and corner.) Since we can make all the construction above invariant under the $\Gamma(\text{\bf x})$ action, the space $V(\text{\bf x})$ has a $\Gamma(\text{\bf x})$ action. (Note that we may choose $N_{a,i}$ so that $\\{N_{a,i}\mid a,i\\}$ is invariant under the action of $\Gamma(\text{\bf x})$.) The obstruction bundle $E$ is the space $\bigoplus_{a}E_{0,a}$ at x and $\bigoplus_{a}I_{(\text{\bf v},w^{\prime})}(E_{0,a})$ at in $(\text{\bf v},w^{\prime})$. We omit the construction of coordinate changes. See [FOOO3] section 7.1 (= [FOOO2] section 29) which in turn is similar to section 15 [FO]. The Kuranishi map is given by $((\Sigma^{\prime},\vec{z}^{\prime}),w^{\prime})\mapsto\overline{\partial}w^{\prime}\in\bigoplus_{a}E_{0,a}.$ (15.6) We have thus finished our review of the construction of Kuranishi charts. Now we assume that $X$ has a $T^{n}$ action, which preserves both the complex and the symplectic structures on $X$. We also assume that $L$ is a $T^{n}$ orbit. We want to construct a family of vector spaces $(\text{\bf v},w^{\prime})\mapsto\bigoplus_{a}I_{(\text{\bf v},w^{\prime})}(E_{0,a})$ so that it is invariant under the $T^{n}$ action. We need to slightly modify the above construction for this purpose. In fact it is not totally obvious to make the condition (15.5) $T^{n}$-invariant. For this purpose we proceed in the following way. We fix a point $p_{0}\in L$ and consider an element $\text{\bf x}\in ev_{0}^{-1}(p_{0})\cap\mathcal{M}_{k+1,\ell}(\beta).$ We are going to construct a $T^{n}$ equivariant Kuranishi neighborhood of the $T^{n}$ orbit of x. Let $\text{\bf v}\in\mathfrak{V}$ and $w^{\prime}:(\Sigma(\text{\bf v}),\partial\Sigma(\text{\bf v}))\to(X,L)$ be as before. We replace Condition 15.9 by the following. ###### Condition 15.10. Let $z_{0}$ be the 0-th boundary marked point and let $g\in T^{n}$ be the unique element satisfying $w^{\prime}(z_{0})=g(p_{0})$. There exists $\epsilon>0$ depending only on x such that the following holds. 1. (1) $\sup_{x\in\Sigma_{0}}\operatorname{dist}(w^{\prime}(i_{\text{\bf v}}(x)),g(w(x)))\leq\epsilon$. 2. (2) For any connected component $D_{c}$ of $\Sigma(\text{\bf v})\setminus\text{\rm Im}(i_{\text{\bf v}})$, the diameter of $w^{\prime}(D_{c})$ in $X$ (with a fixed Riemannian metric on it) is smaller than $\epsilon$. Now we define an embedding $I_{(\text{\bf v},w^{\prime})}:\bigoplus_{a}E_{0,a}\longrightarrow w^{\prime*}(TX)\otimes\Lambda^{0,1}(\Sigma(\text{\bf v}))$ as follows : We first use the $g$ action to define an isomorphism $g_{*}:T_{w(x)}X\otimes\Lambda^{0,1}_{x}(\Sigma)\cong T_{g(w(x))}X\otimes\Lambda^{0,1}_{x}(\Sigma).$ Then we use the parallel transport in the same way as before to define $\bigoplus_{a}g_{*}(E_{0,a})\longrightarrow w^{\prime*}(TX)\otimes\Lambda^{0,1}(\Sigma(\text{\bf v})).$ By composing the two we obtain the embedding $I_{(\text{\bf v},w^{\prime})}$. Now we consider the equation $\overline{\partial}w^{\prime}\equiv 0\mod\bigoplus_{a}I_{(\text{\bf v},w^{\prime})}(E_{0,a}),$ (15.7) together with the additional conditions $w^{\prime}(z_{a,i})\in g(N_{a,i})$ (15.8) as before. Clearly these equations are $T^{n}$ invariant. It is also $\Gamma(\text{\bf x})$ invariant. Note the action of the automorphism group $\Gamma(\text{\bf x})$ of x of our Kuranishi structure, which is a finite group, acts on the source while $T^{n}$ acts on the target. Therefore it is obvious that two actions commute. By definition the obstruction bundle has a $T^{n}$ action. Moreover (15.6) is $T^{n}$ equivariant. It is fairly obvious from the construction that coordinate changes of the constructed Kuranishi structure is also $T^{n}$ equivariant. The proof of Proposition 15.7 is now complete. ∎ ###### Remark 15.11. 1. (1) From the above construction, it is easy to see that our $T^{n}$ equivariant Kuranishi structure is compatible with the gluing at the boundary marked points. Namely under the embedding $\mathcal{M}_{k_{1}+1}(\beta_{1}){}_{ev_{0}}\times_{ev_{i}}\mathcal{M}_{k_{2}+1}(\beta_{2})\subset\mathcal{M}_{k_{1}+k_{2}}(\beta_{1}+\beta_{2})$ the restriction of the Kuranishi structure of the right hand side coincides with the fiber product of the Kuranishi structure of the left hand side. More precisely speaking, we can construct the system of our Kuranishi structures so that this statement holds : this Kuranishi structure is constructed inductively over the number of disc components and the energy of $\beta$. 2. (2) On the other hand, when we construct the $T^{n}$ invariant Kuranishi structure in the way we described above it may not be compatible with the gluing at the interior marked point. Namely by the embedding $\mathcal{M}_{1}(\alpha)\times_{X}\mathcal{M}_{k+1,1}(\beta)\subset\mathcal{M}_{k+1}(\beta+\alpha)$ (15.9) the restriction of the Kuranishi structure of the right hand sides may not coincide with the fiber product of the Kuranishi structure of the left hand side. This compatibility is not used in this paper and hence not required. See Remark 11.4 where similar point is discussed for the choice of multisections. However contrary to the choice of multisections mentioned in Remark 11.4, we remark that it is possible to construct $T^{n}$ equivariant Kuranishi structure compatible with (15.9). Since we do not use this point in the paper, we will not elaborate it here. We next review on the multisections. (See section 3 [FO].) Let $(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})$ be a Kuranishi chart of $\mathcal{M}$. For $x\in V_{\alpha}$ we consider the fiber $E_{\alpha,x}$ of the bundle $E_{\alpha}$ at $x$. We take its $l$ copies and consider the direct product $E_{\alpha,x}^{l}$. We take the quotient thereof by the action of symmetric group of order $l!$ and let $\mathcal{S}^{l}(E_{\alpha,x})$ be the quotient space. There exists a map $tm_{m}:\mathcal{S}^{l}(E_{\alpha,x})\to\mathcal{S}^{lm}(E_{\alpha,x}),$ which sends $[a_{1},\cdots,a_{l}]$ to $[\,\underbrace{a_{1},\cdots,a_{1}}_{\text{$m$ copies}},\cdots,\underbrace{a_{l},\cdots,a_{l}}_{\text{$m$ copies}}].$ A smooth multisection $s$ of the bundle $E_{\alpha}\to V_{\alpha}$ consists of an open covering $\bigcup_{i}U_{i}=V_{\alpha}$ and $s_{i}$ which maps $x\in U_{i}$ to $s_{i}(x)\in\mathcal{S}^{l_{i}}(E_{\alpha,x})$. They are required to have the following properties : ###### Condition 15.12. 1. (1) $U_{i}$ is $\Gamma_{\alpha}$-invariant. $s_{i}$ is $\Gamma_{\alpha}$-equivariant. (We remark that there exists an obvious map $\gamma:\mathcal{S}^{l_{i}}(E_{\alpha,x})\to\mathcal{S}^{l_{i}}(E_{\alpha,\gamma x})$ for each $\gamma\in\Gamma_{\alpha}$.) 2. (2) If $x\in U_{i}\cap U_{j}$ then we have $tm_{l_{j}}(s_{i}(x))=tm_{l_{i}}(s_{j}(x))\in\mathcal{S}^{l_{i}l_{j}}(E_{\alpha,\gamma x}).$ 3. (3) $s_{i}$ is liftable and smooth in the following sense. For each $x$ there exists a smooth section $\tilde{s}_{i}$ of $\underbrace{E_{\alpha}\oplus\cdots\oplus E_{\alpha}}_{\text{$l_{i}$ times}}$ in a neighborhood of $x$ such that $\tilde{s}_{i}(y)=(s_{i,1}(y),\cdots,s_{i,l_{i}}(y)),\quad s_{i}(y)=[s_{i,1}(y),\cdots,s_{i,l_{i}}(y)].$ (15.10) We identify two multisections $(\\{U_{i}\\},\\{s_{i}\\},\\{l_{i}\\})$, $(\\{U^{\prime}_{i}\\},\\{s^{\prime}_{i}\\},\\{l^{\prime}_{i}\\})$ if $tm_{l_{i}}(s_{i}(x))=tm_{l^{\prime}_{j}}(s^{\prime}_{j}(x))\in\mathcal{S}^{l_{i}l^{\prime}_{j}}(E_{\alpha,\gamma x})$ on $U_{i}\cap U^{\prime}_{j}$. We say $s_{i,j}$ to be a branch of $s_{i}$ in the situation of (15.10). We next prove the following lemma which we used in section 11. ###### Lemma 15.13. Let $\mathcal{M}$ have a $T^{n}$ action and a $T^{n}$ equivariant Kuranishi structure. Suppose that the $T^{n}$ action on each of the Kuranishi neighborhood is free. Then we can descend the Kuranishi structure to $\mathcal{M}/T^{n}$ in a canonical way. ###### Proof. Let $(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})$ be a Kuranishi chart. Since $T^{n}$ action on $V_{\alpha}$ is free $V_{\alpha}/T^{n}$ is a smooth manifold and $E_{\alpha}/T^{n}\to V_{\alpha}/T^{n}$ is a vector bundle. Since $\Gamma_{\alpha}$ action commutes with $T^{n}$ action, it follows that it acts on this vector bundle. The $T^{n}$ equivariance of $s_{\alpha}$ implies that we have a section $\overline{s}_{\alpha}$ of $E_{\alpha}/T^{n}\to V_{\alpha}/T^{n}$. Moreover since $\psi_{\alpha}:s_{\alpha}^{-1}(0)\to\mathcal{M}$ is $T^{n}$ equivariant it induces a map $\overline{s}_{\alpha}^{-1}(0)\to\mathcal{M}/T^{n}$. Thus we obtain a Kuranishi chart. It is easy to define the coordinate changes. ∎ ###### Definition 15.14. In the situation of Proposition 15.13, we say that a system of multisections of the Kuranishi structure of $\mathcal{M}$ is $T^{n}$ equivariant, if it is induced from the multisection of the Kuranishi structure on $\mathcal{M}/T^{n}$ in an obvious way. ###### Corollary 15.15. In the situation of Proposition 15.13, we assume that we have a $T^{n}$ equivariant multisection at the boundary of $\mathcal{M}$, which is transversal to $0$. Then it extends to a $T^{n}$ equivariant multisection of $\mathcal{M}$. This is an immediate consequence of Lemma 3.14 [FO], that is the non- equivariant version. ## 16\. Appendix 3 : Smooth correspondence via the zero set of multisection In this section we explain the way how we use the zero set of multisection to define smooth correspondence, when appropriate submersive properties are satisfied. Such a construction is a special case of the techniques of using a continuous family of multisections and integration along the fiber on their zero sets so that smooth correspondence by spaces with Kuranishi structure induces a map between de Rham complex. This (more general) technique is not new and is known to some experts. In fact [Ru], section 16 [Fu2] use a similar technique and section 7.5 [FOOO3] (= section 33 [FOOO2]), [Fu4], [Fu5] contain the details of this more general technique. We explain the special case (namely the case we use a single multisection) in our situation of toric manifold, for the sake of completeness and of reader’s convenience. Let $\mathcal{M}$ be a space with Kuranishi structure and $ev_{s}:\mathcal{M}\to L_{s}$, $ev_{t}:\mathcal{M}\to L_{t}$ be strongly continuous smooth maps. (Here $s$ and $t$ stand for source and target, respectively.) We assume our smooth manifolds $L_{s},L_{t}$ are compact and oriented without boundary. We also assume $\mathcal{M}$ has a tangent bundle and is oriented in the sense of Kuranishi structure. Suppose that $L_{t}$ has a free and transitive $T^{n}$ action, and $L_{s}$ and $\mathcal{M}$ have $T^{n}$ action. We assume that the Kuranishi structure on $\mathcal{M}$ is $T^{n}$ equivariant and the maps $ev_{s}$, $ev_{t}$ are $T^{n}$ equivariant. Let $\\{(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})\\}$ be a $T^{n}$ equivariant Kuranishi coordinate system (good coordinate system) of $\mathcal{M}$. We use Corollary 15.15 to find a $T^{n}$ equivariant (system) of multisection $\mathfrak{s}_{\alpha}:V_{\alpha}\to E_{\alpha}$ which is transversal to $0$. Let $\theta_{\alpha}$ be a smooth differential form of compact support on $V_{\alpha}$. We assume that $\theta_{\alpha}$ is $\Gamma_{\alpha}$-invariant. Let $f_{\alpha}:V_{\alpha}\to L_{s}$ be a $\Gamma_{\alpha}$ equivariant submersion. (The $\Gamma_{\alpha}$ action on $L_{s}$ is trivial.) We next define integration along the fiber $((V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha}),\mathfrak{s}_{\alpha},f_{\alpha})_{*}(\theta_{\alpha})\in\Omega^{\deg\theta_{\alpha}+\mathop{\kern 0.0pt{\rm dim}}\nolimits L_{t}-\mathop{\kern 0.0pt{\rm dim}}\nolimits\mathcal{M}}(L_{t}).$ We first fix $\alpha$. Let $(U_{i},\mathfrak{s}_{\alpha,i})$ be a representative of $\mathfrak{s}_{\alpha}$. Namely $\\{U_{i}\mid i\in I\\}$ is an open covering of $V_{\alpha}$ and $\mathfrak{s}_{\alpha}$ is represented by $\mathfrak{s}_{\alpha,i}$ on $U_{i}$. By the definition of the multisection, $U_{i}$ is $\Gamma_{\alpha}$-invariant. We may shrink $U_{i}$, if necessary, so that there exists a lifting $\tilde{\mathfrak{s}}_{\alpha,i}=(\tilde{\mathfrak{s}}_{\alpha,i,1},\cdots,\tilde{\mathfrak{s}}_{\alpha,i,l_{i}})$ as in (15.10). Let $\\{\chi_{i}\mid i\in I\\}$ be a partition of unity subordinate to the covering $\\{U_{i}\mid i\in I\\}$. By replacing $\chi_{i}$ with its average over $\Gamma_{\alpha}$ we may assume $\chi_{i}$ is $\Gamma_{\alpha}$-invariant. We put $\tilde{\mathfrak{s}}_{\alpha,i,j}^{-1}(0)=\\{x\in U_{i}\mid\tilde{\mathfrak{s}}_{\alpha,i,j}(x)=0\\}.$ (16.1) By assumption $\tilde{\mathfrak{s}}_{\alpha,i,j}^{-1}(0)$ is a smooth manifold. Since the $T^{n}$ action on $L_{t}$ is free and transitive it follows that $ev_{t,\alpha}|_{\tilde{\mathfrak{s}}_{\alpha,i,j}^{-1}(0)}:\tilde{\mathfrak{s}}_{\alpha,i,j}^{-1}(0)\to L_{t}$ (16.2) is a submersion. ###### Definition 16.1. We define a differential form on $L_{t}$ by $\displaystyle((V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha}),\mathfrak{s}_{\alpha},ev_{t,\alpha})_{*}(\theta_{\alpha})$ (16.3) $\displaystyle=\frac{1}{\\#\Gamma_{\alpha}}\sum_{i=1}^{I}\sum_{j=1}^{l_{i}}\frac{1}{l_{i}}(ev_{t,\alpha})_{!}\left(\chi_{i}\theta_{\alpha}|_{\tilde{\mathfrak{s}}_{\alpha,i,j}^{-1}(0)}\right).$ Here $(ev_{t,\alpha})_{!}$ is the integration along fiber of the smooth submersion (16.2). ###### Lemma 16.2. The right hand side of $(\ref{intfibermainformula})$ depends only on $(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})$, $\mathfrak{s}_{\alpha}$, $ev_{t,\alpha}$, and $\theta_{\alpha}$ but independent of the following choices : 1. (1) The choice of representatives $(\\{U_{i}\\},\mathfrak{s}_{\alpha,i})$ of $\mathfrak{s}_{\alpha}$. 2. (2) The partition of unity $\chi_{i}$. ###### Proof. The proof is straightforward generalization of the proof of well-definedness of integration on manifold, which can be found in the text book of manifold theory, and is left to the leader. ∎ So far we have been working on one Kuranishi chart $(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})$. We next describe the compatibility conditions of multisections for various $\alpha$. During the construction we need to shrink $V_{\alpha}$ a bit several times. We will not explicitly mention this point henceforth. Let $\alpha_{1}<\alpha_{2}$. For $\alpha_{1}<\alpha_{2}$, we take the normal bundle $N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}$ of $\varphi_{\alpha_{2},\alpha_{1}}(V_{\alpha_{2},\alpha_{1}})$ in $V_{\alpha_{2}}$. We use an appropriate $\Gamma_{\alpha_{2}}$ invariant Riemannian metric on $V_{\alpha_{2}}$ to define the exponential map $\text{Exp}_{\alpha_{2},\alpha_{1}}:B_{\varepsilon}N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}\to V_{2}.$ (16.4) (Here $B_{\varepsilon}N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}$ is the $\epsilon$ neighborhood of the zero section of $N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}$.) Using $\text{Exp}_{\alpha_{2},\alpha_{1}}$, we identify $B_{\varepsilon}N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}/{\Gamma_{\alpha_{1}}}$ to an open subset of $V_{\alpha_{1}}/{\Gamma_{\alpha_{1}}}$ and denote it by $U_{\epsilon}({V_{\alpha_{2},\alpha_{1}}}/{\Gamma_{\alpha_{1}}})$. Using the projection $\text{\rm Pr}_{V_{\alpha_{2},\alpha_{1}}}:U_{\epsilon}({V_{\alpha_{2},\alpha_{1}}}/{\Gamma_{\alpha_{1}}})\to{V_{\alpha_{2},\alpha_{1}}}/{\Gamma_{\alpha_{1}}}$ we extend the orbibundle $E_{\alpha_{1}}$ to $U_{\epsilon}({V_{\alpha_{2},\alpha_{1}}}/{\Gamma_{\alpha_{1}}})$. Also we extend the embedding $E_{\alpha_{1}}\to{\widehat{\varphi}_{\alpha_{2},\alpha_{1}}^{*}E_{\alpha_{2}}}$, (which is induced by $\widehat{\varphi}_{\alpha_{2},\alpha_{1}}$) to $U_{\epsilon}({V_{\alpha_{2},\alpha_{1}}}/{\Gamma_{\alpha_{1}}})$. We fix a $\Gamma_{\alpha}$-invariant inner product of the bundles $E_{\alpha}$. We then have a bundle isomorphism $E_{\alpha_{2}}\cong E_{\alpha_{1}}\oplus\frac{\widehat{\varphi}_{\alpha_{2},\alpha_{1}}^{*}E_{\alpha_{2}}}{E_{\alpha_{1}}}$ (16.5) on $U_{\varepsilon}(V_{\alpha_{2},\alpha_{1}}/\Gamma_{\alpha_{1}})$. We can use Condition 15.3 to modify $\text{Exp}_{\alpha_{2},\alpha_{1}}$ in (16.4) so that the following is satisfied. ###### Condition 16.3. If $y=\text{Exp}_{\alpha_{2},\alpha_{1}}(\tilde{y})\in U_{\epsilon}({V_{\alpha_{2},\alpha_{1}}}/{\Gamma_{\alpha_{1}}})$ then $ds_{\alpha_{2}}(\tilde{y}\mod TV_{\alpha_{1}})\equiv s_{\alpha_{2}}(y)\mod E_{\alpha_{1}}.$ (16.6) Let us explain the notation of (16.6). We remark that $\tilde{y}\in T_{{\varphi}_{\alpha_{2},\alpha_{1}}(x)}V_{\alpha_{2}}$ for $x=\text{\rm Pr}(\tilde{y})\in V_{\alpha_{2},\alpha_{1}}$. Hence $\tilde{y}\mod TV_{\alpha_{1}}\in\frac{T_{{\varphi}_{\alpha_{2},\alpha_{1}}(x)}V_{\alpha_{2}}}{T_{x}V_{\alpha_{1}}}.$ Therefore $ds_{\alpha_{2}}(\tilde{y}\mod TV_{\alpha_{1}})\in\frac{(E_{\alpha_{2}})_{{\varphi}_{\alpha_{2},\alpha_{1}}(x)}}{(E_{\alpha_{1}})_{x}}.$ (16.6) claims that it coincides with $s_{\alpha_{2}}$ modulo $(E_{\alpha_{1}})_{x}$. We remark that Condition 15.3 implies that $[\tilde{y}]\mapsto ds_{\alpha_{2}}(\tilde{y}\mod TV_{\alpha_{1}}):\frac{T_{{\varphi}_{\alpha_{2},\alpha_{1}}(x)}V_{\alpha_{2}}}{T_{x}V_{\alpha_{1}}}\longrightarrow\frac{(E_{\alpha_{2}})_{{\varphi}_{\alpha_{2},\alpha_{1}}(x)}}{(E_{\alpha_{1}})_{x}}$ is an isomorphism. Therefore we can use the implicit function theorem to modify $\text{Exp}_{\alpha_{2},\alpha_{1}}$ so that Condition 16.3 holds. ###### Definition 16.4. A multisection $\mathfrak{s}_{\alpha_{2}}$ of $V_{\alpha_{2}}$ is said to be compatible with $\mathfrak{s}_{\alpha_{1}}$ if the following holds for each $y=\text{Exp}_{\alpha_{2},\alpha_{1}}(\tilde{y})\in U_{\epsilon}({V_{\alpha_{2},\alpha_{1}}}/{\Gamma_{\alpha_{1}}})$. $\mathfrak{s}_{\alpha_{2}}(y)=\mathfrak{s}_{\alpha_{1}}(\text{\rm Pr}(\tilde{y}))\oplus ds_{\alpha_{2}}(\tilde{y}\mod TV_{\alpha_{1}}).$ (16.7) We remark that $\mathfrak{s}_{\alpha_{1}}(w,\text{\rm Pr}(\tilde{y}))$ is a multisection of $\pi_{\alpha_{1}}^{*}E_{\alpha_{1}}$ and $ds_{\alpha_{2}}(\tilde{y}\mod TV_{\alpha_{1}})$ is a (single valued) section. Therefore via the isomorphism (16.5) the right hand side of (16.7) defines an element of $\mathcal{S}^{l_{i}}(E_{\alpha_{2}})_{x}$ ($x=\text{\rm Pr}(\tilde{y})$), and hence is regarded as a multisection of $\pi_{\alpha_{2}}^{*}E_{\alpha_{2}}$. In other words, we omit $\widehat{\varphi}_{\alpha_{2},\alpha_{1}}$ in (16.7). We next choose a partition of unity $\chi_{\alpha}$ subordinate to our Kuranishi charts. To define the notion of partition of unity, we need some notation. Let $\text{Pr}_{\alpha_{2},\alpha_{1}}:N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}\to V_{\alpha_{2},\alpha_{1}}$ be the projection. We fix a $\Gamma_{\alpha_{1}}$-invariant positive definite metric of $N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}$ and let $r_{\alpha_{2},\alpha_{1}}:N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}\to[0,\infty)$ be the norm with respect to this metric. We fix a sufficiently small $\delta$ and let $\chi^{\delta}:{\mathbb{R}}\to[0,1]$ be a smooth function such that $\chi^{\delta}(t)=\begin{cases}0&t\geq\delta\\\ 1&t\leq\delta/2.\end{cases}$ Let $U_{\delta}(V_{\alpha_{2},\alpha_{1}}/\Gamma_{\alpha_{1}})$ be the image of the exponential map. Namely $U_{\delta}(V_{\alpha_{2},\alpha_{1}}/\Gamma_{\alpha_{1}})=\\{\text{Exp}(v)\mid v\in N_{V_{\alpha_{2},\alpha_{1}}}V_{\alpha_{2}}/\Gamma_{\alpha_{1}}\mid r_{\alpha_{2},\alpha_{1}}(v)<\delta\\}.$ We push out our function $r_{\alpha_{2},\alpha_{1}}$ to $U_{\delta}(V_{\alpha_{2},\alpha_{1}}/\Gamma_{\alpha_{1}})$ and denote it by the same symbol. We call $r_{\alpha_{2},\alpha_{1}}$ a tubular distance function. We require $r_{\alpha_{2},\alpha_{1}}$ to satisfy the compatibility conditions for various tubular neighborhoods and tubular distance functions, which is formulated in sections 5 and 6 in [Ma]. Let $x\in V_{\alpha}$. We put $\displaystyle\mathfrak{A}_{x,+}$ $\displaystyle=\\{\alpha_{+}\mid x\in V_{\alpha_{+},\alpha},\,\alpha_{+}>\alpha\\}$ $\displaystyle\mathfrak{A}_{x,-}$ $\displaystyle=\\{\alpha_{-}\mid[x\mod\Gamma_{\alpha}]\in U_{\delta}(V_{\alpha,\alpha_{-}}/\Gamma_{\alpha_{-}}),\,\alpha_{-}<\alpha\\}.$ For $\alpha_{-}\in\mathfrak{A}_{x,-}$ we take $x_{\alpha_{-}}$ such that $\text{Exp}(x_{\alpha_{-}})=x$. ###### Definition 16.5. A system $\\{\chi_{\alpha}\mid\alpha\in\mathfrak{A}\\}$ of $\Gamma_{\alpha}$-equivariant smooth functions $\chi_{\alpha}:V_{\alpha}\to[0,1]$ of compact support is said to be a partition of unity subordinate to our Kuranishi chart if : $\chi_{\alpha}(x)+\sum_{\alpha_{-}\in\mathfrak{A}_{x,-}}\chi^{\delta}(r_{\alpha,\alpha_{-}}(x_{\alpha_{-}}))\chi_{\alpha_{-}}(\text{Pr}_{\alpha,\alpha_{-}}(x_{\alpha_{-}}))+\sum_{\alpha_{+}\in\mathfrak{A}_{x,+}}\chi_{\alpha_{+}}(\varphi_{\alpha_{+},\alpha}(x))=1.$ ###### Lemma 16.6. There exists a partition of unity subordinate to our Kuranishi chart. We may choose them so that they are $T^{n}$ equivariant. ###### Proof. We may assume that $\mathfrak{A}$ is a finite set since $\mathcal{M}$ is compact. By shrinking $V_{\alpha}$ if necessary we may assume that there exists $V^{-}_{\alpha}$ such that $V^{-}_{\alpha}$ is a relatively compact subset of $V_{\alpha}$ and that $E_{\alpha}$, $\varphi_{\alpha_{2},\alpha_{1}}$, $s_{\alpha}$, etc restricted to $V^{-}_{\alpha}$ still defines a good coordinate system. We take a $\Gamma_{\alpha}$ invariant smooth function $\chi^{\prime}_{\alpha}$ on $V_{\alpha}$ which has compact support and satisfies $\chi^{\prime}_{\alpha}=1$ on $V^{-}_{\alpha}$. We define $h_{\alpha}(x)=\chi^{\prime}_{\alpha}(x)+\sum_{\alpha_{-}\in\mathfrak{A}_{x,-}}\chi^{\delta}(r_{\alpha,\alpha_{-}}(x_{\alpha_{-}}))\chi^{\prime}_{\alpha_{-}}(\text{Pr}_{\alpha,\alpha_{-}}(x_{\alpha_{-}}))+\sum_{\alpha_{+}\in\mathfrak{A}_{x,+}}\chi^{\prime}_{\alpha_{+}}(\varphi_{\alpha_{+},\alpha}(x)).$ Using compatibility of tubular neighborhoods and tubular distance functions, we can show that $h_{\alpha}$ is $\Gamma_{\alpha}$ invariant and $h_{\alpha_{2}}(\varphi_{\alpha_{2},\alpha_{1}}(x))=h_{\alpha_{1}}(x)$ if $x\in V_{\alpha_{2},\alpha_{1}}$. Therefore $\chi_{\alpha}(x)=\chi^{\prime}_{\alpha}(x)/h_{\alpha}(x)$ has the required properties. ∎ Now we consider the situation we start with. Namely we have two strongly continuous $T^{n}$ equivariant smooth maps $ev_{s}:\mathcal{M}\to L_{s},\qquad ev_{t}:\mathcal{M}\to L_{t}$ and $ev_{t}$ is weakly submersive. (In fact $T^{n}$ action on $L_{t}$ is transitive and free.) Let a differential form $h$ on $L_{s}$ be given. We choose a $T^{n}$ equivariant good coordinate system $\\{(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})\\}$ of $\mathcal{M}$ and a $T^{n}$ equivariant multisection represented by $\\{\mathfrak{s}_{\alpha}\\}$ in this Kuranishi chart. Assume that the multisection is transversal to zero. We also choose a partition of unity $\\{\chi_{\alpha}\\}$ subordinate to our Kuranishi chart. Then we put $\theta_{\alpha}=\chi_{\alpha}(ev_{s,\alpha})^{*}h$ (16.8) which is a differential form on $V_{\alpha}$. ###### Definition 16.7. Define $(\mathcal{M};ev_{s},ev_{t})_{*}(h)=\sum_{\alpha}((V_{\alpha},\Gamma_{\alpha},E_{\alpha},\psi_{\alpha},s_{\alpha}),\mathfrak{s}_{\alpha},ev_{t,\alpha})_{*}(\theta_{\alpha}).$ (16.9) This is a smooth differential form on $L_{t}$. It is $T^{n}$ equivariant if $h$ is $T^{n}$ equivariant. ###### Remark 16.8. 1. (1) Actually the right hand side of (16.9) depends on the choice of $(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})$, $\mathfrak{s}_{\alpha}$. We write $\mathfrak{s}$ to demonstrate this choice and write $(\mathcal{M};ev_{s},ev_{t},\mathfrak{s})_{*}(h)$. 2. (2) The right hand side of (16.9) is independent of the choice of partition of unity. The proof is similar to the well-definiedness of integration on manifolds. In case $\mathcal{M}$ has a boundary $\partial\mathcal{M}$, the choices $(V_{\alpha},E_{\alpha},\Gamma_{\alpha},\psi_{\alpha},s_{\alpha})$, $\mathfrak{s}_{\alpha}$ on $\mathcal{M}$ induces one for $\partial\mathcal{M}$. We then have the following : ###### Lemma 16.9 (Stokes’ theorem). We have $d((\mathcal{M};ev_{s},ev_{t},\mathfrak{s})_{*}(h))=(\mathcal{M};ev_{s},ev_{t},\mathfrak{s})_{*}(dh)+(\partial\mathcal{M};ev_{s},ev_{t},\mathfrak{s})_{*}(h).$ (16.10) We will discuss the sign at the end of this section. ###### Proof. Using the partition of unity $\chi_{\alpha}$ it suffices to consider the case when $\mathcal{M}$ has only one Kuranishi chart $V_{\alpha}$. We use the open covering $U_{i}$ of $V_{\alpha}$ and the partition of unity again to see that we need only to study on one $U_{i}$. In that case (16.10) is immediate from the usual Stokes’ formula. ∎ We next discuss composition of smooth correspondences. We consider the following situation. Let $ev_{s;st}:\mathcal{M}_{st}\to L_{s},\qquad ev_{t;st}:\mathcal{M}_{st}\to L_{t}$ be as before such that $T^{n}$ action on $L_{t}$ is free and transitive. Let $ev_{r;rs}:\mathcal{M}_{rs}\to L_{r},\qquad ev_{s;rs}:\mathcal{M}_{rs}\to L_{s}$ be a similar diagram such that $T^{n}$ on $L_{s}$ is free and transitive. Using the fact that $ev_{s;rs}$ is weakly submersive, we define the fiber product $\mathcal{M}_{rs}\,{}_{ev_{s;rs}}\times_{ev_{s;st}}\mathcal{M}_{st}$ as a space with Kuranishi structure. We write it as $\mathcal{M}_{rt}$. We have a diagram of strongly continuous smooth maps $ev_{r;rt}:\mathcal{M}_{rt}\to L_{r},\qquad ev_{t;rt}:\mathcal{M}_{rt}\to L_{t}.$ We next make choices $\mathfrak{s}^{st}$, $\mathfrak{s}^{rs}$ for $\mathcal{M}_{st}$ and $\mathcal{M}_{rs}$. It is easy to see that it determines a choice $\mathfrak{s}^{rt}$ for $\mathcal{M}_{rt}$. Now we have : ###### Lemma 16.10 (Composition formula). We have the following formula for each differential form $h$ on $L_{r}$. $\displaystyle(\mathcal{M}_{rt};ev_{r;rt},ev_{t;,rt},\mathfrak{s}^{rt})_{*}(h)$ (16.11) $\displaystyle=((\mathcal{M}_{st};ev_{s;st},ev_{t;,st},\mathfrak{s}^{st})_{*}\circ(\mathcal{M}_{rs};ev_{r;rs},ev_{s;,rs},\mathfrak{s}^{rs})_{*})(h).$ ###### Proof. Using a partition of unity it suffices to study locally on $\mathcal{M}_{rs}$, $\mathcal{M}_{st}$. In that case it suffices to consider the case of usual manifold, which is well-known. ∎ We finally discuss the signs in Lemmas 16.9 and 16.10. It is rather cumbersome to fix appropriate sign convention and show those lemmata with sign. So, instead, we use the trick of subsection 8.10.3 [FOOO3] (= section 53.3 [FOOO2]) (see also section 13 [Fu5]) to reduce the orientation problem to the case which is already discussed in Chapter 8 [FOOO3] (= Chapter 9 [FOOO2]), as follows. The correspondence $h\mapsto(\mathcal{M};ev_{s},ev_{t},\mathfrak{s})_{*}(h)$ extends to the currents $h$ that satisfy appropriate transversality properties about its wave front set. (See [Hö].) We can also represent the smooth form $h$ by an appropriate average (with respect to certain smooth measure) of a family of currents realized by smooth singular chains. So, as far as sign concerns, it suffices to consider a current realized by a smooth singular chain. Then the right hand side of (16.3) turn out to be a current realized by a smooth singular chain which is obtained from a smooth singular chain on $L_{s}$ by a transversal smooth correspondence. In fact, we may assume that all the fiber products appearing here are transversal, since it suffices to discuss the sign in the generic case. Thus the problem reduces to find a sign convention (and orientation) for the correspondence of the singular chains by a smooth manifold. In the situation of our application, such sign convention (singular homology version) was determined and analyzed in detail in Chapter 8 [FOOO3] (= Chapter 9 [FOOO2]). Especially, existence of an appropriate orientation that is consistent with the sign appearing in $A_{\infty}$ formulae etc. was proved there. Therefore we can prove that there is a sign (orientation) convention which induces all the formulae we need with sign, in our de Rham version, as well. See subsection 8.10.3 [FOOO3] (= section 53.3 [FOOO2]) or section 13 [Fu5] for the detail. ## References * [Ab] M. Abouzaid, Morse homology, tropical geometry, and homological mirror symmetry for toric varieties, Geom. Topol. 10 (2006), 1097–1157 (electronic). * [Au] M. Audin, Torus Actions on Symplectic Manifolds, Progress in Mathematics 93, Birkhäuser Verlag, Basel, 2004. * [Aur] D. Auroux, Mirror symmetry and T-duality in the complement of an anticanonical divisor, J. Gökova Geom. Topol. 1 (2007), 51-91 (electronic), arXiv:0706.3207v2. * [AKO] D. Auroux, L. Katzarkov and D. Orlov, Mirror symmetry for weighted projective planes and their noncommutative deformations, Ann. of Math. 167 (2008) 867–943, mathAG/0404281. * [B1] V. Batyrev, Quamtum cohomology rings of toric manifolds, Journees de Geometrie Algebrique d’Orsay, 1992, Asterisque 218, 9-34. * [B2] V. Batyrev, Dual polyhedra and mirror symmetry for Calabi-Yau hypersurfaces in toric varieties, J. Alge. Geom. 3 (1994), no. 3, 493–535. * [BGR] S. Bosch, U. Güntzer and R. Remmert, Non-Archimedean Analysis. Asystematic Approach to Rigid Analytic Geometry, Grundlehren der Mathematischen Wissenschaften 261, Springer-Verlag, Berline, 1984. * [CL] K. Chan and N.C. Leung, Mirror symmetry for toric Fano manifolds via SYZ transformations, arXiv:0801.2830. * [Che] Y.V. Chekanov, Lagrangian intersections, symplectic energy, and areas of holomorphic curves, Duke Math. J. 95 (1998), no. 1, 213–226. * [Cho] C.-H. Cho, Non-displaceable Lagrangian submanifolds and Floer cohomology with non-unitary line bundle, J. Geom. Phys. 58 (2008), 1465–1476, arXiv:0710.5454. * [CO] C.-H. Cho and Y.-G. Oh, Floer cohomology and disc instantons of Lagrangian torus fibers in Fano toric manifolds, Asian J. Math. 10 (2006), 773–814. * [CK] D. A. Cox, and S. Katz, Mirror Symmetry and Algebraic Geometry, Mathematical Surveys and Monographs, 68. American Mathematical Society, Providence, RI, 1999. * [CS] K. Cieliebak and D. Salamon, Wall crossing for symplectic vortices and quantum cohomology, Math. Ann. 335 (2006), no. 1, 133–192. * [De] T. Delzant, Hamiltoniens périodiques et image convexe de l’application moment, Bull. Soc. Math. France 116 (1988), 315 – 339. * [Ei] D. Eisenbud, Commutative Algebra with a View Toward Algebraic Geometry, Graduate Text in Math. 150, Springer-Verlag, Berline, 1994. * [EP1] M. Entov and L. Polterovich, Quasi-states and symplectic intersections, Comment. Math. Helv. 81 (2006), 75–99. * [EP2] M. Entov and L. Polterovich, Symplectic quasi-states and semi-simplicity of quantum homology, in ‘Toric topology’, Contemp. Math. 460, 47–70, arXiv:0705.3735. * [EP3] M. Entov and L. Polterovich, Rigid subsets of symplectic manifolds, Compos. Math., 145 (2009) 773–826 , arXiv:0704.0105. * [Fl] A. Floer, Morse theory for Lagrangian intersections, J. Differ. Geom. 28 (1988), 513–547. * [Fu1] K. Fukaya, Morse homotopy, $A_{\infty}$-category and Floer homologies, in Proceedings of GARC Workshop on Geometry and Topology, ed by H. J. Kim, Seoul National University, Korea 1933. * [Fu2] K. Fukaya, Mirror symmetry of abelian variety and multi theta function, J. Algebraic Geom. 11 393–512 2002 * [Fu3] K. Fukaya, Floer homology for families– a progress report, Integrable systems, topology, and physics (Tokyo 2000), Contemp. Math. 309, Amer. Math. Soc. 2002, 33–68. * [Fu4] K. Fukaya, Application of Floer homology of Lagrangian submanifolds to symplectic topology. in ‘Morse Theoretic Methods in Nonlinear Analysis and in Symplectic Topology”, 231–276, NATO Sci. Ser. II Math. Phys. Chem., 217, Springer, Dordrecht, 2006. * [Fu5] K. Fukaya, Differentiable operad, Kuranishi correspondence,and Foundation of topological field theories based on pseudo-holomorphic curve, to appear in ‘Arithmetic and Geometry around Quantization’ Prog. Math., 279, Birkhäuser. * [FOOO1] K. Fukaya, Y.-G. Oh, H. Ohta and K. Ono, Lagrangian intersection Floer theory-anomaly and obstruction, Kyoto University preprint, 2000. * [FOOO2] K. Fukaya, Y.-G. Oh, H. Ohta and K. Ono, Lagrangian intersection Floer theory-anomaly and obstruction, expanded version of [FOOO1], 2006 & 2007\. * [FOOO3] K. Fukaya, Y.-G. Oh, H. Ohta and K. Ono, Lagrangian intersection Floer theory-anomaly and obstruction, AMS/IP Studies in Advanced Mathematics, vol 46, in press, AMS/International Press. * [FOOO4] K. Fukaya, Y.-G. Oh, H. Ohta and K. Ono, Lagrangian Floer theory on compact toric manifolds II : bulk deformations, preprint 2008, arXiv:0810.5654. * [FOOO5] K. Fukaya, Y.-G. Oh, H. Ohta and K. Ono, Canonical models of filtered $A_{\infty}$-algebras and Morse complexes, to appear in ‘New Perspectives and Challenges in Symplectic Field Theory’, CRM Proc. and Lecture Notes, 49, Amer. Math. Soc., arXiv:0812.1963. * [FO] K. Fukaya and K. Ono, Arnold conjecture and Gromov-Witten invariant, Topology 38 (1999), no. 5, 933–1048. * [Ful] W. Fulton, Introduction to Toric Varieties, Annals of Math. Studies, 131, Princeton University Press, Princeton, 1993. * [Gi1] A. Givental, Homological geometry and mirror symmetry, in “Proceeding of the ICM-1994, Zürich, Basel, 1995, 472 - 480. * [Gi2] A. Givental, Equivariant Gromov-Witten invariants, Internat. Math. Res. Notices, 13 1996, 613 – 663. * [GrPa] T. Graber and R. Pandharipande, Localization of virtual classes, Invent. Math. 135 (1999), 487–518 * [Gu] V. Guillemin, Kähler structures on toric varieties, J. Differ. Geom. 43 (1994), 285–309. * [H] H. Hofer, On the topological properties of symplectic maps, Proc. Royal Soc. Edinburgh 115 (1990), 25–38. * [Ho] K. Hori, Linear models in supersymmetric D-branes, Proceedings of the KIAS conference-Mirror Symmetry and Symplectic Geometry- (Seoul, 2000), eds by K. Fukaya, Y.-G. Oh, K. Ono and G. Tian, World Sci. Publishing, River Edge, New Jersey, 2001. * [HV] K. Hori and C. Vafa, Mirror symmetry, preprint, 2000, hep-th/0002222. * [MIRROR] K. Hori, S Katz, A. Klemm, R. Pandharipande, R. Thomas, C. Vafa, R. Vakil and E. Zaslow, Mirror Symmetry, Clay Mathematics Monographs, Amer. Math. Soc., 2003. * [Hö] L. Hörmander Fourier integral operators I, Acta Math., 127 (1971), 79 - 183. * [Iri1] H. Iritani, Quantum D-modules and equivariant Floer theory for free loop spaces, Math. Z. 252 (2006), 577 – 622. * [Iri2] H. Iritani, Convergence of quantum cohomology by quantum Lefschetz, J. Reine Angew. Math. 610 (2007), 29–69. * [Ko] M. Kontsevich, Homological algebra of mirror symmetry, ICM-1994 Proceedings, Zürich, Birkhäuser, 1995. * [Ku] A.G. Kushnirenko, Polyédres de Newton et nombre de Milnor. Invent. Math. 32 (1976) 1 – 31. * [LiRu] A.-M. Li and Y. Ruan, Symplectic surgery and Gromov-Witten invariants of Calabi-Yau 3-folds, Invent. Math. 145 (2001), no. 1, 151–218. * [Man] Y.I. Manin, Frobenius Manifolds, Quantum Cohomology, and Moduli Spaces, AMS Colloquium Publications, 47, American Mathematical Society, Providence, RI, 1999. * [Ma] J. Mather, Stratifications and mappings, Dynamical systems (Proc. Sympos., Univ. Bahia, Salvador, 1971) 41\. 195–232. Academic Press. New York. 1973\. * [Mat] H. Matsumura, Commutative Algebra, Benjamin, 1970. * [MW] J. Marsden and A. Weinstein, Reduction of symplectic manifolds with symmetry, Reports of Math. Phys. 5, (1974) 121–130. * [MT] D. McDuff and S. Tolman, Topological properties of Hamiltonian circle actions, Int. Math. Res. Pap. 2006, 72826, 1 – 77. * [Mc] D. McDuff, Displacing Lagrangian toric fibers via probes, preprint, arXiv:0904.1686. * [Mi] J. Milnor, Singular Points of Complex Hypersurfaces, Ann. Math. Studies Princeton Univ. Press, Princeton N.J., 1968 * [Na] M. Nagata, Local rings. Interscience Tracts in Pure and Applied Math. 13, Interscience Publishers a division of John Wiley & Sons New York-London 1962 * [Oh1] Y.-G. Oh, Floer cohomology of Lagrangian intersections and pseudo-holomorphic discs I, Comm. Pure Appl. Math. 46 (1993), 949-994 addenda, ibid, 48 (1995), 1299–1302. * [Oh2] Y.-G. Oh, Symplectic topology as the geometry of action functional II, Commun. Anal. Geom. 7 (1999), 1 – 55 * [Oh3] Y.-G. Oh, Construction of spectral invariants of Hamiltonian paths on closed symplectic manifolds, in “The Breadth of Symplectic and Poisson Geometry”, Prog. Math. 232, 525 – 570, Birkhäuser, Boston, 2005. * [Oh4] Y.-G. Oh, Floer mini-max theory, the Cerf diagram, and the spectral invariants, J. Korean Math. Soc. 46 (2009), 363 – 447, mathSG/0406449 (v4 2007). * [Ohta] H. Ohta, Obstruction to and deformation of Lagrangian intersection Floer cohomology, Symplectic Geometry and Mirror Symmetry (Seoul, 2000), K. Fukaya, Y.-G. Oh, K. Ono and G. Tian ed., World Sci. Pulishing, River Edge, 2001, pp. 281–309. * [Ru] Y. Ruan, Virtual neighborhood and pseudoholomorphic curve. Turkish J. Math. 1999 161–231 * [Sa] K. Saito, Period mapping associated to a primitive form, Publ. R.I.M.S. 19 (1983), 1231 –1261. * [Sc] M. Schwarz, On the action spectrum for closed symplectically aspherical manifolds, Pacific J. Math. 193 (2000), 419–461. * [Se1] P. Seidel, $\pi_{1}$ of symplectic automorphism groups and invertibles in quantum homology rings, Geom. Funct. Anal. 7 (1997), no. 6, 1046–1095. * [Se2] P. Seidel, Fukaya Categories and Picard-Lefschetz Theory, European Mathematical Society (EMS), Zürich, 2008. * [Ue] K. Ueda, Homological mirror symmetry for toric del Pezzo surfaces, Comm. Math. Phys. 264 (2006), 71 – 85. * [Us] M. Usher, Spectral numbers in Floer theories, Compositio Math. 144 (2008), 1581-1592, arXiv:0709.1127. * [Vi] C. Viterbo, Symplectic topology as the geometry of generating functions, Math. Ann. 292 (1992), 685–710.

arxiv-papers

2024-09-04T02:48:53.709684

0802.1760

# Doping and energy dependent microwave conductivity of kinetic energy driven superconductors with extended impurities Zhi Wang Department of Physics, Beijing Normal University, Beijing 100875, China Huaiming Guo Department of Physics, Capital Normal University, Beijing 100037, China Shiping Feng∗ Department of Physics, Beijing Normal University, Beijing 100875, China ###### Abstract Within the framework of the kinetic energy driven superconducting mechanism, the effect of the extended impurity scatterers on the quasiparticle transport of cuprate superconductors in the superconducting state is studied based on the nodal approximation of the quasiparticle excitations and scattering processes. It is shown that there is a cusplike shape of the energy dependent microwave conductivity spectrum. At low temperatures, the microwave conductivity increases linearly with increasing temperatures, and reaches a maximum at intermediate temperature, then decreases with increasing temperatures at high temperatures. In contrast with the dome shape of the doping dependent superconducting gap parameter, the minimum microwave conductivity occurs around the optimal doping, and then increases in both underdoped and overdoped regimes. ###### pacs: 74.25.-h, 74.25.Fy, 74.25.Nf ## I Introduction After over 20 years extensive studies, it has become clear that superconductivity in doped cuprates results when charge carriers pair up into Cooper pairs tsuei as in the conventional superconductors bcs , then these charge carrier Cooper pairs condensation reveals the superconducting (SC) ground-state. However, as a natural consequence of the unconventional SC mechanism that is responsible for the high SC transition temperatures anderson , the charge carrier Cooper pairs in cuprate superconductors have a dominated d-wave symmetry tsuei ; shen . In particular, this d-wave SC state remains one of the cornerstones of our understanding of the physics in cuprate superconductors. However, in spite of the unconventional SC mechanism, the angle-resolved photoemission spectroscopy (ARPES) experimental results have unambiguously established the Bogoliubov-quasiparticle nature of the sharp SC quasiparticle peak in cuprate superconductors matsui ; campuzano1 , then the SC coherence of the low energy quasiparticle peak is well described by the simple Bardeen-Cooper-Schrieffer (BCS) formalism bcs with the d-wave SC gap function $\Delta({\bf k})=\Delta({\rm cos}k_{x}-{\rm cos}k_{y})/2$. In this d-wave case, the characteristic feature is the existence of four nodal points $[\pm\pi/2,\pm\pi/2]$ (in units of inverse lattice constant) in the Brillouin zone tsuei ; shen , where the SC gap function vanishes, then the most physical properties of cuprate superconductors in the SC state are controlled by the quasiparticle excitations around the nodes turner . In particular, the key signature of the nodal quasiparticle transport appears in the microwave conductivity $\sigma(\omega,T)$ turner , which is essentially electromagnetic absorption by the quasiparticles excited out of the condensation (either thermally excited quasiparticles or excitations created by the impurity scattering). Understanding the role of impurities in cuprate superconductors has taken many years of great effort palee ; hirschfeld . This follows from that the physical properties of cuprate superconductors in the SC state are extreme sensitivity to the impurity effect than the conventional superconductors due to the finite angular-momentum charge carrier Cooper pairing hirschfeld1 . Experimentally, By virtue of systematic studies using the microwave conductivity measurements, some essential features of the evolution of the quasiparticle transport of cuprate superconductors with energy and temperature in the SC state have been established turner ; bonn ; lee ; hosseini ; harris : (1) at low temperatures, the experimental results show the existence of the very long-live excitation deep in the SC state, as evidenced by the sharp cusplike energy dependent microwave conductivity spectrum, where the width of the sharp peak is nearly temperature independent, and main behaviors of the microwave conductivity are governed by thermally excited quasiparticles being scattered by impurities or other defects turner ; harris ; (2) at low energies, the temperature dependent microwave conductivity increases linearly with increasing temperatures at low temperatures, and reaches a maximum (a large broad peak) around intermediate temperature, then decreases with increasing temperatures at high temperatures bonn ; hosseini ; harris . In particular, this broad peak shifts to higher temperatures as the energy is increased. Theoretically, an agreement has emerged that the BCS formalism with the d-wave SC gap function is useful in the phenomenological description of the quasiparticle transport of cuprate superconductors in the SC state palee ; hirschfeld ; quinlan ; hettler ; durst ; duffy ; nunner , although the SC pairing mechanism is beyond the conventional electron-phonon mechanism. In this case, the microwave conductivity of cuprate superconductors has been phenomenologically discussed in the zero temperature and energy by including the contributions of the vertex corrections durst . Recently, these discussions have been generalized to study the temperature and energy dependence of the quasiparticle transport of cuprate superconductors in the SC state nunner . To the best of our knowledge, the microwave conductivity of cuprate superconductors has not been treated starting from a microscopic SC theory, and no explicit calculations of the doping dependence of the microwave conductivity has been made so far. In this paper, we start from the kinetic energy driven SC mechanism feng1 , and study the effect of the extended impurity scatterers on the microwave conductivity of cuprate superconductors. We evaluate explicitly the microwave conductivity of cuprate superconductors within the nodal approximation of the quasiparticle excitations and scattering processes, and qualitatively reproduced some main features of the microwave conductivity measurements on cuprate superconductors in the SC state turner ; bonn ; lee ; hosseini ; harris . It is shown that there is a cusplike shape of the energy dependent microwave conductivity spectrum. At low temperatures, the microwave conductivity increases linearly with increasing temperatures, and reaches a maximum at intermediate temperature, then decreases with increasing temperatures at high temperatures. In contrast with the dome shape of the doping dependent SC gap parameter, the minimum microwave conductivity occurs around the optimal doping, and then increases in both underdoped and overdoped regimes. This paper is organized as follows. We present the basic formalism in Sec. II, and then discuss the energy, temperature, and doping dependence of the quasiparticle transport of cuprate superconductors in the SC state in Sec. III, where we show that the quasiparticle transport of cuprate superconductors in the SC state can be qualitatively understood within the framework of the kinetic energy driven SC mechanism by considering the effect of the extended impurity scatterers. Finally, we give a summary in Sec. IV. ## II Formalism In cuprate superconductors, the single common feature is the presence of the two-dimensional CuO2 plane, and it is believed that the unconventional physical properties of cuprate superconductors is closely related to the doped CuO2 planes shen ; kastner . It has been argued that the $t$-$J$ model on a square lattice captures the essential physics of the doped CuO2 plane anderson ; shen , $H=-t\sum_{i\hat{\eta}\sigma}\hat{C}^{\dagger}_{i\sigma}\hat{C}_{i+\hat{\eta}\sigma}+\mu\sum_{i\sigma}\hat{C}^{\dagger}_{i\sigma}\hat{C}_{i\sigma}+J\sum_{i\hat{\eta}}{\bf S}_{i}\cdot{\bf S}_{i+\hat{\eta}},$ (1) where $\hat{\eta}=\pm\hat{x},\pm\hat{y}$, ${\bf S}_{i}=\hat{C}^{\dagger}_{i}{\boldsymbol{\tau}}\hat{C}_{i}/2$ is spin operator with ${\boldsymbol{\tau}}=(\tau_{1},\tau_{2},\tau_{3})$ as Pauli matrices, the constrained electron operator $\hat{C}_{i\sigma}=C_{i\sigma}(1-n_{i-\sigma})$ with $n_{i\sigma}=C^{\dagger}_{i\sigma}C_{i\sigma}$, and $\mu$ is the chemical potential. In the constrained electron operator, the operators $C^{\dagger}_{i\sigma}$ and $C_{i\sigma}$ are to be thought of as operating within the full Hilbert space, while the constrained electron operator $\hat{C}^{\dagger}_{i\sigma}$ ($\hat{C}_{i\sigma}$) does not create (destroy) any doubly occupied sites, and therefore represents physical creation (annihilation) operator acting in the restricted Hilbert space without double electron occupancy rice ; anderson1 . The strong electron correlation in the $t$-$J$ model manifests itself by the restriction of the motions of the electrons in the restricted Hilbert space without double electron occupancy anderson , which can be treated properly in analytical calculations within the charge-spin separation (CSS) fermion-spin theory feng2 , where the constrained electron operators are decoupled as $\hat{C}_{i\uparrow}=h^{\dagger}_{i\uparrow}S^{-}_{i}$ and $\hat{C}_{i\downarrow}=h^{\dagger}_{i\downarrow}S^{+}_{i}$, with the spinful fermion operator $h_{i\sigma}=e^{-i\Phi_{i\sigma}}h_{i}$ describes the charge degree of freedom together with some effects of spin configuration rearrangements due to the presence of the doped hole itself (dressed holon), while the spin operator $S_{i}$ describes the spin degree of freedom (spin), then the motions of electrons are restricted in the restricted Hilbert space without double electron occupancy in analytical calculations. In particular, it has been shown that under the decoupling scheme, this CSS fermion-spin representation is a natural representation of the constrained electron defined in the restricted Hilbert space without double electron occupancy feng3 . Moreover, these dressed holon and spin are gauge invariant feng2 , and in this sense, they are real and can be interpreted as the physical excitations laughlin . In this CSS fermion-spin representation, the low-energy behavior of the $t$-$J$ model (1) can be expressed as, $\displaystyle H$ $\displaystyle=$ $\displaystyle t\sum_{i\hat{\eta}}(h^{\dagger}_{i+\hat{\eta}\uparrow}h_{i\uparrow}S^{+}_{i}S^{-}_{i+\hat{\eta}}+h^{\dagger}_{i+\hat{\eta}\downarrow}h_{i\downarrow}S^{-}_{i}S^{+}_{i+\hat{\eta}})$ (2) $\displaystyle-$ $\displaystyle\mu\sum_{i\sigma}h^{\dagger}_{i\sigma}h_{i\sigma}+J_{{\rm eff}}\sum_{i\hat{\eta}}{\bf S}_{i}\cdot{\bf S}_{i+\hat{\eta}},$ with $J_{{\rm eff}}=(1-\delta)^{2}J$, and $\delta=\langle h^{\dagger}_{i\sigma}h_{i\sigma}\rangle=\langle h^{\dagger}_{i}h_{i}\rangle$ is the hole doping concentration. As an important consequence, the kinetic energy term in the $t$-$J$ model has been transferred as the dressed holon- spin interaction, which reflects that even the kinetic energy term in the $t$-$J$ Hamiltonian has strong Coulombic contribution due to the restriction of the motions of electrons in the restricted Hilbert space without double electron occupancy. Recently, we have developed a kinetic energy driven SC mechanism feng1 based on the CSS fermion-spin theory feng2 , where the dressed holon-spin interaction from the kinetic energy term in the $t$-$J$ model (2) induces the dressed holon pairing state with the d-wave symmetry by exchanging spin excitations, then the electron Cooper pairs originating from the dressed holon pairing state are due to the charge-spin recombination, and their condensation reveals the d-wave SC ground-state. Moreover, this d-wave SC state is controlled by both SC gap function and quasiparticle coherence, then the maximal SC transition temperature occurs around the optimal doping, and decreases in both underdoped and overdoped regimes. In particular, we have shown that this SC state is the conventional BCS like with the d-wave symmetry feng3 ; guo1 , so that the basic BCS formalism with the d-wave SC gap function is still valid in quantitatively reproducing all main low energy features of the ARPES experimental measurements on cuprate superconductors, although the pairing mechanism is driven by the kinetic energy by exchanging spin excitations, and other exotic magnetic scattering dai ; arai is beyond the BCS formalism. Following our previous discussions feng1 ; feng3 ; guo1 , the full dressed holon Green’s function in the SC state can be obtained in the Nambu representation as, $\displaystyle\tilde{g}({\bf k},\omega)$ $\displaystyle=$ $\displaystyle Z_{hF}{1\over\omega^{2}-E^{2}_{h{\bf k}}}\left(\begin{array}[]{cc}{\omega+\bar{\xi}_{{\bf k}}}&{\bar{\Delta}_{hZ}({\bf k})}\\\ {\bar{\Delta}_{hZ}({\bf k})}&{\omega-\bar{\xi}_{{\bf k}}}\end{array}\right)$ (5) $\displaystyle=$ $\displaystyle Z_{hF}{\omega\tau_{0}+\bar{\Delta}_{hZ}({\bf k})\tau_{1}+\bar{\xi}_{{\bf k}}\tau_{3}\over\omega^{2}-E^{2}_{h{\bf k}}},$ (6) where $\tau_{0}$ is the unit matrix, the renormalized dressed holon excitation spectrum $\bar{\xi}_{{\bf k}}=Z_{hF}\xi_{\bf k}$, with the mean-field (MF) dressed holon excitation spectrum $\xi_{{\bf k}}=Zt\chi\gamma_{{\bf k}}-\mu$, the spin correlation function $\chi=\langle S_{i}^{+}S_{i+\hat{\eta}}^{-}\rangle$, $\gamma_{{\bf k}}=(1/Z)\sum_{\hat{\eta}}e^{i{\bf k}\cdot\hat{\eta}}$, $Z$ is the number of the nearest neighbor sites, the renormalized dressed holon d-wave pair gap function $\bar{\Delta}_{hZ}({\bf k})=Z_{hF}\bar{\Delta}_{h}({\bf k})$, where the effective dressed holon d-wave pair gap function $\bar{\Delta}_{h}({\bf k})=\bar{\Delta}_{h}\gamma^{(d)}_{{\bf k}}$ with $\gamma^{(d)}_{{\bf k}}=({\rm cos}k_{x}-{\rm cos}k_{y})/2$, and the dressed holon quasiparticle spectrum $E_{h{\bf k}}=\sqrt{\bar{\xi}^{2}_{{\bf k}}+\mid\bar{\Delta}_{hZ}({\bf k})\mid^{2}}$, while the dressed holon quasiparticle coherent weight $Z_{hF}$ and effective dressed holon gap parameters $\bar{\Delta}_{h}$ are determined by the following two equations feng1 ; feng3 ; guo1 , $\displaystyle 1$ $\displaystyle=$ $\displaystyle{(Zt)^{2}\over N^{3}}\sum_{{\bf k,p,p^{\prime}}}\gamma^{2}_{{\bf p+k}}\gamma^{(d)}_{{\bf k-p^{\prime}+p}}\gamma^{(d)}_{{\bf k}}{Z^{2}_{hF}\over E_{h{\bf k}}}{B_{{\bf p}}B_{{\bf p^{\prime}}}\over\omega_{{\bf p}}\omega_{{\bf p^{\prime}}}}\left({F^{(1)}_{1}({\bf k,p,p^{\prime}})\over(\omega_{{\bf p^{\prime}}}-\omega_{{\bf p}})^{2}-E^{2}_{h{\bf k}}}-{F^{(2)}_{1}({\bf k,p,p^{\prime}})\over(\omega_{{\bf p^{\prime}}}+\omega_{{\bf p}})^{2}-E^{2}_{h{\bf k}}}\right),$ (7a) $\displaystyle{1\over Z}_{hF}$ $\displaystyle=$ $\displaystyle 1+\left({Zt\over N}\right)^{2}\sum_{{\bf p,p^{\prime}}}\gamma^{2}_{{\bf p}+{\bf k}_{0}}Z_{hF}{B_{{\bf p}}B_{{\bf p^{\prime}}}\over 4\omega_{{\bf p}}\omega_{{\bf p^{\prime}}}}\left({F^{(1)}_{2}({\bf p,p^{\prime}})\over(\omega_{{\bf p}}-\omega_{{\bf p^{\prime}}}-E_{h{\bf p-p^{\prime}+k_{0}}})^{2}}+{F^{(2)}_{2}({\bf p,p^{\prime}})\over(\omega_{{\bf p}}-\omega_{{\bf p^{\prime}}}+E_{h{\bf p-p^{\prime}+k_{0}}})^{2}}\right.$ (7b) $\displaystyle+$ $\displaystyle\left.{F^{(3)}_{2}({\bf p,p^{\prime}})\over(\omega_{{\bf p}}+\omega_{{\bf p^{\prime}}}-E_{h{\bf p-p^{\prime}+k_{0}}})^{2}}+{F^{(4)}_{2}({\bf p,p^{\prime}})\over(\omega_{{\bf p}}+\omega_{{\bf p^{\prime}}}+E_{h{\bf p-p^{\prime}+k_{0}}})^{2}}\right),$ respectively, where ${\bf k}_{0}=[\pi,0]$, $B_{{\bf p}}=\lambda[(2\epsilon\chi^{z}+\chi)\gamma_{{\bf p}}-(2\chi^{z}+\epsilon\chi)]$, $\lambda=2ZJ_{{\rm eff}}$, $\epsilon=1+2t\phi/J_{{\rm eff}}$, the dressed holon’s particle-hole parameter $\phi=\langle h^{\dagger}_{i\sigma}h_{i+\hat{\eta}\sigma}\rangle$, the spin correlation function $\chi^{z}=\langle S_{i}^{z}S_{i+\hat{\eta}}^{z}\rangle$, $F^{(1)}_{1}({\bf k,p,p^{\prime}})=(\omega_{{\bf p^{\prime}}}-\omega_{{\bf p}})[n_{B}(\omega_{{\bf p}})-n_{B}(\omega_{{\bf p^{\prime}}})][1-2n_{F}(E_{h{\bf k}})]+E_{h{\bf k}}[n_{B}(\omega_{{\bf p^{\prime}}})n_{B}(-\omega_{{\bf p}})+n_{B}(\omega_{{\bf p}})n_{B}(-\omega_{{\bf p^{\prime}}})]$, $F^{(2)}_{1}({\bf k,p,p^{\prime}})=(\omega_{{\bf p^{\prime}}}+\omega_{{\bf p}})[n_{B}(-\omega_{{\bf p^{\prime}}})-n_{B}(\omega_{{\bf p}})][1-2n_{F}(E_{h{\bf k}})]+E_{h{\bf k}}[n_{B}(\omega_{{\bf p^{\prime}}})n_{B}(\omega_{{\bf p}})+n_{B}(-\omega_{{\bf p^{\prime}}})n_{B}(-\omega_{{\bf p}})]$, $F^{(1)}_{2}({\bf p,p^{\prime}})=n_{F}(E_{h{\bf p-p^{\prime}+k_{0}}})[n_{B}(\omega_{{\bf p^{\prime}}})-n_{B}(\omega_{{\bf p}})]-n_{B}(\omega_{{\bf p}})n_{B}(-\omega_{{\bf p^{\prime}}})$, $F^{(2)}_{2}({\bf p,p^{\prime}})=n_{F}(E_{h{\bf p-p^{\prime}+k_{0}}})[n_{B}(\omega_{{\bf p}})-n_{B}(\omega_{{\bf p^{\prime}}})]-n_{B}(\omega_{{\bf p^{\prime}}})n_{B}(-\omega_{{\bf p}})$, $F^{(3)}_{2}({\bf p,p^{\prime}})=n_{F}(E_{h{\bf p-p^{\prime}+k_{0}}})[n_{B}(\omega_{{\bf p^{\prime}}})-n_{B}(-\omega_{{\bf p}})]+n_{B}(\omega_{{\bf p}})n_{B}(\omega_{{\bf p^{\prime}}})$, $F^{(4)}_{2}({\bf p,p^{\prime}})=n_{F}(E_{h{\bf p-p^{\prime}+k_{0}}})[n_{B}(-\omega_{{\bf p^{\prime}}})-n_{B}(\omega_{{\bf p}})]+n_{B}(-\omega_{{\bf p}})n_{B}(-\omega_{{\bf p^{\prime}}})$, $n_{B}(\omega_{{\bf p}})$ and $n_{F}(E_{h{\bf k}})$ are the boson and fermion distribution functions, respectively, and the MF spin excitation spectrum, $\displaystyle\omega^{2}_{{\bf p}}$ $\displaystyle=$ $\displaystyle\lambda^{2}[(A_{1}-\alpha\epsilon\chi^{z}\gamma_{{\bf p}}-{1\over 2Z}\alpha\epsilon\chi)(1-\epsilon\gamma_{{\bf p}})$ (8) $\displaystyle+$ $\displaystyle{1\over 2}\epsilon(A_{2}-{1\over 2}\alpha\chi^{z}-\alpha\chi\gamma_{{\bf p}})(\epsilon-\gamma_{{\bf p}})],$ where $A_{1}=\alpha C^{z}+(1-\alpha)/(4Z)$, $A_{2}=\alpha C+(1-\alpha)/(2Z)$, and the spin correlation functions $C=(1/Z^{2})\sum_{\hat{\eta},\hat{\eta^{\prime}}}\langle S_{i+\hat{\eta}}^{+}S_{i+\hat{\eta^{\prime}}}^{-}\rangle$ and $C^{z}=(1/Z^{2})\sum_{\hat{\eta},\hat{\eta^{\prime}}}\langle S_{i+\hat{\eta}}^{z}S_{i+\hat{\eta^{\prime}}}^{z}\rangle$. In order to satisfy the sum rule of the correlation function $\langle S^{+}_{i}S^{-}_{i}\rangle=1/2$ in the case without the antiferromagnetic long- range-order, an important decoupling parameter $\alpha$ has been introduced in the above MF calculation feng1 ; feng3 ; guo1 , which can be regarded as the vertex correction. These two equations in Eqs. (4a) and (4b) must be solved simultaneously with other self-consistent equations, then all order parameters, decoupling parameter $\alpha$, and chemical potential $\mu$ are determined by the self-consistent calculation. In the CSS fermion-spin theory feng2 ; feng3 , the electron Green’s function is a convolution of the spin Green’s function and dressed holon Green’s function. Following our previous discussions feng1 ; feng3 ; guo1 , we can obtain the electron diagonal and off-diagonal Green’s functions in the SC state as, $\displaystyle G({\bf k},\omega)$ $\displaystyle=$ $\displaystyle{1\over N}\sum_{{\bf p}}Z_{hF}{B_{{\bf p}}\over 4\omega_{{\bf p}}}\left\\{{\rm coth}[{1\over 2}\beta\omega_{{\bf p}}]\left({U^{2}_{h{\bf p+k}}\over\omega+E_{h{\bf p+k}}-\omega_{{\bf p}}}+{U^{2}_{h{\bf p+k}}\over\omega+E_{h{\bf p+k}}+\omega_{{\bf p}}}\right.\right.$ (9a) $\displaystyle+$ $\displaystyle\left.{V^{2}_{h{\bf p+k}}\over\omega-E_{h{\bf p+k}}+\omega_{{\bf p}}}+{V^{2}_{h{\bf p+k}}\over\omega-E_{h{\bf p+k}}-\omega_{{\bf p}}}\right)+{\rm tanh}[{1\over 2}\beta E_{h{\bf p+k}}]\left({U^{2}_{h{\bf p+k}}\over\omega+E_{h{\bf p+k}}+\omega_{{\bf p}}}\right.$ $\displaystyle-$ $\displaystyle{U^{2}_{h{\bf p+k}}\over\omega+E_{h{\bf p+k}}-\omega_{{\bf p}}}+\left.\left.{V^{2}_{h{\bf p+k}}\over\omega-E_{h{\bf p+k}}-\omega_{{\bf p}}}-{V^{2}_{h{\bf p+k}}\over\omega-E_{h{\bf p+k}}+\omega_{{\bf p}}}\right)\right\\},$ $\displaystyle\Gamma^{\dagger}({\bf k},\omega)$ $\displaystyle=$ $\displaystyle{1\over N}\sum_{{\bf p}}Z_{hF}{\bar{\Delta}_{hZ}({\bf p+k})\over 2E_{h{\bf p+k}}}{B_{{\bf p}}\over 4\omega_{{\bf p}}}\left\\{{\rm coth}[{1\over 2}\beta\omega_{{\bf p}}]\left({1\over\omega-E_{h{\bf p+k}}-\omega_{{\bf p}}}+{1\over\omega-E_{h{\bf p+k}}+\omega_{{\bf p}}}\right.\right.$ (9b) $\displaystyle-$ $\displaystyle\left.{1\over\omega+E_{h{\bf p+k}}+\omega_{{\bf p}}}-{1\over\omega+E_{h{\bf p+k}}-\omega_{{\bf p}}}\right)+{\rm tanh}[{1\over 2}\beta E_{h{\bf p+k}}]\left({1\over\omega-E_{h{\bf p+k}}-\omega_{{\bf p}}}\right.$ $\displaystyle-$ $\displaystyle\left.\left.{1\over\omega-E_{h{\bf p+k}}+\omega_{{\bf p}}}-{1\over\omega+E_{h{\bf p+k}}+\omega_{{\bf p}}}+{1\over\omega+E_{h{\bf p+k}}-\omega_{{\bf p}}}\right)\right\\},$ respectively, where the dressed holon quasiparticle coherence factors $U^{2}_{h{\bf k}}=(1+{\bar{\xi}_{{\bf k}}/E_{h{\bf k}}})/2$ and $V^{2}_{h{\bf k}}=(1-{\bar{\xi}_{{\bf k}}/E_{h{\bf k}}})/2$. These convolutions of the spin Green’s function and dressed holon diagonal and off-diagonal Green’s functions reflect the charge-spin recombination anderson2 . Since the spins center around the $[\pi,\pi]$ point in the MF level feng1 ; feng3 ; guo1 , then the main contributions for the spins comes from the $[\pi,\pi]$ point. In this case, the electron diagonal and off-diagonal Green’s functions in Eqs. (6a) and (6b) can be approximately reduced as the BCS formalism with the d-wave SC gap function in terms of $\omega_{{\bf p}=[\pi,\pi]}\sim 0$ and the equation feng1 ; feng2 $1/2=\langle S_{i}^{+}S_{i}^{-}\rangle=(1/N)\sum_{{\bf p}}B_{{\bf p}}{\rm coth}(\beta\omega_{{\bf p}}/2)/(2\omega_{{\bf p}})$, $\displaystyle G({\bf k},\omega)$ $\displaystyle\approx$ $\displaystyle Z_{F}\left({U^{2}_{{\bf k}}\over\omega-E_{{\bf k}}}+{V^{2}_{{\bf k}}\over\omega+E_{{\bf k}}}\right),$ (10a) $\displaystyle\Gamma^{\dagger}({\bf k},\omega)$ $\displaystyle\approx$ $\displaystyle Z_{F}{\bar{\Delta}_{Z}({\bf k})\over 2E_{{\bf k}}}\left({1\over\omega-E_{{\bf k}}}+{1\over\omega+E_{{\bf k}}}\right),$ (10b) where the electron quasiparticle coherent weight $Z_{F}=Z_{hF}/2$, the electron quasiparticle coherence factors $U^{2}_{{\bf k}}\approx V^{2}_{h{\bf k+k_{A}}}=(1+{\bar{\varepsilon}_{{\bf k}}/E_{{\bf k}}})/2$ and $V^{2}_{{\bf k}}\approx U^{2}_{h{\bf k+k_{A}}}=(1-{\bar{\varepsilon}_{{\bf k}}/E_{{\bf k}}})/2$, with ${\bar{\varepsilon}_{{\bf k}}=Z_{F}\varepsilon}_{{\bf k}}$, $\varepsilon_{{\bf k}}=Zt\chi\gamma_{{\bf k}}+\mu$, and ${\bf k_{A}}=[\pi,\pi]$, $\bar{\Delta}_{Z}({\bf k})=\bar{\Delta}_{hZ}({\bf k})/2$, and electron quasiparticle spectrum $E_{{\bf k}}\approx E_{h{\bf k+k_{A}}}=\sqrt{\bar{\varepsilon}^{2}_{{\bf k}}+\mid\bar{\Delta}_{Z}({\bf k})\mid^{2}}$, i.e., the hole-like dressed holon quasiparticle coherence factors $V_{h{\bf k}}$ and $U_{h{\bf k}}$ and hole-like dressed holon quasiparticle spectrum $E_{h{\bf k}}$ have been transferred into the electron quasiparticle coherence factors $U_{{\bf k}}$ and $V_{{\bf k}}$ and electron quasiparticle spectrum $E_{{\bf k}}$, respectively, by the convolutions of the spin Green’s function and dressed holon Green’s functions due to the charge- spin recombination. This means that within the kinetic energy driven SC mechanism, the dressed holon pairs condense with the d-wave symmetry in a wide range of the hole doping concentration, then the electron Cooper pairs originating from the dressed holon pairing state are due to the charge-spin recombination, and their condensation automatically gives the electron quasiparticle character. For the convenience in the following discussions, these electron Green’s functions in Eq. (7) in the SC state can be expressed in the Nambu representation as, $\displaystyle\tilde{G}({\bf k},\omega)$ $\displaystyle=$ $\displaystyle Z_{F}{1\over\omega^{2}-E^{2}_{{\bf k}}}\left(\begin{array}[]{cc}{\omega+\bar{\varepsilon}_{{\bf k}}}&{\bar{\Delta}_{Z}({\bf k})}\\\ {\bar{\Delta}_{Z}({\bf k})}&{\omega-\bar{\varepsilon}_{{\bf k}}}\end{array}\right)$ (13) $\displaystyle=$ $\displaystyle Z_{F}{\omega\tau_{0}+\bar{\Delta}_{Z}({\bf k})\tau_{1}+\bar{\varepsilon}_{{\bf k}}\tau_{3}\over\omega^{2}-E_{{\bf k}}^{2}}.$ (14) With the helps of this BCS formalism under kinetic energy driven SC mechanism, now we can discuss the effect of the extended impurity scatterers on the quasiparticle transport in cuprate superconductors. In the presence of impurities, the unperturbed electron Green’s function in Eq. (8) is dressed via the impurity scattering durst ; nunner , $\displaystyle\tilde{G}_{I}({\bf k},\omega)^{-1}=\tilde{G}({\bf k},\omega)^{-1}-\tilde{\Sigma}({\bf k},\omega),$ (15) with the self-energy function $\tilde{\Sigma}({\bf k},\omega)=\sum_{\alpha}\Sigma_{\alpha}({\bf k},\omega)\tau_{\alpha}$. It has been shown that all but the scalar component of the self-energy function can be neglected or absorbed into $\bar{\Delta}_{Z}({\bf k})$ durst ; nunner . In this case, the dressed electron Green’s function in Eq. (9) can be explicitly rewritten as, $\displaystyle\tilde{G}_{I}({\bf k},\omega)=Z_{F}{[\omega-\Sigma_{0}({\bf k},\omega)]\tau_{0}+\bar{\Delta}_{Z}({\bf k})\tau_{1}+[\bar{\varepsilon}_{{\bf k}}+\Sigma_{3}({\bf k},\omega)]\tau_{3}\over[\omega-\Sigma_{0}({\bf k},\omega)]^{2}-\bar{\varepsilon}^{2}_{{\bf k}}-\bar{\Delta}^{2}_{Z}({\bf k})}.$ (16) Based on the phenomenological d-wave BCS-type electron Green’s function durst , the energy and temperature dependence of the microwave conductivity of cuprate superconductors has been fitted nunner , where the electron self- energy functions $\Sigma_{0}({\bf k},\omega)$ and $\Sigma_{3}({\bf k},\omega)$ have been treated within the framework of the T-matrix approximation. Following their discussions durst ; nunner , the electron self-energy function $\tilde{\Sigma}({\bf k},\omega)$ can be obtained approximately as, $\displaystyle\tilde{\Sigma}({\bf k},\omega)=\rho_{i}\tilde{T}_{{\bf k}{\bf k}}(\omega),$ (17) with $\rho_{i}$ is the impurity concentration, and $\tilde{T}_{{\bf k}{\bf k}}(\omega)$ is the diagonal element of the T-matrix, $\displaystyle\tilde{T}_{{\bf k}{\bf k}^{\prime}}(\omega)=V_{{\bf k}{\bf k}^{\prime}}\tau_{3}+\sum_{{\bf k}^{\prime\prime}}V_{{\bf k}{\bf k}^{\prime\prime}}\tau_{3}\tilde{G}_{I}({\bf k}^{\prime\prime},\omega)\tilde{T}_{{\bf k}^{\prime\prime}{\bf k}^{\prime}}(\omega),$ (18) where $V_{{\bf k}{\bf k}^{\prime}}$ is the impurity scattering potential. As mentioned in Sec. I, there is no gap to the quasiparticle excitations at the four nodes for the d-wave SC state of cuprate superconductors, therefore the quasiparticles are generated only around these four nodes. It has been shown durst that this characteristic feature is very useful when considering the impurity scattering, since the initial and final momenta of a scattering event must always be approximately equal to the ${\bf k}$-space location of one of the four nodes in the zero temperature and zero energy, while the impurity scattering potential $V_{{\bf k}{\bf k}^{\prime}}$ varies slowly over the area of a node. In this case, a general scattering potential $V_{{\bf k}{\bf k}^{\prime}}$ need only be evaluated in three possible case: the intranode impurity scattering $V_{{\bf k}{\bf k}^{\prime}}=V_{1}$ (${\bf k}$ and ${\bf k}^{\prime}$ at the same node), the adjacent-node impurity scattering $V_{{\bf k}{\bf k}^{\prime}}=V_{2}$ (${\bf k}$ and ${\bf k}^{\prime}$ at the adjacent nodes), and the opposite-node impurity scattering $V_{{\bf k}{\bf k}^{\prime}}=V_{3}$ (${\bf k}$ and ${\bf k}^{\prime}$ at the opposite nodes), then the impurity scattering potential $V_{{\bf k}{\bf k}^{\prime}}$ in the T-matrix can be effectively reduced as durst , $\displaystyle V_{{\bf k}{\bf k}^{\prime}}\rightarrow\underline{V}=\left(\begin{array}[]{cccc}V_{1}&V_{2}&V_{3}&V_{2}\\\ V_{2}&V_{1}&V_{2}&V_{3}\\\ V_{3}&V_{2}&V_{1}&V_{2}\\\ V_{2}&V_{3}&V_{2}&V_{1}\end{array}\right)\,.$ (23) At the zero temperature and zero energy, these nodes reduce to points. In this case, this nodal approximation for the impurity potential can reproduce any impurity potential. However, at finite temperatures and energies, there is a limitation on the forward scattering character of the impurity potential because this nodal approximation assumes the Brillouin zone quadrant around a particular node durst . It has been shown nunner that although the strict forward scattering limit can therefore not be reached at finite temperatures and energies, this nodal approximation is still appropriate to treat the intermediate range scatters. Therefore in the following discussions, we employ the simplified impurity scattering potential in Eq. (13) to study the impurity scattering effect on the quasiparticle transport of cuprate superconductors. Substituting Eq. (13) into Eq. (12), the T-matrix can be obtained as a $4\times 4$-matrix around the nodal points, $\displaystyle\tilde{T}_{jj^{\prime}}(\omega)=V_{jj^{\prime}}\tau_{3}+\tilde{I}_{G}(\omega)\tau_{3}\sum_{j^{\prime\prime}}V_{jj^{\prime\prime}}\tilde{T}_{j^{\prime\prime}j^{\prime}}(\omega),$ (24) where $\tilde{I}_{G}(\omega)$ is the integral of the electron Green’s function, and can be obtained as, $\displaystyle\tilde{I}_{G}(\omega)={1\over N}\sum_{{\bf k}}\tilde{G}_{I}({\bf k},\omega)\approx\tilde{G}_{I0}(\omega)\tau_{0}+\tilde{G}_{I3}(\omega)\tau_{3},$ (25) with $\tilde{G}_{I0}(\omega)$ and $\tilde{G}_{I3}(\omega)$ are given by, $\displaystyle\tilde{G}_{I0}(\omega)$ $\displaystyle=$ $\displaystyle{1\over N}\sum_{{\bf k}}Z_{F}{\omega-\Sigma_{0}(\omega)\over\omega^{2}-E^{2}_{{\bf k}}},$ (26a) $\displaystyle\tilde{G}_{I3}(\omega)$ $\displaystyle=$ $\displaystyle{1\over N}\sum_{{\bf k}}Z_{F}{\bar{\varepsilon}_{{\bf k}}+\Sigma_{3}(\omega)\over\omega^{2}-E^{2}_{{\bf k}}},$ (26b) and the self-energy functions $\Sigma_{0}(\omega)$ and $\Sigma_{0}(\omega)$ are evaluated as, $\displaystyle\Sigma_{0}(\omega)$ $\displaystyle=$ $\displaystyle{\rho_{i}\over 4}\left({2\tilde{G}_{I0}(\omega)V_{13}^{2}\over[1-\tilde{G}_{I3}(\omega)V_{13}]^{2}-[\tilde{G}_{I0}(\omega)V_{13}]^{2}}+{\tilde{G}_{I0}(\omega)(V^{-}_{123})^{2}\over[1-\tilde{G}_{I3}(\omega)V^{-}_{123}]^{2}-[\tilde{G}_{I0}(\omega)V^{-}_{123}]^{2}}\right.$ (27a) $\displaystyle+$ $\displaystyle\left.{\tilde{G}_{I0}(\omega)(V^{+}_{123})^{2}\over[1-\tilde{G}_{I3}(\omega)V^{+}_{123}]^{2}-[\tilde{G}_{I0}(\omega)V^{+}_{123}]^{2}}\right),$ $\displaystyle\Sigma_{3}(\omega)$ $\displaystyle=$ $\displaystyle{\rho_{i}\over 4}\left({2V_{13}[1-\tilde{G}_{I3}(\omega)V_{13}]\over[1-\tilde{G}_{I3}(\omega)V_{13}]^{2}-[\tilde{G}_{I0}(\omega)V_{13}]^{2}}+{V^{-}_{123}[1-\tilde{G}_{I3}(\omega)V^{-}_{123}]\over[1-\tilde{G}_{I3}(\omega)V^{-}_{123}]^{2}-[\tilde{G}_{I0}(\omega)V^{-}_{123}]^{2}}\right.$ (27b) $\displaystyle+$ $\displaystyle\left.{V^{+}_{123}[1-\tilde{G}_{I0}(\omega)V^{+}_{123}]\over[1-\tilde{G}_{I3}(\omega)V^{+}_{123}]^{2}-[\tilde{G}_{I0}(\omega)V^{+}_{123}]^{2}}\right),$ where $V_{13}=V_{1}-V_{3}$, $V^{-}_{123}=V_{1}-2V_{2}+V_{3}$, and $V^{+}_{123}=V_{1}+2V_{2}+V_{3}$. In the framework of the linear response theory, the microwave conductivity of cuprate superconductors can be calculated by means of the Kubo formula as mahan , $\displaystyle\sigma(\omega,T)=-{{\rm Im}\Pi(\omega,T)\over\omega},$ (28) with $\Pi(\omega,T)$ is the electron current-current correlation function in the SC state, and can be expressed as, $\displaystyle\Pi(\tau-\tau^{\prime})=-<T_{\tau}{\bf J}(\tau)\cdot{\bf J}(\tau^{\prime})>.$ (29) In the CSS fermion-spin representation feng2 ; feng3 , the electron polarization operator can be evaluated as, $\displaystyle{\bf P}=\sum_{i}{\bf R}_{i}\hat{n}_{i}=\sum_{i,\sigma}{\bf R}_{i}\hat{C}^{{\dagger}}_{i,\sigma}\hat{C}_{i,\sigma}={1\over 2}\sum_{i,\sigma}{\bf R}_{i}h_{i,\sigma}h^{{\dagger}}_{i,\sigma},~{}~{}~{}$ (30) then within the $t$-$J$ model (2), the current density of electrons is obtained by the time derivation of this polarization operator using the Heisenberg’s equation of motion as, $\displaystyle{\bf J}=ie[H,{\bf P}]=i{1\over 2}et\sum_{i\hat{\eta}}\hat{\eta}(h^{\dagger}_{i+\hat{\eta}\uparrow}h_{i\uparrow}S_{i}^{+}S^{-}_{i+\hat{\eta}}+h^{\dagger}_{i+\hat{\eta}\downarrow}h_{i\downarrow}S^{\dagger}_{i}S^{-}_{i+\eta})=-i{1\over 2}et\sum_{i\eta\sigma}\hat{\eta}\hat{C}^{\dagger}_{i\sigma}\hat{C}_{i+\hat{\eta}\sigma}\approx{ev_{f}\over\sqrt{2}}\sum_{{\bf k}\sigma}{\bf k}\hat{C}^{\dagger}_{{\bf k}\sigma}\hat{C}_{{\bf k}\sigma},$ (31) with $v_{f}=\sqrt{2}t$ is the electron velocity at the nodal points. According to this current density (21), the current-current correlation function in Eq. (19) can be obtained as, $\displaystyle\Pi(i\omega_{n},T)$ $\displaystyle=$ $\displaystyle{e^{2}v_{f}^{2}\over 2}{1\over N}\sum_{{\bf k}}{1\over\beta}\sum_{i\omega_{n}^{\prime}}{\rm Tr}[\tilde{G}({\bf k},i\omega_{n}^{\prime})$ (32) $\displaystyle\times$ $\displaystyle\tilde{G}({\bf k},i\omega_{n}^{\prime}+i\omega_{n})\tilde{\Gamma}({\bf k},i\omega_{n}^{\prime},i\omega_{n})],$ with $\tilde{\Gamma}({\bf k},\omega^{\prime},\omega)$ is the vertex function, and can be evaluated explicitly as the sum of ladder diagrams, $\displaystyle\Pi(i\omega_{n})=e^{2}v^{2}_{f}{1\over\beta}\sum_{i\omega_{n}^{\prime}}J(i\omega_{n},i\omega_{n}^{\prime}),$ (33) where the kernel function $J(\omega,\omega^{\prime})$ is expressed as, $\displaystyle J(\omega,\omega^{\prime})={I^{(0)}_{0}+L_{A}[I^{(0)}_{0}I_{3}^{(3)}+I_{0}^{(3)}I_{3}^{(0)}]\over[1-(L_{A}I_{0}^{(0)}+L_{B}I^{(0)}_{3})][1-(L_{A}I_{3}^{(3)}+L_{B}I^{(3)}_{0})]-[L_{A}I_{0}^{(3)}+L_{B}I^{(3)}_{3}][L_{A}I_{3}^{(0)}+L_{B}I^{(0)}_{0}]},$ (34) with the functions, $\displaystyle L_{A}(\omega,\omega^{\prime})$ $\displaystyle=$ $\displaystyle[T_{11}^{(0)}(\omega)T_{11}^{(0)}(\omega+\omega^{\prime})+T_{11}^{(3)}(\omega)T_{11}^{(3)}(\omega+\omega^{\prime})-T_{13}^{(0)}(\omega)T_{13}^{(0)}(\omega+\omega^{\prime})-T_{13}^{(3)}(\omega)T_{13}^{(3)}(\omega+\omega^{\prime})],$ (35a) $\displaystyle L_{B}(\omega,\omega^{\prime})$ $\displaystyle=$ $\displaystyle[T_{11}^{(0)}(\omega)T_{11}^{(3)}(\omega+\omega^{\prime})+T_{11}^{(3)}(\omega)T_{11}^{(0)}(\omega+\omega^{\prime})-T_{13}^{(0)}(\omega)T_{13}^{(3)}(\omega+\omega^{\prime})-T_{13}^{(3)}(\omega)T_{13}^{(0)}(\omega+\omega^{\prime})],$ (35b) while the functions $I^{(0)}_{0}(\omega,\omega^{\prime})$ and $I^{(0)}_{3}(\omega,\omega^{\prime})$ are give by, $\displaystyle I^{(0)}_{0}(\omega,\omega^{\prime})\tau_{0}$ $\displaystyle+$ $\displaystyle I^{(3)}_{0}(\omega,\omega^{\prime})\tau_{3}={1\over N}\sum_{{\bf k}}\tilde{G}_{I}({\bf k},\omega)\tilde{G}_{I}({\bf k},\omega+\omega^{\prime}),$ (36a) $\displaystyle I^{(0)}_{3}(\omega,\omega^{\prime})\tau_{0}$ $\displaystyle+$ $\displaystyle I^{(3)}_{3}(\omega,\omega^{\prime})\tau_{3}={1\over N}\sum_{{\bf k}}\tilde{G}_{I}({\bf k},\omega)\tilde{\tau}_{3}\tilde{G}_{I}({\bf k},\omega+\omega^{\prime}).$ (36b) Substituting Eq. (23) into Eq. (18), the microwave conductivity of cuprate superconductors is obtained explicitly as, $\displaystyle\sigma(\omega,T)$ $\displaystyle=$ $\displaystyle e^{2}v^{2}_{f}\int^{\infty}_{-\infty}{d\omega^{\prime}\over 2\pi}{n_{F}(\omega^{\prime})-n_{F}(\omega^{\prime}+\omega)\over\omega^{\prime}}[{\rm Re}J(\omega^{\prime}-i\delta,\omega^{\prime}+\omega+i\delta)-{\rm Re}J(\omega^{\prime}+i\delta,\omega^{\prime}+\omega+i\delta)].$ (37) We emphasize that based on the nodal approximation of the quasiparticle excitations and scattering processes durst ; nunner , this microwave conductivity of cuprate superconductors in Eq. (27) is obtained within the kinetic energy driven SC mechanism, although its expression is similar to that obtained within the phenomenological BCS formalism with the d-wave SC gap function durst ; nunner . ## III Microwave conductivity of cuprate superconductors In cuprate superconductors, although the values of $J$ and $t$ is believed to vary somewhat from compound to compound, however, as a qualitative discussion, the commonly used parameters in this paper are chosen as $t/J=2.5$, with an reasonably estimative value of $J\sim 1000$K. We are now ready to discuss the doping, energy, and temperature dependence of the quasiparticle transport of cuprate superconductors in the SC state with extended impurities. We have performed a calculation for the microwave conductivity $\sigma(\omega,T)$ in Eq. (27), and the results of $\sigma(\omega,T)$ as a function of energy with temperature $T=0.002J=2$K (solid line), $T=0.004J=4$K (dashed line), $T=0.008J=8$K (dash-dotted line), and $T=0.01J=10$K (dotted line) under the slightly strong impurity scattering potential with $V_{1}=58J$, $V_{2}=49.32J$, and $V_{3}=40.6J$ at the impurity concentration $\rho=0.000014$ for the doping concentration $\delta=0.15$ are plotted in Fig. 1 in comparison with the corresponding experimental results of cuprate superconductors in the SC state harris (inset). Obviously, the energy evolution of the microwave conductivity of cuprate superconductor harris is qualitatively reproduced. In particular, a low temperature cusplike shape of the microwave conductivity is obtained for cuprate superconductors in the presence of the impurity scattering. At the low energy regime ($\omega<0.0002J$), this low temperature microwave conductivity $\sigma(\omega,T)$ rises rapidly from the universal zero-temperature limit to a much larger microwave conductivity. However, this low temperature microwave conductivity $\sigma(\omega,T)$ becomes smaller and varies from weakly energy dependence at the intermediate energy regime ($0.0002J<\omega<0.0008J$), to the almost energy independence at the high energy regime ($\omega>0.0008J$). Figure 1: The microwave conductivity as a function of energy with $T=0.002J=2$K (solid line), $T=0.004J=4$K (dashed line), $T=0.008J=8$K (dash- dotted line), and $T=0.01J=10$K (dotted line) at $\rho=0.000014$ for $t/J=2.5$, $V_{1}=58J$, $V_{2}=49.32J$, and $V_{3}=40.6J$ in $\delta=0.15$. Inset: the corresponding experimental result of cuprate superconductors in the SC state taken from Ref. [14]. Figure 2: The microwave conductivity as a function of doping with $T=0.002J=$K and $\omega=0.000087J\approx 1.81$GHz at $\rho=0.000014$ for $t/J=2.5$, $V_{1}=58J$, $V_{2}=49.32J$, and $V_{3}=40.6J$. The dashed line is the corresponding result of the superconducting gap parameter. For a better understanding of the physical properties of the microwave conductivity $\sigma(\omega,T)$ in cuprate superconductors, we have studied the doping evolution of the microwave conductivity, and the result of $\sigma(\omega,T)$ as a function of doping with temperature $T=0.002J=2$K and energy $\omega=0.000087J\approx 1.81$GHz under the slightly strong impurity scattering potential with $V_{1}=58J$, $V_{2}=49.32J$, and $V_{3}=40.6J$ at the impurity concentration $\rho=0.000014$ is plotted in Fig. 2 (solid line). For comparison, the corresponding result of the SC gap parameter of cuprate superconductors is also shown in the same figure (dashed line). Our result shows that in contrast to the dome shape of the doping dependent SC gap parameter huefner ; feng1 ; feng3 ; guo1 , the microwave conductivity $\sigma(\omega,T)$ decreases with increasing doping in the underdoped regime, and reaches a minimum in the optimal doping, then increases in the overdoped regime. This doping dependent behavior of the low energy microwave conductivity $\sigma(\omega,T)$ at low temperatures is also qualitatively consistent with the universal microwave conductivity limit palee $\sigma\propto 1/\Delta$ at low energy as temperature $T\rightarrow 0$, if this SC gap parameter $\Delta$ in the phenomenological BCS formalism palee has the similar dome shape doping dependence. Figure 3: The microwave conductivity as a function of temperature with energy $\omega\approx 1.14$GHz (solid line), $\omega\approx 2.28$GHz (long dashed line), $\omega\approx 13.4$GHz (dash-dotted line), $\omega\approx 22.8$GHz (dotted line), and $\omega\approx 75.3$GHz (short dashed line) at $\rho=0.000014$ for $t/J=2.5$, $V_{1}=58J$, $V_{2}=49.32J$, and $V_{3}=40.6J$ in $\delta=0.15$. Inset: the corresponding experimental result of cuprate superconductors in the SC state taken from Ref. [14]. In the above discussions, we mainly study the effect of the extended impurity scatterers on the quasiparticle transport at low temperatures ($T\ll T_{c}$) and low energies ($\omega\ll\Delta$). Now we discuss the temperature dependence of the quasiparticle transport of cuprate superconductors, where $T$ may approach to $T_{c}$ from low temperature side. In this case, it has been shown nunner that the inelastic scattering process should be considered at higher temperatures, such as the quasiparticle-quasiparticle scattering. This is followed from the fact that the inelastic quasiparticle-quasiparticle scattering process is suppressed at low temperatures due to the large SC gap parameter in the quasiparticle excitation spectrum. However, the contribution from this inelastic quasiparticle-quasiparticle scattering process is increased rapidly when $T$ approaches to $T_{c}$ from low temperature side, since there is a small SC gap parameter near $T_{c}$. In particular, it has been pointed out walker that the contribution from the quasiparticle- quasiparticle scattering process to the transport lifetime is exponentially suppressed at low temperatures, and therefore the effect of this inelastic quasiparticle-quasiparticle scattering can be considered by adding the inverse transport lifetime duffy $\tau^{-1}_{{\rm inel}}(T)$ to the imaginary part of the self-energy function $\Sigma_{0}(\omega)$ in Eq. (17a), then the total self-energy function $\Sigma^{{\rm tot}}_{0}(\omega)$ can be expressed as walker ; nunner , $\displaystyle\Sigma^{{\rm tot}}_{0}(\omega)=\Sigma_{0}(\omega)-i[2\tau_{{\rm inel}}(T)]^{-1},$ (38) with $\tau_{{\rm inel}}(T)$ has been chosen as $[2\tau_{{\rm inel}}(T)]^{-1}=91.35(T-0.005)^{4}J$. Using this total self-energy function $\Sigma^{{\rm tot}}_{0}(\omega)$ to replace $\Sigma_{0}(\omega)$ in Eq. (27), we have performed a calculation for the microwave conductivity $\sigma(\omega,T)$, and the results of $\sigma(\omega,T)$ as a function of temperature $T$ with energy $\omega=0.0000547J\approx 1.14$GHz (solid line), $\omega=0.0001094J\approx 2.28$GHz (long dashed line), $\omega=0.0006564J\approx 13.4$GHz (dash-dotted line), $\omega=0.001094J\approx 22.8$GHz (dotted line), and $\omega=0.0036102J\approx 75.3$GHz (short dashed line) under the slightly strong impurity scattering potential with $V_{1}=58J$, $V_{2}=49.32J$, and $V_{3}=40.6J$ at the impurity concentration $\rho=0.000014$ for the doping concentration $\delta=0.15$ are plotted in Fig. 3 in comparison with the corresponding experimental results of cuprate superconductors in the SC state harris (inset). Our results show that the temperature dependent microwave conductivity $\sigma(\omega,T)$ increases rapidly with increasing temperatures to a large broad peak around temperature $T\approx 0.25J\approx 25$K for energy $\omega=0.0000547J\approx 1.14$GHz, and then falls roughly linearly. However, this broad peak in the microwave conductivity spectrum resulting from the temperature dependent impurity scattering rate in Eq. (28) is energy dependent, and moves to higher temperatures with increasing energies, in qualitative agreement with the experimental data harris . Within the framework of the kinetic energy driven d-wave cuprate superconductivity feng1 , our present results of the energy and temperature dependence of the microwave conductivity by considering the effect of the extended impurity scatterers are qualitatively similar to the earlier attempts to fit the experimental data by using a phenomenological d-wave BCS formalism palee ; hirschfeld ; quinlan ; hettler ; durst ; duffy ; nunner . Establishing this agreement is important to confirming the nature of the SC phase of cuprate superconductors as the d-wave BCS-like SC state within the kinetic energy driven SC mechanism. It has been shown turner ; bonn ; lee ; hosseini ; harris that there are some subtle differences for different families of cuprate superconductors, and these subtle differences may be induced by the other effects except the impurity scattering. However, we in this paper are primarily interested in exploring the general notion of the effect of the extended impurity scatterers on the kinetic energy driven cuprate superconductors in the SC state. The qualitative agreement between the present theoretical results and experimental data also show that the presence of impurities has a crucial effect on the microwave conductivity of cuprate superconductors. ## IV Summary In conclusion we have shown very clearly in this paper that if the effect of the extended impurity scatterers is taken into account in the framework of the kinetic energy driven d-wave superconductivity, the microwave conductivity of the $t$-$J$ model calculated based on the nodal approximation of the quasiparticle excitations and scattering processes per se can correctly reproduce some main features found in the microwave conductivity measurements on cuprate superconductor in the SC state turner ; bonn ; lee ; hosseini ; harris , including the energy and temperature dependence of the microwave conductivity spectrum. The theory also predicts a V-shaped doping dependent microwave conductivity, which is in contrast with the dome shape of the doping dependent SC gap parameter, and therefore should be verified by further experiments. ###### Acknowledgements. This work was supported by the National Natural Science Foundation of China under Grant No. 10774015, and the funds from the Ministry of Science and Technology of China under Grant Nos. 2006CB601002 and 2006CB921300. ## References * (1) To whom correspondence should be addressed. * (2) See, e.g., the review, C. C. Tsuei and J. R. Kirtley, Rev. Mod. Phys. 72, 969 (2000). * (3) J. Bardeen, L. N. Cooper, and J. R. Schrieffer, Phys. Rev. 108, 1175 (1957); J. R. Schrieffer, Theory of Superconductivity, Benjamin, New York, 1964. * (4) P. W. Anderson, in Frontiers and Borderlines in Many Particle Physics, edited by R. A. Broglia and J. R. Schrieffer (North-Holland, Amsterdam, 1987), p. 1; Science 235, 1196 (1987). * (5) See, e.g., the review, A. Damascelli, Z. Hussain, and Z. X. Shen, Rev. Mod. Phys. 75, 473 (2003), and references therein. * (6) H. Matsui, T. Sato, T. Takahashi, S. C. Wang, H. B. Yang, H. Ding, T. Fujii, T. Watanabe, and A. Matsuda, Phys. Rev. Lett. 90, 217002 (2003). * (7) J. C. Campuzano, H. Ding, M. R. Norman, M. Randeira, A. F. Bellman, T. Yokoya, T. Takahashi, H. Katayama-Yoshida, T. Mochiku, and K. Kadowaki, Phys. Rev. B 53, R14737 (1996). * (8) P. J. Turner, R. Harris, Saeid Kamal, M. E. Hayden, D. M. Broun, D. C. Morgan, A. Hosseini, P. Dosanjh, G. K. Mullins, J. S. Preston, Ruixing Liang, D. A. Bonn, and W. N. Hardy, Phys. Rev. Lett. 90, 237005 (2003). * (9) P. A. Lee, Phys. Rev. Lett. 71, 1887 (1993). * (10) P. J. Hirschfeld, W. O. Putikka, and D. J. Scalapino, Phys. Rev. Lett. 71, 3705 (1993); Phys. Rev. B 50, 10250 (1994). * (11) See, e.g., the review, H. Alloul, J. Bobroff, M. Gabay, and P. J. Hirschfeld, arXiv: 0711.0877 (unpublished). * (12) D. A. Bonn, R. Liang, T. M. Riseman, D. J. Baar, D. C. Morgan, K. Zhang, P. Dosanjh, T. L. Duty, A. MacFarlane, G. D. Morris, J. H. Brewer, W. N. Hardy, C. Kallin, and A. J. Berlinsky, Phys. Rev. B 47, 11314 (1993). * (13) Shih-Fu Lee, D. C. Morgan, R. J. Ormeno, D. M. Broun, R. A. Doyle, J. R. Waldram, and K. Kadowaki, Phys. Rev. Lett. 77, 735 (1996). * (14) A. Hosseini, R. Harris, Saeid Kamal, P. Dosanjh, J. Preston, Ruixing Liang, W. N. Hardy, and D. A. Bonn, Phys. Rev. B 60, 1349 (1999). * (15) R. Harris, P. J. Turner, Saeid Kamal, A. R. Hosseini, P. Dosanjh, G. K. Mullins, J. S. Bobowski, C. P. Bidinosti, D. M. Broun, Ruixing Liang, W. N. Hardy, and D. A. Bonn, Phys. Rev. B 74, 104508 (2006). * (16) S. M. Quinlan, P. J. Hirschfeld, and D. J. Scalapino, Phys. Rev. B 53, 8575 (1996). * (17) M. H. Hettler and P. J. Hirschfeld, Phys. Rev. B 59, 9606 (1999); Phys. Rev. B 61, 11313 (2000). * (18) A. C. Durst and P. A. Lee, Phys. Rev. B 62, 1270 (2000). * (19) D. Duffy, P. J. Hirschfeld, and D. J. Scalapino, Phys. Rev. B 64, 224522 (2001). * (20) Tamara S. Nunner and P. J. Hirschfeld, Phys. Rev. B 72, 014514 (2005). * (21) Shiping Feng, Phys. Rev. B 68, 184501 (2003); Shiping Feng, Tianxing Ma, and Huaiming Guo, Physica C 436, 14 (2006); Shiping Feng and Tianxing Ma, in Superconductivity Research Horizons, edited by Eugene H. Peterson (Nova Science Publishers, New York, 2007) chapter 5, p. 129, and references therein. * (22) See, e.g., the review, M. A. Kastner, R. J. Birgeneau, G. Shirane, and Y. Endoh, Rev. Mod. Phys. 70, 897 (1998), and references therein. * (23) C. Gros, R. Joynt, and T.M. Rice, Phys. Rev. B 36, 381 (1987). * (24) P. W. Anderson, arXiv:0709.0656 (unpublished). * (25) Shiping Feng, Jihong Qin, and Tianxing Ma, J. Phys. Condens. Matter 16, 343 (2004); Shiping Feng, Tianxing Ma, and Jihong Qin, Mod. Phys. Lett. B 17, 361 (2003). * (26) See, e.g., the review, Shiping Feng, Huaiming Guo, Yu Lan, and Li Cheng, Int. J. Mod. Phys. B, in press, arXiv: 0710.5258. * (27) R. B. Laughlin, Phys. Rev. Lett. 79, 1726 (1997); J. Low. Tem. Phys. 99, 443 (1995). * (28) Huaiming Guo and Shiping Feng, Phys. Lett. A 361, 382 (2007); Shiping Feng and Tianxing Ma, Phys. Lett. A 350, 138 (2006); Yu Lan, Jihong Qin, and Shiping Feng, Phys. Rev. B 76, 014533 (2007). * (29) P. Dai, H. A. Mook, R. D. Hunt, and F. Dog̃an, Phys. Rev. B 63, 054525 (2001); P. Bourges, B. Keimer, S. Pailhés, L. P. Regnault, Y. Sidis, and C. Ulrich, Physica C 424, 45 (2005). * (30) M. Arai, T. Nishijima, Y. Endoh, T. Egami, S. Tajima, K. Tomimoto, Y. Shiohara, M. Takahashi, A. Garret, and S. M. Bennington, Phys. Rev. Lett. 83, 608 (1999); S. M. Hayden, H. A. Mook, P. Dai, T. G. Perring, and F. Dog̃an, Nature 429, 531 (2004); C. Stock, W. J. Buyers, R. A. Cowley, P. S. Clegg, R. Coldea, C. D. Frost, R. Liang, D. Peets, D. Bonn, W. N. Hardy, and R. J. Birgeneau, Phys. Rev. B 71, 024522 (2005). * (31) P. W. Anderson, Phys. Rev. Lett. 67, 2092 (1991); Science 288, 480 (2000); Physica C 341-348, 9 (2000); cond-mat/0108522 (unpublished). * (32) See, e.g., G.D. Mahan, Many Particle Physics, (Plenum Press, New York, 1981). * (33) S. Huefner, M. A. Hossain, A. Damascelli, G. A. Sawatzky, arXiv:0706.4282 (unpublished). * (34) M. B. Walker and M. F. Smith, Phys. Rev. B 61, 11285 (2000).

arxiv-papers

2024-09-04T02:48:53.734311

0802.1763

# Approximating the ground state of fermion system by multiple determinant states: matching pursuit approach Quanlin Jie qljie@whu.edu.cn Department of Physics, Wuhan University, Wuhan 430072, P. R. China ###### Abstract We present a simple and stable numerical method to approximate the ground state of a quantum many-body system by multiple determinant states. This method searches these determinant states one by one according to the matching pursuit algorithm. The first determinant state is identical to that of the Hartree-Fock theory. Calculations for two-dimensional Hubbard model serve as a demonstration. First principle calculation, Quantum many body theory, matching pursuit method, Hartree Fock approximation ###### pacs: 02.70.-c, 31.15.Ar, 71.15.-m Searching a single determinant state to approximate a quantum ground state, namely, the Hartree-Fock (HF) algorithm, plays important role in the understanding of nuclear, atomic, and molecular structures. It is a long standing effort to extend the HF theory into a truly first principle method by searching multiple determinant states to span a quantum state, for recent examples, see 1 ; 2 ; 10a ; 18 ; 7 and references therein. The attracting feature is that this approach is very stable and free from the sign problem. It in principle can apply to a wide variety of systems. However, first principle calculation in terms of multiple determinant states is still a challenge. In fact, including multiple determinant states in the variational treatment, which is the common approach, often results in very complicated formulations. The computation cost is usually impractically demanding. Some realistic implementations impose restrictions on the determinant states. For example, the Multi-configuration Hartree-Fock theory 1 , a time dependent extension to the HF theory, requires single particle states to be orthogonal with each others. Here, we use a new approach to approximate the quantum ground state via multiple determinant states. Here, based on the matching pursuit (MP) algorithm 3 ; 4 ; 5 ; 6 , we show a numerical method to search determinant states to span the ground state of a fermion system. The determinant states are found one by one from all possible determinant states. Searching the first determinant state is identical to the Hartree-Fock theory. A significant feature of the current method is that several tens of basis determinant states are enough for reasonable result, and one can reach high accuracy by searching one or two thousands basis states. These numbers of basis states are several orders smaller than that of the Stochastic diagonalization algorithm 7 , which searches orthogonal determinant basis states stochastically to span a quantum wave function. In comparison with other algorithms of search determinant states to span a fermion ground state, such as the Path-Integral Renormalization Group (PIRG) algorithm 10a ; 18 , the current MP based method is quite simple and efficient. The MP algorithm is originally designed for signal processing 3 . It is now popular on the engineering community for coding, analysis, and compression of video and audio data 4 ; 5 ; 6 . This algorithm searches some basis states from an over-complete basis set to represent a sequence of data. The basis states are found one by one. The convergence of the MP algorithm is proved mathematically. For sufficient redundancy of the over-complete basis set, the convergence can be exponential 6 . The MP algorithm is insensible to the dimension of the data, and thus promises applications in quantum many-body systems. In Ref. 8 , the authors employ this algorithm to propagate quantum wave functions via split operator method in the Gaussian wave packet basis. A encouraging result is that several tens of Gaussian wave packets are able to accurately represent quantum wave function of a 20-dimensional model. The goal of MP algorithm is to obtain a sparse representation of a signal. To represent a quantum many-body wave function, $\psi$, the MP algorithm searches an over-complete basis set and finds some basis states, $\phi_{1}$, $\phi_{2}$, $\cdots$, $\phi_{n}$, such that the combination of the basis states, $\psi_{n}=\alpha_{1}\phi_{1}+\cdots+\alpha_{n}\phi_{n}$ can best approach the state $\psi$. Mathematically, that is to require $\psi_{n}$ has minimum distance with $\psi$, i.e. $|\psi-\psi_{n}|$ reaches minimum. The basis states are found one by one. At $k$-th step, the basis state $\phi_{k}$ is obtained such that the combination of the basis states $\psi_{k}=\alpha_{1}\phi_{1}+\cdots+\alpha_{k}\phi_{k}$ has minimum distance with the state $\psi$, i.e. $|\psi-\psi_{k}|$ has minimum for all possible choice of $\phi_{k}$. Each more step brings the $\psi_{k}$ closer to the target state $\psi$, i.e., the distance $|\psi-\psi_{k}|$ decreases with $k$. The eigenvalue problem is equivalent to find minimum values of the Rayleigh quotient $E={\langle\psi|H|\psi\rangle}/{\langle\psi|\psi\rangle},$ (1) where $H$ is the Hamiltonian and $\psi$ is the trial wave function. Calculation of the ground state by the MP algorithm is to search some basis states to span the ground state. The basis states are found one by one from an over-complete basis set. Each searching process obtains one basis state such that the combination of this basis state and those already found ones minimizes the Rayleigh quotient for all possible choice of the current basis state. This process of finding a new basis state continues until convergence of $E$. Without loss of generality, in the following discussions, we focus on fermion systems, and use all possible Slater determinant states as over- complete basis set. Note that, for fermion systems, the first step is to find a Slater determinant state that minimizes the Rayleigh quotient. This is just the well known Hartree-Fock approximation. We employ an iterative method to search a new determinant state, including the first one. We denote the single particle basis states as $|i\rangle$, ($i=1,\cdots,n$), and $a^{+}_{i}$ ($a_{i}$) the operator for creation (annihilation) of the state $|i\rangle$, i.e., $|i\rangle=a^{+}_{i}|0\rangle$ with $|0\rangle$ the vacuum state. A determinant state can be expressed as $|\phi\rangle=\prod_{j=1}^{m}F^{+}_{j}|0\rangle,$ (2) where $m$ is particle number and $F^{+}_{j}$ ($F_{j}$) is creation (annihilation) operator for single particle state, $F^{+}_{j}=c_{1j}a^{+}_{1}+\cdots+c_{nj}a^{+}_{n}$. Searching for the determinant state $|\phi\rangle$ is equivalent to find the coefficients $\\{c_{ij}\\}$ (or the operators $\\{F^{+}_{j}\\}$). We use an iterative relaxation procedure to search the operators $\\{F^{+}_{j}\\}$. From an initial trial state in the form of (2) which can be chosen randomly, we optimize $F^{+}_{1}$, $F^{+}_{2}$, $\cdots$, $F^{+}_{m}$ consecutively. Each step of the optimization lowers the Rayleigh quotient. This iteration continues until the convergence of the Rayleigh quotient. Note that the determinant state (2) is a multi-linear function of the coefficients $\\{c_{ij}\\}$. For a fixed $j$, $|\phi\rangle$ is just a linear function of $c_{1j}$, $\cdots$, $c_{nj}$: $|\phi\rangle=\sum_{i}c_{ij}|\phi_{ij}\rangle,$ (3) where $|\phi_{ij}\rangle={\partial|\phi\rangle}/{\partial c_{ij}}$. Thus an approximate ground state $\Psi_{k}=\sum_{i}\alpha_{i}\phi^{(i)}+\alpha\phi$ can be written as $\Psi_{k}=\sum_{i}\alpha_{i}\phi^{(i)}+\sum_{i}\alpha c_{ij}\phi_{ij}.$ (4) This means that we can improve $\Psi_{k}$ and hence update the operator $F^{+}_{j}$ by finding the lowest eigenstate of the Hamiltonian in the subspace spanned by $\\{|\phi_{ij}\rangle$, $i=1,\cdots,n\\}$ and those previously found determinant states $|\phi^{(i)}\rangle$. Such relaxation procedure to update the operators $F^{+}_{j}$ is the key ingredient of this contribution. Suppose we have already obtained $k-1$ determinant states $|\phi^{(i)}\rangle$, the searching process for $k$-th determinant state $|\phi^{(k)}\rangle=|\phi\rangle$ in the form (2) involves the following iteration: (1) randomly generate a determinant state $|\phi\rangle$. (2) For $j=1,2,\cdots,m$, do the following iteration loop to update $|\phi\rangle$: (2a) Calculate the matrix elements of the Hamiltonian in the subspace $\Xi^{(k)}_{j}$ spanned by $\\{|\phi^{(1)}\rangle,\cdots,|\phi^{(k-1)}\rangle$, $|\phi_{1j}\rangle,\cdots,|\phi_{nj}\rangle\\}$; (2b) Find the ground state $\Psi_{k}^{j}$ of the Hamiltonian in the above subspace $\Xi^{(k)}_{j}$, $\Psi_{k}^{j}=\sum_{i}\alpha_{i}\phi^{(i)}+\sum_{i}\beta_{ij}^{k}\phi_{ij}$; (2c) Update $F^{+}_{j}$ by setting $c_{ij}=\beta^{k}_{ij}$ ($i=1,\cdots,n$); Then make $F^{+}_{j}|0\rangle$ orthogonal to other single particle states $\\{F^{+}_{l}|0\rangle,\ l\neq j\\}$, and restore $F^{+}_{j}|0\rangle$ to unit length by a normalization procedure. (3) Check the convergence of the Rayleigh quotient. Repeat the step (2) until reaching convergence. In case of $k=1$, the above searching process of finding the first determinant state is the same as the Hartree-Fock algorithm. A randomly generated initial trial state needs several tens of iteration rounds to converge. Each of subsequent determinant states needs about similar rounds of iteration. Here, the main numeric cost is the step (2a) for calculation of the matrix elements of the Hamiltonian between basis states. The step (2b) of finding lowest eigenstate in the subspace can be implemented efficiently via iteration algorithm 9 ; 10 that needs only small portion of the computation cost. Starting from $k=2$, the number of iteration to obtain $\phi_{k}$ depends on the initial choice. If a trial state has large overlap with the state $(H-E_{k-1})|\Psi_{k-1}\rangle$, one may reach convergence by just a few rounds of iteration. Here, $E_{k-1}$ and $\Psi_{k-1}=\sum_{i=1}^{k-1}\alpha_{i}\phi^{(i)}$ are the approximate ground state energy and wave function obtained in the previous step. One can understand this property by considering minimization of the Rayleigh quotient in the two dimensional subspace spanned by $\Psi_{k-1}$ and the trial state $\phi$ 7 . From this observation, we perform a preparing treatment of the trial state before step (2) of the above iteration procedure. The preparing treatment of the initial trial state $\phi$ is to modify the state $\phi$ so that it has maximum overlap with the state $(H-E_{k-1})|\Psi_{k-1}\rangle$. This procedure is easy to carry out by exploiting the fact that state $\phi$, or the overlap $\langle\phi|H-E_{k-1}|\Psi_{k-1}\rangle$, is a multi-linear function of the coefficients $c_{ij}$ (or the operators $F^{+}_{j}$). We maximize the overlap iteratively by updating the operators $F^{+}_{j}$, $(j=1,\cdots,m)$, consecutively. Usually, 3 to 5 rounds of the iteration are enough. After such preparing treatment, one usually needs about 2 to 3 iteration rounds of the searching process to minimize the Rayleigh quotient. Thus, such preparing treatment makes the overall procedure about 5 to 10 times faster. As an optional choice to achieve high accuracy, one can perform backward optimization after reaching convergence in the above procedure. This procedure updates the already found basis states one by one (One can also choose to update some selected basis states 4 ). The operation to update a basis state is the same as searching a new basis state. It is numerically expansive to perform the backward optimization. In fact, searching the basis states one by one is a kind of restriction on the determinant states, and the backward optimization means removing such constraint. At first sight, the current method shares some features with the Path-Integral Renormalization Group (PIRG) algorithm 10a ; 18 . However, based on different strategies, the PIRG and the current method are two different methods of searching basis determinant states. The PIRG filters out the ground state by repeatedly expanding $e^{-\tau H}|\psi\rangle$ into summation of determinant states and keeping some of the determinant states as new basis states to update the trial ground state $|\psi\rangle$. At each step, the PIRG must update whole basis states, while the current method only adds (or updates) one basis state via relaxation method and exploiting the multi-linearity of the determinant states. Updating basis states in PIRG, i.e., choosing some determinant states from those ones that span the state $e^{-\tau H}|\psi\rangle$, involves diagonalization of many sizable matrices. The diagonalization in subspace is a major numeric cost of PIRG, while it takes only a small portion of numeric operations in current method. The current method only calculates matrix elements of the Hamiltonian between determinant states, this is much easier and more efficient than expanding $e^{-\tau H}|\psi\rangle$ (or $H|\psi\rangle$) into summation of determinant states. We test the above method via the two dimensional fermionic Hubbard mode on a $N=L\times L$ square lattice with periodic condition. The Hamiltonian reads $H=-t\sum_{\langle ij\rangle}(c^{+}_{i\sigma}c_{j\sigma}+c^{+}_{j\sigma}c_{i\sigma})+U\sum_{i}n_{i\uparrow}n_{i\downarrow}.$ (5) Here $c^{+}_{j\sigma}$ ($c_{j\sigma}$) is the creation (annihilation) operator of an electron with spin $\sigma$ at $j$-th site and $n_{j\sigma}=c^{+}_{j\sigma}c_{j\sigma}$. $U$ is the on-site Coulomb energy. The summation $\langle ij\rangle$ runs over nearest-neighbor sites. Table 1: Ground state energies of some the $4\times 4$ systems. system | $U/t$ | CPMC | SD | MP | N | Exact ---|---|---|---|---|---|--- $10/16$ | 4 | -19.5808 | -19.58 | -19.5775 | 1874 | -19.5808 $14/16$ | 4 | -15.7296 | -15.49 | -15.7107 | 1865 | -15.7446 $16/16$ | 4 | | -13.59 | -13.5963 | 1622 | -13.6219 $10/16$ | 8 | -17.4800 | -17.40 | -17.4625 | 2000 | -17.5104 $14/16$ | 8 | -11.648 | | -11.6763 | 2000 | -11.8688 $16/16$ | 8 | | | -8.41565 | 1850 | -8.46889 $14/16$ | 12 | -9.696 | | -9.79500 | 2000 | -10.0515 $16/16$ | 12 | | | -5.95238 | 1950 | -5.99222 Table 1 shows ground state energies (in the unit of $t$) of some $4\times 4$ systems. The column “system” indicates the number of electrons versus the lattice number. $N$ is number of basis determinant states to span the ground state of the current method (MP). We list the results of the Constrained Path Quantum Monte Carlo (CPMC) 11 , Stochastic Diagonalization (SD) 7 , and exact diagonalization 12 ; 13 ; 14 ; 15 for comparison. The accuracy of our method is almost unchanged for various interaction strength $U$ and filling number. This demonstrates the stability of the method. All initial trial states for searching basis determinant states are randomly generated without any symmetry consideration. We use convergence rate $\epsilon=2|E_{n}-E_{n-1}|/|E_{n}+E_{n-1}|$ to determine the number of basis states, where $E_{n}$ and $E_{n-1}$ are ground state energies obtained with $n$ and $n-1$ basis states, respectively. The searching for basis states stops if $\epsilon$ is smaller than a criteria $\epsilon_{0}$, or maximum acceptable number of basis states is reached. Usually, $\epsilon_{0}=10^{-5}$ is enough to obtain quite reasonable result. At this setting, one usually needs several hundreds basis states which increases slowly with $U$. If $\epsilon_{0}=10^{-6}$, one needs several thousands of basis states for convergence. Roughly speaking, result from about 100 basis states is quite well. The interest point is that the beginning several tens of determinant states, usually less than 60 basis states, make dominant contribution. And the first one, i.e., the HF approximation, contributes most. The number of dominant basis states increases slowly with the interaction strength $U$. After the dominant basis states, the contribution from each of following ones drops rapidly. Table 2: Correlation functions of some the $4\times 4$ systems. Method | system | $U/t$ | $\rho(2,1)$ | $S_{m}(\pi,\pi)$ | $S_{d}(\pi,\pi)$ ---|---|---|---|---|--- CPMC | $10/16$ | 4 | -0.0556 | 0.731 | 0.504 MP(800) | $10/16$ | 4 | -0.0559 | 0.7315 | 0.5089 Exact | $10/16$ | 4 | -0.0556 | 0.73 | 0.506 QL | $16/16$ | 4 | -0.0363 | 3.42 | 0.3946 MP(437) | $16/16$ | 4 | -0.0478 | 3.490 | 0.3888 Exact | $16/16$ | 4 | -0.0475 | 3.64 | 0.385 CPMC | $10/16$ | 8 | -0.0462 | 0.761 | 0.4403 MP(800) | $10/16$ | 8 | -0.0493 | 0.7645 | 0.4412 Exact | $10/16$ | 8 | -0.0485 | 0.75 | 0.443 Table 2 shows ground state’s correlation functions of some $4\times 4$ systems. The comparing results of CPMC, Quantum Langevin (QL), and exact diagonalization are from 11 , 15 , and 12 ; 15 , respectively. Here $S_{m}$ and $S_{d}$ are magnetic and density structure factors 11 , respectively; and $\rho(\mathbf{r})$ is the one body density matrix. The number in the column “method” is the number of basis states of our method (MP). The current method obtains the ground state energy and wave function at the same time. Then calculation of the correlation functions and other related quantities is a trivial task that simply reads the wave function. In comparison with the exact result, we see that the wave functions are almost in the same accuracy as the correspondent energies. Again, the beginning several tens of basis states make major contribution. To demonstrate this property, as indicated in the parentheses, we use only several hundreds of basis states to calculate the correlation functions in table 2. Since our program does not perform any symmetry treatment, we can only compare non-degenerated ground states with the exact result. In fact, the present method can take into account of symmetries. After the searching process of finding basis determinant states, the resultant wave function is usually a combination of ground states with different symmetries. One may employ, e.g., the project technique 16 to filter out the target symmetry. This may further improve the accuracy. Table 3: Ground state energies (in the unit of $t$) of some large systems with $U=4t$. system | QMC | VMC | SD | MP | $N$ ---|---|---|---|---|--- $26/6\times 6$ | -42.32 | | -40.77 | -41.0757 | 400 $34/6\times 6$ | -33.30 | -32.76 | | -32.7323 | 940 $36/6\times 6$ | -30.96 | -30.384 | | -30.5166 | 905 $50/8\times 8$ | -72.80 | | -67.00 | -68.5029 | 630 $54/8\times 8$ | -67.55 | | | -63.8981 | 560 $62/8\times 8$ | -57.70 | | | -55.5255 | 490 $64/8\times 8$ | -55.23 | | | -53.583 | 472 $100/10\times 10$ | -86.70 | | | -82.9549 | 152 Table 3 shows ground state energies for large system size that exact diagonalization is impossible. Here $N$ is the number of basis states of the current method (MP). Our result is quite close to that of the SD 7 , and the variational quantum Monte-Carlo (VMC) 18 . It is worth to note that the number of basis states of our method is several orders smaller than that of the SD algorithm. As a consequence, our method needs much less memory, and there is no need for external storage. There are several percent of discrepancy with the Quantum Monte Carlo (QMC) result 17 , this disagreement increases with the system size. This needs further investigations. The discrepancy between QMC result and the strict variational result is also found by other authors, see, e.g., 10a ; 16 ; 18 . For practical applications, the extrapolation method introduced in PIRG’s implementation is an useful tool to handle the discrepancy with QMC results 10a ; 16 . Similar to the $4\times 4$ cases, the beginning several tens of basis states make dominant contribution. For a fixed interaction strength $U$, the contribution from the dominant basis states increases with the system size. This means that one needs less basis states for larger system size. On the other hand, the computation cost to search a basis state scales about quadratically with the system size. The role of preparing step for searching a basis state is more significant for larger system size. With out this preparing step, after several tens of dominant basis states, the overlap between a randomly generated trial state $\phi$ and the state $(H-E_{k-1})|\Psi_{k-1}\rangle$ almost vanishes. One must substantially increase accuracy requirement for following steps, which in turn increases numeric cost. The computation cost scales about quadratically with the number of basis states. Figure 1: Relative error $E_{r}$ versus number of basis states $n$ for various filling numbers and system sizes. Fig. 1 shows the relative error $E_{r}=|(E_{n}-E_{0})/E_{0}|$ versus the number of basis states $n$, where $E_{n}$ is the ground state energy obtained with $n$ basis states, and $E_{0}$ is the converged ground state energy (or exact ground state energy if available). This illustrates the overall properties of the method. The first basis state makes most important contribution. It accounts for about $80\%$ to $95\%$ of the ground state energy depending on system size and correlation strength. Roughly speaking, the mean field effect increases with the system size, and decreases with the correlation between electrons of the system. For system size $N_{0}=4\times 4$ with $U=12t$, the contribution of the first basis state is about $80\%$. As the system size reaching $N_{0}=10\times 10$ with $U=4t$, the contribution of the first basis state is more than $90\%$. As a consequence, for a same accuracy requirement, the number of necessary basis states decreases with system size. The convergence rate is fast for the beginning several tens of dominant basis states. These basis states contribute more than $95\%$ to the ground state for moderate correlation. Then contribution from each of the following basis states drops rapidly. However, the major computation cost for Fig. 1 is to search the remaining basis states. In practical calculations, one usually needs several tens of basis states for a reasonable accuracy. We perform backward optimization for some cases to improve the accuracy. There is very limited improvement from the backward optimization if the MP method is converged. The improvement is usually less than $0.5\%$. If backward optimization is performed before the convergence of MP process, the improvement can be more than $1.5\%$ for some cases. Our calculation is performed on single PC (AMD Opteron(tm) Processor 248). Parallel implementation is easy for the current method. Since the major numeric cost is the computation of the matrix elements of the Hamiltonian during the search of the basis states, parallel implementation can be simply realized by requiring each node handling some matrix elements of the Hamiltonian. There are many possible ways to improve the current method. For example, it is worth to explore other type of basis states. In the present form, our method is an extension to the mean field HF approximation. Mathematically, the redundancy of the over-complete basis states is crucial for the convergence speed of the MP algorithm 6 . By increase the redundancy of the over-complete basis states, i.e., enlarging the searching space, it is possible to speed up the convergence of MP method for searching the basis states. On the other hand, for particles moving in 3D space, storage of a single particle state needs sizable memory, one may choose the basis states for single particles as product of one dimensional wave functions. In principle, this method is able to compute the exited states. With some modifications, it may be feasible to calculate the low-lying exited states. In summary, the current method is stable and free from the sign problem. It can apply to any system that can apply Hartree-Fock algorithm, and can be regarded as an extension to the Hartree-Fock algorithm. Several tens of determinant states are usually enough for meaningful result. This method may offer an alternative to explore quantum effects of Many body systems. This work is supported by the National Science Foundation of China (Grant No. 10375042). We acknowledge beneficial discussions with Prof. W. Wang. ## References * (1) M. Nest, R. Padmanaban, and P. Saalfrank, J. Chem. Phys. 126, 214106 (2007); M. Nest, T. Klamroth, and P. Saalfrank, J. Chem. Phys. 122, 124102 (2005); M. H. Beck and H.-D. Meyer, J. Chem. Phys. 114, 2036 (2001). * (2) Gregory Beylkin, Martin J. Mohlenkamp, Fernando Pérez, arXiv:0708.2896v1 [math-ph]. * (3) M. Imada and T. Kashima, J. Phys. Soc. Jpn. 69, 2723 (2000); T. Kashima and M. Imada, ibid. 70, 2287 (2001). * (4) T. Yanagisawa, Phys. Rev. B 75, 224503 (2007). * (5) H. Raedt, and M. Frick, Phys. Rep. 231, 107 (1993). * (6) S. Mallat, and Z. Zhang, IEEE Trans. Signal Process. 41, 3397 (1993). * (7) Laura Rebollo-Neira, and David Lowe, IEEE Signal Process. Lett. 9, 137 (2002); Miroslav Andrle, Laura Rebollo-Neira, and Evengelos Sagianos, IEEE Signal Process. Lett. 11, 137 (2004). * (8) P. Huggins, and W. Zucker, IEEE Trans. Signal Process. 55, 3760, (2007). * (9) R. Gribonval, and P. Vandergheynst, IEEE Trans. Signal Process. 52, 255 (2006). * (10) Yinghua Wu and Victor S. Batista, J. Chem. Phys. 118, 6720 (2003); 121, 1676 (2004). * (11) F. Andreozzi, A. Porrino, and N.L. Iudice, J. Phys. A 35, L61 (2002). * (12) Quanlin Jie, and Dunhuan Liu, Phys. Rev. E 68, 056706 (2003). * (13) S. Zhang, J. Carlson, and J. E. Gubernatis, Phys. Rev. Lett. 74, 3652 (1995); ibid, Phys. Reb. B 55, 7464 (1997). * (14) A. Parola, S. Sorella, M. Parrinello, and E. Tosatti, Phys. Reb. B 43, 6190 (1991). * (15) H. Q. Lin, Phys. Reb. B 44, 7151 (1991). * (16) G. Fano, F. Ortolani, and A. Parola, Phys. Rev. B 42, 6877 (1990). * (17) A. Parola, S. Sorella, R. Car, M. Parrinello, and E. Tosatti, Physica C 162-164, 771 (1989). * (18) T. Mizusaki, and M. Imada, Phys. Rev. B 69, 125110 (2004). * (19) N. Furukawa, and M. Imada, J. Phys. Soc. Jpn. 61, 3331 (1992).

arxiv-papers

2024-09-04T02:48:53.740489

0802.1768

# LENS MAPPING OF DARK MATTER SUBSTRUCTURE WITH VSOP-2 Shigenori Ohashi,1 Masashi Chiba,1 and Kaiki Taro Inoue 2 ###### Abstract Hierarchical clustering models of cold dark matter (CDM) predict that about 5% - 10% of a galaxy-sized halo with mass $\sim 10^{12}$ solar masses ($M_{\odot}$) resides in substructures (CDM subhalos) with masses $\mathrel{\hbox{\hbox to0.0pt{\hbox{\lower 4.0pt\hbox{$\sim$}}\hss}\hbox{$<$}}}10^{8}M_{\odot}$. To directly identify such substructures, we propose to observe radio continuum emission from multiply imaged QSOs using VSOP-2 with a high angular resolution. 11footnotetext: Astronomical Institute, Tohoku University, Aoba-ku, Sendai, 980-8578, Japan22footnotetext: Department of Science and Engineering, Kinki University, Higashi-Osaka, 577-8502, Japan ## 1\. Introduction The currently standard framework for understanding cosmological structure is based on hierarchical clustering of CDM in the accelerating Universe. Low mass systems collapse and form earlier and later they merge to form more massive gravitationally bound systems. The standard model is successful for explaining a wide variety of astrophysical observations such as the temperature fluctuations in the Cosmic Microwave Background and large-scale distribution of galaxies. Despite the great success of the concordance model, there are several open issues. One of them is so-called ”the missing satellite problem”. Some authors have argued that this discrepancy could be resolved by considering some suppressing process for star formation, such as gas heating by an intergalactic UV background. In whatever models relying on the suppression of galaxy formation, a typical galaxy-sized halo should contain numerous dark subhalos. It is thus important to constrain the presence or absence of many invisible satellites around a galaxy like the Milky Way in the form of dark subhalos. To obtain the direct evidence for such dark subhalos, we propose to observe the multiply imaged QSOs with a high angular resolution. Irrespective of the presence or absence of associated luminous components, it is possible to probe these subhalos in a foreground lensing galaxy by means of gravitational lensing. ## 2\. Theory of Gravitational Lensing The phenomena that light rays are deflected by gravity is one of the consequences of Einstein’s General Theory of Relativity. This is referred to as Gravitational lensing. Gravitational lensing is a powerful tool to determine the mass of lens objects which deflect and distort a source image in the background. In a gravitational lens system, Einstein radius is a useful indicator of the scale of image distortion. In the case of an Singular Isothermal Sphere (SIS) with density distribution of $\rho\propto r^{-2}$, Einstein radius is written as, $\theta_{E}\sim 1.5\left(\frac{M_{t}}{5\times 10^{7}M_{\odot}}\right)\left({\frac{r_{t}}{1.0\mathrm{kpc}}}\right)^{-1}\left(\frac{D_{ls}/D_{os}}{0.5}\right)[\mathrm{mas}]$ (1) where $r_{t}$ is a supposed tidal radius of an SIS lens and $M_{t}$ is a total mass enclosed within $r_{t}$. $D_{ls}$ and $D_{os}$ are angular diameter distances between lens and source and between observer and source, respectively. Several observations of lens systems in a galactic scale have revealed that the density distribution of a galaxy lens, being composed of both baryonic and dark matter, is well represented by an SIS profile. However, actual density profiles in subhalos are yet to be settled. According to N-body simulations based on the CDM model, i.e., without taking into account dissipative baryonic matter, all dark halos appear to show a shallower density profile at their central parts, expressed as $\rho\propto r^{\gamma}$ with $-1.5<\gamma<-1$ (e.g., Navarro et al. 2004). If this is the case, then a dark halo with this profile has a smaller Einstein radius than an SIS case, so the resulting lensing signal to be observed by VSOP-2 would be biased further in favor of more massive subhalos. How baryonic components, if there are any, affect a dark halo profile is yet uncertain: an adiabatic contraction of a halo as driven by gas cooling yields a steeper density profile, whereas starburst activity and associated heating by supernovae explosion may make a gravitational potential shallower owing to galactic wind. Whatever effects of baryonic components are at work, an SIS profile would provide a possibly upper limit to the lensing signal of subhalos. As is shown in equation (1), the size of an Einstein radius depends on the distance ratio, $D_{ls}/D_{os}$. We plot, in Figure 1, this dependence for a subhalo of $M_{t}=5\times 10^{7}M_{\odot}$ and $r_{t}=1.0$ kpc, using known lensing systems taken from CASTLES. Figure 1.: The dependence of an Einstein radius on lens and source redshift in the case of $M_{t}=5\times 10^{7}M_{\odot}$ and $r_{t}=1.0$kpc. Filled circles: known lensing systems taken from CASTLES. Solid lines: the size of an Einstein angle in units of mas or that of an Einstein raidus projected in a lens plane in units of pc. ## 3\. Simulated Images In Figure 2, we show simulated images of a gravitational lensing system using the lens parameters of MG0414+0534, where a source image and a lensing galaxy are at redshifts of 2.64 and 0.96, respectively. We assume an SIS profile for the density distribution of a subhalo. Figure 2.: Simulated lensed images perturbed by a subhalo with $M_{t}=5\times 10^{7}M_{\odot}$ and $r_{t}=1.0$ kpc, where the angular resolution is assumed to be 10, 1 and 0.1 [mas] for left, middle, and right panels, respectively. In this lensing system, a source image and a lensing galaxy are at redshifts 2.64 and 0.96, respectively. The locally distorted feature for an arc-like lensed image in Figure 2 can be caused by not only subhalo lensing but also intrinsic substructure within a source image. To distinguish the effect of subhalo lensing alone, we should select a lens system showing multiple images as a target: intrinsic substructure in a source would be seen in all multiple images, whereas subhalos yield particular features in each image. Thus it is possible to extract the lensing signals of subhalos by subtracting commonly-observed features in lensed images. ## References * Inoue K.T. & Chiba M. (2003) Inoue K.T. & Chiba M. 2003, ApJ, 591,L83 * Inoue K.T. & Chiba M. (2005) Inoue K.T. & Chiba M. 2005, ApJ, 633,23 * Inoue K.T. & Chiba M. (2005) Inoue K.T. & Chiba M. 2005, ApJ, 634,70 * Navaro (2004) Navarro J.F. et al. 2004, MNRAS, 349, 1039

arxiv-papers

2024-09-04T02:48:53.744754

0802.1813

# Taub-NUT Black Holes in Third order Lovelock Gravity S. H. Hendi1,3111email address: hendi@mail.yu.ac.ir and M. H. Dehghani2,3 222email address: mhd@shirazu.ac.ir 1Physics Department, College of Sciences, Yasouj University, Yasouj 75914, Iran 2Physics Department and Biruni Observatory, College of Sciences, Shiraz University, Shiraz 71454, Iran 3Research Institute for Astrophysics and Astronomy of Maragha (RIAAM), Maragha 55134, Iran ###### Abstract We consider the existence of Taub-NUT solutions in third order Lovelock gravity with cosmological constant, and obtain the general form of these solutions in eight dimensions. We find that, as in the case of Gauss-Bonnet gravity and in contrast with the Taub-NUT solutions of Einstein gravity, the metric function depends on the specific form of the base factors on which one constructs the circle fibration. Thus, one may say that the independence of the NUT solutions on the geometry of the base space is not a robust feature of all generally covariant theories of gravity and is peculiar to Einstein gravity. We find that when Einstein gravity admits non-extremal NUT solutions with no curvature singularity at $r=N$, then there exists a non-extremal NUT solution in third order Lovelock gravity. In $8$-dimensional spacetime, this happens when the metric of the base space is chosen to be $\mathbb{CP}^{3}$. Indeed, third order Lovelock gravity does not admit non-extreme NUT solutions with any other base space. This is another property which is peculiar to Einstein gravity. We also find that the third order Lovelock gravity admits extremal NUT solution when the base space is $T^{2}\times T^{2}\times T^{2}$ or $S^{2}\times T^{2}\times T^{2}$. We have extended these observations to two conjectures about the existence of NUT solutions in Lovelock gravity in any even-dimensional spacetime. ###### pacs: 04.70.Bw, 04.20.Jb, 04.70.-s ## I Introduction The question as to why the Planck and electroweak scales differ by so many orders of magnitude remains mysterious. In recent years, attempts have been made to address this hierarchy issue within the context of theories with extra spatial dimensions. In higher-dimensional spacetimes even with the assumption of Einstein – that the left-hand side of the field equations is the most general symmetric conserved tensor containing no more than two derivatives of the metric – the field equations need to be generalized. This generalization has been done by Lovelock Lov , and he found a second rank symmetric conserved tensor in $d$ dimensions which contains upto second order derivative of the metric. Other higher curvature gravities which have higher derivative terms of the metric, e.g., terms with quartic derivatives, have serious problems with the presence of tachyons and ghosts as well as with perturbative unitarity, while the Lovelock gravity is free of these problems Zw . Many authors have considered the possibility of higher curvature terms in the field equations and how their existence would modify the predictions about the gravitating system. Here, we are interested in the properties of the black holes, and we want to know which properties of the black holes are peculiar to Einstein gravity, and which are robust features of all generally covariant theories of gravity. This fact provide a strong motivation for considering new exact solutions of Lovelock gravity. We show that some properties of NUT solutions are peculiar to Einstein gravity and not robust feature of all generally covariant theories of gravity. Although the nonlinearity of the field equations causes to have a few exact black hole solutions in Lovelock gravity, there are many papers on this subject All GB Third ; DehMan ; DehHen . In this letter we introduce Taub-NUT metrics in third order Lovelock gravity, and investigate which properties of these kinds of solutions will be modified by considering higher curvature terms in the field equations. The original four-dimensional solution TaubNUT is only locally asymptotic flat. The spacetime has as a boundary at infinity a twisted $S^{1}$ bundle over $S^{2}$, instead of simply being $S^{1}\times S^{2}$. There are known extensions of the Taub-NUT solutions to the case when a cosmological constant is present. In this case the asymptotic structure is only locally de Sitter (for positive cosmological constant) or anti-de Sitter (for negative cosmological constant) and the solutions are referred to as Taub-NUT-(A)dS metrics. In general, the Killing vector that corresponds to the coordinate that parameterizes the fibre $S^{1}$ can have a zero-dimensional fixed point set (called a NUT solution) or a two-dimensional fixed point set (referred to as a ‘bolt’ solution). Generalizations to higher dimensions follow closely the four-dimensional case BaisPA ; Awad ; Lorenzo . Also, these kinds of solutions have been generalized in the presence of electromagnetic field and their thermodynamics have been investigated Mann12 ; DehKhod . It is therefore natural to suppose that the generalization of these solutions to the case of Lovelock gravity, which is the low energy limit of supergravity, might provide us with a window on some interesting new corners of M-theory moduli space. The outline of this letter is as follows. We give a brief review of the field equations of third order Lovelock gravity in Sec. II. In Sec. III, we obtain Taub-NUT solutions of third order Lovelock gravity in eight dimensions and then we check the conjectures given in Ref. DehMan . We finish this letter with some concluding remarks. ## II Field Equations The vacuum gravitational field equations of third order Lovelock gravity may be written as: $\alpha_{0}g_{\mu\nu}+\alpha_{1}G_{\mu\nu}^{(1)}+\alpha_{2}G_{\mu\nu}^{(2)}+\alpha_{3}G_{\mu\nu}^{(3)}=0,$ (1) where $\alpha_{i}$’s are Lovelock coefficients, $G_{\mu\nu}^{(1)}$ is just the Einstein tensor, and $G_{\mu\nu}^{(2)}$ and $G_{\mu\nu}^{(3)}$ are the second and third order Lovelock tensors given as $G_{\mu\nu}^{(2)}=2(R_{\mu\sigma\kappa\tau}R_{\nu}^{\phantom{\nu}\sigma\kappa\tau}-2R_{\mu\rho\nu\sigma}R^{\rho\sigma}-2R_{\mu\sigma}R_{\phantom{\sigma}\nu}^{\sigma}+RR_{\mu\nu})-\frac{1}{2}\mathcal{L}_{2}g_{\mu\nu},$ (2) $\displaystyle G_{\mu\nu}^{(3)}$ $\displaystyle=$ $\displaystyle-3(4R^{\tau\rho\sigma\kappa}R_{\sigma\kappa\lambda\rho}R_{\phantom{\lambda}{\nu\tau\mu}}^{\lambda}-8R_{\phantom{\tau\rho}{\lambda\sigma}}^{\tau\rho}R_{\phantom{\sigma\kappa}{\tau\mu}}^{\sigma\kappa}R_{\phantom{\lambda}{\nu\rho\kappa}}^{\lambda}+2R_{\nu}^{\phantom{\nu}{\tau\sigma\kappa}}R_{\sigma\kappa\lambda\rho}R_{\phantom{\lambda\rho}{\tau\mu}}^{\lambda\rho}$ (3) $\displaystyle-R^{\tau\rho\sigma\kappa}R_{\sigma\kappa\tau\rho}R_{\nu\mu}+8R_{\phantom{\tau}{\nu\sigma\rho}}^{\tau}R_{\phantom{\sigma\kappa}{\tau\mu}}^{\sigma\kappa}R_{\phantom{\rho}\kappa}^{\rho}+8R_{\phantom{\sigma}{\nu\tau\kappa}}^{\sigma}R_{\phantom{\tau\rho}{\sigma\mu}}^{\tau\rho}R_{\phantom{\kappa}{\rho}}^{\kappa}$ $\displaystyle+4R_{\nu}^{\phantom{\nu}{\tau\sigma\kappa}}R_{\sigma\kappa\mu\rho}R_{\phantom{\rho}{\tau}}^{\rho}-4R_{\nu}^{\phantom{\nu}{\tau\sigma\kappa}}R_{\sigma\kappa\tau\rho}R_{\phantom{\rho}{\mu}}^{\rho}+4R^{\tau\rho\sigma\kappa}R_{\sigma\kappa\tau\mu}R_{\nu\rho}+2RR_{\nu}^{\phantom{\nu}{\kappa\tau\rho}}R_{\tau\rho\kappa\mu}$ $\displaystyle+8R_{\phantom{\tau}{\nu\mu\rho}}^{\tau}R_{\phantom{\rho}{\sigma}}^{\rho}R_{\phantom{\sigma}{\tau}}^{\sigma}-8R_{\phantom{\sigma}{\nu\tau\rho}}^{\sigma}R_{\phantom{\tau}{\sigma}}^{\tau}R_{\mu}^{\rho}-8R_{\phantom{\tau}{\sigma\mu}}^{\tau\rho}R_{\phantom{\sigma}{\tau}}^{\sigma}R_{\nu\rho}-4RR_{\phantom{\tau}{\nu\mu\rho}}^{\tau}R_{\phantom{\rho}\tau}^{\rho}$ $\displaystyle+4R^{\tau\rho}R_{\rho\tau}R_{\nu\mu}-8R_{\phantom{\tau}{\nu}}^{\tau}R_{\tau\rho}R_{\phantom{\rho}{\mu}}^{\rho}+4RR_{\nu\rho}R_{\phantom{\rho}{\mu}}^{\rho}-R^{2}R_{\nu\mu})-\frac{1}{2}\mathcal{L}_{3}g_{\mu\nu}.$ In Eqs. (2) and (3) $\mathcal{L}_{2}=R_{\mu\nu\gamma\delta}R^{\mu\nu\gamma\delta}-4R_{\mu\nu}R^{\mu\nu}+R^{2}$ is the Gauss-Bonnet Lagrangian and $\displaystyle\mathcal{L}_{3}$ $\displaystyle=$ $\displaystyle 2R^{\mu\nu\sigma\kappa}R_{\sigma\kappa\rho\tau}R_{\phantom{\rho\tau}{\mu\nu}}^{\rho\tau}+8R_{\phantom{\mu\nu}{\sigma\rho}}^{\mu\nu}R_{\phantom{\sigma\kappa}{\nu\tau}}^{\sigma\kappa}R_{\phantom{\rho\tau}{\mu\kappa}}^{\rho\tau}+24R^{\mu\nu\sigma\kappa}R_{\sigma\kappa\nu\rho}R_{\phantom{\rho}{\mu}}^{\rho}$ (4) $\displaystyle+3RR^{\mu\nu\sigma\kappa}R_{\sigma\kappa\mu\nu}+24R^{\mu\nu\sigma\kappa}R_{\sigma\mu}R_{\kappa\nu}+16R^{\mu\nu}R_{\nu\sigma}R_{\phantom{\sigma}{\mu}}^{\sigma}-12RR^{\mu\nu}R_{\mu\nu}+R^{3}$ is the third order Lovelock Lagrangian. Equation (1) does not contain the derivative of the curvatures, and therefore the derivatives of the metric higher than two do not appear. In order to have the contribution of all the above terms in the field equation, the dimension of the spacetime should be equal or larger than seven. Here, for simplicity, we consider the NUT solutions of the dimensionally continued gravity in eight dimensions. The dimensionally continued gravity in $D$ dimensions is a special class of the Lovelock gravity, in which the Lovelock coefficients are reduced to two by embedding the Lorentz group $SO(D-1,1)$ into a larger AdS group $SO(D-1,2)$ Zan . By choosing suitable unit, the remaining two fundamental constants can be reduced to one fundamental constant $l$. Thus, the Lovelock coefficients $\alpha_{i}$’s can be written as $\alpha_{0}=-\frac{21}{l^{2}},\text{ \ \ }\alpha_{1}=3,\text{ \ \ }\alpha_{2}=\frac{3l^{2}}{20},\text{ \ \ \ }\alpha_{3}=\frac{l^{4}}{120}.$ ## III Eight-dimensional Solutions In this section we study the eight-dimensional Taub-NUT solutions of third order Lovelock gravity. In constructing these metrics the idea is to regard the Taub-NUT spacetime as a $U(1)$ fibration over a $6$-dimensional base space endowed with an Einstein-Kähler metric $d\Omega{{}_{B}}^{2}$. Then the Euclidean section of the $8$-dimensional Taub-NUT spacetime can be written as: $ds^{2}=F(r)(d\tau+N\mathcal{A})^{2}+F^{-1}(r)dr^{2}+(r^{2}-N^{2})d\Omega{{}_{B}}^{2},$ (5) where $\tau$ is the coordinate on the fibre $S^{1}$ and $A$ has a curvature $F=dA$, which is proportional to some covariantly constant 2-form. Here $N$ is the NUT charge and $F(r)$ is a function of $r$. The solution will describe a ‘NUT’ if the fixed point set of the $U(1)$ isometry $\partial/\partial\tau$ (i.e. the points where $F(r)=0$) is less than $6$-dimensional and a ‘bolt’ if the fixed point set is $6$-dimensional. Here, we consider only the cases where all the factor spaces of $\mathcal{B}$ have zero or positive curvature. Thus, the base space $\mathcal{B}$ can be the 6-dimensional space $\mathbb{CP}^{3}$, a product of three 2-dimensional spaces ($T^{2}$ or $S^{2}$), or the product of a 4-dimensional space $\mathbb{CP}^{2}$ with a 2-dimensional one. The $1$-forms and metrics of $S^{2}$, $T^{2}$, $\mathbb{CP}^{2}$ and $\mathbb{CP}^{3}$ are Pop $\displaystyle\mathcal{A}_{S^{2}}=2\cos\theta_{i}d\phi_{i},\hskip 14.22636ptd\Omega_{S^{2}}^{2}=d\theta_{i}^{2}+\sin^{2}\theta_{i}d\phi_{i}^{2},$ (6) $\displaystyle\mathcal{A}_{T^{2}}=2\eta_{i}d\zeta_{i},\hskip 34.14322ptd\Omega_{T^{2}}^{2}=d\eta_{i}^{2}+d\zeta_{i}^{2},$ (7) $\displaystyle\mathcal{A}_{\mathbb{CP}^{2}}=6\sin^{2}\theta_{2}(d\phi_{2}+\sin^{2}\theta_{1}d\phi_{1}),$ (8) $\displaystyle d\Omega^{2}{{}_{\mathbb{CP}^{2}}}=6\\{d{\theta_{2}}^{2}+\sin^{2}\theta_{2}\cos^{2}\theta_{2}(d\phi_{2}+\sin^{2}\theta_{1}d\phi_{1})^{2}$ $\displaystyle\hskip 48.36958pt+sin^{2}\theta_{2}({d\theta_{1}}^{2}+\sin^{2}\theta_{1}\cos^{2}\theta_{1}{d\phi_{1}}^{2})\\},$ (9) $\displaystyle\mathcal{A}_{\mathbb{CP}^{3}}=\frac{1}{2}\left(\frac{1}{2}(\cos^{2}\theta_{3}-\sin^{2}\theta_{3})d\phi_{3}-\cos^{2}\theta_{3}\cos\theta_{1}d\phi_{1}-\sin^{2}\theta_{3}\cos\theta_{2}d\phi_{2}\right),$ $\displaystyle d\Omega^{2}{{}_{\mathbb{CP}^{3}}}=8\\{d\theta_{3}^{2}+\frac{1}{4}\sin^{2}\theta_{3}\cos^{2}\theta_{3}(d\phi_{3}-\cos\theta_{1}d\phi_{1}+\cos\theta_{2}d\phi_{2})^{2}$ $\displaystyle\hskip 48.36958pt+\frac{1}{4}\cos^{2}\theta_{3}(d\theta_{1}^{2}+\sin^{2}\theta_{1}d\phi_{1}^{2})+\frac{1}{4}\sin^{2}\theta_{3}(d\theta_{2}^{2}+\sin^{2}\theta_{2}d\phi_{2}^{2})\\},$ (10) respectively. To find the metric function $F(r)$, one may use any components of Eq. (1). After some calculation, we find that the metric function $F(r)$ for any base space $\mathcal{B}$ can be written as $F(r)=\frac{1}{\Gamma l^{2}}\left(\frac{\Psi^{1/3}}{2\beta}-\frac{3\left(r^{2}-N^{2}\right)^{4}B+\left(r^{2}-N^{2}\right)^{2}l^{4}E}{3\Psi^{1/3}}+\Omega\right),$ (11) where $\displaystyle\Psi$ $\displaystyle=$ $\displaystyle\left(r^{2}-N^{2}\right)\left(4C+\sqrt{16C^{2}+8\beta^{3}\left(r^{2}-N^{2}\right)^{4}\left[\left(r^{2}-N^{2}\right)^{6}B^{3}+\frac{l^{12}}{27}E^{3}\right]}\right),$ $\displaystyle\Omega$ $\displaystyle=$ $\displaystyle\frac{p}{6}\left(r^{4}-N^{4}\right)l^{2}+\left(r^{2}-N^{2}\right)\left(5r^{2}+3N^{2}\right),$ $\displaystyle\Gamma$ $\displaystyle=$ $\displaystyle 5r^{4}+6N^{2}r^{2}+5N^{4},$ and $p$ is the dimension of the curved factor spaces of $\mathcal{B}$. The constant $\beta$ and the functions $B$, $E$ and $C$ depend on the choice of the base space as: $\mathcal{B}$ | $\beta$ | $E$ | $B$ ---|---|---|--- ${{\mathbb{CP}^{3}}}$ | $2$ | $0$ | $\left(8N^{2}-l^{2}\right)^{2}$ $S^{2}\times\mathbb{CP}^{2}$ | $9$ | $\ 2\left(13r^{4}-6r^{2}N^{2}+13N^{4}\right)$ | $72N^{2}\left(4N^{2}-l^{2}\right)$ $S^{2}\times S^{2}\times S^{2}$ | $3$ | $12\left(r^{4}+N^{4}\right)$ | $24N^{2}\left(4N^{2}-l^{2}\right)$ $T^{2}\times\mathbb{CP}^{2}$ | $9$ | $\ -4\left(r^{4}+6r^{2}N^{2}+N^{4}\right)$ | $48N^{2}\left(6N^{2}-l^{2}\right)$ $T^{2}\times S^{2}\times S^{2}$ | $3$ | $0$ | $\frac{2}{3}\left(12N^{2}-l^{2}\right)^{2}$ $T^{2}\times T^{2}\times S^{2}$ | $3$ | $-2\left(r^{2}+N^{2}\right)^{2}$ | $8N^{2}\left(12N^{2}-l^{2}\right)$ $T^{2}\times T^{2}\times T^{2}$ | $1$ | $0$ | $32N^{4}$ and $\mathcal{B}$ | $C$ ---|--- $\begin{array}[]{c}\mathbb{CP}{{{}^{3}}}\end{array}$ | $\left(l^{2}-8N^{2}\right)^{3}\left(r^{2}+N^{2}\right)\Upsilon+mr\Gamma^{2}$ $\begin{array}[]{c}S^{2}\times\mathbb{CP}^{2}\\\ \text{ \ \ \ }\end{array}$ | $\begin{array}[]{c}324\left(r^{2}+N^{2}\right)\\{18N^{4}\left(3l^{2}-8N^{2}\right)\Upsilon+l^{6}\left[7\left(r^{8}+N^{8}\right)+38r^{2}N^{2}\left(N^{4}+r^{2}N^{2}+r^{4}\right)\right]\\\ -l^{4}N^{2}(93N^{8}+652r^{2}N^{6}+558r^{4}N^{4}+652r^{6}N^{2}+93r^{8})\\}+81mr\Gamma^{2}\end{array}$ $\begin{array}[]{c}S^{2}\times S^{2}\times S^{2}\\\ \text{ \ \ }\end{array}$ | $\begin{array}[]{c}72\left(r^{2}+N^{2}\right)\\{3N^{4}\left(3l^{2}-8N^{2}\right)\Upsilon+l^{6}\left(2N^{8}+9r^{2}N^{6}+10r^{4}N^{4}+9r^{6}N^{2}+2r^{8}\right)\\\ -6l^{4}N^{2}\left(3N^{4}+2r^{2}N^{2}+3r^{4}\right)\left(N^{4}+6r^{2}N^{2}+3r^{4}\right)\\}+9mr\Gamma^{2}\end{array}$ $\begin{array}[]{c}T^{2}\times\mathbb{CP}^{2}\\\ \text{ \ \ }\end{array}$ | $\begin{array}[]{c}108\left(r^{2}+N^{2}\right)\\{108N^{4}\left(l^{2}-4N^{2}\right)\Upsilon-l^{6}\left(r^{2}-N^{2}\right)^{4}-6l^{4}N^{2}(9N^{8}+92r^{2}N^{6}\\\ +54r^{4}N^{4}+92r^{6}N^{2}+9r^{8})\\}+81mr\Gamma^{2}\end{array}$ $\begin{array}[]{c}T^{2}\times S^{2}\times S^{2}\\\ \text{ \ \ }\end{array}$ | $\begin{array}[]{c}-2\left(r^{2}+N^{2}\right)\\{18N^{2}\left(l^{4}-12l^{2}N^{2}+48N^{4}\right)\Upsilon+l^{6}(r^{2}+N^{2})\left(r^{2}-N^{2}\right)^{2}(7N^{4}\\\ +2r^{2}N^{2}+7r^{4})\\}+9mr\Gamma^{2}\end{array}$ ${\,}T^{2}\times T^{2}\times S^{2}$ | $4\left(r^{2}+N^{2}\right)\left[108N^{4}\left(l^{2}-8N^{2}\right)\Upsilon+l^{6}\left(r^{4}-N^{4}\right)^{2}+18l^{4}N^{2}(r^{2}-N^{2})^{4}\right]+27mr\Gamma^{2}$ ${\,}\begin{array}[]{c}T^{2}\times T^{2}\times T^{2}\end{array}$ | $-64N^{6}\left(r^{2}+N^{2}\right)\Upsilon+mr\Gamma^{2}$ where $\Upsilon=(11r^{8}+84N^{2}r^{6}+66N^{4}r^{4}+84N^{6}r^{2}+11N^{8}).$ Although we have written the metric function $F(r)$ for specific values of Lovelock coefficients belonging to dimensionally continued Lovelock gravity in $8$ dimensions, the form of $F(r)$ for arbitrary values of $\alpha_{i}$’s is the same as Eq. (11) with more complicated $E$, $B$ and $C$. ### III.1 Taub-NUT Solutions: The solutions given in Eq. (11) describe NUT solutions, if (i) $F(r=N)=0$ and (ii) $F^{\prime}(r=N)=1/(4N)$. The first condition comes from the fact that all the extra dimensions should collapse to zero at the fixed point set of $\partial/\partial\tau$, and the second one is to avoid conical singularity with a smoothly closed fiber at $r=N$. Using these conditions, one finds that the third order Lovelock gravity, in eight dimensions admits NUT solutions only with $\mathbb{CP}^{3}$ base space when the mass parameter is fixed to be $m_{n}=2N\left(8N^{2}-l^{2}\right)^{3}.$ (12) Computation of the Kretschmann scalar at $r=N$ for the solutions in eight dimensions shows that the spacetimes with base spaces $S^{2}\times S^{2}\times T^{2}$, $S^{2}\times S^{2}\times S^{2}$, $T^{2}\times CP^{2}$ or $S^{2}\times CP^{2}$ have a curvature singularity at $r=N$ in Einstein gravity, while the spacetime with $\mathcal{B}=\mathbb{CP}^{3}$ has no curvature singularity at $r=N$. Thus, the conjecture given in DehMan is confirmed for third order Lovelock gravity too. Here we generalize this conjecture for Lovelock gravity as “For the non-extremal NUT solutions of Einstein gravity which have no curvature singularity at $r=N$, the Lovelock gravity admits NUT solutions, while the Lovelock gravity does not admit non-extremal NUT when the spacetime has curvature singularity at $r=N$.” Indeed, we have non-extreme NUT solutions in $2+2k$ dimensions with non-trivial fibration when the $2k$-dimensional base space is chosen to be $\mathbb{CP}^{k}$. Although we have not written the solutions in $2+2k$ dimensions and with arbitrary $\alpha_{i}$’s, but calculations confirm the above conjecture. ### III.2 Extreme Taub-NUT Solution The solutions (11) with the base space $\mathcal{B}_{2}=T^{2}\times T^{2}\times T^{2}$ and $\mathcal{B}_{3}=S^{2}\times T^{2}\times T^{2}$ satisfy the extremal NUT solutions provided the mass parameter is fixed to be $\displaystyle m_{n}^{B_{2}}$ $\displaystyle=$ $\displaystyle 128N^{7},$ (13) $\displaystyle m_{n}^{B_{3}}$ $\displaystyle=$ $\displaystyle 16N^{5}\left(8N^{2}-l^{2}\right).$ (14) Indeed for these two cases $F^{\prime}(r=N)=0$, and therefore the NUT solutions should be regarded as extremal solutions. Computing the Kretschmann scalar, we find that there is a curvature singularity at $r=N$ for the spacetime with $\mathcal{B}_{3}=S^{2}\times T^{2}\times T^{2}$, while the spacetime with $\mathcal{B}_{2}=T^{2}\times T^{2}\times T^{2}$ has no curvature singularity at $r=N$. This leads us to the generalization of second conjecture of Ref. DehMan : “ Lovelock gravity has extremal NUT solutions whenever the base space is a product of 2-torii with at most one $2$-dimensional space of positive curvature”. Indeed, calculations in other dimensions show that when the base space has at most one two dimensional curved space as one of its factor spaces, then Lovelock gravity admits an extreme NUT solution even though there exists a curvature singularity at $r=N$. ## IV Concluding Remarks We considered the existence of Taub-NUT solutions in $8$-dimensional third order Lovelock gravity with cosmological constant. Although one can do the calculations for any arbitrary values of Lovelock coefficients, we chose them as those of dimensionally continued Lovelock gravity in eight dimensions to have more compact form for the solutions. It is worthwhile to mention that this choice of Lovelock coefficients has no effect on the properties if the solutions. These solutions are constructed as circle fibrations over even dimensional spaces that in general are products of Einstein-Kähler spaces. We found that as in the case of Gauss-Bonnet gravity, the function $F(r)$ of the metric depends on the specific form of the base factors on which one constructs the circle fibration. In other words we found that the solutions are sensitive to the geometry of the base space, in contrast to Einstein gravity where the metric in any dimension is independent of the specific form of the base factors. Thus, one may say that the sensitivity of the NUT solutions on the geometry of the base space is a common feature of higher order Lovelock gravity, which does not happen in Einstein gravity. We have found that when Einstein gravity admits non-extremal NUT solutions with no curvature singularity at $r=N$, then there exists a non-extremal NUT solution in third order Lovelock gravity. In $8$-dimensional spacetime, this happens when the metric of the base space is chosen to be $\mathbb{CP}^{3}$. Indeed, third order Lovelock gravity does not admit non-extreme NUT solutions with any other base spaces. Although we have not written the NUT solutions in other dimensions, we found that in any dimension $2k+2$, we have only one non- extremal NUT solution with $\mathbb{CP}^{k}$ as the base space. That is, the Lovelock gravity singles out the preferred non-singular base space $\mathbb{CP}^{k}$ in $2+2k$ dimensions. We also found that only when the base space is $T^{2}\times T^{2}\times T^{2}$ or $S^{2}\times T^{2}\times T^{2}$, eight-dimensional third order Lovelock gravity admits extremal NUT solution. Calculations show that the extremal NUT black holes exist for the base spaces $T^{2}\times T^{2}\times T^{2}.....\times T^{2}$ and $S^{2}\times T^{2}\times T^{2}.....\times T^{2}$. We have extended these observations to two conjectures about the existence of NUT solutions in Lovelock gravity. The study of thermodynamic properties of these solutions remains to be carried out in future. ###### Acknowledgements. This work has been supported financially by Research Institute for Astronomy and Astrophysics of Maragha. ## References * (1) D. Lovelock, J. Math. Phys. 12, 498 (1971); N. Deruelle and L. Farina-Busto, Phys. Rev. D 41, 3696 (1990); G. A. Mena Marugan, _ibid_. 46, 4320 (1992); _ibid_. 4340 (1992). * (2) B. Zwiebach, Phys. Lett. B 156, 315 (1985); B. Zumino, Phys. Reports 137, 109 (1986). * (3) D. G. Boulware and S. Deser, Phys. Rev. Lett. 55, 2656 (1985); J. T. Wheeler, Nucl. Phys. B 268, 737 (1986); J. T. Wheeler, _ibid_. 273, 732 (1986); D. L. Wiltshire, Phys. Lett. B 169, 36 (1986); R. G. Cai, Phys. Rev. D 65, 084014 (2002); R. G. Cai and Q. Guo _ibid_. 69, 104025 (2004); R. G. Cai, Phys. Lett. B 582, 237 (2004); M. Cvetic, S. Nojiri, S. D. Odintsov, Nucl. Phys. B 628, 295 (2002); M. H. Dehghani, Phys. Rev. D 67, 064017 (2003); _ibid_. 69, 064024 (2004); _ibid_. 70, 064019 (2004); M. H. Dehghani, G. H. Bordbar and M. Shamirzaie, _ibid_. 74, 064023 (2006); M. H. Dehghani and M. Shamirzaie, _ibid_. 72, 124015 (2005); M. H. Dehghani and R. B. Mann, _ibid_. 73, 104003 (2006); M. H. Dehghani and S. H. Hendi, Int. J. Mod. Phys. D 16, 1829 (2007); M. H. Dehghani, N. Bostani and A. Sheykhi, Phys. Rev. D 73, 104013 (2006); M. H. Dehghani, N. Alinejadi and S. H. Hendi, _ibid_. 77, 104025 (2008). * (4) M. H. Dehghani and R. B. Mann, Phys. Rev. D 72, 124006 (2005). * (5) M. H. Dehghani and S. H. Hendi, Phys. Rev. D 73, 084021 (2006). * (6) A. H. Taub, Annal. Math. 53, 472 (1951); E. Newman, L. Tamburino and T. Unti, J. Math. Phys. 4, 915 (1963). * (7) F. A. Bais and P. Batenburg, Nucl. Phys. B 253, 162 (1985); D. N. Page and C. N. Pope, Class. Quant. Grav. 4, 213 (1987); M. M. Taylor-Robinson, [hep-th/9809041]; M. M. Akbar and G. W. Gibbons, Class. Quant. Grav. 20, 1787 (2003). * (8) A. Awad and A. Chamblin, Class. Quant. Grav. 19, 2051 (2002). * (9) R. Clarkson, L. Fatibene and R. B. Mann, Nucl. Phys. B 652 348 (2003). * (10) Robert Mann and Cristian Stelea, Class. Quant. Grav. 21, 2937 (2004); R. B. Mann and C. Stelea, Phys. Lett. B 634, 448 (2006). * (11) M. H. Dehghani and A. Khodam-Mohammadi, Phys. Rev. D 73, 124039 (2006). * (12) M. Banados, M. Henneaux, C. Teitelboim, and J. Zanelli, Phys. Rev. D 48, 1506 (1993); M. Banados, C. Teitelboim, and J. Zanelli, _ibid_. 49, 975 (1994). * (13) P. Hoxha, R. R. Martinez-Acosta, C. N. Pope, Class. Quant. Grav. 17, 4207 (2000).

arxiv-papers

2024-09-04T02:48:53.748901

0802.1852

# Knudsen Diffusion in Silicon Nanochannels Simon Gruener and Patrick Huber p.huber@physik.uni-saarland.de Faculty of Physics and Mechatronics Engineering, Saarland University, D-66041 Saarbrücken, Germany ###### Abstract Measurements on helium and argon gas flow through an array of parallel, linear channels of $12$ nm diameter and $200$ $\rm\mu$m length in a single crystalline silicon membrane reveal a Knudsen diffusion type transport from $10^{\rm 2}$ to $10^{\rm 7}$ in Knudsen number Kn. The classic scaling prediction for the transport diffusion coefficient on temperature and mass of diffusing species, $D_{\rm He}\propto\sqrt{T}$ is confirmed over a $T$-range from 40 K to 300 K for He and for the ratio of $D_{\rm He}/D_{\rm Ar}\propto\sqrt{m_{\rm Ar}/m_{\rm He}}$. Deviations of the channels from a cylindrical form, resolved with electron microscopy down to subnanometer scales, quantitatively account for a reduced diffusivity as compared to Knudsen diffusion in ideal tubular channels. The membrane permeation experiments are described over 10 orders of magnitude in Kn, encompassing the transition flow regime, by the unified flow model of Beskok and Karniadakis. ###### pacs: 47.45. n, 47.61.-k, 47.56.+r Knudsen diffusion (KD) refers to a gas transport regime where the mean free path between particle-particle collisions $\lambda$ is significantly larger than at least one characteristic spatial dimension $d$ of the system considered Knudsen (1909). By virtue of the negligible mutual particle collisions transport in such systems takes place via a series of free flights and statistical flight direction changes after collisions with the confining walls. Because of the dependency of $\lambda$ on pressure $p$ and temperature $T$ given by the kinetic theory of gases, $\lambda\propto T/p$, this transport regime is only observable in macroscopic systems at very low $p$ or elevated $T$. By contrast, for transport in spatially nanoconfined systems with $d=\mathcal{O}(10$ nm$)$, the Knudsen number $Kn=\lambda/d$, which quantifies the gas rarefaction, is larger than 1 even for ambient pressures and temperatures, e.g. $\lambda(\text{He})=118$ nm at $p=1$ bar and $T=297$ K. Thus, for many processes involving gas transport in restricted geometries, e.g. gas catalysis and storageMRS (2006) or equilibration phenomena via gas flow in meso- and nanoporesWallacher (2004); Huber and Knorr (1999), KD plays a crucial role. Figure 1: (color online) He pressure relaxations in R1 and R2 at $T=297$ K for starting pressures $p_{\rm s}=0.1$ mbar (stars) and $100$ mbar (triangles) without (a) and with SiNC membrane (b). (c) Pressure relaxation time $\tau$ as determined from measurements without (squares) and with SiNC membrane (circles) versus Knudsen number in the capillaries $Kn_{\rm c}$ (bottom axis) and in the SiNCs $Kn$ (top axis). The lines in (a), (b) and (c) represent calculated $p(t)$ and $\tau(Kn)$ values. Insets in (c): TEM cross section views of two SiNCs in comparison with circular perimeters of 12 nm diameter (dashed lines). Albeit the phenomenon of KD has been known for almost a hundred years, details on its microscopic mechanisms are still at debate, in particular how wall roughness down to atomic scales and particle-wall interaction parameters affect the statistics of the diffusion process and thus the value of the KD transport coefficient $D$ Theo (2005). In the past, a comparison between theory and experiment has often been hampered by complex pore network topologies, tortuous transport paths and poorly characterized pore wall structures in available nanoporous matrices. Nowadays, membranes with better defined geometries, such as carbon nanotube bundlesHinds et al. (2004); Holt et al. (2006), porous alumina Roy et al. (2003), and tailored pores in mesoporous silicon Wallacher (2004); Striemer (2007), allow us to address such fundamental questions in more detail. A better understanding of, and more in- depth information on, transport in such systems is not only of academic interest, but also of importance for the architecture and functionality in the emerging field of nanoelectromechanical and nanofluidic systems Karniadakis et al. (2005); Whitby and Quirke (2007). In this Letter, we report the first rigorous experimental study on transport of simple rare gases, i.e. He and Ar through an array of parallel aligned, linear channels with $\sim 12$ nm diameter and $200$ $\mu$m length in single crystalline Si over a wide temperature ($40\text{ K}<T<300$ K) and $Kn$-range ($10^{\rm 2}<Kn<10^{\rm 7}$). We explore the classical scaling predictions given by Knudsen for $D$ on $T$, mass of diffusing species, $m$, and explore its dependency on $Kn$. Attention shall also be paid to the morphology of the channel walls, resolved with transmission electron microscopy (TEM), and how it affects the diffusion. Linear, non-interconnected channels oriented along the $\langle 100\rangle$ Si crystallographic direction (SiNCs) are electrochemically etched into the surface of a Si (100) wafer Lehmann and Gösele (1991). After the channels reach the desired length $L$ of $200\pm 5~{}\mu$m, the anodization current is increased by a factor of 10 with the result that the SiNC array is detached from the bulk wafer Henschel et al. (2007). The crystalline Si walls call for irregular channel perimeters as one can see in the TEM pictures of Fig. 1(c), recorded from a membrane part Ar ion milled to $\sim 5$ nm thickness. In first approximation, however, the SiNC’s cross-section can be described as circular with a diameter of $\sim 12$ nm, in accordance with the analysis of an Ar sorption experiment at $T=86$ K, which yields additionally a SiNCs density of $\Phi=(1.5\pm 0.1)\cdot 10^{\rm 11}$ cm-2. Our experimental setup consists of a copper cell with an inlet and outlet opening MFA2007 . Inlet and outlet are connected via stainless steel capillaries of radius $r_{\rm c}=0.7$ mm and length $l_{\rm c}=70$ cm with two gas reservoirs, R1 and R2, of an all-metal gas handling. Additionally two pneumatic valves V1 and V2 are used to open and close the connections between the sample cell and R1 and the sample cell and R2. Four thermostatted capacitive pressure gauges allow us to measure the gas pressures in R1 and R2, $p_{\rm 1}$ and $p_{\rm 2}$, resp., over a wide pressure range (5$\cdot 10^{\rm-3}$ mbar $<p<$1 bar) with an accuracy of $10^{\rm-3}$ mbar. The cell is mounted in a closed-cycle He cryostat to control the temperature between 40 K and 300 K with an accuracy of $1$ mK. Our goal of studying gas transport dynamics over a large $Kn$ range necessitates a thorough understanding of the intrinsic flow characteristics of our apparatus. Therefore, we start with measurements on He flow through the macroscopic capillaries and the empty sample cell at room temperature $T=297$ K. We record the equilibration of $p_{\rm 1}(t)$ and $p_{\rm 2}(t)$ towards a pressure $p_{\rm eq}$ as a function of time $t$ after initial conditions, $p_{\rm 1}=p_{\rm s}>p_{\rm 2}=0$ mbar at $t=0$ s. In Fig. 1 (a) $p_{\rm 1}(t)/p_{\rm s}$ and $p_{\rm 2}(t)/p_{\rm s}$ are plotted for starting pressures of $p_{\rm s}=0.1$ and 100 mbar. We observe pressure equilibrations towards $p_{\rm eq}={0.4\cdot p_{\rm s}}$. The value 0.4 is dictated by the volume ratio of R1 to R2. From the $p(t)$-curves we derive characteristic relaxation times $\tau$ according to the recipe $p_{\rm 1}(t=\tau)-p_{\rm 2}(t=\tau)\equiv 1/10\cdot p_{\rm s}$. It is understood that $p$ is monotonically decreasing downstream from R1 to R2. In order to quantify the gas rarefaction we resort, therefore, to a calculation of a mean Knudsen number $Kn_{\rm c}=\lambda(\overline{p})/r_{\rm c}$ in the capillaries assuming a mean pressure, $\overline{p}=p_{\rm eq}$. This simplification is justified by the analysis provided below and by theoretical studies which indicate differences of less than 1% between flow rates calculated with an exact and an averaged treatment for $Kn$ Sharipov, Seleznev (1994); Karniadakis et al. (2005). In Fig. 1 (c) we plot measured $\tau$ values versus $Kn_{\rm c}$ corresponding to a $p_{\rm s}$ variation from $0.01$ to $100$ mbar. For $Kn_{\rm c}<0.1$, $\tau$ increases with increasing $Kn_{\rm c}$. In an intermediate $Kn_{\rm c}$ range, $0.1<Kn_{\rm c}<0.6$, we observe a cross- over regime towards a saturation plateau with $\tau\sim 140$ s that extends to the largest $Kn_{\rm c}$ studied. These changes in the $p$ relaxation and hence flow dynamics are reminiscent of the three distinct transport regimes known to occur for gases as a function of their rarefaction Karniadakis et al. (2005); Tabeling (2005): For $Kn_{\rm c}<0.1$ the number of interparticle collisions still predominates over the number of particle-wall collisions. Hagen-Poiseuille’s law is valid and predicts a decreasing flow rate and hence increasing $\tau$ due to the $1/Kn$-scaling of the particle number density in gas flows. For $Kn_{\rm c}>1$ we enter the pure KD regimeKnudsen (1909), where theory predicts a He KD transport coefficient $D_{\rm c}^{\rm He}$ dependent, however a $Kn_{\rm c}$ independent particle flow rate, $\dot{n}_{\rm Kn}=\pi\frac{r_{\rm c}^{2}}{l_{\rm c}}\frac{p_{\rm o}-p_{\rm i}}{k_{\rm B}T}D_{\rm c}^{\rm He},$ (1) which is responsible for the plateau in $\tau$ for the larger $Kn_{\rm c}$ investigated. In Eq. 1 $p_{\rm i}$, $p_{\rm o}$ refers to the inlet and outlet pressure of the capillary considered, resp., and $k_{\rm B}$ to the Boltzmann factor. In the intermediate $Kn$-range, the interparticle collisions occur as often as particle-wall collisions which gives rise to the cross-over behavior found for $\tau$. We elucidate this behavior in more detail by dividing the flow path within our apparatus into two flow segments (up- and downstream capillary) and calculate the particle number changes along the flow path and the resulting $p_{\rm 1}(t)$, $p_{\rm 2}(t)$ with a $1$ ms resolution using the unified flow model of Beskok and Karniadakis (BK-model)Karniadakis et al. (2005); Beskok (1999); Grüner (2006) and a local $Kn$-number for each flow segment, $Kn_{\rm l}=Kn((p_{\rm i}+p_{\rm o})/2)$: $\dot{n}_{\rm BK}=\frac{\pi r_{\rm c}^{4}(p_{\rm o}^{\rm 2}-p_{\rm i}^{\rm 2})}{16\,l_{\rm c}\,k_{\rm B}\,T}\frac{1+\alpha\,Kn_{\rm l}}{\mu(T)}\left(1+\frac{4\,Kn_{\rm l}}{1-b\,\,Kn_{\rm l}}\right).$ (2) Eq. 2 comprises a Hagen-Poiseuille term, a term which treats the transition of the transport coefficient from continuum-like, i.e. the bare dynamic viscosity $\mu(T)$, to the KD transport coefficient, $D_{\rm c}^{\rm He}$ with $\alpha=2.716/\pi\tan^{\rm-1}(\alpha_{\rm 1}Kn^{\beta})$, and a generalized velocity slip term, which is second-order accurate in $Kn$ in the slip and early transition flow regimes ($Kn<0.5$). The model captures for $Kn\rightarrow 0$ the no-slip Hagen-Poiseuille limit, whereas it transforms to Eq. 1 for $Kn\gg 1$. As verified by comparison of the BK-model with Direct Simulation Monte Carlo and solutions of the Boltzmann equation, a choice of $b=-1$ for the slip parameter results in the correct velocity profile and flowrate, as well as a proper pressure and shear stress distribution in a wide $Kn$-range, including the transition flow regime. After optimizing the free parameters in Eq. 2, $\alpha_{\rm 1}$ and $\beta$, we arrive at the $p(t)$\- and $\tau(Kn_{\rm c})$-curves depicted in Fig. 1 (a) and (c), resp. The good agreement of the BK-model predictions with our measurements is evident and the extracted parameters $\alpha_{\rm 1}=10\pm 0.2$ and $\beta=0.5\pm 0.05$ ($D_{\rm c}^{\rm He}=0.412\pm 0.06$ m2/s) agree with BK-modelling of He gas flow as a function of rarefaction Karniadakis et al. (2005); Tison (1993). An analogous analysis for Ar gas flow yields $\alpha_{\rm 1}=1.55\pm 0.05$ and $\beta=0.1\pm 0.02$ ($D_{\rm c}^{\rm Ar}=0.134\pm 0.02$ m2/s). Figure 2: (color online) Pressure relaxation time $\tau$ for He (circles) and Ar (triangles) measured with built-in SiNC membrane at $T=297$ K versus Knudsen number in the SiNCs $Kn$ (bottom axis) and in the capillaries $Kn_{\rm c}$ (top axis). The lines represent calculated $\tau(Kn)$-curves. Inset: TEM cross section of a SiNC in comparison with a circular channel perimeter of 12 nm diameter (dashed line). The SiNC’s perimeter is highlighted by a line. Having achieved a detailed understanding of the intrinsic flow characteristics of our apparatus, we can now turn to measurements with the SiNC membrane. The membrane is epoxy-sealed in a copper ring MFA2007 and special attention is paid to a careful determination of the accessible membrane area, $A=0.79\pm 0.016$ cm2 in order to allow for a reliable determination of the number of SiNCs inserted into the flow path, $N=\Phi\cdot A=(11.8\pm 1)\cdot 10^{\rm 10}$. As expected the pressure equilibration is significantly slowed down after installing the membrane - compare panel (a) and (b) in Fig. 1. Choosing selected $p_{\rm s}$ within the range 0.005 mbar to 100 mbar, which corresponds to a variation of $Kn$ in the SiNCs from $10^{\rm 2}$ to $10^{\rm 7}$, we find an increase in $\tau$ of $\sim 170$ s, when compared to the empty cell measurements - see Fig. 1 (c). From the bare offset in $\tau$ for $Kn_{\rm c}>1$, we could calculate $D_{\rm He}$ in the SiNCs. We are, however, interested in the $D_{\rm He}$ behavior in a wide $Kn$-range, therefore we modify our flow model by inserting a segment with KD transport mechanism characteristic of $N$ tubular channels with $d=12$ nm and $L=200$ $\rm\mu$m in between the capillary flow segments. Adjusting the single free parameter in our simulation, the value of the He diffusion coefficient $D_{\rm He}$ in a single SiNC, we arrive at the $p(t)$\- and $\tau(Kn)$-curves presented as solid lines in Fig. 1 (b) and (c), resp. The agreement with our measurement is excellent and the model yields a $Kn$ independent $D_{\rm He}=3.76\pm 0.8$ mm2/s. It is worthwhile to compare this value with Knudsen’s prediction for $D$. In his seminal paper he derives an expression for $D$ with two contributions, a factor $G$, characteristic of the KD effectivity of the channel’s shape, and a factor solely determined by the mean thermal velocity of the particles $\overline{v}$ given by the kinetic theory of gases, $D=\frac{1}{3}\;G\;\overline{v}=\frac{1}{3}\;G\;\sqrt{\frac{8k_{\rm B}T}{\pi m}}.$ (3) Interestingly, by an analysis of the number of particles crossing a given section of a channel in unit time after completely diffuse reflection from an arbitrary element of wall surface and while assuming an equal collision accessibility of all surface elements, Knudsen derived an analytical expression for $G$ for a channel of arbitrary shape. Knudsen’s second assumption is, however, only strictly valid for circular channel cross- sections, as pointed out by v. Smoluchowski Knudsen (1909) and elaborated for fractal pore wall morphologies by Coppens and Froment Theo (2005). Given the roughly circular SiNCs’ cross section shapes, we nevertheless resort to Knudsen’s formula, here quoted normalized to the KD form factor for a perfect cylinder: $\frac{1}{G}=\frac{1}{4L}\;\int_{0}^{L}{\frac{o(l)}{A(l)}\;dl}.$ (4) The integral in Eq. 4 depends on the ratio of perimeter length $o(l)$ and cross sectional area $A(l)$ along the channel’s long axis direction $l$, only. This ratio is optimized by a circle, accordingly the most effective KD channel shape is a cylinder, provided one calls for a fixed $A$ along the channel. Eq. 4 yields due to our normalization just $G=d$ and we would expect a $D_{\rm He}$ of $5$ mm2/s for He KD in a SiNC, if it were a cylindrical channel of $d=12$ nm. A value which is $\sim 30\%$ larger than our measured one. If one recalls the TEM pictures, which clearly indicate non-circular cross-sections this finding is not surprising. In fact these pictures deliver precisely the information needed for an estimation of the influence of the SiNC’s irregularities on the KD dynamics. We determine the ratio $o/A$ of 20 SiNC cross sections, and therefrom values of $G$. The values of the SiNC in Fig. 1 (top), (bottom) and Fig. 2 corresponds to $G$s of 9.1 nm, 10.3 nm, and 10.6 nm, resp. Tacitly assuming that the irregularities exhibited on the perimeter are of similar type as along the channel’s long axis we take an ensemble average over the 20 $G$ values and arrive at a mean $G$ of 9.9 nm, which yields a $D_{\rm He}$ of $4.1$ mm2/s, a value which agrees within the error margins with our measured one. To further explore the KD transport dynamics in the SiNCs we now focus on the $m$ and $T$ dependency of $D$. In Fig. 2 the $\tau(Kn)$ curve recorded for Ar and He at $T=297$ K are presented. Both exhibit a similar form, the one of Ar is, however, shifted up markedly towards larger $\tau$. Our computer model can quantitatively account for this slowed-down dynamics by a factor $2.97\pm 0.25$ decrease in $D$ as compared to the He measurements, which confirms the prediction of Eq. 3, $D_{\rm He}/D_{\rm Ar}=\sqrt{m_{\rm Ar}/m_{\rm He}}=3.16$. For the exploration of the $T$ behavior we choose again He as working fluid due to its negligible physisorption tendency, even at low $T$. We perform measurements at selected $T$s from $297$ K down to $40$ K, shown in Fig. 3. We again perform computer calculations assuming KD in the SiNC array superimposed to the transport in the supply capillaries, presented as lines in Fig. 3, and optimize the single free parameter $D(T)$ in order to match the observed $\tau(Kn,T)$ behavior. Despite unresolvable deviations at larger $Kn_{\rm c}$ and decreasing $T$, which we attribute to thermal creep effects, characteristic of $T$ gradients along the flow path Karniadakis et al. (2005), we find, in agreement with the experiment, increasingly faster dynamics with decreasing $T$. Note, this counter-intuitive finding for a diffusion process results from the $1/T$ scaling in Eq. 1, which reflects the $T$-dependency of the particle number density in gas flows. By contrast, the $D_{\rm He}(T)$, determined by our simulations, scales in excellent agreement with $\sqrt{T}$, see inset in Fig. 3, confirming Knudsen’s classic result down to the lowest $T$ investigated. Figure 3: (color online) Pressure relaxation time $\tau$ for He gas flow through SiNC membrane (symbols) and simulated $\tau$(Kn) (lines) versus Knudsen number in the SiNC membrane $Kn$ (bottom axis) and in the supply capillaries $Kn_{\rm c}$ (top axis) at selected temperatures $T$. Inset: He diffusion coefficient $D_{\rm He}$ in SiNCs in comparison with the $\sqrt{T}$ prediction of Eq. 3 (line) plotted versus $\sqrt{T}$. We find no hints of anomalous fast KD here, as was recently reported for a bundle of linear carbon nanotubes Holt et al. (2006) and explained by an highly increased fraction of specularly reflected particles upon wall collisions Knudsen (1909). The altered collision statistics was attributed to the crystalline structure and the atomical smoothness of the nanotube walls. The SiNC walls are also crystalline, however, not atomical flat, as can be seen from our TEM analysis. Along with the formation of a native oxide layer, typical of Si surfaces, which renders the near surface structure amorphous, silica like, this, presumably, accounts for the normal KD observed here contrasting the one found for the graphitic walls of carbon nanotubes. We presented here the first detailed study of gas transport in linear SiNCs. Our conclusions are drawn from a correct treatment of gas flow over 10 orders of magnitude in gas rarefaction, which is, to the best of our knowledge, the largest $Kn$ range ever explored experimentally. The characteristic properties of KD, an independency of $D$ on $Kn$, its scaling predictions on $m$, $T$, and on details of the channel’s structure, resolved with sub-nm resolution, are clearly exhibited by our measurements. ###### Acknowledgements. We thank A. Beskok for helpful discussions and the German Research Foundation (DFG) for support within the priority program 1164, Nano- & Microfluidics (Hu 850/2). ## References * Knudsen (1909) M. Knudsen, Ann. Phys. (Leipzig) 333, 75 (1909), M. v. Smoluchowski, ibid. 338, 1559 (1910). * MRS (2006) J. Kärger and D. Ruthven, _Diffusion in Zeolites and Microporous Solids_ (Wiley & Sons, New York, 1992), S. M. Auerbach, Int. Rev. Phys. Chem. 19, 155 (2000). * Wallacher (2004) D. Wallacher et al., Phys. Rev. Lett. 92, 195704 (2004). * Huber and Knorr (1999) P. Huber and K. Knorr, Phys. Rev. B 60, 12657 (1999); R. Paul and H. Rieger, J. Chem. Phys. 123, 024708 (2005); R. Valiullin et al., Nature 443, 965 (2006). * Theo (2005) M. O. Coppens and G. F. Froment, Fractals 3, 807 (1995); A.I. Skoulidas et al., Phys. Rev. Lett. 89, 185901 (2002); G. Arya, H.C. Chang, and E.J. Maginn, ibid. 91, 026102 (2003); S. Russ et al., Phys. Rev. E 72, 030101(R) (2005). * Hinds et al. (2004) B. J. Hinds et al., Science 303, 62 (2004). * Holt et al. (2006) J. K. Holt et al., Science 312, 1034 (2006). * Roy et al. (2003) S. Roy et al., J. Appl. Phys. 93, 4870 (2003); K.J. Alvine et al.. Phys. Rev. Lett. 97, 175503 (2006). * Striemer (2007) C.C. Striemer et al., Nature 445, 749 (2007). * Karniadakis et al. (2005) G. Karniadakis, A. Beskok, and N. Aluru, _Microflows and Nanoflows_ (Springer, 2005). * Whitby and Quirke (2007) H.A. Stone, A.D. Stroock, and A. Ajdari, Ann. Rev. Fluid Mech. 36, 381 (2004); P. Huber et al., Eur. Phys. J. Spec. Top. 141, 101 (2007); M. Whitby and N. Quirke, Nature Nanotechnology 2, 87 (2007). * Lehmann and Gösele (1991) V. Lehmann and U. Gösele, Appl. Phys. Lett. 58, 856 (1991); V. Lehmann, R. Stengl, and A. Luigart, Mat. Sc. Eng. B 69-70, 11 (2000). * Henschel et al. (2007) A. Henschel et al., Phys. Rev. E 75, 021607 (2007); P. Kumar et al., J. Appl. Phys. 103, 024303 (2008). * (14) See EPAPS Document No. E-PRLTAO-100-019807 for a raytracing illustration of the gas flow apparatus employed for the membrane permeation measurements. For more information on EPAPS, see http://www.aip.org/pubservs/epaps.html. * Sharipov, Seleznev (1994) F. M. Sharipov and V. D. Seleznev, J. Vac. Sc. Technol. A 12, 2933 (1994). * Tabeling (2005) P. Tabeling _Introduction to Microfluidics_ (Oxford University Press, New York, 2005). * Beskok (1999) A. Beskok and G. Karniadakis, Nanosc. Microsc. Thermophys. Eng. 3, 43 (1999). * Grüner (2006) S. Grüner, Diploma thesis, Saarland University (2006). * Tison (1993) S. Tison, Vacuum 44, 1171 (1993).

arxiv-papers

2024-09-04T02:48:53.753421

0802.2095

# The Nascent Red-Sequence at ${\mathbf{z}\sim 2}$11affiliation: Based on observations with the NASA/ESA Hubble Space Telescope, obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555 Andrew W. Zirm22affiliation: Johns Hopkins University, 3400 N. Charles Street, Baltimore, MD, 21218; azirm@pha.jhu.edu , S.A. Stanford33affiliation: Institute of Geophysics and Planetary Physics, LLNL, L-413, Livermore, CA 94550 44affiliation: Physics Dept., University California at Davis, One Shields Avenue, Davis, CA 95616 , M. Postman55affiliation: Space Telescope Science Institute, 3700 San Martin Drive, Baltimore, MD 21218 , R.A. Overzier22affiliation: Johns Hopkins University, 3400 N. Charles Street, Baltimore, MD, 21218; azirm@pha.jhu.edu , J.P. Blakeslee66affiliation: Department of Physics and Astronomy, Washington State University, Pullman, WA 99164 77affiliation: Herzberg Institute of Astrophysics, NRC of Canada, 5071 W. Saanich Road, Victoria, BC, V9E 2E7, Canada , P. Rosati88affiliation: ESO- European Southern Observatory, Karl-Schwarzschild-Strasse 2, D-85748, Garching bei München, Germany , J. Kurk99affiliation: Max-Planck-Institut für Astronomie, D-69117 Heidelberg, Germany , L. Pentericci1010affiliation: Osservatorio Astronomico di Roma, I-00040 Monte Porzio Catone, Italy , B. Venemans1111affiliation: IOA, Cambridge University, Madingley Road, Cambridge, CB3 0HA, UK , G.K. Miley1212affiliation: Leiden Observatory, Leiden University, Postbus 9513, NL-2300 RA Leiden, The Netherlands , H.J.A. Röttgering, M. Franx, A. van der Wel22affiliation: Johns Hopkins University, 3400 N. Charles Street, Baltimore, MD, 21218; azirm@pha.jhu.edu , R. Demarco22affiliation: Johns Hopkins University, 3400 N. Charles Street, Baltimore, MD, 21218; azirm@pha.jhu.edu , W. van Breugel33affiliation: Institute of Geophysics and Planetary Physics, LLNL, L-413, Livermore, CA 94550 1313affiliation: University of California, Merced, PO Box 2039, Merced, CA 95344 ###### Abstract We present new constraints on the evolution of the early-type galaxy color- magnitude relation (CMR) based on deep near-infrared imaging of a galaxy protocluster at $z=2.16$ obtained using NICMOS on-board the Hubble Space Telescope. This field contains a spectroscopically confirmed space-overdensity of Lyman-$\alpha$ and H-$\alpha$ emitting galaxies which surrounds the powerful radio galaxy MRC 1138-262. Using these NICMOS data we identify a significant surface-overdensity ($=6.2\times$) of red $J_{110}-H_{160}$ galaxies in the color-magnitude diagram (when compared with deep NICMOS imaging from the HDF-N and UDF). The optical-NIR colors of these prospective red-sequence galaxies indicate the presence of on-going dust-obscured star- formation or recently formed ($\mathrel{\hbox to0.0pt{\lower 3.0pt\hbox{$\mathchar 536\relax$}\hss}\raise 2.0pt\hbox{$\mathchar 316\relax$}}1.5$ Gyr) stellar populations in a majority of the red galaxies. We measure the slope and intrinsic scatter of the CMR for three different red galaxy samples selected by a wide color cut, and using photometric redshifts both with and without restrictions on rest-frame optical morphology. In all three cases both the rest-frame $U-B$ slope and intrinsic color scatter are considerably higher than corresponding values for lower redshift galaxy clusters. These results suggest that while some relatively quiescent galaxies do exist in this protocluster both the majority of the galaxy population and hence the color-magnitude relation are still in the process of forming, as expected. ###### Subject headings: galaxies: evolution — galaxies: formation — galaxies: high-redshift — galaxies: stellar content — galaxies: clusters: individual: MRC 1138-262 ††journal: Accepted for publication in the Astrophysical Journal ## 1\. Introduction The color-magnitude diagram is a powerful diagnostic of galaxy evolution and formation. The presence, as early as $z\sim 1.5$, of a prominent and low- scatter, ‘red-sequence’ (RS) in galaxy clusters places useful constraints on the possible evolutionary pathways in galaxy color and luminosity (Mullis et al., 2005; Stanford et al., 2005, 2006; Bell et al., 2004; Faber et al., 2007). The red colors of the primarily early-type RS galaxies are due to the observed filters spanning the 4000Å spectral break. The universality and prominence of the RS in appropriately chosen filters have been used to discover high-redshift clusters (e.g., Gladders & Yee, 2005). Moreover, the defining characteristic of galaxy clusters, i.e., the large numbers of galaxies all at the same redshift, allows the slope and intrinsic scatter of the RS to be measured with great precision. Based on studies of galaxy clusters at $z<1.3$, the slope of the RS does not appear to evolve and therefore is more likely the by-product of the mass-metallicity relation as observed in local galaxy samples (e.g., Tremonti et al., 2004) rather than the result of a mass-age trend. The scatter, however, is likely due to the fractional age differences between the RS galaxies (e.g., Blakeslee et al., 2003). By constructing a set of model galaxies with different star-formation histories and timescales it is possible to fit for the mean epoch of last significant star-formation by matching the intrinsic scatter of the RS. Such studies at $z\sim 1$ have derived formation redshifts of $z_{\rm form}\sim 2.0-2.5$ (e.g., Blakeslee et al., 2006; van Dokkum & van der Marel, 2007). At redshifts beyond $z\sim 1.5$, however, the 4000Å-break moves into the near- infrared and galaxy clusters, and therefore the RS, have not been observed closer to the inferred epoch of formation for early-type galaxies. Hence, to uncover the younger or forming red-sequence at higher redshifts requires deep near-infrared imaging of suspected (or, preferably, confirmed) protocluster fields. Figure 1.— $J_{110}-H_{160}$ vs. $H_{160}$ color-magnitude diagram for the MRC 1138-262 NICMOS field (open black circles). The large yellow star is the radio galaxy itself. The blue background points are from the NICMOS data of the Ultra Deep Field and Hubble Deep Field North. The deep fields cover 2.5x the area of our observations. Also shown are the spectroscopically confirmed $H\alpha$ (orange filled circles) and Lyman-$\alpha$ (green filled circles) emitters. The three dot-dash lines show where the CMRs of lower redshift clusters would lie under different assumptions. The top line is the Coma cluster with no evolution, simply redshifted to $z=2.16$ and observed through the NICMOS filters. The next line down is the same but for the $z=1.24$ cluster RDCS1252. Finally, if we passively de-evolve RDCS1252 to redshift $z=2.16$ (almost exactly two Gyr), assuming a median age for the 1252 galaxies of about 3 Gyr (or $z_{\rm form}\sim 3$), we get the third line. We have undertaken a NICMOS imaging program to study the red galaxy population in a protocluster at $z=2.16$. Broad and narrow-band imaging, both in the optical and near-infrared, of the field surrounding the powerful radio galaxy MRC 1138-262 ($z=2.16$) have identified more than 100 candidate companion galaxies. This target served as the proof-of-concept for the successful VLT Large Program summarized in Venemans et al. (2007). There are surface- overdensities of both line-emitting candidates (Lyman-$\alpha$ and H$\alpha$), X-ray point sources, sub-mm selected galaxies and red optical–near-infrared galaxies (Pentericci et al., 2002; Kurk, 2003; Kurk et al., 2004a, b; Croft et al., 2005; Stevens et al., 2003). Fifteen of the Ly$\alpha$ and 9 of the H$\alpha$ emitters have now been spectroscopically confirmed to lie at the same redshift as the radio galaxy. The $I-K$-selected Extremely Red Objects (EROs; $I-K>4.3$ Vega) seem concentrated around the RG but have no spectroscopic redshifts at this time. However, by obtaining deep images through the NICMOS $J_{110}$ and $H_{160}$ filters, which effectively span the 4000Å-break at $z=2.16$, accurate and precise colors and basic morphological parameters can be measured for the red galaxy population. In this paper we present the first results from this project. The article is organized as follows: in Section §2 we describe the data and their reductions, in Section §5 we present the comparison between the red galaxy counts in this field and in deep field data, in Section §6 we present the full color-magnitude diagram and our fits to the “red sequence.” We use a $(\Omega_{\Lambda},\Omega_{M})=(0.73,0.27)$, $H_{0}=71$ ${\rm km}$ ${\rm s^{-1}}$ ${\rm Mpc^{-1}}$ cosmology throughout. At $z=2.16$ one arcsecond is equivalent to 8.4 kpc. All magnitudes are referenced to the AB system (Oke, 1974) unless otherwise noted. ## 2\. Observations, Data Reductions and Photometry Figure 2.— $g_{435}-I_{814}$ vs. $J_{110}-H_{160}$ color-color diagram using the ACS and NICMOS data. Arrows represent limits where the galaxy is only detected in a single band for that color. Filled circles indicate spectroscopically-confirmed Lyman-$\alpha$ (green) and H-$\alpha$ (orange) emitting protocluster members. The yellow star is the radio galaxy. The blue, green and red grids indicate the regions occupied by galaxies with an exponentially-decaying star-formation rate $\tau=0.15$ Gyr (red), $\tau=0.4$ Gyr (green) and a $\tau=1000$ Gyr (blue) at $z=2.16$ for three ages (0.1, 1 and 3 Gyr) and two different extinctions ($E(B-V)=0.0,1.0$). The NICMOS instrument on-board HST is capable of deep near-infrared imaging more quickly than from the ground but with a relatively small field-of-view ($51\arcsec\times 51\arcsec$). In the case of MRC 1138-262, we know that galaxies are overdense on the scale of a few arcminutes (Kurk et al., 2004a; Croft et al., 2005) and are thus well-suited for observations with NICMOS camera 3 on HST. We used 30 orbits of HST time to image 10 of the 24 confirmed members and $\sim 70$ of the candidate (narrow-band excess sources and EROs) protocluster members in both the $J_{110}$ and $H_{160}$ filters. We used seven pointings of NICMOS camera 3 in both filters and one additional pointing in $H_{160}$ alone. This single ‘outrigger’ $H_{160}$ pointing was included to obtain rest-frame optical morphological information for a small concentration of candidate members. These observations reach an AB limiting magnitude ($m_{10\sigma}$; 10$\sigma$, $0\farcs 5$ diameter circular aperture) of $m_{10\sigma}=24.9$ mag in $J_{110}$ and $m_{10\sigma}=25.1$ mag in $H_{160}$. The same field was imaged in the $g_{475}$ ($m_{10\sigma}=27.5$ mag) and $I_{814}$ ($m_{10\sigma}=26.8$ mag) filters using the Wide-Field Channel of the Advanced Camera for Surveys on HST as part of a Guaranteed Time program (# 10327; Miley et al. 2006). These optical data are useful for their higher angular resolution and their coverage of the rest-frame far-UV, thus extending the observed SEDs of candidate protocluster members to shorter wavelengths where young stars and on-going star-formation dominate the emitted spectrum. In particular, the $g_{475}$ and $I_{814}$ data allow us to partially differentiate obscured star-formation from evolved stellar populations in the candidate RS galaxies. The NICMOS images were reduced using the on-the-fly reductions from the HST archive, the IRAF task PEDSKY and the dither/drizzle package to combine the images in a mosaic. The dither offsets were calculated using image cross- correlation and were refined with one further iteration of cross-correlation. Alignment of the pointings relative to each other was accomplished using a rebinned version of the ACS $I_{814}$ image as a reference. The final mosaic has a pixel scale of $0\farcs 1$. Galaxies were selected using the $H_{160}$-band image for detection within SExtractor (Bertin & Arnouts, 1996). We used a $2.2\sigma$ detection threshold with a minimum connected area of 10 pixels. We also corrected the NICMOS data for the count-rate dependent non- linearity (de Jong, 2006). Total galaxy magnitudes were estimated by using the MAG_AUTO values from SExtractor. The $J_{110}-H_{160}$ colors were determined by running SExtractor (Bertin & Arnouts, 1996) in two-image mode using the $H_{160}$ image for object detection and isophotal apertures. The $J_{110}$ image was PSF-matched to the $H_{160}$ band. The resulting colors and magnitudes are shown in Figure 1. For galaxies which are not detected at $2\sigma$ significance in the $J_{110}$-band (those to the right of the thick dashed line, representing $J_{110,tot}>26.7$, in Fig. 1) we consider the color to be a lower limit. We also measured similarly PSF-matched, isophotal colors using the two ACS bands and have used them to construct a $g_{475}-I_{814}$ versus $J_{110}-H_{160}$ color-color diagram (Figure 2). We compared these colors to those of model SEDs for different ages, star-formation histories and dust extinctions. Using the 2007 Charlot & Bruzual population synthesis models we have constructed spectral energy distributions for galaxies with an exponentially-decaying star-formation rate with time constants of $\tau=0.15,0.4,1000.0$ Gyr (the red, green and blue grids in Fig. 2 respectively). Each model’s colors are calculated for ages of 0.1, 1 and 3 Gyr and for $E(B-V)=0.0$ and $1.0$. Aging of the population moves primarily the $J_{110}-H_{160}$ color to the red while the dust extinction significantly reddens the $g_{475}-I_{814}$ color. From this analysis it appears that the $\tau=0.4$ Gyr model represents well the colors of a majority of the red $J_{110}-H_{160}$ galaxies. To extend the wavelength coverage for the protocluster galaxies we also incorporated ground-based $U_{n}$-band data from LRIS-B on the Keck telescope, $K_{s}$-band imaging from VLT/ISAAC and three band IRAC imaging (the 3.6, 4.5 and 5.8 $\mu$m bands) from the Spitzer Space Telescope. The Keck $U$-band data (PI W. van Breugel) were obtained in late January and early February of 2003. The ISAAC data (PI G. Miley) were taken in Period 73 in service mode. The Spitzer data are from the IRAC Guaranteed Time program (PI G. Fazio, Program #17). We have smoothed the imaging data for all bands, apart from the IRAC data, to match the resolution of the $U_{n}$-band image (approximated by a FWHM $\sim 1\arcsec$ Gaussian). We then used SExtractor to measure galaxy magnitudes within a $0\farcs 5$ radius circular aperture for each smoothed image. To incorporate the IRAC data, which has much poorer angular resolution, we derived aperture magnitudes which were then corrected to match the smoothed data. These aperture corrections were derived using the photometric curves-of- growth for 20 stars in the field. The resulting catalog was used to generate photometric redshift estimates as described below in Section § 3. Figure 3.— upper panel: Distribution of high-confidence ($>95\%$) photometric redshifts and their selection function assuming a uniform $N(z)$ for our model template galaxies (yellow curve) for $1.1\leq(J_{110}-H_{160})\leq 2.1$ galaxies in the MRC 1138-262 NICMOS field. The peak between $z=2.1$ and $z=2.4$ is statistically highly significant. middle panel: Sum of redshift probability distributions for all the galaxies in the upper panel. 38.5% of the total probability is contained in the redshift interval from 2 to 2.3. lower 2 panels: two examples of the probability distribution function for individual galaxies ## 3\. Photometric Redshifts We have used the ACS ($g_{475}$, $I_{814}$), NICMOS ($J_{110}$, $H_{160}$), ground-based $U_{n}$-band from Keck/LRIS-B, $K_{S}$-band imaging from VLT/ISAAC and Spitzer/IRAC imaging to estimate photometric redshifts for our $H_{160}$-band selected sample. We input a catalog of aperture galaxy magnitudes, based on the matched, smoothed images described above, into the Bayesian photometric redshift code (BPZ) of Benítez (2000) using a uniform prior. We felt that the default prior, based on optical galaxy selection and spectroscopy in the HDF-N, would not necessarily represent the redshift distribution for our near-infrared selected galaxies. We generated our own extensive set of template spectral energy distributions using the models of Charlot & Bruzual (2007). All these SEDs are $\tau$ models with values for $\tau=[0.15,0.4,1.0,2.0,1000.0]$ Gyr and ages $=[0.05,0.1,0.5,1.0,2.0,3.0]$ Gyr. We also included models with internal dust extinction ranging from $E(B-V)=[0.0,0.1,0.3,0.5,0.75,1.0]$ mag and metallicity of $(Z/Z_{\odot})=[0.3,1.0,2.5]$. We focused particular attention on the $J_{110}-H_{160}$ selected surface-overdensity. In the upper panel of Figure 3 we present the high confidence ($>95\%$) photo-$z$ distribution for the NIR- color selected ($1.1\leq(J_{110}-H_{160})\leq 2.1$) subsample. We ran extensive simulations by redshifting our template set, adding appropriate photometric errors and using BPZ to recover the redshifts. The yellow curve represents the redshift selection function for this color cut, template set and filters assuming that these model galaxies follow a uniform $N(z)$ over this redshift interval. The simulation results were free of significant systematic errors and the random errors are estimated to be $\delta z/z\sim 0.1$. Based on these SED fits, the approximate luminosity-weighted ages of the red galaxies lie between 1 and 2.5 Gyrs and their stellar masses are of order a few $\times 10^{10}M_{\odot}$. These stellar masses are reasonable as are the absolute magnitudes (see Figure 6). More detailed SED modeling is deferred to a future paper. There is a clear excess of galaxies between $z=2.0$ and $z=2.5$. For each galaxy fit by BPZ we have generated the full redshift probability distribution. In the lower panel of Fig. 3 we show the $H_{160}$-band weighted-average of these probability distributions. There is a clear peak (containing 38.5% of the total probability compared to only 17% of the total selection function in the same redshift interval) between $z=2.0$ and $z=2.3$, consistent with the significant peak in the redshift histogram itself. ## 4\. NICMOS Galaxy Morphologies Figure 4.— Distribution of inferred stellar ages (in terms of $\tau$) for both the concentrated ($n\geq 2.5$, red line) and diffuse ($n<2.5$, blue line) galaxies which are well-resolved in the NICMOS data. Constant star-formation models, for which the $e$-folding time is infinite, are placed at the left- hand edge of the plot. The blue and red distributions are quite different. Of particular note is that the most evolved galaxies generally have high $n$ while the low $n$ galaxies dominate the star-forming population. NICMOS camera 3 provides good angular resolution over its entire field-of- view. The FWHM of the PSF in our final mosaic is $\approx 0\farcs 27$. To exploit this resolution we have used the GALFIT code (Peng et al., 2002) to fit analytic Sérsic surface-brightness profiles (Sersic, 1968) to all the $H_{160}\leq 24.5$ sources in our $H_{160}$-band mosaic. A model point-spread function was created for each of these galaxies individually by generating a TinyTim simulated PSF (Krist, 1993) at the galaxies’ positions in each exposure and then drizzling these PSFs together in exactly the same fashion as for the data themselves (see Zirm et al. 2007). We restricted the Sérsic index, $n$, to be between 1 and 5. We will present a full analysis of the morphologies of these galaxies in a future paper. For the current work, we use these derived sizes and profile shapes to assist us in selecting the morphological “early-type” members of the red galaxy population. Figure 5.— Histogram of the color distributions for the 1138 and deep fields (blue). The deep field data has been normalized by total area to the 1138 data. Note the clear excess of red galaxies in the 1138 field. At $1.1\leq(J_{110}-H_{160})\leq 2.1$ (horizontal dotted lines) for galaxies brighter than the 2$\sigma$ $J_{110}$-band limit (dashed line) there is an overdensity of a factor of 6.2 in the 1138 field. In Figure 4 we show the distribution of galaxy ages derived via these SED fits as parametrized by the $\tau$ value for the best-fitting model for those galaxies with high and low Sérsic index ($n\geq 2.5$, red line, and $n<2.5$, blue line). It is clear that while there is substantial overlap between these distributions they are not identical and that they differ in the sense that one might expect, namely, that the concentrated galaxies appear to be comprised of older stellar populations. This trend gives us some confidence in trying to select the “early-type” galaxies using these data which is important for our discussion of the color-magnitude relation in Section 6. ## 5\. Surface Overdensity of Red Galaxies Figure 6.— Linear fits to the rest-frame $U-B$ (Vega) color-magnitude diagrams for three different sub-samples of the $H_{160}$-band selected NICMOS sample. Panel (A) shows the fit (solid line) and intrinsic scatter ($1\sigma$; dotted lines) for a sample selected to have $1.1\leq(J_{110}-H_{160})\leq 2.1$. The crossed-out point are those which are rejected as outliers in more than half of the realizations (see §6). Both the observed and intrinsic scatter are smaller than the initial color cut. Panel (B) shows the fit and intrinsic scatter for a photometric redshift selected sample with $2.0<z_{\rm phot}<2.5$. The stars indicate galaxies whose preferred photometric template has an age $<4\tau$, while circles represent galaxies with older than this limit. Panel (C) shows the fit and scatter for those galaxies which meet the same redshift cut but also are well-resolved with a high Sérsic index ($n>2.5$) and best-fit by an age $\geq 4\tau$ template. To compare this protocluster field to more generic ‘blank’ field data we have compiled catalogs for the public NICMOS data in both the Hubble Deep Field North (HDF-N) and the Ultra Deep Field (UDF). Figure 1 shows the $J_{110}-H_{160}$ color-magnitude diagram (open black circles) and the color distributions for both the MRC 1138-262 and the combined HDF-N and UDF galaxy catalogs (blue circles). The deep field data were also $H_{160}$-band selected. The area of the two deep fields is roughly 2.5 times the area of our protocluster observations. We have applied no correction to the deep field number counts to account for clustering in those fields. The color histogram in Figure 5 shows the area-normalized galaxy counts from the two deep fields (blue line) and from the 1138 field to the same ($2\sigma$) limiting magnitude of $J_{110}=26.7$ mag (AB). The red dashed line shows the difference between the two color distributions. It is clear that the radio galaxy field is overdense in red galaxies by a large factor. For sources with colors between $1.1\leq(J_{110}-H_{160})\leq 2.1$, the horizontal(vertical) dotted lines in Fig.1(5), and brighter than our $J_{110}$-band $2\sigma$ limit ($26.7$), we calculate an area-normalized overdensity of $6.2$ when compared to the deep fields data. We note that the exact value of the measured overdensity is rather sensitive to systematic color offsets between the protocluster and deep field data. A redward shift of 0.05 for the deep field galaxies would lower the measured overdensity to $5.0$. However, we are confident that these systematic offsets remain small ($<0.05$ mag) since we have used the same instrument, filters, selection technique and photometric code with very similar input parameters for both the deep field and 1138 datasets. Looking back at Figure 2 we can see that many of the spectroscopically-confirmed line emitters (filled blue circles) and red galaxies in the overdensity are well- represented by the $\tau=0.4$ Gyr model (green lines) at different ages and extinctions. This current work is not the first to observe red galaxies in this field. Kurk et al. (2004) identified a small ($\sim 1.5\times$) surface overdensity of extremely red objects (EROs; $I-K>4.3$ Vega magnitudes) using ground-based $I$ and $K$ band data. Many of these EROs are also identified as red in the NICMOS $J_{110}-H_{160}$ color. More recently Kodama et al. (2007) observed this field using the wide-field NIR imager, MOIRCS, on the Subaru telescope. These authors found several bright (presumably massive) red galaxies over a wider field-of-view but to shallower depths than the NICMOS data presented here. 24 of their color-selected protocluster candidates are within our NICMOS mosaic. 23 of the 24 are identified in our data as being red in $J_{110}-H_{160}$. Furthermore, 18 of the 94 galaxies which satisfy our color criteria (and have $J_{110}<26.67$) are also identified by Kodama et al. as protocluster candidates. The much larger number of red galaxies in the NICMOS data is primarily due to fainter galaxies detected at high significance in our deeper data. ## 6\. An Emergent Red-Sequence? To study the colors and magnitudes of these galaxies in more detail and to possibly identify a red-sequence in the 1138 field we have split the galaxies into three sub-samples defined by $J_{110}-H_{160}$ color, photometric redshift and morphology (Sérsic index). The first sample (Sample A) comprises all 56 galaxies with $1.1\leq(J_{110}-H_{160})\leq 2.1$ and $H_{160}<24.5$ and includes the radio galaxy itself. Sample B is made up of all 28 galaxies with a robust photometric redshift between 2.0 and 2.5 and $J_{110}-H_{160}>0.75$ and $H_{160}<26.0$. This liberal color cut is included to select galaxies which comprise the large observed surface-overdensity. Finally, sample C contains seven galaxies with the same photometric redshift cut but which also have well-resolved $H_{160}$-band surface-brightness profiles with Sérsic index $n>2.5$. All of these galaxies’ SEDs are also best-fit by models with relatively little on-going star-formation. We use a limit of (age $\geq 4\times\tau$, cf. Grazian et al 2007 ). Therefore, sample C mimics the color, morphological and photometric redshift selection of early-type galaxies in clusters at $z\mathrel{\hbox to0.0pt{\lower 3.0pt\hbox{$\mathchar 536\relax$}\hss}\raise 2.0pt\hbox{$\mathchar 316\relax$}}1$. The photometry, photo-$z$s and sizes of the sample C galaxies are listed in Table 1, their rest-frame color-magnitude distributions are shown in Figure 6 and the two- dimensional spatial distribution of the Sample A galaxies is plotted in Figure 7. We note that because the measured overdensity is a factor of 6, we statistically expect one of every seven sample A galaxies to be a field galaxy. However, this should not effect our results significantly. For these three sample selections we have fit a line and measured the intrinsic scatter about that best-fit line (see Fig. 6). For comparison to lower redshift galaxy clusters we have transformed our observed $J_{110}-H_{160}$ color and $H_{160}$ magnitudes into rest-frame $U-B$ and $B$ (Vega), respectively, using the following expressions: $(U-B)_{\rm rest}=0.539\times(J_{110}-H_{160})_{\rm obs}-0.653$ (1) $M_{B,{\rm rest}}=H_{160,{\rm obs}}-0.170\times(J_{110}-H_{160})_{\rm obs}-43.625$ (2) The small color corrections used in these relations were derived using a family of $\tau$-models with a range of ages (0.1-12 Gyr), $\tau$ (0.1-5 Gyr) and three metallicities ($0.4,1$ and $2.5Z_{\odot}$). Figure 7.— Spatial distribution of galaxies (red circles) relative to the radio galaxy MRC 1138-262 (yellow star) for the color-selected sample defined in the text (A) and shown in the first panel of Fig. 6. The irregular black outline encloses the coverage of the 7 NICMOS camera 3 pointing with both $J_{110}$ and $H_{160}$ imaging. To fit the “CMR” we used a bootstrap re-sampling technique to estimate the error on the fitted slope. Then, by assuming that all the red galaxies lie on this fit line, we ran Monte Carlo realizations of the contribution of the photometric errors to the observed color scatter about the fit line, i.e., by fixing a color-magnitude relation we calculate the measurement scatter with zero intrinsic scatter. We then calculate the intrinsic scatter by subtracting (in quadrature) the estimated measurement scatter from the observed scatter. We show these fits (solid line) and the intrinsic scatters (dotted lines) for the three samples (A, B and C) in Figure 6. The fits to both Sample’s A and B have nearly identical rest-frame $U-B$ slopes, $0.027$ and $0.026$ respectively, and intrinsic scatters ($0.10$ and $0.12$). While these slopes are comparable to those found for the well-populated lower redshift cluster CMRs, the intrinsic scatters are considerably higher. However, the scatter measured for the eight galaxy Sample C is comparable to that of the lower redshift samples but with a much steeper slope ($0.130$). When these scatters are compared to model predictions based on lower redshift clusters (specifically RDCS 1252.9-2927 at $z=1.24$; Gobat et al. 2008) we find that the 1138 protocluster has lower than predicted scatter. This may suggest that the 1138 protocluster is in a more advanced evolutionary state than RDCS 1252 was at $z=2.2$. We have calculated three representative color-magnitude relations for comparison to the colors and magnitudes of the red galaxies (three dot-dash lines in Fig. 1). We have taken two lower redshift clusters, Coma at $z=0.023$ and RDCS 1252.9-2927 at $z=1.24$, and transformed them to the observed filters and $z=2.16$ under the assumption that the colors do not evolve. In this no evolution case (the two dot-dash lines in Fig. 1) the CMRs appear at the red edge of the observed overdensity. There is almost exactly 2 Gyr of cosmic time between $z=2.16$ and $z=1.24$ in our adopted cosmology. From Blakeslee et al. (2003) we know that the median redshift of last significant star-formation for the RDCS 1252 galaxies is between $z=2.7-3.6$. Therefore, if we observe those galaxies at $z=2.16$ they will be significantly younger and hence bluer. In fact, this passively de-evolved line (bluest dot-dash line in Fig. 1, labeled ‘$z_{\rm form}\sim 3$’) does fall blueward of the red galaxy overdensity. We discuss the implications for these comparisons in Section 7. We have also translated the Kodama et al. ground-based $J-K$ colors to our NICMOS filters assuming all the red galaxies lie at $z=2.16$. These bright galaxies also fall along the passively de-evolved line with the radio galaxy. We have used our suite of SED models to estimate the color transformation from their ground-based $J-K_{S}$ to our NICMOS $J_{110}-H_{160}$ color. Roughly, the Kodama et al. bright red galaxies fall where the RDCS 1252 passive line crosses our color cut at $J_{110}-H_{160}=1.1$. This result hints at a possible bi-modality in the red galaxy population of this protocluster. Namely, that there are faint red galaxies that are inconsistent with passively-evolving cluster members either due to large amounts of dust, or due to higher redshifts of formation but that the more luminous protocluster members may have already finished forming and seem consistent with passive evolution to the present-day. ## 7\. Discussion We have identified a (6.2$\times$) surface-overdensity and a corresponding photometric redshift ‘spike’ of red $J_{110}-H_{160}$ galaxies which are likely associated with a known protocluster at $z=2.16$. The optical-NIR spectral energy distributions of these sources suggest that they comprise both evolved galaxies as well as dust-obscured star-forming galaxies. Based on our SED fits from the photometric redshift determinations, the approximate luminosity-weighted ages of these sources lie between 1 and 2.5 Gyrs and their stellar masses are of order a few $\times 10^{10}M_{\odot}$. Detailed modeling of the SEDs for the protocluster population, along with their morphologies, is reserved for a future paper. Comparison with the CMRs of lower redshift clusters shows that the red galaxy overdensity primarily lies blueward of the no-evolution predictions. That the red galaxies in 1138 are also redder than the $z_{\rm form}\sim 3$ case suggests both that there are galaxies with significant dust content, an assertion supported by the SED fits, and also that they were perhaps formed at higher redshift than the RDCS1252 galaxies. Of course, without a classical, low-scatter red-sequence to use as a baseline there remains considerable uncertainty in the age of the population as a whole. The results of Steidel et al. (2005) suggest that protocluster galaxies are older than their “field” counterparts at $z\sim 2.3$ and that these ages and stellar masses were broadly consistent with evolution to lower redshift cluster galaxies. However, their protocluster members were all UV-selected and star-forming. With future spectroscopy of our red galaxy sample it will be possible to see if these differences persist when looking at a more varied galaxy sample. For three samples of galaxies drawn from the full $H_{160}$-band selected dataset we have fit a color-magnitude relation and estimated the intrinsic scatter. The CMR at $z=2.16$ is not as well-defined as at $z\sim 1$ or 0. For sample C, made up of 8 galaxies, the color, best-fit spectral template, morphology and photo-$z$ all point towards them being (proto-)elliptical galaxies within the protocluster. For this small sample, the estimated intrinsic scatter is rather low and may suggest that these galaxies represent the forming red-sequence in this protocluster. The slope of this relation is extremely steep compared to lower redshift clusters. The slope of the CMR is generally assumed to be a manifestation of the mass-metallicity relation and would therefore flatten at higher redshift. The major caveat regarding the steep slope of Sample C is that none of these galaxies are spectroscopically confirmed protocluster members. Therefore, this “relation” may just be a random, although somewhat unlikely, coincidence rather than a nascent CMR. However, further deep NIR imaging coverage of this field is required to identify additional members of this proto-elliptical galaxy class. Support for program # 10404 was provided by NASA through a grant (GO-10404.01-A) from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555. The work of SAS was performed in part under the auspices of the U.S. Department of Energy, National Nuclear Security Administration by the University of California, Lawrence Livermore National Laboratory under contract No. W-7405-Eng-48. JK is financially supported by the DFG, grant SFB 439\. WvB acknowledges support for radio galaxy studies at UC Merced, including the work reported here, with the Hubble Space Telescope and the Spitzer Space Telescope via NASA grants HST # 10127, SST # 1264353, SST # 1265551, SST # 1279182. ## References * Bell et al. (2004) Bell, E. F., et al. 2004, ApJ, 608, 752 * Benítez (2000) Benítez, N. 2000, ApJ, 536, 571 * Bertin & Arnouts (1996) Bertin, E. & Arnouts, S. 1996, A&AS, 117, 393 * Blakeslee et al. (2003) Blakeslee, J. P., et al. 2003, ApJ, 596, L143 * Blakeslee et al. (2006) Blakeslee, J. P., et al. 2006, ApJ, 644, 30 * Bruzual & Charlot (2003) Bruzual, G. & Charlot, S. 2003, MNRAS, 344, 1000 * Charlot & Bruzual (2007) Charlot, S. & Bruzual, G. 2007, in prep * Croft et al. (2005) Croft, S., Kurk, J., van Breugel, W., Stanford, S. A., de Vries, W., Pentericci, L., & Röttgering, H. 2005, AJ, 130, 867 * de Jong (2006) de Jong, R. 2006. Correcting the nicmos count-rate dependent non-linearity. Technical report, STScI * Faber et al. (2007) Faber, S. M., et al. 2007, ApJ, 665, 265 * Gladders & Yee (2005) Gladders, M. D. & Yee, H. K. C. 2005, ApJS, 157, 1 * Gobat et al. (2008) Gobat et al., 2008, submitted to A&A * Grazian et al. (2007) Grazian, A., et al. 2007, A&A, 465, 393 * Kodama et al. (2007) Kodama, T., Tanaka, I., Kajisawa, M., Kurk, J., Venemans, B., De Breuck, C., Vernet, J., & Lidman, C. 2007, MNRAS, 377, 1717 * Krist (1993) Krist, J. 1993, in ASP Conf. Ser. 52: Astronomical Data Analysis Software and Systems II 536 * Kurk (2003) Kurk, J. D. 2003, May. The cluster environments and gaseous halos of distant radio galaxies. Ph. D. thesis, Leiden University * Kurk et al. (2004a) Kurk, J. D., Pentericci, L., Röttgering, H. J. A., & Miley, G. K. 2004a, A&A, 428, 793 * Kurk et al. (2004b) Kurk, J. D., Pentericci, L., Overzier, R. A., Röttgering, H. J. A., & Miley, G. K. 2004b, A&A, 428, 817 * Miley et al. (2006) Miley, G. K., et al. 2006, ApJ, 650, L29 * Mullis et al. (2005) Mullis, C. R., Rosati, P., Lamer, G., Böhringer, H., Schwope, A., Schuecker, P., & Fassbender, R. 2005, ApJ, 623, 85 * Oke (1974) Oke, J. B. 1974, ApJS, 27, 21 * Peng et al. (2002) Peng, C. Y., Ho, L. C., Impey, C. D., & Rix, H.-W. 2002, AJ, 124, 266 * Pentericci et al. (2002) Pentericci, L., Kurk, J. D., Carilli, C. L., Harris, D. E., Miley, G. K., & Röttgering, H. J. A. 2002, A&A, 396, 109 * Sersic (1968) Sersic, J. L. 1968, Atlas de galaxias australes Cordoba, Argentina: Observatorio Astronomico, 1968 * Stanford et al. (2005) Stanford, S. A., et al. 2005, ApJ, 634, 129 * Stanford et al. (2006) Stanford, S. A., et al. 2006, ApJ, 646, 13 * Steidel et al. (2005) Steidel, C. C., Adelberger, K. L., Shapley, A. E., Erb, D. K., Reddy, N. A., & Pettini, M. 2005, ApJ, 626, 44 * Stevens et al. (2003) Stevens, J. A., et al. 2003, Nature, 425, 264 * Tremonti et al. (2004) Tremonti, C. A., et al. 2004, ApJ, 613, 898 * van Dokkum & van der Marel (2007) van Dokkum, P. G. & van der Marel, R. P. 2007, ApJ, 655, 30 * Venemans et al. (2007) Venemans, B. P., et al. 2007, A&A, 461, 823 * Zirm et al. (2007) Zirm, A. W., et al. 2007, ApJ, 656, 66 Table 1Properties of Red Galaxies in the MRC 1138-262 Field: $H_{160}<24.5\wedge 1.1\leq(J_{110}-H_{160})\leq 2.1$ ID | $z_{\rm phot}$aaThe errors on these photometric redshifts are approximately 0.1-0.2 | OddsbbThe ODDS parameter approaches unity when the probability distribution function has a single narrow peak | SED TypeccBest-fit broad-band template type: 0: dusty and star-forming 1: quiescent | Sérsic | $r_{e}$ | $r_{e}$ | $H_{160}$ddTotal magnitudes based on the SExtractor MAG_AUTO parameter | $U_{n}-g_{475}$ | $g_{475}-I_{814}$ | $I_{814}-J_{110}$ | $J_{110}-H_{160}$ ---|---|---|---|---|---|---|---|---|---|---|--- | | (BPZ) | | Index | ($\arcsec$) | (kpc) | (AB) | (AB) | (AB) | (AB) | (AB) | | | | ($n$) | | | | | | | 606 | 2.04 | 0.98 | 1.00 | 3.4 $\pm$ 0.4 | 0.21 $\pm$ 0.01 | 1.79 $\pm$ 0.10 | 23.09 $\pm$ 0.03 | 1.10 | 0.79 | 1.18 | 1.06 507 | 2.09 | 1.00 | 1.00 | 5.0 $\pm$ 0.4 | 0.24 $\pm$ 0.01 | 1.99 $\pm$ 0.08 | 22.50 $\pm$ 0.01 | 28.88 | -27.16 | 2.22 | 1.52 314 | 2.10 | 1.00 | 1.00 | 5.0 $\pm$ 0.7 | 0.15 $\pm$ 0.01 | 1.28 $\pm$ 0.05 | 23.03 $\pm$ 0.01 | 1.92 | 1.38 | 1.89 | 1.29 586 | 2.10 | 1.00 | 1.00 | 5.0 $\pm$ 0.3 | 0.40 $\pm$ 0.03 | 3.34 $\pm$ 0.23 | 21.77 $\pm$ 0.01 | 1.80 | 1.89 | 2.22 | 1.61 547 | 2.23 | 1.00 | 1.00 | 5.0 $\pm$ 0.6 | 0.12 $\pm$ 0.00 | 1.02 $\pm$ 0.04 | 22.70 $\pm$ 0.01 | 1.30 | 1.13 | 1.67 | 1.36 493 | 2.25 | 1.00 | 1.00 | 3.5 $\pm$ 0.3 | 0.56 $\pm$ 0.04 | 4.72 $\pm$ 0.36 | 22.75 $\pm$ 0.02 | 0.63 | 1.28 | 2.19 | 1.64 312 | 2.42 | 1.00 | 1.00 | 5.0 $\pm$ 0.3 | 0.38 $\pm$ 0.02 | 3.20 $\pm$ 0.19 | 22.00 $\pm$ 0.01 | -28.02 | 1.45 | 1.94 | 1.55 127 | 2.44 | 1.00 | 1.00 | 2.7 $\pm$ 0.6 | 0.12 $\pm$ 0.02 | 1.01 $\pm$ 0.18 | 22.64 $\pm$ 0.01 | 0.34 | 1.04 | 1.37 | 1.40

arxiv-papers

2024-09-04T02:48:53.762505

0802.2164

Could dark matter or neutrinos discriminate between the enantiomers of a chiral molecule? Pedro Bargueño Instituto de Matemáticas y Física Fundamental, CSIC, Madrid, Spain Departamento de Química Física, Universidad de Salamanca, Spain Antonio Dobado and Isabel Gonzalo Departamento de Física Teórica I and Departamento de Óptica Universidad Complutense de Madrid, 28040 Madrid, Spain ABSTRACT We examine the effect of cold dark matter on the discrimination between the two enantiomers of a chiral molecule. We estimate the energy difference between the two enantiomers due to the interaction between fermionic WIMPs (weak interacting massive particles) and molecular electrons on the basis that electrons have opposite helicities in opposite enantiomers. It is found that this energy difference is completely negligible. Dark matter could then be discarded as an inductor of chiroselection between enantiomers and then of biological homochirality. However, the effect of cosmological neutrinos, revisited with the currently accepted neutrino density, would reach, in the most favorable case, an upper bound of the same order of magnitude as the energy difference obtained from the well known electroweak electron-nucleus interaction in some molecules. PACS: 33.15.Bh, 13.15.+g, 95.35.+d ## 1 Introduction The origin of biological homochirality, that is, the almost exclusive one- handedness of chiral molecules in biological organisms, is a fundamental problem for which there is not yet a convincing solution. Several mechanisms have been proposed to explain chiroselection among the two possible enantiomers of a chiral molecule (see for example [1, 2, 3] and references therein). These mechanisms involve chance, $\beta$-radiolysis [4], circularly polarized light [5, 6, 7], magnetic fields [8, 9], and violation of parity in the weak interaction (see below). The discovery of an excess of L-amino acids in meteorites [10, 11] has reinforced the idea of an extraterrestrial origin of biological homochirality [1, 12]. In this context, universal mechanisms of chiroselection such as parity violation in weak interactions would acquire special interest in spite their tiny effects, without of course underestimating other mechanisms. The effect of electroweak interactions between electrons and nuclei mediated by the Z0, have been extensively studied and observed in atoms (see the review [13]), and only predicted in molecules, where an energy difference between the two enantiomers of chiral molecules has been estimated to be between $10^{-16}$ and $10^{-21}$ eV [14, 15, 16, 17, 18]. In the laboratory, no conclusive energy difference have been reported in experimental spectroscopic studies reaching an energy resolution of about $10^{-15}$ eV [19]. The above tiny energy difference would require a powerful mechanism of amplification in order to induce a real enantioselective effect. Otherwise the small energy difference would be masked by the natural broadening of the energy levels of the molecule, thermal fluctuations and environment interactions, which do not discriminate, in average, between L and D enantiomers. There is active research on amplification mechanisms in which a permanent although very small interaction acting always in the same enantioselective direction, and under appropriate conditions, could lead to an effective enantioselection. Some mechanisms are based on nonlinear autocatalytic processes of polymerization or crystallization along a large period of time [20, 21]. Another one involves a second-order phase transition below a certain critical temperature [22] that could work at low temperatures such as those of the interstellar space. However, theoretical or experimental conclusive results from the diverse mechanisms to amplify enantioselection based on electroweak energy difference, are not yet at hand (e.g.,[23, 24]). Another universal mechanism that could discriminate between the two enantiomers of a chiral molecule and that involves weak interaction is neutrino-electron axial-vector interaction mediated by charged electroweak bosons W±. This process discriminates on the basis of an asymmetry between the number of neutrinos and antineutrinos, and that electrons of opposite (L,R) enantiomers have opposite helicity. Considering the cosmological relic sea of neutrinos, the estimated energy difference between the two enantiomers was found to be even lower than the value $10^{-21}$ eV mentioned above [25]. However, it could increase significantly under bigger neutrino fluxes, as in supernova remnants [26] assumed that big molecules could survive in the surroundings. Revisited assumptions about the number density of cosmological neutrinos in the relic sea lead to an increase of the mentioned energy difference, as we shall see in the next section. Looking for other universal mechanisms acting also outside of the Earth, we analyze here the possible enantioselective effect of chiral dark matter on chiral molecules. The existence of dark matter is inferred from astrophysical observations in light of studies of the dynamics of stars in the local disk environment, rotation curves for a large number of spiral galaxies, gravitational lensing by clusters of galaxies and some large scale studies of the Universe (for a recent review of experimental searches for dark matter see for example [27]). A vast variety of candidates have been proposed for dark matter content, from baryonic to non-baryonic matter. The non-baryonic candidates are basically postulated elementary particles beyond the Standard Model which have not been discovered yet, like axions, WIMPs (Weak Interacting Massive Particles) and other exotic candidates. The baryonic candidates are the Massive Compact Halo Objects (Macho) [28]. Another important difference is the hot versus cold dark matter. A dark matter candidate is called hot if it was moving at relativistic speeds at the time when galaxies could just start to form, and cold if it was moving non-relativistically at that time. The problem is that hot dark matter cannot reproduce correctly the observed structure of the Universe. Therefore we focus our attention on cold dark matter. The fact that dark matter interacts weakly with matter makes its detection very difficult [27]. However many experiments are currently in progress in order to reach this goal. Here we estimate the energy difference between the two enantiomers of a chiral molecule, due to the weak type interaction between non-baryonic cold dark matter (specifically WIMPs) and molecular electrons with non zero helicity. Experimental results on dark matter are used. Given the resemblance in the procedure with the energy difference induced by cosmological neutrinos, estimated in a previous work [25], we first recall this procedure at the time we improve the result we obtained in that work. ## 2 Energy difference between opposite enantiomers induced by cosmological neutrinos Following a previous work [25], we consider neutrino-electron interactions mediated by the axial-vector Hamiltonian density $H=\frac{G_{F}}{\sqrt{2}}\bar{e}\gamma^{\mu}(g_{V}-g_{A}\gamma_{5})e\bar{\nu}\gamma_{\mu}(1-\gamma_{5})\nu,$ (1) where $G_{F}$ is Fermi’s constant, $e(\nu)$ denotes the electron (neutrino) spinor field, $\bar{e}(\bar{\nu})$ is its adjoint spinor, $\gamma^{\mu}$ are the Dirac matrices (regarded as a four-vector), $\gamma^{5}=i\gamma^{0}\gamma^{1}\gamma^{2}\gamma^{3}$, and $g_{V,A}$ are suitable coupling constants that parameterize the strength of the interactions. As it was discussed for example in [29], in the non-relativistic limit it is possible to make the following approximations for the dominant temporal components of the four-vectors appearing in the above Hamiltonian, $\displaystyle\bar{e}\gamma^{\mu}\gamma_{5}e$ $\displaystyle\sim$ $\displaystyle\vec{\sigma}_{e}\cdot\vec{v}_{e},\qquad\ $ $\displaystyle\bar{\nu}\gamma_{\mu}\nu$ $\displaystyle\sim$ $\displaystyle n_{\nu}-n_{\bar{\nu}}\qquad(\textrm{Dirac neutrinos}),$ $\displaystyle\bar{\nu}\gamma_{\mu}\gamma_{5}\nu$ $\displaystyle\sim$ $\displaystyle n_{\nu_{l}}-n_{\nu_{r}}\qquad(\textrm{Majorana neutrinos}),$ (2) where the number density differences, $n_{\nu}-n_{\bar{\nu}}$ and $n_{\nu_{l}}-n_{\nu_{r}}$ refer to neutrino-antineutrino and left-right helicity eigenstates respectively. Obviously they are not zero only in the case where there is a net lepton number or helicity in the cosmic neutrinos background. We recall from [25] and [30] that for Dirac neutrinos the energy splitting obtained for the electron is $\Delta E\sim G_{F}|(n_{\nu}-n_{\bar{\nu}})\langle\vec{\sigma}_{e}\cdot\vec{v}_{e}\rangle|,$ (3) where the expected value of the electron helicity $\langle\vec{\sigma}_{e}\cdot\vec{v}_{e}\rangle$ takes opposite signs for the two opposite enantiomers, as we can see from a simplified chiral molecule model [31]. In this model, a dominant axial symmetry around axis Z, with a left(right)-handed perturbative potential of period a, is assumed, so that the electronic molecular states can be described by superposition of eigenstates of both angular momentum $L_{z}$ and linear momentum $P_{z}$ (eigenvalues $\hbar n(2\pi/a)$), i.e. $|M_{L},n\rangle$, in the form $\displaystyle\Phi_{L}$ $\displaystyle=$ $\displaystyle c_{0}|0,0\rangle+c_{1}|+1,-1\rangle+c_{2}|-1,+1\rangle\,,$ (4) $\displaystyle\Phi_{R}$ $\displaystyle=$ $\displaystyle c_{0}|0,0\rangle+c_{1}|+1,+1\rangle+c_{2}|-1,-1\rangle\,,$ (5) with $|c_{1}|^{2}=|c_{2}|^{2}\equiv C$. These L and R states have then opposite helicities: $\langle\Phi_{L}|L_{z}P_{z}|\Phi_{L}\rangle=-4C\pi/a=-\langle\Phi_{R}|L_{z}P_{z}|\Phi_{R}\rangle\,.$ (6) Notice that we are using all the time natural units where $\hbar=c=1$. The spin of the electron can be taken into consideration by replacing $L_{z}$ with $J_{z}=L_{z}+S_{z}$. In a realistic chiral molecule the electronic states would not be eigenstates of the helicity, but its mean value would have opposite sign for L and R enantiomers. The parameter $C<1/2$ accounts for the degree of chirality. We note that the velocity of the molecule carrier (interstellar grains, meteorites, the Earth…) does not contribute to the helicity of the electrons: If $\vec{P}_{T}$ is the translational momentum of the carrier, the electronic wave function $\Phi_{L(R)}$ must include the factor $e^{i\vec{P}_{T}\cdot\vec{R}}$ (here $\vec{R}$ is the position of the molecule), and the contribution of $\vec{P}_{T}$ to the electron helicity is then $\displaystyle\langle\Phi_{L(R)}e^{i\vec{P}_{T}\cdot\vec{R}}|\vec{L}\cdot\vec{P}_{T}|\Phi_{L(R)}e^{i\vec{P}_{T}\cdot\vec{R}}\rangle$ $\displaystyle=\langle\Phi_{L(R)}|\vec{L}|\Phi_{L(R)}\rangle\cdot\vec{P}_{T}=0\,,$ (7) since $\langle\Phi_{L}|\vec{L}|\Phi_{L}\rangle=\langle\Phi_{R}|\vec{L}|\Phi_{R}\rangle=0$ as can be seen from Eqs. (4,5). We also remark that the particle flux is assumed to be isotropic, thus, its interaction with the electrons of a chiral molecule is the same irrespective of the orientation of the molecule. The energy difference that we obtained, assuming complete neutrino- antineutrino asymmetry, with number density of about $10^{-2}\mathrm{cm}^{-3}$, $C=1/2$, $a\sim 1$ Angstrom and the electron helicity given by Eq. (6), was of the order of $10^{-26}$ eV [25]. However, it has been recently suggested [29, 32] that, in scenarios beyond the standard model, the neutrino-antineutrino density asymmetry $n_{\nu}-n_{\bar{\nu}}$ could be up to the order of $\sim 10-1050\ \mathrm{cm}^{-3}$. Although the extreme upper bound density asymmetry seems to be excluded by considerations of primordial nucleosynthesis [33], we consider it to estimate an upper bound of the energy difference. If we take the value $n_{\nu}-n_{\bar{\nu}}\sim 1000\ \mathrm{cm}^{-3}$, we then obtain an upper bound for the energy difference between enantiomers of the order of $10^{-21}$ eV, per molecular electron with non zero helicity. Evidently this tiny energy needs massive amplification mechanisms as those mentioned at the beginning in order to induce an effective enantioselection. ## 3 Energy difference between opposite enantiomers induced by fermionic cold dark matter In a similar way to the neutrino-electron interaction above considered, we are now to estimate the electron energy splitting induced by the axial-vector interaction between a fermionic dark matter candidate (typically a WIMP) and an electron. The relevant Hamiltonian density can be written as $H=\sum_{i}d_{i}\bar{\chi}\gamma_{\mu}(1-\gamma_{5})\chi\bar{\psi}_{i}\gamma^{\mu}\gamma_{5}\psi_{i},$ (8) where $\chi$ is the dark matter spinor which can be Dirac or Majorana. The index $i$ runs through $i=e,u,d,s$, i.e., we are considering also the interaction between the dark matter particle and the $u,d$ and $s$ quarks. This will be important later in order to use the present experiments trying to measure the WIMP flux on Earth to set some bounds on the possible effect of dark matter on opposite enantiomers. Therefore $d_{e},d_{u},d_{d}$ and $d_{s}$ are the coupling of the $\chi$ field to the different matter fields $e=\psi_{e},u=\psi_{u},d=\psi_{d}$ and $s=\psi_{s}$. As WIMPs are typical examples of cold dark matter and heavy by definition, we can invoke again the non-relativistic limit. Thus, as it was the case of neutrinos, for Dirac WIMPs the term $\bar{\chi}\gamma_{\mu}\chi$ dominates with the temporal component of this vector being proportional to $n_{\chi}-n_{\bar{\chi}}$. For the Majorana case only the axial vector $\bar{\chi}\gamma_{\mu}\gamma_{5}\chi$ remains and its temporal component becomes proportional to $n_{\chi_{l}}-n_{\chi_{r}}$, as in Eq. (2). The expression for the corresponding electron energy splitting is similar to that of Eq. (3), $\Delta E\sim d_{e}|\Delta n\langle\vec{\sigma}_{e}\cdot\vec{v}_{e}\rangle|,$ (9) where $\Delta n$ is the appropriate number density difference corresponding to the Dirac or the Majorana case. In principle these differences depend on the unknown nature of the dark matter and its evolution along the universe history. In the following we will write these differences as $|\Delta n|=\alpha n$ where $n$ is the total WIMP number density. Clearly the parameter $\alpha$ is a measure of the degree of particle-antiparticle or left-right asymmetry present in the the dark matter respectively. For example, in the case of Dirac dark matter, $\alpha=1$ indicates that all WIMPs are particles with no antiparticles present and $\alpha=0$ means a complete particle- antiparticle symmetry. As in the neutrino case, interactions between molecular electrons with non zero helicity and cold dark matter could lead to an energy difference between the two enantiomers of a chiral molecule whenever the parameter $\alpha$ is different from zero. To have an estimation of the energy difference, we consider the interaction between WIMPs and an electron of a chiral molecule. Let $\rho=nM_{\chi}$ be the energy density of those WIMPs, with $M_{\chi}$ being their mass and $n$ their number density. The density of WIMPs trapped in the gravitational potential wall of the galaxy is expected to be of the order of $\rho\sim$ 0.3 GeV $\mathrm{cm}^{-3}$. Then the energy splitting can be written as $\Delta E\sim d_{e}\alpha\ \frac{\rho}{M_{\chi}}|\langle\vec{\sigma}_{e}\cdot\vec{v}_{e}\rangle|\,.$ (10) In order to see how important this splitting could be, we need to know which values of the coupling constant $d_{e}$ are acceptable. In principle there is no any available experimental information about $d_{e}$. However one reasonable assumption that could be done is that all the $d_{i}$ couplings are at least of the same order of magnitude. In the absence of a theory of WIMPs this seems to be not so bad assumption since WIMPs does not interact strongly with matter. If this is the case, one can then use the present bounds on the elastic cross-section proton$-\chi$ to get some information about the size of the $d_{i}$ couplings. In order to compute this cross section, one needs to relate the quark$-\chi$ couplings with the proton$-\chi$ coupling. This can be done by using the effective Hamiltonian (see [34] and references therein) $H=-a_{p}2\sqrt{2}\bar{\chi}\gamma_{\mu}\gamma_{5}\chi\bar{p}s^{\mu}p\,,$ (11) where $p$ is the proton spinor and $s^{\mu}$ is its spin vector (here we are considering the Majorana case but the Dirac case can be treated in a similar way). The coupling $a_{p}$ is defined as $a_{p}=\frac{1}{\sqrt{2}}\sum_{i=u,d,s}d_{i}\Delta q_{i}^{(p)}\,.$ (12) The constants $\Delta q_{i}^{(p)}$ (with $q_{1}=u,q_{2}=d$ and $q_{3}=s$) are introduced trough the proton matrix element $\langle p|\bar{\psi}_{i}\gamma^{\mu}\gamma_{5}\psi_{i}|p\rangle=2s^{\mu}\Delta q^{(p)}.$ (13) Experimentally we have $\Delta u^{(p)}\simeq 0.78$, $\Delta d^{(p)}\simeq-0.5$ and $\Delta s^{(p)}\simeq-0.16$. Then, by using standard methods, it is straightforward to compute the elastic proton$-\chi$ cross-section, which is given in the proton rest frame by $\frac{d\sigma}{dq^{2}}=\frac{\sigma_{n}}{4v_{\chi}\mu^{2}}\,,$ (14) where $\vec{q}$ is the momentum transfer, $v_{\chi}$ is the $\chi$ velocity, $\mu$ is the proton $-\chi$ reduced mass and $\sigma_{n}=\frac{12a_{p}^{2}m_{p}^{2}M_{\chi}^{2}}{\pi(m_{p}+M_{\chi})^{2}}$ (15) ($m_{p}$ being the proton mass) is just the non-relativistic cross section for vanishing momentum transfer. Now days there are many experiments around the world trying to detect WIMPs directly (visit the webpage [35] for complete and upgraded report of their main results). Usually they set exclusion regions on the plane $\sigma_{n}-M_{\chi}$. From the recent XENON10 2007 [35] we learn for example that, for $M_{\chi}\simeq 100$ GeV, $\sigma_{n}$ must be lesser than $10^{-43}$ cm2 and, for $M_{\chi}\simeq 1000$ GeV, lesser than $10^{-42}$ cm2. Assuming for simplicity all the quark couplings to be the same, i.e. $d_{q}\simeq d_{u}\simeq d_{d}\simeq d_{s}$, we have $a_{p}\simeq 0.0072d_{q}^{2}$. Then we get that for $M_{\chi}\simeq 100$ GeV, $d_{q}^{2}<10^{-14}$ GeV-4 and for $M_{\chi}\simeq 1000$ GeV, $d_{q}^{2}<10^{-13}$ GeV-4. As discussed above we now assume $d_{e}\sim d_{q}$. Then it is possible to set a bound on the energy splitting which turn to be very tiny even in best case corresponding to $M_{\chi}\simeq 100$ GeV. We obtain in this case, with an electron velocity about $10^{-2}$, $\Delta E\leq\alpha 10^{-44}$ eV. ## 4 Conclusion We have analyzed the effect of cold dark matter on the discrimination between the two enantiomers of a chiral molecule whose external electrons have opposite helicities in the respective opposite enantiomers. The estimated energy difference between the two enantiomers, due to WIMP-electron interaction, is found to be extremely small, several orders of magnitude lower than that induced by electron-nuclei weak interaction. Hence, dark matter would be discarded as inductor of chiroselection between enantiomers and then of biological homochirality. By contrast, the enantioselective effect of the cosmological relic sea of neutrinos acquires relevance with the current assumptions about the number density of cosmological neutrinos. In this case we obtain an energy difference between $10^{-23}$ and $10^{-21}$ eV for the two opposite enantiomers per molecular electron with non zero helicity. The upper bound of the energy difference, although could be excluded by reasons previously mentioned, reaches the same order of magnitude as the energy difference induced by the well known electron-nucleus electroweak interaction in some molecules. acknowledgments This work is supported by DGICYT (Spain) project BPA2005-02327, by the Universidad Complutense/CAM projects 910309 and CCG06-UCM/ESP-137, and by the MEC (Spain) projects CTQ2005-09185-C02-02 and FIS2004-03267. The work of P. Bargueño was supported by the FPI grant BES-2006-11976 from the Spanish MEC. The authors would like to thank A. L. Maroto and R. Pérez de Tudela for useful discussions. ## References * [1] Bonner, W.A., Orig. Life Evol. Biosph. 21, 59 (1991) . * [2] A. Jorissen and C. Cerf, Orig. Life Evol. Biosph., 32, 129 (2002). * [3] D. B. CLine (Ed.), Physical Origin of Homochirality in Life, AIP press, (1996) * [4] R.A. Hegstrom, A. Rich and J. Van House, Nature, 313, 391 (1985). * [5] J. Bailey, A. Chrysostomou, J.H. Hough et al., Science, 281, 672 (1998) * [6] J. Bailey, Orig. Life Evol. Biosph., 31, 167 (2001). * [7] M. Buscherm hle, D.C.B. Whittet, A. Chrysostomou et al., Astroph. Journal, 624, 821 (2005). * [8] G.L.J.A. Rikken and E. Raupach, Nature, 405, 932 (2000). * [9] T. Ruchon, M. Vallet, D. Chauvat et al., J. Chem. Phys., 125, 084104 (2006). * [10] M.H. Engel and S.A. Macko, Nature, 389, 265 (1997). * [11] S. Pizzarello and J.R. Cronin, Geochim. Cosmochim. Acta, 64, 329 (2000). * [12] W.A. Bonner, Orig. Life Evol. Biosph., 22, 407 (1992). * [13] A.M. Bouchiat and C.C. Bouchiat, Rep. Prog. Phys., 60, 1351 (1997). * [14] V. S. Letokhov, Phys. Lett., 53A, 275 (1975). * [15] D. W. Rein, R.A. Hegstrom and P.G.H. Sandars, J. Chem. Phys., 73, 2329 (1980). * [16] R. Zanasi, A. Lazzeretti and A. Soncini, Phys. Rev. E, 59, 3382 (1999) * [17] J.K. Laerdahl and P. Schwerdtfeger, Phys. Rev. Lett, 84, 3811 (2000). * [18] P. Soulard et al., Phys. Chem. Chem. Phys., 8, 79 (2006). * [19] J. Crassous et al., Org. Biomol. Chem., 3, 2218 (2005). * [20] D.K. Kondepudi and G.M. Nelson, Nature, 314, 438 (1985). * [21] R. Plasson, D. K. Kondepudi, H. Bersini, A. Commeyras and K. Asakura, Chirality, 19, 589 (2007). * [22] A. Salam, J. Mol. Evol., 33,105 (1991). * [23] M. Avalos, R. Babiano, P. Cintas, J. L. Jiménez and J. C. Palacios, Tetr. Asymm., 11, 2845 (2000). * [24] S. Chandrasekhar,Chirality, 20, 84 (2008). * [25] P. Bargueño and I. Gonzalo, Orig. Life Evol. Biosph., 36, 171 (2006). * [26] P. Bargueño and R. Pérez de Tudela, Orig. Life Evol. Biosph., 37, 253 (2007). * [27] C. Muñoz, Int. J. Mod. Phys., A19, 3093 (2004). * [28] B. Sadoulet, Rev. Mod. Phys., 71, 197 (1999). * [29] G. B. Gelmini, Phys. Script., T121, 131, 2005. * [30] L. Stodolsky, Phys. Rev. Lett., 34, 110, 1975. * [31] J. L. Pérez-Díaz, V. M. Pérez-García and I. Gonzalo, Phys. Lett. A, 160, 453 (1991). * [32] G. Duda, G. Gelmini and S. Nussinov, Phys. Rev. D, 64, 122001 (2001). * [33] A. D. Dolgov et al., Nucl.Phys. B, 632, 363 (2002). * [34] P. Binetruy, Supersymmetry: Theory, Experiment and Cosmology, Oxford University Press, (2006). * [35] http://dmtools.berkeley.edu

arxiv-papers

2024-09-04T02:48:53.768462

0802.2206

# Large–Scale Structure of the Molecular Gas in Taurus Revealed by High Linear Dynamic Range Spectral Line Mapping Paul F. Goldsmith11affiliation: Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena CA, Paul.F.Goldsmith@jpl.nasa.gov , Mark Heyer22affiliation: FCRAO, Department of Astronomy, University of Massachusetts, Amherst MA , Gopal Narayanan22affiliation: FCRAO, Department of Astronomy, University of Massachusetts, Amherst MA , Ronald Snell22affiliation: FCRAO, Department of Astronomy, University of Massachusetts, Amherst MA , Di Li33affiliation: Jet Propulsion Laboratory, California Institute of Technology, 4800 Oak Grove Drive, Pasadena CA , and Chris Brunt44affiliation: University of Exeter ###### Abstract We report the results of a 100 square degree survey of the Taurus Molecular Cloud region in the J = 1 $\rightarrow$ 0 transition of 12CO and of 13CO. The image of the cloud in each velocity channel includes $\simeq$ 3$\times$106 Nyquist- sampled pixels, sampled on a 20′′ grid. The high sensitivity and large linear dynamic range of the maps in both isotopologues reveal a very complex, highly structured cloud morphology. There are large scale correlated structures evident in 13CO emission having very fine dimensions, including filaments, cavities, and rings. The 12CO emission shows a quite different structure, with particularly complex interfaces between regions of greater and smaller column density defining the boundaries of the largest–scale cloud structures. The axes of the striations seen in the 12CO emission from relatively diffuse gas are aligned with the direction of the magnetic field. We have developed a statistical method for analyzing the pixels in which 12CO but not 13CO is detected, which allows us to determine the CO column in the diffuse portion of the cloud as well as in the denser regions in which we detect both isotopologues. Using a column density–dependent model for the CO fractional abundance, we derive the mass of the region mapped to be 2.4$\times$104 $M_{\odot}$. This is more than a factor of two greater than would be obtained using a canonical fixed fractional abundance of 13CO and a factor three greater than would be obtained using this fractional abundance restricted to the high column density regions. We determine that half the mass of the cloud is in regions having column density below 2.1$\times$1021 cm-2. The distribution of young stars in the region covered is highly nonuniform, with the probability of finding a star in a pixel with a specified column density rising sharply for $N(H_{2})$ = 6$\times$1021 cm-2. We determine a relatively low star formation efficiency (mass of young stars/mass of molecular gas), between 0.3 and 1.2 percent, and an average star formation rate during the past 3 Myr of 8$\times$10-5 stars yr-1. ISM: molecules – individual (carbon monoxide) ISM: structure; ISM:individual (Taurus) ## 1 INTRODUCTION The close association of young stars and concentrations within molecular clouds indicates that stars form in cloud cores, which are regions of increased density within the bulk of molecular clouds (cf. Beichman et al., 1986). While the evolution from cloud core to protostar is dominated by gravity, the physics controlling the process in which the cores themselves, and the clouds in which they are embedded, are formed and evolve is still quite controversial. While on the scale of pc to tens of pc molecular clouds are close to satisfying virial equilibrium between gravitational and kinetic energies, the significance of this equality is not entirely clear. Furthermore, the role of magnetic field, while often postulated to be significant, remains uncertain (Shu et al., 1987; Heiles & Crutcher, 2005). Finally, the formation of molecular clouds themselves, and their lifetime, remains very much a matter of discussion (e.g. Hartmann et al., 2001) Molecular clouds may be formed by compression of atomic gas, with the increased density and extinction enhancing the formation rate of molecules, starting with H2, for which self–shielding enables the buildup of a substantial fraction of the total hydrogen density even when the visual extinction AV is only a fraction of a magnitude. It has also been suggested that the large molecular cloud presence in galactic spiral arms is the result of the agglomeration of molecular material existing in the interarm region, as discussed by Pringle, Allen, & Lubow (2001). While one viewpoint has held that molecular clouds have relatively long lifetimes, and are disrupted only by the energy injected by massive star formation and evolution, another picture is that molecular clouds are relatively transient objects, with the denser regions representing only turbulent fluctuations of density rather than well- defined gravitationally bound condensations (see e.g. review by Vázquez–Semadeni, 2007). These issues have been discussed on global scale, addressing the distribution of clouds and the apportioning of molecular and atomic gas in the Galaxy. They are also very relevant to studies of specific molecular cloud complexes, with one of the best–studied of these being that in Taurus. The structure of the interstellar gas in atomic and molecular form, the stellar population, the issue of star formation rate, and the role of different physical processes have all been the subject of numerous papers focused on the Taurus region, primarily because its proximity (140 pc; Elias, 1978)111This value, from Elias (1978), is so entrenched in the literature that we will use it despite the plausible suggestion by Hartigan & Kenyon (2003) that the distance should be reduced by about 10%, to $\sim$126 pc. allows very detailed studies of the morphology of the gas and the relationship between gas and stars. The sheer volume of the data that have been obtained and the number of analyses that have been carried out preclude giving a complete listing of the references to Taurus, so we will have to be selective rather than comprehensive, recognizing that we may have omitted many valuable contributions. The very closeness of Taurus means that available instrumentation, particularly at radio frequencies, has faced a challenge to cover the entire region with angular resolution sufficient to reveal the morphology of the gas. The result has been that previous large–scale surveys of molecular line emission at millimeter wavelengths have been limited to quite low angular resolution (Ungerechts & Thaddeus, 1987). The survey of Ungerechts & Thaddeus (1987) covers essentially all of Taurus and part of Perseus, but the 30′ angular resolution of the map (obtained by averaging multiple telescope pointings to obtain a larger effective beam size) yields only 3000 pixels in the 750 square degree region mapped. The pixel size corresponds to a linear size of 1.2 pc at a distance of 140 pc, which is sufficiently large to blur out structure at important astrophysical scales. In fact, the maps of Ungerechts & Thaddeus (1987), while delineating the large–scale structure quite well, show an almost complete absence of fine detail. This is in part due to the use exclusively of 12CO, which is sufficiently optically thick that significant variations in column densities can be entirely hidden, as well as to the low angular resolution. There have been a number of investigations of molecular gas in the Taurus region with higher angular resolution, but these have typically been limited to small subregions within the overall gas distribution. These studies, with $\simeq$1′ to 2′ angular resolution include a few thousand to $\simeq$ 30,000 spatial pixels (Schloerb & Snell, 1984; Duvert, Cernicharo, & Baudry, 1986; Heyer et al., 1987; Mizuno et al., 1995). These studies, with the combination of higher angular resolution and use of the J = 1$\rightarrow$0 transition of 13CO do reveal considerable structure in the molecular gas, but have not elucidated its relationship to larger–scale features in the molecular gas distribution. A number of other studies have utilized yet higher angular resolution and different tracers to probe gas having different characteristic properties over limited regions. Some examples include Langer et al. (1995) employing CCS, Onishi et al. (1996) and Onishi et al. (1998) using C18O, Onishi et al. (2002) using H13CO+, and Tatematsu et al. (2004) employing N2H+. Many individual cores have been observed in ammonia, a tracer in which they appear relatively well–defined, as indicated by compilation of Jijina, Myers, & Adams (1999). Most of the regions covered by these studies have been pre-selected based on the large–scale surveys discussed above. In these maps, we see indications of finer–scale structure, but the emission is generally quite spatially restricted compared to that seen in the more abundant isotopologues of carbon monoxide. In this paper we present the initial results from a large–scale high angular resolution study of the Taurus molecular clouds using 12CO and 13CO. The data cover approximately 100 square degrees on the sky (11.5° in R.A. by 8.5° in decl.) corresponding to a region 28 pc by 21 pc. The reduced maps include 3.2$\times$106 Nyquist–sampled pixels in each isotopologue, with pixel size 20′′ corresponding to 0.014 pc. The linear dynamic range (LDR, defined as map size divided by Nyquist–sampled interval) of the maps thus exceeds 1000, which is the largest of any molecular cloud study carried out to date. The good angular resolution and large LDR together allow us to examine in detail the relationship between the relatively fine structures seen, especially in 13CO, with the large–scale distribution of the molecular material, the young stars in the region, and the magnetic field. The region of Taurus studied here has been observed using a variety of other tracers. The Leiden/Dwingeloo 21 cm study (Burton & Hartmann, 1994) traced the atomic hydrogen in this direction, but with an angular resolution of 35′. One investigation (Shuter et al., 1987) used the Arecibo radio telescope having an angular resolution of 4′, but included only $\sim$ 1300 positions to probe the self–absorption seen in the 21 cm HI line. This cold atomic hydrogen appears to be associated with molecular gas (Li & Goldsmith, 2003; Goldsmith & Li, 2005), but the limited sampling of Shuter et al. does not reveal much about its morphology. The far–infrared emission from Taurus has been studied by Abergel et al. (1995), who also compared it to moderate resolution maps of 12CO J = 1$\rightarrow$0 emission. The dust column density distribution has been examined by Padoan et al. (2002) and does bear a quite close resemblance to the integrated intensity of 13CO and thus to the column density of gas in relatively high extinction regions. We discuss the observations and data reduction procedure in §2. Derivation of the column density in the different portions of the maps is presented in §3, in which we also discuss the distribution of column density and mass in the region. We present a brief discussion of the large–scale gas kinematics in §4. We address the relationship of the molecular material and the magnetic field in §5, and discuss the relationship of the gas and the young stars in the region in §6. We discuss some of the interesting features of the morphology of the gas in §7. We summarize our results in §8. ## 2 OBSERVATIONS The observations were taken between 2003 November and 2005 May using the 13.7m radome–enclosed Quabbin millimeter wave telescope. The 32 pixel SEQUOIA focal plane array222A 16 pixel single–polarization version of the array is described in Erickson et al. (1999). receiver observed the J = 1$\rightarrow$0 transition of 12CO and 13CO simultaneously. Since the receiver uses amplifiers for the first stage, there is no issue of the sideband gain uncertainty and its effect on calibration. Sixteen pixels are arranged in a 4 x 4 array in two orthogonal linear polarizations. The main beam of the antenna pattern had a full width to half maximum angular width of 45′′ for 12CO and 47′′ for 13CO. The data were obtained using an on–the–fly (OTF) mapping technique. A standard position was observed using position switching several times per observing session to verify calibration consistency. Details of the data–taking, data reduction, and calibration procedures are given by Narayanan et al. (2007). The signals from a band of frequencies around each spectral line were sent to an autocorrelation spectrometer with 1024 lags covering 25 MHz for each spectral line. The lag spacing of the spectrometer system corresponds to 0.068 km s-1 for 13CO and 0.065 km s-1 for 12CO. The data cube of each isotopologue employed in the subsequent analysis included 76 spectral channels for 13CO and 80 channels for 12CO covering approximately -5 km s-1 to +14.9 km s-1 and thus included 2.4$\times$108 voxels. As discussed in detail by Narayanan et al. (2007), the overall quality of the data was excellent. After calibration and combination of the 30′ by 30′ submaps which were the units in which the data was taken, the data were resampled onto a uniform grid of 20′′ spacing, which is very close to the Nyquist sampling interval $\lambda/2D$ for the 13.7 m diameter telescope operating at a wavelength of 2.6 mm. The images produced by the combination of the submaps and regridding were 2069 pixels in RA by 1529 pixels in decl., thus comprising 3,163,501 spatial pixels resampled onto a uniform 20′′ grid. The final data set has a well–behaved distribution of noise with a mean rms antenna temperature equal to 0.125 K for 13CO and 0.28 K for 12CO in channel widths of 0.27 km s-1 and 0.26 km s-1, respectively. Figure 1: Antenna temperature of the 13CO J = 1$\rightarrow$0 transition integrated over the velocity range 2 km s-1 to 9 km s-1. The scale is shown in the bar at the right; values have not been corrected for antenna efficiency. We show the basic 13CO data in Figure 1, which gives the intensity of the 13CO J = 1$\rightarrow$0 transition integrated over the velocity range 2 km s-1 to 9 km s-1. This interval encompasses almost all of the emission in the Taurus region, with the exception of some isolated areas with gas at $\simeq$ 10 km s-1, which may well not be associated with Taurus, and a limited amount of emission in the velocity range 1 km s-1 to 2 km s-1. Figure 2 displays the 12CO J = 1$\rightarrow$0 peak emission within this same velocity interval. Note that in both of these figures, the emission is not corrected for the antenna efficiency. Narayanan et al. (2007) present images of the emission of both isotopologues in 1 km s-1 bins covering the range 0 km s-1 to 13 km s-1. It is evident that the 12CO is detectable over a significantly larger area than is the 13CO. Particularly in the northeast portion of the map, we see very extended 12CO emission, where there is relatively little 13CO. There are also two interesting regions of quite strong 12CO emission, at 4h22m+28∘30′ and 4h48m+29∘40′, which are among the warmest regions observed, and yet which do not show up as significant local maxima in the 13CO (and hence column density). In general, the warmer gas as traced by 12CO is seen in regions of high column density, but the amount of structure seen in the optically thick 12CO with our angular resolution, sampling, and sensitivity, is very impressive. Figure 2: Maximum antenna temperature of the 12CO J = 1$\rightarrow$0 transition over the velocity range 2 km s-1 to 9 km s-1. The antenna temperature has not been corrected for the antenna efficiency. ## 3 COLUMN DENSITY, COLUMN DENSITY DISTRIBUTIONS, AND CLOUD MASS ### 3.1 Mask Regions In order to facilitate analysis of the data to determine column densities, we have broken the Taurus data up into 4 regions, according to the detection or nondetection of 12CO and 13CO. The detection thresholds are defined by the requirement that the integrated intensity over the velocity range extending from 0 km s-1 to 12 km s-1 be a minimum of 3.5 times larger than the rms noise in an individual pixel over this 12 km s-1 velocity interval. The median values are $\sigma_{T_{int}}$ = 0.18 K kms-1 for 13CO and $\sigma_{T_{int}}$ = 0.40 K kms-1 for 12CO. Since the peak values of the integrated intensity are 6 K kms-1 for 13CO and 18 K kms-1 for 12CO, the peak integrated intensities are 30 to 50 $\sigma_{T_{int}}$. We define mask 0 to be the region in which neither 12CO nor 13CO is detected, mask 1 to be the region in which 12CO is detected but 13CO is not, mask 2 to be the region in which both isotopologues are detected, and mask 3 to be the region in which 13CO is detected but 12CO is not. The different regions and the number of pixels in each are given in Table 1. Table 1: Mask Regions in the Taurus Map Mask Region | Characteristics | Number of Pixels ---|---|--- 0 | neither 12CO nor 13CO | string944,802 1 | 12CO but not 13CO | 1,212,271 2 | both 12CO and 13CO | 1,002,955 3 | 13CO but not 12CO | string3,473 The average spectra of mask 0, mask 1, and mask 2 regions are shown in Figure 3. These profiles are valuable for deducing general characteristics of the regions, but it must be kept in mind that the characteristics of the average profile are quite different from those of individual profiles. The difference is primarily due to systematic velocity shifts across the cloud; these result in the average spectra being much weaker and broader than individual spectra. The line width of the averaged mask 1 spectra is close to a factor of 2 greater than the average line width of spectra in this region. For mask 2, the ratio is $\simeq$1.5. Along with this, the peak intensities are much weaker than those seen in individual spectra or even in spectra averaged over a restricted region. Consequently, in determining characteristics of the molecular gas, we have used individual spectra wherever possible to derive physical quantities. Figure 3: Spectra averaged in each of the three significant mask regions of the Taurus map, arranged from mask 0 (bottom) to mask 2 (top). In each panel, the 12CO J= 1$\rightarrow$0 spectrum is the more intense (blue), while the weaker 13CO spectrum is plotted in red. The antenna temperature scale for the 12CO is on the left hand side, while that for the 13CO is on the right hand side, expanded by a factor of three relative to that of 13CO. The dashed vertical lines delineate the velocity range used to define the material in the Taurus region. As expected, the lines are strongest in mask 2. The 12CO to 13CO ratio at the line peak in mask 2 is just over 3, consistent with relatively high optical depth in the more abundant isotopologue. We do see that when an average over $\sim$ 106 pixels of mask 1 is formed, we readily see emission in 13CO as well as 12CO. The ratio of peak intensities is significantly larger in mask 1 than in mask 2. The value, about 10, is still much less than the presumed abundance ratio [12CO]/13CO], suggesting that the 12CO in mask 1, while optically thick, typically has lower opacity than in mask 2. The mask 0 12CO and 13CO spectra show two or three peaks, including velocities for which the emission in mask 2 is very weak compared to that in the range of the peak emission, 5 km s-1 to 8 km s-1. In particular, the 10 km s-1 emission feature comes from a fairly extended region in the northern portion of our map, but is so weak that only when averaging over modest-sized ($\sim$1 square degree) regions in mask 0 can it be detected. Emission in this velocity range can be quite clearly seen in the mask 1 spectrum, but hardly can be detected in mask 2. This is consistent with it being relatively low average column density material, which is extended over quite large areas. Thus, even in what we consider largely “empty” regions between the major, well–known subunits of the Taurus molecular cloud complex, there is molecular gas. This is discussed further in the following section. The overall composition of the mask 0 region, particularly the presence of atomic gas, is the subject of another study. The mask 0, mask 1, and mask 2 regions have close to equal numbers of pixels. Their distribution, however, is very different. Figure 4 shows the four mask regions. It is evident that the mask 1 predominantly surrounds mask 2, which is consistent with the expectation that both isotopologues are detected in the regions of highest column density (mask 2) while in the periphery of these regions we detect in individual pixels the 12CO but not the 13CO emission. Figure 4: Image showing the mask regions (see text for definitions) in the Taurus molecular cloud. Mask 0 is shown in black, mask 1 in blue, and mask 2 in red. The relatively few mask 3 pixels are not shown. The pixels in mask 3 are unusual inasmuch as they exhibit detectable 13CO emission but not 12CO. There are evidently very few such pixels ($\simeq$ 0.1% of the total), although this number is considerably larger than would be expected purely on the basis of Gaussian noise statistics. On close inspection of these spectra, it appears that the problem is due to very low level baseline imperfections partially canceling the 12CO integrated intensity, resulting in a “non–detection” of this isotopologue. We thus ignore the mask 3 pixels in further analysis of the emission from Taurus. ### 3.2 Calculation of the Column Density We wish to exploit the large linear dynamic range of our map to examine the structure in the column density, and thus wish to determine the column density for as many pixels as possible. This is also important for accurately determining the total molecular mass of the region. In what follows we divide the problem into two parts. The first is determination of the carbon monoxide column density. While subject to its own uncertainties due to excitation, optical depth, and limited signal to noise ratio, we can carry out this step of the analysis based only on data in hand. The second step is conversion of the carbon monoxide column densities to molecular hydrogen column densities, and finally to total cloud mass. This is evidently dependent on the processes which determine the fractional abundance of the various isotopologues observed. Since the additional uncertainties in the second step are large, we present results first in terms of the carbon monoxide distribution and subsequently give results for the molecular hydrogen distribution and the total molecular mass. This second step should benefit significantly from combination of our data with dust column density determined from e.g. 2MASS data. This effort is in progress and will be reported in a subsequent publication. #### 3.2.1 Carbon Monoxide Column Density The three different different regions of the cloud, defined by the detectability of each isotopologue, require different schemes to determine the carbon monoxide column density. We ignore mask 3 in determining the column density and to the mass of the cloud as we cannot readily correct for the artificial non–detections of 12CO (discussed above). Its extremely small area and weak 13CO emission make its contribution negligible. Mask 2 represents the portion of the cloud that is most conventional in terms of column density determination. Since we have both 12CO and 13CO in each pixel, we determine the kinetic temperature from the peak value of the 12CO (with appropriate correction for antenna efficiency). Here (as well as for other mask regions), we use the maximum antenna temperature of 12CO in the velocity interval between 0 km s-1 and 12 km s-1. The kinetic temperature is distributed from 3 K to 21 K, but with the vast majority of positions having kinetic temperatures between 6 K and 12 K. Since mask 2 is the densest portion of the cloud, we assume that the 13CO levels are populated in LTE at the kinetic temperature, but we calculate a nominal value for the optical depth from the ratio of the peak 13CO and 12CO intensities using the usual equation of radiative transfer in a uniform medium. We assume a 12CO to 13CO abundance ratio of 65, very close to the average value for local clouds found by Langer & Penzias (1993). We use the value of optical depth obtained to make a saturation correction to the 13CO column density derived assuming optically thin emission, with the usual formula $N(^{13}\rm{CO~{}corrected})=N(^{13}\rm{CO~{}assumed~{}optically~{}thin)}\frac{\tau}{1-exp(-\tau)}\>\>.$ (1) Mask 1 presents the greatest challenge in terms of column density determination since it encompasses approximately one third of the area mapped and has reasonably strong 12CO emission. However, since the 13CO is not detected in individual pixels, we need a different scheme to extract the column density. We have developed a statistical approach, which should be applicable to other large maps in which only the more abundant isotopologue is detected in individual pixels. The procedure assumes that the 12CO is optically thick at its peak, and that the value of the antenna temperature can directly be converted to the excitation temperature of the 12CO. Since mask 1 points lie at the periphery of the regions of high extinction and greater molecular column density (as witnessed by the detection of 13CO in each mask 2 pixel), they encompass lower column density gas which is presumably characterized by lower volume density. Therefore we cannot assume that LTE applies as it does in mask 2. Approximately half of the mask 1 positions have an excitation temperature $\leq$ 7.5 K, and if in LTE the gas would have to be unusually cold. It is thus reasonable to assume that this gas is subthermally excited. To analyze positions in mask 1 we use a simple excitation/radiative transfer analysis employing a spherical cloud large velocity gradient (LVG) code to compute the line intensities (e.g. Snell, 1981; Goldsmith, Young, & Langer, 1983). We are using an LVG model largely as a tool to characterize the effect of trapping, which is important for excitation of CO at lower density. We do not believe it necessarily represents any statement about the detailed kinematics of the gas. The sensitivity of our results to the details of the velocity field should be quite small. We have assumed that the kinetic temperature of the mask 1 region is uniformly 15 K, somewhat higher than well-shielded dense gas, which is plausible in view of increased heating in the peripheral regions surrounding regions of high extinction. (e.g. Li, Goldsmith, & Menten, 2003). We take advantage of the large number of pixels in our map, and bin the data according to the excitation temperature of the 12CO determined as described above. In each bin, we have a sufficient number of pixels that the 13CO J = 1$\rightarrow$0 line is detected with good signal to noise ratio. For each $T_{ex}$ bin, we then have the 12CO excitation temperature and the observed 12CO/13CO integrated intensity ratio. The data generally have the observed intensity ratio decreasing with increasing $T_{ex}$, from $\simeq$ 22 for $T_{ex}$ = 4.5 K to $\simeq$ 13 for $T_{ex}$ = 2.5 K. The free parameters are the 12CO column density, the H2 density, and the 12CO/13CO abundance ratio. The latter cannot be assumed to be a fixed value (e.g. 65), due to the complicating presence of isotopic enhancement due to chemical and/or photo effects (e.g. Watson et al., 1976; Bally & Langer, 1982; Chu & Watson, 1983; Van Dishoeck & Black, 1988). We thus consider $R$ = 12CO/13CO between 25 and 65. Table 2: 12CO Excitation Temperature Bins in Mask 1 and Best Estimates of Their Characteristics11The last three columns are model values. Tex | 12CO/13CO | Number of Pixels | n(H2) | N(12CO)/$\delta$v | 12CO/13CO ---|---|---|---|---|--- K | Observed | | cm-3 | 1016 cm-2/kms-1 | Abundance Ratio 4.5 | 21.7 | string32321 | string125 | 0.7 | 30 5.5 | 21.7 | 113923 | string200 | 1.0 | 35 6.5 | 19.6 | 202328 | string250 | 1.4 | 38 7.5 | 16.7 | 245949 | string280 | 2.0 | 40 8.5 | 14.9 | 211649 | string325 | 2.7 | 42 9.5 | 13.9 | 175431 | string425 | 3.1 | 45 10.5 | 13.4 | 122423 | string550 | 3.6 | 50 11.5 | 13.0 | string65428 | string850 | 3.7 | 55 12.5 | 12.8 | string27387 | 1200 | 4.3 | 65 With three free parameters and only two observables, we cannot uniquely determine the properties of the gas in mask 1. Rather, we compute for each $T_{ex}$ bin, a family of $R$, density and CO column density per unit line width solutions. If we knew a priori the value of $R$, then we could compute a unique density and CO column density per unit line width for each $T_{ex}$. With no knowledge of $R$, then the values of density and CO column density per unit line width span a range of approximately a factor of 4, with density and CO column density per unit line width inversely correlated. The family of solutions for the physical parameters of the gas show some significant general characteristics. First, for higher values of $T_{ex}$, only solutions with $R\geq 50$ fit the data. This is encouraging as the higher excitation gas has on average the largest column density and we do not expect significant fractionation in the more shielded regions. On the other hand, for lower values of $T_{ex}$, values of $R$ as large as 65 are excluded, and the range of acceptable solutions gradually shifts from $R$ $\leq$ 50 at $T_{ex}$ = 7.5 K to values of $R$ $\leq$ 30 at $T_{ex}$ = 4.5 K. Correspondingly, the allowable solutions for the gas density and CO column density per unit line width decrease with decreasing excitation temperature. This trend again is consistent with increasing fractionation in the less well–shielded regions at the periphery of the clouds (see Liszt 2007) for a discussion of this effect in diffuse clouds). These regions dominate the positions found within our mask 1. This result agrees with the behavior found in previous observational studies (e.g. Goldsmith et al., 1980; Langer et al., 1980; Young et al., 1982; Langer et al., 1989; Goldsmith & Li, 2005; Kainulainen et al., 2006). It is not possible to model the mask 1 observations with a fixed value of the in situ carbon monoxide isotopic ratio but rather require that the value of $R$ vary significantly with excitation temperature. We have chosen solutions such that $R$ varies smoothly from a value of 65 at $T_{ex}$ $\geq$ 12.5 K to a value of 30 for $T_{ex}$ = 4.5 K. With this choice of $R$, we find that the gas density and CO column density per unit line width both increase monotonically with increasing excitation temperature. The solutions we have chosen are shown in Table 2 and in Figure 5. We emphasize that these solutions are not unique, but depend on our choice of $R$. However, the general behavior of the solutions are physically plausible, given that we expect the excitation temperature to increase as one moves from the cloud interior to the cloud periphery. This suggests that binning by $T_{ex}$ is a useful approach, and gives us a reasonable handle on how the physical conditions vary as a function of excitation temperature and position in the cloud. Our assumption of 15 K for the kinetic temperature is a potential source of error in determining the carbon monoxide column density. To assess this, we have carried out some calculations using a kinetic temperature of 25 K which seems an upper limit to what one might expect in a cloud edge in a region with modest UV intensity. We find that for this value of the kinetic temperature, the column density per unit velocity gradient is approximately a factor 1.5 larger than for a kinetic temperature of 15 K, and the derived H2 density is a factor of 2.5 lower, for an assumed value of $R$. The same trends of carbon monoxide column density and H2 density as a function of $R$ are seen for the higher kinetic temperature as for the lower. The uncertainty resulting from the assumption of a fixed kinetic temperature is thus of the same order as resulting from our choosing a best value of $R$, and combining these could yield a factor of 2 uncertainty in $N$(CO). Observations of multiple transitions of carbon monoxide isotopologues would provide a more accurate estimate of the molecular column density. However, observations of these transitions over a region of comparable size would pose a formidable challenge for currently available telescopes and receiver systems. To obtain the column density for each line of sight, we utilize an analytic fit to the relationship between the CO column density per unit line width and the excitation temperature obtained for the set of $T_{ex}$ bins, $N(^{12}CO)/\delta v=(-1.473\times 10^{16}+4.672\times 10^{15}T_{ex})$. We multiply the results by the observed FWHM 12CO line width $\Delta v$ from the data. The use of the LVG model introduces some uncertainty because the carbon monoxide excitation is quite subthermal, and the excitation temperature does depend on the optical depth, and is quite different for 12CO and 13CO. Nevertheless, the likely error in the trapping predicted by the LVG and other models is relatively modest compared to other uncertainties inherent in this analysis. Figure 5: Parameters of mask 1 pixels binned by 12CO excitation temperature $T_{ex}$. The bottom panel shows the observed 12CO/13CO intensity ratio (left hand scale; triangles), and the number of pixels in each excitation temperature bin (right hand scale; squares). The most common excitation temperatures are between 6 K and 10 K. The middle panel shows the H2 density (left hand scale; triangles) and 12CO column density assuming a line width of 1 km s-1 (right hand scale; squares). These are obtained from the 12CO and 13CO intensities. The top panel shows the derived 12CO/13CO abundance ratio. The H2 density, 13CO column density, and the derived 12CO13CO ratio all increase monotonically as a function of 12CO excitation temperature. Figure 6: Histogram of 12CO column density distributions in mask 0, 1, and 2 regions mapped in Taurus. The mask 0 data are indicated by the vertical dashed line, which represents the approximately 106 pixels in this region. The distribution for mask 1 is shown by the solid (blue) curve, that for mask 2 by dot–dashed (red) curve, and the combination of the two regions by the long–dashed (black) curve. The abscissa is the logarithm of the 12CO column density, defined in bins of width 0.1 dex, except for mask 0 which is a single value. The ordinate is the logarithm of the number of pixels in each column density bin. The total number of pixels included is close to 3.1 million. In mask 0, after averaging $\simeq$ 106 spatial pixels, we are able to detect both isotopologues, and we thus analyze the emission for the region as if it were a single spatial entity. The general analysis follows the procedure described above for mask 1. The fact that the integrated 12CO/13CO ratio is $\simeq$ 19 indicates that the 13CO is almost certainly optically thin. This is also the case for mask 1, and here as well results in the 12CO and 13CO having quite different excitation temperatures due to the radiative trapping for the more abundant isotopologue. Again, we fix the kinetic temperature to be 15 K, reflecting increased heating in regions of low extinction, and assume that the average line width is 2 km s-1, similar to that observed for the low excitation gas of mask 1. Note that the average mask 0 spectrum (Figure 3) is much broader than 2 km s-1, but the large value of the line width reflects changes in the line center velocity over the entire region observed. Following the trend of $R$ from mask 1, we assume this ratio to have a value of 20. The mask 0 data cannot be fit satisfactorily by larger values of $R$ thus confirming that relatively strong isotopic selective effects are at work in the low density/low column density regions of Taurus. With these assumptions, the parameters we derive, although again not unique as described above, are $n(H_{2})$ = 75 cm-3, and $N$(12CO) = 7.5$\times$1015 cm-2. The carbon monoxide excitation in this region is evidently highly subthermal, consistent with the low derived H2 density and the modest 12CO optical depth. This very low value for the density of the mask 0 region gives a reasonably low column density for the extended component of the gas in Taurus. Taking a representative dimension for mask 0 of 1.5$\times$1019 cm-2, we obtain $N(H_{2})$ = 1.1$\times$1021 cm-2. This corresponds to Av $\simeq$ 1 for the extended component of the cloud, consistent with that determined from stellar reddening (Cernicharo & Guélin, 1987). The spatial distribution of column densities from the three mask regions is shown in Figure 6. The column density for mask 0 is a single value $<N(^{12}CO)>$ = 7.5$\times$1015 cm-2 as given above. The column density distribution in the mask 1 region is a relatively symmetric, fairly Gaussian distribution with a mean value $<N(^{12}CO)>$ = 3.6$\times$1016 cm-2. The column density distribution in the mask 2 region is flat–topped with a mean value $<N(^{12}CO)>$ = 1.3$\times$1017 cm-2. The distribution of carbon monoxide in the Taurus region is shown in Figure 7. This figure dramatically illustrates the complexity of the molecular gas distribution. The impression given is quite different from that of studies with low angular resolution, in that instead of an ensemble of “relaxed”, fairly smooth condensations one sees a great deal of highly filamentary structure, a strong suggestion of cavities and surrounding regions with enhanced column densities. The large size of the region covered also suggests relationships between the different portions of the Taurus molecular region. The most striking of these points will be addressed briefly later in this paper. Figure 7: Distribution of carbon monoxide in Taurus region expressed as the logarithm of the 12CO column density (cm-2) in mask 1 and mask 2 regions. The scale is indicated by the bar on the right hand side. The 12CO column density in the mask 0 region (indicated by light grey density) is 7.5$\times$1015 cm-2, as discussed in the text. The maximum 12CO column density is a factor of $\simeq$ 100 larger. #### 3.2.2 Molecular Hydrogen Column Density and Mass Most studies of molecular regions using carbon monoxide have emphasized regions in which the column density is sufficiently large that dust shielding plus self–shielding result in an “asymptotic” 12CO abundance between 0.9$\times$10-4 and 3.0$\times$10-4 relative to H2 (see e.g. Frerking, Langer, & Wilson, 1982; Lacy et al., 1994). In our study of Taurus, only the mask 2 region is plausibly consistent with this assumption. The remainder of the cloud is characterized by lower densities and column densities, and the fractional abundance of carbon monoxide must be regarded as being significantly uncertain and likely to be dependent on the extinction. There is considerable value in trying to make a self–consistent model for the carbon monoxide as a tracer of total molecular (H2) column density. To this end, we have used the theoretical modeling by Van Dishoeck & Black (1988). We have utilized the curve for $I_{UV}$ = 1.0 (in units of Habings), carbon depletion $\delta_{C}$ = 0.1, and models T1–T6, which correspond to temperature range 40 K to 15 K and nH = 500 cm-3 to 1000 cm-3 throughout the model slab being considered. We have used a polynomial fit to the data from the appropriate curve in Figure 8 of Van Dishoeck & Black (1988) for the relationship between CO and H2 column densities. This value of carbon depletion is recommended by Van Dishoeck & Black (1988) as agreeing with the available Taurus data. We also note that the carbon monoxide fractional abundance as given by these models of Van Dishoeck & Black (1988) agrees well at low column densities with the UV measurements of Sonnentrucker et al. (2007) and Burgh et al. (2007). The lower CO lines in absorption from diffuse clouds lying in front of millimeter continuum sources have been observed by Liszt & Lucas (1998). The clouds, analyzed by Liszt (2007) have a range of H2 column density (determined by UV absorption; Federman et al. 1994) which extends from 5$\times$1020 cm-2 to just above 1021 cm-2, and thus includes our mask 0 (and very low end of mask 1) results. While there is considerable scatter among various clouds having the same hydrogen column density, the best fit relationship gives $X$(12CO) = 5$\times$10-6 for $N$(H2) = 1021 cm-2. This is quite close to our results and again reinforces the general applicability of a reduced carbon monoxide fractional abundance for low extinction cloud material. The specific parameters we have adopted have been chosen, in addition to being consistent with the measurements of low column density diffuse clouds, to give good agreement at high column densities with the mm emission measurements of Bachiller & Cernicharo (1986), Cernicharo & Guélin (1987), and Alves, Lada, & Lada (1999). The strong dependence of CO column density on H2 column density reflects the onset of self–shielding when N(CO) reaches $\simeq$1015 cm-2. This produces a rapidly increasing CO fractional abundance as a function of H2 column density in the range covered by the mask 0 and mask 1 regions of our study, and a gradual leveling out of N(CO)/N(H2) in mask 2. The most significant difference is that using this approach we find that the low CO column densities correspond to considerably larger H2 column densities than would be found if a constant fractional abundance of CO were adopted. We convert our CO distribution to a molecular hydrogen distribution using the nonlinear relationship, and the result is given in histogram form in Figure 8. Figure 8: Histogram of H2 column density distribution in Taurus. The distribution of column density in the mask 0 region is given by the vertical (dashed) line, in the mask 1 region by the solid (blue) curve, in the mask 2 region by the dot–dashed (red) curve, and in the combination of the two regions by the long–dashed (black) curve. When compared to Figure 6, it is evident that the varying fractional abundance has resulted in a significant compression in converting the carbon monoxide to H2 column densities. The drop in X(CO) in regions of lower extinction and lower density means that the relatively weak emission that we observe there implies a greater H2 column density than would be derived assuming a constant fractional abundance. Taking mask 0 as an example, the CO column density of 7.5$\times$1015 cm-2, with fractional abundance 7.0$\times$10-6 corresponds to an H2 column density equal to 1.1$\times$1021 cm-2 using the variable fractional abundance, more than an order of magnitude larger than would be obtained using the canonical high–extinction fractional abundance of 10-4. This suggests that the majority of the area within the Taurus molecular cloud complex has a visual extinction from molecular hydrogen on the order of 1 magnitude. This is consistent with the hydrogen column density of mask 1 discussed in the previous section, as well as with the ”halo” component of the HCL2 region discussed by Cernicharo & Guélin (1987). There is certainly a high column density tail which reaches 1022 cm-2, but this includes only a very small fraction of the cloud area and mass. While 13CO is not the ideal tracer of the densest component of the cloud, this study makes it clear that only about 10-3 of the pixels with 12CO detectable have $A_{v}$ $\geq$ 5. Despite the relatively low density in mask 0 and mask 1 regions, the time scale to arrive at the the low fractional abundance of carbon monoxide found there is quite modest. Using the expression from Section 4.1 of Liszt (2007), we find that if we start with $X(e)$ = 10-5 and $n$(H2) = 100 cm-3, the characteristic time to reach $X$(CO) = 10-5 is only $\sim$ 105 yr. This is consistent with results obtained using explicit time–dependent models with CO formation and destruction by E. Bergin (private communication). Thus, whatever the history of the diffuse surroundings of dense clouds, the low but significant abundance of carbon monoxide found there appears entirely plausible. We show the spatial distribution of H2 column density in the Taurus region mapped in Figure 9. The contributions of individual pixels in mask 1 and mask 2 are included. Approximately 50 percent of the total molecular mass of the region is in directions in which 13CO cannot readily be detected in an individual map pixel. From the masses in each mask region, we compute the total mass of the region of Taurus mapped in the present study. The results (including correction for He and heavy elements) are given in Table 3. For mask 0, we have considered the entire area it comprises to be characterized by the single set of conditions derived in the previous subsection, while the contribution of mask 3 has been neglected. Table 3 shows that assuming the physically plausible variable fractional abundance of carbon monoxide gives a total mass of the region a factor approximately 2.5 times larger than that obtained using a uniform high abundance characteristic of well–shielded regions. We also see that the contributions from the low column density mask 0 and mask 1‘ regions are considerably enhanced and that their contribution to the total mass is no longer negligible as would be the case if a constant fractional abundance obtained. Figure 9: Image showing the molecular hydrogen column density distribution derived from mask 0, mask 1, and mask 2 regions in Taurus. The scale is indicated by the bar on the right, expressed as the logarithm of the derived H2 column density in cm-2. Table 3: Mass of Region in Taurus Mapped Mask Region | Mass (103 $M_{\odot}$) | ---|---|--- | a | b 0 | 0.1 | string4.1 1 | 1.7 | string7.7 2 | 7.8 | 11.8 Total | 9.6 | 23.6 ### 3.3 Cloud Structure Valuable insight into the structure of the cloud can be obtained by examining the cumulative distribution of cloud mass and area as a function of column density. This information in shown in Figure 10. Our survey focused on the region of the Taurus molecular cloud known to have most prominent high density regions with exceptional chemical diversity (TMC-1; Pratap et al., 1997) and prominent star formation (e.g. L1495). Nevertheless, we see that half of the cloud’s mass is in material with N(H2) less than 2.1$\times$1021 cm-2. Only about 5% of the cloud’s mass occurs at H2 column densities above 5$\times$1021 cm-2, or visual extinction greater than 5. The column density we derive may be modestly underestimated due to incomplete correction for saturation in our 13CO observations for large column densities, and as a consequence of molecular depletion at high densities, but even together these effects are unlikely to increase this fraction by a factor of 2 (see e.g. Alves, Lada, & Lada, 1999). The fraction of the cloud area with N(H2) $\geq$ 5$\times$1021 cm-2 is only 0.02. Figure 10: Cumulative fraction of Taurus as a function of molecular hydrogen column density. The solid (blue) curve gives the fraction of the area characterized by column density less than specified value, and the dotted (red) curve gives the fraction of the mass similarly characterized. Half the area mapped has column density below 1.6$\times$1021 cm-2, and half the mass of the cloud is included in regions having column density below 2.1$\times$1021 cm-2. Another view of the mass distribution can be obtained by attempting to dissect the cloud by extracting the well–recognized high column density regions from the remainder of the gas. In Figure 11 we show the division into eight regions, which together include approximately 25% of the area of the map. We have generally followed the region limits and designations given in Fig. 3 of Onishi et al. (1996). Figure 11: Image showing H2 column density overlaid with the eight well-known regions of high column density as designated by Onishi et al. (1996). These regions define the masses and areas given in Table 4. We give the mass of each of these regions in Table 4. The total mass contained in these regions, 9807 $M_{\odot}$, is 42% of the total mass included in the region we have studied, and their combined area is 21% of that of the region we have mapped. However, since we have made an unbiased map of 13CO rather than a map restricted to regions of strong intensity (as Mizuno et al., 1995, did in their 13CO survey), we include somewhat larger areas. The masses we derive for L1495/B213 and for B18 are approximately a factor of 3 larger than those obtained by Mizuno et al. (1995), and that for HCl2 is a factor of 2 larger. It is evident that a large fraction of the mass even within the boundaries shown in Figure 11 is in relatively low–density gas. Table 4: Mass of High–Density Regions in TaurusaaRegions defined in Figure 11 Region | Mass bbUsing H2/CO ratio with I(UV) = 1.0 and $\delta_{C}$ = 0.1 from Van Dishoeck & Black (1988) | Area ---|---|--- | $M_{\odot}$ | pc2 L1495 | 2616 | string31.7 B213 | 1095 | string13.7 L1521 | 1584 | string17.6 HCl2 | 1513 | string15.8 L1498 | string373 | string5.7 L1506 | string491 | string7.7 B18 | 1157 | string14.5 L1536 | string978 | string16.6 Total | 9807 | 123.3 bbfootnotetext: Includes correction for He Having a well–sampled 12CO map of a large region and a mass determination allows us to examine the application of a CO luminosity to mass conversion factor (Dickman, Snell, & Schloerb, 1986) to Taurus. In Table 5 we show the results for the different mask regions and the total. The entries in the third column are obtained using a conversion factor M($M_{\odot}$) = 4.1LCO(K km s-1 pc2). This value is obtained using the Egret $\gamma$–ray data (Strong & Mattox, 1996), and a factor 1.36 for the total mass per H2 molecule (including He and metals) in the gas. For mask 0, the CO luminosity drastically underestimates the mass, due to highly subthermal excitation of the CO and its modest optical depth. For the denser regions, the agreement is much better. The surprisingly close agreement for the complete Taurus region may, to a certain extent, be fortuitous, but it suggests that use of the 12CO luminosity to derive total mass of molecular regions does appear to work reasonably well for regions with only low–mass young stars, as well as for regions with young high–mass stars. Table 5: Comparison of Masses Determined from 12CO and 13CO With Those Derived from CO Luminosity Region | Mass from | 12CO Luminosity | Mass from ---|---|---|--- | 12CO and 13CO | | 12CO Luminosity | ($M_{\odot}$) | (K km s-1 pc2) | ($M_{\odot}$) mask 0 | string4081 | string193 | string791 mask 1 | string7699 | 2052 | string8413 mask 2 | 11752 | 3305 | 13550 Total | 23532 | 5550 | 22754 ## 4 LARGE SCALE KINEMATICS OF THE MOLECULAR GAS Previous studies have revealed a variety of motions on different scales within the Taurus complex. These include velocity gradients along individual filaments possibly indicative of rotation, along with a systematic East–West velocity difference as one moves across the region. In Figure 12 we show a color–coded image of the integrated intensities in three velocity intervals for the two isotopologues. There is a great deal of structure seen even in this relatively crude representation of the velocity field. Certain regions, and particularly the edges of particular regions, show up as having significantly shifted velocities relative to the surrounding gas. This coarsely divided integrated intensity does not give the full measure of the complexity of the 13CO and 12CO line profiles in Taurus. An indication of this can be seen in Fig. 20 of Narayanan et al. (2007), in which it is evident that in general the regions with multiply–peaked lines exhibit this characteristic in both 12CO and 13CO. Since the visibility of the multiple peaks is approximately equal in the two isotopologues, it is unlikely to be a result of self–absorption, but rather an indication of multiple, kinematically distinct components. These are most prominent in several regions of Taurus, notably the western part of B18, north of L1521, in B213 and west thereof, and in the southern part of Heiles’ Cloud 2. This indicates that some regions are characterized by a considerably greater degree of velocity multiplicity along lines of sight. There does not appear to be any correlation of this characteristic with e.g. star formation. Figure 12: Color–coded image of the integrated intensities in three velocity intervals of carbon monoxide in Taurus, with emission at velocities between 3 km s-1 and 5 km s-1 coded blue, between 5 km s-1 and 7 km s-1 coded green, and between 7 km s-1 and 9 km s-1 coded red. Upper panel: 12CO with blue– and red–coded intervals scaled to 7.5 K kms-1 and the green–coded interval scaled to 15 K kms-1. Lower panel: 13CO with blue– and red–coded intervals scaled to 5.0 K kms-1 and the green–coded interval scaled to 2.5 K kms-1. ## 5 MOLECULAR GAS AND THE MAGNETIC FIELD The Taurus Molecular Cloud has long been a target for investigations of the interstellar magnetic field and its role within the dynamics of the molecular gas component (Moneti et al., 1984; Heyer et al., 1987; Heyer, 1988; Goodman et al., 1992; Troland et al., 1996; Crutcher & Troland, 2000). Many of these studies have compared the distribution of gas and dust with respect to the magnetic field geometry inferred from optical polarization measurements of background stars. The relationship of the cloud geometry to the magnetic field morphology is an essential aspect of models that have been developed for the formation of Taurus (Gomez de Castro & Pudritz, 1992; Ballesteros-Paredes, Hartmann, & Vázquez–Semadeni, 1999). These have hypothesized an initial alignment of a more diffuse cloud with the Galactic magnetic field as part of the initial conditions for formation of the dense cloud, with the gas streaming along magnetic field lines. Observationally, at intermediate scales ($\sim$1 pc), the situation has become more complex. In particular, toward the western end of the Taurus cloud, the long axis of the L1506 filament is oriented along the field in contrast to alignments of Heiles’ Cloud 2 and the B216 and B217 filaments for which the field is essentially perpendicular to the axis of the filaments (Goodman et al., 1992). Note that the latter structure is denoted B213 in Figure 11. From this departure from rigorous alignment, Goodman et al. (1992) conclude that either the magnetic field does not dominate the cloud structure at these scales and densities, or that the optical polarization measurements probe a volume that is spatially distinct from the dense filaments. Goodman et al. (1992) demonstrate that polarization by selective absorption at optical and infrared wavelengths is produced by dust grains within the outer, low column density envelopes of the molecular clouds and provides little or no information on the magnetic field direction within the high density filaments. The 12CO and 13CO data presented in this study afford an opportunity to extend these comparisons to lower column densities than these previous investigations. We have used the data assembled by Heiles (2000), taken from other sources, and superimposed this on a figure showing the integrated intensities of 12CO and 13CO. Figure 13 shows the results. Figure 13: Upper panel: magnetic field direction in Taurus superimposed on the 12CO antenna temperature distribution integrated over the 5 km s-1 to 8 km s-1 velocity range, chosen to emphasize striations and other fine structure. The line segments indicate the direction of the magnetic field derived from observations of absorption by polarized dust grains; their length is proportional to the fractional polarization. The horizontal bar at lower left in each panel indicates 5% fractional polarization. Lower panel: magnetic field direction superimposed on the distribution of 13CO antenna temperature integrated over the same velocity interval. The integrated intensities for each panel in K kms-1 are indicated by the bar at the right. This figure highlights the relationship between the field direction and the morphology of the dense filaments of gas discussed in the references given above. We can use the 12CO emission to probe the relationship between the lower column density portions of Taurus and the magnetic field. This comparison is shown in the top half of Figure 13. Within the faint, low surface brightness 12CO emission, we see marked striations, which are discussed in more detail in §7.2.6. Remarkably, these features within the Taurus Cloud follow the local orientation of the magnetic field even as the polarization angles vary from a mean of 53 degrees within the northeast corner of the surveyed area, to 81 degrees within the southwest corner. The alignment of these faint features points toward a strong coupling of the gas with the interstellar magnetic field. Such strong coupling may be expected in these low column density regions that are more exposed to the ambient, UV radiation field, which maintains a higher degree of ionization. The origin of these threadlike features and the mechanism whereby they are aligned with the magnetic field are not established, but we can speculate on several processes that may be responsible. The channel maps of the molecular line emission identify regions of systematic motions over scales from the resolution limit up to 30′ to 60′. If the magnetic field is well coupled to the neutral gas by frequent ion-neutral collisions but the magnetic energy is small with respect to the kinetic energy of the gas, then the field can be carried by these large scale flows within the cloud. Correspondingly, the field lines would be stretched along the direction of the flow. Alternatively, the narrow emission threads may arise from successive compressions and rarefactions of the gas and magnetic field produced by magnetosonic waves that propagate perpendicular to the field. Within the subthermally excited regime, which likely prevails within these regions of low surface brightness, these column density perturbations would produce corresponding variations in the 12CO intensity. ## 6 MOLECULAR GAS AND YOUNG STARS IN TAURUS The distribution of young stellar objects with respect to the molecular gas may offer valuable insights to the formation of stars within a dense interstellar cloud. For comparison with our molecular images, we adopted the set of pre-main sequence stars in the Taurus regions from S. Kenyon (2007 private communication, to be published in 2008). This list is comprised of data from many surveys in optical and infrared wavebands 333We obtain essentially the same results using the data compiled by F. Palla, which was also provided to us as a private communication.. The pre–main sequence stars are divided into three populations according to their colors. If the R-K magnitude is larger than eight, the star is categorized as likely to be a Class I or younger source. If R-K is smaller than eight, the source is likely to be a T-Tauri star. If the source is not detected in either R or K, it is is likely to be extended/nebulous, in which case it is probably still a protostar, younger than a T-Tauri star. In the region covered by our map, there are a total of 230 stars, 18 of which are Class I or younger, 44 are extended, and 168 are likely to be T-Tauri stars. The stars are shown overlaid on the distribution of the H2 column density in Figure 14. The distribution of pre–main sequence stars generally follows that of the dense gas, although a many of the stars in the older category are located in regions with only diffuse gas emission. As noted by Hartmann (2002), the young stars are grouped in three nearly parallel bands that are associated with Heiles’ Cloud 2/L1521/B213/L1495, B18/L1506 and L1536. Figure 14: Locations of young stars in Taurus superimposed on map of the H2 column density. The stellar positions are from Kenyon (2007). The diamonds indicate diffuse or extended sources (of which there are 44 in the region mapped), the squares indicate Class I or younger stars (18), and the asterisks indicate T-Tauri stars (168). It is evident that the diffuse and younger sources are almost without exception coincident with regions of relatively large column density, while the older stars show a much larger probability of being found in regions of lower column density. The relationship between H2 column density and stellar population is examined further in Fig. 15. Figure 15: Upper left: number of stars as a function of column density. The same bins of column density are used in all panels. Upper right: fraction of pixels in each column density bin. Lower left: surface density of stars obtained by dividing the number of stars in each bin by the area of the map corresponding to each bin. Lower right: surface density of stars in each bin for likely class I and diffuse sources (solid line) and likely T-Tauri stars (dashed line). The division of sources is discussed in the text. Roughly equal number of stars can be found in each of the column density bins spanning the range from 0 to $6.5\times 10^{21}$ cm-2 (upper-left panel). Although the number of stars drops towards higher column density regions, such direct examination of the distribution of stars is somewhat misleading inasmuch as our map includes a substantial area with very weak or no carbon monoxide emission, as shown in the upper right panel of Figure 15. The surface density of stars versus column density is plotted in the lower left panel. A significant jump in the surface density occurs at around $N(H_{2})=6\times 10^{21}$ cm-2, or roughly, Av = 6, suggestive of a threshold for star formation. Note that the same trend is visible even in a sample of mostly T-Tauri stars (lower right panel). In Taurus, neither the dispersion of gas due to star formation nor the dispersion of stars due to stellar motion is likely to have altered the collocation of very dense gas and highly extincted young stars. The threshold in column density for star formation is consistent with the conclusion of Mizuno et al. (1995) with the difference being our finding a higher threshold of $6\times 10^{21}$ cm-2 instead of $3\times 10^{21}$ cm-2. Given the larger number of pre–main sequence stars available for the present work, the significance of the change in the stellar surface density is also higher. With our rather complete coverage of gas and stars, we can examine the relationship of the stellar mass to the gas mass, which defines the star formation efficiency (or SFE). From a very simplified point of view of the time evolution of the star formation process, we can define the star formation efficiency in three ways. In the first, the SFE is defined as the mass of all known young (pre–main sequence) stars divided by the total gas mass. Assuming an average mass of 0.6 solar mass for each of the stars in our sample (following Palla & Stahler, 2000) and the total molecular mass of 2.4$\times 10^{4}$ $M_{\odot}$ (Table 3), the star formation efficiency thus defined is 0.6 percent. In the second, we define the SFE more strictly for the current epoch, i.e., counting only the mass of protostars and of dense gas (that in our mask 2 region). The SFE thus defined in this more restricted sense is about 0.3 percent. For the third method, we adopt a less physically motivated but procedurally simple approach of defining the star formation efficiency to be the mass of all pre–main sequence stars divided by the mass of dense gas, we obtain an SFE equal to 1.2 percent. These low values confirm that Taurus is a region of relatively low star formation efficiency. Since star formation is an ongoing process in Taurus the SFE as defined will evolve with time. A more meaningful quantity is the star formation rate per unit molecular gas mass. The star formation history of Taurus is a topic of some controversy (cf. Palla & Stahler, 2000; Hartmann et al., 2001; Palla & Stahler, 2002), particularly the issue regarding whether the star formation rate is presently accelerating or has already reached a peak and is declining. Nevertheless, there does seem to be agreement that star formation has been rapid. Star formation in Taurus began over 10 Myr ago, but most of the identified pre-main sequence stars have formed in the past 3 Myr (Palla & Stahler, 2002). The average star formation rate over the past 3 Myr within the region of Taurus included in this study has been $\simeq$ 8$\times 10^{-5}$ stars yr-1. Assuming as before an average mass of 0.6 solar masses, we derive a star formation rate of 5$\times 10^{-5}$ $M_{\odot}$ yr-1. Thus, the star formation rate per unit molecular gas mass is approximately 2$\times 10^{-9}$ $M_{\odot}$ per year per solar mass of molecular gas. If this rate were to continue, the gas consumption timescale would be over 400 Myr. However, most of the dense gas is likely to be dispersed by the winds from the newly forming stars long before a significant fraction of the cloud mass is converted into stars. It is intriguing that the star formation rate per unit molecular gas mass in Taurus is very similar to that found globally in the Milky Way (assuming a total molecular mass of 2$\times 10^{9}$ $M_{\odot}$ and a star formation rate of 3 $M_{\odot}$ yr-1). ## 7 MORPHOLOGY OF THE MOLECULAR GAS ### 7.1 General Structure of the Gas ### 7.2 Regions of Interest In this section we discuss several of the regions of particular interest that stand out in the carbon monoxide emission from Taurus. These are to some degree reflections of the complex structure seen on a large scale, but highlight some of the varied structures that can easily be identified. The present discussion is by no means complete but does illustrate the varied and complex structures found in this region in which only low mass star formation is taking place. These are grouped together by location within the cloud so that they can be highlighted by detailed images, but this does not necessarily reflect any physical relationship between different features. #### 7.2.1 Filamentary Structure Within the Dense Gas A very striking feature of the molecular gas within the dense portion of Taurus is the fact that the 13CO emission is highly structured even in integrated intensity, as can be seen in Figure 1. An impressive example is shown in Figure 16 which shows a several approximately parallel filaments at 4h27m+26∘45′, having a southeast to northwest orientation. The filaments are $\simeq$ 20’ to 25’ (0.8 pc to 1.0 pc) long, with a $\simeq$ 6:1 length to width ratio. These filaments are readily visible in individual velocity images (Narayanan et al., 2007) as well as the 13CO integrated intensity image, but are invisible in the 12CO data. The peak H2 column density of the filaments is 3$\times$1021 cm-2, about a factor of two greater than that of the region between them. Another very interesting feature visible in Figure 16 is the almost complete ring–like structure centered at 4h31m +28∘01′. It is fairly circular, having an angular diameter of 18′, corresponding to 0.73 pc. The molecular hydrogen column density is typically 3$\times$1021 cm-2 around the periphery of the ring and 1.8$\times$1021 cm-2 in the center. This ring shows up quite clearly in the 12CO integrated intensity image in which $\int T_{A}dv$ increases from 7.5 K kms-1 in the center to $\simeq$ 11 K kms-1 on the periphery. This features is not discernible in the 12CO maximum intensity image, indicating that it is showing increased line width, although distinct kinematic structure is not evident. Figure 16: Enlarged image of 13CO integrated intensity showing filamentary structures located near 4h27m+26∘45′. A nearly circular ring–like structure centered at 4h31m+28∘ is also visible. The young stars in this region are indicated by the symbols defined in Fig. 14. The diamonds indicate diffuse or extended young stellar sources, the square indicates a Class I or diffuse source, and the asterisks indicate T-Tauri stars. #### 7.2.2 Cometary Globules and Ring in Large Cavity A structure that appears to be a large cavity is visible at the eastern end of B213, just to the north of B18, visible in the 12CO image, but more clearly in the 13CO integrated intensity (Figure 1). An enlarged image is shown in Figure 17. The center of the cavity is approximately 4h29m+25∘30′. Although the cavity is still clearly visible, it is considerably smaller, 40′ (1.6 pc) in 12CO compared to 70′ (2.9 pc) in 13CO. The minimum H2 column density of the cavity is 1.4$\times$1021 cm-3 (it is included in mask 1), but the 13CO is detected when averaged over a reasonable number of pixels. Figure 17: This enlarged 13CO integrated intensity image shows the cavity between B213 and B18. At the northwest, three cometary globules are evident, having properties presented in Table 6 and discussed in the text. A nearly complete ring is also seen at the eastern edge of the cavity. The diamonds indicate diffuse or extended young stellar sources, the squares indicate Class I or younger stars, and the asterisks indicate T-Tauri stars. The boundary of this cavity contains an impressive number of young stars, which in fact nearly completely surround it. To the north, these seem to be distributed around the periphery of the cavity, but at its western edge (the eastern end of B213), there are three prominent condensations, looking remarkably like cometary globules, projecting into the cavity. Some properties of the condensations are shown in Table 6. The globules are undistinguished in terms of maximum 12CO temperature. The maximum column density of each of the globules is close to 4$\times$1021 cm-2. We have not been able to identify any source that would be responsible for forming the cavity, but this may be a result of its relatively great age. Table 6: Properties of Cometary Globules In B213–B18 Cavity Globule Number | RA(J2000) | Decl(J2000) | Mass | Embedded Star ---|---|---|---|--- | | | $M_{\odot}$ | 1 | 4h26m49s.8 | 25∘39′06′′ | 5.9 | DF Tau 2 | 4h27m06s.3 | 26∘06′07′′ | 8.3 | DG Tau | | | | FV Tau 3 | 4h29m25s.5 | 26∘14′42′′ | 4.1 | FW Tau As indicated in Table 6 (see also Figure 14), each of the globules contains a T Tauri star, with Globule 2 containing two stars. DF Tau is located slightly inwards (toward the cavity center) relative to Globule 1, while the stars in Globules 2 and 3 are located 3′ away from the cavity center compared to the tip of the globule. There does not appear to be any readily discernible kinematic signature giving clues to the origin of the globules, or revealing an effect of the star formation. For example, although the star DG Tau B in Globule 2 has an optical jet which is presumed to be driving the observed red–shifted molecular outflow (Mitchell, Sargent, & Mannings, 1997), we do not see an effect on the quiescent gas distribution. The stars in question range from 0.2 $L_{\odot}$ to 2.2 $L_{\odot}$, and have ages between 0.6 Myr (DG Tau) to 1.2 Myr (FW Tau). Stars of this age may well have moved a significant distance since their formation, so that it is not surprising that if they were formed in these globules by e.g. radiative implosion (Bertoldi & McKee, 1990), they may now appear displaced from their formation sites. #### 7.2.3 Irregular Filament or Boundary in L1536 Figure 18: Filament running to northwest from L1536 seen in 13CO integrated emission. The condensations have a maximum H2 column density of 5$\times$1021 cm-2. The diamonds indicate diffuse or extended young stellar sources and the asterisk indicates a T-Tauri star. A very long filament having one end in the south–central portion of L1536 and extending to the northwest is visible in the 13CO emission, shown in an enlarged view in Figure 18. The filament center is at 4h23m +23∘45′, and its length is 2∘, corresponding to 4.9 pc. The morphology of the filament is suggestive of its being a boundary between regions of lower (to the south) and higher (to the north) column density. The form of the filament is somewhat suggestive of a helix, but it could simply have an irregular shape. The H2 column density along the filament is typically 3$\times$1021 cm-2, but reaches 5$\times$1021 cm-2 in the regions of strongest emission. The region surrounding the filament has a H2 column density of 1.3 to 1.5 $\times$1021 cm-2, only slightly greater than our minimum value defined by mask 0 of 1.1$\times$1021 cm-2. This filament, is roughly parallel to the structure formed by B18 and L1506, to the filamentary part of B213, and also to the less well–defined but still quite flattened structure formed by Heiles’ Cloud 2 and L1521. This thin filament is the most southerly and furthest from the Galactic plane of all of these structures. The position angle of all four of these filamentary/elongated clouds is approximately 45∘ relative to the plane of the Milky Way. #### 7.2.4 Molecular Ring and Planar Boundary Figure 19 includes several different structures. The first is the “molecular ring”, studied in detail by Schloerb & Snell (1984). This ring, 30′ (1.2 pc) in diameter, centered at 4h40m30s +25∘45′, contains at least 6 dense condensations visible in the 13CO integrated intensity image. The best–studied of these is the chemically very interesting TMC-1 ridge, observed in detail by Pratap et al. (1997) and many others. The ridge (the NH3 peak is at 4h41m21s+25∘48′) is not very prominent in the 13CO integrated intensity image, which is presumably a result of the significant optical depth in the ring material, which may not be corrected for entirely by the simple process (described in §3.2.1) employed here. The peak H2 column density we derive is 7$\times$1021 cm-2 which is somewhat less than half of that which would be derived from the C18O observations of Pratap et al. (1997). Given the difficulties expected in deriving the column density in regions of optically thick emission in which significant temperature gradients may be present, this difference is not unreasonable. The ridge is more visible in our 12CO map than in that of Schloerb & Snell (1984) due to the better sampling in the present work. Figure 19: 13CO integrated intensity image showing the molecular ring in Heiles’ Cloud 2 centered at 4h40m30s +25∘45′. We also see the very well–defined planar boundary centered at 4h38m30s +26∘50′. The 12CO boundary in this region is not at all sharp, having a hair-like appearance as seen most clearly in Figure 13. The diamonds indicate diffuse or extended young stellar sources, the squares indicate Class I or younger stars, and the asterisks indicate T-Tauri stars. The second noticeable feature in Figure 19 is the very straight boundary of the molecular emission seen in 13CO centered at 4h38m30s +26∘50′ and extending for over 1 degree (2.4 pc). The questions of the formation of this interface and how it is maintained are intriguing. In this region, the 12CO emission extends significantly beyond that of the 13CO away from the high column density portion of the cloud, typically by 0.5 pc. As can be seen in Figures 2 and 13, the 12CO emission is highly structured, particularly perpendicular to the interface direction. This behavior is not restricted to this portion of the cloud boundary, but in fact is a general characteristic of the 12CO emission in the mask 1 region surrounding the high column density portion of the cloud (mask 2) where 13CO is detected in individual spectra. Finally, we note the intriguing feature to the west of the better–known ring discussed above. With a center at 4h37m+26∘45′, this is again a slightly non circular ring having a diameter of 30′ (1.2 pc). Given the complexity of the structure observed in our study of the molecular gas in Taurus, this could certainly be a superposition of filaments rather than a ring. #### 7.2.5 L1495 and B213 The L1495 region contains the greatest concentration of young stars within the region of the Taurus molecular cloud that we have mapped. Figure 20 shows the eastern part of L1495; the western part (seen in Figure 1) is more diffuse. The enlarged image also shows the very narrow B213 filament which extends to the southeast from L1495. The 13CO emission and the H2 column density we derive from it, are relatively continuous over the high column density portion of L1495 and the B213 filament. In C18O (Onishi et al., 1996) individual dense cores are better resolved, and in HCO+ (Onishi et al., 2002) they stand out yet more clearly. The central part of of L1495 contains over 20 young stars in Palla’s compilation (Palla, 2008), and has a maximum H2 column density of 1022 cm-2, which is the highest we see in our map. The mass of the L1495 region is (Table 4) 2.6$\times$103 $M_{\odot}$, but a significant fraction of this is in the spatially extended, lower density material. The B213 filament is approximately 75′ or 3 pc in length, and only 4.5′ or 0.2 pc thick. One of the curious features about this structure is that while there are dense cores seen along its entire length (Onishi et al., 1996, 2002), young stars have apparently not yet formed in the northwestern 30′ (1.2 pc) long portion closest to L1495. The magnetic field orientation at the boundaries of this filament is strikingly oriented perpendicular to its long axis, as seen dramatically in Figure 13, and discussed in §5. Figure 20: Enlarged 13CO integrated intensity image of the western portion of L1495 (4h17m30s +28∘20′) and the filamentary region B213 (running to the southeast from 4h17m30s +27∘40′. The diamonds indicate diffuse or extended young stellar sources, the squares indicate Class I or younger stars, and the asterisks indicate T-Tauri stars. #### 7.2.6 Striations in 12CO Emission One of the surprising features in the map of 12CO is the prominent striations (or threads, or strands) seen in the lower level emission seen away from the main molecular condensations. These can be recognized in Figure 2, but this effect is more visible in the enlarged image shown in Figure 21. Another region in which this is very prominent is located at 4h15m +24∘30′. These are similar to structures seen within some infrared cirrus clouds. The striations are visible in images of maximum antenna temperature and also integrated antenna temperature. The characteristic values are $T_{A}$ = 3 K on the striations and 2 K between them, while $\int T_{A}dv$ drops from $\simeq$ 2.8 K kms-1 on the striations to between 1 and 1.5 K kms-1 between them. Given that the density in these regions is low, the 12CO emission is almost certainly subthermally excited so that it is difficult to determine the kinetic temperature. Based on the procedure described in §3.2.1, which assumes $T_{kin}$ equal to 15 K, the H2 column density of the striated features is 2$\times$1021 cm-2, approximately double that of the background emission. A striking feature of the striations is their alignment parallel to the direction of the magnetic field measured by optical starlight polarization, as shown in Figure 13 and discussed in §5. Figure 21: One of the regions showing prominent striations in the 12CO emission, displayed here in an image of the maximum 12CO antenna temperature. The diamonds indicate diffuse or extended young stellar sources, the squares indicate Class I or younger stars, and the asterisks indicate T-Tauri stars. ## 8 SUMMARY AND CONCLUSIONS We have carried out a large–scale survey of the molecular gas in Taurus by mapping a 100 square degree region with the 13.7 m Five College Radio Astronomy Observatory millimeter telescope. The J = 1 $\rightarrow$ 0 transition of 12CO and of 13CO were observed simultaneously using the 32 pixel Sequoia focal plane array receiver. The observing and data reduction techniques are discussed by Narayanan et al. (2007). In this overview, we have discussed some of the highlights of the data that we have obtained, deferring detailed analyses to future papers. The combination of an unbiased, high sensitivity survey with coverage of a relatively large area allows us to study the structure and properties of the molecular gas in new ways. With approximately 3 million independent spatial pixels, we have a linear dynamic range which is unequaled in previous studies of the Taurus region. While our angular resolution is inferior to that obtained with larger/higher frequency telescopes or interferometers, the strength of the present work is to show the relationship between structures on scales ranging from $\simeq$ 1′ or 0.04 pc to 10 degrees (approximately 25 pc). Our observations are sensitive to a range of column densities equivalent to a range in visual extinction between 1 and 10 magnitudes. Cloud Morphology One of our key conclusions is that the morphology of this region is very complex. In contrast to earlier large–scale surveys carried out with low angular resolution in which clouds appeared largely smooth–edged and having little structure, we find an astoundingly rich range of structures including filaments, ridges, blobs, and holes. The internal structure is more striking in 13CO than in 12CO which is not surprising given the large optical depth of the former isotopologue. The filaments have lengths up to 3 pc, and axial to transverse dimension ratios as large as 15:1. Holes in the molecular emission appear on a large range of scales extending from 0.1 pc to 3 pc. The edges of the dense molecular regions are generally very irregular, with structures on the order of 0.1 pc in size visible especially in 12CO which traces cloud boundaries which are more extended than seen in the 13CO. This “hair–like” edge structure is found to be common in 12CO while the 13CO cloud boundaries are relatively sharper but still quite irregular. There is one notable exception in which we find a sharp, straight boundary in 13CO almost 2.5 pc in length. Cloud Mass and Mass Distribution Having both the 12CO and 13CO detected in regions of relatively large column density (mask 2, comprising about 1/3 of the map pixels), we have used the standard method to derive the kinetic temperature and molecular column density, including a correction for saturation of the 13CO which becomes significant for the regions of greatest column density. To analyze portions of the image (mask 1 comprising about 1/3 of total area of the cloud mapped) in which we detect 12CO but not 13CO in individual pixels we use a different approach. With $\simeq$ 1 million such pixels available, we have binned them by 12CO excitation temperature $T_{ex}$. When spectra within a bin are averaged, the 13CO as well as the 12CO is readily detectable, and we obtain the H2 density and the CO column density. We thus have a relationship which gives us n(H2) and N(CO) as a function of $T_{ex}$(12CO). Since the excitation temperature is available for each pixel, we can derive the CO column density for each line of sight. Averaging together all the pixels in mask 0 (in any one of which neither 12CO nor 13CO was detectable), we detect both isotopologues, and use the two spectra to derive the average density and column density for mask 0, the final third of the map. This procedure allows us to determine the CO column density throughout the region mapped, including even regions of relatively low column density. To convert $N(CO)$ to total column density, we have used the results of Van Dishoeck & Black (1988) which are appropriate for Taurus. The essential point is that the fractional abundance of carbon monoxide drops as the total H2 column density is reduced, as a result of reduced dust shielding and self–shielding. Inverting this argument, the column density of H2 corresponding to a low column density of carbon monoxide is larger than would be obtained assuming a constant fractional abundance for CO. The result is that the total mass for the region of Taurus mapped is close to 2.4$\times$104 $M_{\odot}$, compared to less than 1$\times$104 $M_{\odot}$ that would be found using a standard, uniform fractional abundance. We find that half the mass of the cloud is contained in regions having column density below 2.1$\times$1021 cm-2. This result reduces the fraction of mass found in dense cores by a factor greater than 2, and also confirms the presence of significant external pressure in the regions external to the dense regions. The total mass for the region we have mapped thus obtained agrees well with that predicted from the CO luminosity, 5.55$\times$103 K km s-1 pc2, and a standard conversion M($M_{\odot}$) = 4.1 LCO (K km s-1 pc2). It seems likely that our conclusion that a significant component of diffuse molecular gas accompanies the more widely studied high density regions is not restricted to Taurus. It reinforces the importance of observations which can study this diffuse molecular material, which is not readily detected in individual spectra with the sensitivity typically available in large–scale molecular cloud surveys. Cloud Structure and Star Formation The structural complexity over a wide range of scale sizes hints at the richness of the physical processes which underly the formation and evolution of molecular cloud complexes such as Taurus. The present data set, both in terms of morphology and mass distribution, constitutes a potentially valuable resource for comparison with outputs from simulations of cloud formation. The large scale kinematic structure that we see confirms that identified in earlier studies. Along with the complexity of the line profiles observed along many lines of sight, this poses a real challenge for any detailed theoretical model of this region. We see a varied relationship between the magnetic field as measured by polarization of background stars, and the distribution of the gas. In the more diffuse regions traced by 12CO we see large–scale alignment between the field direction and striated structure in the gas. Although we have not been able to measure any kinematic signature, the appearance is strongly suggestive of flows along the field lines. In several of the very elongated filaments seen in the denser gas traced by 13CO, the magnetic field is oriented perpendicular, or nearly perpendicular, to the major axes of the filaments. Combined with the hair–like appearance of the boundaries of these filaments seen in 13CO but more prominent in 12CO, this again suggests that motions of material along the field lines have been responsible for building up the regions of higher density within the overall molecular cloud. The surface density of very young and moderately young stars shows a rapid increase at a H2 column density of 6$\times$1021 cm-2, confirming the existence of a threshold for star formation. We have used new compilations of young stars in the Taurus region to calculate the star formation efficiency (SFE). Our large value for the gas mass, especially in regions of lower column density, results in the SFE, taken to be the mass of all young stars in the region divided by the total molecular mass, to be 0.6 percent. Taking the SFE for most recent star formation by comparing the mass of only the embedded protostars with that of the dense gas, gives an SFE equal to 0.3 percent. If we consider all of the young stars (whether embedded protostars or T-Tauri stars) in the region of high column density, we obtain a SFE equal to 1.2 percent. The average star formation rate over the past 3 Myr within the region of Taurus included in this study has been $\simeq$ 8$\times 10^{-5}$ stars yr-1, corresponding to a mass going into new stars of 5$\times 10^{-5}$ $M_{\odot}$ yr-1. This work was supported in part by the National Science Foundation through grant AST-0407019 to Cornell University, and by the Jet Propulsion Laboratory, California Institute of Technology. The Five College Radio Astronomy Observatory is operated with support from the National Science Foundation through NSF grant AST 05 40852 and with permission of the Metropolitan District Commission. We thank Yvonne Tang for contributions to data taking and analysis of dense condensations in Taurus, and Marko Krco for assistance with observations. We thank Pierre Hily–Blant for the suggestion to compare the magnetic field and integrated intensity maps in Taurus, and for many useful conversations about this and other topics. We are indebted to Francesco Palla and Scott Kenyon for providing compilations of young stars in the Taurus region and their properties. We thank Ted Bergin for carrying out time–dependent calculations of the CO abundance in diffuse regions. We thank the anonymous reviewer for very carefully reading the lengthy manuscript, noting some problems, and making some suggestions for further work which has improved this study. This research has made use of NASA’s Astrophysics Data System. ## References * Abergel et al. (1995) Abergel, A., Boulanger, F., Fukui, Y., & Mizuno, A. 1995, A&A, 111, 483 * Alves, Lada, & Lada (1999) Alves, J., Lada, C.J., & Lada, E.A. 1999, ApJ, 515, 265 * Bachiller & Cernicharo (1986) Bachiller, R. & Cernicharo, J. 1986, A&A, 168, 262 * Ballesteros-Paredes, Hartmann, & Vázquez–Semadeni (1999) Ballesteros–Paredes, J., Hartmann, L., & Vázquez–Semadeni, E. 1999, ApJ, 527, 285 * Bally & Langer (1982) Bally, J. & Langer, W.D. 1982, ApJ, 255, 143 * Beichman et al. (1986) Beichman, C.A., Myers, P.C., Emerson, J.P., Harris, S., Mathieu, R., Benson, P.J., & Jennings, R.E. 1986, ApJ, 307, 377 * Bertoldi & McKee (1990) Bertoldi, F. & McKee, C.F. 1990, ApJ, 354, 529 * Burgh et al. (2007) Burgh, E.B., France, K., & McCandliss, S.R. 2007, ApJ, 658, 446 * Burton & Hartmann (1994) Burton, W.B. & Hartmann, D. 1994, in Unveiling Large–Scale Structures Behind the Milky Way, ASP Conf. Series, Vol. 67, C. Balkowski & R.C. Kraan-Kortweg eds. (San Francisco: Astronomical Society of the Pacific), 31 * Cernicharo & Guélin (1987) Cernicharo, J., & Guélin, M. 1987, A&A, 176, 299 * Chu & Watson (1983) Chu, W.–H. & Watson, W.D. 1983, ApJ, 267, 151 * Crutcher & Troland (2000) Crutcher, R. M., & Troland, T.H. 2000, ApJ, 537, L139 * Dickman, Snell, & Schloerb (1986) Dickman, R.L., Snell, R.L., & Schloerb, F.P. 1986, ApJ, 309, 326 * Duvert, Cernicharo, & Baudry (1986) Duvert, G., Cernicharo, J., & Baudry, A. 1986, A&A, 164, 349 * Elias (1978) Elias, J.H. 1978, ApJ, 224, 857 * Erickson et al. (1999) Erickson, N.R., Grosslein, R.M., Erickson, R.B., & Weinreb, S. 1999, IEEE Trans. Microwave Theory Tech., 47(12), 2212 * Federman et al. (1994) Federman, S.R., Strom, C.J., Lambert, D.L., Cardelli, J.A., Smith, V.V., & Joseph, C.L. 1994, ApJ, 424, 772 * Frerking, Langer, & Wilson (1982) Frerking, M.A., Langer, W.D., & Wilson, R.W. 1982, ApJ, 262, 590 * Goldsmith et al. (1980) Goldsmith, P.F., Langer, W.D., Carlson, R.E., & Wilson, R.W. 1980, in Interstellar Molecules, IAU Symp. 87, B.H. Andrew ed. (Dordrecht: Reidel), 417 * Goldsmith, Young, & Langer (1983) Goldsmith, P.F., Young, J.S., & Langer, W.D. 1983, ApJS, 51, 203 * Goldsmith & Li (2005) Goldsmith, P.F. & Li, D. 2005, ApJ, 622, 938 * Gomez de Castro & Pudritz (1992) Gomez de Castro, A.I. & Pudritz, R.E. 1992, ApJ, 395, 501 * Goodman et al. (1992) Goodman, A.A., Jones, J.T., Lada, E.A., & Myers, P.C. 1992, ApJ, 399, 108 * Hartigan & Kenyon (2003) Hartigan, P., & Kenyon, S.J. 2003, ApJ, 583, 334 * Hartmann et al. (2001) Hartmann, L., Ballesteros–Paredes, J., & Bergin, E. A. 2001, ApJ,562, 852 * Hartmann (2002) Hartmann, L. 2002, ApJ, 578, 914 * Heiles (2000) Heiles, C. 2000, AJ, 119, 923 * Heiles & Crutcher (2005) Heiles, C. & Crutcher, R. 2005, in Cosmic Magnetic Fields, R. Wielebinski & R. Beck ed. (Berlin: Springer), 137 * Heyer et al. (1987) Heyer, M.H., Vrba, F.J., Snell, R.L., Schloerb, F.P., Strom, S.E., Goldsmith, P.F., & Strom, K.M. 1987, ApJ, 321, 855 * Heyer (1988) Heyer, M.H. 1988, ApJ, 324, 311 * Jijina, Myers, & Adams (1999) Jijina, J., Myers, P.C., & Adams, F.C. 1999, ApJS, 125, 161 * Kainulainen et al. (2006) Kainulainen, J., Lehtinen, K., & Harju, J. 2006, A&A, 447, 597 * Kenyon (2007) Kenyon, S. 2007, private communication; to appear in The Handbook of Star Forming Regions, ASP Conference Series, B. Reipurth, ed., 2008 * Lacy et al. (1994) Lacy, J.H., Knacke, R., Geballe, T.R., & Tokunaga, A.T. 1994, ApJ, 428, L69 * Langer et al. (1980) Langer, W.D., Goldsmith, P.F., Carlson, E.R., & Wilson, R.W. 1980, ApJ, 235, L39 * Langer et al. (1989) Langer, W.D., Wilson, R.W., Goldsmith, P.F., & Beichman, C.A. 1989, ApJ, 337, 355 * Langer & Penzias (1993) Langer, W.D. & Penzias, A.A. 1993, ApJ, 408, 539 * Langer et al. (1995) Langer, W.D., Velusamy, T., Kuiper, T.B.H., Levin, S., Olsen, E., & Migenes, V. 1995, ApJ, 453, 293 * Li & Goldsmith (2003) Li, D. & Goldsmith, P.F. 2003, ApJ, 585, 823 * Li, Goldsmith, & Menten (2003) Li, D., Goldsmith, P.F., & Menten, K.M. 2003, ApJ, 587, 262 * Liszt & Lucas (1998) Liszt, H.S. & Lucas, R. 1998, A&A, 339, 561 * Liszt (2007) Liszt, H.S. 2007, A&A, 476, 291 * Mitchell, Sargent, & Mannings (1997) Mitchell, G.F., Sargent, A.I., & Mannings, V. 1997, ApJ, 483, L127 * Mizuno et al. (1995) Mizuno, A., Onishi, T., Yonekura, Y., Nagahama, T., Ogawa, H., & Fukui, Y. 1995, ApJ, 445, L161 * Moneti et al. (1984) Moneti, A., Pipher, J.L., Helfer, H.L., McMillan, R.S., & Perry, M.L. 1984, ApJ, 282, 508 * Narayanan et al. (2007) Narayanan, G., Heyer, M., Brunt, C., Snell, R.L., Goldsmith, P.F., & Li, D. 2007, submitted to ApJ * Onishi et al. (1996) Onishi, T., Mizuno, A., Kawamura, A., Ogawa, H., & Fukui, Y. 1996, ApJ, 465, 815 * Onishi et al. (1998) Onishi, T., Mizuno, A., Kawamura, A., Ogawa, H., & Fukui, Y. 1998, ApJ, 502, 296 * Onishi et al. (2002) Onishi, T., Mizuno, A., Kawamura, A., Tachihara, K., & Fukui, Y. 2002, ApJ, 575, 950 * Padoan et al. (2002) Padoan, P., Cambrésy, L., & Langer, W.D. 1992, ApJ, 580, L57 * Palla & Stahler (2000) Palla, F. & Stahler, S.W. 2000, ApJ, 540, 255 * Palla & Stahler (2002) Palla, F. & Stahler, S.W. 2002, ApJ, 58, 1194 * Palla (2008) Palla, F. 2008, private communication * Pratap et al. (1997) Pratap, P., Dickens, J.E., Snell, R. L., Miralles, M.P., Bergin, E.A., Irvine, W.M., & Schloerb, F.P. 1997, ApJ, 486, 862 * Pringle, Allen, & Lubow (2001) Pringle, J.E., Allen, R.J., & Lubow, S.H. 2001, MNRAS, 327, 663 * Schloerb & Snell (1984) Schloerb, F.P., & Snell, R.L. 1984, ApJ, 283, 129 * Shu et al. (1987) Shu, F.H., Adams, F.C., & Lizano, S. 1987, ARA&A, 25, 23 * Shuter et al. (1987) Shuter, W.L.H., Dickman, R.L., & Klatt, C. 1987, ApJ, 322, L103 * Snell (1981) Snell, R.L. 1981, ApJS, 45, 121 * Sonnentrucker et al. (2007) Sonnentrucker, P., Welty, D.E., Thorburn, J.A., & York, D.G. 2007, ApJS, 168, 58 * Strong & Mattox (1996) Strong, A.W. & Mattox, J.R. 1996, A&A, 308, L21 * Tamura et al. (1991) Tamura, M., Gatley, I., Wall, W., & Werner, M.W. 1991, ApJ, 374, L25 * Tatematsu et al. (2004) Tatematsu, K., Umemoto, T., Kandori, R., & Sekimoto, Y. 2004, ApJ, 606, 333 * Troland et al. (1996) Troland, T., Crutcher, R.M., Goodman, A.A., Heiles, C., Kazès, I., & Myers, P.C. 1996, ApJ, 471, 302 * Ungerechts & Thaddeus (1987) Ungerechts, H. & Thaddeus, P. 1987, ApJS, 63, 645 * Van Dishoeck & Black (1988) Van Dishoeck, E.F. & Black, J.H. 1988, ApJ, 334, 771 * Vázquez–Semadeni (2007) Vázquez–Semadeni, E. 2007, in Triggered Star Formation in a Turbulent ISM, Proc. IAU Symp. 237, B.G. Elmegreen & J. Palous, eds. (Cambridge: Cambridge University Press), 292 * Watson et al. (1976) Watson, W.D., Anicich, V.G., & Huntress, W.T. Jr. 1976, ApJ, 205, L165 * Young et al. (1982) Young, J.S., Goldsmith, P.F., Langer, W.D., & Wilson, R.W. 1982, ApJ, 261, 513

arxiv-papers

2024-09-04T02:48:53.773995

0802.2211

# Boundary effects on the dynamics of chains of coupled oscillators D. Bambusi, A. Carati, T. Penati Università degli Studi di Milano, Dipartimento di Matematica “F. Enriques”, Via Saldini 50, 20133 Milano, Italy ###### Abstract We study the dynamics of a chain of coupled particles subjected to a restoring force (Klein-Gordon lattice) in the cases of either periodic or Dirichlet boundary conditions. Precisely, we prove that, when the initial data are of small amplitude and have long wavelength, the main part of the solution is interpolated by a solution of the nonlinear Schrödinger equation, which in turn has the property that its Fourier coefficients decay exponentially. The first order correction to the solution has Fourier coefficients that decay exponentially in the periodic case, but only as a power in the Dirichlet case. In particular our result allows one to explain the numerical computations of the paper [BMP07]. ## 1 Introduction. The dynamics of chains of coupled particles has been the object of a huge number of studies, but only recently some numerical works (see [BG08, BMP07]) have shown that the boundary conditions have some relevance on FPU type investigations. The goal of the present paper is to study analytically the effects of the boundary conditions (BC) on the dynamics of a simple 1-dimensional model, namely the so called Klein Gordon lattice (coupled particles subjected to an on site restoring force). Precisely, we concentrate on the cases of periodic and of Dirichlet boundary conditions, and use the methods of normal form to study the dynamics. This leads to a quite clear understanding of the role of the boundary conditions and to an explanation of the numerical results of [BMP07]. On the contrary, our theory does not allow one to explain the results of [BG08]. More precisely, we study the dynamics of a large lattice corresponding to small amplitude initial data with long wavelength; we show that if the size $N$ is large enough and the amplitude $\epsilon$ of the initial excitation is of order $\mu:=\frac{1}{N}$, then the solution $z$ has the form $z=\mu z^{a}(t)+\mu^{2}z_{1}(t)\ ,$ (1.1) up to times $|t|\leq\mathcal{O}(\mu^{-2})$. In (1.1) $z^{a}$ is interpolated by a solution of the nonlinear Schrödinger equation (NLS) and has a behaviour which is independent of the BC. On the contrary $z_{1}$ depends on the BC. Precisely, its Fourier coefficients decrease exponentially in the periodic case, but only as $|k|^{-3}$ in the Dirichlet case. The theory we develop in order to give the representation (1.1) provides a clear interpretation of the phenomenon. Indeed, it turns out that the normal form of the system is independent of the BC (and coincides with the NLS), but the coordinate transformation introducing the normal form has properties which are different in the periodic and in the Dirichlet case. In particular, in the Dirichlet case it maps sequences which decay fast into sequences which decay as $|k|^{-3}$. This introduces the slow decay in the Dirichlet case. It should be pointed out that our result still depends on the size $N$ of the lattice.111However our normal form holds in the region of the phase points with small energy density $\epsilon^{2}$, independently of N. The limitation $\epsilon\sim\mu$ comes from the fact that we are only able to study the dynamics of the NLS in this situation. Nevertheless, we think that (within the range of validity of our result) we clearly show the role of the boundary conditions and provide a good interpretation of the numerical results. The present situation has many similarities with the one occurring in the theory of the Navier Stokes equation (see e.g. [Tem91]), where it is well known that the spectrum of the solution depends on the boundary conditions. Moreover, we recall that a power law decay of localized object has been previously observed in nonlinear lattice dynamics in [DP03, Pey04, Fla98, GF05] and that the connection between the nonlinear Schrödinger equation and the dynamics of long chains of particles was studied in many papers (see e.g. [Kal89, KSM92, Sch98]). The paper is organized as follows: in sect. 2 we present the model, state our main result and discuss its relation with numerical computations. In sect. 3 we give the proof of the normal form construction. In sect. 4 we give the proof of the decomposition (1.1). Some technical details are deferred to the appendix. Each section is split into several subsections. Acknowledgments. This work was partially supported by MIUR under the project COFIN2005 “Sistemi dinamici nonlineari e applicazioni fisiche”. We thank Antonio Ponno, Luigi Galgani and Simone Paleari for many discussion and suggestions during the preparation of this paper. We also thank Sergej Flach and Michel Peyrard for some interesting discussion. ## 2 Main result In this chapter we present the model, we recall some numerical simulations (see [BMP07]) which clearly show the dependence of the metastable Fourier decay on the boundary conditions and we finally state the main theoretical result and use it to explain the numerics. ### 2.1 The model. We consider a chain of particles described by the Hamiltonian function $\displaystyle H({p},{q})$ $\displaystyle=$ $\displaystyle\sum_{j}\frac{p_{j}^{2}}{2}+\sum_{j}V(q_{j})+\sum_{j}W(q_{j}-q_{j-1})$ (2.1) where $j$ runs from $0$ to $N$ in the case of Dirichlet boundary conditions (DBC), namely $q_{0}=q_{N+1}=0$, while it runs from $-(N+1)$ to $N$ in the case of periodic boundary conditions (PBC), i.e. $q_{-N-1}=q_{N+1}$. The corresponding Hamilton equations are $\ddot{q}_{j}=-V^{\prime}(q_{j})-W^{\prime}(q_{j}-q_{j-1})+W^{\prime}(q_{j+1}-q_{j})\ .$ (2.2) We recall that the standard Fermi Pasta Ulam model is obtained by taking $V\equiv 0$ and $W(x)=\frac{x^{2}}{2}+\alpha\frac{x^{3}}{3}+\beta\frac{x^{4}}{4}$. Here instead we will take $V(x)=\frac{1}{2}x^{2}+\frac{1}{3}\alpha x^{3}+\frac{1}{4}\beta x^{4}\ ,\quad W(x)=\frac{1}{2}ax^{2}\ ,\ \alpha,\beta,a\geq 0\ .$ (2.3) Explicitly our Hamiltonian has the form $\displaystyle H$ $\displaystyle=$ $\displaystyle H_{0}+H_{1}+H_{2},$ (2.4) $\displaystyle H_{0}(p,q)$ $\displaystyle:=$ $\displaystyle\sum_{j}\frac{p_{j}^{2}+q_{j}^{2}}{2}+a\frac{(q_{j}-q_{j-1})^{2}}{2},$ (2.5) $\displaystyle\quad H_{1}(p,q)$ $\displaystyle:=$ $\displaystyle\alpha\sum_{j}\frac{q_{j}^{3}}{3}\ ,\quad H_{2}(p,q):=\beta\sum_{j}\frac{q_{j}^{4}}{4}\ .$ (2.6) ###### Remark 2.1. In the case where $V(x)=V(-x)$ (2.7) the equations (2.2) with PBC are invariant under the involution $q_{j}\mapsto- q_{-j}$, $p_{j}\mapsto-p_{-j}$. As a consequence the submanifold of the periodic sequences which are also skew-symmetric, is invariant under the dynamics. For this reason when (2.7) is fulfilled the case of DBC is just a subcase of the case of PBC. This happens in the standard FPU model and also in the case of the Hamiltonian (2.1) with the potential (2.3) and $\alpha=0$. The case $\alpha\not=0$ is the simplest one where a difference between DBC and PBC is possible. Consider the vectors ${\hat{e}}_{k}(j)=\begin{cases}\frac{\delta_{PD}}{\sqrt{N+1}}\sin{\left(\frac{jk\pi}{N+1}\right)},\qquad k=1,\ldots,N,\cr\frac{1}{\sqrt{N+1}}\cos{\left(\frac{jk\pi}{N+1}\right)},\qquad k=-1,\ldots,-N,\cr\frac{1}{\sqrt{2N+2}},\qquad\qquad\qquad k=0,\cr\frac{(-1)^{j}}{\sqrt{2N+2}},\qquad\qquad\qquad k=-N-1,\cr\end{cases}$ (2.8) then the Fourier basis is formed by $\hat{e}_{k}$, $k=1,\ldots,N$ and $\delta_{PD}=\sqrt{2}$ in the case of DBC, and by $\hat{e}_{k}$, $k=-N-1,\ldots,N$ and $\delta_{PD}=1$ in the case of PBC. Here we will treat in a unified way both the cases of DBC and PBC, thus we will not specify the set where the indexes $j$ and $k$ vary. Introducing the rescaled Fourier variables $(\hat{p}_{k},\hat{q}_{k})$ defined by $\displaystyle p_{j}=\sum_{k}{\sqrt{\omega_{k}}{\hat{p}}_{k}{\hat{e}}_{k}({j})},\qquad q_{j}=\sum_{k}{\frac{{\hat{q}}_{k}}{\sqrt{\omega_{k}}}{\hat{e}}_{k}({j})},$ (2.9) where the frequencies are defined by $\omega_{k}=\sqrt{1+4a{\sin^{2}{\left(\frac{k\pi}{2N+2}\right)}}},$ (2.10) the Hamiltonian $H_{0}$ is changed to $H_{0}=\sum_{k}{\omega_{k}\frac{{\hat{p}}^{2}_{k}+{\hat{q}}^{2}_{k}}{2}}\ .$ (2.11) ### 2.2 The phenomenon and its numerical evidence. Figure 1: Averaged harmonic energies distribution. DBC (dots) and PBC (crosses) with $N=511,\,a=.5,\,{\mathcal{E}}=0.001,\,T=10^{5}$. Panel (a): $\alpha$ model with $\alpha=0.25$. Panel (b): $\beta$ model with $\beta=0.25$. Figure 2: DBC with parameters $N=511,\,a=0.5,\,\alpha=0.1,\,T=10^{5}$. Panel (a): distribution of ${\langle E_{k}\rangle}$ in semi-log scale. Energy densities: ${\mathcal{E}}=0.05,0.025,0.01,0.005,0.001$. Panel (b): distribution of ${\langle E_{k}\rangle}$ in log-log scale. Energy densities: ${\mathcal{E}}=0.01,0.001,0.0001,0.00001$. Let us define the energy of a normal mode and its time average by $E_{k}:=\omega_{k}\frac{\hat{p}_{k}^{2}+\hat{q}_{k}^{2}}{2},\qquad\qquad\langle E_{k}\rangle(t):=\frac{1}{t}\int_{0}^{t}E_{k}(s)ds\ .$ In the case of PBC the oscillators of index $k$ and $-k$ are in resonance, so the relevant quantity to be observed is the average $\overline{\langle E_{k}\rangle}=\frac{1}{2}(\langle E_{k}\rangle+\langle E_{-k}\rangle)$. Take an initial datum with all the energy concentrated on the first Fourier mode with energy density ${\mathcal{E}}\equiv H_{0}/N=0.001$. Integrating the system numerically one can see that after a short transient the time averages of the harmonic energies relax to well defined steady values, which persist for very long times. In figures 1 we plot in a semi-log scale the time-average energies $\langle E_{k}\rangle(T)$ (or $\overline{\langle E_{k}\rangle}(t)$) at time $T=10^{5}$ (subsequent to the relaxation time) as a function of the index $k$. The parameters in the two panels are $\alpha=0.25,\,\beta=0$ and $\alpha=0,\,\beta=0.25$ respectively. In both the cases $a=0.5$. In each distribution the dots refer to the DBC case while the crosses pertain the PBC one222Actually we plot only modes with odd index, since, as shown in [BMP07] the dynamics involves only modes with odd index t. While in panel (b) one clearly observes a perfect overlapping of the exponential part of the decays, in panel (a) a sharp difference arises. Indeed, while the PBC solution is once more characterized by an exponential distribution, in the case of DBC one sees a richer behavior: at an energy approximately equal to $10^{-8}$ there is crossover and a new regime appears. Nevertheless, a striking similarity among the exponential part of the two dynamics is evident. To describe more carefully the situation in the case of DBC we plot in figures 2 four different distributions of the quantities ${\langle E_{k}\rangle}$, in a semi-log and in a log-log scale respectively. They correspond to different values of the energy density (see the caption). In the first panel we plot the first part of the distribution: we notice that by decreasing the energy density, the slope of the exponential decay of the low frequencies increases. In the second figure, instead, we focus our attention on the second part of the distribution: we see that the corresponding curves are parallel. So a change of energy only induces a translation. Except for the last part, that we will interpret as due to discreteness effects, the curves are very well interpolated by a straight line giving a power decay with an exponent close to $-6$. A similar behavior is also obtained if one excites a few modes of large wave length. ### 2.3 Explanation of the phenomenon. In order to state our main result we need a topology in the phase space. ###### Definition 2.1. Let us define the spaces $\ell^{2}_{s,\sigma}$ of the sequences $p=\\{\hat{p}_{k}\\}$ s.t. $\left\|p\right\|^{2}_{s,\sigma}:=\mu\sum_{k}[k]^{2s}{\rm e}^{2\sigma|k|}\left|\hat{p}_{k}\right|^{2}<\infty\ ,\qquad[k]:=\max\left\\{1,|k|\right\\}\ ,$ (2.12) and the phase spaces ${\mathcal{P}}_{s,\sigma}:=\ell^{2}_{s,\sigma}\times\ell^{2}_{s,\sigma}\ni(p,q)$. The main part of the solution will be described by NLS, so we consider a a smooth solution $\varphi(x,\tau)$ of the nonlinear Schrödinger equation $-{\rm i}\partial_{\tau}\varphi=-\partial_{xx}\varphi+\gamma\varphi|\varphi|^{2}\ ,\quad\gamma:=\frac{3}{8a}\left(\beta-\frac{10}{9}\alpha^{2}\right)\ ,\quad x\in\mathbb{T}:={\mathbb{R}}/2\pi{\mathbb{Z}}$ (2.13) For fixed $\tau$ we will measure the size of a function $\varphi$ by the norm $\left\|\varphi\right\|_{s,\sigma}^{2}:=\sum_{k}[k]^{2s}{\rm e}^{2\sigma|k|}|\hat{\varphi}_{k}|^{2}$ (2.14) where $\hat{\varphi}_{k}$ are the Fourier coefficients of $\varphi$, which are defined by $\varphi(x):=\sum_{k}\hat{\varphi}_{k}\hat{e}^{c}_{k}(x)\ ;$ (2.15) here $\hat{e}^{c}_{k}(x)$ is the continuous Fourier basis, $\hat{e}^{c}_{k}:=\left\\{\begin{matrix}\frac{1}{\sqrt{\pi}}\cos kx&k>0\\\ \frac{1}{\sqrt{2\pi}}&k=0\\\ \frac{\delta_{PD}}{\sqrt{\pi}}\sin(-kx)&k<0\end{matrix}\right.\ .$ (2.16) ###### Remark 2.2. The dynamics of (2.13) is well known [FT87, GK03]. Precisely, if $\tilde{\gamma}<\tilde{\gamma}^{*}$ then $\forall\sigma\geq 0$ there exists $0\leq\sigma^{\prime}<\sigma$ such that from $\left\|\varphi(x,0)\right\|_{s,\sigma}=1$ it follows $\left\|\varphi^{a}(x,t)\right\|_{s,\sigma^{\prime}}\leq C_{s}\sim 1$ for all times. Corresponding to $\varphi(x,\tau)$ we define an approximate solution of the original model by $z^{a}_{j}(t)\equiv\left(p_{j}^{a}(t),q_{j}^{a}(t)\right):=\left(\mathop{\rm Re}\nolimits\left(e^{it}\varphi(\mu j,a\mu^{2}t)\right)\ ,\ \mathop{\rm Im}\nolimits\left(e^{it}\varphi(\mu j,a\mu^{2}t)\right)\right)$ (2.17) Our main result concerns the comparison between $z^{a}(t)$ and the solution $z(t)$ of the original system with initial datum $z(0):=\mu z^{a}(0)$ ###### Theorem 2.1. Assume $\left\|\varphi^{0}(x)\right\|_{s,\sigma}\leq 1$, then $\forall T>0$ there exists $\mu^{*}>0$ with the following properties: if $\mu<\mu_{*}$ then there exists $z_{1}(t)$ defined for $|t|\leq T\mu^{-2}$ such that $z(t)=\mu z^{a}(t)+\mu^{2}z_{1}(t)$ (2.18) where $z^{a}$ is the approximate solution just defined and, $\left\|z_{1}(t)\right\|_{s,\sigma^{\prime}}\leq C$ (2.19) with $\begin{matrix}\frac{1}{2}<s\ ,\quad&\sigma^{\prime}>0&\ \text{if\ PBC}\\\ \frac{1}{2}<s<\frac{5}{2}\ ,&\sigma^{\prime}=0&\ \text{if\ DBC}\end{matrix}$ (2.20) The above result gives an upper estimate of the error $z_{1}(t)$. We want now to compute it, at least for short times. To this end we assume, for simplicity, that $\varphi^{0}$ is purely imaginary, i.e. that the initial datum has zero velocity. Consider now the complex function $\psi_{10}$ with Fourier coefficients given by $(\psi_{10})^{\wedge}_{k}:=\frac{\alpha}{6\sqrt{2}}\left[4e^{{\rm i}\omega_{k}t}-3e^{2{\rm i}t}-e^{-2{\rm i}t}-6{\rm i}e^{{\rm i}\omega_{k}t}+6{\rm i}\right]\hat{\Phi}_{k}$ (2.21) where $\Phi(x):=-{\rm i}\varphi^{0}(x)^{2}$, and let $z_{10}\equiv(p_{10},q_{10})$ be the sequence with $\frac{p_{10,j}+{\rm i}q_{10,j}}{\sqrt{2}}=\psi_{10}(\mu j)\ .$ We have the following ###### Theorem 2.2. For any $0<b<1$, one has $\left\|z_{1}(t)-z_{10}(t)\right\|_{s,\sigma^{\prime}}\leq C\mu^{1-b}\ ,\quad|t|\leq\frac{T}{\mu^{b}}$ (2.22) The above theorems provide the interpretation for the numerical results of the previous paragraph. Referring to FIG. 1, we identify the exponential part of the distribution as due to the main part of the solution, namely $z^{a}$, which is the same for both the boundary conditions. In the case of PBC, the inequality (2.19) implies that also the error $z_{1}$ is exponentially decreasing; thus the whole solution is in particular exponentially localized in Fourier space. In the case of DBC the situation is different. Indeed the correction, namely $z_{1}$, is ensured to have coefficients such that the series with general term $|k|^{2s}|\hat{z}_{1,k}(t)|^{2}$ is convergent; this is very close to say that $|\hat{z}_{1,k}(t)|^{2}<\frac{C}{|k|^{2s+1}}$ which, taking $s$ very close to $5/2$ essentially gives a power law decay like $|k|^{-6}$. Then theorem 2.2 shows that this is actually optimal, as seen by taking $\varphi^{0}(x)={\rm i}\sin x$ (2.23) (as in the numerical computations). Indeed in such a case one has $\hat{\Phi}_{k}=\left(\sin^{2}x\right)^{\wedge}_{k}\sim\frac{1}{k^{3}}$ (2.24) so that (2.22) shows that after a time of order 1 the energy of the $k$-th mode is of order $\mu^{2}/k^{6}$ as shown by the numerics. ## 3 The normal form construction. In this part we introduce and use the methods of the normal form theory for the proof of our main result. Accordingly one looks for a canonical transformation putting the system in a simpler form. ### 3.1 Preliminaries and main claim. We first need to introduce some notations: * • $z$ will denote a phase point. In particular a phase point can be represented using the coordinates $(p_{j},q_{j})$ of the lattice’s particles or the Fourier coordinates $(\hat{p}_{k},\hat{q}_{k})$. * • In the phase space we will also use coordinates $\psi$ defined by $\hat{\psi}_{k}:=\frac{\hat{p}_{k}+{\rm i}\hat{q}_{k}}{\sqrt{2}}$ and, in real space $\psi_{j}:=\sum_{k}{\hat{\psi}}_{k}{\hat{e}}_{k}({j})$ * • Given a Hamiltonian function $H$, we will denote by $X_{H}$ the corresponding Hamiltonian vector field. Thus if one uses for example the variables $(p_{j},q_{j})$, one has $X_{H}(p,q)=\left(-\frac{\partial H}{\partial q_{j}},\frac{\partial H}{\partial p_{j}}\right)\ \quad\text{or}\quad X_{H}(\psi):=\left({\rm i}\frac{\partial H}{\partial\bar{\psi}_{j}}\right)$ Correspondingly we will write the Hamilton equations of a Hamiltonian function $H$ by $\dot{z}=X_{H}(z)\ .$ * • The Lie transform $\Phi^{1}_{\chi}$ generated by a Hamiltonian function $\chi$ is the time one flow of the corresponding Hamilton equations, namely $\Phi^{1}_{\chi}:=\Phi^{t}_{\chi}\big{|}_{t=1}\ ,\quad\frac{d}{dt}\Phi^{t}_{\chi}(z)=X_{\chi}\left(\Phi^{t}_{\chi}(z)\right)\ ,\quad\Phi^{t}_{\chi}\big{|}_{t=0}=\mathbb{I}\ .$ * • The Poisson bracket $\left\\{f,g\right\\}$ of two function $f,g$ is defined by $\left\\{f,g\right\\}:=dfX_{g}\equiv\sum_{j}\left(\frac{\partial f}{\partial q_{j}}\frac{\partial g}{\partial p_{j}}-\frac{\partial f}{\partial p_{j}}\frac{\partial g}{\partial q_{j}}\right).$ The normalizing transformation will be constructed by composing two Lie transforms ${\mathcal{T}}_{1}=\Phi^{1}_{\chi_{1}}$ and ${\mathcal{T}}_{2}=\Phi^{1}_{\chi_{2}}$ generated by two functions $\chi_{1}$ and $\chi_{2}$. Taking $\chi_{1}$ and $\chi_{2}$ to be homogeneous polynomials of degree 3 and 4 respectively, an elementary computations shows that $\displaystyle H\circ{\mathcal{T}}_{1}\circ{\mathcal{T}}_{2}$ $\displaystyle=$ $\displaystyle H_{0}$ (3.1) $\displaystyle+$ $\displaystyle\left\\{\chi_{1},H_{0}\right\\}+H_{1}$ (3.2) $\displaystyle+$ $\displaystyle\left\\{\chi_{2},H_{0}\right\\}+\frac{1}{2}\left\\{\chi_{1},\left\\{\chi_{1},H_{0}\right\\}\right\\}+\left\\{\chi_{1},H_{1}\right\\}+H_{2}$ (3.3) $\displaystyle+$ $\displaystyle{\rm h.o.t.}$ (3.4) where the term (3.2) is a homogeneous polynomial of degree 3, the term (3.3) has degree 4 and h.o.t denotes higher order terms. We will construct a function $\chi_{1}$ such that (3.2) vanishes and we will show that there exists a $\chi_{2}$ such that (3.3) is reduced to normal form. To make precise the construction we need to split $H_{0}$ as follows $\displaystyle H_{0}$ $\displaystyle=$ $\displaystyle H_{00}+H_{01}\ ,\qquad H_{00}(z):=\sum_{k}\frac{\hat{p}_{k}^{2}+\hat{q}_{k}^{2}}{2},$ (3.5) $\displaystyle H_{01}(z)$ $\displaystyle:=$ $\displaystyle\sum_{k}\nu_{k}\frac{\hat{p}_{k}^{2}+\hat{q}_{k}^{2}}{2}\ ,\qquad\nu_{k}:=\omega_{k}-1=a\frac{4{\sin^{2}{\left(\frac{k\pi}{2N+2}\right)}}}{\omega_{k}+1}\leq 2a$ ###### Definition 3.1. A polynomial $Z$ will be said to be in normal form if it Poisson commutes with $H_{00}$, i.e. if $\left\\{H_{00},Z\right\\}\equiv 0\ .$ (3.6) ###### Remark 3.1. In order to study the system with DBC we will always extend the system to a system defined for $j=-(N+1),...,N$ with PBC, which is invariant under the involution $q_{j}\mapsto-q_{-j}$, $p_{j}\mapsto-p_{-j}$. The extension of the $\beta$-model is obtained without modifying the equations, while the extension of the $\alpha$-$\beta$ model is given by the system $\dot{q}_{j}=p_{j}\ ,\quad\dot{p}_{j}=-q_{j}-a(\Delta_{1}q)_{j}-\alpha s_{j}q_{j}^{2}-\beta q_{j}^{3},$ (3.7) where $s_{j}$ is a discretization of the step function given by $s_{j}:=\left\\{\begin{matrix}1&\text{if}\ j\geq 1\\\ 0&\text{if}\ j=0\\\ -1&\text{if}\ j\leq 1\end{matrix}\right.$ (3.8) and $(\Delta_{1}q)_{j}=2q_{j}-q_{j+1}-q_{j-1},$ (3.9) is the discrete Laplacian. The need of the introduction of the sequence $s_{j}$ is at the origin of the finite smoothness of the solution in the DBC case. We are going to prove the following ###### Theorem 3.1. Assume $a<1/3$. Then, there exists an analytic canonical transformation $z={\mathcal{T}}(\psi)$, defined in a neighborhood of the origin $H\circ{\mathcal{T}}=H_{0}(\psi)+Z(\psi)+{\mathcal{R}}(\psi)\ ,$ (3.10) where $Z$ is in normal form and the following holds true * 1) The remainder, the normal form and the canonical transformation are estimated by $\displaystyle\left\|X_{{\mathcal{R}}}(\psi)\right\|_{s,\sigma}$ $\displaystyle\leq$ $\displaystyle C_{\mathcal{R}}\left\|\psi\right\|_{s,\sigma}^{4},$ (3.11) $\displaystyle\left\|X_{Z}(\psi)\right\|_{s,\sigma}$ $\displaystyle\leq$ $\displaystyle C_{Z}\left\|\psi\right\|_{s,\sigma}^{3}\ ,$ (3.12) $\displaystyle\left\|\psi-{\mathcal{T}}(\psi)\right\|_{s,\sigma}$ $\displaystyle\leq$ $\displaystyle C_{\mathcal{T}}\left\|\psi\right\|_{s,\sigma}^{2}\ $ (3.13) where $\begin{matrix}\frac{1}{2}<s\ ,\quad&\sigma\geq 0&\ \text{if\ PBC}\\\ \frac{1}{2}<s<\frac{5}{2}\ ,&\sigma=0&\ \text{if\ DBC}\end{matrix}$ (3.14) and $C_{{\mathcal{R}},Z,{\mathcal{T}}}$ are constants independent of $N$. * 2) One has ${\mathcal{T}}=\mathbb{I}+X_{\chi_{01}}+{\mathcal{R}}_{{\mathcal{T}}}$ with $\chi_{01}$ given by $\chi_{01}(\psi)=-\frac{\alpha}{6\sqrt{2}}\sum_{j}{\left(\frac{1}{3}\psi_{j}^{3}-\frac{{\rm i}}{3}\overline{\psi}_{j}^{3}+3\psi_{j}\overline{\psi}_{j}^{2}-3{\rm i}\psi_{j}^{2}\overline{\psi}_{j}\right)}.$ (3.15) and $\left\|{\mathcal{R}}_{{\mathcal{T}}}(\psi)\right\|_{s-s_{1},\sigma}\leq C_{{\mathcal{R}}_{{\mathcal{T}}}}\mu^{s_{1}}\left\|\psi\right\|_{s,\sigma}^{2}$ (3.16) where the parameters vary in the range $\begin{matrix}0\leq s_{1}<s-\frac{1}{2}<2,\ \sigma=0\ \text{for\ DBC}\\\ 0\leq s_{1}<s-\frac{1}{2}\ ,s_{1}\leq 2\ ,\ \sigma\geq 0\ \text{for\ PBC}\end{matrix}$ (3.17) * 3) The normal form has the following structure $Z(\psi,\bar{\psi})=Z_{0}+Z_{r}\ $ where $Z_{0}:=\tilde{\gamma}\sum_{j}\left|\psi_{j}\right|^{4},\quad\tilde{\gamma}:=\frac{3}{8}\left(\beta-\frac{10}{9}\alpha^{2}\right)$ (3.18) and $\left\|X_{Z_{r}}(\psi)\right\|_{s-s_{1},\sigma}\leq\mu^{s_{1}}C_{r}\left\|\psi\right\|_{s,\sigma}^{3}\ $ (3.19) with the parameters varying in the range (3.17). The proof of this theorem is divided into three parts. In the first one we will prove an abstract normal form theorem under the assumption that the non- linearity corresponding to the system (2.4) is smooth. In the second part we will prove such a smoothness assumption. In the third part we will compute the first order term of the normal form and of the transformation and we will estimate the corresponding errors. We point out that there are 3 delicate points in the proof: the first one is to solve the homological equation (see lemma 3.3); the second one is to prove smoothness of the perturbation in the optimal space, and the third one is the actual computation of the main part of the normal form and of the canonical transformation. ### 3.2 An abstract normal for theorem. First we recall that a homogeneous polynomial map $F:{\mathcal{P}}_{s,\sigma}\to{\mathcal{P}}_{s,\sigma}$ of degree $r$ is continuous and also analytic if and only if it is bounded, i.e. if there exists a constant $C$ such that $\left\|F(z)\right\|_{s,\sigma}\leq C\left\|z\right\|^{r}_{s,\sigma}\qquad\qquad\forall z\in{\mathcal{P}}_{s,\sigma}.$ (3.20) ###### Definition 3.2. The best constant such that (3.20) holds is called the norm of $F$, and will be denoted by $\left\bracevert F\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}$. One has $\left\bracevert F\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}:=\sup_{\left\|z\right\|_{s,\sigma}=1}\left\|F(z)\right\|_{s,\sigma}\ .$ (3.21) ###### Definition 3.3. A polynomial function $f$, homogeneous of degree $r+2$, will be said to be of class ${\mathcal{H}}_{s,\sigma}^{r}$ if its Hamiltonian vector field $X_{f}$ is bounded as a map from ${\mathcal{P}}_{s,\sigma}$ to itself. ###### Theorem 3.2. Let $H_{0}$ be as above (see (2.11)) with $a<1/3$. Assume that $H_{j}\in{\mathcal{H}}_{s,\sigma}^{j}$ for some fixed $s,\sigma$ and for $j=1,2$; then there exists an analytic canonical transformation ${\mathcal{T}}={\mathcal{T}}_{1}\circ{\mathcal{T}}_{2}$, defined in a neighborhood of the origin in ${\mathcal{P}}_{s,\sigma}$ such that $H\circ{\mathcal{T}}=H_{0}+Z+{\mathcal{R}}\ ,$ (3.22) where $Z$ is in normal form and the remainder, the normal form and the canonical transformation are bounded by $\displaystyle\left\|X_{{\mathcal{R}}}(z)\right\|_{s,\sigma}$ $\displaystyle\leq$ $\displaystyle C_{\mathcal{R}}\left\|z\right\|_{s,\sigma}^{4}\ ,$ (3.23) $\displaystyle\left\|X_{Z}(z)\right\|_{s,\sigma}$ $\displaystyle\leq$ $\displaystyle C_{Z}\left\|z\right\|_{s,\sigma}^{3}\ ,$ (3.24) $\displaystyle\left\|z-{\mathcal{T}}(z)\right\|_{s,\sigma}$ $\displaystyle\leq$ $\displaystyle C_{\mathcal{T}}\left\|z\right\|_{s,\sigma}^{2},\ $ (3.25) with constants $C_{{\mathcal{R}},Z,{\mathcal{T}}}$ depending only on $a,\left\bracevert X_{H_{1}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma},\left\bracevert X_{H_{2}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}$. The rest of this subsection will be occupied by the proof of theorem 3.2. First we need some simple estimates. ###### Lemma 3.1. Let $f\in{\mathcal{H}}_{s,\sigma}^{r}$ and $g\in{\mathcal{H}}_{s,\sigma}^{r_{1}}$, then $\left\\{f,g\right\\}\in{\mathcal{H}}_{s,\sigma}^{r+r_{1}}$, and $\left\bracevert X_{\\{f;g\\}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\leq(r+r_{1}+2)\left\bracevert X_{f}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\left\bracevert X_{g}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}$ (3.26) ###### Proof. First remember that $X_{\left\\{f,g\right\\}}=[X_{f};X_{g}]=dX_{f}X_{g}-dX_{g}X_{f}\ .$ (3.27) We recall now that, given a polynomial $X$ of degree $r+1$, there exists a unique $(r+1)-$linear symmetric form $\widetilde{X}$ such that $X(z)=\widetilde{X}(z,...,z)\ ,$ then, (3.27) is explicitly given by $X_{\\{f;g\\}}=(r+1)\widetilde{X}_{f}(X_{g}(z),z,z,...,z)-(r_{1}+1)\widetilde{X}_{g}(X_{f}(z),z,z,...,z)$ (3.28) moreover from (3.21) one has $\left\|\widetilde{X}(z_{1},...,z_{r+1})\right\|_{s,\sigma}\leq\left\bracevert X\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\left\|z_{1}\right\|_{s,\sigma}...\left\|z_{r+1}\right\|_{s,\sigma}$ (3.29) from which the thesis immediately follows. ∎ ###### Remark 3.2. Let $f\in{\mathcal{H}}_{s,\sigma}^{r}$ then the corresponding vector field generates a flow in a neighborhood of the origin in ${\mathcal{P}}_{s,\sigma}$. ###### Lemma 3.2. Let $\chi$ be of class ${\mathcal{H}}_{s,\sigma}^{r}$, and let $f\in{\mathcal{H}}^{r_{1}}_{s,\sigma}$. Let $\Phi^{1}_{\chi}$ be the Lie transform generated by $\chi$, then each term of the Taylor expansion of $f\circ\Phi^{1}_{\chi}$ is a polynomial with bounded vector field. ###### Proof. Iterating the relation $\frac{d}{dt}(f\circ\Phi^{t}_{\chi})=\left\\{\chi,f\right\\}\circ\Phi^{t}_{\chi}$ one gets that the Taylor expansion of $f\circ\Phi_{\chi}^{1}$ is given by $\displaystyle f\circ\Phi^{1}_{\chi}$ $\displaystyle=$ $\displaystyle\sum_{l\geq 0}{f_{l}}$ $\displaystyle f_{0}$ $\displaystyle=$ $\displaystyle f,\qquad\qquad f_{l}=\frac{1}{l}\left\\{\chi,f_{l-1}\right\\},\qquad l\geq 1.$ Then, the thesis follows from lemma 3.1∎ A key role in the proof of theorem 3.2 is played by the so called homological equation, namely $\left\\{\chi_{j};H_{0}\right\\}+f_{j}=Z_{j}$ (3.30) where $f_{j}\in{\mathcal{H}}_{s,\sigma}^{j}$ is a given polynomial, and $\chi_{j}\in{\mathcal{H}}_{s,\sigma}^{j}$, $Z_{j}\in{\mathcal{H}}_{s,\sigma}^{j}$ are to be determined with the property that $Z_{j}$ is in normal form. ###### Lemma 3.3. Consider the Homological equation (3.30) with $f_{j}$ of class ${\mathcal{H}}_{s,\sigma}^{j}$, with $j=1,2$. Assume that $a<\frac{1}{3}$, then (3.30) admits a solution $\chi_{j}$, $Z_{j}\in{\mathcal{H}}_{s,\sigma}^{j}$ with $\left\bracevert X_{\chi_{j}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\leq\frac{1}{2(1-3a)}\left\bracevert X_{f_{j}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}.$ (3.31) ###### Proof. First we rewrite the homological equation as $(L_{0}+L_{1})\chi_{j}=f_{j}-Z_{j}$ (3.32) where the operators $L_{0}$ and $L_{1}$ are defined by $L_{0}\chi_{j}:=\left\\{H_{00},\chi_{j}\right\\}\ ,\quad L_{1}\chi_{j}:=\left\\{H_{01},\chi_{j}\right\\}$ and $H_{00}$ and $H_{01}$ are defined by (3.5). We will invert $L_{0}$ and solve (3.32) by Neumann series (see [BDGS07]). We begin by showing that the space ${\mathcal{H}}_{s,\sigma}^{j}$, $j\leq 2$ decompose into the sum of the kernel ${\rm Ker}(L_{0})$ of $L_{0}$ and of its range ${\rm Im}(L_{0})$. Moreover, we show that $L_{0}$ is invertible on its range. Given $f\in{\mathcal{H}}_{s,\sigma}^{j}$ with $j=1,2$ define $\displaystyle Z$ $\displaystyle:=$ $\displaystyle\frac{1}{T}\int_{0}^{T}{f\left(\Psi^{t}(z)\right)dt}\ ,$ (3.33) $\displaystyle\chi$ $\displaystyle:=$ $\displaystyle\frac{1}{T}\int_{0}^{T}{\left[tf\left(\Psi^{t}(z)\right)-Z\left(\Psi^{t}(z)\right)\right]dt}\ ,$ (3.34) where $\Psi^{t}$ is the flow of $X_{H_{00}}$ and $T=1$ is its period. Then an explicit computation shows that $Z\in$Ker$(L_{0})$, and that (see [BG93]) $L_{0}\chi=f-Z\ .$ (3.35) Thus denoting by $Q$ the projector on the kernel of $L_{0}$, and $P=\mathbb{I}-Q$ the projector on the range, one sees that the (3.33) is a concrete definition of $Q$, while (3.34) is the definition of $L_{0}^{-1}$ restricted to Im($L_{0}$). It remains to show that $Z,\chi\in{\mathcal{H}}_{s,\sigma}^{j}$. Remark that, since $\Psi^{t}$ is a canonical transformation one has $\displaystyle X_{Z}(z)$ $\displaystyle\equiv$ $\displaystyle X_{Qf}(z)=\frac{1}{T}\int_{0}^{T}\left(\Psi^{-t}\circ X_{f}\circ\Psi^{t}\right)(z)dt\ ,$ (3.36) $\displaystyle X_{\chi}(z)$ $\displaystyle\equiv$ $\displaystyle X_{L_{0}^{-1}Pf}(z)=\frac{1}{T}\int_{0}^{T}\left(\Psi^{-t}\circ X_{Pf}\circ\Psi^{t}\right)(z)tdt\ ,$ (3.37) From which it follows that $\displaystyle\left\bracevert X_{Qf}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}$ $\displaystyle\leq$ $\displaystyle\left\bracevert X_{f}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\ ,$ (3.38) $\displaystyle\left\bracevert X_{{L_{0}^{-1}Pf}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}$ $\displaystyle\leq$ $\displaystyle\frac{1}{2}\left\bracevert X_{Pf}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\leq\frac{1}{2}\left\bracevert X_{f}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\ \Longrightarrow\left\|L_{0}^{-1}\right\|\leq\frac{1}{2}$ (3.39) where the last norm is the norm of $L_{0}^{-1}$ as a linear operator acting on the space ${\mathcal{H}}_{s,\sigma}^{j}$, and thus $Qf\in{\mathcal{H}}_{s,\sigma}^{j}$, $L_{0}^{-1}Pf\in{\mathcal{H}}_{s,\sigma}^{j}$. We come now to the true homological equation (3.32). We look for a solution $\chi_{j}=P\chi_{j}$ and $Z_{j}=QZ_{j}$. Applying $P$ or $Q$ to (3.32), remarking that since $[L_{0},L_{1}]=0$ one has $[P,L_{1}]=[Q,L_{1}]=0$, we get $(L_{0}+L_{1})\chi_{j}=Pf_{j}\ ,\quad Qf_{j}=Z_{j}\ .$ (3.40) The first of (3.40) is formally solved by Neumann series, i.e. defining $(L_{0}+L_{1})^{-1}:=\sum_{k\geq 0}(-1)^{k}\left(L_{0}^{-1}L_{1}\right)^{k}L_{0}^{-1}\ ,\quad\chi_{j}:=(L_{0}+L_{1})^{-1}Pf_{j}\ .$ (3.41) To show the convergence of the series in operator norm we need an estimate of $\left\|L_{1}\right\|$. To this end remark that, for any $s,\sigma$ one has $\left\bracevert X_{H_{01}}\right\bracevert\hbox{}\kern-5.0pt_{s\sigma}\leq 2a\ ,$ which using lemma 3.1 implies $\left\|L_{1}\right\|\leq 2a(j+1)\leq 6a$. It follows that the series (3.41) converges provided $a<1/3$, which is our assumption, and that $\left\|(L_{0}+L_{1})^{-1}\right\|\leq\frac{1}{2(1-3a)},$ (3.42) which concludes the proof. ∎ End of the proof of Theorem 3.2. From lemma 3.3 one has that the solution $\chi_{1}$ of the homological equation with $f_{1}\equiv H_{1}$ is well defined provided $H_{1}\in{\mathcal{H}}_{s,\sigma}^{1}$ for some $s,\sigma$. Then $\chi_{1}$ generates a Lie transform ${\mathcal{T}}_{1}$ which puts the system in normal form up to order 4. Then the part of degree four of $H\circ{\mathcal{T}}_{1}$ takes the form $f_{2}:=\frac{1}{2}\left\\{\chi_{1},\left\\{\chi_{1},H_{0}\right\\}\right\\}+\left\\{\chi_{1},H_{1}\right\\}+H_{2}\equiv\frac{1}{2}\left\\{\chi_{1},H_{1}\right\\}+H_{2}$ (3.43) which is of class ${\mathcal{H}}_{s,\sigma}^{2}$. It follows that one can use the homological equation with such a known term and determine a $\chi_{2}$ which generates the Lie transformation putting the system in normal form up to order 4. This concludes the proof of theorem 3.2.∎ ### 3.3 Proof of the smoothness properties of the nonlinearity In this subsection we prove the following lemma ###### Lemma 3.4. Let $H_{j}$, $j=1,2$ be given by (2.4). Consider the vector fields $X_{H_{j}}$ of the cubic and of the quartic terms of the Hamiltonian: they fulfill the estimates $\left\|X_{H_{1}}(z)\right\|_{s,\sigma}\leq G_{1}\left\|z\right\|_{s,\sigma}^{2}\ ,\quad\left\\{\begin{matrix}\frac{1}{2}<s,&\sigma\geq 0&\text{if\ PBC}\\\ \frac{1}{2}<s<\frac{5}{2},&\sigma=0&\text{if\ DBC}\end{matrix}\right.$ (3.44) and $\left\|X_{H_{2}}(z)\right\|_{s,\sigma}\leq G_{2}\left\|z\right\|_{s,\sigma}^{3}\ ,\quad\frac{1}{2}<s\ ,\ \sigma\geq 0\,\ \text{both\ cases}$ (3.45) where we set $G_{j}:=\left\bracevert X_{H_{j}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}$. The proof will be split into two parts. First we show that it is possible to prove the result working on interpolating functions, and then we show that the “interpolating nonlinearities” have a smooth vector field when the parameters $s,\sigma$ vary in the considered range. ###### Remark 3.3. Define $T_{l}$ as the map $q_{j}\mapsto q_{j}^{l+1}$ in the case of PBC, and $[T_{1}(q)]_{j}=q_{j}^{2}s_{j}$ and $[T_{2}(q)]_{j}:=q_{j}^{3}$ in the case of DBC. Then the vector field $X_{H_{l}}$ has only $p$ components, moreover the norms are defined in terms of the Fourier variables, so we have to estimate the map constructed as follows $\displaystyle\hat{q}_{k}\mathop{\mapsto}^{1}\hat{q}_{k}^{S}:=\frac{\hat{q}_{k}}{\sqrt{\omega_{k}}}\mathop{\mapsto}^{{\mathcal{F}}}q_{j}:=\sum_{k}\hat{q}_{k}^{S}{\hat{e}}_{k}({j})\mathop{\mapsto}T_{l}(q_{j})\mathop{\mapsto}^{{\mathcal{F}}^{-1}}$ $\displaystyle\mathop{\mapsto}^{{\mathcal{F}}^{-1}}p_{k}^{S}:=\sum_{j}q_{j}^{l+1}{\hat{e}}_{k}({j})\mathop{\mapsto}^{2}\hat{p}_{k}:=\sqrt{\omega_{k}}p_{k}^{S}$ It is immediate to realize that the maps 1 and 2 are smooth (the frequencies are between 1 and 3) so it is enough to estimate the remaining maps. In turns the remaining maps essentially coincide with the map $T_{l}(q)$ read in terms of standard Fourier variables (without the factors $\sqrt{\omega_{k}}$). These are the maps we will estimate. All along this section we will use a definition of the Fourier coefficients of a sequence not including the factors $\sqrt{\omega_{k}}$, namely we define $\hat{q}_{k}$ by $q_{j}=\sum_{k}\hat{q}_{k}\hat{e}_{k}(j)\ .$ We start by showing how to use the interpolation in order to make estimates. To this end we define an interpolation operator $I$ by $[I(q)](x):=\sum_{k}\sqrt{\mu}\hat{q}_{k}\hat{e}_{k}^{c}(x)$ (3.46) We also define a restriction operator $R$ that to a function associates the corresponding sequence, by $[R(u)]_{j}:=u(\mu j)$ (3.47) We remark that the operator $R$ is defined on functions which do not necessarily have finitely many non-vanishing Fourier coefficients. ###### Remark 3.4. With the definition (2.14) one has $\left\|Iq\right\|_{s,\sigma}=\left\|q\right\|_{s,\sigma}\ .$ (3.48) ###### Lemma 3.5. For any $s>1/2$ there exists a constant $C_{6}(s)$ such that one has $\left\|Ru\right\|_{s,\sigma}\leq C_{6}\left\|u\right\|_{s,\sigma}$ (3.49) ###### Proof. Denote $q_{j}=(Ru)_{j}$; using the formula $\hat{e}^{c}_{k}(\mu j)=\hat{e}^{c}_{k+2(N+1)m}(\mu j)=\frac{1}{\sqrt{\mu}}\hat{e}_{k}(j)$ (3.50) one gets $q_{j}=\sum_{k\in{\mathbb{Z}}}\hat{u}_{k}\hat{e}^{c}_{k}(\mu j)=\sum_{k=-(N+1)}^{N}{\sqrt{\mu}}{\hat{e}^{c}_{k}(\mu j)}\sum_{m\in{\mathbb{Z}}}\frac{\hat{u}_{k+2(N+1)m}}{\sqrt{\mu}}=\sum_{k=-(N+1)}^{N}{\hat{e}_{k}(j)\hat{q}_{k}}\ ,$ from which $\hat{q}_{k}=\sum_{m\in{\mathbb{Z}}}\frac{\hat{u}_{k+2(N+1)m}}{\sqrt{\mu}}.$ (3.51) Let’s define $\gamma_{m,k}(s,\sigma):=\frac{[k]^{s}e^{\sigma|k|}}{[k+2(N+1)m]^{s}e^{\sigma|k+2(N+1)m|}}=\left(\frac{[k]}{[k+2(N+1)m]}\right)^{s}\frac{e^{\sigma|k|}}{e^{\sigma|k+2(N+1)m|}}$ and replace (3.51) in the norm $\left\|q\right\|_{s,\sigma}^{2}$, then we get $\displaystyle\left\|q\right\|_{s,\sigma}^{2}$ $\displaystyle=$ $\displaystyle\mu\sum_{k=-(N+1)}^{N}{[k]^{2s}e^{2\sigma|k|}|\hat{q}_{k}|^{2}}=$ $\displaystyle=$ $\displaystyle\sum_{k=-(N+1)}^{N}[k]^{2s}e^{2\sigma|k|}\left|\sum_{m\in{\mathbb{Z}}}\hat{u}_{k+2(N+1)m}\right|^{2}=$ $\displaystyle=$ $\displaystyle\sum_{k=-(N+1)}^{N}[k]^{2s}e^{2\sigma|k|}\left|\sum_{m\in{\mathbb{Z}}}\frac{\gamma_{m,k}(s,\sigma)}{\gamma_{m,k}(s,\sigma)}\hat{u}_{k+2(N+1)m}\right|^{2}\leq$ $\displaystyle\leq$ $\displaystyle\sum_{k=-(N+1)}^{N}\left(\sum_{m\in{\mathbb{Z}}}{\gamma_{m,k}^{2}(s,\sigma)}\right)\sum_{m\in{\mathbb{Z}}}{\frac{[k]^{2s}e^{2\sigma|k|}}{\gamma_{m,k}^{2}(s,\sigma)}}|\hat{u}_{k+2(N+1)m}|^{2}\leq$ $\displaystyle\leq$ $\displaystyle C_{6}(s)\left\|u\right\|^{2}_{s,\sigma}\ .$ Indeed, since $k=-(N+1),\ldots,N$, we can estimates the two factors of $\gamma_{m,k}$ as follows: * • $|2(N+1)m+k|\geq|2(N+1)|m|-|k||\geq\begin{cases}|k|,\qquad m=0,\\\ N+1\geq|k|,\qquad m\not=0,\end{cases}$ which gives $\frac{e^{\sigma|k|}}{e^{\sigma|k+2(N+1)m|}}\leq 1.$ * • $\frac{[k]}{[k+2(N+1)m]}=\begin{cases}\frac{1}{[2(N+1)m]}<\frac{1}{[2m]},\qquad\qquad m=0,\\\ \frac{|k|}{|k+2(N+1)m|}\leq\frac{1}{1+2m}<\frac{1}{[2m]},\qquad m\not=0\end{cases}$ which gives $\left(\frac{[k]}{[k+2(N+1)m]}\right)^{2s}\leq\frac{1}{[2m]^{2s}}.$ ∎ ###### Corollary 3.3. Let $T:{\mathbb{R}}^{2(N+1)}\to{\mathbb{R}}^{2(N+1)}$ be a polynomial map, assume that there exists an “interpolating polynomial map $T^{c}$” such that $T=RT^{c}I$. If the map $T^{c}$ is bounded in some space $H^{s,\sigma}$, with $s>1/2$, then $T$ is bounded in $\ell^{2}_{s,\sigma}$. Moreover one has $\left\bracevert T\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\leq C_{6}\left\bracevert T^{c}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\ .$ (3.52) First we define the interpolating maps we have to study. They are $T^{c}_{l}(u):=u^{l+1}$ in the case of PBC and $T^{c}_{1}(u)={\rm sgn}(x)u^{2}(x)$ and $T^{c}_{2}(u):=u^{3}$ in the case of DBC. Here we introduced the function ${\rm sgn}(x):=\left\\{\begin{matrix}1&\text{if}\ x>0\\\ 0&\text{if}\ x=0\\\ -1&\text{if}\ x<1\end{matrix}\right.$ The estimates (3.44) in the case of PBC and (3.45) for both boundary conditions are proved in Lemma A.1 by a standard argument on the Sobolev norm of the product of two functions. We come to the estimate of $T^{c}_{1}$ in the case of DBC. We will denote by $H^{s}_{o}$ the subspace of $H^{s,0}$ composed by the odd functions $u(x)$ on $[-\pi,\pi]$. ###### Lemma 3.6. For any $1/2<s<5/2$, The operator $T^{c}_{1}(u):=u^{2}\mathrm{sgn}(x)$ is smooth from $H^{s}_{o}$ in itself and there exists $C_{7}(s)$ such that $\left\|T^{c}_{1}u\right\|_{s}\leq C_{7}\left\|u\right\|^{2}_{s}.$ ###### Proof. We begin with the case $2\leq s<\frac{5}{2}$. First, observe that the function $T^{c}_{1}(u)$ is odd when $u$ is odd. we will prove the thesis by showing that the second weak derivative $d^{2}T^{c}_{1}(u)$ of $T^{c}_{1}(u)$ is in $H_{o}^{s-2}$. First remark that, by an explicit computation which exploits the fact that $u(0)=0$ one has $\displaystyle d^{2}T_{1}^{c}(u)=2[{\rm sgn}(x)(u^{\prime 2})+{\rm sgn}(x)ud^{2}u]\ .$ We show now that both terms are in $H^{s-2}_{o}$. The second term can be considered as the product of the function $d^{2}u\in H^{s-2}_{o}$ and of ${\rm sgn}(x)u$. This last function is of class $H^{1}$, as it is seen by computing its derivative, namely $d\left({\rm sgn}(x)u\right)={\rm sgn}(x)du(x)+\delta(x)u(x)={\rm sgn}(x)du(x)$ which clearly belongs to $L^{2}$. From lemma A.1 it follows that the product ${\rm sgn}(x)u[d^{2}u]\in H^{s-2}$. Concerning the term ${\rm sgn}(x)(u^{\prime 2})$, it can be considered as the product of $u^{\prime 2}\in H^{s-1}$ and of ${\rm sgn}(x)$, which is of class $H^{r}_{o}$ for all $r<1/2$, as it can be seen by explicitly computing its Fourier coefficients. Thus lemma A.1 gives the result. The case $1\leq s<2$ is easier and works in a very similar way. Indeed, since $u(x){\rm sgn}(x)\in H^{1}$ and $du(x)\in H^{s}$ with $0\leq s<1$, the derivative $d(u^{2}(x){\rm sgn}(x))=2u(x)(du(x)){\rm sgn}(x)$ belongs to $H^{s}$ with $0\leq s<1$, which gives the thesis. Quite different is the case $\frac{1}{2}<s<1$, since by hypothesis no weak derivative exists for the function $u(x)$. We exploit the following equivalent definition of the norm of the Sobolev space $H^{s}_{o}([-\pi,\pi])$ with real exponent $s$ $\left\|u\right\|^{2}_{s}:=\int_{-\pi}^{\pi}\int_{-\pi}^{\pi}{\frac{|u(x)-u(y)|^{2}}{|x-y|^{1+2s}}dxdy},$ (3.53) and the skew-symmetry of the periodic function $u(x)\in H^{s}_{o}$. We want to prove that $g(x):=u(x){\rm sgn}(x)\in H^{s}$ with $\frac{1}{2}<s<1$; more precisely $\left\|u(x){\rm sgn}(x)\right\|^{2}_{s}<4\left\|u(x)\right\|^{2}_{s},\qquad\qquad\frac{1}{2}<s<1.$ (3.54) The symmetries of $g(x)$ on the given domain allow to simplify the integral in (3.53) $\displaystyle\left\|g\right\|^{2}_{s}$ $\displaystyle=$ $\displaystyle 2\int_{0}^{\pi}\int_{-\pi}^{\pi}{\frac{|g(x)-g(y)|^{2}}{|x-y|^{1+2s}}dxdy}=$ $\displaystyle=$ $\displaystyle 2\int_{0}^{\pi}\int_{0}^{\pi}\left[{\frac{|g(x)-g(y)|^{2}}{|x-y|^{1+2s}}}+\frac{|g(x)-g(y)|^{2}}{|x+y|^{1+2s}}\right]dxdy=$ $\displaystyle=$ $\displaystyle 2\int_{0}^{\pi}\int_{0}^{\pi}\left[{\frac{|u(x)-u(y)|^{2}}{|x-y|^{1+2s}}}+\frac{|u(x)-u(y)|^{2}}{|x+y|^{1+2s}}\right]dxdy\leq$ $\displaystyle\leq$ $\displaystyle 4\int_{0}^{\pi}\int_{0}^{\pi}{\frac{|u(x)-u(y)|^{2}}{|x-y|^{1+2s}}}dxdy<4\left\|u\right\|^{2}_{s}.$ ∎ ### 3.4 Computation of the normal form and of the transformation. In this section we will concentrate on the case of DBC which is the most difficult one. Consider again the Hamiltonian (2.4), introduce complex variables $\xi_{j},\eta_{j}$ defined by $\xi_{j}=\frac{p_{j}+{\rm i}q_{j}}{\sqrt{2}}\ ,\qquad\eta_{j}=\frac{p_{j}-{\rm i}q_{j}}{{\rm i}\sqrt{2}}$ (3.55) and split $H_{0}={\mathcal{H}}_{00}+{\mathcal{H}}_{01}$ with $\displaystyle{\mathcal{H}}_{00}$ $\displaystyle:=$ $\displaystyle\sum_{j}\frac{p_{j}^{2}+q_{j}^{2}}{2}\equiv{\rm i}\left\langle\xi;\eta\right\rangle_{\ell^{2}}\ ,$ (3.56) $\displaystyle{\mathcal{H}}_{01}$ $\displaystyle:=$ $\displaystyle\sum_{j}\frac{a}{2}(q_{j}-q_{j-1})^{2}\equiv\frac{a}{2}\left\langle\frac{\xi-{\rm i}\eta}{{\rm i}\sqrt{2}};-\Delta_{1}\frac{\xi-{\rm i}\eta}{{\rm i}\sqrt{2}}\right\rangle_{\ell^{2}}\ ,$ (3.57) where $\left\langle\xi;\eta\right\rangle_{\ell^{2}}:=\sum_{j}\xi_{j}\eta_{j}$. ###### Remark 3.5. The above splitting is different from the one introduced in (3.5) which had been used in the proof of lemma 3.3, and which was based on the Fourier variables. In particular one has $\left\\{H_{00};H_{01}\right\\}\equiv 0$, but $\left\\{{\mathcal{H}}_{00};{\mathcal{H}}_{01}\right\\}\not\equiv 0$. In the variables $(\xi,\eta)$ the flow $\Phi^{t}$ of ${\mathcal{H}}_{00}$ acts as follows $\Phi^{t}(\xi,\eta):=\begin{cases}\xi_{j}\mapsto e^{{\rm i}t}\xi_{j},\\\ \eta_{j}\mapsto e^{-{\rm i}t}\eta_{j}.\end{cases}$ (3.58) The third order part of the Hamiltonian takes the form $H_{1}=\frac{\alpha}{3}\sum_{j}\left(\frac{\xi_{j}-{\rm i}\eta_{j}}{{\rm i}\sqrt{2}}\right)^{3}s_{j}$ (3.59) where $s_{j}$ is the discrete step function defined in (3.8). The form of $H_{2}$ will be given below. Denoting $\mathcal{L}_{0}:=\left\\{{\mathcal{H}}_{00},.\right\\}\ ,\quad\mathcal{L}_{1}:=\left\\{{\mathcal{H}}_{01};.\right\\}$ (3.60) we rewrite the homological equation for $\chi_{1}$ as follows $(\mathcal{L}_{0}+\mathcal{L}_{1})\chi_{1}=H_{1}$ (3.61) which is solvable since the Kernel of $\mathcal{L}_{0}+\mathcal{L}_{1}=L_{0}+L_{1}$ on polynomials of third order is empty. The solution $\chi_{1}$ of (3.61) is unique and, as shown in Lemma 3.3, exists. By a direct computation one has $\chi_{1}=\mathcal{L}_{0}^{-1}H_{1}-(\mathcal{L}_{0}+\mathcal{L}_{1})^{-1}\mathcal{L}_{1}\mathcal{L}_{0}^{-1}H_{1};$ (3.62) we are going to show that the second term is much smaller than the first one. Before starting, a couple of remarks are in order. ###### Remark 3.6. The discrete Laplacian is $\ell^{2}$-symmetric on periodic sequences $<\Delta_{1}\xi,\eta>_{\ell^{2}}=<\xi,\Delta_{1}\eta>_{\ell^{2}}.$ (3.63) This is an immediate consequence of the fact that in Fourier variables it acts as a multiplier by a real factor. ###### Remark 3.7. In Fourier coordinates the discrete Laplacian $\Delta_{1}$ defined in (3.9) acts as a multiplier by $\sin^{2}k\mu$. It follows that it has norm 1 when acting on anyone of the spaces $\ell^{2}_{s,\sigma}$. Moreover, since $|\sin^{2}(k\mu)|\leq k^{s_{1}}\mu^{s_{1}},\qquad s_{1}\in[0,2],\qquad k\mu\in\left[0,\pi\right]$ one also has $\left\|\Delta_{1}\xi\right\|_{s-s_{1},\sigma}\leq\mu^{s_{1}}\left\|\xi\right\|_{s,\sigma},\qquad s_{1}\in[0,2].$ (3.64) From (3.64) and (3.57) it follows $\left\|X_{{\mathcal{H}}_{01}}(\xi,\eta)\right\|_{s-s_{1},0}\leq C(a)\mu^{s_{1}}\left\|(\xi,\eta)\right\|_{s,0},\qquad 0\leq s_{1}<s-\frac{1}{2}.$ (3.65) ###### Lemma 3.7. Assume $a<\frac{1}{3}$, then $\chi_{1}=\chi_{10}+\chi_{1r}$ with $\chi_{10}(\xi,\eta)=\frac{\alpha}{6\sqrt{2}}\sum_{j}{\left(\frac{1}{3}\xi_{j}^{3}-\frac{{\rm i}}{3}\eta_{j}^{3}+3\xi_{j}\eta_{j}^{2}-3{\rm i}\xi_{j}^{2}\eta_{j}\right)},$ and there exists $C_{8}(a,G_{1})$ such that $\left\|X_{\chi_{1r}}(\xi,\eta)\right\|_{s-s_{1},\sigma}\leq C_{8}\mu^{s_{1}}\left\|(\xi,\eta)\right\|_{s,\sigma}^{2},$ (3.66) with $\left\\{\begin{matrix}0\leq s_{1}<s-\frac{1}{2}<2,\ \sigma=0\ \text{for\ DBC}\\\ 0\leq s_{1}<s-\frac{1}{2},\,s_{1}\leq 2\,\ \sigma\geq 0\ \text{for\ PBC}\end{matrix}\right.$ ###### Proof. According to (3.62), let’s define $\chi_{10}:=\mathcal{L}_{0}^{-1}H_{1},\qquad\qquad\chi_{1r}:=-(\mathcal{L}_{0}+\mathcal{L}_{1})^{-1}\mathcal{L}_{1}\chi_{10}=-(L_{0}+L_{1})^{-1}\mathcal{L}_{1}\chi_{10}.$ Since $\chi_{10}$ solves the homological equation $\left\\{{\mathcal{H}}_{00},\chi_{10}\right\\}=H_{1}$, it can be explicitly computed by $\displaystyle\chi_{10}(\xi,\eta)$ $\displaystyle=$ $\displaystyle\int_{0}^{1}{H_{1}\left(\Phi^{t}(\xi,\eta)\right)dt}=$ $\displaystyle=$ $\displaystyle-\frac{\alpha}{6\sqrt{2}}\sum_{j}{s_{j}\left(-\frac{1}{3}\xi_{j}^{3}+\frac{{\rm i}}{3}\eta_{j}^{3}-3\xi_{j}\eta_{j}^{2}+3{\rm i}\xi_{j}^{2}\eta_{j}\right)}$ which also implies $\left\bracevert X_{\chi_{10}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}\leq\frac{1}{2}\left\bracevert X_{H_{1}}\right\bracevert\hbox{}\kern-5.0pt_{s,\sigma}.$ The thesis follows from (3.28), (3.42) and (3.64).∎ We move to the second Homological equation $\left\\{\chi_{2},H_{0}\right\\}+\tilde{H}_{2}=Z$ (3.67) where $\tilde{H}_{2}=H_{2}+\frac{1}{2}\left\\{\chi_{1},H_{1}\right\\}$ can be split according to (3.62) into $\tilde{H}_{2}=\tilde{H}_{20}+\tilde{H}_{21}$ with $\tilde{H}_{20}:=H_{2}+\frac{1}{2}\left\\{\chi_{10},H_{1}\right\\},\qquad\tilde{H}_{21}:=\frac{1}{2}\left\\{\chi_{1r},H_{1}\right\\}.$ (3.68) More explicitly, the leading term $\tilde{H}_{20}$ is composed of $\displaystyle\frac{1}{2}\left\\{\chi_{10},H_{1}\right\\}$ $\displaystyle=$ $\displaystyle\frac{\alpha^{2}}{24}\sum_{j}{\left(\xi_{j}^{4}+4{\rm i}\xi_{j}^{3}\eta_{j}+10\xi_{j}^{2}\eta_{j}^{2}-4{\rm i}\xi_{j}\eta_{j}^{3}+\eta_{j}^{4}\right)},$ $\displaystyle H_{2}$ $\displaystyle=$ $\displaystyle\frac{\beta}{16}\sum_{j}{\left(\xi_{j}^{4}-4{\rm i}\xi_{j}^{3}\eta_{j}-6\xi_{j}^{2}\eta_{j}^{2}+4{\rm i}\xi_{j}\eta_{j}^{3}+\eta_{j}^{4}\right)}.$ Before proceeding, it is useful to perform the change of variables $\hat{\psi}_{k}=\frac{\hat{p}_{k}+{\rm i}\hat{q}_{k}}{\sqrt{2}},\qquad\qquad\psi_{j}=\sum_{k}{\hat{\psi}_{k}\hat{e}_{k}(j)}\ .$ (3.69) which puts the quadratic part into diagonal form. This implies a modification of $\tilde{H}_{20}$, which however is of higher order and therefore will be included into the remainder terms. Indeed, the Lemma below shows that the difference between $\xi$ and $\psi$ is small ###### Lemma 3.8. For any $2\geq s_{1}\geq 0$ it holds true $\displaystyle\left\|\xi-\psi\right\|_{s,\sigma}\leq a\mu^{s_{1}}\left\|\psi\right\|_{s+s_{1},\sigma}\ ,\quad\left\|{\rm i}\eta-\bar{\psi}\right\|_{s,\sigma}\leq a\mu^{s_{1}}\left\|\psi\right\|_{s+s_{1},\sigma}\ ,$ (3.70) $\displaystyle s>\frac{1}{2},\,\sigma\geq 0.$ ###### Proof. By definition $\displaystyle\left\|\xi-\psi\right\|_{s,\sigma}^{2}$ $\displaystyle=$ $\displaystyle\sum_{k}{[k]^{2s}e^{2\sigma k}(\sqrt{\omega}_{k}-1)^{2}(\frac{|\hat{\tilde{q}}_{k}|^{2}}{\omega_{k}}+|\hat{\tilde{p}}_{k}|^{2})}\leq$ $\displaystyle\leq$ $\displaystyle\sum_{k}{[k]^{2s}e^{2\sigma k}(\sqrt{\omega}_{k}-1)^{2}(|\hat{\tilde{q}}_{k}|^{2}+|\hat{\tilde{p}}_{k}|^{2})};$ a Taylor expansion of the frequencies $\omega_{k}$ gives $|\sqrt{\omega}_{k}-1|^{2}\leq\frac{1}{4}a^{2}{4\sin^{4}{\left(\frac{k\pi}{2N+2}\right)}}\leq Ca^{2}\mu^{2s_{1}}k^{2s_{1}}$ which is the thesis. ∎ ###### Corollary 3.4. In terms of the variables $\psi,\bar{\psi}$ one has $\tilde{H}_{2}=H_{20}+H_{21}$ where $\displaystyle H_{20}(\psi,\overline{\psi})$ $\displaystyle:=$ $\displaystyle\frac{\alpha^{2}}{24}\sum_{j}{\left(\psi_{j}^{4}+4{\rm i}\psi_{j}^{3}\overline{\psi}_{j}-10|\psi_{j}|^{4}+4{\rm i}\psi_{j}\overline{\psi}_{j}^{3}+\overline{\psi}_{j}^{4}\right)}+$ (3.71) $\displaystyle+$ $\displaystyle\frac{\beta}{16}\sum_{j}{\left(\psi_{j}^{4}-4{\rm i}\psi_{j}^{3}\overline{\psi}_{j}+6|\psi_{j}|^{4}-4{\rm i}\psi_{j}\overline{\psi}_{j}^{3}+\overline{\psi}_{j}^{4}\right)},$ and there exists $C_{9}(a,G_{1})$ such that $\left\|X_{H_{21}}(\psi,\bar{\psi})\right\|_{s-s_{1},\sigma}\leq C_{9}\mu^{s_{1}}\left\|\psi\right\|_{s,\sigma}^{3}$ (3.72) Just averaging (3.71) with respect to the flow $\Phi^{t}$ it is now immediate to get the following Corollary. ###### Corollary 3.5. The normal form $Z$ is composed of two terms, $Z=Z_{0}+Z_{r}$, where the leading term $Z_{0}$ is smooth and reads $Z_{0}(\psi)=\tilde{\gamma}\sum_{j}{|\psi_{j}|^{4}},\qquad\qquad\tilde{\gamma}:=\frac{3}{8}\left(\beta-\frac{10}{9}\alpha^{2}\right).$ (3.73) while the remainder is small $\left\|X_{Z_{r}}(\psi)\right\|_{s-s_{1},\sigma}\leq C_{9}\mu^{s_{1}}\left\|\psi\right\|_{s,\sigma}^{3}\quad\left\\{\begin{matrix}0\leq s_{1}<s-\frac{1}{2}<2\ ,\ \sigma=0\ \text{for\ DBC}\\\ 0\leq s_{1}<s-\frac{1}{2},\,s_{1}\leq 2,\,\sigma\geq 0\ \text{for\ PBC}\end{matrix}\right.$ (3.74) Thus we have proved that the formula for $Z_{0}$ holds. The formula for $\chi_{10}$ implies that the canonical transformation has the structure (3.15) and this concludes the proof. ## 4 Proof of Theorem 2.1 To discuss this issue we first write the equations of motion of the first part of the normal form, namely of $H_{0}+Z_{0}$, in the form ${\rm i}\dot{\psi}_{j}=(A\psi)_{j}-\tilde{\gamma}\psi_{j}|\psi_{j}|^{2}\ ,$ (4.1) where $A$ is a linear operator which in the Fourier variables acts as a multiplier by $\omega_{k}=1+\frac{a}{2}\mu^{2}k^{2}+O(\mu^{4}k^{4})$, namely $(\widehat{A\psi})_{k}=\omega_{k}\hat{\psi}_{k}=(1+\frac{a}{2}\mu^{2}k^{2})\psi_{k}+\mathcal{O}(\mu^{4})\equiv(\widehat{A_{NLS}\psi})_{k}+\mathcal{O}(\mu^{4})\ .$ (4.2) Take now an interpolating function for $\psi$, in other words a function $u$ such that $\psi_{j}=\epsilon u(\mu j)\ ,$ (4.3) where $\epsilon$ is a small parameter representing the amplitude. Then, up to corrections of higher order, $u$ fulfills the equation $-{\rm i}u_{t}=u+a\mu^{2}u_{xx}-\tilde{\gamma}\epsilon^{2}u|u|^{2}\ ,$ (4.4) which, up to a Gauge transformation and a scaling of the time introduced by $u(x,t)=e^{{\rm i}t}\varphi(x,\tau)\ ,\qquad\qquad\tau:=a\mu^{2}t,$ (4.5) gives the NLS equation ${\rm i}\varphi_{\tau}=-\varphi_{xx}+\gamma\varphi|\varphi|^{2},\qquad\qquad\gamma:=\frac{\tilde{\gamma}}{a}\frac{\epsilon^{2}}{\mu^{2}}.$ (4.6) In order to get a bounded value of $\gamma$, from now on we take $\epsilon=\mu$. We now compare an approximate solution constructed through NLS and the true solution of our Hamiltonian system. More explicitly, corresponding to a solution $\varphi^{a}(t)$ of the NLS with analytic initial datum, we define an approximate solution $\psi^{a}$ of the original model by $\psi_{j}^{a}(t)=\mu e^{{\rm i}t}\varphi^{a}(\mu j,a\mu^{2}t)\ .$ (4.7) We also consider the true solution $\psi(t)$ of the Hamilton equation of the original model, with initial datum $\psi_{j}^{0}=\mu\varphi^{a}(\mu j,0)$. From now on we will restrict to the case of DBC which is the complicate one. We first work in the variables in which $H$ is reduced to the normal form $H=H_{0}+Z_{0}+Z_{r}+{\mathcal{R}}.$ (4.8) ###### Lemma 4.1. Let $\psi$ be the solution of the equations of motion of (4.8) with initial datum $\psi(0)={\mathcal{T}}^{-1}(\psi^{0})$ and let $\psi^{a}$ be as defined in (4.7), then $\psi=\psi^{a}+\psi_{1}$ with $\left\|\psi_{1}(t)\right\|_{s,0}\leq C\mu^{2},\quad\frac{3}{2}<s<\frac{5}{2}\ ,\qquad|t|\leq\frac{T}{\mu^{2}}.$ ###### Proof. Observe that the NLS equation for $\psi^{a}$ may be rewritten as $\dot{\psi}^{a}=X_{H_{NLS,0}}(\psi^{a})+X_{Z}(\psi^{a})+X_{\mathcal{R}}(\psi^{a})-X_{{\mathcal{R}}_{1}}(\psi^{a})$ (4.9) where $\displaystyle{{\mathcal{R}}_{1}}$ $\displaystyle:=$ $\displaystyle\left({H_{0}}-{H_{NLS,0}}\right)+{Z_{r}}+{\mathcal{R}}$ $\displaystyle{H_{NLS,0}}(\hat{\psi})$ $\displaystyle:=$ $\displaystyle\sum_{k}\left(1+a\frac{1}{2}\mu^{2}k^{2}\right)\left|\hat{\psi}_{k}\right|^{2}\ ,$ so that $X_{{\mathcal{R}}_{1}}$ fulfills the estimate $\displaystyle\left\|X_{{\mathcal{R}}_{1}}(\psi^{a})\right\|_{s,0}$ $\displaystyle\leq$ $\displaystyle C_{1}\mu^{4}\left\|\psi^{a}\right\|_{s+4,0}+C_{r}\mu^{s_{1}}\left\|\psi^{a}\right\|^{3}_{s+s_{1},0}+C_{\mathcal{R}}\left\|\psi^{a}\right\|^{4}_{s,0}\leq$ $\displaystyle\leq$ $\displaystyle C_{4}\mu^{4}\ ,\qquad\qquad\frac{1}{2}<s<\frac{5}{2}\ .$ We compare $\psi^{a}$ with the full solution $\psi$ of the equation $\dot{\psi}=X_{H_{0}}(\psi)+X_{Z}(\psi)+X_{\mathcal{R}}(\psi)$ with initial datum $\psi_{0}={\mathcal{T}}^{-1}(\psi^{0})$, whose difference from $\psi^{0}$ (initial datum for $\psi^{a}$) is controlled by $\left\|\psi_{1}(0)\right\|_{s,0}=\left\|\psi^{0}-{\mathcal{T}}^{-1}(\psi^{0})\right\|_{s,0}\leq C_{\mathcal{T}}\left\|\psi^{0}\right\|_{s,0}^{2}\leq C_{\mathcal{T}}\mu^{2}.$ So we apply the Gronwall lemma (see lemma A.3) with $A:=X_{H_{0}}\ ,\qquad P=X_{Z}+X_{{\mathcal{R}}}\ ,\qquad R=X_{{\mathcal{R}}_{1}},$ obtaining that the error $\psi_{1}:=\psi-\psi^{a}$ from the NLS dynamics satisfies $\dot{\psi}_{1}=A\psi_{1}+\left[P(\psi^{a}+\psi_{1})-P(\psi^{a})\right]+X_{{\mathcal{R}}_{1}}(\psi^{a})$ (4.10) and is estimated by $\left\|\psi_{1}(t)\right\|_{s,\sigma^{\prime}}\leq C_{\mathcal{T}}\mu^{2}e^{C_{6}\mu^{2}t}+\frac{C_{4}\mu^{4}}{C_{6}\mu^{2}}\left(e^{C_{6}\mu^{2}t}-1\right)\leq C_{7}\mu^{2},\qquad|t|\leq T/\mu^{2}$ (4.11) where $C_{6}:=6C_{Z}$. ∎ When we go back to the original variables, the solution $z$ may be split as $z={\mathcal{T}}(\psi)={\mathcal{T}}(\psi^{a}+\psi_{1})=\mu z^{a}+\mu^{2}z_{1}+\mu^{3}z_{2}$ where we have defined $z^{a}:=\frac{\psi^{a}}{\mu},\qquad z_{1}:=\frac{\psi_{1}+X_{\chi_{01}}(\psi^{a})}{\mu^{2}},\qquad\qquad z_{2}:=\frac{z-\mu z^{a}-\mu^{2}z_{1}}{\mu^{3}}.$ (4.12) More precisely, we claim that it holds ###### Lemma 4.2. We have $\left\|z_{1}\right\|_{s,0}\leq C_{1},\qquad\qquad\left\|z_{2}\right\|_{s,0}\leq C_{2},$ up to times s.t. $|t|\leq\frac{T}{\mu^{2}}$. ###### Proof. The first inequality comes directly from the lemma 4.1. Concerning the second one, we remark that $\displaystyle z$ $\displaystyle=$ $\displaystyle{\mathcal{T}}(\psi^{a}+\psi_{1})=\psi^{a}+\psi_{1}+\left[{\mathcal{T}}(\psi^{a}+\psi_{1})-(\psi^{a}+\psi_{1})\right]=$ $\displaystyle=$ $\displaystyle\psi^{a}+\psi_{1}+X_{\chi_{1}}(\psi^{a}+\psi_{1})+{\mathcal{O}}(\left\|\psi^{a}+\psi_{1}\right\|_{s,0}^{3})=$ $\displaystyle=$ $\displaystyle\psi^{a}+\psi_{1}+X_{\chi_{1}}(\psi^{a})+{\mathcal{O}}(\left\|\psi_{1}\right\|_{s,0}\left\|\psi^{a}\right\|_{s,0}+\left\|\psi^{a}+\psi_{1}\right\|_{s,0}^{3})$ (by differentiability of $X_{\chi_{1}}$ and Lagrange mean value theorem). Finally from (3.66) with $s_{1}=1$ we have $X_{\chi_{1}}(\psi^{a})=X_{\chi_{01}}(\psi^{a})+{\mathcal{O}}(\mu\left\|\psi^{a}\right\|^{2}_{s+1,0})$ ∎ We now analyze the first correction $z_{1}$. To this end we analyze separately its two terms. First remark that, from (4.10) one has $\psi_{1}=\psi_{10}+{\mathcal{O}}(|t|\left\|\psi^{a}\right\|^{3}_{s,0})$, where $\psi_{10}$ solves the equation $\dot{\psi}_{10}=A\psi_{10}$ with initial datum ${\mathcal{T}}^{-1}(\psi^{0})-\psi^{0}$. Thus we have $\psi_{10}={\rm e}^{At}({\mathcal{T}}^{-1}(\psi^{0})-\psi^{0})=-{\rm e}^{At}X_{\chi_{01}}(\psi^{0})+{\mathcal{O}}(\mu\left\|\psi^{0}\right\|^{2}_{s+1,0})$ (4.13) We now analyze the other term. To this end, with the aim of considering the short time dynamics, we rewrite the equation (4.9) as $\dot{\psi}^{a}=i\psi^{a}+{\mathcal{O}}(\mu^{3})\quad\iff\quad\psi^{a}={\rm e}^{{\rm i}t}\psi^{0}+{\mathcal{O}}(\mu^{3}|t|)\ .$ Thus exploiting the differentiability of $X_{\chi_{01}}$ we have $\mu^{2}z_{1}=-{\rm e}^{At}X_{\chi_{01}}(\psi^{0})+X_{\chi_{01}}\left({\rm e}^{{\rm i}t}\psi^{0}\right)+{\mathcal{O}}(\mu^{3}|t|)\ .$ (4.14) Since $\displaystyle{\rm e}^{At}X_{\chi_{01}}(\psi^{0})$ $\displaystyle=$ $\displaystyle-\frac{\alpha}{6\sqrt{2}}\left[3e^{At}\psi_{0}^{2}+6{\rm i}e^{At}|\psi_{0}|^{2}+e^{At}\overline{\psi}_{0}^{2}\right],$ $\displaystyle X_{\chi_{01}}\left({\rm e}^{{\rm i}t}\psi^{0}\right)$ $\displaystyle=$ $\displaystyle-\frac{\alpha}{6\sqrt{2}}\left[3\psi_{0}^{2}e^{2{\rm i}t}+6{\rm i}|\psi_{0}|^{2}+e^{-2{\rm i}t}\overline{\psi}_{0}^{2}\right]$ which yields to $\mu^{2}z_{1}=-\frac{\alpha}{6\sqrt{2}}\left[3(e^{At}-e^{2{\rm i}t})\psi_{0}^{2}+6{\rm i}(e^{At}-1)|\psi_{0}|^{2}+(e^{At}-e^{-2{\rm i}t})\overline{\psi}_{0}^{2}\right]+{\mathcal{O}}(\mu^{3}|t|).$ In the case of $\psi_{0}={\rm i}\mu z_{0}$ (zero velocity initial datum) we have $z_{1}=\frac{\alpha}{6\sqrt{2}}\left[4e^{At}-3e^{2{\rm i}t}-e^{-2{\rm i}t}-6{\rm i}e^{At}+6{\rm i}\right]z_{0}^{2}+{\mathcal{O}}(\mu|t|)$ which gives immediately the thesis. ## Appendix A Appendix: a few technical lemmas. ###### Lemma A.1. Let $u\in H^{r,\sigma}$ and $v\in H^{s,\sigma}$ with $s>\frac{1}{2}$ and $s\geq r\geq 0$, $\sigma\geq 0$. Then there exists $C=C(r,s)$ such that the following inequality holds $\left\|uv\right\|_{r,\sigma}\leq C\left\|u\right\|_{r,\sigma}\left\|v\right\|_{s,\sigma}$ (A.1) ###### Proof. In this proof it is useful to use the expansion of $u$ and $v$ on the complex exponentially. Thus we will write $u(x)=\sum_{k\in{\mathbb{Z}}}\hat{u}_{k}{\rm e}^{{\rm i}kx}/\sqrt{2\pi}$, and remark that if in the definition of the norm cf (2.14) we substitute such coefficients to the coefficients on the real Fourier basis, nothing changes. This is due to the fact that both the basis of the complex exponentials and the real Fourier basis are orthonormal. The advantage is that in terms of the complex exponentials the product is mapped into the convolution of the Fourier coefficients, thus we have simply to estimate the norm of the function whose Fourier coefficients are $(\hat{u}*\hat{v})_{k}=\sum_{j}\hat{u}_{j-k}\hat{v}_{k}\ .$ (A.2) As a preliminary fact we define the quantities $\gamma_{j,k}=\frac{[j-k][k]^{\frac{s}{r}}}{[j]}\ ,$ and prove that there exists a constant $C(s,r)$ such that $\sum_{k}{\frac{1}{\gamma_{j,k}^{2r}}}<C$ (A.3) To obtain (A.3) we need some preliminary inequalities. For any positive $a$ and $b$ one has $(a+b)^{2r}\leq 2^{2r}\max\left\\{a;b\right\\}^{2r}<4^{r}(a^{2r}+b^{2r})$ (A.4) and for any $j$ and $k$ in ${\mathbb{Z}}$ $\frac{1}{[j-k]^{2r}[k]^{2s-2r}}<\frac{1}{(\min\\{[j-k],[k]\\})^{2s}}<\frac{1}{[j-k]^{2s}}+\frac{1}{[k]^{2s}}.$ (A.5) From (A.4), (A.5) and $[j]<[j-k]+[k]$ it follows $\displaystyle\frac{1}{\gamma_{j,k}^{2r}}$ $\displaystyle\leq$ $\displaystyle\left(\frac{[j-k]+[k]}{[j-k][k]^{\frac{s}{r}}}\right)^{2r}\leq\left(\frac{1}{[k]^{\frac{s}{r}}}+\frac{1}{[j-k][k]^{\frac{s}{r}-1}}\right)^{2r}\leq$ $\displaystyle\leq$ $\displaystyle 4^{r}\left(\frac{1}{[k]^{2s}}+\frac{1}{[j-k]^{2r}[k]^{2s-2r}}\right)\leq$ $\displaystyle\leq$ $\displaystyle 4^{r}\left(\frac{1}{[k]^{2s}}+\frac{1}{[j-k]^{2s}}+\frac{1}{[k]^{2s}}\right)$ which gives (A.3) with $C=3\times 4^{r}\times\sum_{k\in{\mathbb{Z}}}\frac{1}{[k]^{2s}}.$ Hence $\displaystyle\left\|uv\right\|_{r,\sigma}^{2}$ $\displaystyle=$ $\displaystyle\sum_{j}{[j]^{2r}e^{2\sigma|j|}|\sum_{k}{\hat{u}_{j-k}\hat{v}_{k}}|^{2}}\leq$ $\displaystyle\leq$ $\displaystyle\sum_{j}{[j]^{2r}e^{2\sigma|j|}\left(\sum_{k}{\frac{1}{\gamma_{j,k}^{2r}}}\right)\left(\sum_{k}{\gamma_{j,k}^{2r}|\hat{u}_{j-k}\hat{v}_{k}|^{2}}\right)}\leq$ $\displaystyle\leq$ $\displaystyle C^{2}\sum_{j,k}{[j-k]^{2r}e^{2\sigma|j-k|}|\hat{u}_{k-j}|^{2}[k]^{2s}e^{2\sigma|k|}|\hat{v}_{k}|^{2}}\leq$ $\displaystyle\leq$ $\displaystyle C^{2}\left(\sum_{l}{[l]^{2r}e^{2\sigma|l|}|\hat{u}_{l}|^{2}}\right)\left(\sum_{k}{[k]^{2r}e^{2\sigma|k|}|\hat{u}_{k}|^{2}}\right),$ which concludes the proof. ∎ We state here a version of the Gronwall Lemma which is suited for our estimates. First we recall the following lemma. ###### Lemma A.2. Let $x:[0,T]\rightarrow{\mathcal{P}}$ be a differentiable function and ${\mathcal{P}}$ a Banach space. Assume that $\forall t\in[0,T]$ it fulfills the integral inequality $\left\|x(t)\right\|\leq K+\int_{0}^{t}{\left(a\left\|x(s)\right\|+b\right)ds}$ (A.6) with $a,b$ real and non negative parameters, then $\left\|x(t)\right\|\leq e^{at}K+\frac{b}{a}\left(e^{at}-1\right)\ .$ (A.7) The lemma we use in sect. 4 is the following one. ###### Lemma A.3. Let $z(t),z^{a}(t)\in{\mathcal{P}}$, $t\in[-T,T]$ be respectively the solutions of $\begin{cases}\dot{z}=Az+P(z),\\\ z(0)=z_{0}\end{cases},\qquad\qquad\begin{cases}\dot{z}^{a}=Az^{a}+P(z^{a})-R(z^{a}),\\\ z^{a}(0)=z^{a}_{0}\end{cases}$ where $A$ is the generator of a unitary group in ${\mathcal{P}}$, and $\left\|z^{a}(t)\right\|\leq C$. Assume also that the non-linearity $P$ has a zero of third order at the origin and that for all $t\in[-T,T]$ and all $z$ with $\left\|z\right\|\leq 2C$ $\left\|P(z)\right\|\leq\rho_{1}\left\|z\right\|^{3},\qquad\qquad\left\|dP(z)\right\|\leq 3\rho_{1}\left\|z\right\|^{2},$ and the remainder $R$ is estimated by $\left\|R(z^{a}(t))\right\|\leq\rho_{2},\qquad\forall t\in[-T,T]\ .$ Let $\delta:=z-z^{a}$, then the following estimate holds $\left\|\delta(t)\right\|\leq\left\|\delta(0)\right\|e^{\rho_{1}t}+\frac{\rho_{2}}{\rho_{1}}\left(e^{\rho_{1}t}-1\right),\qquad\qquad t\in[-T,T]\ .$ (A.8) ###### Proof. The difference $\delta(t)$ is solution of the differential equation $\dot{\delta}=A\delta+\left(P(z^{a}(t)+\delta)-P(z^{a}(t))\right)+R(z^{a}(t))\ ,$ by Duhamel formula one has $\delta(t)=e^{At}\delta(0)+e^{At}\int_{0}^{t}e^{-As}\left[\left(P(z^{a}(s)+\delta)-P(z^{a}(s))\right)+R(z^{a}(s))\right]ds$ Using Lagrange mean value theorem to estimate $P(z^{a}(s)+\delta)-P(z^{a}(s))$ and the fact that $A$ is unitary one has $\left\|\delta(t)\right\|\leq\left\|\delta(0)\right\|+\int_{0}^{t}{\left[3\rho_{1}\left\|\delta(s)\right\|+\rho_{2}\right]ds},$ which fulfills (A.6), from which the thesis follows.∎ ## References * [BDGS07] D. Bambusi, J.M. Delort, B. Grébert, and J. Szeftel, _Almost global existence for Hamiltonian semi-linear Klein-Gordon equations with small Cauchy data on Zoll manifolds_ , Comm. Pure Appl. Math. 60 (2007), no. 11, 1665,1690. * [BG93] D. Bambusi and A. Giorgilli, _Exponential stability of states close to resonance in infinite-dimensional Hamiltonian systems_ , J. Statist. Phys. 71 (1993), no. 3-4, 569–606. * [BG08] G. Benettin and G. Gradenigo, _A study of the Fermi Pasta Ulam problem in dimension two_ , Chaos (2008). * [BMP07] D. Bambusi, D. Muraro, and T. Penati, _Numerical studies on boundary effects on the FPU paradox_ , Phys. Lett. A (2007). * [DP03] Isabelle Daumont and Michel Peyrard, _One-dimensional “turbulence” in a discrete lattice_ , Chaos 13 (2003), no. 2, 624–636. * [Fla98] S. Flach, _Breathers on lattices with long range interaction_ , Phys. Rev. E 58 (1998), no. 4, R4116–R4119. * [FT87] L. D. Faddeev and L. A. Takhtajan, _Hamiltonian methods in the theory of solitons_ , Springer Series in Soviet Mathematics, Springer-Verlag, Berlin, 1987\. * [GF05] A. V. Gorbach and S. Flach, _Compactlike discrete breathers in systems with nonlinear and nonlocal dispersive terms_ , Phys. Rev. E (3) 72 (2005), no. 5, 056607, 9. * [GK03] B. Grébert and T. Kappeler, _Perturbation of the defocusing Nonlinear Schrödinger Equation_ , Milan J. of Math. 71 (2003), 141–174. * [Kal89] L. A. Kalyakin, _Long-wave asymptotics. Integrable equations as the asymptotic limit of nonlinear systems_ , Uspekhi Mat. Nauk 44 (1989), no. 1(265), 5–34, 247. * [KSM92] Pius Kirrmann, Guido Schneider, and Alexander Mielke, _The validity of modulation equations for extended systems with cubic nonlinearities_ , Proc. Roy. Soc. Edinburgh Sect. A 122 (1992), no. 1-2, 85–91. * [Pey04] Michel Peyrard, _The statistical distributions of one-dimensional “turbulence”_ , Phys. D 193 (2004), no. 1-4, 265–277. * [Sch98] G. Schneider, _Justification of modulation equations for hyperbolic systems via normal forms_ , NoDEA Nonlinear Differential Equations Appl. 5 (1998), no. 1, 69–82. * [Tem91] Roger Temam, _Approximation of attractors, large eddy simulations and multiscale methods_ , Proc. Roy. Soc. London Ser. A 434 (1991), no. 1890, 23–39, Turbulence and stochastic processes: Kolmogorov’s ideas 50 years on.

arxiv-papers

2024-09-04T02:48:53.783061

0802.2314

# Injectivity on the set of conjugacy classes of some monomorphisms between Artin groups Eon-Kyung Lee and Sang-Jin Lee Department of Mathematics, Sejong University, Seoul, 143-747, Korea eonkyung@sejong.ac.kr Department of Mathematics, Konkuk University, Seoul, 143-701, Korea sangjin@konkuk.ac.kr ###### Abstract. There are well-known monomorphisms between the Artin groups of finite type $\mathbf{A}_{n}$, $\mathbf{B}_{n}=\mathbf{C}_{n}$ and affine type $\tilde{\mathbf{A}}_{n-1}$, $\tilde{\mathbf{C}}_{n-1}$. The Artin group $A(\mathbf{A}_{n})$ is isomorphic to the $(n+1)$-strand braid group $B_{n+1}$, and the other three Artin groups are isomorphic to some subgroups of $B_{n+1}$. The inclusions between these subgroups yield monomorphisms $A(\mathbf{B}_{n})\to A(\mathbf{A}_{n})$, $A(\tilde{\mathbf{A}}_{n-1})\to A(\mathbf{B}_{n})$ and $A(\tilde{\mathbf{C}}_{n-1})\to A(\mathbf{B}_{n})$. There are another type of monomorphisms $A(\mathbf{B}_{d})\to A(\mathbf{A}_{md-1})$, $A(\mathbf{B}_{d})\to A(\mathbf{B}_{md})$ and $A(\mathbf{B}_{d})\to A(\mathbf{A}_{md})$ which are induced by isomorphisms between Artin groups of type $\mathbf{B}$ and centralizers of periodic braids. In this paper, we show that the monomorphisms $A(\mathbf{B}_{d})\to A(\mathbf{A}_{md-1})$, $A(\mathbf{B}_{d})\to A(\mathbf{B}_{md})$ and $A(\mathbf{B}_{d})\to A(\mathbf{A}_{md})$ induce injective functions on the set of conjugacy classes, and that none of the monomorphisms $A(\mathbf{B}_{n})\to A(\mathbf{A}_{n})$, $A(\tilde{\mathbf{A}}_{n-1})\to A(\mathbf{B}_{n})$ and $A(\tilde{\mathbf{C}}_{n-1})\to A(\mathbf{B}_{n})$ does so. _Key words:_ Artin group, braid group, conjugacy class, Nielsen-Thurston classification _MSC:_ 20F36, 20F10 ## 1\. Introduction Let $M$ be a symmetric $n\times n$ matrix with entries $m_{ij}\in\\{2,3,\ldots,\infty\\}$ for $i\neq j$, and $m_{ii}=1$ for $1\leq i\leq n$. The _Artin group_ of type $M$ is defined by the presentation $A(M)=\langle s_{1},\ldots,s_{n}\mid\underbrace{s_{i}s_{j}s_{i}\cdots}_{m_{ij}}=\underbrace{s_{j}s_{i}s_{j}\cdots}_{m_{ij}}\quad\mbox{for all $i\neq j$, $m_{ij}\neq\infty$}\rangle.$ The _Coxeter group_ $W(M)$ of type $M$ is the quotient of $A(M)$ by the relation $s_{i}^{2}=1$. The Artin group $A(M)$ is said to be _of finite type_ if $W(M)$ is finite, and _of affine type_ if $W(M)$ acts as a proper, cocompact group of isometries on some Euclidean space with the generators $s_{1},\ldots,s_{n}$ acting as affine reflections. Artin groups are usually described by _Coxeter graphs_ , whose vertices are numbered $1,\ldots,n$ and which has an edge labelled $m_{ij}$ between the vertices $i$ and $j$ whenever $m_{ij}\geq 3$ or $m_{ij}=\infty$. The label 3 is usually suppressed. There are well-known monomorphisms between the Artin groups of finite type $\mathbf{A}_{n}$, $\mathbf{B}_{n}=\mathbf{C}_{n}$, and affine type $\tilde{\mathbf{A}}_{n-1}$, $\tilde{\mathbf{C}}_{n-1}$. The Coxeter graphs associated to these Artin groups are as follows. $\mathbf{A}_{n}$ | | | $\tilde{\mathbf{A}}_{n-1}$ | ---|---|---|---|--- $\mathbf{B}_{n}$ | | | $\tilde{\mathbf{C}}_{n-1}$ | In this paper, we investigate whether these monomorphisms induce injective functions on the set of conjugacy classes or not. Before stating our results, let us review briefly braid groups and monomorphisms between these Artin groups. Let $D^{2}=\\{z\in\mathbb{C}\mid|z|\leq n+1\\}$, and let $D_{n}$ be the $n$-punctured disk $D^{2}\setminus\\{1,\ldots,n\\}$. The braid group $B_{n}$ is defined as the group of self-diffeomorphisms of $D_{n}$ that fix the boundary pointwise, modulo isotopy relative to the boundary. Equivalently, $n$-braids can be defined as isotopy classes of collections of pairwise disjoint $n$ strands $l=l_{1}\cup\cdots\cup l_{n}$ in $D^{2}\times[0,1]$ such that $l\cap(D^{2}\times\\{t\\})$ consists of $n$ points for each $t\in[0,1]$, and $l\cap(D^{2}\times\\{0,1\\})=\\{1,\ldots,n\\}\times\\{0,1\\}$. The admissible isotopies lie in the interior of $D^{2}\times[0,1]$. The $n$-braid group $B_{n}$ has the well-known Artin presentation [Art25]: $B_{n}=\left\langle\sigma_{1},\ldots,\sigma_{n-1}\biggm{|}\begin{array}[]{ll}\sigma_{i}\sigma_{j}=\sigma_{j}\sigma_{i}&\mbox{if }|i-j|\geq 2\\\ \sigma_{i}\sigma_{j}\sigma_{i}=\sigma_{j}\sigma_{i}\sigma_{j}&\mbox{if }|i-j|=1\end{array}\right\rangle.$ ###### Definition 1.1. For $\alpha\in B_{n}$, let $\pi_{\alpha}$ denote its induced permutation on $\\{1,\ldots,n\\}$. If $\pi_{\alpha}(i)=i$, we say that $\alpha$ is _$i$ -pure_, or the $i$-th strand of $\alpha$ is _pure_. For $P\subset\\{1,\ldots,n\\}$, we say that $\alpha$ is _$P$ -pure_ if $\alpha$ is $i$-pure for each $i\in P$. ###### Definition 1.2. Let $B_{n,1}$ denote the subgroup of $B_{n}$ consisting of 1-pure braids. Let $\operatorname{lk}:B_{n,1}\to\mathbb{Z}$ be the homomorphism measuring the linking number of the first strand with the other strands: $B_{n,1}$ is generated by $\sigma_{1}^{2},\sigma_{2},\ldots,\sigma_{n-1}$, and $\operatorname{lk}$ is defined by $\operatorname{lk}(\sigma_{1}^{2})=1$ and $\operatorname{lk}(\sigma_{i})=0$ for $i\geq 2$. A braid $\alpha$ is said to be _1-unlinked_ if it is 1-pure and $\operatorname{lk}(\alpha)=0$. Let $\nu:B_{n,1}\to B_{n-1}$ be the homomorphism deleting the first strand. For example, the braid in Figure 1 is $\\{1,4,5\\}$-pure and 1-unlinked. Note that $\\{1,\ldots,n\\}$-pure braids are nothing more than pure braids in the usual sense, and that $\operatorname{lk}$ is a conjugacy invariant of 1-pure braids because $\operatorname{lk}(\beta\alpha\beta^{-1})=\operatorname{lk}(\beta)+\operatorname{lk}(\alpha)-\operatorname{lk}(\beta)=\operatorname{lk}(\alpha)$ for $\alpha,\beta\in B_{n,1}$. Figure 1. This braid is $\\{1,4,5\\}$-pure and 1-unlinked. From the presentation of $B_{n}$, it is obvious that the Artin group $A(\mathbf{A}_{n})$ is isomorphic to $B_{n+1}$. The other Artin groups $A(\mathbf{B}_{n})$, $A(\tilde{\mathbf{A}}_{n-1})$ and $A(\tilde{\mathbf{C}}_{n-1})$ are known to be isomorphic to subgroups of $B_{n+1}$ as follows [All02, CC05, BM07]. $\displaystyle A(\mathbf{B}_{n})$ $\displaystyle\simeq$ $\displaystyle\\{\alpha\in B_{n+1}\mid\mbox{$\alpha$ is 1-pure}\\}=B_{n+1,1}\subset B_{n+1};$ $\displaystyle A(\tilde{\mathbf{A}}_{n-1})$ $\displaystyle\simeq$ $\displaystyle\\{\alpha\in B_{n+1}\mid\mbox{$\alpha$ is 1-unlinked}\\}\subset B_{n+1,1};$ $\displaystyle A(\tilde{\mathbf{C}}_{n-1})$ $\displaystyle\simeq$ $\displaystyle\\{\alpha\in B_{n+1}\mid\mbox{$\alpha$ is $\\{1,n+1\\}$-pure}\\}\subset B_{n+1,1}.$ From now on, we identify $A(\mathbf{A}_{n})$ with $B_{n+1}$, and the Artin groups $A(\mathbf{B}_{n})$, $A(\tilde{\mathbf{A}}_{n-1})$ and $A(\tilde{\mathbf{C}}_{n-1})$ with the corresponding subgroups of $B_{n+1}$ induced by the above isomorphisms. Then the inclusions between the subgroups of $B_{n+1}$ yield monomorphisms $\psi_{1}:A(\mathbf{B}_{n})\to A(\mathbf{A}_{n})$, $\psi_{2}:A(\tilde{\mathbf{A}}_{n-1})\to A(\mathbf{B}_{n})$ and $\psi_{3}:A(\tilde{\mathbf{C}}_{n-1})\to A(\mathbf{B}_{n})$. See Figure 2. $\textstyle{A(\mathbf{A}_{n})}$$\textstyle{A(\mathbf{B}_{n})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\psi_{1}}$$\textstyle{A(\tilde{\mathbf{A}}_{n-1})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\psi_{2}}$$\textstyle{A(\tilde{\mathbf{C}}_{n-1})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\psi_{3}}$ --- $\textstyle{A(\mathbf{A}_{md-1})}$$\textstyle{A(\mathbf{A}_{md})}$$\textstyle{A(\mathbf{B}_{md})\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\nu}$$\scriptstyle{\psi_{1}}$$\textstyle{A(\mathbf{B}_{d})\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\psi_{4}}$$\scriptstyle{\psi_{5}}$$\scriptstyle{\psi_{6}}$ Figure 2. $\psi_{1},\ldots,\psi_{6}$ are monomorphisms between Artin groups. There are another type of monomorphisms which can be described by isomorphisms between Artin groups of type $\mathbf{B}$ and centralizers of periodic braids. The well-known Nielsen-Thurston classification of mapping classes of surfaces (possibly with punctures) with negative Euler characteristic into periodic, reducible and pseudo-Anosov ones [Thu88] yields an analogous classification of braids. An $n$-braid $\alpha$ is said to be _periodic_ if some power of it is a power of $\Delta^{2}$, where $\Delta=\sigma_{1}(\sigma_{2}\sigma_{1})\cdots(\sigma_{n-1}\cdots\sigma_{1})$; _reducible_ if there is an essential curve system in $D_{n}$ whose isotopy class is invariant under the action of $\alpha$; _pseudo-Anosov_ if there is a pseudo-Anosov diffeomorphism $f$ defined on the interior of $D_{n}$ representing $\alpha$ modulo $\Delta^{2}$, that is, there exist a pair of transverse measured foliations $(F^{s},\mu^{s})$ and $(F^{u},\mu^{u})$ and a real $\lambda>1$, called _dilatation_ , such that $f(F^{s},\mu^{s})=(F^{s},\lambda^{-1}\mu^{s})$ and $f(F^{u},\mu^{u})=(F^{u},\lambda\mu^{u})$. Every braid belongs to one of the following three mutually disjoint classes: periodic; pseudo-Anosov; non- periodic and reducible. We say that two braids are _of the same Nielsen- Thurston type_ if both of them belong to the same class. Let $\delta=\sigma_{n-1}\cdots\sigma_{1}$ and $\epsilon=\delta\sigma_{1}$, then $\delta^{n}=\Delta^{2}=\epsilon^{n-1}$. If we need to specify the braid index $n$, we will write $\delta=\delta_{(n)}$, $\epsilon=\epsilon_{(n)}$ and $\Delta=\Delta_{(n)}$. The braids $\delta$ and $\epsilon$ are induced by rigid rotations of the punctured disk as in Figure 3 when the punctures are at the origin or evenly distributed on a round circle centered at the origin. Due to L.E.J. Brouwer [Bro19], B. de Kérékjártó [Ker19] and S. Eilenberg [Eil34] (see also [CK94]), it is known that an $n$-braid is periodic if and only if it is conjugate to a power of $\delta$ or $\epsilon$. | | ---|---|--- (a) $\delta_{(n)}=\sigma_{n-1}\sigma_{n-2}\cdots\sigma_{1}\in B_{n}$ | | (b) $\epsilon_{(n)}=\delta_{(n)}\sigma_{1}\in B_{n}$ Figure 3. The braids $\delta$ and $\epsilon$ are represented by rigid rotations of punctured disks. For a group $G$ and an element $g\in G$, let $Z_{G}(g)$ denote the centralizer of $g$ in $G$, that is, $Z_{G}(g)=\\{h\in G:gh=hg\\}$. If the group $G$ is clear from the context, we write simply $Z(g)$. The following theorem is a consequence of the result of D. Bessis, F. Digne and J. Michel [BDM02] on irreducible complex reflection groups. See also [GW04]. ###### Theorem 1.3 ([BDM02]). Let $\omega\in B_{n}$ be conjugate to $\delta^{k}$ (resp. $\epsilon^{k}$), where $k$ is not a multiple of $n$ (resp. $n-1$). Then the centralizer $Z(\omega)$ is isomorphic to the Artin group $A(\mathbf{B}_{d})$, where $d=\gcd(k,n)$ (resp. $d=\gcd(k,n-1)$). Let $m\geq 2$ and $d\geq 1$. The centralizer $Z(\epsilon_{(md+1)}^{d})$ consists of 1-pure braids because the first strand of $\epsilon_{(md+1)}^{d}$ is the only pure strand. By theorem 1.3, we have $\displaystyle A(\mathbf{B}_{d})$ $\displaystyle\simeq$ $\displaystyle Z(\epsilon_{(md+1)}^{d})\subset B_{md+1,1}=A(\mathbf{B}_{md})\subset B_{md+1}=A(\mathbf{A}_{md});$ $\displaystyle A(\mathbf{B}_{d})$ $\displaystyle\simeq$ $\displaystyle Z(\delta_{(md)}^{d})\subset B_{md}=A(\mathbf{A}_{md-1}).$ Compositions of the above isomorphisms and inclusions yield monomorphisms $\psi_{4}:A(\mathbf{B}_{d})\to A(\mathbf{B}_{md})$, $\psi_{5}:A(\mathbf{B}_{d})\to A(\mathbf{A}_{md-1})$ and $\psi_{6}:A(\mathbf{B}_{d})\to A(\mathbf{A}_{md})$. The monomorphisms $\psi_{4}$ and $\psi_{5}$ in [BDM02] and [GW04] are such that $\psi_{4}(\epsilon_{(d+1)})=\epsilon_{(md+1)}$, $\psi_{5}(\epsilon_{(d+1)})=\delta_{(md)}$ and $\psi_{5}=\nu\circ\psi_{4}$, where $\nu:B_{md+1,1}\to B_{md}$ is the homomorphism deleting the first strand (see Definition 1.2). By definition, $\psi_{6}=\psi_{1}\circ\psi_{4}$. See Figure 2. In [BGG06] J. Birman, V. Gebhardt and J. González-Meneses described explicitly in terms of standard generators the monomorphism $\psi_{5}:A(\mathbf{B}_{d})\to A(\mathbf{A}_{md-1})$ when $m=2$. The following theorem is the main result of this paper. ###### Theorem 1.4. Let $\omega$ be a periodic $n$-braid, and let $\alpha,\beta\in Z(\omega)$. If $\alpha$ and $\beta$ are conjugate in $B_{n}$, then they are conjugate in $Z(\omega)$. We prove the above theorem, relying on the Nielsen-Thurston classification of braids. In the case of pseudo-Anosov braids, we obtain a stronger result: if $\alpha,\beta\in Z(\omega)$ are pseudo-Anosov and conjugate in $B_{n}$, then any conjugating element from $\alpha$ to $\beta$ belongs to $Z(\omega)$ (see Proposition 5.1). The case of periodic braids is easy to prove due to the uniqueness of roots up to conjugacy for Artin groups of type $\mathbf{B}$ [LL07], hence most part of this paper is devoted to the case of reducible braids. Using the characterization of $Z(\omega)$ as the Artin group of type $\mathbf{B}$, we obtain the following corollary which is equivalent to the above theorem (see Lemma 4.3 for the equivalence). ###### Corollary 1.5. For $m\geq 2$ and $d\geq 1$, the monomorphisms $\psi_{5}:A(\mathbf{B}_{d})\to A(\mathbf{A}_{md-1})$ and $\psi_{6}:A(\mathbf{B}_{d})\to A(\mathbf{A}_{md})$ induce injective functions on the set of conjugacy classes. The above corollary implies that $\psi_{4}$ is also injective on the set of conjugacy classes because $\psi_{5}=\nu\circ\psi_{4}$ and $\psi_{6}=\psi_{1}\circ\psi_{4}$. We also investigate the injectivity for the well-known monomorphisms of $A(\tilde{\mathbf{A}}_{n-1})$ and $A(\tilde{\mathbf{C}}_{n-1})$ into $A(\mathbf{B}_{n})$, and a further monomorphism of $A(\mathbf{B}_{n})$ into $A(\mathbf{A}_{n})$, and have the following. ###### Theorem 1.6. For $n\geq 3$, none of the monomorphisms $\psi_{1}:A(\mathbf{B}_{n})\to A(\mathbf{A}_{n})$, $\psi_{2}:A(\tilde{\mathbf{A}}_{n-1})\to A(\mathbf{B}_{n})$ and $\psi_{3}:A(\tilde{\mathbf{C}}_{n-1})\to A(\mathbf{B}_{n})$ induces an injective function on the set of conjugacy classes. For the proof of the above theorem, we present three examples which show that the monomorphisms $\psi_{1}$, $\psi_{2}$ and $\psi_{3}$ are not injective on the set of conjugacy classes. For instance, the first example is a pair of elements in $B_{n+1,1}$ that are conjugate in $B_{n+1}$, but not conjugate in $B_{n+1,1}$. We close this section with a remark on a monomorphism from $A(\mathbf{A}_{n-1})$ to $A(\mathbf{D}_{n})$. The Coxeter graph of type $\mathbf{D}_{n}$ is as follows. $\mathbf{D}_{n}$ | ---|--- Let $\sigma_{1},\ldots,\sigma_{n-1}$ and $t_{1},\ldots,t_{n}$ be the standard generators of $A(\mathbf{A}_{n-1})$ and $A(\mathbf{D}_{n})$. Let $\iota:A(\mathbf{A}_{n-1})\to A(\mathbf{D}_{n})$ and $\pi:A(\mathbf{D}_{n})\to A(\mathbf{A}_{n-1})$ be the homomorphisms defined by $\iota(\sigma_{i})=t_{i+1}$ for $1\leq i\leq n-1$; $\pi(t_{1})=\sigma_{1}$; $\pi(t_{i})=\sigma_{i-1}$ for $i\geq 2$. Because $\pi\circ\iota$ is the identity, $\pi$ is an epimorphism and $\iota$ is a monomorphism, further, $\iota$ is injective on the set of conjugacy classes. It is known by J. Crisp and L. Paris that $A(\mathbf{D}_{n})$ is isomorphic to a semidirect product $F_{n-1}\rtimes A(\mathbf{A}_{n-1})$, where $F_{n-1}$ is a free group of rank $n-1$ [CP05]. ## 2\. Preliminaries In this section, we review the uniqueness of roots up to conjugacy in some Artin groups, and some tools useful in handling reducible braids such as canonical reduction system and standard reduction system. ### 2.1. Periodic braids The center of $B_{n}$ is the infinite cyclic group generated by $\Delta^{2}$. An $n$-braid is periodic if and only if it is of finite order in the central quotient $B_{n}/\langle\Delta^{2}\rangle$. Note that $\delta$ generates a cyclic group of order $n$ in $B_{n}/\langle\Delta^{2}\rangle$, hence $\delta^{a}$ and $\delta^{b}$ generate the same group in $B_{n}/\langle\Delta^{2}\rangle$ if and only if $\gcd(a,n)=\gcd(b,n)$. In fact, we have the following. ###### Lemma 2.1. Let $a$ and $b$ be integers. In $B_{n}$, the following hold. * (i) $Z(\delta^{a})=Z(\delta^{b})$ if and only if $\gcd(a,n)=\gcd(b,n)$. * (ii) $Z(\epsilon^{a})=Z(\epsilon^{b})$ if and only if $\gcd(a,n-1)=\gcd(b,n-1)$. * (iii) $Z(\delta^{a})=Z(\delta^{\gcd(a,n)})$ and $Z(\epsilon^{a})=Z(\epsilon^{\gcd(a,n-1)})$. In the above, (iii) is a direct consequence of (i) and (ii). The central quotient $B_{n}/\langle\Delta^{2}\rangle$ is the group of isotopy classes of self-diffeomorphisms of $D_{n}$ without the condition of fixing $\partial D_{n}$ pointwise on self-diffeomorphisms and isotopies. If a diffeomorphism $f:D_{n}\to D_{n}$ does not fix $\partial D_{n}$ pointwise, it determines a braid modulo $\Delta^{2}$. ### 2.2. Uniqueness of roots up to conjugacy in some Artin groups It is known that the $k$-th roots of a braid are unique up to conjugacy by J. González-Meneses [Gon03], and this is generalized to the Artin groups $A(\mathbf{B}_{n})=A(\mathbf{C}_{n})$, $A(\tilde{\mathbf{A}}_{n})$ and $A(\tilde{\mathbf{C}}_{n})$ [LL07]. ###### Theorem 2.2 (J. González-Meneses [Gon03]). Let $\alpha$ and $\beta$ be $n$-braids such that $\alpha^{k}=\beta^{k}$ for some nonzero integer $k$. Then $\alpha$ and $\beta$ are conjugate in $B_{n}$. ###### Theorem 2.3 (E.-K. Lee and S.-J. Lee [LL07]). Let $G$ denote one of the Artin groups of finite type $\mathbf{A}_{n}$, $\mathbf{B}_{n}=\mathbf{C}_{n}$ and affine type $\tilde{\mathbf{A}}_{n-1}$, $\tilde{\mathbf{C}}_{n-1}$. If $\alpha,\beta\in G$ are such that $\alpha^{k}=\beta^{k}$ for some nonzero integer $k$, then $\alpha$ and $\beta$ are conjugate in $G$. ###### Corollary 2.4. Let $G$ denote one of the Artin groups of finite type $\mathbf{A}_{n}$, $\mathbf{B}_{n}=\mathbf{C}_{n}$ and affine type $\tilde{\mathbf{A}}_{n-1}$, $\tilde{\mathbf{C}}_{n-1}$. Let $\alpha$ and $\beta$ be elements of $G$, and let $k$ be a nonzero integer. Then, $\alpha$ and $\beta$ are conjugate in $G$ if and only if so are $\alpha^{k}$ and $\beta^{k}$. ### 2.3. Canonical reduction system of reducible braids Abusing notation, we use the same symbol for a curve and its isotopy class. Hence, for curves $C_{1}$ and $C_{2}$, $C_{1}=C_{2}$ means that $C_{1}$ and $C_{2}$ are isotopic, unless stated otherwise explicitly. A curve system $\mathcal{C}$ in $D_{n}$ means a finite collection of disjoint simple closed curves in $D_{n}$. It is said to be _essential_ if each component is homotopic neither to a point nor to a puncture nor to the boundary. For an $n$-braid $\alpha$ and a curve system $\mathcal{C}$ in $D_{n}$, let $\alpha*\mathcal{C}$ denote the left action of $\alpha$ on $\mathcal{C}$. Recall that an $n$-braid $\alpha$ is reducible if $\alpha*\mathcal{C}=\mathcal{C}$ for some essential curve system $\mathcal{C}$ in $D_{n}$, called a _reduction system_ of $\alpha$. For a reduction system $\mathcal{C}$ of an $n$-braid $\alpha$, let $D_{\mathcal{C}}$ be the closure of $D_{n}\setminus N(\mathcal{C})$ in $D_{n}$, where $N(\mathcal{C})$ is a regular neighborhood of $\mathcal{C}$. The restriction of $\alpha$ induces a self-diffeomorphism on $D_{\mathcal{C}}$ that is well defined up to isotopy. Due to J. Birman, A. Lubotzky and J. McCarthy [BLM83] and N. V. Ivanov [Iva92], for any $n$-braid $\alpha$, there is a unique _canonical reduction system_ $\operatorname{\mathcal{R}}(\alpha)$ with the following properties. 1. (i) $\operatorname{\mathcal{R}}(\alpha^{m})=\operatorname{\mathcal{R}}(\alpha)$ for all $m\neq 0$. 2. (ii) $\operatorname{\mathcal{R}}(\beta\alpha\beta^{-1})=\beta*\operatorname{\mathcal{R}}(\alpha)$ for all $\beta\in B_{n}$. 3. (iii) The restriction of $\alpha$ to each component of $D_{\operatorname{\mathcal{R}}(\alpha)}$ is either periodic or pseudo-Anosov. A reduction system with this property is said to be _adequate_. 4. (iv) If $\mathcal{C}$ is an adequate reduction system of $\alpha$, then $\operatorname{\mathcal{R}}(\alpha)\subset\mathcal{C}$. Note that a braid $\alpha$ is non-periodic and reducible if and only if $\operatorname{\mathcal{R}}(\alpha)\neq\emptyset$. Let $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ denote the collection of outermost components of $\operatorname{\mathcal{R}}(\alpha)$. Then $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ satisfies the properties (i) and (ii). ### 2.4. Reducible braids with a standard reduction system Here we introduce some notions in [LL08] that will be used in computations involving reducible braids. ###### Definition 2.5. An essential curve system in $D_{n}$ is said to be _standard_ if each component is isotopic to a round circle centered at the real axis as in Figure 4 (a), and _unnested_ if none of its components encloses another component as in Figure 4 (b). Two curve systems $\mathcal{C}_{1}$ and $\mathcal{C}_{2}$ in $D_{n}$ are said to be _of the same type_ if there is a diffeomorphism $f:D_{n}\to D_{n}$ such that $f(\mathcal{C}_{1})$ is isotopic to $\mathcal{C}_{2}$. | | ---|---|--- (a) a standard curve system | | (b) the unnested standard curve system $\mathcal{C}_{(1,1,2,1,2,3)}$ Figure 4. Some standard curve systems in $D_{10}$ $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ is unnested for any non-periodic reducible braid $\alpha$. If $\alpha$ and $\beta$ are conjugate, $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$ are of the same type. Recall that an ordered $k$-tuple $\mathbf{n}=(n_{1},\ldots,n_{k})$ is a $k$-composition of $n$ if $n=n_{1}+\cdots+n_{k}$ and $n_{i}\geq 1$ for each $i$. Unnested standard curve systems in $D_{n}$ are in one-to-one correspondence with $k$-compositions of $n$ for $2\leq k\leq n-1$. The $k$-braid group $B_{k}$ acts on the set of $k$-compositions as $\alpha*(n_{1},\ldots,n_{k})=(n_{\theta^{-1}(1)},\ldots,n_{\theta^{-1}(k)}),$ where $\alpha\in B_{k}$ and $\theta=\pi_{\alpha}$. ###### Definition 2.6. For a composition $\mathbf{n}=(n_{1},\ldots,n_{k})$ of $n$, let $\mathcal{C}_{\mathbf{n}}$ denote the curve system $\cup_{n_{i}\geq 2}C_{i}$ in $D_{n}$, where $C_{i}$ with $n_{i}\geq 2$ is a round circle enclosing the punctures $\\{m:\sum_{j=1}^{i-1}n_{j}<m\leq\sum_{j=1}^{i}n_{j}\\}$. For example, Figure 4 (b) shows $\mathcal{C}_{\mathbf{n}}$ for $\mathbf{n}=(1,1,2,1,2,3)$. For compositions $\mathbf{n}_{1}$ and $\mathbf{n}_{2}$ of $n$, $\mathcal{C}_{\mathbf{n}_{1}}$ and $\mathcal{C}_{\mathbf{n}_{2}}$ are of the same type if and only if $\mathbf{n}_{1}$ and $\mathbf{n}_{2}$ induce the same partition of $n$ such as $\mathbf{n}_{1}=(2,2,1)$ and $\mathbf{n}_{2}=(2,1,2)$. ###### Definition 2.7. Let $\mathbf{n}=(n_{1},\cdots,n_{k})$ be a composition of $n$. * (i) Let $\hat{\alpha}=l_{1}\cup\cdots\cup l_{k}$ be a $k$-braid, where the right endpoint of $l_{i}$ is at the $i$-th position from bottom. We define $\langle\hat{\alpha}\rangle_{\mathbf{n}}$ as the $n$-braid obtained from $\hat{\alpha}$ by taking $n_{i}$ parallel copies of $l_{i}$ for each $i$. See Figure 5 (a). * (ii) Let $\alpha_{i}\in B_{n_{i}}$ for $i=1,\ldots,k$. We define $(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}$ as the $n$-braid $\alpha_{1}^{\prime}\alpha_{2}^{\prime}\cdots\alpha_{k}^{\prime}$, where each $\alpha_{i}^{\prime}$ is the image of $\alpha_{i}$ under the homomorphism $B_{n_{i}}\to B_{n}$ defined by $\sigma_{j}\mapsto\sigma_{(n_{1}+\cdots+n_{i-1})+j}$. See Figure 5 (b). We will use the notation $\alpha=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}$ frequently. See Figure 5 (c). Let $\bigoplus_{i=1}^{m}\alpha_{i}$ denote $\alpha_{1}\oplus\cdots\oplus\alpha_{m}$, and let $m\alpha$ denote $\bigoplus_{i=1}^{m}\alpha$. For example, $\begin{array}[]{rcl}(\alpha_{0}\oplus{\textstyle\bigoplus_{i=1}^{d}}(\alpha_{i,1}\oplus\cdots\oplus\alpha_{i,m}))_{\mathbf{n}}&=&(\alpha_{0}\oplus(\alpha_{1,1}\oplus\cdots\oplus\alpha_{1,m})\oplus\cdots\oplus(\alpha_{d,1}\oplus\cdots\oplus\alpha_{d,m}))_{\mathbf{n}},\\\ (\alpha_{0}\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}}&=&(\alpha_{0}\oplus(\underbrace{\alpha_{1}\oplus\cdots\oplus\alpha_{1}}_{m})\oplus\cdots\oplus(\underbrace{\alpha_{r}\oplus\cdots\oplus\alpha_{r}}_{m}))_{\mathbf{n}}.\end{array}$ | | | | ---|---|---|---|--- (a) $\langle\sigma_{1}^{-1}\sigma_{2}\rangle_{\mathbf{n}}$ | | (b) $(\sigma_{1}^{3}\oplus\sigma_{1}^{-2}\sigma_{2}^{3}\oplus 1)_{\mathbf{n}}$ | | (c) $\langle\sigma_{1}^{-1}\sigma_{2}\rangle_{\mathbf{n}}(\sigma_{1}^{3}\oplus\sigma_{1}^{-2}\sigma_{2}^{3}\oplus 1)_{\mathbf{n}}$ Figure 5. $\mathbf{n}=(2,3,1)$ ###### Lemma 2.8 ([LL08, Lemmas 3.5 and 3.6]). Let $\mathbf{n}=(n_{1},\ldots,n_{k})$ be a composition of $n$. 1. (i) The expression $\alpha=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}$ is unique, that is, if $\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}=\langle\hat{\beta}\rangle_{\mathbf{n}}(\beta_{1}\oplus\cdots\oplus\beta_{k})_{\mathbf{n}}$, then $\hat{\alpha}=\hat{\beta}$ and $\alpha_{i}=\beta_{i}$ for $1\leq i\leq k$. 2. (ii) For $\alpha\in B_{n}$, $\alpha*\mathcal{C}_{\mathbf{n}}$ is standard if and only if $\alpha$ can be expressed as $\alpha=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}$. In this case, $\alpha*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\hat{\alpha}*\mathbf{n}}$. 3. (iii) $\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}=(\alpha_{\theta^{-1}(1)}\oplus\cdots\oplus\alpha_{\theta^{-1}(k)})_{\hat{\alpha}\ast\mathbf{n}}\langle\hat{\alpha}\rangle_{\mathbf{n}}$, where $\theta=\pi_{\hat{\alpha}}$. 4. (iv) $\langle\hat{\alpha}\hat{\beta}\rangle_{\mathbf{n}}=\langle\hat{\alpha}\rangle_{\hat{\beta}*\mathbf{n}}\langle\hat{\beta}\rangle_{\mathbf{n}}$. 5. (v) $(\langle\hat{\alpha}\rangle_{\mathbf{n}})^{-1}=\langle\hat{\alpha}^{-1}\rangle_{\hat{\alpha}*\mathbf{n}}$. 6. (vi) $(\alpha_{1}\beta_{1}\oplus\cdots\oplus\alpha_{k}\beta_{k})_{\mathbf{n}}=(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}(\beta_{1}\oplus\cdots\oplus\beta_{k})_{\mathbf{n}}$ 7. (vii) $(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}^{-1}=(\alpha_{1}^{-1}\oplus\cdots\oplus\alpha_{k}^{-1})_{\mathbf{n}}$. 8. (viii) $\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}=\Delta_{(n)}$ if and only if $\hat{\alpha}=\Delta_{(k)}$ and $\alpha_{i}=\Delta_{(n_{i})}$ for $1\leq i\leq k$. Lemma 2.8 (iii)–(vii) are similar to the computations in the wreath product $G\wr B_{k}$, where $k$-braids act on the $k$-cartesian product of $G$ by permuting coordinates. Let $n=mk$ with $m,k\geq 2$, and let $\mathbf{n}=(m,\ldots,m)$ be a composition of $n$. Let $B_{n}(\mathcal{C}_{\mathbf{n}})=\\{\alpha\in B_{n}\mid\alpha*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}\\}$. Then Lemma 2.8 shows that $B_{n}(\mathcal{C}_{\mathbf{n}})$ is indeed isomorphic to $B_{m}\wr B_{k}$. ###### Definition 2.9. Let $\mathbf{n}=(n_{1},\ldots,n_{k})$ be a composition of $n$, and let $\alpha\in B_{n}$ be such that $\alpha*\mathcal{C}_{\mathbf{n}}$ is standard. Then $\alpha$ is uniquely expressed as $\alpha=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{1}\oplus\cdots\oplus\alpha_{k})_{\mathbf{n}}$ by Lemma 2.8. We call $\hat{\alpha}$ the _$\mathbf{n}$ -exterior braid_ of $\alpha$, and denote it by $\operatorname{Ext}_{\mathbf{n}}(\alpha)$. The braids $\alpha_{1},\ldots,\alpha_{k}$ are called _$\mathbf{n}$ -interior braids_ of $\alpha$. If $\operatorname{Ext}_{\mathbf{n}}(\alpha)$ is the identity, $\alpha$ is called an _$\mathbf{n}$ -split braid_. The lemma below follows from the above lemma and properties of $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\cdot)$. ###### Lemma 2.10. Let $\alpha$ and $\beta$ be $n$-braids such that $\alpha*\mathcal{C}_{\mathbf{n}}=\beta*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$ for a composition $\mathbf{n}$ of $n$. * (i) $\operatorname{Ext}_{\mathbf{n}}(\alpha\beta)=\operatorname{Ext}_{\mathbf{n}}(\alpha)\operatorname{Ext}_{\mathbf{n}}(\beta)$ and $\operatorname{Ext}_{\mathbf{n}}(\alpha^{-1})=\operatorname{Ext}_{\mathbf{n}}(\alpha)^{-1}$. * (ii) If $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)=\mathcal{C}_{\mathbf{n}}$ and $\gamma$ is an $n$-braid with $\beta=\gamma\alpha\gamma^{-1}$, then $\gamma*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$ and $\operatorname{Ext}_{\mathbf{n}}(\beta)=\operatorname{Ext}_{\mathbf{n}}(\gamma)\operatorname{Ext}_{\mathbf{n}}(\alpha)\operatorname{Ext}_{\mathbf{n}}(\gamma)^{-1}$. ## 3\. Proof of Theorem 1.6 In this section, we prove Theorem 1.6 by presenting three examples which show that none of the monomorphisms $\psi_{1}:A(\mathbf{B}_{n})\to A(\mathbf{A}_{n})$, $\psi_{2}:A(\tilde{\mathbf{A}}_{n-1})\to A(\mathbf{B}_{n})$ and $\psi_{3}:A(\tilde{\mathbf{C}}_{n-1})\to A(\mathbf{B}_{n})$ is injective on the set of conjugacy classes for $n\geq 3$. Recall that $\begin{array}[]{l}A(\mathbf{A}_{n})=B_{n+1};\\\ A(\mathbf{B}_{n})=B_{n+1,1};\end{array}\quad\begin{array}[]{l}A(\tilde{\mathbf{A}}_{n-1})=\\{\alpha\in B_{n+1}\mid\mbox{$\alpha$ is 1-unlinked}\\};\\\ A(\tilde{\mathbf{C}}_{n-1})=\\{\alpha\in B_{n+1}\mid\mbox{$\alpha$ is $\\{1,n+1\\}$-pure}\\}.\end{array}$ ###### Proof of Theorem 1.6. Consider the following $(n+1)$-braids for $n\geq 3$. See Figure 6. $\biggl{\\{}\begin{array}[]{l}\alpha_{1}=\sigma_{1}^{2}\sigma_{2}^{4},\\\ \beta_{1}=\sigma_{2}^{2}\sigma_{1}^{4},\end{array}\qquad\biggl{\\{}\begin{array}[]{l}\alpha_{2}=\sigma_{n}^{2}\sigma_{n-1}^{4},\\\ \beta_{2}=\sigma_{n-1}^{2}\sigma_{n}^{4},\end{array}\qquad\mbox{and}\qquad\biggl{\\{}\begin{array}[]{l}\alpha_{3}=\sigma_{n}\cdots\sigma_{2},\\\ \beta_{3}=\sigma_{1}^{-2}\alpha_{3}\sigma_{1}^{2}.\end{array}$ | | | | | | ---|---|---|---|---|---|--- (a) $\alpha_{1}=\sigma_{1}^{2}\sigma_{2}^{4}$ | | (b) $\alpha_{2}=\sigma_{n}^{2}\sigma_{n-1}^{4}$ | | (c) $\alpha_{3}=\sigma_{n}\cdots\sigma_{2}$ | | (d) $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha_{3})$ | | | | | | (e) $\beta_{1}=\sigma_{2}^{2}\sigma_{1}^{4}$ | | (f) $\beta_{2}=\sigma_{n-1}^{2}\sigma_{n}^{4}$ | | (g) $\beta_{3}=\sigma_{1}^{-2}\alpha_{3}\sigma_{1}^{2}$ | | (h) $\chi=(\sigma_{1}\cdots\sigma_{n})(\sigma_{n}\cdots\sigma_{1})$ Figure 6. The examples for Theorem 1.6 when $n=4$ First, let us consider $\alpha_{1}$ and $\beta_{1}$. Because $\alpha_{1}$ and $\beta_{1}$ are 1-pure, they belong to $B_{n+1,1}$. They are conjugate in $B_{n+1}$ because $\beta_{1}=(\sigma_{1}\sigma_{2}\sigma_{1})\alpha_{1}(\sigma_{1}\sigma_{2}\sigma_{1})^{-1}$. However they are not conjugate in $B_{n+1,1}$ because $\operatorname{lk}(\alpha_{1})=1$ and $\operatorname{lk}(\beta_{1})=2$. This shows that $\psi_{1}:A(\mathbf{B}_{n})\to A(\mathbf{A}_{n})$ is not injective on the set of conjugacy classes. The second pair $(\alpha_{2},\beta_{2})$ is similar to the first one. Notice that $n\geq 3$. Both $\alpha_{2}$ and $\beta_{2}$ are $\\{1,n+1\\}$-pure, hence they belong to $A(\tilde{\mathbf{C}}_{n-1})$. They are conjugate in $B_{n+1,1}$ because $\sigma_{n}\sigma_{n-1}\sigma_{n}$ is 1-pure and $\beta_{2}=(\sigma_{n}\sigma_{n-1}\sigma_{n})\alpha_{2}(\sigma_{n}\sigma_{n-1}\sigma_{n})^{-1}$. However $\alpha_{2}$ cannot be conjugated to $\beta_{2}$ by an $\\{1,n+1\\}$-pure braid. An easy way to see this is to consider the linking number, say $\operatorname{lk}^{\prime}$, of $(n+1)$-st strand with the other strands, which is a well-defined homomorphism from $A(\tilde{\mathbf{C}}_{n-1})$ to $\mathbf{Z}$. Because $\operatorname{lk}^{\prime}(\alpha_{2})=1$ and $\operatorname{lk}^{\prime}(\beta_{2})=2$, the braids $\alpha_{2}$ and $\beta_{2}$ are not conjugate in $A(\tilde{\mathbf{C}}_{n-1})$. This shows that $\psi_{3}:A(\tilde{\mathbf{C}}_{n-1})\to A(\mathbf{B}_{n})$ is not injective on the set of conjugacy classes. Now, let us consider the last pair $(\alpha_{3},\beta_{3})$. Because both $\alpha_{3}$ and $\beta_{3}$ are 1-unlinked, they belong to $A(\tilde{\mathbf{A}}_{n-1})$. They are conjugate in $B_{n+1,1}$ because $\sigma_{1}^{2}$ is 1-pure and $\beta_{3}=\sigma_{1}^{-2}\alpha_{3}\sigma_{1}^{2}$. The claim below shows the structure of $Z_{B_{n+1}}(\alpha_{3})$. It is a direct consequence of the main result of J. González-Meneses and B. Wiest in [GW04], however we provide a proof for the reader’s convenience. _Claim._ $Z_{B_{n+1}}(\alpha_{3})$ is generated by $\alpha_{3}$ and $\chi=(\sigma_{1}\cdots\sigma_{n})(\sigma_{n}\cdots\sigma_{1})$. In particular, every element $\zeta$ in $Z_{B_{n+1}}(\alpha_{3})$ is 1-pure with $\operatorname{lk}(\zeta)\equiv 0\pmod{n}$. ###### Proof of Claim. Let $\mathbf{n}=(1,n)$ be a composition of $n+1$, and let $Z(\alpha_{3})=Z_{B_{n+1}}(\alpha_{3})$. Note that $\alpha_{3}=(1\oplus\delta_{(n)})_{\mathbf{n}}$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha_{3})=\mathcal{C}_{\mathbf{n}}$. See Figure 6 (d). If $\gamma\in Z(\alpha_{3})$, then $\gamma*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$, hence $\gamma$ is of the form $\gamma=\langle\hat{\gamma}\rangle_{\mathbf{n}}(1\oplus\gamma_{1})_{\mathbf{n}}$ for some $\hat{\gamma}\in B_{2}$ and $\gamma_{1}\in B_{n}$. By direct computation, we can see that $\gamma\in Z(\alpha_{3})$ if and only if $\hat{\gamma}=\sigma_{1}^{2k}$ for some $k\in\mathbf{Z}$ and $\gamma_{1}$ commutes with $\delta_{(n)}$. Because $Z_{B_{n}}(\delta_{(n)})$ is infinite cyclic generated by $\delta_{(n)}$ by [BDM02], $Z(\alpha_{3})$ is generated by $\langle\sigma_{1}^{2}\rangle_{\mathbf{n}}=\chi$ and $(1\oplus\delta_{(n)})_{\mathbf{n}}=\alpha_{3}$. Because $\operatorname{lk}(\chi)=n$ and $\operatorname{lk}(\alpha_{3})=0$, every element in $Z(\alpha_{3})$ has linking number a multiple of $n$. ∎ Assume that $\alpha_{3}$ and $\beta_{3}$ are conjugate in $A(\tilde{\mathbf{A}}_{n-1})$. Then there exists a 1-unlinked $(n+1)$-braid $\gamma$ such that $\alpha_{3}=\gamma^{-1}\beta_{3}\gamma=\gamma^{-1}\sigma_{1}^{-2}\alpha_{3}\sigma_{1}^{2}\gamma$. Therefore $\sigma_{1}^{2}\gamma\in Z_{B_{n+1}}(\alpha_{3})$. Because $\operatorname{lk}(\sigma_{1}^{2}\gamma)=\operatorname{lk}(\sigma_{1}^{2})+\operatorname{lk}(\gamma)=1+0=1$, $\operatorname{lk}(\sigma_{1}^{2}\gamma)$ is not a multiple of $n\geq 3$. It is a contradiction. This shows that $\psi_{2}:A(\tilde{\mathbf{A}}_{n-1})\to A(\mathbf{B}_{n})$ is not injective on the set of conjugacy classes. ∎ ## 4\. Braids in the centralizer of a periodic braid In this section, we review isomorphisms between Artin groups of type $\mathbf{B}$ and centralizers of periodic braids, and make some tools useful in proving Theorem 1.4. ### 4.1. Discussions One of the difficulties in proving Theorem 1.4 lies in dealing with reducible braids. Let $(\alpha,\beta,\omega)$ be a triple of $n$-braids such that $\omega$ is periodic and both $\alpha$ and $\beta$ are non-periodic reducible braids in $Z(\omega)$ which are conjugate in $B_{n}$. If $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$ are standard, it would be very convenient for computations, but they are not necessarily standard. Our strategy is to construct another such triple $(\alpha^{\prime},\beta^{\prime},\omega^{\prime})$ such that both $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha^{\prime})$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta^{\prime})$ are standard. This new triple is related to the original one by conjugations so that if $\alpha^{\prime}$ and $\beta^{\prime}$ are conjugate in $Z(\omega^{\prime})$ then $\alpha$ and $\beta$ are conjugate in $Z(\omega)$. This construction is the main theme of this section. The following lemma is obvious, hence we omit the proof. ###### Lemma 4.1. Let $\omega$ be a periodic $n$-braid, and let $\alpha$ and $\beta$ belong to $Z(\omega)$. * (i) Let $\chi\in B_{n}$. Then $\chi\alpha\chi^{-1}$ and $\chi\beta\chi^{-1}$ belong to $Z(\chi\omega\chi^{-1})$. Furthermore, $\chi\alpha\chi^{-1}$ and $\chi\beta\chi^{-1}$ are conjugate in $Z(\chi\omega\chi^{-1})$ if and only if $\alpha$ and $\beta$ are conjugate in $Z(\omega)$. * (ii) Let $\chi_{1},\chi_{2}\in Z(\omega)$. Then $\chi_{1}\alpha\chi_{1}^{-1}$ and $\chi_{2}\beta\chi_{2}^{-1}$ belong to $Z(\omega)$. Furthermore, $\chi_{1}\alpha\chi_{1}^{-1}$ and $\chi_{2}\beta\chi_{2}^{-1}$ are conjugate in $Z(\omega)$ if and only if $\alpha$ and $\beta$ are conjugate in $Z(\omega)$. From the above lemma, one can think of two ways in order to make $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$ standard. First, one can expect that there exists $\chi\in B_{n}$ such that both $\chi*\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\chi*\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$ are standard. If it is true, $(\chi\alpha\chi^{-1},\chi\beta\chi^{-1},\chi\omega\chi^{-1})$ can be taken as a desired triple by Lemma 4.1 (i). Second, one can expect that there exist $\chi_{1},\chi_{2}\in Z(\omega)$ such that $\chi_{1}*\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\chi_{2}*\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$ are standard. If it is true, $(\chi_{1}\alpha\chi_{1}^{-1},\chi_{2}\beta\chi_{2}^{-1},\omega)$ can be taken as a desired triple by Lemma 4.1 (ii). However the following example shows that each of these ideas does not work for itself. Our construction will be as $(\chi\chi_{1}\alpha\chi_{1}^{-1}\chi^{-1},\chi\chi_{2}\beta\chi_{2}^{-1}\chi^{-1},\chi\omega\chi^{-1})$, where $\chi_{1},\chi_{2}\in Z(\omega)$ and $\chi\in B_{n}$ such that both $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\chi\chi_{1}\alpha\chi_{1}^{-1}\chi^{-1})$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\chi\chi_{2}\beta\chi_{2}^{-1}\chi^{-1})$ are standard and $\chi\omega\chi^{-1}$ is a special kind of periodic braid (see Corollary 4.14). ###### Example 4.2. Consider the 4-braids $\omega=\delta^{2}=(\sigma_{3}\sigma_{2}\sigma_{1})^{2},\qquad\alpha=\sigma_{2}\sigma_{1}\sigma_{2}^{-1}\quad\mbox{and}\quad\beta=\sigma_{3}\sigma_{2}\sigma_{3}^{-1}.$ The element $\omega$ is a periodic 4-braid which corresponds to the $\pi$-rotation of the punctured disk when punctures are distributed as in Figure 3 (a). $\alpha$ and $\beta$ are reducible braids in $Z(\omega)$ which are conjugate in $B_{4}$. Let $C=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $C^{\prime}=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$. They enclose two punctures as in Figure 7 (a) and (b). | | | | ---|---|---|---|--- | | | | (a) $C$ | (b) $C^{\prime}$ | (c) | (d) | (e) Figure 7. The figures in two rows show the same curves with different location of punctures. There are three standard curves in $D_{4}$ enclosing two punctures as in Figure 7 (c)–(e) and the geometric intersection numbers between two of them are either 0 or 2. Because the geometric intersection number between $C$ and $C^{\prime}$ is 4, there is no 4-braid $\chi$ such that $\chi*C$ and $\chi*C^{\prime}$ are standard. Assume that there exists $\chi_{1}\in Z(\omega)$ such that $\chi_{1}*C$ is standard. Because $C$ encloses two punctures, so does $\chi_{1}*C$. Therefore $\chi_{1}*C$ is one of the three curves in Figure 7 (c)–(e). Moreover, $\chi_{1}*C$ is invariant under the action of $\omega$ because $\omega*(\chi_{1}*C)=\chi_{1}*(\omega*C)=\chi_{1}*C$. Because none of the curves in Figure 7 (c)–(e) is invariant under the action of $\omega$, it is a contradiction. Similarly, there is no $\chi_{2}\in Z(\omega)$ such that $\chi_{2}*C^{\prime}$ is standard. The monomorphisms $\psi_{5}$ and $\psi_{6}$ are compositions of an isomorphism $\varphi$ and the inclusion $\iota$: $\textstyle{B_{d+1,1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\varphi}$$\scriptstyle{\simeq}$$\textstyle{Z(\omega)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\iota}$$\textstyle{B_{n},}$ where $n=md$ or $md+1$ with $m\geq 2$; $\omega$ is a periodic $n$-braid conjugate to $\delta^{k}$ with $d=\gcd(k,n)$ or to $\epsilon^{k}$ with $d=\gcd(k,n-1)$. Observe the following. * (i) $\iota$ is injective on the set of conjugacy classes if and only if so is $\iota\circ\varphi$. * (ii) Let $\omega^{\prime}$ be a periodic $n$-braid which is conjugate to $\omega$, and let $\iota^{\prime}:Z(\omega^{\prime})\to B_{n}$ be the inclusion. Then $\iota$ and $\iota^{\prime}$ are related by an inner automorphism, say $I_{\chi}$, of $B_{n}$ and its restriction to $Z(\omega)$ as follows. $\textstyle{Z(\omega)\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\simeq}$$\scriptstyle{I_{\chi}}$$\scriptstyle{\iota}$$\textstyle{B_{n}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\simeq}$$\scriptstyle{I_{\chi}}$$\textstyle{Z(\omega^{\prime})\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\iota^{\prime}}$$\textstyle{B_{n}}$ Therefore, $\iota$ is injective on the set of conjugacy classes if and only if so is $\iota^{\prime}$. * (iii) Consequently, the injectivity of $\iota\circ\varphi$ on the set of conjugacy classes depends only on the conjugacy class of $\omega$. Therefore Theorem 1.4 states that $\iota:Z(\omega)\to B_{n}$ is injective on the set of conjugacy classes for any periodic $n$-braid $\omega$. And Corollary 1.5 is equivalent to $\iota:Z(\omega)\to B_{n}$ being injective on the set of conjugacy classes for any noncentral periodic $n$-braid $\omega$. Because Theorem 1.4 obviously holds for central braids $\omega$, we have the following. ###### Lemma 4.3. Theorem 1.4 and Corollary 1.5 are equivalent. We remark that there are infinitely many isomorphisms from $B_{d+1,1}$ to $Z(\omega)$, and any two of them are related by an automorphism of $B_{d+1,1}=A(\mathbf{B}_{d})$. See [CC05] for the classification of automorphisms of Artin groups of type $\mathbf{A}$, $\mathbf{B}$, $\tilde{\mathbf{A}}$ and $\tilde{\mathbf{C}}$. ### 4.2. Periodic braids and their centralizers Let $m$ and $d$ be integers with $m\geq 2$ and $d\geq 1$, and let $n=md+1$. Here we construct a periodic $n$-braid $\mu_{m,d}$ conjugate to $\epsilon^{d}$, and an isomorphism $B_{d+1,1}\simeq Z(\mu_{m,d})$ following the work in [BDM02]. Let $Z_{n}^{(m)}=\\{z_{0}\\}\cup\\{z_{i,j}\mid 1\leq i\leq d,~{}1\leq j\leq m\\}$, where $z_{0}$ is the origin and $z_{i,j}$ is the complex number with $|z_{i,j}|=i$ and $\arg(z_{i,j})=2\pi(j-1)/m$. Let $D_{n}^{(m)}=\\{z\in\mathbb{C}\mid|z|\leq n+1\\}\setminus Z_{n}^{(m)}.$ $\scriptstyle{\eta_{m,d}\qquad\mbox{}}$$\scriptstyle{\phi_{m,d}\mbox{}}$ $D_{d+1}$ $D_{n}^{(m)}$ and $\rho_{m}$ $D_{n}$ and $\eta_{m,d}\circ\rho_{m}\circ\eta_{m,d}^{-1}$ Figure 8. The punctured spaces $D_{n}^{(m)}$ and $D_{d+1}$, where $n=13$, $m=3$ and $d=4$ Now we construct a covering map $\phi_{m,d}:D_{n}^{(m)}\to D_{d+1}$ and a diffeomorphism $\eta_{m,d}:D_{n}^{(m)}\to D_{n}$ as in Figure 8. Let $\rho_{m}$ be the clockwise $(2\pi/m)$-rotation of $D_{n}^{(m)}$. The identification space $D_{n}^{(m)}/\langle\rho_{m}\rangle$ is diffeomorphic to the $(d+1)$-punctured disk $D_{d+1}$. We write the natural projection by $\phi_{m,d}:D_{n}^{(m)}\to D_{d+1}.$ $\phi_{m,d}$ is a covering map whose group of covering transformations is the cyclic group generated by $\rho_{m}$. We construct a diffeomorphism $\eta_{m,d}:D_{n}^{(m)}\to D_{n},$ which fixes $\partial D_{n}^{(m)}=\partial D_{n}$ pointwise, as follows. First, cut $D_{n}^{(m)}$ and $D_{n}$ by $d$ round circles into one once- punctured disk and $d$ annuli with $m$ punctures as in Figure 9 (a). Then, define $\eta_{m,d}$ on each punctured annulus as in Figure 9 (b) such that it coincides at the intersections of two adjacent annuli, and extend $\eta_{m,d}$ to the once-punctured disk in an obvious way. Such a diffeomorphism $\eta_{m,d}$ is unique up to isotopy. $\scriptstyle{\eta_{m,d}}$ | $\scriptstyle{\eta_{m,d}}$ ---|--- (a) | (b) Figure 9. The diffeomorphism $\eta_{m,d}:D_{n}^{(m)}\to D_{n}$ ###### Definition 4.4. Let $m$ and $d$ be integers with $m\geq 2$ and $d\geq 0$. We define an $(md+1)$-braid $\mu_{m,d}$ recursively as follows: $\mu_{m,0}$ is the unique braid with one strand; for $d\geq 1$, $\mu_{m,d}=\mu_{m,d-1}(\sigma_{dm}\sigma_{dm-1}\cdots\sigma_{2}\sigma_{1})(\sigma_{1}\sigma_{2}\cdots\sigma_{(d-1)m+1}),$ where $\mu_{m,d-1}$ is regarded as an $(md+1)$-braid by using the inclusion $B_{m(d-1)+1}\to B_{md+1}$ defined by $\sigma_{i}\mapsto\sigma_{i}$ for $1\leq i\leq m(d-1)$. See Figure 10 for $\mu_{m,d}$ with $m=3$ and $1\leq d\leq 4$. | | | ---|---|---|--- $\mu_{3,1}$ | $\mu_{3,2}$ | $\mu_{3,3}$ | $\mu_{3,4}$ Figure 10. The periodic braid $\mu_{3,d}$ for $d=1,2,3,4$ From the construction, we can see the following. * (i) $\eta_{m,d}\circ\rho_{m}\circ\eta_{m,d}^{-1}:D_{n}\to D_{n}$ represents the braid $\mu_{m,d}$ modulo $\Delta^{2}$. * (ii) An $n$-braid belongs to $Z(\mu_{m,d})$ if and only if it is represented by $\eta_{m,d}\circ h\circ\eta_{m,d}^{-1}$ for a $\rho_{m}$-equivariant diffeomorphism $h:D_{n}^{(m)}\to D_{n}^{(m)}$ that fixes the boundary pointwise. Because $(\mu_{m,d})^{m}=\Delta_{(n)}^{2}=(\epsilon_{(n)}^{d})^{m}$, $\mu_{m,d}$ is conjugate to $\epsilon_{(n)}^{d}$, hence $\nu(\mu_{m,d})$ is conjugate to $\delta_{(n-1)}^{d}$. Recall that the monomorphisms $\psi_{4}$, $\psi_{5}$ and $\psi_{6}$ are induced by the isomorphisms $B_{d+1,1}\simeq Z(\epsilon_{(n)}^{d})$ and $B_{d+1,1}\simeq Z(\delta_{(n-1)}^{d})$ in [BDM02, GW04]. In the same way, we define monomorphisms $\psi_{4}^{\prime}$, $\psi_{5}^{\prime}$ and $\psi_{6}^{\prime}$ by using isomorphisms $B_{d+1,1}\simeq Z(\mu_{m,d})$ and $B_{d+1,1}\simeq Z(\nu(\mu_{m,d}))$. ###### Definition 4.5. Let $m\geq 2$ and $d\geq 1$, and let $n=md+1$. We define an isomorphism $\varphi:B_{d+1,1}\to Z(\mu_{m,d})$ as follows. Let $\alpha\in B_{d+1,1}$, that is, $\alpha$ is a 1-pure $(d+1)$-braid. Then $\alpha$ is represented by a diffeomorphism $f:D_{d+1}\to D_{d+1}$ that fixes the first puncture together with all the boundary points. Let $\tilde{f}:D_{n}^{(m)}\to D_{n}^{(m)}$ be the lift of $f$ (with respect to $\phi_{m,d}$) that fixes $\partial D_{n}^{(m)}$ pointwise. Then $\tilde{f}$ is $\rho_{m}$-equivariant, hence $\eta_{m,d}\circ\tilde{f}\circ\eta_{m,d}^{-1}$ represents a braid in $Z(\mu_{m,d})$. We define $\varphi(\alpha)$ as the braid represented by $\eta_{m,d}\circ\tilde{f}\circ\eta_{m,d}^{-1}$. --- $\textstyle{D_{n}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\eta_{m,d}^{-1}}$$\textstyle{D_{n}^{(m)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\phi_{m,d}}$$\scriptstyle{\tilde{f}}$$\textstyle{D_{n}^{(m)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\phi_{m,d}}$$\scriptstyle{\eta_{m,d}}$$\textstyle{D_{n}}$$\textstyle{D_{d+1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{f}$$\textstyle{D_{d+1}}$ The monomorphisms $\psi_{4}^{\prime}$ and $\psi_{6}^{\prime}$ are compositions of $\varphi$ with the inclusions $Z(\mu_{m,d})\subset B_{md+1,1}$ and $Z(\mu_{m,d})\subset B_{md+1}$, respectively. The monomorphism $\psi_{5}^{\prime}$ is the composition of $\varphi$ with the homomorphism $\nu:B_{md+1,1}\to B_{md}$. See Figure 11. | ---|--- $\textstyle{B_{md}}$$\textstyle{Z(\nu(\mu_{m,d}))\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{Z(\mu_{m,d})\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\mbox{}\quad\nu}$$\textstyle{B_{md+1,1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\textstyle{B_{md+1}}$$\textstyle{B_{d+1,1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\varphi}$$\scriptstyle{\simeq}$$\scriptstyle{\psi_{5}^{\prime}}$$\scriptstyle{\simeq}$$\scriptstyle{\psi_{4}^{\prime}}$$\scriptstyle{\psi_{6}^{\prime}}$ Figure 11. Monomorphisms $\psi_{4}^{\prime}$, $\psi_{5}^{\prime}$ and $\psi_{6}^{\prime}$ By the construction, $\psi_{4}^{\prime}(\Delta_{(d+1)}^{2})=\psi_{6}^{\prime}(\Delta_{(d+1)}^{2})=\mu_{m,d}$. The following lemma looks obvious, but we include a sketchy proof for completeness. ###### Lemma 4.6. Let $\psi$ denote either $\psi_{4}^{\prime}:B_{d+1,1}\to B_{md+1,1}$, $\psi_{5}^{\prime}:B_{d+1,1}\to B_{md}$ or $\psi_{6}^{\prime}:B_{d+1,1}\to B_{md+1}$. The monomorphism $\psi$ preserves the Nielsen-Thurston type, that is, for $\alpha\in B_{d+1,1}$ * (i) $\alpha$ is periodic if and only if $\psi(\alpha)$ is periodic; * (ii) $\alpha$ is pseudo-Anosov if and only if $\psi(\alpha)$ is pseudo-Anosov; * (iii) $\alpha$ is reducible and non-periodic if and only if $\psi(\alpha)$ is reducible and non-periodic. ###### Proof. Assume $\psi=\psi_{6}^{\prime}$. The case of $\psi_{4}^{\prime}$ is the same, and the case of $\psi_{5}^{\prime}$ can be proved similarly. Because the three classes {periodic braids}, {pseudo-Anosov braids} and {non-periodic reducible braids} are mutually disjoint, it suffices to prove (i) together with (ii)′ and (iii)′ where * (ii)′ If $\alpha$ is pseudo-Anosov, then $\psi(\alpha)$ is pseudo-Anosov; * (iii)′ If $\alpha$ is reducible and non-periodic, then $\psi(\alpha)$ is reducible and non-periodic. (i) $\alpha$ is periodic $\Leftrightarrow$ $\alpha^{k}=(\Delta^{2})^{\ell}$ for some $k\neq 0$ and $\ell\in\mathbf{Z}$ $\Leftrightarrow$ $\psi(\alpha)^{k}=\psi(\alpha^{k})=\psi(\Delta^{2\ell})=\mu_{m,d}^{\ell}$ for some $k\neq 0$ and $\ell\in\mathbf{Z}$ $\Leftrightarrow$ $\psi(\alpha)$ is periodic (ii)′ Suppose that $\alpha\in B_{d+1,1}$ is pseudo-Anosov. There is a pseudo- Anosov diffeomorphism, say $f$, defined on the interior of $D_{d+1}$ representing $\alpha$ modulo $\Delta_{(d+1)}^{2}$. Let $\tilde{f}$ be the lift of $f$ (with respect to $\phi_{m,d}$) defined on the interior of $D_{md+1}^{(m)}$. Then $\tilde{f}$ is also a pseudo-Anosov diffeomorphism because invariant measured foliations of $f$ lift to invariant measured foliations of $\tilde{f}$ with the same dilatation. Therefore $\eta_{m,d}\circ\tilde{f}\circ\eta_{m,d}^{-1}$ is a pseudo-Anosov diffeomorphism. Since $\eta_{m,d}\circ\tilde{f}\circ\eta_{m,d}^{-1}$ represents $\psi(\alpha)$ modulo $\Delta_{(md+1)}^{2}$, $\psi(\alpha)$ is pseudo-Anosov. (iii)′ Suppose that $\alpha\in B_{d+1,1}$ is reducible and non-periodic. Then $\psi(\alpha)$ is non-periodic by (i). Because $\alpha$ is reducible, there exists a diffeomorphism $f:D_{d+1}\to D_{d+1}$, which represents $\alpha$, together with an essential curve system $\mathcal{C}$ such that $f(\mathcal{C})=\mathcal{C}$. Let $\tilde{\mathcal{C}}=\phi_{m,d}^{-1}(\mathcal{C})$, and let $\tilde{f}$ be a lift of $f$ with respect to $\phi_{m,d}$. Then $\tilde{\mathcal{C}}$ is an invariant essential curve system of $\tilde{f}$, hence $\eta_{m,d}(\tilde{\mathcal{C}})$ is a reduction system of $\psi(\alpha)$. ∎ ###### Corollary 4.7. Let $m\geq 2$ and $d\geq 1$. Let $\omega$ be a 1-pure braid conjugate to $\epsilon_{(md+1)}^{d}$, and let $\alpha\in Z(\omega)$. Then $\alpha$ and $\nu(\alpha)$ have the same Nielsen-Thurston type, that is, * (i) $\alpha$ is periodic if and only if $\nu(\alpha)$ is periodic; * (ii) $\alpha$ is pseudo-Anosov if and only if $\nu(\alpha)$ is pseudo-Anosov; * (iii) $\alpha$ is reducible and non-periodic if and only if $\nu(\alpha)$ is reducible and non-periodic. ###### Proof. Without loss of generality, we may assume that $\omega=\mu_{m,d}$. Then there exists $\bar{\alpha}\in B_{d+1,1}$ such that $\alpha=\psi_{4}^{\prime}(\bar{\alpha})$ and $\nu(\alpha)=\psi_{5}^{\prime}(\bar{\alpha})$. By Lemma 4.6, $\alpha$ and $\bar{\alpha}$ (resp. $\nu(\alpha)$ and $\bar{\alpha}$) have the same Nielsen- Thurston type. Therefore $\alpha$ and $\nu(\alpha)$ have the same Nielsen- Thurston type. ∎ From Lemma 4.6 and its proof, we can see that the monomorphisms $\psi_{4}^{\prime}$, $\psi_{5}^{\prime}$ and $\psi_{6}^{\prime}$ send canonical reduction systems to canonical reduction systems as follows. ###### Corollary 4.8. Let $m\geq 2$ and $d\geq 1$, and let $\alpha\in B_{d+1,1}$ be reducible and non-periodic. If $\mathcal{C}$ is the canonical reduction system of $\alpha$, then $\eta_{m,d}(\phi_{m,d}^{-1}(\mathcal{C}))$ is the canonical reduction system of both $\psi_{4}^{\prime}(\alpha)$ and $\psi_{6}^{\prime}(\alpha)$, and the canonical reduction system of $\psi_{5}^{\prime}(\alpha)$ is obtained from $\eta_{m,d}(\phi_{m,d}^{-1}(\mathcal{C}))$ by forgetting the first puncture. In particular, $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\psi_{4}^{\prime}(\alpha))=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\psi_{6}^{\prime}(\alpha))=\eta_{m,d}(\phi_{m,d}^{-1}(\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)))$, and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\psi_{5}^{\prime}(\alpha))$ is obtained from $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\psi_{4}^{\prime}(\alpha))$ by forgetting the first puncture. ###### Proof. We prove the corollary only for the case of $\psi_{4}^{\prime}$. The other cases are similar. Recall that the canonical reduction system of a braid is the smallest adequate reduction system of it. Let $\mathcal{C}\subset D_{d+1}$ be the canonical reduction system of $\alpha\in B_{d+1,1}$. Following the construction in Lemma 4.6, let $\tilde{\mathcal{C}}=\phi_{m,d}^{-1}(\mathcal{C})$ and $\mathcal{D}=\eta_{m,d}(\tilde{\mathcal{C}})$. From the proof of Lemma 4.6, $\mathcal{D}$ is a reduction system of $\psi_{4}^{\prime}(\alpha)$. Because $\mathcal{C}$ is adequate, $\mathcal{D}$ is adequate by Lemma 4.6 (i) and (ii). Hence there is a subset $\mathcal{D}^{\prime}$ of $\mathcal{D}$ which is the canonical reduction system of $\psi_{4}^{\prime}(\alpha)$. Let $\tilde{\mathcal{C}}^{\prime}=\eta_{m,d}^{-1}(\mathcal{D}^{\prime})$ and $\mathcal{C}^{\prime}=\phi_{m,d}(\tilde{\mathcal{C}}^{\prime})$. Note that $\tilde{\mathcal{C}}^{\prime}\subset\tilde{\mathcal{C}}$ and $\mathcal{C}^{\prime}\subset\mathcal{C}$. --- $\textstyle{D_{n}^{(m)}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\phi_{m,d}}$$\scriptstyle{\eta_{m,d}}$$\textstyle{D_{n}}$$\textstyle{D_{d+1}}$ --- $\textstyle{\tilde{\mathcal{C}}^{\prime}\subset\tilde{\mathcal{C}}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\phi_{m,d}}$$\scriptstyle{\eta_{m,d}}$$\textstyle{\mathcal{D}^{\prime}\subset\mathcal{D}}$$\textstyle{\mathcal{C}^{\prime}\subset\mathcal{C}}$ We claim that $\tilde{\mathcal{C}}^{\prime}=\phi_{m,d}^{-1}(\mathcal{C}^{\prime})$ and that $\mathcal{C}^{\prime}$ is a reduction system of $\alpha$. Since $\mathcal{D}^{\prime}$ is the canonical reduction system of $\psi_{4}^{\prime}(\alpha)$ and since $\psi_{4}^{\prime}(\alpha)$ commutes with $\mu_{m,d}$, $\mu_{m,d}*\mathcal{D}^{\prime}=\mu_{m,d}*\operatorname{\mathcal{R}}(\psi_{4}^{\prime}(\alpha))=\operatorname{\mathcal{R}}(\mu_{m,d}\,\psi_{4}^{\prime}(\alpha)\,\mu_{m,d}^{-1})=\operatorname{\mathcal{R}}(\psi_{4}^{\prime}(\alpha))=\mathcal{D}^{\prime}.$ Because the diffeomorphism $\eta_{m,d}\circ\rho_{m}\circ\eta_{m,d}^{-1}$ represents $\mu_{m,d}\bmod\Delta^{2}$, the above formula implies that $\rho_{m}(\tilde{\mathcal{C}}^{\prime})=\tilde{\mathcal{C}}^{\prime}$, that is, if $\tilde{\mathcal{C}}^{\prime}$ contains a component of $\tilde{\mathcal{C}}$ then it contains all the components of $\tilde{\mathcal{C}}$ in the $\rho_{m}$-orbit of this component. Since the covering transformation group of $\phi_{m,d}$ is generated by $\rho_{m}$, $\tilde{\mathcal{C}}^{\prime}=\phi_{m,d}^{-1}(\mathcal{C}^{\prime})\quad\mbox{and hence}\quad\mathcal{D}^{\prime}=\eta_{m,d}(\phi_{m,d}^{-1}(\mathcal{C}^{\prime})).$ Let $f:D_{d+1}\to D_{d+1}$ be a diffeomorphism representing $\alpha$ such that $f(\mathcal{C})=\mathcal{C}$. Let $\tilde{f}$ be a lift of $f$ with respect to $\phi_{m,d}$. Then $\tilde{f}(\tilde{\mathcal{C}})=\tilde{\mathcal{C}}$ because $\tilde{\mathcal{C}}=\phi_{m,d}^{-1}(\mathcal{C})$. The diffeomorphism $\eta_{m,d}\circ\tilde{f}\circ\eta_{m,d}^{-1}$ represents the braid $\psi_{4}^{\prime}(\alpha)$. Since $\mathcal{D}^{\prime}=\eta_{m,d}(\tilde{\mathcal{C}}^{\prime})$ is a reduction system of $\psi_{4}^{\prime}(\alpha)$, one has $\tilde{f}(\tilde{\mathcal{C}}^{\prime})=\tilde{\mathcal{C}}^{\prime}$. Therefore $f(\mathcal{C}^{\prime})=\mathcal{C}^{\prime}$ because $\tilde{\mathcal{C}}^{\prime}=\phi_{m,d}^{-1}(\mathcal{C}^{\prime})$ and $\tilde{f}$ is a lift of $f$ with respect to $\phi_{m,d}$. Hence $\mathcal{C}^{\prime}$ is a reduction system of $\alpha$. Since $\mathcal{D}^{\prime}$ is adequate, the reduction system $\mathcal{C}^{\prime}$ is also adequate by Lemma 4.6 (i) and (ii). Because $\mathcal{C}$ is the canonical reduction system of $\alpha$, it is the smallest adequate reduction system of $\alpha$, hence $\mathcal{C}\subset\mathcal{C}^{\prime}$. Because $\mathcal{C}^{\prime}\subset\mathcal{C}$, this implies $\mathcal{C}=\mathcal{C}^{\prime}$ and hence $\mathcal{D}=\mathcal{D}^{\prime}$. So far, we have shown that if $\mathcal{C}$ is the canonical reduction system of $\alpha$ then $\eta_{m,d}(\phi_{m,d}^{-1}(\mathcal{C}))$ is the canonical reduction system of $\psi_{4}^{\prime}(\alpha)$. If $\mathcal{C}^{\prime\prime}$ is the collection of the outermost components of $\mathcal{C}$, then $\eta_{m,d}(\phi_{m,d}^{-1}(\mathcal{C}^{\prime\prime}))$ is the collection of the outermost components of $\eta_{m,d}(\phi_{m,d}^{-1}(\mathcal{C}))$, hence $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\psi_{4}^{\prime}(\alpha))=\eta_{m,d}(\phi_{m,d}^{-1}(\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)))$. ∎ We remark that the above argument actually shows that there is a one-to-one correspondence $\\{\mbox{curve systems $\mathcal{C}\subset D_{d+1}$}\\}\stackrel{{\scriptstyle 1-1}}{{\longleftrightarrow}}\\{\mbox{curve systems $\mathcal{D}\subset D_{md+1}$ such that $\mu_{m,d}*\mathcal{D}=\mathcal{D}$}\\}$ sending $\mathcal{C}\subset D_{d+1}$ to $\eta_{m,d}(\phi_{m,d}^{-1}(\mathcal{C}))$. This correspondence satisfies the property that $\mathcal{C}$ is a reduction system (resp. an adequate reduction system, the canonical reduction system) of $\alpha\in B_{d+1}$ if and only if $\eta_{m,d}(\phi_{m,d}^{-1}(\mathcal{C}))$ is a reduction system (resp. an adequate reduction system, the canonical reduction system) of $\psi_{4}^{\prime}(\alpha)$. For $1\leq i\leq d$ and $1\leq j\leq m$, let $x_{0}$ and $x_{i,j}$ be the integers such that $x_{0}=1$ and $x_{i,j}=(i-1)m+j+1$. In other words, $(1,2,\ldots,md+1)=(x_{0},x_{1,1},\ldots,x_{1,m},\ldots,x_{d,1},\ldots,x_{d,m})$. Note that $\eta_{m,d}(z_{0})=x_{0}$ and $\eta_{m,d}(z_{i,j})=x_{i,j}$ for $1\leq i\leq d$ and $1\leq j\leq m$. The induced permutation of $\mu_{m,d}$ is $\pi_{\mu_{m,d}}=(x_{0})(x_{1,m},\ldots,x_{1,2},x_{1,1})\cdots(x_{d,m},\ldots,x_{d,2},x_{d,1}).$ ###### Lemma 4.9. Let $m\geq 2$ and $d\geq 1$, and let $\theta$ be a permutation on $\\{1,\ldots,md+1\\}$. Then the following conditions are equivalent. * (i) $\theta$ is an induced permutation of an element of $Z(\mu_{m,d})$. * (ii) $\theta$ commutes with the induced permutation of $\mu_{m,d}$ * (iii) $\theta$ is such that $\theta(x_{0})=x_{0}$ and $\theta(x_{i,j})=x_{k_{i},j-\ell_{i}}$ for $1\leq i\leq d$ and $1\leq j\leq m$, where $0\leq\ell_{i}<m$ and $k_{1},\ldots,k_{d}$ is a rearrangement of $1,\ldots,d$. (Hereafter we regard the second index $j-\ell_{i}$ of $(k_{i},j-\ell_{i})$ as being taken modulo $m$.) ###### Proof. (i) $\Rightarrow$ (ii) $\Leftrightarrow$ (iii) is clear. Let us show that (ii) $\Rightarrow$ (i). Suppose that (ii) is true. The centralizer of $\pi_{\mu_{m,d}}$ is generated by two types of permutations $\theta_{i}$ for $1\leq i\leq d-1$ and $\tau_{i}$ for $1\leq i\leq d$, where * (i) the permutation $\theta_{i}$ interchanges the $i$-th and the $(i+1)$-st cycles of $\pi_{\mu_{m,d}}$, more precisely, $\theta_{i}(x_{i,k})=x_{i+1,k}$, $\theta_{i}(x_{i+1,k})=x_{i,k}$ and $\theta_{i}(x_{j,k})=x_{j,k}$ for $1\leq k\leq m$ and $j\not\in\\{i,i+1\\}$; * (ii) the permutation $\tau_{i}$ shifts the $i$-th cycle of $\pi_{\mu_{m,d}}$ by one in a way that $\tau_{i}(x_{i,k})=x_{i,k-1}$ and $\tau_{i}(x_{j,k})=x_{j,k}$ for $1\leq k\leq m$ and $j\neq i$. Notice that $\theta_{i}$ and $\tau_{i}$ are induced permutations of elements of $Z(\mu_{m,d})$. Let the punctures be located as in Figure 12 (a) so that $\mu_{m,d}$ corresponds to the clockwise $2\pi/m$-rotation. Figure 12 (b) and (c) illustrate diffeomorphisms whose induced permutations interchange two cycles or shift a cycle by one. ∎ | | | | ---|---|---|---|--- (a) | | (b) | | (c) Figure 12. (a) shows the punctured disk $D_{13}^{(3)}$. (b) and (c) illustrate diffeomorphisms inducing $\theta_{2}$ and $\tau_{3}$, respectively. ### 4.3. Reducible braids in the centralizer of a periodic braid Let $m$ and $d$ be integers with $m\geq 2$ and $d\geq 0$, and let $n=md+1$. Let $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ be a composition of $d+1$. We define a composition $L_{m}(\mathbf{d})$ of $n$ as $L_{m}(\mathbf{d})=(md_{0}+1,\underbrace{d_{1},\ldots,d_{1}}_{m},\ldots,\underbrace{d_{r},\ldots,d_{r}}_{m}).$ For example, if $\mathbf{d}=(2,2,1)$ then $L_{3}(\mathbf{d})=(4,2,2,2,1,1,1)$, and if $\mathbf{d}=(3,2,1)$ then $L_{3}(\mathbf{d})=(7,2,2,2,1,1,1)$. Let $\mathcal{C}_{m,\mathbf{d}}$ denote the curve system $\phi_{m,d}^{-1}(\mathcal{C}_{\mathbf{d}})$ in $D_{n}^{(m)}$. See Figure 13. $\scriptstyle{\eta_{m,\mathbf{d}}}$$\scriptstyle{\phi_{m,d}}$ $\mathcal{C}_{\mathbf{d}}$ in $D_{d+1}$ $\mathcal{C}_{m,\mathbf{d}}$ in $D_{n}^{(m)}$ $\mathcal{C}_{L_{m}(\mathbf{d})}$ in $D_{n}$ Figure 13. The curve system $\mathcal{C}_{\mathbf{d}}$ in $D_{d+1}$ is lifted to $\mathcal{C}_{m,\mathbf{d}}$ in $D_{n}^{(m)}$, and then mapped to $\mathcal{C}_{L_{m}(\mathbf{d})}$ in $D_{n}$. Here $n=13$, $m=3$, $d=4$ and $\mathbf{d}=(2,2,1)$. We construct a diffeomorphism $\eta_{m,\mathbf{d}}:D_{n}^{(m)}\to D_{n}$ similarly to $\eta_{m,d}$ such that it fixes the boundary $\partial D_{n}^{(m)}=\partial D_{n}$ pointwise and sends $\mathcal{C}_{m,\mathbf{d}}$ in $D_{n}^{(m)}$ to $\mathcal{C}_{L_{m}(\mathbf{d})}$ in $D_{n}$. First, we cut the punctured disk $D_{n}^{(m)}$ (resp. $D_{n}$) by $r$ round circles disjoint from $\mathcal{C}_{m,\mathbf{d}}$ (resp. $\mathcal{C}_{L_{m}(\mathbf{d})}$) into one $(md_{0}+1)$-punctured disk and $r$ annuli such that each annulus contains $m$ punctures or $m$ components of $\mathcal{C}_{m,\mathbf{d}}$ (resp. $\mathcal{C}_{L_{m}(\mathbf{d})}$) as in Figure 14 (a). Then, on each annulus, define $\eta_{m,\mathbf{d}}$ as in Figure 14 (b), and on the $(md_{0}+1)$-punctured disk, define $\eta_{m,\mathbf{d}}$ as $\eta_{m,d_{0}}$. Such a diffeomorphism $\eta_{m,\mathbf{d}}$ is unique up to isotopy. $\scriptstyle{\eta_{m,\mathbf{d}}}$ | $\scriptstyle{\eta_{m,\mathbf{d}}}$ ---|--- (a) | (b) Figure 14. The diffeomorphism $\eta_{m,\mathbf{d}}:D_{n}^{(m)}\to D_{n}$ where $n=13$, $m=3$, $d=4$ and $\mathbf{d}=(2,2,1)$. ###### Definition 4.10. Let $n=md+1$ with $m\geq 2$ and $d\geq 1$. Let $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ be a composition of $d+1$ and $\mathbf{n}=L_{m}(\mathbf{d})$. We define an $n$-braid $\mu_{m,\mathbf{d}}$ as $\mu_{m,\mathbf{d}}=\langle\mu_{m,r}\rangle_{\mathbf{n}}(\mu_{m,d_{0}}\oplus(\underbrace{\Delta_{(d_{1})}^{2}\oplus 1\oplus\cdots\oplus 1}_{m})\oplus\cdots\oplus(\underbrace{\Delta_{(d_{r})}^{2}\oplus 1\oplus\cdots\oplus 1}_{m}))_{\mathbf{n}}.$ See Figure 15 (b) for $\mu_{m,\mathbf{d}}$ with $m=3$ and $\mathbf{d}=(2,2,1)$ and compare it with $\mu_{m,r}$ in Figure 15 (a). $\mu_{m,\mathbf{d}}*\mathcal{C}_{L_{m}(\mathbf{d})}=\mathcal{C}_{L_{m}(\mathbf{d})}$ and $\mu_{m,\mathbf{d}}$ is conjugate to $\epsilon_{(n)}^{d}$. Observe that $\alpha\in B_{n}$ belongs to $Z(\mu_{m,\mathbf{d}})$ if and only if it is represented by $\eta_{m,\mathbf{d}}\circ h\circ\eta_{m,\mathbf{d}}^{-1}$ for a $\rho_{m}$-equivariant diffeomorphism $h:D_{n}^{(m)}\to D_{n}^{(m)}$ which fixes the boundary pointwise. | | | | ---|---|---|---|--- (a) $\mu_{3,2}$ | | (b) $\mu_{3,\mathbf{d}}$ | | (c) $\mathcal{C}_{L_{3}(\mathbf{d})}$ Figure 15. The braid $\mu_{3,\mathbf{d}}$ with $\mathbf{d}=(2,2,1)$ in (b) is obtained from the braid $\mu_{3,2}$ in (a) by taking four parallel copies of the first strand and two parallel copies of the next three strands, and then by concatenating some interior braids. This braid has the standard reduction system $\mathcal{C}_{L_{3}(\mathbf{d})}=\mathcal{C}_{(4,2,2,2,1,1,1)}$ as in (c). ###### Lemma 4.11. Let $n=md+1$ with $m\geq 2$ and $d\geq 1$. Let $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ be a composition of $d+1$, and let $\mathbf{n}=L_{m}(\mathbf{d})$. Let $\alpha$ be an $n$-braid with $\alpha*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$. Then, $\alpha\in Z(\mu_{m,\mathbf{d}})$ if and only if $\alpha$ is of the form $\alpha=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{0}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\underbrace{\alpha_{i}\Delta_{(d_{i})}^{2}\oplus\cdots\oplus\alpha_{i}\Delta_{(d_{i})}^{2}}_{\ell_{i}}\oplus\underbrace{\alpha_{i}\oplus\cdots\oplus\alpha_{i}}_{m-\ell_{i}}))_{\mathbf{n}}$ such that $\hat{\alpha}\in Z(\mu_{m,r})$, $\alpha_{0}\in Z(\mu_{m,d_{0}})$ and $d_{k_{i}}=d_{i}$ for $1\leq i\leq r$, where $k_{i}$’s and $\ell_{i}$’s are integers such that $\pi_{\hat{\alpha}}(x_{i,j})=x_{k_{i},j-\ell_{i}}$ as in Lemma 4.9. ###### Proof. Since $\alpha*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$, $\alpha$ is of the form $\alpha=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{0}\oplus\underbrace{\alpha_{1,1}\oplus\cdots\oplus\alpha_{1,m}}_{m}\oplus\cdots\oplus\underbrace{\alpha_{r,1}\oplus\cdots\oplus\alpha_{r,m}}_{m})_{\mathbf{n}}.$ Suppose that $\alpha\in Z(\mu_{m,\mathbf{d}})$. Since $\alpha\ast\mathcal{C}_{\mathbf{n}}=\mu_{m,\mathbf{d}}\ast\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$, we have $\hat{\alpha}\mu_{m,r}=\operatorname{Ext}_{\mathbf{n}}(\alpha\mu_{m,\mathbf{d}})=\operatorname{Ext}_{\mathbf{n}}(\mu_{m,\mathbf{d}}\alpha)=\mu_{m,r}\hat{\alpha},$ hence $\hat{\alpha}\in Z(\mu_{m,r})$. By Lemma 4.9, $\pi_{\hat{\alpha}}$ is of the form $\pi_{\hat{\alpha}}(x_{0})=x_{0}$ and $\pi_{\hat{\alpha}}(x_{i,j})=x_{k_{i},j-\ell_{i}}$, where $1\leq k_{i}\leq r$ and $0\leq\ell_{i}<m$. Since $\pi_{\hat{\alpha}}(x_{i,j})=x_{k_{i},j-\ell_{i}}$ and $\mathbf{n}=L_{m}(\mathbf{d})=(md_{0}+1,\underbrace{d_{1},\ldots,d_{1}}_{m},\ldots,\underbrace{d_{r},\ldots,d_{r}}_{m}),$ one has $\hat{\alpha}^{-1}*\mathbf{n}=(md_{0}+1,\underbrace{d_{k_{1}},\ldots,d_{k_{1}}}_{m},\ldots,\underbrace{d_{k_{r}},\ldots,d_{k_{r}}}_{m}).$ Therefore $d_{k_{i}}=d_{i}$ for $1\leq i\leq r$ because $\hat{\alpha}*\mathbf{n}=\mathbf{n}$. Now we compute $\alpha\mu_{m,\mathbf{d}}$ and $\mu_{m,\mathbf{d}}\alpha$ straightforwardly: $\displaystyle\alpha\cdot\mu_{m,\mathbf{d}}$ $\displaystyle=$ $\displaystyle\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{0}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\alpha_{i,1}\oplus\cdots\oplus\alpha_{i,m}))_{\mathbf{n}}$ $\displaystyle\mbox{}\quad\cdot\langle\mu_{m,r}\rangle_{\mathbf{n}}(\mu_{m,d_{0}}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\Delta_{(d_{i})}^{2}\oplus 1\oplus\cdots\oplus 1))_{\mathbf{n}}$ $\displaystyle=$ $\displaystyle\langle\hat{\alpha}\rangle_{\mathbf{n}}\langle\mu_{m,r}\rangle_{\mathbf{n}}\cdot(\alpha_{0}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\alpha_{i,m}\oplus\alpha_{i,1}\oplus\cdots\oplus\alpha_{i,m-1}))_{\mathbf{n}}$ $\displaystyle\mbox{}\quad\cdot(\mu_{m,d_{0}}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\Delta_{(d_{i})}^{2}\oplus 1\oplus\cdots\oplus 1))_{\mathbf{n}}$ $\displaystyle=$ $\displaystyle\langle\hat{\alpha}\mu_{m,r}\rangle_{\mathbf{n}}(\alpha_{0}\mu_{m,d_{0}}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\alpha_{i,m}\Delta_{(d_{i})}^{2}\oplus\alpha_{i,1}\oplus\cdots\oplus\alpha_{i,m-1}))_{\mathbf{n}},$ $\displaystyle\mu_{m,\mathbf{d}}\cdot\alpha$ $\displaystyle=$ $\displaystyle\langle\mu_{m,r}\rangle_{\mathbf{n}}(\mu_{m,d_{0}}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\Delta_{(d_{i})}^{2}\oplus 1\oplus\cdots\oplus 1))_{\mathbf{n}}$ $\displaystyle\mbox{}\qquad\cdot\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{0}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\alpha_{i,1}\oplus\cdots\oplus\alpha_{i,m}))_{\mathbf{n}}$ $\displaystyle=$ $\displaystyle\langle\mu_{m,r}\rangle_{\mathbf{n}}\langle\hat{\alpha}\rangle_{\mathbf{n}}\cdot(\mu_{m,d_{0}}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\underbrace{1\oplus\cdots\oplus 1}_{\ell_{i}}\oplus\Delta_{(d_{i})}^{2}\oplus\underbrace{1\oplus\cdots\oplus 1}_{m-\ell_{i}-1}))_{\mathbf{n}}$ $\displaystyle\mbox{}\qquad\cdot(\alpha_{0}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\alpha_{i,1}\oplus\cdots\oplus\alpha_{i,m}))_{\mathbf{n}}$ $\displaystyle=$ $\displaystyle\langle\mu_{m,r}\hat{\alpha}\rangle_{\mathbf{n}}(\mu_{m,d_{0}}\alpha_{0}\oplus$ $\displaystyle\mbox{}\qquad{\textstyle\bigoplus_{i=1}^{r}}(\alpha_{i,1}\oplus\cdots\oplus\alpha_{i,\ell_{i}}\oplus\Delta_{(d_{i})}^{2}\alpha_{i,\ell_{i}+1}\oplus\alpha_{i,\ell_{i}+2}\oplus\cdots\oplus\alpha_{i,m}))_{\mathbf{n}}.$ Comparing the above two formulae, we have $\alpha_{0}\in Z(\mu_{m,d_{0}})$. In addition, for $1\leq i\leq r$, $\displaystyle\alpha_{i,m}\Delta_{(d_{i})}^{2}=\alpha_{i,1}=\alpha_{i,2}=\cdots=\alpha_{i,\ell_{i}}=\Delta_{(d_{i})}^{2}\alpha_{i,\ell_{i}+1},$ $\displaystyle\alpha_{i,\ell_{i}+1}=\alpha_{i,\ell_{i}+2}=\cdots=\alpha_{i,m}.$ Let $\alpha_{i}=\alpha_{i,m}$, then $\alpha$ has the desired form. The other direction is straightforward. ∎ ###### Corollary 4.12. Let $n=md+1$ with $m\geq 2$ and $d\geq 1$. Let $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ be a composition of $d+1$, and let $\mathbf{n}=L_{m}(\mathbf{d})$. Let $\alpha$ be an $n$-braid with $\alpha*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$, and let the $\mathbf{n}$-exterior braid of $\alpha$ be pure. Then, $\alpha\in Z(\mu_{m,\mathbf{d}})$ if and only if $\alpha$ is of the form $\alpha=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{0}\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}}$ where $\hat{\alpha}\in Z(\mu_{m,r})$ and $\alpha_{0}\in Z(\mu_{m,d_{0}})$. ###### Proof. Since $\operatorname{Ext}_{\mathbf{n}}(\alpha)$ is pure, $k_{i}=i$ and $\ell_{i}=0$ for $i=1,\ldots,r$ in Lemma 4.11. ∎ ###### Lemma 4.13. Let $n=md+1$ with $m\geq 2$ and $d\geq 1$. Let $\mathcal{C}$ be an unnested curve system in $D_{n}$ such that $\mu_{m,d}*\mathcal{C}=\mathcal{C}$. Then there exist $\zeta\in Z(\mu_{m,d})$ and a composition $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ of $d+1$ with $d_{1}\geq\cdots\geq d_{r}$ such that $(\chi\zeta)*\mathcal{C}=\mathcal{C}_{L_{m}(\mathbf{d})},$ where $\chi$ is an $n$-braid represented by $\eta_{m,\mathbf{d}}\circ\eta_{m,d}^{-1}$. ###### Proof. The construction of $\mathbf{d}$ and $\zeta$ can be viewed in Figure 16. Let $\mathcal{C}^{\prime}=\eta_{m,d}^{-1}(\mathcal{C})$. It is a curve system in $D_{n}^{(m)}$ such that $\rho_{m}(\mathcal{C}^{\prime})$ is isotopic to $\mathcal{C}^{\prime}$. --- $\mathcal{C}$$\scriptstyle{h}$$\scriptstyle{f}$$\mathcal{C}_{L_{m}(\mathbf{d})}$ $\mathcal{C}^{\prime}$$\scriptstyle{g}$$\scriptstyle{\phi_{m,d}}$$\scriptstyle{\eta_{m,d}}$$\mathcal{C}_{m,\mathbf{d}}$$\scriptstyle{\phi_{m,d}}$$\scriptstyle{\eta_{m,d}}$$\scriptstyle{\eta_{m,\mathbf{d}}}$ $\bar{\mathcal{C}}^{\prime}$$\scriptstyle{\bar{g}}$$\mathcal{C}_{\mathbf{d}}$ Figure 16. This figure shows the constructions in the proof of Lemma 4.13. _Claim._ _Without loss of generality, we may assume that $\rho_{m}(\mathcal{C}^{\prime})=\mathcal{C}^{\prime}$._ ###### Proof. Give a complete hyperbolic geometry on the interior of $D_{d+1}$, and lift it to the interior of $D_{n}^{(m)}$ by using the covering $\phi_{m,d}:D_{n}^{(m)}\to D_{d+1}$. Then $\rho_{m}$ is an isometry. Let $\mathcal{C}^{\prime\prime}$ be the unique geodesic curve system isotopic to $\mathcal{C}^{\prime}$. Because $\rho_{m}(\mathcal{C}^{\prime\prime})$ is isotopic to $\mathcal{C}^{\prime\prime}$, we have $\rho_{m}(\mathcal{C}^{\prime\prime})=\mathcal{C}^{\prime\prime}$. ∎ Now we assume that $\rho_{m}(\mathcal{C}^{\prime})=\mathcal{C}^{\prime}$. Let $\bar{\mathcal{C}}^{\prime}=\phi_{m,d}(\mathcal{C}^{\prime})$. Because $\bar{\mathcal{C}}^{\prime}$ is an unnested curve system in $D_{d+1}$, there exists a diffeomorphism $\bar{g}:D_{d+1}\to D_{d+1}$ fixing the first puncture and the points at $\partial D_{d+1}$ such that $\bar{g}(\bar{\mathcal{C}}^{\prime})$ is a standard unnested curve system, hence $\bar{g}(\bar{\mathcal{C}}^{\prime})=\mathcal{C}_{\mathbf{d}}$ for a composition $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ of $d+1$. Note that $\bar{g}$ can be chosen so that $d_{1}\geq d_{2}\geq\cdots\geq d_{r}$. Let $g:D_{n}^{(m)}\to D_{n}^{(m)}$ be the lift of $\bar{g}$ that fixes $\partial D_{n}^{(m)}$ pointwise. Then $g$ is a $\rho_{m}$-equivariant diffeomorphism with $g(\mathcal{C}^{\prime})=\phi_{m,d}^{-1}(\mathcal{C}_{\mathbf{d}})=\mathcal{C}_{m,\mathbf{d}}$. Let $h=\eta_{m,d}\circ g\circ\eta_{m,d}^{-1}$ and $f=\eta_{m,\mathbf{d}}\circ\eta_{m,d}^{-1}$. Let $\zeta$ and $\chi$ be $n$-braids represented by $h$ and $f$, respectively. Because $g$ is $\rho_{m}$-equivariant, $\zeta\in Z(\mu_{m,d})$. By the construction, $(f\circ h)(\mathcal{C})=\mathcal{C}_{L_{m}(\mathbf{d})}$, hence $(\chi\zeta)*\mathcal{C}=\mathcal{C}_{L_{m}(\mathbf{d})}$. ∎ ###### Corollary 4.14. Let $n=md+1$ with $m\geq 2$ and $d\geq 1$. Let $\mathcal{C}_{1}$ and $\mathcal{C}_{2}$ be unnested curve systems in $D_{n}$ that are of the same type and invariant under the action of $\mu_{m,d}$. Then there exist $\chi\in B_{n}$, $\chi_{1},\chi_{2}\in Z(\mu_{m,d})$ and a composition $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ of $d+1$ such that $\chi\mu_{m,d}\chi^{-1}=\mu_{m,\mathbf{d}}\quad\mbox{and}\quad(\chi\chi_{1})*\mathcal{C}_{1}=(\chi\chi_{2})*\mathcal{C}_{2}=\mathcal{C}_{L_{m}(\mathbf{d})}.$ ###### Proof. Applying Lemma 4.13 to $\mathcal{C}_{1}$ and $\mathcal{C}_{2}$, we have $\chi_{1},\chi_{2}\in Z(\mu_{m,d})$ and compositions $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ and $\mathbf{d}^{\prime}=(d_{0}^{\prime}+1,d_{1}^{\prime},\ldots,d_{r^{\prime}}^{\prime})$ of $d+1$ with $d_{1}\geq\cdots\geq d_{r}$ and $d_{1}^{\prime}\geq\cdots\geq d_{r^{\prime}}^{\prime}$ such that $(\chi\chi_{1})*\mathcal{C}_{1}=\mathcal{C}_{L_{m}(\mathbf{d})}\quad\mbox{and}\quad(\chi^{\prime}\chi_{2})*\mathcal{C}_{2}=\mathcal{C}_{L_{m}(\mathbf{d}^{\prime})},$ where $\chi$ and $\chi^{\prime}$ are represented by $\eta_{m,\mathbf{d}}\circ\eta_{m,d}^{-1}$ and $\eta_{m,\mathbf{d}^{\prime}}\circ\eta_{m,d}^{-1}$, respectively. Because $\mathcal{C}_{1}$ and $\mathcal{C}_{2}$ are of the same type, so are $\mathcal{C}_{L_{m}(\mathbf{d})}$ and $\mathcal{C}_{L_{m}(\mathbf{d}^{\prime})}$. This implies $\mathbf{d}=\mathbf{d}^{\prime}$ because $d_{1}\geq\cdots\geq d_{r}$ and $d_{1}^{\prime}\geq\cdots\geq d_{r^{\prime}}^{\prime}$. Therefore $\chi=\chi^{\prime}$. Notice that the diffeomorphisms $(\eta_{m,\mathbf{d}}\circ\eta_{m,d}^{-1})\circ(\eta_{m,d}\circ\rho_{m}\circ\eta_{m,d}^{-1})\circ(\eta_{m,\mathbf{d}}\circ\eta_{m,d}^{-1})^{-1}$ and $\eta_{m,\mathbf{d}}\circ\rho_{m}\circ\eta_{m,\mathbf{d}}^{-1}$ represent $\chi\mu_{m,d}\chi^{-1}$ and $\mu_{m,\mathbf{d}}$, respectively, modulo $\Delta^{2}$. Since the two diffeomorphisms are the same, $\chi\mu_{m,d}\chi^{-1}=\mu_{m,\mathbf{d}}\bmod\Delta^{2}$. Therefore $\chi\mu_{m,d}\chi^{-1}=\mu_{m,\mathbf{d}}$ because $\chi\mu_{m,d}\chi^{-1}$ and $\mu_{m,\mathbf{d}}$ are conjugate. ∎ ###### Definition 4.15. Let $\Lambda$ be a choice function for the set $\cup_{k\geq 1}B_{k}$, hence $\Lambda(A)\in A$ for $A\subset\cup_{k\geq 1}B_{k}$. For $\alpha\in B_{k}$, let $[\alpha]$ denote the conjugacy class of $\alpha$ in $B_{k}$. Let $m\geq 2$ and $t\geq 0$. We define $\lambda=\lambda_{m,t}:\cup_{k\geq 1}B_{k}\to\cup_{k\geq 1}B_{k}$ as $\lambda(\alpha)=\Lambda(Z(\mu_{m,t})\cap[\alpha])$ if $\alpha\in B_{mt+1}$ and $[\alpha]\cap Z(\mu_{m,t})\neq\emptyset$, and $\lambda(\alpha)=\Lambda([\alpha])$ otherwise. We call $\lambda(\alpha)$ the _conjugacy representative_ of $\alpha$ with respect to $(m,t)$. Then $\lambda$ has the following properties. * (i) $\lambda(\alpha)$ is conjugate to $\alpha$. Two braids $\alpha$ and $\beta$ are conjugate if and only if $\lambda(\alpha)=\lambda(\beta)$. * (ii) If $\alpha\in Z(\mu_{m,t})$, then $\lambda(\alpha)\in Z(\mu_{m,t})$. ###### Lemma 4.16. Let $n=md+1$ with $m\geq 2$ and $d\geq 1$. Let $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ be a composition of $d+1$, and let $\mathbf{n}=L_{m}(\mathbf{d})$. Let $\alpha=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{0}\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}}$ belong to $Z(\mu_{m,\mathbf{d}})$ such that $\hat{\alpha}$ is a pure braid. Let $\lambda=\lambda_{m,d_{0}}$. Suppose that $\alpha_{0}$ and $\lambda(\alpha_{0})$ are conjugate in $Z(\mu_{m,d_{0}})$. Then there exists an $\mathbf{n}$-split braid $\gamma$ in $Z(\mu_{m,\mathbf{d}})$ such that $\gamma\alpha\gamma^{-1}=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\lambda(\alpha_{0})\oplus m\lambda(\alpha_{1})\oplus\cdots\oplus m\lambda(\alpha_{r}))_{\mathbf{n}}.$ ###### Proof. Since $\alpha\in Z(\mu_{m,\mathbf{d}})$, one has $\alpha_{0}\in Z(\mu_{m,d_{0}})$ by Lemma 4.11, hence $\lambda(\alpha_{0})\in Z(\mu_{m,d_{0}})$. By the hypothesis, there exists $\gamma_{0}\in Z(\mu_{m,d_{0}})$ such that $\gamma_{0}\alpha_{0}\gamma_{0}^{-1}=\lambda(\alpha_{0})$. For $1\leq i\leq r$, choose $\gamma_{i}\in B_{d_{i}}$ such that $\gamma_{i}\alpha_{i}\gamma_{i}^{-1}=\lambda(\alpha_{i})$. Using $\gamma_{i}$’s, define an $\mathbf{n}$-split braid $\gamma$ as $\gamma=\langle 1\rangle_{\mathbf{n}}(\gamma_{0}\oplus m\gamma_{1}\oplus\cdots\oplus m\gamma_{r})_{\mathbf{n}}.$ By Corollary 4.12, $\gamma\in Z(\mu_{m,\mathbf{d}})$. Because the $\mathbf{n}$-exterior braids of $\alpha$ and $\gamma$ are pure braids, we can compute the conjugation $\gamma\alpha\gamma^{-1}$ component-wise as follows. $\displaystyle\gamma\alpha\gamma^{-1}$ $\displaystyle=$ $\displaystyle\langle 1\rangle_{\mathbf{n}}\langle\hat{\alpha}\rangle_{\mathbf{n}}\langle 1\rangle_{\mathbf{n}}^{-1}(\gamma_{0}\alpha_{0}\gamma_{0}^{-1}\oplus m(\gamma_{1}\alpha_{1}\gamma_{1}^{-1})\oplus\cdots\oplus m(\gamma_{r}\alpha_{r}\gamma_{r}^{-1}))_{\mathbf{n}}$ $\displaystyle=$ $\displaystyle\langle\hat{\alpha}\rangle_{\mathbf{n}}(\lambda(\alpha_{0})\oplus m\lambda(\alpha_{1})\oplus\cdots\oplus m\lambda(\alpha_{r}))_{\mathbf{n}}.$ This completes the proof. ∎ ## 5\. Proof of Theorem 1.4 In this section we prove Theorem 1.4. Let us explain our strategy. Let $\omega$ be a periodic $n$-braid, and let $\alpha,\beta\in Z(\omega)$ be conjugate in $B_{n}$. If $\alpha$ and $\beta$ are periodic, it is easy to prove the theorem because we may assume that $\alpha$ and $\beta$ are central by Corollary 2.4. If $\alpha$ and $\beta$ are pseudo-Anosov, then we obtain a stronger result in Proposition 5.1 that any conjugating element between $\alpha$ and $\beta$ must be contained in $Z(\omega)$. (This result will be crucial in dealing with reducible braids.) However a far more laborious task is required when $\alpha$ and $\beta$ are reducible. Let us say that a periodic braid is _$\delta$ -type_ (resp. _$\epsilon$ -type_) if it is conjugate to a power of $\delta$ (resp. a power of $\epsilon$). The periodic braid $\omega$ is either $\delta$-type or $\epsilon$-type, and Theorem 1.4 can be proved similarly for the two cases. One possible strategy for escaping repetition of similar arguments would be writing an explicit proof for one case and then showing that the other case follows easily from that. However this strategy hardly works. It seems that (when $\alpha$ and $\beta$ are reducible) the proof for $\delta$-type $\omega$ must contain most of the whole arguments needed in the proof for $\epsilon$-type $\omega$, and that the theorem for $\delta$-type $\omega$ is not a consequence of that for $\epsilon$-type $\omega$. Proposition 5.2 is a key step in the proof of the theorem when $\alpha$ and $\beta$ are reducible. The proposition may look rather unnatural at a glance, but it is written in that form in order to minimize repetition of similar arguments. ###### Proposition 5.1. Let $\omega$ be a periodic $n$-braid, and let $\alpha$ and $\beta$ be pseudo- Anosov braids in $Z(\omega)$. If $\gamma$ is an $n$-braid with $\beta=\gamma\alpha\gamma^{-1}$, then $\gamma\in Z(\omega)$. ###### Proof. Let $I_{\omega}:B_{n}\to B_{n}$ be the inner automorphism such that $I_{\omega}(\chi)=\omega^{-1}\chi\omega$ for $\chi\in B_{n}$. Since $\alpha,\beta\in Z(\omega)$, we have $I_{\omega}(\alpha)=\alpha$ and $I_{\omega}(\beta)=\beta$. Therefore $\gamma\alpha\gamma^{-1}=\beta=I_{\omega}(\beta)=I_{\omega}(\gamma\alpha\gamma^{-1})=I_{\omega}(\gamma)I_{\omega}(\alpha)I_{\omega}(\gamma^{-1})=I_{\omega}(\gamma)\alpha I_{\omega}(\gamma)^{-1}.$ Let $\gamma_{0}=\gamma^{-1}I_{\omega}(\gamma)$. Then $\gamma_{0}\in Z(\alpha)$. It is known by J. González-Meneses and B. Wiest [GW04] that the centralizer of a pseudo-Anosov braid is a free abelian group of rank two. Since $\alpha$ is pseudo-Anosov and $\gamma_{0},\omega\in Z(\alpha)$, $\gamma_{0}$ commutes with $\omega$, hence $I_{\omega}^{k}(\gamma_{0})=\gamma_{0}$ for all $k$. Take a positive integer $m$ such that $\omega^{m}$ is central. Since $I_{\omega}^{m}(\gamma)=\gamma$, $\displaystyle\gamma_{0}^{m}$ $\displaystyle=$ $\displaystyle\gamma_{0}\gamma_{0}\cdots\gamma_{0}=\gamma_{0}\cdot I_{\omega}(\gamma_{0})\cdots I_{\omega}^{m-1}(\gamma_{0})$ $\displaystyle=$ $\displaystyle\gamma^{-1}I_{\omega}(\gamma)\cdot I_{\omega}(\gamma^{-1})I_{\omega}^{2}(\gamma)\cdots I_{\omega}^{m-1}(\gamma^{-1})I_{\omega}^{m}(\gamma)$ $\displaystyle=$ $\displaystyle\gamma^{-1}I_{\omega}^{m}(\gamma)=\gamma^{-1}\gamma=1.$ Since braid groups are torsion free (for example see [Deh98]), we have $\gamma_{0}=1$, hence $\gamma=I_{\omega}(\gamma)$. This means that $\gamma\in Z(\omega)$. ∎ We remark that, in the above proposition, the pseudo-Anosov condition on $\alpha$ and $\beta$ is necessary. Let $\omega$ be a non-central periodic braid, and let $\gamma$ be a braid which does not belong to $Z(\omega)$. Let $\alpha=\beta=\Delta^{2}$. Because $\Delta^{2}$ is central, we have $\alpha,\beta\in Z(\omega)$ and $\gamma\alpha\gamma^{-1}=\beta$. However, $\gamma\not\in Z(\omega)$. The following proposition is a key step in the proof of Theorem 1.4. Recall that $\nu:B_{n,1}\to B_{n-1}$ is the homomorphism deleting the first strand. ###### Proposition 5.2. Let $m\geq 2$ and $d\geq 1$. Let $\omega$ be a 1-pure braid conjugate to $\epsilon_{(md+1)}^{d}$. Let $\alpha$ and $\beta$ be pure braids in $Z(\omega)$ such that either $\alpha$ and $\beta$ are conjugate in $B_{md+1}$ or $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{md}$. Then $\alpha$ and $\beta$ are conjugate in $Z(\omega)$. Before proving the above proposition, we prove Theorem 1.4 assuming the proposition. ###### Proof of Theorem 1.4. Let $\omega$ be a periodic $n$-braid, and let $\alpha,\beta\in Z(\omega)$ be conjugate in $B_{n}$. We will show that $\alpha$ and $\beta$ are conjugate in $Z(\omega)$. _Claim._ Without loss of generality, we may assume that $\omega$ is non- central, and both $\alpha$ and $\beta$ are pure braids. ###### Proof of Claim. If $\omega$ is central, then $Z(\omega)=B_{n}$, hence there is nothing to prove. Hence we may assume that $\omega$ is non-central. Assume that Theorem 1.4 holds for pure braids. Choose a positive integer $k$ such that $\alpha^{k}$ and $\beta^{k}$ are pure. By assumption, $\alpha^{k}$ and $\beta^{k}$ are conjugate in $Z(\omega)$. Because $Z(\omega)$ is isomorphic to $A(\mathbf{B}_{\ell})$ for some $\ell\geq 1$, $\alpha$ and $\beta$ are conjugate in $Z(\omega)$ by Corollary 2.4. This shows that it suffices to prove Theorem 1.4 for the case where $\alpha$ and $\beta$ are pure braids. ∎ Now we assume that $\omega$ is noncentral, and both $\alpha$ and $\beta$ are pure braids. By Lemma 4.1 (i), we may assume that $\omega=\delta^{d}$ or $\omega=\epsilon^{d}$ for some $d$. By Lemma 2.1, we may further assume that $d$ is a positive integer which is a divisor of $n$ if $\omega=\delta^{d}$ and a divisor of $n-1$ if $\omega=\epsilon^{d}$. Therefore there is a positive integer $m$ such that $n=md$ if $\omega=\delta^{d}$ and $n=md+1$ if $\omega=\epsilon^{d}$. Note that $m\geq 2$ because $\omega$ is central if $m=1$. If $\omega=\epsilon^{d}$, then $\alpha$ and $\beta$ are conjugate in $Z(\omega)$ by Proposition 5.2. Now, suppose that $\omega=\delta^{d}$. Let $\bar{\alpha}$ and $\bar{\beta}$ be the elements of $B_{d+1,1}$ such that $\psi_{5}(\bar{\alpha})=\alpha$ and $\psi_{5}(\bar{\beta})=\beta$. Let $\alpha^{\prime}=\psi_{4}(\bar{\alpha})$ and $\beta^{\prime}=\psi_{4}(\bar{\beta})$. Then $\alpha^{\prime}$ and $\beta^{\prime}$ are $(n+1)$-braids in $Z(\epsilon_{(n+1)}^{d})$ such that $\nu(\alpha^{\prime})=\alpha$ and $\nu(\beta^{\prime})=\beta$ are conjugate in $B_{n}$. --- $\textstyle{B_{n}\supset Z(\delta_{(n)}^{d})}$$\textstyle{Z(\epsilon_{(n+1)}^{d})\subset B_{n+1,1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\nu}$$\textstyle{B_{d+1,1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\psi_{5}}$$\scriptstyle{\psi_{4}}$ --- $\textstyle{\alpha,~{}\beta}$$\textstyle{\alpha^{\prime},~{}\beta^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\nu}$$\textstyle{\bar{\alpha},~{}\bar{\beta}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\psi_{5}}$$\scriptstyle{\psi_{4}}$ Since $\alpha$ and $\beta$ are pure, so are $\alpha^{\prime}$ and $\beta^{\prime}$. Applying Proposition 5.2 to $(\alpha^{\prime},\beta^{\prime},\epsilon_{(n+1)}^{d})$, we conclude that $\alpha^{\prime}$ and $\beta^{\prime}$ are conjugate in $Z(\epsilon_{(n+1)}^{d})$. This implies that $\bar{\alpha}$ and $\bar{\beta}$ are conjugate in $B_{d+1,1}$, hence $\alpha$ and $\beta$ are conjugate in $Z(\delta^{d})=Z(\omega)$. ∎ ###### Proof of Proposition 5.2. Let $n=md+1$. Recall that $m\geq 2$ and $d\geq 1$; $\omega$ is a 1-pure braid conjugate to $\epsilon_{(n)}^{d}$; $\alpha$ and $\beta$ are pure braids in $Z(\omega)$ such that either $\alpha$ and $\beta$ are conjugate in $B_{n}$ or $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$. We will show that $\alpha$ and $\beta$ are conjugate in $Z(\omega)$ by using induction on $d\geq 1$. By Lemma 4.1, we may assume that $\omega=\mu_{m,d}$. _Claim 1._ * (i) $\alpha$ and $\beta$ are conjugate in $Z(\omega)$ if and only if $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $Z(\nu(\omega))$. * (ii) The braids $\alpha$, $\beta$, $\nu(\alpha)$ and $\nu(\beta)$ have the same Nielsen-Thurston type. * (iii) The proposition holds if $\alpha$ is periodic or pseudo-Anosov, or if $d=1$. * (iv) If $\alpha$ is non-periodic and reducible, then $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$ are of the same type. ###### Proof. Let $\bar{\alpha}$ and $\bar{\beta}$ be the elements of $B_{d+1,1}$ such that $\alpha=\psi_{4}^{\prime}(\bar{\alpha})$ and $\beta=\psi_{4}^{\prime}(\bar{\beta})$, hence $\nu(\alpha)=\psi_{5}^{\prime}(\bar{\alpha})$ and $\nu(\beta)=\psi_{5}^{\prime}(\bar{\beta})$. We have the following commutative diagram, where the isomorphisms preserve the Nielsen-Thurston type by Lemma 4.6 and Corollary 4.7. --- $\textstyle{Z(\nu(\omega))}$$\textstyle{Z(\omega)\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\nu}$$\scriptstyle{\simeq}$$\textstyle{B_{d+1,1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\psi_{5}^{\prime}}$$\scriptstyle{\simeq}$$\scriptstyle{\psi_{4}^{\prime}}$$\scriptstyle{\simeq}$ --- $\textstyle{\nu(\alpha),~{}\nu(\beta)}$$\textstyle{\alpha,~{}\beta\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\nu}$$\textstyle{\bar{\alpha},~{}\bar{\beta}\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}$$\scriptstyle{\psi_{5}^{\prime}}$$\scriptstyle{\psi_{4}^{\prime}}$ (i) This follows from the fact that $\nu:Z(\omega)\to Z(\nu(\omega))$ is an isomorphism. (ii) Because $\nu$ preserves the Nielsen-Thurston type, $(\alpha,\nu(\alpha))$ and $(\beta,\nu(\beta))$ are pairs of braids with the same Nielsen-Thurston type. By the hypothesis, either $(\alpha,\beta)$ or $(\nu(\alpha),\nu(\beta))$ is a pair of braids with the same Nielsen-Thurston type. Therefore the braids $\alpha$, $\beta$, $\nu(\alpha)$ and $\nu(\beta)$ have the same Nielsen- Thurston type. (iii) Suppose that $\alpha$ is pseudo-Anosov. Then $\beta$, $\nu(\alpha)$ and $\nu(\beta)$ are pseudo-Anosov by (ii). If $\alpha$ and $\beta$ are conjugate in $B_{n}$, they are conjugate in $Z(\omega)$ by Proposition 5.1. If $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$, they are conjugate in $Z(\nu(\omega))$ by Proposition 5.1, hence $\alpha$ and $\beta$ are conjugate in $Z(\omega)$ by (i). Suppose that $\alpha$ is periodic. Then $\beta$, $\nu(\alpha)$ and $\nu(\beta)$ are periodic by (ii). Since $\alpha$, $\beta$, $\nu(\alpha)$ and $\nu(\beta)$ are periodic and pure, they are central. If $\alpha$ and $\beta$ are conjugate in $B_{n}$, then $\alpha=\beta$. If $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$, then $\nu(\alpha)=\nu(\beta)$, hence $\alpha=\beta$. Suppose that $d=1$. Because $\bar{\alpha}\in B_{2,1}=\\{\sigma_{1}^{2k}\mid k\in\mathbf{Z}\\}=\\{\Delta_{(2)}^{2k}\mid k\in\mathbf{Z}\\}$, $\bar{\alpha}$ is periodic. Hence $\alpha$ is periodic. From the above discussion, $\alpha$ and $\beta$ are conjugate in $Z(\omega)$. (iv) The braids $\nu(\alpha)$, $\beta$ and $\nu(\beta)$ are non-periodic and reducible by (ii). If $\alpha$ and $\beta$ are conjugate in $B_{n}$, then $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$ are of the same type. Suppose that $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$. Taking conjugates of $\bar{\alpha}$ and $\bar{\beta}$ by elements of $B_{d+1,1}$ if necessary, we may assume that $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\bar{\alpha})$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\bar{\beta})$ are standard. Let $\mathbf{d}_{1}=(d_{0}+1,d_{1},\ldots,d_{r})$ and $\mathbf{d}_{2}=(e_{0}+1,e_{1},\ldots,e_{s})$ be the compositions of $d+1$ such that $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\bar{\alpha})=\mathcal{C}_{\mathbf{d}_{1}}$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\bar{\beta})=\mathcal{C}_{\mathbf{d}_{2}}$. Let $\begin{array}[]{*7{l}}\mathbf{n}_{1}&=&(md_{0}+1,\underbrace{d_{1},\ldots,d_{1}}_{m},\ldots,\underbrace{d_{r},\ldots,d_{r}}_{m}),&&\mathbf{n}_{1}^{\prime}&=&(md_{0},\underbrace{d_{1},\ldots,d_{1}}_{m},\ldots,\underbrace{d_{r},\ldots,d_{r}}_{m}),\\\ \mathbf{n}_{2}&=&(me_{0}+1,\underbrace{e_{1},\ldots,e_{1}}_{m},\ldots,\underbrace{e_{s},\ldots,e_{s}}_{m}),&&\mathbf{n}_{2}^{\prime}&=&(me_{0},\underbrace{e_{1},\ldots,e_{1}}_{m},\ldots,\underbrace{e_{s},\ldots,e_{s}}_{m}),\end{array}$ where $md_{0}$ (resp. $me_{0}$) is deleted if $d_{0}=0$ (resp. $e_{0}=0$). Let us write “$\mathcal{C}_{1}\approx\mathcal{C}_{2}$” if two curve systems $\mathcal{C}_{1}$ and $\mathcal{C}_{2}$ are of the same type. By Corollary 4.8, $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)\approx\mathcal{C}_{\mathbf{n}_{1}},\quad\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)\approx\mathcal{C}_{\mathbf{n}_{2}},\quad\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\alpha))\approx\mathcal{C}_{\mathbf{n}_{1}^{\prime}},\quad\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\beta))\approx\mathcal{C}_{\mathbf{n}_{2}^{\prime}}.$ Since $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$, one has $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\alpha))\approx\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\beta))$, and hence $\mathcal{C}_{\mathbf{n}_{1}^{\prime}}\approx\mathcal{C}_{\mathbf{n}_{2}^{\prime}}$. Thus the following equality holds between multisets consisting of integers: $\\{md_{0},\underbrace{d_{1},\ldots,d_{1}}_{m},\ldots,\underbrace{d_{r},\ldots,d_{r}}_{m}\\}=\\{me_{0},\underbrace{e_{1},\ldots,e_{1}}_{m},\ldots,\underbrace{e_{s},\ldots,e_{s}}_{m}\\}.$ Since $m\geq 2$, one has $r=s$ and $d_{0}=e_{0}$ and there is an $r$-permutation $\theta$ such that $d_{i}=e_{\theta(i)}$ for $1\leq i\leq r$. Therefore $\mathcal{C}_{\mathbf{n}_{1}}$ and $\mathcal{C}_{\mathbf{n}_{2}}$ are of the same type, hence so are $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$. ∎ By the above claim, the proposition holds if $d=1$, or if $\alpha$ is periodic or pseudo-Anosov. Thus we may assume that $d\geq 2$ and that $\alpha$ is non- periodic and reducible. _Claim 2._ Without loss of generality, we may assume that $\omega=\mu_{m,\mathbf{d}}$ for a composition $\mathbf{d}=(d_{0}+1,d_{1},\ldots,d_{r})$ of $d+1$, and that $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)=\mathcal{C}_{\mathbf{n}}$ where $\mathbf{n}=L_{m}(\mathbf{d})$. ###### Proof of Claim 2. Recall that we have assumed that $\omega=\mu_{m,d}$. By Claim 1 (iv), $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$ are of the same type. They are invariant under the action of $\omega$. By Corollary 4.14, there exist $\chi\in B_{n}$, $\chi_{1},\chi_{2}\in Z(\omega)$ and a composition $\mathbf{d}$ of $d+1$ such that $\chi\omega\chi^{-1}=\mu_{m,\mathbf{d}}$ and $(\chi\chi_{1})*\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)=(\chi\chi_{2})*\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)=\mathcal{C}_{L_{m}(\mathbf{d})}$. Notice that $(\chi\chi_{1})*\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\chi\chi_{1}\alpha\chi_{1}^{-1}\chi^{-1})$ and $(\chi\chi_{2})*\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\chi\chi_{2}\beta\chi_{2}^{-1}\chi^{-1})$. By Lemma 4.1, it suffices to show that $\chi\chi_{1}\alpha\chi_{1}^{-1}\chi^{-1}$ and $\chi\chi_{2}\beta\chi_{2}^{-1}\chi^{-1}$ are conjugate in $Z(\chi\omega\chi^{-1})=Z(\mu_{m,\mathbf{d}})$. ∎ Let us assume the hypothesis stated in Claim 2. By Corollary 4.12, $\alpha$ and $\beta$ are written as $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{0}\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}},$ $\displaystyle\beta$ $\displaystyle=$ $\displaystyle\langle\hat{\beta}\rangle_{\mathbf{n}}(\beta_{0}\oplus m\beta_{1}\oplus\cdots\oplus m\beta_{r})_{\mathbf{n}},$ where $\hat{\alpha},\hat{\beta}\in Z(\mu_{m,r})$ and $\alpha_{0},\beta_{0}\in Z(\mu_{m,d_{0}})$. Let $\lambda=\lambda_{m,d_{0}}$. By the induction hypothesis, $\alpha_{0}$ and $\lambda(\alpha_{0})$ (resp. $\beta_{0}$ and $\lambda(\beta_{0})$) are conjugate in $Z(\mu_{m,d_{0}})$. By Lemmas 4.16 and 4.1 (ii), we may assume that $\alpha_{i}$’s and $\beta_{i}$’s are conjugacy representatives, that is, $\alpha_{i}=\lambda(\alpha_{i})$ and $\beta_{i}=\lambda(\beta_{i})$ for $0\leq i\leq r$. The braids $\nu(\alpha)$ and $\nu(\beta)$ are written differently according to whether $d_{0}\geq 1$ or $d_{0}=0$. Let $\mathbf{n}^{\prime}$ be a compositions of $n-1$ such that $\begin{array}[]{lcll}\mathbf{n}^{\prime}&=&(md_{0},\underbrace{d_{1},\ldots,d_{1}}_{m},\ldots,\underbrace{d_{r},\ldots,d_{r}}_{m})&\mbox{if $d_{0}\geq 1$},\\\ \mathbf{n}^{\prime}&=&(\underbrace{d_{1},\ldots,d_{1}}_{m},\ldots,\underbrace{d_{r},\ldots,d_{r}}_{m})&\mbox{if $d_{0}=0$}.\end{array}$ By Corollary 4.8, $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\alpha))$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\beta))$ are obtained from $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)$ and $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)$, respectively, by forgetting the first puncture. Since $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)=\mathcal{C}_{\mathbf{n}}$, one has $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\alpha))=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\beta))=\mathcal{C}_{\mathbf{n}^{\prime}}.$ If $d_{0}\geq 1$, then $\nu(\mu_{m,\mathbf{d}})$, $\nu(\alpha)$ and $\nu(\beta)$ are written as $\displaystyle\nu(\mu_{m,\mathbf{d}})$ $\displaystyle=$ $\displaystyle\langle\mu_{m,r}\rangle_{\mathbf{n}^{\prime}}(\nu(\mu_{m,d_{0}})\oplus{\textstyle\bigoplus_{i=1}^{r}}(\underbrace{\Delta_{(d_{i})}^{2}\oplus 1\oplus\cdots\oplus 1}_{m}))_{\mathbf{n}^{\prime}},$ $\displaystyle\nu(\alpha)$ $\displaystyle=$ $\displaystyle\langle\hat{\alpha}\rangle_{\mathbf{n}^{\prime}}(\nu(\alpha_{0})\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}^{\prime}},$ $\displaystyle\nu(\beta)$ $\displaystyle=$ $\displaystyle\langle\hat{\beta}\rangle_{\mathbf{n}^{\prime}}(\nu(\beta_{0})\oplus m\beta_{1}\oplus\cdots\oplus m\beta_{r})_{\mathbf{n}^{\prime}}.$ In this case, $\nu(\alpha_{0}),\nu(\beta_{0})\in Z(\nu(\mu_{m,d_{0}}))$. If $d_{0}=0$, then $\nu(\mu_{m,\mathbf{d}})$, $\nu(\alpha)$ and $\nu(\beta)$ are written as $\displaystyle\nu(\mu_{m,\mathbf{d}})$ $\displaystyle=$ $\displaystyle\langle\nu(\mu_{m,r})\rangle_{\mathbf{n}^{\prime}}({\textstyle\bigoplus_{i=1}^{r}}(\underbrace{\Delta_{(d_{i})}^{2}\oplus 1\oplus\cdots\oplus 1}_{m}))_{\mathbf{n}^{\prime}},$ $\displaystyle\nu(\alpha)$ $\displaystyle=$ $\displaystyle\langle\nu(\hat{\alpha})\rangle_{\mathbf{n}^{\prime}}(m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}^{\prime}}$ $\displaystyle\nu(\beta)$ $\displaystyle=$ $\displaystyle\langle\nu(\hat{\beta})\rangle_{\mathbf{n}^{\prime}}(m\beta_{1}\oplus\cdots\oplus m\beta_{r})_{\mathbf{n}^{\prime}}.$ In this case, $\nu(\hat{\alpha}),\nu(\hat{\beta})\in Z(\nu(\mu_{m,r}))$. _Claim 3._ $\hat{\alpha}$ and $\hat{\beta}$ are conjugate in $B_{mr+1}$. ###### Proof of Claim 3. If $\alpha$ and $\beta$ are conjugate in $B_{n}$, then the $\mathbf{n}$-exterior braids $\hat{\alpha}$ and $\hat{\beta}$ are conjugate in $B_{mr+1}$ by Lemma 2.10. Suppose that $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$. If $d_{0}\geq 1$, then $\hat{\alpha}$ and $\hat{\beta}$ are conjugate in $B_{mr+1}$ by Lemma 2.10 because $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\alpha))=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\beta))=\mathcal{C}_{\mathbf{n}^{\prime}}$ and the $\mathbf{n}^{\prime}$-exterior braids of $\nu(\alpha)$ and $\nu(\beta)$ are $\hat{\alpha}$ and $\hat{\beta}$. Similarly, if $d_{0}=0$, then $\nu(\hat{\alpha})$ and $\nu(\hat{\beta})$ are conjugate in $B_{mr}$, hence $\hat{\alpha}$ and $\hat{\beta}$ are conjugate in $Z(\mu_{m,r})$ by the induction hypothesis. ∎ Since $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)=\mathcal{C}_{\mathbf{n}}$ and $\operatorname{Ext}_{\mathbf{n}}(\alpha)=\hat{\alpha}$, the $\mathbf{n}$-exterior braid $\hat{\alpha}$ is either periodic or pseudo- Anosov. Case 1. $\hat{\alpha}$ is periodic. Let $k=mr+1$. Since $\hat{\alpha}$ and $\hat{\beta}$ are pure and periodic, they are central. Because they are conjugate by the above claim, $\hat{\alpha}=\hat{\beta}=\Delta_{(k)}^{2u}$ for some integer $u$. _Claim 4._ $\alpha_{0}=\beta_{0}$ and there is an $r$-permutation $\theta$ such that $\beta_{i}=\alpha_{\theta(i)}$ for $1\leq i\leq r$. ###### Proof of Claim 4. If $\alpha$ and $\beta$ are conjugate in $B_{n}$, then the following equality holds between multisets consisting of conjugacy classes by J. González-Meneses [Gon03, Proposition 3.2]: $\Bigl{\\{}[\alpha_{0}],\underbrace{[\alpha_{1}],\ldots,[\alpha_{1}]}_{m},\ldots,\underbrace{[\alpha_{r}],\ldots,[\alpha_{r}]}_{m}\Bigr{\\}}=\Bigl{\\{}[\beta_{0}],\underbrace{[\beta_{1}],\ldots,[\beta_{1}]}_{m},\ldots,\underbrace{[\beta_{r}],\ldots,[\beta_{r}]}_{m}\Bigr{\\}}.$ Since $m\geq 2$, $[\alpha_{0}]=[\beta_{0}]$ and there is an $r$-permutation $\theta$ such that $[\beta_{i}]=[\alpha_{\theta(i)}]$ for $1\leq i\leq r$. By the induction hypothesis, $\alpha_{0}$ and $\beta_{0}$ are conjugate in $Z(\mu_{m,d_{0}})$. If $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$ and $d_{0}\geq 1$, then the following equality holds between multisets consisting of conjugacy classes: $\Bigl{\\{}[\nu(\alpha_{0})],\underbrace{[\alpha_{1}],\ldots,[\alpha_{1}]}_{m},\ldots,\underbrace{[\alpha_{r}],\ldots,[\alpha_{r}]}_{m}\Bigr{\\}}=\Bigl{\\{}[\nu(\beta_{0})],\underbrace{[\beta_{1}],\ldots,[\beta_{1}]}_{m},\ldots,\underbrace{[\beta_{r}],\ldots,[\beta_{r}]}_{m}\Bigr{\\}}.$ Since $m\geq 2$, $[\nu(\alpha_{0})]=[\nu(\beta_{0})]$ and there is an $r$-permutation $\theta$ such that $[\beta_{i}]=[\alpha_{\theta(i)}]$ for $1\leq i\leq r$. By the induction hypothesis, $\alpha_{0}$ and $\beta_{0}$ are conjugate in $Z(\mu_{m,d_{0}})$. If $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$ and $d_{0}=0$, then the following equality holds between multisets consisting of conjugacy classes: $\Bigl{\\{}\underbrace{[\alpha_{1}],\ldots,[\alpha_{1}]}_{m},\ldots,\underbrace{[\alpha_{r}],\ldots,[\alpha_{r}]}_{m}\Bigr{\\}}=\Bigl{\\{}\underbrace{[\beta_{1}],\ldots,[\beta_{1}]}_{m},\ldots,\underbrace{[\beta_{r}],\ldots,[\beta_{r}]}_{m}\Bigr{\\}}.$ Then there is an $r$-permutation $\theta$ such that $[\beta_{i}]=[\alpha_{\theta(i)}]$ for $1\leq i\leq r$. In this case, $\alpha_{0}$ and $\beta_{0}$ are the unique braid with one strand. From the above three cases, we can see that $\alpha_{0}$ is conjugate to $\beta_{0}$ in $Z(\mu_{m,d_{0}})$ and there is an $r$-permutation $\theta$ such that $\beta_{i}$ is conjugate to $\alpha_{\theta(i)}$ for $1\leq i\leq r$. Because we have assumed that each $\mathbf{n}$-interior braid of $\alpha$ and $\beta$ is the conjugacy representative with respect to $(m,d_{0})$, one has $\alpha_{0}=\beta_{0}$ and $\beta_{i}=\alpha_{\theta(i)}$ for $1\leq i\leq r$. ∎ In the above claim, $\theta$ must satisfy $d_{i}=d_{\theta(i)}$ for all $1\leq i\leq r$. By the claim, we have $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle\langle\Delta_{(k)}^{2u}\rangle_{\mathbf{n}}(\alpha_{0}\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}},$ $\displaystyle\beta$ $\displaystyle=$ $\displaystyle\langle\Delta_{(k)}^{2u}\rangle_{\mathbf{n}}(\alpha_{0}\oplus m\alpha_{\theta(1)}\oplus\cdots\oplus m\alpha_{\theta(r)})_{\mathbf{n}}.$ By Lemma 4.9, there exists $\hat{\zeta}\in Z(\mu_{m,r})$ such that the induced permutation of $\hat{\zeta}$ fixes $x_{0}$ and sends $x_{i,j}$ to $x_{\theta(i),j}$ for $1\leq i\leq r$ and $1\leq j\leq m$ under the notation of the lemma. Let $\zeta=\langle\hat{\zeta}\rangle_{\mathbf{n}}$. Then $\zeta\ast\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$ because $d_{i}=d_{\theta(i)}$ for all $1\leq i\leq r$, hence $\zeta\in Z(\mu_{m,\mathbf{d}})$ by Lemma 4.11. On the other hand, $\displaystyle\zeta\beta\zeta^{-1}$ $\displaystyle=$ $\displaystyle\langle\hat{\zeta}\rangle_{\mathbf{n}}\langle\Delta_{(k)}^{2u}\rangle_{\mathbf{n}}(\alpha_{0}\oplus m\alpha_{\theta(1)}\oplus\cdots\oplus m\alpha_{\theta(r)})_{\mathbf{n}}\langle\hat{\zeta}\rangle_{\mathbf{n}}^{-1}$ $\displaystyle=$ $\displaystyle\langle\hat{\zeta}\rangle_{\mathbf{n}}\langle\Delta_{(k)}^{2u}\rangle_{\mathbf{n}}\langle\hat{\zeta}\rangle_{\mathbf{n}}^{-1}(\alpha_{0}\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}}$ $\displaystyle=$ $\displaystyle\langle\Delta_{(k)}^{2u}\rangle_{\mathbf{n}}(\alpha_{0}\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}}=\alpha.$ Therefore $\alpha$ and $\beta$ are conjugate in $Z(\mu_{m,\mathbf{d}})$. Case 2. $\hat{\alpha}$ is pseudo-Anosov. _Claim 5._ The braids $\alpha$ and $\beta$ are conjugate in $B_{n}$. ###### Proof of Claim 5. If $\alpha$ and $\beta$ are conjugate in $B_{n}$, there is nothing to prove. Therefore, assume that $\nu(\alpha)$ and $\nu(\beta)$ are conjugate in $B_{n-1}$. Let $\gamma$ be an $(n-1)$-braid with $\nu(\beta)=\gamma^{-1}\nu(\alpha)\gamma$. Since $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\alpha))=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\nu(\beta))=\mathcal{C}_{\mathbf{n}^{\prime}}$, one has $\gamma*\mathcal{C}_{\mathbf{n}^{\prime}}=\mathcal{C}_{\mathbf{n}^{\prime}}$ and $\operatorname{Ext}_{\mathbf{n}^{\prime}}(\nu(\beta))=\operatorname{Ext}_{\mathbf{n}^{\prime}}(\gamma)^{-1}\operatorname{Ext}_{\mathbf{n}^{\prime}}(\nu(\alpha))\operatorname{Ext}_{\mathbf{n}^{\prime}}(\gamma)$ by Lemma 2.10. First, suppose that $d_{0}\geq 1$. Because $\gamma*\mathcal{C}_{\mathbf{n}^{\prime}}=\mathcal{C}_{\mathbf{n}^{\prime}}$, $\gamma$ is of the form $\gamma=\langle\hat{\gamma}\rangle_{\mathbf{n}^{\prime}}(\gamma_{0}\oplus\underbrace{\gamma_{1,1}\oplus\cdots\oplus\gamma_{1,m}}_{m}\oplus\cdots\oplus\underbrace{\gamma_{r,1}\oplus\cdots\oplus\gamma_{r,m}}_{m})_{\mathbf{n}^{\prime}}.$ In this case, $\hat{\beta}=\hat{\gamma}^{-1}\hat{\alpha}\hat{\gamma}$. Because $\hat{\alpha}$ and $\hat{\beta}$ are pseudo-Anosov braids in $Z(\mu_{m,r})$, one has $\hat{\gamma}\in Z(\mu_{m,r})$ by Proposition 5.1. In particular, $\hat{\gamma}$ is 1-pure by Lemma 4.9. Since $\nu(\beta)=\gamma^{-1}\nu(\alpha)\gamma$ and since $\hat{\alpha}$, $\hat{\beta}$ and $\hat{\gamma}$ are all 1-pure, we have $\nu(\beta_{0})=\gamma_{0}^{-1}\nu(\alpha_{0})\gamma_{0}$. By the induction hypothesis, there exists $\gamma_{0}^{\prime}\in Z(\mu_{m,d_{0}})$ such that $\beta_{0}=\gamma_{0}^{\prime-1}\alpha_{0}\gamma_{0}^{\prime}$. Let $\gamma^{\prime}$ be the $n$-braid obtained from $\gamma$ by replacing $\mathbf{n}^{\prime}$ with $\mathbf{n}$ and $\gamma_{0}$ with $\gamma_{0}^{\prime}$, that is, $\gamma^{\prime}=\langle\hat{\gamma}\rangle_{\mathbf{n}}(\gamma_{0}^{\prime}\oplus\underbrace{\gamma_{1,1}\oplus\cdots\oplus\gamma_{1,m}}_{m}\oplus\cdots\oplus\underbrace{\gamma_{r,1}\oplus\cdots\oplus\gamma_{r,m}}_{m})_{\mathbf{n}}.$ We can easily show $\beta=\gamma^{\prime-1}\alpha\gamma^{\prime}$ by using the fact that $\nu(\beta)=\gamma^{-1}\nu(\alpha)\gamma$, $\hat{\gamma}$ is 1-pure, and $\beta_{0}=\gamma_{0}^{\prime-1}\alpha_{0}\gamma_{0}^{\prime}$. Now, suppose that $d_{0}=0$. Because $\gamma*\mathcal{C}_{\mathbf{n}^{\prime}}=\mathcal{C}_{\mathbf{n}^{\prime}}$, $\gamma$ is of the form $\gamma=\langle\hat{\gamma}\rangle_{\mathbf{n}^{\prime}}(\underbrace{\gamma_{1,1}\oplus\cdots\oplus\gamma_{1,m}}_{m}\oplus\cdots\oplus\underbrace{\gamma_{r,1}\oplus\cdots\oplus\gamma_{r,m}}_{m})_{\mathbf{n}^{\prime}}.$ In this case, $\nu(\hat{\beta})=\hat{\gamma}^{-1}\nu(\hat{\alpha})\hat{\gamma}$. Since $\hat{\alpha}$ and $\hat{\beta}$ are pseudo-Anosov braids in $Z(\mu_{m,r})$, $\nu(\hat{\alpha})$ and $\nu(\hat{\beta})$ are pseudo-Anosov braids in $Z(\nu(\mu_{m,r}))$ by Corollary 4.7. Thus $\hat{\gamma}\in Z(\nu(\mu_{m,r}))$ by Proposition 5.1. Because $\nu:Z(\mu_{m,r})\to Z(\nu(\mu_{m,r}))$ is an isomorphism, there exists $\hat{\gamma}^{\prime}\in Z(\mu_{m,r})$ with $\nu(\hat{\gamma}^{\prime})=\hat{\gamma}$. Then $\hat{\beta}=\hat{\gamma}^{\prime-1}\hat{\alpha}\hat{\gamma}^{\prime}$. In addition, $\hat{\gamma}^{\prime}$ is 1-pure by Lemma 4.9. Since $d_{0}=0$, both $\alpha_{0}$ and $\beta_{0}$ are the unique braid with one strand. Let $\gamma^{\prime}$ be the $n$-braid obtained from $\gamma$ by replacing $\mathbf{n}^{\prime}$ with $\mathbf{n}$ and $\hat{\gamma}$ with $\hat{\gamma}^{\prime}$ and by inserting the trivial 1-braid as follows: $\gamma^{\prime}=\langle\hat{\gamma}^{\prime}\rangle_{\mathbf{n}}(1\oplus\underbrace{\gamma_{1,1}\oplus\cdots\oplus\gamma_{1,m}}_{m}\oplus\cdots\oplus\underbrace{\gamma_{r,1}\oplus\cdots\oplus\gamma_{r,m}}_{m})_{\mathbf{n}}.$ We can easily show $\beta=\gamma^{\prime-1}\alpha\gamma^{\prime}$ by using the fact that $\nu(\beta)=\gamma^{-1}\nu(\alpha)\gamma$, $\hat{\beta}=\hat{\gamma}^{\prime-1}\hat{\alpha}\hat{\gamma}^{\prime}$, and $\hat{\gamma}^{\prime}$ is 1-pure. ∎ Recall that $\alpha$ and $\beta$ are of the form $\displaystyle\alpha$ $\displaystyle=$ $\displaystyle\langle\hat{\alpha}\rangle_{\mathbf{n}}(\alpha_{0}\oplus m\alpha_{1}\oplus\cdots\oplus m\alpha_{r})_{\mathbf{n}},$ $\displaystyle\beta$ $\displaystyle=$ $\displaystyle\langle\hat{\beta}\rangle_{\mathbf{n}}(\beta_{0}\oplus m\beta_{1}\oplus\cdots\oplus m\beta_{r})_{\mathbf{n}},$ where $\hat{\alpha},\hat{\beta}\in Z(\mu_{m,r})$, $\alpha_{0},\beta_{0}\in Z(\mu_{m,d_{0}})$, $\alpha_{i}=\lambda(\alpha_{i})$ and $\beta_{i}=\lambda(\beta_{i})$ for $0\leq i\leq r$. Because $\alpha$ and $\beta$ are conjugate in $B_{n}$ by the above claim, there is an $n$-braid $\gamma$ such that $\beta=\gamma\alpha\gamma^{-1}.$ Since $\operatorname{\mathcal{R}}_{\operatorname{ext}}(\alpha)=\operatorname{\mathcal{R}}_{\operatorname{ext}}(\beta)=\mathcal{C}_{\mathbf{n}}$, we have $\gamma*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$, hence $\gamma$ is of the form $\gamma=\langle\hat{\gamma}\rangle_{\mathbf{n}}(\gamma_{0}\oplus\underbrace{\gamma_{1,1}\oplus\cdots\oplus\gamma_{1,m}}_{m}\oplus\cdots\oplus\underbrace{\gamma_{r,1}\oplus\cdots\oplus\gamma_{r,m}}_{m})_{\mathbf{n}}.$ Since $\hat{\alpha}$ and $\hat{\beta}$ are pseudo-Anosov braids in $Z(\mu_{m,r})$ and $\hat{\beta}=\operatorname{Ext}_{\mathbf{n}}(\beta)=\operatorname{Ext}_{\mathbf{n}}(\gamma\alpha\gamma^{-1})=\hat{\gamma}\hat{\alpha}\hat{\gamma}^{-1}$, we have $\hat{\gamma}\in Z(\mu_{m,r})$ by Proposition 5.1. Let $\theta$ be the $(mr+1)$-permutation induced by $\hat{\gamma}$. By Lemma 4.9, $\theta(x_{0})=x_{0}$ and there exist integers $1\leq k_{i}\leq r$ and $0\leq\ell_{i}<m$ such that $\theta(x_{i,j})=x_{k_{i},j-\ell_{i}}$ for $1\leq i\leq r$ and $1\leq j\leq m$. Define $\chi=\langle\hat{\gamma}\rangle_{\mathbf{n}}(1\oplus{\textstyle\bigoplus_{i=1}^{r}}(\underbrace{\Delta_{(d_{i})}^{2}\oplus\cdots\oplus\Delta_{(d_{i})}^{2}}_{\ell_{i}}\oplus\underbrace{1\oplus\cdots\oplus 1}_{m-\ell_{i}}))_{\mathbf{n}}.$ Since $\chi\ast\mathcal{C}_{\mathbf{n}}=\langle\hat{\gamma}\rangle_{\mathbf{n}}*\mathcal{C}_{\mathbf{n}}=\gamma*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$, one has $d_{i}=d_{k_{i}}$ for $1\leq i\leq r$. Thus $\chi\in Z(\mu_{m,\mathbf{d}})$ by Lemma 4.11. Let $\beta^{\prime}=\chi^{-1}\beta\chi.$ Since $\beta,\chi\in Z(\mu_{m,\mathbf{d}})$ with $\beta*\mathcal{C}_{\mathbf{n}}=\chi*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$, one has $\beta^{\prime}\in Z(\mu_{m,\mathbf{d}})$ with $\beta^{\prime}*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$. Since $\operatorname{Ext}_{\mathbf{n}}(\beta^{\prime})=\operatorname{Ext}_{\mathbf{n}}(\chi^{-1}\beta\chi)=\hat{\gamma}^{-1}\hat{\beta}\hat{\gamma}=\hat{\alpha}$ is pure, $\beta^{\prime}$ is of the following form by Corollary 4.12: $\beta^{\prime}=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\beta_{0}^{\prime}\oplus m\beta_{1}^{\prime}\oplus\cdots\oplus m\beta_{r}^{\prime})_{\mathbf{n}},\qquad\beta_{0}^{\prime}\in Z(\mu_{m,d_{0}}).$ Since both $\operatorname{Ext}_{\mathbf{n}}(\beta)$ and $\operatorname{Ext}_{\mathbf{n}}(\chi)$ are 1-pure, one has $\beta_{0}^{\prime}=\beta_{0}$, hence $\beta_{0}^{\prime}=\beta_{0}=\lambda(\beta_{0})=\lambda(\beta_{0}^{\prime})$. By Lemma 4.16, there exists an $\mathbf{n}$-split braid $\chi^{\prime}$ in $Z(\mu_{m,\mathbf{d}})$ such that (1) $\beta^{\prime\prime}=\chi^{\prime-1}\beta^{\prime}\chi^{\prime}=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\beta_{0}^{\prime\prime}\oplus m\beta_{1}^{\prime\prime}\oplus\cdots\oplus m\beta_{r}^{\prime\prime})_{\mathbf{n}},$ where $\beta_{i}^{\prime\prime}$’s are conjugacy representatives, that is, $\beta_{i}^{\prime\prime}=\lambda(\beta_{i}^{\prime\prime})$ for $0\leq i\leq r$. Now we will show that $\beta^{\prime\prime}=\alpha$. Let $\chi^{\prime\prime}=\gamma^{-1}\chi$. Then $\chi^{\prime\prime}*\mathcal{C}_{\mathbf{n}}=\mathcal{C}_{\mathbf{n}}$. Because $\operatorname{Ext}_{\mathbf{n}}(\chi^{\prime\prime})=\operatorname{Ext}_{\mathbf{n}}(\gamma)^{-1}\operatorname{Ext}_{\mathbf{n}}(\chi)=\hat{\gamma}^{-1}\hat{\gamma}=1$, $\chi^{\prime\prime}$ is an $\mathbf{n}$-split braid. Note that $\beta^{\prime\prime}=\chi^{\prime-1}\beta^{\prime}\chi^{\prime}=\chi^{\prime-1}\chi^{-1}\beta\chi\chi^{\prime}=\chi^{\prime-1}\chi^{-1}\gamma\alpha\gamma^{-1}\chi\chi^{\prime}=(\chi^{\prime\prime}\chi^{\prime})^{-1}\alpha(\chi^{\prime\prime}\chi^{\prime}).$ Because both $\chi^{\prime}$ and $\chi^{\prime\prime}$ are $\mathbf{n}$-split braids, so is $\chi^{\prime\prime}\chi^{\prime}$. Therefore $\chi^{\prime\prime}\chi^{\prime}$ is of the form $\chi^{\prime\prime}\chi^{\prime}=\langle 1\rangle_{\mathbf{n}}(\chi_{0}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\chi_{i,1}\oplus\cdots\oplus\chi_{i,m}))_{\mathbf{n}}.$ Because the $\mathbf{n}$-exterior braids of $\alpha$ and $\chi^{\prime\prime}\chi^{\prime}$ are pure braids, the conjugation $(\chi^{\prime\prime}\chi^{\prime})^{-1}\alpha(\chi^{\prime\prime}\chi^{\prime})$ can be computed component-wise: (2) $(\chi^{\prime\prime}\chi^{\prime})^{-1}\alpha(\chi^{\prime\prime}\chi^{\prime})=\langle\hat{\alpha}\rangle_{\mathbf{n}}(\chi_{0}^{-1}\alpha_{0}\chi_{0}\oplus{\textstyle\bigoplus_{i=1}^{r}}(\chi_{i,1}^{-1}\alpha_{i}\chi_{i,1}\oplus\cdots\oplus\chi_{i,m}^{-1}\alpha_{i}\chi_{i,m}))_{\mathbf{n}}.$ Comparing $(\ref{eq:1})$ and $(\ref{eq:2})$, we can see that $\alpha_{i}$ and $\beta_{i}^{\prime\prime}$ are conjugate for $0\leq i\leq r$. Because $\alpha_{i}$ and $\beta_{i}^{\prime\prime}$ are conjugacy representatives with respect to $(m,d_{0})$, we have $\alpha_{i}=\beta_{i}^{\prime\prime}$ for $0\leq i\leq r$, hence $\alpha=\beta^{\prime\prime}$. Because $\alpha=\beta^{\prime\prime}=(\chi\chi^{\prime})^{-1}\beta(\chi\chi^{\prime})$ and $\chi,\chi^{\prime}\in Z(\mu_{m,\mathbf{d}})$, $\alpha$ and $\beta$ are conjugate in $Z(\mu_{m,\mathbf{d}})$. ∎ We remark that in the above proof we can show by a straightforward computation that $\beta_{i}^{\prime}=\beta_{k_{i}}$ for $i=1,\ldots,r$. In particular, the $\mathbf{n}$-interior braids $\beta_{0}^{\prime},\ldots,\beta_{r}^{\prime}$ are conjugacy representatives with respect to $(m,d_{0})$, hence we can take $\chi^{\prime}=1$ and $\beta^{\prime\prime}=\beta^{\prime}$. This means that $\alpha=\chi^{-1}\beta\chi$, where $\chi$ depends only on $\mathbf{n}$ and $\hat{\gamma}$. However we introduced $\chi^{\prime}$ and $\beta^{\prime\prime}$ because it is much simpler than showing $\beta_{i}^{\prime}=\beta_{k_{i}}$. ### Acknowledgements The authors are grateful to the anonymous referee for valuable comments and suggestions. This work was done partially while the authors were visiting Department of Mathematical and Computing Sciences, Tokyo Institute of Technology in February 2008. We thank Sadayoshi Kojima and Eiko Kin for their hospitality and interest in this work. This work was supported by the Korea Research Foundation Grant funded by the Korean Government (MOEHRD, Basic Research Promotion Fund) (KRF-2008-331-C00039). ## References * [All02] D. Allcock, _Braid pictures for Artin groups_ , Trans. Amer. Math. Soc. 354 (2002), 3455–3474. * [Art25] E. Artin, _Theorie der Zöpfe_ , Hamburg Abh. 4 (1925), 47–72. * [BM07] R.W. Bell and D. Margalit, _Injections of Artin groups_ , Comment. Math. Helv. 82 (2007), 725–751. * [BDM02] D. Bessis, F. Digne and J. Michel, _Springer theory in braid groups and the Birman-Ko-Lee monoid_ , Pacific J. Math. 205 (2002), 287–309. * [BGG06] J.S. Birman, V. Gebhardt and J. González-Meneses, _Conjugacy in Garside Groups III: Periodic braids_ , J. Algebra 316 (2007), 246–276. * [BLM83] J.S. Birman, A. Lubotzky and J. McCarthy, _Abelian and solvable subgroups of maping class groups_ , Duke Math. J. 50 (1983), 1107–1120. * [Bro19] L.E.J. Brouwer, _Über die periodischen Transformationen der Kugel_ , Math. Ann. 80 (1919), 39–41. * [CC05] R. Charney and J. Crisp, _Automorphism groups of some affine and finite type Artin groups_ , Math. Res. Lett. 12 (2005), 321–333. * [CK94] A. Constantin and B. Kolev, _The theorem of Kerékjártó on periodic homeomorphisms of the disc and the sphere_ , Enseign. Math. (2) 40 (1994), 193–204. * [Cri99] J. Crisp, _Injective maps between Artin groups_ , Geometric group theory down under (Canberra, 1996), 119–137, de Gruyter, Berlin, 1999. * [CP05] J. Crisp and L. Paris, _Artin groups of type $B$ and $D$_, Adv. Geo. 5 (2005), 607–636. * [Deh98] P. Dehornoy, _Gaussian groups are torsion free_ , J. Algebra 210 (1998), 291–297. * [Eil34] S. Eilenberg, _Sur les transformations périodiques de la surface de la sphére_ , Fund. Math. 22 (1934), 28–41. * [Gon03] J. González-Meneses, _The $n$th root of a braid is unique up to conjugacy_, Algebr. Geom. Topol. 3 (2003), 1103–1118. * [GW04] J. González-Meneses and B. Wiest, _On the structure of the centralizer of a braid_ , Ann. Sci. Ecole Norm. Sup. (4) 37 (2004), 729–757. * [Iva92] N.V. Ivanov, _Subgroups of Teichmüller modular group_ , Transl. Math. Monogr. 115, AMS, 1992. * [Ker19] B. de Kerékjártó, _Über die periodischen Transformationen der Kreisscheibe und der Kugelfläche_ , Math. Annalen 80 (1919), 3–7. * [LL07] E.-K. Lee and S.-J. Lee, _Uniqueness of roots up to conjugacy for some affine and finite type Artin groups_ , arXiv:0711.0091, to appear in Math. Z. * [LL08] E.-K. Lee and S.-J. Lee, _A Garside-theoretic approach to the reducibility problem in braid groups_ , J. Algebra 320 (2008), 783–820. * [Thu88] W. Thurston, _On the isotopy and geometry of diffeomorphisms of surfaces_ , Bull. Amer. Math. Soc. 19 (1988), 109–140.

arxiv-papers

2024-09-04T02:48:53.792681

0802.2338

11footnotetext: This work is partially supported by the Emmy Noether Research Institute for Mathematics and the Minerva Foundation of Germany and the Israel Science Foundation grant # 8008/02-3 (Excellency Center “Group Theoretic Methods in the Study of Algebraic Varieties”). # On non Fundamental Group Equivalent Surfaces Michael Friedman Michael Friedman, Department of Mathematics, Bar-Ilan University, 52900 Ramat Gan, Israel fridmam@macs.biu.ac.il and Mina Teicher1 Mina Teicher, Department of Mathematics, Bar-Ilan University, 52900 Ramat Gan, Israel teicher@macs.biu.ac.il ###### Abstract. In this paper we present an example of two polarized K3 surfaces which are not Fundamental Group Equivalent (their fundamental groups of the complement of the branch curves are not isomorphic; denoted by FGE) but the fundamental groups of their related Galois covers are isomorphic. For each surface, we consider a generic projection to $\mathbb{CP}^{2}$ and a degenerations of the surface into a union of planes - the “pillow” degeneration for the non-prime surface and the “magician” degeneration for the prime surface. We compute the Braid Monodromy Factorization (BMF) of the branch curve of each projected surface, using the related degenerations. By these factorizations, we compute the above fundamental groups. It is known that the two surfaces are not in the same component of the Hilbert scheme of linearly embedded K3 surfaces. Here we prove that furthermore they are not FGE equivalent, and thus they are not of the same Braid Monodromy Type (BMT) (which implies that they are not a projective deformation of each other). ###### Key words and phrases: fundamental group, generic projection, curves and singularities, branch curve. ###### 2000 Mathematics Subject Classification: 14F35, 14H20, 14H30, 14J28, 14Q05, 14Q10, 20F36, 57M12 ## 1\. Introduction Given $X\subset\mathbb{CP}^{n}$ a smooth algebraic surface of degree $m$, one can obtain information on $X$ by considering it as a branched cover of $\mathbb{CP}^{2}$. It is well–known that for $X\longrightarrow\mathbb{CP}^{2}$ a generic projection, the branch locus is a plane curve $\bar{S}\subset\mathbb{CP}^{2}$ which is, in general, singular, and its singularities are nodes and cusps. Let $S\subset\mathbb{C}^{2}\subset\mathbb{CP}^{2}$ be a generic affine portion of $\bar{S}$. It was proven in [8] that if the Braid Monodromy Factorizations (BMF) of the branch loci of two surfaces $X_{1}$ and $X_{2}$ are Hurwitz-equivalent, then the surfaces are diffeomorphic. Moreover, if the factorizations are not Hurwitz-equivalent, then $X_{1}$ and $X_{2}$ are not projectively deformation equivalent. Therefore, the BMT invariant (the equivalence class of a BMF) is really in the “middle”, i.e., between the diffeomorphism equivalence and the projectively deformation equivalence. We need to find an algorithm that decides whether two BMFs are equivalent. In general, it was shown in [9] that there is no finite algorithm which determines whether two positive factorizations are Hurwitz- equivalent. However, [9] did not examine the particular case of the BMFs. Therefore, we have to extract the information contained in the braid monodromy factorization via the introduction of more manageable (but less powerful) invariants. Two discrete invariants are induced from the BMF of the branch curve – $S$: the fundamental group of the complement of the branch curve (see [2],[7],[11],[17]) and its subquotient: the fundamental group of the Galois Cover of $X$ (see [10],[12],[13]). We say that two surfaces are Fundamental Group Equivalent (FGE) if their fundamental groups of the complement of the branch curve are isomorphic. In this article we present two surfaces, which are embeddings of a K3 surface with respect to two different linear systems; therefore they are diffeomorphic. Due to the nature of the particular linear systems, these embedded surfaces are not projectively deformation equivalent. It is also known that any two K3 surfaces can be abstractly deformed one into the other. Thus one can raise the questions: Are the surfaces FGE? Are the fundamental groups of the corresponding Galois covers isomorphic? Here we prove that although the latter groups are isomorphic, the surfaces are not FGE. Therefore, these surfaces are also not BMT–equivalent, which means that the surfaces are not in the same component of the Hilbert scheme of linearly embedded K3 surfaces. Acknowledgments: We are very grateful to Ciro Ciliberto, for pointing out the degeneration used for the “magician” K3 surface and for other helpful discussions. We also wish to thank Meirav Amram for her fruitful remarks. ## 2\. Preliminaries: The K3 surfaces and the BMT invariant In this section we recall the main definitions and constructions regarding the two embeddings of the $K3$ surface, and the braid monodromy factorization (=BMF) related to a (branch) curve. We begin with the introduction of the two embeddings of a $K3$ surface. ### 2.1. Two embeddings of a $K3$ surface Recall that the surfaces with Kodaira dimension which equals to 0 ,that are simply connected, have in fact trivial canonical bundle, and are called $K3$ surfaces. The invariants for such surfaces are $p_{g}=1$, $q=0$, $e=24$. The moduli space of all $K3$ surfaces is $20$-dimensional. Most $K3$ surfaces are not algebraic; the algebraic ones are classified by an infinite collection (depending on an integer $g\geq 2$) of $19$-dimensional moduli spaces. The general member of the family has a rank one Picard group, generated by an ample class $H$ with $H^{2}=2g-2$; the general member of the linear system $|H|$ is a smooth curve of genus $g$, and this linear system maps the $K3$ surface to $\mathbb{P}^{g}$ as a surface of degree $2g-2$. For example, a $K3$ surface is a smooth quartic surface in $\mathbb{P}^{3}$. The quartic surfaces in $\mathbb{P}^{3}$ form the family with $g=3$. The integer $g$ is called the _genus_ of the family. The first embedded surface is a $K3$ surface of genus 9, embedded in $\mathbb{CP}^{9}$ by the pillow (2,2)-pillow degeneration (see [6] for details). The resulting embedding can be degenerated into a union of 16 planes, such that the whole degenerated object would “resemble a pillow” (see figure 1 for clarification). We denote by $X_{1}$ the embedded $K3$ surface, and by $(X_{1})_{0}$ the degenerated surface (see [17] for an explicit definition of a degeneration). Figure 1: $(X_{1})_{0}$ – the (2,2)-pillow degeneration: every triangle denotes a plane Figure 2: A 2-dimensional figure of $(X_{1})_{0}$: the boundaries are identified (top to top, bottom to bottom, side to side) The degeneration process has a “local inverse” – the regeneration process (see an explanation in the following subsection), and for it we need to fix a numeration of vertices (and the lines; see [1] for details). This is done in the following way (see figure 3): Figure 3: The numeration of the singular points of $(X_{1})_{0}$ The 16 planes meet each other along a total of 24 lines, each joining 2 of the 10 coordinate points. We numerate the lines as follows: if $L$ has endpoints $a<b$ and $M$ has endpoints $c<d$, then $L<M$ if $b<d$ or $b=d$ and $a<c$. This gives a total ordering of the lines, which we interpret as a numbering from 1 to 24, as shown in figure 4. Figure 4: The numeration of the intersection lines of $(X_{1})_{0}$ Under a general projection $\pi:(X_{1})_{0}\rightarrow\mathbb{CP}^{2}$, each of the 16 planes is mapped isomorphically to $\mathbb{CP}^{2}$. The ramification locus $R_{1}$ of $\pi$ is a local isomorphism. Here $R_{1}$ is exactly the 24 lines. Let $(S_{1})_{0}=\pi(R_{1})$ be the degenerated branch curve. It is a line arrangement, composed of the image of the 24 lines. The second embedded surface is also an embedded $K3$ surface of genus 9 in $\mathbb{CP}^{9}$. We call this surface the “magician” surface, since its degeneration “resembles” a magician’s hat. The surface and its degeneration into a union of 16 planes are described in [5]. The dual graph of the degenerated surface is presented explicitly in [5, pg. 430] \- and from it we can build the degenerated surface (see figure 5). Figure 5: Every point in the dual graph represents a plane; every plane represents a point Denote by $X_{2}$ this embedded surface, and by $(X_{2})_{0}$ the degenerated surface. We can depict a 2-dimensional graph of $(X_{2})_{0}$, where the boundaries are identified (see figure 6): Figure 6: $(X_{2})_{0}$ \- the boundaries are identified (top to bottom) Once again, we numerate the vertices and then the edges. We note that the extreme edges of the graph $(X_{2})_{0}$ are actually 4–points: singular points in the degenerated surface which are the intersection of four planes. In order to regenerate it (see [19] for the possible degenerations of this point), we need to numerate the vertices in such a way that the number of “entering” and “exiting” lines from these points will be equal. Therefore, we numerate them as vertices 5 and 6. Following the symmetry appearing in the graph, we numerate the other vertices as follows (see figure 7): Figure 7: Numeration of $(X_{2})_{0}$ Note that $(X_{2})_{0}$ also contains 24 intersection lines and 10 singular points. We denote by $(S_{2})_{0}=\pi_{2}(R_{2})$ the degenerated branch curve with respect to a generic projection $\pi_{2}:(X_{2})_{0}\rightarrow\mathbb{CP}^{2}$. Since every two $K3$ surfaces are diffeomorphic, $X_{1}$ and $X_{2}$ are also diffeomorphic. Note that the Hilbert scheme of embedded linearly normal K3 surfaces can be reducible. This is indeed the case here – the Picard group Pic$X_{1}$ is generated by $\frac{1}{2}H$ (where $H$ is the hyperplane class; see [6]) and Pic$X_{2}$ is generated by $H$ (see [5]). Two polarized K3 surfaces are projectively deformation equivalent if and only if there is a diffeomorphism which carries the hyperplane class to the hyperplane class. As indicated above, this is not the case. We show in the following sections that these surfaces are also not BMT–equivalent, and that the fundamental groups of complement of the branch curve can also be used in order to differentiate between irreducible components of the Hilbert scheme. Thus it is a topological invariant that arises in algebro-geometric considerations. ### 2.2. The braid group and the BMF Recall that computing the braid monodromy is the main tool to compute fundamental groups of complements of curves. The reader who is familiar with this subject can skip the following definitions. We begin by defining the braid monodromy associated to a curve. Let $D$ be a closed disk in $\mathbb{R}^{2},$ $K\subset Int(D),$ $K$ finite, $n=\\#K$. Recall that the braid group $B_{n}[D,K]$ can be defined as the group of all equivalent diffeomorphisms $\beta$ of $D$ such that $\beta(K)=K\,,\,\beta|_{\partial D}=\text{Id}\left|{}_{\partial D}\right.$. Definition: $H(\sigma)$, half-twist defined by $\sigma$ Let $a,b\in K,$ and let $\sigma$ be a smooth simple path in $Int(D)$ connecting $a$ with $b$ s.t. $\sigma\cap K=\\{a,b\\}.$ Choose a small regular neighborhood $U$ of $\sigma$ contained in $Int(D),$ s.t. $U\cap K=\\{a,b\\}$. Denote by $H(\sigma)$ the diffeomorphism of $D$ which switches $a$ and $b$ by a counterclockwise $180^{\circ}$ rotation and is the identity on $D\setminus U$ . Thus it defines an element of $B_{n}[D,K],$ called the half-twist defined by $\sigma$ . Denote $[A,B]=ABA^{-1}B^{-1},\,\langle A,B\rangle=ABAB^{-1}A^{-1}B^{-1}$. We recall the Artin presentation of the braid group: ###### Theorem 2.1. $B_{n}$ is generated by the half-twists $H_{i}$ of a frame $H_{i}$ and all the relations between $H_{1},...,H_{n-1}$ follow from: $[H_{i},H_{j}]=1\,\,$ if $|i-j|>1$ $\langle H_{i},H_{j}\rangle=1\,\,\text{if}\,\,\,\,|i-j|=1$. Assume that all of the points of $K$ are on the $X$-axis (when considering $D$ in $\mathbb{R}^{2}$). In this situation, if $a,b\in K$, and $z_{a,b}$ is a path that connects them, then we denote it by $Z_{a,b}=H(z_{a,b})$. If $z_{a,b}$ is a path that goes below the $X$-axis, then we denote it by $\underline{Z}_{a,b}$, or just $Z_{a,b}$. If $z_{a,b}$ is a path that goes above the $x$-axis, then we denote it by $\overline{Z}_{a,b}$. We also denote by $\overset{(c-d)}{\underline{Z}_{a,b}}$ ($\underset{(c-d)}{\bar{Z}_{a,b}}$) the braid induced from a path connecting the points $a$ and $b$ below (resp. above) the $X$-axis, going above (resp. below) it from the point $c$ till point $d$. Definition: The braid monodromy w.r.t. $S,\pi,u$ Let $S$ be a curve, $S\subseteq\mathbb{C}^{2}$ . Let $\pi:S\to\mathbb{C}^{1}$ be defined by $\pi(x,y)=x.$ We denote $\deg\pi$ by $m.$ Let $N=\\{x\in\mathbb{C}^{1}\bigm{|}\\#\pi^{-1}(x)<m\\}.$ Take $u\notin N,$ s.t. $\Re(x)\ll u$ $\forall x\in N.$ Let $\mathbb{C}^{1}_{u}=\\{(u,y)\\}.$ There is a naturally defined homomorphism $\pi_{1}(\mathbb{C}^{1}-N,u)\xrightarrow{\varphi}B_{m}[\mathbb{C}_{u}^{1},\mathbb{C}_{u}^{1}\cap S]$ which is called the braid monodromy w.r.t. $S,\pi,u,$ where $B_{m}$ is the braid group. We sometimes denote $\varphi$ by $\varphi_{u}$. In fact, denoting by $E$, a big disk in $\mathbb{C}^{1}$ s.t. $E\supset N$, we can also take the path in $E\setminus N$ not to be a loop, but just a non-self-intersecting path. This induces a diffeomorphism between the models $(D,K)$ at the two ends of the considered path, where $D$ is a big disk in $\mathbb{C}^{1}_{u}$, and $K=\mathbb{C}_{u}^{1}\cap S\subset D$. Definition: $\underline{\psi_{T},\ \text{Lefschetz diffeomorphism induced by a path}\ T}$ Let $T$ be a path in $E\setminus N$ connecting $x_{0}$ with $x_{1}$, $T:[0,1]\rightarrow E\setminus N$. There exists a continuous family of diffeomorphisms $\psi_{(t)}:D\rightarrow D,\ t\in[0,1],$ such that $\psi_{(0)}=Id$, $\psi_{(t)}(K(x_{0}))=K(T(t))$ for all $t\in[0,1]$, and $\psi_{(t)}(y)=y$ for all $y\in\partial D$. For emphasis we write $\psi_{(t)}:(D,K(x_{0}))\rightarrow(D,K(T(t))$. A Lefschetz diffeomorphism induced by a path $T$ is the diffeomorphism $\psi_{T}=\psi_{(1)}:(D,K(x_{0}))\underset{\sim}{\rightarrow}(D,K(x_{1})).$ Since $\psi_{(t)}\left(K(x_{0})\right)=K(T(t))$ for all $t\in[0,1]$, we have a family of canonical isomorphisms $\psi_{(t)}^{\nu}:B_{p}\left[D,K(x_{0})\right]\underset{\sim}{\rightarrow}B_{p}\left[D,K({T(t)})\right],\ \quad\text{for all}\,\,t\in[0,1].$ We recall Artin’s theorem on the presentation of the Dehn twist of the braid group as a product of braid monodromy elements of a geometric-base (a base of $\pi=\pi(\mathbb{C}^{1}-N,u)$ with certain properties; see [14] for definitions). ###### Theorem 2.2. Let $S$ be a curve transversal to the line in infinity, and $\varphi$ is a braid monodromy of $S,\varphi:\pi\rightarrow B_{m}$. Let $\delta_{i}$ be a geometric (free) base (g-base) of $\pi,$ and $\Delta^{2}$ is the generator of Center($B_{m}$). Then: $\Delta^{2}=\prod\varphi(\delta_{i}).$ This product is also defined as the braid monodromy factorization (BMF) related to a curve $S$. Note that if $x_{1},...,x_{n-1}$ are the generators of $B_{n}$, then we know that $\Delta^{2}=(x_{1}\cdot\ldots\cdot x_{n-1})^{n}$ and thus deg($\Delta^{2}$) = $n(n-1)$. So in order to find out what is the braid monodromy factorization of $\Delta_{p}^{2}$, we have to find out what are $\varphi(\delta_{i}),\,\forall i$. We refer the reader to the definition of a skeleton (see [15]) $\lambda_{x_{j}},x_{j}\in N$, which is a model of a set of paths connecting points in the fiber, s.t. all those points coincide when approaching $A_{j}=$($x_{j},y_{j}$)$\in S$, when we approach this point from the right. To describe this situation in greater detail, for $x_{j}\in N$, let $x_{j}^{\prime}=x_{j}+\alpha$. So the skeleton in $x_{j}$ is defined as a system of paths connecting the points in $K(x_{j}^{\prime})\cap D(A_{j},\varepsilon)$ when $0<\alpha\ll\varepsilon\ll 1$, $D(A_{j},\varepsilon)$ is a disk centered in $A_{j}$ with radius $\varepsilon$. For a given skeleton, we denote by $\Delta\langle\lambda_{x_{j}}\rangle$ the braid which rotates by $180^{\circ}$ counterclockwise a small neighborhood of the given skeleton. Note that if $\lambda_{x_{j}}$ is a single path, then $\Delta\langle\lambda_{x_{j}}\rangle=H(\lambda_{x_{j}})$. We also refer the reader to the definition of $\delta_{x_{0}}$, for $x_{0}\in N$ (see [15]), which describes the Lefschetz diffeomorphism induced by a path going below $x_{0}$, for different types of singular points (tangent, node, branch; for example, when going below a node, a half-twist of the skeleton occurs and when going below a tangent point, a full-twist occurs). We define, for $x_{0}\in N$, the following number: $\varepsilon_{x_{0}}=1,2,4$ when ($x_{0},y_{0}$) is a branch / node / tangent point (respectively). So we have the following statement (see [15, Prop. 1.5]): Let $\gamma_{j}$ be a path below the real line from $x_{j}$ to $u$, s.t. $\ell(\gamma_{j})=\delta_{j}$. So $\varphi_{u}(\delta_{j})=\varphi(\delta_{j})=\Delta\bigg{\langle}(\lambda_{x_{j}})\bigg{(}\prod\limits_{m=j-1}^{1}\delta_{x_{m}}\bigg{)}\bigg{\rangle}^{\varepsilon_{x_{j}}}.$ When denoting $\xi_{x_{j}}=(\lambda_{x_{j}})\bigg{(}\prod\limits_{m=j-1}^{1}\delta_{x_{m}}\bigg{)}$ we get – $\varphi(\delta_{j})=\Delta\langle(\xi_{x_{j}})\rangle^{\varepsilon_{x_{j}}}.$ Note that the last formula gives an algorithm to compute the needed factorization. For a detailed explanation of the braid monodromy, see [14]. We shall now define an equivalence relation on the BMF. Definition: _Hurwitz moves_ : Let $\vec{t}=(t_{1},\ldots,t_{m})\in G^{m}$ . We say that $\vec{s}=(s_{1},\ldots,s_{m})\in G^{m}$ is obtained from $\vec{t}$ by the Hurwitz move $R_{k}$ (or $\vec{t}$ is obtained from $\vec{s}$ by the Hurwitz move $R^{-1}_{k}$) if $s_{i}=t_{i}\quad\text{for}\ i\neq k\,,\,k+1\,,\\\ s_{k}=t_{k}t_{k+1}t^{-1}_{k}\,,\\\ s_{k+1}=t_{k}\,.$ Definition: _Hurwitz move on a factorization_ Let $G$ be a group $t\in G.$ Let $t=t_{1}\cdot\ldots\cdot t_{m}=s_{1}\cdot\ldots\cdot s_{m}$ be two factorized expressions of $t.$ We say that $s_{1}\cdot\ldots\cdot s_{m}$ is obtained from $t_{1}\cdot\ldots\cdot t_{m}$ by a Hurwitz move $R_{k}$ if $(s_{1},\ldots,s_{m})$ is obtained from $(t_{1},\ldots,t_{m})$ by a Hurwitz move $R_{k}$ . Definition: _Hurwitz equivalence of factorization_ Two factorizations are Hurwitz equivalent if they are obtained from each other by a finite sequence of Hurwitz moves. Definition: _Braid monodromy type of curves (BMT)_ Two curves $S_{1}$ and $S_{2}$ are of the same BMT (denoted by $\cong$) if they have related BMF’s that are equivalent. In 1998, the following theorem was proved ([8]) : ###### Theorem 2.3. If $S_{1}\cong S_{2}$, then $S_{1}$ is isotopic to $S_{2}$ (when $S_{1},S_{2}$ are any curves). Thus, an invariant of surfaces can be derived from the BMT of the branch curve of a surface. Definition: _Braid monodromy type of surfaces (BMT)_ The BMT of a projective surface is the BMT of the branch curve of a generic projection of the surface embedded in a projective space by means of a complete linear system. Consequently, the following was proved ([8]): ###### Theorem 2.4. The BMT of a projective surface $X$ determines the diffeomorphism type of $X$. We recall now the regeneration methods. The regeneration methods are actually, locally, the reverse process of the degeneration method. When regenerating a singular configuration consisting of lines and conics, the final stage in the regeneration process involves doubling each line, so that each point of $K$ corresponding to a line labelled $i$ is replaced by a pair of points, labelled $i$ and $i^{\prime}$. The purpose of the regeneration rules is to explain how the braid monodromy behaves when lines are doubled in this manner. We denote by $Z_{i,j}=H(z_{i,j})$ where $z_{i,j}$ is a path connecting points in $K$. The rules are (see [16, pg. 336-337]): 1. (1) First regeneration rule: The regeneration of a branch point of any conic: A factor of the braid monodromy of the form $Z_{i,j}$ is replaced in the regeneration by $Z_{i^{\prime},j}\cdot\overset{(j)}{\underline{Z}}_{i,j^{\prime}}$ 2. (2) Second regeneration rule: The regeneration of a node: A factor of the form $Z^{2}_{ij}$ is replaced by a factorized expression $Z^{2}_{ii^{\prime},j}:=Z^{2}_{i^{\prime}j}\cdot Z^{2}_{ij}$ , $Z^{2}_{i,jj^{\prime}}:=Z^{2}_{ij^{\prime}}\cdot Z^{2}_{ij}$ or by $Z^{2}_{ii^{\prime},jj^{\prime}}:=Z^{2}_{i^{\prime}j^{\prime}}\cdot Z^{2}_{ij^{\prime}}Z^{2}_{i^{\prime}j}\cdot Z^{2}_{ij}$. 3. (3) Third regeneration rule: The regeneration of a tangent point: A factor of the form $Z^{4}_{ij}$ in the braid monodromy factorized expression is replaced by $Z^{3}_{i,jj^{\prime}}:=(Z^{3}_{ij})^{Z_{jj^{\prime}}}\cdot(Z^{3}_{ij})\cdot(Z^{3}_{ij})^{Z^{-1}_{jj^{\prime}}}$. As a result, we get a factorized expression, which, by [8], determines the diffeomorphism type of our surface, and, by [20], determines $\pi_{1}(\mathbb{CP}^{2}-\overline{S})$. This is explained in the following paragraphs. Assume that we have a curve $\bar{S}$ in $\mathbb{CP}^{2}$ and its BMF. Then we can calculate the groups $\pi_{1}(\mathbb{CP}^{2}-\overline{S})$ and $\pi_{1}(\mathbb{C}^{2}-S)$ (where $S=\bar{S}\cap\mathbb{C}^{2}$). Recall that a $g$-base is an ordered free base of $\pi(D\backslash F,v)$, where $D$ is a closed disc, $F$ is a finite set in Int($D$), $v\in\partial D$ which satisfies several conditions; see [14], [15] for the explicit definition. Let $\\{\Gamma_{i}\\}$ be a $g$-base of $G=\pi_{1}(\mathbb{C}_{u}-S,u),$ where $\mathbb{C}_{u}=\mathbb{C}\times u$, and here $S=\mathbb{C}_{u}\cap S$. We cite now the Zariski-Van Kampen Theorem (for cuspidal curves) in order to compute the relations between the generators in $G.$ ###### Theorem 2.5. Zariski-Van Kampen (cuspidal curves version) Let $\overline{S}$ be a cuspidal curve in $\mathbb{CP}^{2}$. Let $S=\mathbb{C}^{2}\cap\overline{S}.$ Let $\varphi$ be a braid monodromy factorization w.r.t. $S$ and $u.$ Let $\varphi=\prod\limits_{j=1}^{p}V_{j}^{\nu_{j}},$ where $V_{j}$ is a half-twist and $\nu_{j}=1,2,3.$ For every $j=1\dots p$, let $A_{j},B_{j}\in\pi_{1}(\mathbb{C}_{u}-S,u)$ be such that $A_{j},B_{j}$ can be extended to a $g$-base of $\pi_{1}(\mathbb{C}_{u}-S,u)$ and $(A_{j})V_{j}=B_{j}.$ Let $\\{\Gamma_{i}\\}$ be a $g$-base of $\pi_{1}(\mathbb{C}_{u}-S,u)$ corresponding to the $\\{A_{i},B_{i}\\}$, where $A_{i},B_{i}$ are expressed in terms of $\Gamma_{i}$. Then $\pi_{1}(\mathbb{C}^{2}-S,u)$ is generated by the images of $\\{\Gamma_{i}\\}$ in $\pi_{1}(\mathbb{C}^{2}-S,u)$ and the only relations are those implied from $\\{V_{j}^{\nu_{j}}\\},$ as follows: $\begin{cases}A_{j}\cdot B_{j}^{-1}&\quad\text{if}\quad\nu_{j}=1\\\ [A_{j},B_{j}]=1&\quad\text{if}\quad\nu_{j}=2\\\ \langle A_{j},B_{j}\rangle=1&\quad\text{if}\quad\nu_{j}=3.\end{cases}$ $\pi_{1}(\mathbb{CP}^{2}-\overline{S},*)$ is generated by $\\{\Gamma_{i}\\}$ with the above relations and one more relation $\prod\limits_{i}\Gamma_{i}=1.$ The following figure illustrates how to find $A_{i},B_{i}$ from the half-twist $V_{i}=H(\sigma)$: Figure 8 So: $A_{V}=\Gamma_{4}^{-1}\Gamma_{6}\Gamma_{4},\,B_{V}=\Gamma_{1}.$ We finish this subsection by recalling the definition of $\tilde{B}_{n}$. Definition: 1. Let $X,Y$ be two half-twists in $B_{n}=B_{n}(D,K)$. We say that $X,Y$ are _transversal_ if they are defined by two simple paths $\xi,\eta$ which intersect transversally in one point different from their ends. 2\. Let $N$ be the normal subgroup of $B_{n}$ generated by conjugates of $[X,Y]$, where $X,Y$ is a transversal pair of half-twists. Define $\tilde{B}_{n}=B_{n}/N$. 3\. Let $Y_{i},\,i=1,..,4$ be four half-twists in $B_{n}$ (resp. $\tilde{B}_{n}$) corresponding to simple paths $\eta_{1},...,\eta_{4}$. Assume that $\eta_{i},\,i=1,...,4$, could be chosen so that they form a quadrangle without self intersections and such that in its interior there are no points of $K$. Then we say that $Y_{1},Y_{2},Y_{3},Y_{4}$ form a _good quadrangle_ in $B_{n}$ (resp. in $\tilde{B}_{n}$). ###### Lemma 2.1. If $y_{1},y_{2},y_{3},y_{4}\in\tilde{B}_{n}$ form a good quadrangle then $y_{1}^{2}y_{3}^{2}=y_{2}^{2}y_{4}^{2}$. ###### Proof. See [11, section 1.1].∎ ## 3\. Computing the BMFs Let $\varphi_{1},\varphi_{2}$ be the BMF of the branch curve of the first (resp. second) K3 surface. Before computing $\varphi_{1},\varphi_{2}$, we need a few notations. Denote the intersection lines on $(X_{i})_{0}$ as $\\{\hat{L}_{i,j}\\}^{24}_{j=1},\linebreak i=1,2$ (recall that $(X_{i})_{0}$ is the degeneration of the K3-surfaces $X_{i},i=1,2)$, and by $\\{\hat{v}_{i,j}\\}^{10}_{j=1},\linebreak i=1,2$ the intersection points of these lines. Take generic projections $\pi_{i}:(X_{i})_{0}\rightarrow\mathbb{C}\mathbb{P}^{2}$, and let $(S_{i})_{0}$ be the branch curve in $\mathbb{C}\mathbb{P}^{2},(\varphi_{i})_{0}$ \- their braid monodromy, and $L_{i,j}=\pi_{i}(\hat{L}_{i,j}),i=1,2,\ j,...,24$. So, $(S_{i})_{0}=\bigcup\limits^{24}_{j=1}L_{i,j},v_{i,j}\doteq\pi_{i}(\widetilde{v}_{i,j}),i=1,2,\ j=1,...,24$ are the singular points of $(S_{i})_{0}$. Let $C_{i}$ be the union of all lines connecting pairs of the $v_{i,j}\in(S_{i})_{0}$. $(S_{i})_{0}$ is a subcurve of $C_{i}$. By [14, Theorem IX], we get a full description of the braid monodromy of $C_{i}:\Delta^{2}_{C_{1}}=\Pi^{1}_{j=10}C_{i,j}\Delta^{2}_{v_{i,j}}(i=1,2)$ with an appropriate description of L.V.C. We use this formula to obtain a description of $(\varphi_{i})_{0}$ by deleting factors that involve lines which do not appear in $(S_{i})_{0}$. Thus, we get $(\varphi_{i})_{0}=\Delta^{2}_{(S_{i})_{0}}=\Pi^{1}_{j=10}\widetilde{C}_{i,j}\widetilde{\Delta}^{2}_{v_{i,j}}$. We describe each factor separately. $\widetilde{C}_{i,j}$ : The factors $\widetilde{C}_{i,j}$ correspond to parasitic intersections; these are intersections created by lines that do not intersect in $\mathbb{C}\mathbb{P}^{9}$ but may intersect in $\mathbb{C}\mathbb{P}^{2}$. By [14] we know that $\widetilde{C}_{i,j}=\prod\limits_{v_{i,j}\in L_{i,t}}\ D_{i,t}$, where $D_{i,t}=\prod\limits_{\stackrel{{\scriptstyle p<t}}{{L_{i,p}\cap L_{i,t}}}=\emptyset}\widetilde{Z}^{2}_{pt}$. For $i=1$, the global BMF, together with the $\widetilde{C}_{1,j}$ is presented in [1, Section 4.1]. For $i=2$, we have (by [14, Thm. X.2.1]): $D_{2,1}=id\quad D_{2,2}=\underline{Z}^{2}_{1,2}\quad D_{2,3}=\underline{Z}^{2}_{2,3}\quad D_{2,4}=\overset{(2)}{\underline{Z}^{2}_{1,4}}$ $D_{2,5}=\bar{Z}^{2}_{2,5}\bar{Z}^{2}_{3,5}\underline{Z}^{2}_{4,5}\quad D_{2,6}=\overset{(2-3)}{\underline{Z}^{2}_{1,6}}\overset{(4)}{\underline{Z}^{2}_{3,6}}\underline{Z}^{2}_{4,6}\quad D_{2,7}=\prod\limits_{i=2,3,4,6}\bar{Z}^{2}_{i,7}\quad$ $D_{2,8}=\prod\limits_{i=1,3,4,5}\underset{(7)}{\bar{Z}^{2}_{i,8}}\quad D_{2,9}=\underset{(7-8)}{\bar{Z}^{2}_{2,9}}\ \overset{(5-6)}{\underline{Z}^{2}_{4,9}}\ \overset{(6)}{\underline{Z}^{2}_{5,9}}\ \underline{Z}^{2}_{6,9}\quad D_{2,10}=\prod\limits_{i=1,2,3,4}\underset{(7-9)}{\bar{Z}^{2}_{i,10}}$ $D_{2,11}=\prod\limits_{\stackrel{{\scriptstyle i=2,3,4,}}{{\scriptscriptstyle{6,8,9,10}}}}\bar{Z}^{2}_{i,11}\quad D_{2,12}=\prod\limits_{\stackrel{{\scriptstyle i=1,3,4,5}}{{\scriptscriptstyle{7,9,10}}}}\underset{(11)}{\bar{Z}^{2}_{i,12}}\quad D_{2,13}=\prod\limits_{\stackrel{{\scriptstyle i=1,3,5,}}{{\scriptscriptstyle{7,\cdots,10}}}}\underset{(11-12)}{\bar{Z}^{2}_{i,13}}$ $D_{2,14}=\prod\limits_{\stackrel{{\scriptstyle i=1,\cdots,4}}{{\scriptscriptstyle{7,8,9}}}}\underset{(11-13)}{\bar{Z}^{2}_{i,14}}\quad D_{2,15}=\prod\limits_{\stackrel{{\scriptstyle i=2,\cdots,14}}{{\scriptscriptstyle{1\neq 5,7,11}}}}\bar{Z}^{2}_{i,15}\quad D_{2,16}=\prod\limits_{\stackrel{{\scriptstyle i=2,\cdots,14}}{{\scriptscriptstyle{i\neq 3,9}}}}\underset{(15)}{\bar{Z}^{2}_{i,16}}$ $D_{2,17}=\prod\limits_{\stackrel{{\scriptstyle i=1,\cdots,14}}{{\scriptscriptstyle{i\neq 2,4,13}}}}\underset{(15-16)}{\bar{Z}^{2}_{i,17}}\quad D_{2,18}=\prod\limits_{\stackrel{{\scriptstyle i=1,\cdots,14}}{{\scriptscriptstyle{i\neq 3,4}}}}\underset{(15-17)}{\bar{Z}^{2}_{i,18}}\quad D_{2,19}=\prod\limits_{i=1,\cdots,10}\underset{(15-18)}{\bar{Z}^{2}_{i,19}}$ $D_{2,20}=\prod\limits_{\stackrel{{\scriptstyle i=2,\cdots,19}}{{\scriptscriptstyle{i\neq 2,6,8,12}}}}\bar{Z}^{2}_{i,20}\quad D_{2,21}=\prod\limits_{\stackrel{{\scriptstyle i=2,\cdots,19}}{{\scriptscriptstyle{i\neq 3,9,16}}}}\underset{(20)}{\bar{Z}^{2}_{i,21}}\quad D_{2,22}=\prod\limits_{\stackrel{{\scriptstyle i=1,\cdots,19}}{{\scriptscriptstyle{i\neq 2,4,13,14}}}}\underset{(20-21)}{\bar{Z}^{2}_{i,22}}$ $D_{2,23}=\prod\limits_{\stackrel{{\scriptstyle i=1,\cdots,19}}{{\scriptscriptstyle{i\neq 3,4,18}}}}\underset{(20-22)}{\bar{Z}^{2}_{i,23}}\quad D_{2,24}=\prod\limits_{\stackrel{{\scriptstyle i=1,\cdots,19}}{{\scriptscriptstyle{i\neq 7,\cdots,10}}}}\underset{(20-23)}{\bar{Z}^{2}_{i,24}}$ and $\widetilde{C}_{2,1}=\prod\limits_{\stackrel{{\scriptstyle t=5,7}}{{\scriptscriptstyle{11,15}}}}D_{2,t}\quad\widetilde{C}_{2,2}=\prod\limits_{\stackrel{{\scriptstyle t=2,6,8}}{{\scriptscriptstyle{12,20}}}}D_{2,t}\quad\widetilde{C}_{2,3}=\prod\limits_{\stackrel{{\scriptstyle t=3,9}}{{\scriptscriptstyle{16,21}}}}D_{2,t}$ $\widetilde{C}_{2,4}=\prod\limits_{\stackrel{{\scriptstyle t=4,13}}{{\scriptscriptstyle{17,22}}}}D_{2,t}\widetilde{C}_{2,5}=\prod\limits_{t=18,23}D_{2,t}\quad\widetilde{C}_{2,6}=\prod\limits_{t=10,14}D_{2,t}$ $\widetilde{C}_{2,7}=D_{2,24}\quad\widetilde{C}_{2,18}=D_{2,19}\quad\widetilde{C}_{2,9}=\widetilde{C}_{2,10}=id$ Recall that a point in a totally degenerated surface is called a $k$-point if it is a singular point which is the intersection of $k$ planes. $\widetilde{\underline{\Delta}}^{2}_{v_{i,j}}$ : In $(S_{1})_{0}$, we have six points, which are 6-point $(v_{1,j},j=2,4,5,7,9,10)$ and four points which are 3-point $(v_{1,j},j=1,3,6,8$ ; note that the regeneration of this 3-point is not similar to the regular 3-point. See [1] for the braid monodromy factorization of the regeneration of our 3-point). In $(S_{2})_{0}$, we have eight points which are 5-point $(v_{2,j},1\leq j\leq 10,j\neq 5,6)$ and two points which are 4-point $(v_{2,j},j=5,6$). Note that the original branch curve, $S_{2}$, has also a few extra branch points. The existence of the extra branch points will be proved later (see Proposition (3.5)). The local braid monodromies, which are $\widetilde{\Delta}^{2}_{v_{2,j}}$, are introduced and regenerated in the following paragraphs. We denote the outcoming local BMF, resulting from the total regeneration $\widetilde{\Delta}^{2}_{v_{2,j}}$, as $\varphi_{2,j}$. Thus after performing a total regeneration to the whole BMF, the resulting BMF will be of the form $\varphi_{2}=\prod\limits^{1}_{i=10}C_{i}\varphi_{i}\prod\limits b_{i}$, where $b_{i}$ are braids corresponding to the extra branch points. Before presenting the expressions for local and global BMFs, we give a few notations. Let $a,b,c,d\in\mathbb{Z}$; denote: $F_{u}(a,b,c,d):=F_{u}(F_{u})_{Z^{-1}_{a,a^{\prime}}Z^{-1}_{d,d^{\prime}}},$ where $\\{b,c\\}<\\{a,d\\}$, and $c<b,a<d$ and $F_{u}=Z^{(3)}_{bb^{\prime},a}Z^{2}_{a^{\prime}d}(Z^{2}_{ad})_{Z^{2}_{bb^{\prime}a}}(Z^{3}_{bb^{\prime},d})_{Z^{2}_{bb^{\prime},a}}(Z_{cb^{\prime}})_{Z^{2}_{b^{\prime}d}Z^{2}_{b^{\prime}a}}(Z_{c^{\prime}b})_{Z^{2}_{bd}Z^{2}_{ba}Z^{2}_{bb^{\prime}}}$ $F_{m}(a,b,c,d):=F_{m}\cdot(F_{m})_{Z^{-1}_{a,a^{\prime}}Z^{-1}_{d,d^{\prime}}}$ where $a<\\{b,c\\}<d$ and $F_{m}=Z^{(3)}_{a^{\prime},cc^{\prime}}\cdot Z^{(3)}_{bb^{\prime},d}\cdot\widetilde{Z}_{c,b^{\prime}}\cdot\widetilde{Z}_{b^{\prime},c}(\underline{Z}^{2}_{a^{\prime},d})_{Z^{2}_{c^{\prime},cc^{\prime}}}\cdot\underline{Z}^{2}_{ad}$ where $\widetilde{Z}_{cb^{\prime}}=(\underline{Z}_{cb^{\prime}})_{Z^{2}_{b^{\prime},d}Z^{2}_{cc^{\prime}}Z^{2}_{a^{\prime}c}}\quad\mbox{and}\quad\widetilde{Z}_{b^{\prime},c}=(\underline{Z}_{b^{\prime},c})_{Z^{2}_{b^{\prime}d}Z^{2}_{a^{\prime}c^{\prime}}}$ $F_{\ell}(a,b,c,d):=F_{\ell}\cdot(F_{\ell})_{Z^{-1}_{a,a^{\prime}}Z^{-1}_{d,d^{\prime}}}$ where $\\{b,c\\}>\\{a,b\\}$ and $F_{\ell}=Z^{2}_{a^{\prime}d}\cdot Z^{(3)}_{d^{\prime},cc^{\prime}}(Z^{2}_{a^{\prime}d^{\prime}})_{Z^{2}_{d^{\prime},cc^{\prime}}Z^{2}_{a^{\prime}d}}(Z^{(3)}_{a^{\prime},cc^{\prime}})_{Z^{2}_{a^{\prime}d}}\cdot(\bar{Z}_{cb^{\prime}})_{Z^{2}_{d^{\prime},c}Z^{2}_{a^{\prime}c}Z^{2}_{a^{\prime}d}}(Z_{cb^{\prime}})_{Z^{2}_{cc^{\prime}}Z^{2}_{d^{\prime}c^{\prime}}Z^{2}_{a^{\prime}c^{\prime}}Z^{2}_{a^{\prime}d}}.$ Note that for $(\varphi_{1})_{0}$ and the singular points of $(S_{1})_{0}$, the regeneration process was already done [1], and thus we have the following: ###### Theorem 3.1. The BMF of the branch curve of $X_{1}$ is $\varphi_{1}=\prod\limits^{1}_{j=10}C_{1,j}\varphi_{1,j}$ where $C_{1,j},\,\varphi_{1,j}$ can be found in [1]. ###### Proof. see [1].∎ Thus, we have to compute the BMF of the branch curve of $X_{2}$. We begin by citing the results about the points $v_{2,5}$ and $v_{2,6}$; these are 4-points and for this type, the BMF of a fully regenerated neighbourhood was computed in [3]. ###### Proposition 3.1. The local braid monodromy $\varphi_{2,5}$ in a small neighbourhood around $v_{2,5}$ has the following form: $\varphi_{2,5}=F_{u}(18,4,3,23)$ and the local braid monodromy $\varphi_{2,6}$ (for $v_{2,6}$) has the same form, when substituting $3\rightarrow 5,\\\ \ 4\rightarrow 6,\ 18\rightarrow 10,\ 14\rightarrow 23$. ###### Proof. See [3].∎ We now move on to compute the local braid monodromy around a small neighbourhood of $v_{2,3}$, which is a 5-point. We will give – for this point – a detailed treatment for the computation of the local BMF, while for the other points $(v_{2,j},j=1,2,4,7,\cdots,10)$ we will just give the final results. We examine the point $v_{2,3}$ in the degenerated surface $(X_{2})_{0}$. Drawing a local neighbourhood of $v_{2,3}$ and numerating the lines $-\,L_{i}(1\leq i\leq 5)$ locally, we get: Figure 9 By the degeneration process, line 3 is regenerated first. By the Claim in [16, Section 2], we know that line $L_{3}$ is regenerated into a conic. More explicitly, we get that after regenerating $V=\bigcup\limits^{5}_{i=1}L_{i}$ in a small neighbourhood $U$ of $v_{2,3}$, $L_{3}$ turns into a conic $Q_{3}$ such that $Q_{3}$ is tangent to $L_{1}$ and $L_{5}$. Denote the resulting branch curve, after the regeneration by $\widetilde{V}$. Thus, the singularities of $T=\widetilde{V}\cap U$ are as in the figure below: Figure 10 ###### Proposition 3.2. The local braid monodromy factorization of the above configuration is $\widetilde{\varphi}=Z^{2}_{2,3}Z^{2}_{3^{\prime},4}Z^{4}_{1,3}\bar{Z}^{4}_{3^{\prime},5}\widetilde{Z}^{2}_{3^{\prime},4}\widetilde{Z}^{2}_{2,3}\widetilde{Z}^{2}_{3,3^{\prime}}(\Delta^{2}\langle 1,2,4,5\rangle)^{Z^{-2}_{3,4}}$ where the braids $\widetilde{Z}_{3^{\prime},4},\widetilde{Z}_{2,3},\widetilde{Z}_{3,3^{\prime}}$ correspond to the following paths: _Figure 11_ ###### Proof. Let $\\{p_{j}\\}^{8}_{j=1}$ be the singular points of a small neighbourhood (that is $U$) of $v_{2,3}$ (see figure 10) with respect to $\pi_{1}$ (the projection to the $X$-axis) as follows: $\\{p_{1},p_{2}\\},\\{p_{2},p_{5}\\}$ – the intersection points of $Q_{3}$ with $L_{2},L_{4}$. $p_{3},p_{4}$ – the tangent points of $Q_{3}$ and $L_{1},L_{5}$. $p_{7}$ – the branch point of $Q_{3}$. $p_{8}$ – the intersection point of $\\{L_{i}\\}_{i=1,2,4,5}$. Let $E$ (resp. $D$) be a closed disk on the $X$-axis (resp. $Y$-axis). Let $N=\\{x(p_{j})=x_{j}|1\leq j\leq 8\\},$ s.t. $N\subset E-\partial E$. Let $M$ be a real point on the $x$-axis, s.t. $x_{j}\ll M,\forall x_{j}\in N,1\leq j\leq 8$. There is a $g$-base $\ell(\gamma_{j})^{8}_{j=1}$ of $\pi_{1}(E-N,u)$, s.t. each path $\gamma_{j}$ is below the real line and the values of $\varphi_{M}$ with respect to this base and $E\times D$ are the ones given in the proposition. We look for $\varphi_{M}(\ell(\gamma_{j}))$ for $j=1,\cdots,8$. Choose a $g$-base $\ell(\gamma_{j})^{8}_{j=1}$ as above and put all the data in the following table: $j$ | $\lambda_{j}$ | $\varepsilon_{j}$ | $\delta_{j}$ ---|---|---|--- 1 | $<2,3>$ | 2 | $\Delta<2,3>$ 2 | $<3^{\prime},4>$ | 2 | $\Delta<3^{\prime},4>$ 3 | $<1,2>$ | 4 | $\Delta^{2}<1,2>$ 4 | $<4,5>$ | 4 | $\Delta^{2}<4,5>$ 5 | $<3^{\prime},4>$ | 2 | $\Delta<3^{\prime},4>$ 6 | $<2,3>$ | 2 | $\Delta<2,3>$ 7 | $<3,3^{\prime}>$ | 1 | $\Delta^{1/2}_{IR}<2>$ 8 | $<1,2,4,5>$ | 2 | – So, we get the following: $\xi_{x_{1}}=z_{2,3}\\\ \varphi_{M}(\ell(\gamma_{1}))=Z^{2}_{2,3}$ $\xi_{x_{2}}=z_{3^{\prime},4}$ ($\Delta\langle 2,3\rangle$ does not affect this path) $\varphi_{M}(\ell(\gamma_{2}))=Z^{2}_{3,4}$ $\xi_{x_{3}}=\,$ $\xrightarrow[\Delta<2,3>]{\Delta<3^{\prime},4>}\,\psfig{}=z_{1,3}\\\ \varphi_{M}(\ell(\gamma_{3}))=Z^{4}_{1,3}\\\ \xi_{x_{4}}=$ $\xrightarrow[\Delta<3^{\prime},4>]{\Delta^{2}<1,2>}\,$ $=\bar{z}_{3,5}$ ($\Delta\langle 2,3\rangle$ does not affect this path) $\varphi_{M}(\ell(\gamma_{4}))=\bar{Z}^{4}_{3^{\prime},5}\\\ \xi_{x_{5}}=$ $\xrightarrow[\Delta^{2}<1,2>\atop\Delta<3^{\prime},4>]{\Delta^{2}<4,5>}\,$ $=\tilde{z}_{3^{\prime},4}$ ($\Delta\langle 2,3\rangle$ does not affect this path) $\varphi_{M}(\ell(\gamma_{5}))=\tilde{Z}^{2}_{3^{\prime},4}\\\ $ $\xi_{x_{6}}=$ $\xrightarrow[\Delta^{2}<4,5>\atop\Delta<1,2>]{\Delta<3^{\prime},4>}\,$ $\xrightarrow[\Delta<2,3>]{\Delta<3^{\prime},4>}\,$ $=\tilde{z}_{2,3}$ $\varphi_{M}(\ell(\gamma_{6}))=\tilde{Z}^{2}_{2,3}\\\ $ $\xi_{x_{7}}=$ $\xrightarrow[\Delta<3^{\prime},4>]{\Delta<2,3>}\,$ $\xrightarrow[\Delta^{2}<1,2>]{\Delta^{2}<4,5>}\,$ $\xrightarrow[\Delta<2,3>]{\Delta<3^{\prime},4>}\\\ $ $=\tilde{z}_{3,3^{\prime}}$ $\varphi_{M}(\ell(\gamma_{7}))=\tilde{Z}_{3,3^{\prime}}\\\ $ $\xi_{x_{8}}=$ $\xrightarrow{\Delta^{\frac{1}{2}}_{IR}<2>}$ $\xrightarrow[\Delta<3^{\prime},4>]{\Delta<2,3>}\,$ $\xrightarrow[\Delta^{2}<1,2>]{\Delta^{2}<4,5>}\,$ $\xrightarrow[\Delta<2,3>]{\Delta<3^{\prime},4>}$ $=\Delta\langle 1,2,4,5\rangle_{Z_{\alpha}}$ where $Z_{\alpha}$ is the braid induced from the motion $\varphi_{M}(\ell(\gamma_{8}))=\Delta^{2}\langle 1,2,4,5\rangle_{Z_{\alpha}}$ ∎ The following regeneration regenerates a small neighborhood of $\bigcup\limits_{i=1,2,4,5}L_{i}$, which is, by definition, a 4-point. Since this type of 4-point and its BMF of its regeneration was treated earlier [3], we can find out what is the BMF of $v_{2,3}$ after the full regeneration. ###### Proposition 3.3. The local BMF $\varphi_{2,3}$ around a small neighborhood of $v_{2,3}$ is: $\varphi_{2,3}=Z^{2}_{2^{\prime},3}Z^{2}_{2,3}Z^{2}_{3^{\prime},4^{\prime}}Z^{2}_{3^{\prime},4}Z^{(3)}_{11^{\prime},3}\cdot\bar{Z}^{(3)}_{3^{\prime},55^{\prime}}\widetilde{Z}^{2}_{3^{\prime},4^{\prime}}\widetilde{Z}^{2}_{3^{\prime},4}\widetilde{Z}^{2}_{2^{\prime},3}\widetilde{Z}^{2}_{2,3}$ $\widetilde{Z}_{3,3^{\prime}}(F_{3}\cdot(F_{3})_{\vartheta})_{Z_{\alpha}}$ where $\vartheta=Z^{-1}_{4,4^{\prime}}\cdot Z^{-1}_{5,5^{\prime}}$, the braids $\widetilde{Z}_{3^{\prime},4^{\prime}},\widetilde{Z}_{3^{\prime},4},\widetilde{Z}_{2^{\prime},3},\widetilde{Z}_{2,3},\widetilde{Z}_{3,3^{\prime}}$ correspond to the following paths: and $Z_{\alpha}$ is the braid induced from the motion: _Figure 12_ and $F_{3}=Z^{(3)}_{22^{\prime},4}Z^{2}_{4^{\prime},5}(Z^{2}_{4,5})_{Z^{2}_{22^{\prime},4}}(Z^{(3)}_{22^{\prime},5})_{Z^{2}_{22^{\prime},4}}(Z_{1,2^{\prime}})_{Z^{2}_{2^{\prime},5}Z^{2}_{2^{\prime},4}}(Z_{1^{\prime},2})_{Z^{2}_{2,5}Z^{2}_{2,4}Z^{2}_{2,2^{\prime}}}$ ###### Proof. Using the regeneration rules, we replace 1. (1) $Z^{2}_{2,3}\,\,\,(Z^{2}_{3`,4},\tilde{Z}^{2}_{3`,4},\tilde{Z}^{2}_{2,3})$ by $Z^{2}_{2\,2^{\prime},3}\,\,\,(\mbox{resp.\,\,}Z^{2}_{3`,4\,4^{\prime}},\tilde{Z}^{2}_{3`,4\,4^{\prime}},\tilde{Z}^{2}_{2\,2^{\prime},3})$ (by the second regeneration rule) 2. (2) $Z^{4}_{1,3}\,\,\,(\bar{Z}^{4}_{3^{\prime},5})$ by $Z^{(3)}_{1\,1^{\prime},3}\,\,\,(\mbox{resp.\,\,}\bar{Z}^{(3)}_{3^{\prime},5\,5^{\prime}})$(by the third regeneration rule) 3. (3) $\Delta^{2}\langle 1,2,4,5\rangle$ by $F_{3}\cdot(F_{3})_{\vartheta}$. ∎ ###### Remark 3.1. Note that the last BMF was given when numerating the lines in the neighbourhood of $v_{2,3}$ locally. So, when numerating globally, we get: $\varphi_{2,3}=Z^{2}_{3^{\prime},9}Z^{2}_{3,9}Z^{2}_{9^{\prime},1\,6^{\prime}}Z^{2}_{9^{\prime},16}Z^{(3)}_{1\,1^{\prime},9}\bar{Z}^{(3)}_{9^{\prime},21\,21^{\prime}}\widetilde{Z}^{2}_{9^{\prime},16^{\prime}}\widetilde{Z}^{2}_{9^{\prime},16}\widetilde{Z}^{2}_{3^{\prime},9}\widetilde{Z}^{2}_{3,9}$ $\widetilde{Z}_{9,9^{\prime}}\cdot(F_{3}\cdot(F_{3})_{\vartheta})_{Z_{\alpha_{3}}}$ where $\vartheta=Z^{-1}_{16,16^{\prime}}Z^{-1}_{21,21^{\prime}}$, $Z_{\alpha_{3}}$ is the braid induced from the motion: _Figure 13_ and $F_{3}=Z^{(3)}_{3\,3^{\prime},16}Z^{2}_{16^{\prime},21}(Z^{2}_{16,21})_{Z^{2}_{3\,3^{\prime},16}}(Z^{(3)}_{3\,3^{\prime},21})_{Z^{2}_{3\,3^{\prime},16}}(Z_{1,3^{\prime}})_{Z^{2}_{3^{\prime},21}Z^{2}_{3^{\prime},16}}\cdot(Z_{1^{\prime},3})_{Z^{2}_{3,21}Z^{2}_{3,16}Z^{2}_{3,3^{\prime}}}.$ We now write the other BMFs. ###### Proposition 3.4. The local braid monodromy $\varphi_{2,1}$ is: $\varphi_{2,1}=Z^{(3)}_{11\,11^{\prime},15}(F_{u}(11,5,1,7))^{Z^{2}_{11\,11^{\prime},15}}\overset{(7-7^{\prime}\atop 5-5^{\prime})}{Z_{11^{\prime},15}}\widetilde{Z}_{15,15^{\prime}}\overset{(7-7^{\prime})}{Z^{2}_{5\,5^{\prime},15\,15^{\prime}}}\widetilde{Z}^{2}_{7\,7^{\prime},15^{\prime}}Z^{2}_{7\,7^{\prime},15}$ where $\widetilde{Z}_{15,15^{\prime}},\widetilde{Z}_{7\,7,15^{\prime}}$ correspond to the following paths: _Figure 14_ The local braid monodromy $\varphi_{2,2}$ is: $\varphi_{2,2}=Z^{2}_{12\,12^{\prime},20}Z^{(3)}_{8\,8^{\prime},20}\widetilde{Z}^{2}_{12\,12^{\prime},20}(F_{u}(8,6,2,12))_{Z_{\alpha_{2}}}\overset{(6-6^{\prime})}{Z^{(3)}_{2\,2,20^{\prime}}}\widetilde{Z}^{2}_{20,20^{\prime}}\widetilde{Z}^{2}_{6\,6^{\prime},20^{\prime}}Z^{2}_{6\,6^{\prime},20}$ where $\widetilde{Z}_{12\,12^{\prime},20},\widetilde{Z}_{20,20^{\prime}},\widetilde{Z}_{6\,6^{\prime},20^{\prime}}$ correspond to the following paths: _Figure 15_ and $Z_{\alpha_{2}}$ is the braid induced from the motion: . The local braid monodromy $\varphi_{2,4}$ is: $\varphi_{2,4}=Z^{2}_{4\,4^{\prime},13}Z^{(3)}_{13^{\prime},17\,17^{\prime}}Z^{(3)}_{2\,2^{\prime},13}Z^{2}_{4\,4^{\prime},13}\widetilde{Z}_{13,13^{\prime}}\overset{(17-17^{\prime})}{\underline{Z}^{2}_{13^{\prime},22\,22^{\prime}}}\widetilde{Z}^{2}_{13,22\,22^{\prime}}(F_{u}(22,4,2,17))_{Z_{\alpha_{4}}}$ where $\widetilde{Z}_{13,13^{\prime}},\widetilde{Z}_{13^{\prime},22\,22^{\prime}}$ correspond to the following paths: _Figure 16_ and $Z_{\alpha_{4}}$ is the braid induced from the motion: . The local braid monodromy $\varphi_{2,7}$ is: $\varphi_{2,7}=Z^{2}_{8\,8^{\prime},9}Z^{2}_{10\,10^{\prime},9^{\prime}}Z^{(3)}_{7\,7^{\prime},9}\overset{{\scriptscriptstyle{(10-10^{\prime})}}}{\underline{Z}^{(3)}_{9^{\prime},24\,24^{\prime}}}\widetilde{Z}^{2}_{9^{\prime},10\,10^{\prime}}\widetilde{Z}^{2}_{8\,8^{\prime},9}\widetilde{Z}_{9,9^{\prime}}(F_{\ell}(7,10,24,8))_{Z_{\alpha_{7}}}$ where $\widetilde{Z}_{9^{\prime},10\,10^{\prime}},\widetilde{Z}_{8\,8^{\prime},9^{\prime}},\widetilde{Z}_{9,9^{\prime}}$ correspond to the following paths: _Figure 17_ and $Z_{\alpha_{7}}$ is the braid induced from the motion: . The local braid monodromy $\varphi_{2,8}$ is: $\varphi_{2,8}=Z^{2}_{13^{\prime},14\,14^{\prime}}Z^{(3)}_{12\,12^{\prime},13}\overset{{\scriptscriptstyle{(14-14^{\prime})}}}{\underline{Z}^{(3)}_{13^{\prime},19\,19^{\prime}}}\widetilde{Z}^{2}_{13^{\prime},14\,14^{\prime}}\widetilde{Z}_{13,13^{\prime}}\widetilde{Z}^{2}_{11\,11^{\prime},13^{\prime}}Z^{2}_{11\,11^{\prime},13}F_{\ell}(12,14,19,11))_{Z_{\alpha_{8}}}$ where $\widetilde{Z}_{13^{\prime},14\,14^{\prime}},\widetilde{Z}_{11\,11^{\prime},13^{\prime}},\widetilde{Z}_{13,13^{\prime}}$ correspond to the following paths: _Figure 18_ and $Z_{\alpha_{8}}$ is the braid induced from the motion: . The local braid monodromy $\varphi_{2,9}$ is: $\varphi_{2,9}=Z^{(3)}_{15^{\prime},16\,16^{\prime}}F_{\ell}(17,19,18,16))_{Z^{2}_{15^{\prime},16\,16^{\prime}}}\overset{{\scriptscriptstyle{(16-16^{\prime})}}}{\underline{Z}^{(3)}_{15^{\prime},19\,19^{\prime}}}\widetilde{Z}_{15\,15^{\prime}}$ $\widetilde{Z}^{2}_{15,18\,18^{\prime}}\overset{{\scriptscriptstyle{(16-16^{\prime})}}}{Z^{2}_{15^{\prime},18\,18^{\prime}}}\bar{Z}^{2}_{15^{\prime},17\,17^{\prime}}\widetilde{Z}^{2}_{15,17\,17^{\prime}}$ where $\widetilde{Z}_{15\,15^{\prime}},\widetilde{Z}_{15\,18\,18^{\prime}},\widetilde{Z}_{15,17\,17^{\prime}}$ correspond to the following paths: _Figure 19_ The local braid monodromy $\varphi_{2,10}$ is: $\varphi_{2,10}=Z^{2}_{20^{\prime},21\,21^{\prime}}\bar{Z}^{(3)}_{20^{\prime},22\,22^{\prime}}\widetilde{Z}^{2}_{20^{\prime},21\,21^{\prime}}(F_{\ell}(21,24,23,22))_{Z_{\alpha_{\\!10}}}\underset{{{\scriptscriptstyle{(23-23^{\prime})}}}}{\bar{Z}^{(3)}_{20^{\prime},24\,24^{\prime}}}$ $\widetilde{Z}_{20\,20^{\prime}}\bar{Z}^{2}_{20^{\prime},23\,23^{\prime}}\widetilde{Z}^{2}_{20,23\,23^{\prime}}$ where $\widetilde{Z}_{20^{\prime},21\,21^{\prime}},\widetilde{Z}_{20\,20^{\prime}},\widetilde{Z}_{20,23\,23^{\prime}}$ correspond to the following paths: _Figure 20_ and $Z_{\alpha_{10}}$ is the braid induced from the motion: . Performing the regeneration affects also the braids induced from the parasitic line intersection. Denote by $C_{2,i}$ the braid, which is created from $\widetilde{C}_{2,i}$ during the regeneration process. Every $\widetilde{C}_{2,i}$ is a product of a 2-degree braid $Z^{2}_{i,j}$, which becomes, as a consequence of the second regeneration rule, an 8-degree braid: $Z^{2}_{ii^{\prime},jj^{\prime}}=Z^{2}_{i^{\prime},j^{\prime}}Z^{2}_{i^{\prime},j}Z^{2}_{i,j^{\prime}}Z^{2}_{i,j}$. If the path representing the braid $Z^{2}_{i,j}$ was above/below a point $p$, then the induced braids would be above/below the points $p$ and $p^{\prime}$. Before we present the global BMF, we have to check if there are extra branch points in the branch curves, that are created during the regeneration of a line $L_{i}$. An extra branch point contributes to a factorization the factor $Z_{i,i^{\prime}}$. (By “contributes” we mean that one should multiply the old factorization $Z_{i,i^{\prime}}$ from the right). $X_{1}$: It was proven in [1, prop. 16] that the factorization $\varphi_{1}=\prod\limits^{1}_{j=10}C_{1,j}\varphi_{1,j}$ is a BMF of the branch curve of $X_{1}$. Thus, there are no missing braids in the factorization above, and therefore there are no extra branch points. $X_{2}$: Denote by $\widetilde{\Delta}=\prod\limits^{1}_{j=10}C_{2,j}\varphi_{2,j}$. If $\widetilde{\Delta}$ was the BMF of the branch curve of $X_{2}$, then deg($\widetilde{\Delta})=\mbox{deg}(\Delta^{2}_{48})=48\cdot(48-1)=2256$. We show that this is not the situation here. $\mbox{deg}(\widetilde{\Delta})=\sum\limits^{10}_{j=1}\mbox{deg}(C_{2,j})+\sum\limits^{10}_{j=1}\mbox{deg}(\varphi_{2,j})$. $\sum\mbox{deg}(C_{2,j})=8\cdot 184=1472\ .\ \mbox{For}\ j=5,6,\ v_{2,j}$ are 4-point, and by [16], deg$(\varphi_{2,5})=\mbox{deg}(\varphi_{2,6})=48\,.\,\mbox{For}\ 1\leq j\leq 10,\ j\neq 5,6\ v_{2,j}$ are 5-point. Although these points have different configurations, their BMFs $-\varphi_{2,j}$ still have 6 factors of degree 3, 8 factors of degree 2, one factor of degree 1, and a factor representing the BMF of the regeneration of a 4-point, whose degree is 48. Thus $\forall\ 1\leq j\leq 10,\ j\neq 5,6,\ \mbox{deg}(\varphi_{2,j})=6\cdot 3+8\cdot 2+1+48=83$. So, deg$(\widetilde{\Delta})=1472+2\cdot 48+8\cdot 83=2232<2256$. Define the forgetting homomorphisms: $1\leq i\leq 24\,\,f_{i}:B_{48}[D,\\{1,1^{\prime},\cdots,24,24^{\prime}\\}]\rightarrow B_{2}[D,\\{i,i^{\prime}\\}].$ It is clear that if $\widetilde{\Delta}$ was a BMF, then $\forall\ i,\mbox{deg}(f_{i}(\widetilde{\Delta}))=2$. However, this is not the case in the current situation. It was proven in [18] (see also [19]), that if deg$(f_{i}(\widetilde{\Delta}))=k<2$, then there are $(2-k)$ extra branch points, and so there is a contribution of the factorization $\prod\limits^{2-k}_{m=1}Z_{i,i^{\prime}}$ to $\widetilde{\Delta}$. ###### Proposition 3.5. 1. _(1)_ The regeneration of the lines $L_{2,j},\ j=3,4,5,6,10,14,18,23$ contributes the factors $Z_{j,j^{\prime}}\cdot Z_{j,j^{\prime}}$ to $\widetilde{\Delta}$. 2. _(2)_ The regeneration of the lines $L_{2,j}\ j=7,8,11,12,16,17,21,22$ contributes the factor $Z_{j,j^{\prime}}$to $\widetilde{\Delta}$. ###### Proof. (1) We prove this case for $j=3;$ the other cases are done using the same method. By Lemma 3.3.3 (or Proposition 3.3.4) in [19], it is enough to prove that deg$(f_{3}(\widetilde{\Delta}))=0$. The braids coming from the parasitic intersection are sent by $f_{3}$ (and by any $f_{i}$, in fact) to $Id$, so it is enough to look only at the factors $\varphi_{2,k},\ 1\leq k\leq 10$ that involve braids, one of whose end points are 3 or $3^{\prime}$. The only suitable $k^{\prime}s$ are $k=5$ and $k=3$. Since $v_{2,3}$ and $v_{2,5}$ are both of 4-point, by [16, Lemma 8, (iv)], deg$(f_{3}(\varphi_{2,3})=\mbox{deg}(f_{3}(\varphi_{2,5}))=0$. Therefore deg$(f_{3}(\widetilde{\Delta}))=0$. (2) We prove for $j=7$; the other cases are done using the same method. It is enough to prove that deg$(f_{7}(\widetilde{\Delta}))=1$ (by [19]). As in (1), we only consider the factors $\varphi_{2,1}$ and $\varphi_{2,7}$. $v_{2,1}$ is a 5-point. The first regeneration is of the line $L_{2,15}$, (which turns into a conic, that intersects the line $L_{2,7}$ at two nodes, which induce braids of the form $Z^{2}_{7,15}$ and $Z^{2}_{7,15^{\prime}}$), which does not contribute to the regeneration factors of the form $Z_{7,7^{\prime}}$. After this regeneration, we are left with the regeneration of a 4-point, and by [16, Lemma 8, (iv)], we get deg$(f_{7}(\varphi_{2,1}))=0$. $v_{2,7}$ is also a 5-point. The first regeneration is of the line $L_{2,9}$, which turns into a conic, $Q_{2,(9,9^{\prime})}$, that is tangent to $L_{2,7}$ (by [16, Claim 1]). This tangency point is regenerated into three cusps (see [15]) which induces the product of three braids - $Z^{3}_{7^{\prime},9}\cdot Z^{3}_{7,9}\cdot(Z^{3}_{7^{\prime},9})_{Z^{-1}_{7,7^{\prime}}}=:Z^{(3)}_{7\,7^{\prime},9}$. By [16, Lemma 2, (i)], we see that deg$(f_{7}(Z^{(3)}_{7\,7^{\prime},9}))=1$. Again, the regeneration afterwards of the 4-point does not contribute a factor of the form $Z_{7,7^{\prime}}$ to the factorization. Thus, we get deg$(f_{7}(\varphi_{2,7}))=1$, and deg$(f_{7}(\widetilde{\Delta}))=1$. ∎ Define an ordered set $\\{i_{n}\\}^{16}_{n=1}:=\\{3,4,5,6,7,8,10,11,12,14,16,17,18,21,22,23\\},$ $\mbox{and for}\ 1\leq n\leq 16$ let: $b_{n}=\left\\{\begin{array}[]{llll}Z_{i_{n},i^{\prime}_{n}}\cdot Z_{i_{n},i^{\prime}_{n}}&i_{n}=\\{3,4,5,6,10,14,18,23\\}\\\ Z_{i_{n},i^{\prime}_{n}}&i_{n}=\\{7,8,11,12,16,17,21,22\\}\end{array}\right.$ ###### Proposition 3.6. $\varphi_{2}=\prod\limits^{1}_{j=10}C_{2,j}\varphi_{2,j}\prod\limits^{16}_{n=1}b_{n}$ is a braid monodromy factorization for $S_{2}$. The proof is divided into a number of lemmas. ###### Lemma 3.1. $\varphi_{2}=\prod\limits^{1}_{j=10}C_{2,j}\widetilde{\varphi}_{2,j}\prod\limits^{16}_{n=1}b_{n}$ is a braid monodromy factorization for $S_{2}$, where $\widetilde{\varphi}_{2,j}=(\varphi_{2,j})_{h_{j}}$ for some $h_{j}\in\langle Z_{kk^{\prime}}|v_{2,j}\in L_{2,k}\rangle.$ ###### Proof. Using Proposition VI.2.1 from [14] on $S_{2}$, we get that $\varphi_{2}=\prod\limits^{1}_{j=10}C_{2,j}\widetilde{\varphi}_{2,j}\prod\limits b_{\ell}$.$h_{j}\in\langle Z_{kk^{\prime}}|\,v_{2,j}\in L_{2,k}\rangle$ are determined by the regeneration of the embedding $B_{k}\hookrightarrow B_{24}$ to $B_{2k}\hookrightarrow B_{48}$ where $k=4$ when $j=5,6$ and $k=5$ otherwise $(1\leq j\leq 10,\ j\neq 5,6;$ see the definition of regeneration of an embedding in [16, Sec. 1]). $b_{\ell}$ are factors that are not converted by $\prod\limits C_{2,j}\widetilde{\varphi}_{2,j}$, and each $b_{n}$ is of the form $Y^{t_{i}}_{i},Y_{i}$, is a positive half-twist, $0\leq t_{i}\leq 3$. Note that deg$(\widetilde{\varphi}_{2,j})=\mbox{deg}(\varphi_{2,j})$. By the previous proposition, we know part of the $b_{\ell}$’s; so we can say that $\varphi_{2}=\prod\limits^{1}_{j=10}C_{2,j}\widetilde{\varphi}_{2,j}\prod\limits^{16}_{n=1}b_{n}\prod b_{\ell}.$ We compute deg$\bigg{(}\prod\limits^{1}_{j=10}C_{2,j}\widetilde{\varphi}_{2,j}\prod\limits^{16}_{n=1}b_{n}\bigg{)}$. By earlier computations and the previous proposition, $\mbox{deg}\bigg{(}\prod\limits^{1}_{j=10}C_{2,j}\widetilde{\varphi}_{2,j}\prod\limits^{16}_{n=1}b_{n}\bigg{)}=2232+2\cdot 8+8=2256=48\cdot 47\ =\mbox{deg}(\Delta^{2}_{48})=\mbox{deg}\varphi_{2}.$ Thus, we have to compute deg$(\prod b_{\ell})$. Since $\forall\ell,b_{\ell}$ is a positive power of a positive half-twist, we get $b_{\ell}=1\,\forall\ell$. So we have $\varphi_{2}=\prod\limits^{1}_{j=10}C_{2,j}\widetilde{\varphi}_{2,j}\prod\limits^{16}_{n=1}b_{n}.$ ∎ ###### Lemma 3.2. $\varphi_{2}=\prod\limits^{1}_{j=10}C_{2,j}\varphi_{2,j}\prod\limits^{16}_{n=1}b_{n}.$ ###### Proof. Recalling the invariance rules for the BMF of 4- and 5- point (see [16] and the appendix (Section (5))), we can apply them as in [16], and get that $\prod\limits^{1}_{j=10}C_{2,j}\varphi_{2,j}\prod\limits^{16}_{n=1}b_{n}$ is also a braid monodromy factorization. Note that although the invariance rules for the 5-point are different from the invariance rules of the standard 4/6 - point, what matters, as can be seen in [16, Section 4] is that the invariance rule regarding the horizontal lines in the 5-point (the two lines that are regenerated last) remains the same in this type of point. ∎ ## 4\. Computing the fundamental groups ### 4.1. Computation for $X_{2}$ By the Van Kampen theorem (Theorem (2.5)), we can compute the relations between the generators in the fundamental group of the complement of the branch curve. We will prove that $\pi_{1}(\mathbb{C}-S_{2})$ is a quotient of $\tilde{B}_{16}$. In order to do so, we have to compute the local relations (or the local fundamental groups of the complement of the branch curve) arising from each singular point of the branch curve. Note that points $v_{2,5},v_{2,6}$ are of the type 4-point, which was investigated in [11], [18]. Thus, we have to look at the remaining 5-points. We focus only on one 5-point – $v_{2,3}$; for the other 5-points, the procedure for deducing the relations is the same, and we state (later) only the relations coming from the branch points for these points. Recall that in the regeneration process, every line is “doubled”, and thus $S_{i}\cap\mathbb{C}$ will contain $48=2\cdot 24$ points. The generators of $\pi_{1}(\mathbb{C}^{2}-S,u)$ (see the Van Kampen Theorem (2.5)) induced from this doubling are denoted as $\\{\Gamma_{i},\Gamma_{i^{\prime}}\\}^{24}_{i=1}$, where each pair $\\{\Gamma_{i},\Gamma_{i^{\prime}}\\}$ originates from the same line. Denote $\bar{\Gamma}_{i}=\Gamma_{i}$ or $\Gamma_{i^{\prime}}$. Before examining $\varphi_{2,3}$ we state the following ###### Remark 4.1. _$\forall_{i,j}$ s.t. $L_{2,i}\bigcap L_{2,j}=\emptyset$, we have the following relations in $\pi_{1}(\mathbb{C}^{2}-S_{2})$:_ $[\bar{\Gamma}_{i},\bar{\Gamma}_{j}]=1.$ The proof of this remark is based on the parasitic intersection braids. From each braid of the expressions $C_{2,i}\,(i=1,...,10)$, using complex conjugation and the Van Kampen Theorem, we can induce the above relations. ###### Proposition 4.1. The following relations in $\pi_{1}(\mathbb{C}^{2}-S_{2})$ are induced from $\varphi_{2,3}$: _(1)_ $<\bar{\Gamma}_{16},\bar{\Gamma}_{1}>=<\bar{\Gamma}_{1},\bar{\Gamma}_{9}>=<\bar{\Gamma}_{9},\bar{\Gamma}_{21}>=<\bar{\Gamma}_{21},\bar{\Gamma}_{3}>=<\bar{\Gamma}_{3},\bar{\Gamma}_{16}>=1$ _(2)_ $[\bar{\Gamma}_{i},\bar{\Gamma}_{j}]=1$ where $L_{i},L_{j}$ do not bound the same triangle, besides $(i=1^{\prime},j=3),\ (i=1,j=3^{\prime})$ and $(i=1,j=21)$ _(3)_ $\Gamma_{1}=(\Gamma_{3^{\prime}})_{\Gamma^{-1}_{16}\Gamma^{-1}_{21}\Gamma_{9}}.$ ###### Proof. In the course of the proof we use the Van Kampen Theorem, the invariance relations of the 5-point, and the complex conjugation method (see [16]). We prove the proposition in several steps. Step 1: By looking at the braids (in $\varphi_{2,3})\ Z^{2}_{3\,3^{\prime},9},Z^{2}_{9\,9^{\prime},16},Z^{(3)}_{1\,1^{\prime},9}$ and $\bar{Z}^{(3)}_{9^{\prime},21\,21^{\prime}}$, we induce immediately (using invariance relations and complex conjugation for the last braid) the following relations: $[\bar{\Gamma}_{9},\bar{\Gamma}_{16}]=[\bar{\Gamma}_{3},\bar{\Gamma}_{9}]=<\\!\bar{\Gamma}_{1},\bar{\Gamma}_{9}\\!>=<\\!\bar{\Gamma}_{9},\bar{\Gamma}_{21}\\!>=1.$ Step 2: Note that the factors in $(F_{3}\cdot(F_{3})_{\vartheta})$ are conjugated by $Z_{\alpha_{3}}$. Denote the corresponding generators induced from $(F_{3}\cdot(F_{3})_{\vartheta})$ (after the conjugation) by $\widetilde{\Gamma}_{i}$. So: $\widetilde{\Gamma}_{3}=\Gamma_{9}\Gamma_{3}\Gamma^{-1}_{9}\stackrel{{\scriptstyle\mathrm{by\ step\ 1}}}{{=}}\Gamma_{3}\quad\quad\widetilde{\Gamma}_{3^{\prime}}=\Gamma_{9}\Gamma_{3^{\prime}}\Gamma^{-1}_{9}=\Gamma_{3^{\prime}}$ $\widetilde{\Gamma}_{1}=\Gamma_{9}\Gamma_{1}\Gamma^{-1}_{9}\quad\quad\quad\,\quad\widetilde{\Gamma}_{1^{\prime}}=\Gamma_{9}\Gamma_{1^{\prime}}\Gamma^{-1}_{9}.$ the other $\widetilde{\Gamma}_{i}$-s are not changed. So, we have, by the braid $Z^{(3)}_{3\,3^{\prime},16}$ in $F_{3}$ the relation $<\bar{\Gamma}_{3},\bar{\Gamma}_{16}>=1$ Step 3: From the braid $Z^{2}_{16^{\prime},21}$ in $F_{3}$, we get the relation: $[\Gamma_{16^{\prime}},\Gamma_{21}]=1$. Looking on the complex conjugate of the braid $(Z^{2}_{16^{\prime},21})_{\vartheta}$, we now get the relation $[\Gamma_{16},\Gamma^{-1}_{21}\Gamma_{21^{\prime}}\Gamma_{21}]=1\stackrel{{\scriptstyle\mathrm{(inv.rel.}\,\rho_{16}\rho_{21})}}{{\longrightarrow}}[\Gamma_{16^{\prime}},\Gamma^{-2}_{21}\Gamma_{21^{\prime}}\Gamma^{2}_{21}]=[\Gamma_{16^{\prime}},\Gamma_{21^{\prime}}]=1.$ By performing another time the invariance relation $(\rho_{16}\rho_{21})$, we get $[\Gamma_{16},\Gamma_{21}]=1$. From $[\Gamma_{16^{\prime}},\Gamma^{-1}_{21}\Gamma_{21^{\prime}}\Gamma_{21}]=1$, we get $[\Gamma_{16},\Gamma_{21^{\prime}}]=1$. So we have the relation $[\bar{\Gamma}_{16},\bar{\Gamma}_{21}]=1$. Step 4: From the braid $(Z^{(3)}_{3\,3^{\prime},2\,1})_{Z^{(3)}_{3\,3^{\prime},16}}$, we get the relation: $<\Gamma_{21},\Gamma_{16}\Gamma_{3^{\prime}}\Gamma^{-1}_{16}>=1$. By Step 3 we get $<\Gamma_{21},\Gamma_{3^{\prime}}>=1$; in the same way, we get $<\Gamma_{21},\Gamma_{3^{\prime}}>=1$ and by invariance relation, we get: $<\bar{\Gamma}_{21},\bar{\Gamma}_{3}>=1$. Step 5: From the braid $((Z_{1,3^{\prime}})_{Z^{2}_{3^{\prime},21}Z^{2}_{3^{\prime},16}})_{Z_{\alpha_{3}}}$, we get the relation: $\Gamma_{1}=(\Gamma_{3^{\prime}})_{\Gamma^{-1}_{16}\Gamma^{-1}_{21}\Gamma_{9}}.$ Thus $\displaystyle<\Gamma_{16},\Gamma_{1}>$ $\displaystyle=$ $\displaystyle<\Gamma_{16},(\Gamma_{3^{\prime}})_{\Gamma^{-1}_{16}\Gamma^{-1}_{21}\Gamma_{9}}>=<(\Gamma_{16})_{\Gamma_{21}},(\Gamma_{3^{\prime}})_{\Gamma^{-1}_{16}}>$ $\displaystyle\stackrel{{\scriptstyle[\Gamma_{16},\Gamma_{21}]=1}}{{=}}$ $\displaystyle<\Gamma_{16},\Gamma_{16}\Gamma_{3^{\prime}}\Gamma^{-1}_{16}>=<\Gamma_{16},\Gamma_{3^{\prime}}>=1.$ By the invariance relations, we get: $<\bar{\Gamma}_{16},\bar{\Gamma}_{1}>=1$. Step 6: We know that $\Gamma_{1}=(\Gamma_{3^{\prime}})_{\Gamma^{-1}_{16}\Gamma^{-1}_{21}\Gamma_{9}}$ and thus $(\Gamma_{1})_{\Gamma^{-1}_{9}\Gamma_{16}}=(\Gamma_{3^{\prime}})_{\Gamma^{-1}_{21}}$ (by $[\Gamma_{16},\Gamma_{21}]=1)$. From the braid $((Z^{2}_{16,21})_{Z^{2}_{3\,3^{\prime},16}})_{Z_{\alpha_{3}}}$ , we get the relation: $\displaystyle[\Gamma_{16},(\Gamma_{21})_{\Gamma_{3^{\prime}}\Gamma_{3}}]=1\ \mbox{or}$ $\displaystyle 1$ $\displaystyle=$ $\displaystyle[\Gamma_{3}\Gamma_{16}\Gamma^{-1}_{3},\Gamma^{-1}_{3^{\prime}}\Gamma_{21}\Gamma_{3^{\prime}}]\stackrel{{\scriptstyle<\Gamma_{3},\Gamma_{16}>=<\Gamma_{3^{\prime}},\Gamma_{21}>=1}}{{=}}$ $\displaystyle[\Gamma^{-1}_{16}\Gamma_{3}\Gamma_{16},\Gamma_{21}\Gamma_{3^{\prime}}\Gamma^{-1}_{21}]=$ $\displaystyle[\Gamma^{-1}_{16}\Gamma_{3}\Gamma_{16},\Gamma^{-1}_{16}\Gamma_{9}\Gamma_{1}\Gamma^{-1}_{9}\Gamma^{-1}_{16}]\stackrel{{\scriptstyle[\Gamma_{9},\Gamma_{16}]=[\Gamma_{9},\Gamma_{3}]=1}}{{=}}[\Gamma_{3},\Gamma_{1}]$ and by invariance we get $[\Gamma_{3^{\prime}},\Gamma_{1^{\prime}}]=1$. ∎ The following proposition proves the missing relations (e.g., $[\bar{\Gamma}_{1},\bar{\Gamma}_{21}]=1)$. The reason for separating this proposition from the former is because we use now relations which are not necessarily from $\varphi_{2,3}$. ###### Proposition 4.2. The following relations in $\pi_{1}(\mathbb{C}^{2}-S_{2})$ hold $[\bar{\Gamma}_{1},\bar{\Gamma}_{3}]=[\bar{\Gamma}_{1},\bar{\Gamma}_{21}]=1.$ ###### Proof. Due to the invariance relations of $v_{2,3}$, it is enough to prove $[\Gamma_{1},\Gamma_{3^{\prime}}]=1$ and $[\Gamma_{1},\Gamma_{21}]=1$. By the braid $Z_{3,3^{\prime}}$ (induced from an extra branch point), we know that $\Gamma_{3}=\Gamma_{3^{\prime}}$. Thus, by the last proposition ((4.1), Step 6), we have $1=[\Gamma_{1},\Gamma_{3}]=[\Gamma_{1},\Gamma_{3^{\prime}}].$ Looking on the local BMF of $v_{2,1}$, we have the following relation from the braid $((Z_{1,5^{\prime}})_{Z^{2}_{5^{\prime},11}Z^{2}_{5^{\prime},7}})_{Z^{2}_{11\,11^{\prime},15}}:$ $\Gamma_{1}=(\Gamma_{5^{\prime}})_{\Gamma^{-1}_{7}\Gamma_{15}\Gamma^{-1}_{11}\Gamma^{-1}_{15}}.$ Since $\Gamma_{21}$ commutes with $\Gamma_{5^{\prime}},\Gamma_{7},\Gamma_{15}$ and $\Gamma_{11}$ (due to the parasitic intersection braids), we have that $[\Gamma_{1},\Gamma_{21}]=1$. ∎ ###### Proposition 4.3. $\forall i,1\leq i\leq 24,i\neq 9,13,15,20,\ \Gamma_{i}=\Gamma_{i^{\prime}}$ in $\pi_{1}(\mathbb{C}^{2}-S_{2})$. We divide the proof into 2 lemmas. ###### Lemma 4.1. For $i=3,...,8,10,11,12,14,16,17,18,21,22,23:\ \Gamma_{i}=\Gamma_{i^{\prime}}$. ###### Proof. The relation $\Gamma_{i}=\Gamma_{i^{\prime}}$ is induced from the braids $Z_{i,i^{\prime}}$ which are created from the extra branch points (by Proposition 3.6).∎ ###### Lemma 4.2. For $i=1,2,19,24:\ \Gamma_{i}=\Gamma_{i^{\prime}}$. ###### Proof. We will prove in details only for $i=1$; the proof for the other $i$’s is the same. We know (from the braid $((Z_{1,5^{\prime}})_{Z^{2}_{5^{\prime},7}Z^{2}_{5^{\prime},11}})_{Z^{2}_{11\,11^{\prime},15}}$ in $\varphi_{2,1}$) the relation: $\Gamma_{1}=\Gamma_{15}\Gamma_{11}\Gamma_{7}\Gamma_{5}\Gamma^{-1}_{7}\Gamma^{-1}_{11}\Gamma^{-1}_{15}$ (we used the relation $\Gamma_{5}=\Gamma_{5`}$). Operating the invariance relations $(\rho_{1}\rho_{5})(\rho_{7}\rho_{11})$ and taking the inverse, we get: $\Gamma^{-1}_{1`}=\Gamma_{15}\Gamma_{11^{\prime}}\Gamma_{7^{\prime}}\Gamma^{-1}_{5^{\prime}}\Gamma^{-1}_{7^{\prime}}\Gamma^{-1}_{11^{\prime}}\Gamma^{-1}_{15}.$ Multiplying the above relations and using Lemma 4.1, we get $\Gamma^{-1}_{1^{\prime}}\Gamma_{1}=1$, or $\Gamma_{1}=\Gamma_{1^{\prime}}$. For $i=2$, we use the braid $((Z_{2,6^{\prime}})_{Z^{2}_{6^{\prime},8}Z^{2}_{6^{\prime},12}})_{Z_{\alpha_{2}}}$ from $\varphi_{2,2}$ and the same method as above. For $i=19,24$, one can use the braids $((\bar{Z}_{10^{\prime},24})_{\cdots})_{Z_{\alpha_{7}}}$ from $\varphi_{2,7}$ (or the braid $((\bar{Z}_{14^{\prime},19})_{\cdots})_{Z_{\alpha_{8}}}$ from $\varphi_{2,8}$) and continue as above. ∎ ###### Remark 4.2. _For each $1\leq i\leq 10$ we denote by $G_{2,i}$ the local fundamental whose generators are $\Gamma_{j}$, such that one of the endpoints of $L_{2,j}$ is $v_{2,i}$. Generalizing Propositions 4.1 and 4.2, it is easy to prove that $\forall_{i,j}$ s.t. $L_{2,i}$ and $L_{2,j}$ do not bound a common triangle, $[\bar{\Gamma}_{i},\bar{\Gamma}_{j}]=1$; and $\forall_{i,j}$ s.t. $L_{2,i}$ and $L_{2,j}$ bound a common triangle, $<\bar{\Gamma}_{i},\bar{\Gamma}_{j}>=1$ (in $\pi_{1}(\mathbb{C}^{2}-S_{2}))$_. ###### Remark 4.3. _It is important to state which braids are coming from the branch points. We list below (for each $\varphi_{2,i}$, for $1\leq i\leq 10,\ i\neq 5,6)$ which braid is induced from a branch point, that is created during the regeneration of the horizontal lines of the 5-point. We use the double and triple relations, and the last proposition, and we obtain:_. $\displaystyle i=1:$ $\displaystyle\quad\Gamma_{1}=(\Gamma_{5})_{\Gamma^{-1}_{7}\Gamma^{-1}_{11}\Gamma^{-1}_{15}}$ $\displaystyle i=2:$ $\displaystyle\quad\Gamma_{2}=(\Gamma_{6})_{\Gamma^{-1}_{8}\Gamma^{-1}_{12}\Gamma^{-1}_{20}}$ $\displaystyle i=3:$ $\displaystyle\quad\Gamma_{1}=(\Gamma_{3})_{\Gamma^{-1}_{16}\Gamma^{-1}_{21}\Gamma_{9}}$ $\displaystyle i=4:$ $\displaystyle\quad\Gamma_{2}=(\Gamma_{4})_{\Gamma^{-1}_{17}\Gamma_{13}\Gamma^{-1}_{22}}$ $\displaystyle i=7:$ $\displaystyle\quad\Gamma_{24}=(\Gamma_{10})_{\Gamma^{-1}_{7}\Gamma^{-1}_{8}\Gamma^{-1}_{9}}$ $\displaystyle i=8:$ $\displaystyle\quad\Gamma_{19}=(\Gamma_{14})_{\Gamma_{11}\Gamma_{12}\Gamma_{13}}$ $\displaystyle i=9:$ $\displaystyle\quad\Gamma_{18}=(\Gamma_{17})_{\Gamma^{-1}_{19}\Gamma_{15^{\prime}}\Gamma^{-1}_{16}}$ $\displaystyle i=10:$ $\displaystyle\quad\Gamma_{23}=(\Gamma_{24})_{\Gamma_{20^{\prime}}\Gamma^{-1}_{22}\Gamma_{21}}.$ ###### Proposition 4.4. For $i=5,6$, there exist a homomorphism $\alpha_{i}:\widetilde{B}_{4}\rightarrow G_{2,i}.$ ###### Proof. This proposition is proven in [18].∎ ###### Proposition 4.5. For $1\leq i\leq 10,\,i\neq 5,6$, there exist a homomorphism $\alpha_{i}:\widetilde{B}_{5}\rightarrow G_{2,i}.$ ###### Proof. Using the Remark (4.3) we prove only for $i=1$, and the proof for the other $i$’s is done in the same way. It is easy to check that $\alpha_{1}:B_{5}\rightarrow G_{2,1}$ is well- defined: $\alpha_{1}(X_{1})=\Gamma_{7}\quad\quad\alpha_{1}(X_{2})=\Gamma_{5}\quad\quad\alpha_{1}(X_{3})=\Gamma_{11}\quad\quad\alpha_{1}(X_{4})=\Gamma_{15}.$ Let $x_{1},...,x_{4}$ be the images of $X_{1},...,X_{4}$ in $\widetilde{B}_{5}$. Consider $T=X_{4}X_{3}X_{1}X_{2}X^{-1}_{1}X^{-1}_{3}X^{-1}_{4}$ in $B_{5}$ (see the following figure): Figure 21 $T$ is transversal to $X_{2}$. Let $t$ be the image of $T$ in $\widetilde{B}_{5}$; by the definition of $\widetilde{B}_{5}$ we have $[t,x_{2}]=1$. To show that $\alpha_{1}:B_{5}\rightarrow G_{2,1}$ defines the desired $\alpha_{1}:\widetilde{B}_{5}\rightarrow G_{2,1}$, it is enough to check that $[\alpha_{1}(T),\alpha(X_{2})]=1.$ We claim that $\alpha(T)=\Gamma_{1}$, because $\alpha(T)=\alpha(X_{4}X_{3}X_{1}X_{2}X^{-1}_{1}X^{-1}_{3}X^{-1}_{4})=(\Gamma_{5})_{\Gamma^{-1}_{7}\Gamma^{-1}_{11}\Gamma^{-1}_{15}}=\Gamma_{1}.$ So we have $[\alpha(T),\alpha(X_{2})]=[\Gamma_{1},\Gamma_{5}]=1.$ The last proposition deals with the relations between $\Gamma_{i^{\prime}}$ and $\Gamma_{i}$ in $\pi_{1}(\mathbb{C}^{2}-S_{2})$, where $i=9,13,15,20$.∎ ###### Proposition 4.6. The following relations in $\pi_{1}(\mathbb{C}^{2}-S_{2})$ hold: $\begin{array}[]{ll}\mbox{\emph{(i)}}\quad\Gamma_{13^{\prime}}=(\Gamma^{2}_{13})_{\Gamma^{2}_{17}}\Gamma^{-1}_{13}&\mbox{\emph{(ii)}}\quad\Gamma_{9^{\prime}}=(\Gamma^{2}_{9})_{\Gamma^{2}_{2\\!1}}\Gamma^{-1}_{9}\\\ \mbox{\emph{(iii)}}\quad\Gamma_{15^{\prime}}=\Gamma_{15}&\mbox{\emph{(iv)}}\quad\Gamma_{20^{\prime}}=\Gamma_{20}\end{array}$ ###### Proof. (i) From the braid $\widetilde{Z}_{13,13^{\prime}}$ in $\varphi_{2,3}$, we induce: $\Gamma_{13}=\Gamma^{-1}_{2}\Gamma^{-1}_{2^{\prime}}\Gamma^{-1}_{4}\Gamma^{-1}_{4^{\prime}}\Gamma^{-1}_{13}\Gamma^{-1}_{17^{\prime}}\Gamma_{17}\Gamma_{13^{\prime}}\Gamma^{-1}_{17}\Gamma^{-1}_{17^{\prime}}\Gamma_{13}\Gamma_{4^{\prime}}\Gamma_{4}\Gamma_{2^{\prime}}\Gamma_{2}.$ Using $[\bar{\Gamma}_{4},\bar{\Gamma}_{2}]=[\bar{\Gamma}_{4},\bar{\Gamma}_{13}]=1$ and $\Gamma_{2}=\Gamma_{2^{\prime}},\Gamma_{17}=\Gamma_{17^{\prime}}$, we get $1=\Gamma^{-2}_{2}\Gamma^{-1}_{13}\Gamma^{2}_{17}\Gamma_{13^{\prime}}\Gamma^{-2}_{17}\Gamma_{13}\Gamma^{2}_{2}\Gamma^{-1}_{13}\quad\mbox{or}:$ $1=\Gamma^{-2}_{2}(\Gamma^{2}_{17})_{\Gamma_{13}}\cdot(\Gamma^{-1}_{13}\Gamma_{13^{\prime}})\Gamma^{-2}_{17}(\Gamma^{2}_{2})_{\Gamma^{-1}_{13}}.$ By $[\bar{\Gamma}_{2},\bar{\Gamma}_{17}]=1$, we get $\Gamma^{-1}_{13}\Gamma_{13^{\prime}}=(\Gamma^{-2}_{17})_{\Gamma_{13}}(\Gamma^{-2}_{2})_{\Gamma^{-1}_{13}}\Gamma^{2}_{2}\Gamma^{2}_{17}=(\Gamma^{-2}_{17})_{\Gamma_{13}}((\Gamma^{-2}_{17})_{\Gamma^{-1}_{13}})^{-1}\underbrace{(\Gamma^{-2}_{17})_{\Gamma^{-1}_{13}}(\Gamma^{-2}_{2})_{\Gamma^{-1}_{13}}\Gamma^{2}_{2}\Gamma^{2}_{17}}_{F_{13}}.$ By Proposition (4.5), the braids in $F_{13}$: $\Gamma_{2},\Gamma_{17},(\Gamma_{2})_{\Gamma^{-1}_{13}},(\Gamma_{17})_{\Gamma^{-1}_{13}}$ are images of a good quadrangle by $\alpha_{4}$ in $\widetilde{B}_{5}$, and thus $F_{13}=1$ (by Lemma (2.1) on good quadrangles in $\widetilde{B}_{n}$). Thus $\Gamma^{-1}_{13}\Gamma_{13^{\prime}}=\Gamma^{-1}_{13}\Gamma^{-2}_{17}\Gamma^{2}_{13}\Gamma^{2}_{17}\Gamma^{-1}_{13}$ or $\Gamma_{13^{\prime}}=(\Gamma^{-2}_{13})_{\Gamma^{-2}_{17}}\cdot\Gamma^{-1}_{13}.$ (ii) We apply the same procedure as in (i) to the braid $\tilde{Z}_{9,9^{\prime}}$ from $\varphi_{2,3}$. (iii) Taking the complex conjugate of $\widetilde{Z}_{15,15^{\prime}}$ in $\varphi_{2,1}$, we induce the relation (using $\Gamma_{1}=\Gamma_{1^{\prime}},\Gamma_{11}=\Gamma_{11^{\prime}})$: $\Gamma_{15^{\prime}}=\Gamma^{-2}_{1}\Gamma^{-2}_{11}\Gamma_{15}\Gamma^{2}_{11}\Gamma^{2}_{1}$ or $1=\Gamma^{2}_{11}\Gamma^{2}_{1}(\Gamma_{15^{\prime}}\Gamma^{-1}_{15})(\Gamma^{-2}_{1})_{\Gamma^{-1}_{15}}(\Gamma^{-2}_{11})_{\Gamma^{-1}_{15}}$ $\Gamma_{15}\cdot\Gamma^{-1}_{15^{\prime}}=(\Gamma^{-2}_{1})_{\Gamma^{-1}_{15}}(\Gamma^{2}_{11})_{\Gamma^{-1}_{15}}\Gamma^{2}_{11}\Gamma^{2}_{1}.$ By the same method as in (i) (using $\alpha_{1}:\widetilde{B}_{5}\rightarrow G_{2,1})$, we get that $\Gamma_{15}\Gamma^{-1}_{15^{\prime}}=1\quad\mbox{or}\quad\Gamma_{15}=\Gamma_{15^{\prime}}.$ (iv) Taking the complex conjugation of $\widetilde{Z}_{20,20^{\prime}}$ in $\varphi_{2,2}$, we induce the relation (using $\Gamma_{2}=\Gamma_{2^{\prime}},\Gamma_{8}=\Gamma_{8^{\prime}})$, and $[\bar{\Gamma}_{12},\bar{\Gamma}_{20}]=1)$: $\Gamma_{20^{\prime}}=\Gamma^{-2}_{2}\Gamma^{-2}_{8}\Gamma_{20}\Gamma^{2}_{8}\Gamma^{2}_{2}$ and we proceed as in (iii). Thus: $\Gamma_{20}=\Gamma_{20^{\prime}}$. ∎ These propositions show that $\pi_{1}(\mathbb{C}^{2}-S_{2})$ is generated only by $\\{\Gamma_{i}\\}^{24}_{i=1}$, since the $\\{\Gamma_{i^{\prime}}\\}^{24}_{i=1}$ can be expressed only in terms of the $(\Gamma_{i^{\prime}})$’s. Our last goal is to prove the following: ###### Theorem 4.1. $G_{2}=\pi_{1}(\mathbb{C}-S_{2})$ is a quotient of $\widetilde{B}_{16}$. ###### Proof. We need to build an epimorphism $\tilde{\alpha}:\widetilde{B}_{16}\rightarrow G_{2}$. But first we build a new representation for $B_{16}$. Consider the geometric model ($D,K$), $\\#K=16$ as in figure 22. Let $\\{t_{i}\\}_{i\in\,I},\linebreak I=\\{1\leq i\leq 24,i\neq 1,3,5,8,11,12,16,17,22,i\in\mathbb{Z}\\}$ segments that connect points in $K$ and $T_{i}$ be the half-twists corresponding to $t_{i}$ (that is, $T_{i}=H(t_{i}),\,i\in I$). Figure 22 ###### Lemma 4.3. There exists a presentation of $B_{16}$ when the generators are $\\{T_{i}\,|\,i\in I\\}$ and the relations are: $\langle T_{i},T_{j}\rangle=1$ if $T_{i},T_{j}$ are consecutive, $[T_{i},T_{j}]=1$ if $T_{i},T_{j}$ are disjoint, $[T_{9},T_{24}T_{21}T^{-1}_{24}]=1$. ###### Proof. This is a standard consequence of the usual presentation of $B_{16}(D,K)$ (see [14]).∎ Let $X,Y\in B_{16}$ be transversal half-twists and let $\widetilde{B}_{16}=B_{16}/\langle[X,Y]\rangle$. By the previous lemma, $\widetilde{B}_{16}$ is generated by $\\{\tilde{T}_{i}\\}_{i\in I}$ (where $\tilde{T}_{i}$ are the images of $T_{i}$ in $\widetilde{B}_{16}$), and has same relations. Using Remark 4.3 we can define the “missing” $T_{j}$’s (where $1\leq j\leq 24,j\notin I$). We begin with $j=8,11,12,16,17,22:$ $T_{8}=(T_{10})_{T^{-1}_{7}T^{-1}_{9}T_{24}}\quad T_{12}=(T_{6})_{T^{-1}_{8}T^{-1}_{20}T_{2}}\quad T_{11}=(T_{14})_{T_{12}T_{13}T^{-1}_{19}}$ (we can use $T_{8}$ and $T_{12}$ since these $T$’s are already defined) $T_{22}=(T_{24})_{T_{20}T_{21}T_{23}}\quad T_{17}=(T_{2})_{T_{22}T^{-1}_{3}T_{4}}\quad T_{16}=(T_{17})_{T^{-1}_{19}T_{15}T_{18}}$ (we used $\Gamma_{20}=\Gamma_{20^{\prime}}$). In order to find out how to define $T_{3}$ (and $T_{5}$), we look at a relation induced from $\varphi_{2,5}$ ($\varphi_{2,6}$). Looking at the braid $(Z_{3,4^{\prime}})_{Z^{2}_{4^{\prime},23}Z^{2}_{4^{\prime},18}}$ from $\varphi_{2,5}$, we get the relation $\Gamma_{3}=(\Gamma_{4})_{\Gamma^{-1}_{18}\Gamma^{-1}_{23}}$. Thus we define $T_{3}=(T_{4})_{T^{-1}_{18}T^{-1}_{23}}$. In the same way we define $T_{5}=(T_{6})_{T^{-1}_{10}T^{-1}_{14}}$. By Remark 4.3, let $T_{1}=(T_{5})_{T^{-1}_{7}T^{-1}_{11}T^{-1}_{15}}$. Denoting by $\\{\tilde{T}_{j}\\}_{j=1}^{24}$ the images of $\\{{T}_{j}\\}_{j=1}^{24}$ in $\widetilde{B}_{16}$, we can say that $\widetilde{B}_{16}$ is generated by $\\{\tilde{T}_{j}\\}_{j=1}^{24}$ with the same relations as above, and when the $\\{\tilde{T}_{j}\\}_{j\notin I}$ are defined as above. Define $\tilde{\alpha}(\tilde{T_{j}})=\Gamma_{j},\,1\leq j\leq 24$. By Remark 4.2, it is easy to see that $\forall i,j$ such that $T_{i}$ and $T_{j}$ are consecutive, $\langle\Gamma_{i},\Gamma_{j}\rangle=1$; and when $T_{i}$ and $T_{j}$ are disjoint, $[T_{i},T_{j}]=1$. The relations induced from the action of taking quotient by $\langle[X,Y]\rangle$ (when $X,Y\in B_{16}$ are transversal) are also preserved, due to Propositions 4.4 and 4.5. Also, $\tilde{\alpha}$ is an epimorphism, since for every generator $\Gamma_{j}$ of $G_{2}$ there exists a $\tilde{T}_{j}$ s.t. $\tilde{\alpha}(\tilde{T})=\Gamma_{j}$. Thus $G_{2}\simeq\widetilde{B}_{16}/\mbox{Ker}\tilde{\alpha}$.∎ ### 4.2. Computation for $X_{1}$ As in subection (4.1), we can compute the local relations induced from each local braid monodromy. However, a quotient of the fundamental group of $X_{1}$ – called the stabilized fundamental group – was already computed in [4]. Noticing that $X_{1}$ can be regarded as a double cover of $\mathbb{C}\mathbb{P}^{1}\times\mathbb{C}\mathbb{P}^{1}$ branched along a smooth algebraic curve of degree (4,4), we can use [4, Theorem 4.6]. Let $\Theta_{1}:\pi_{1}(\mathbb{C}^{2}-S_{1})\rightarrow S_{n}$ be the geometric monodromy representation morphism (here $n=16$). ###### Theorem 4.2. : Let $K_{1}$ be the normal subgroup of $\pi_{1}(\mathbb{C}^{2}-S_{1})$ generated by all commutators $[\gamma_{1},\gamma_{2}],\gamma_{1},\gamma_{2}$ – geometric generators of $\pi_{1}(\mathbb{C}^{2}-S_{1})$, such that $\Theta_{1}(\gamma_{1})$ and $\Theta_{1}(\gamma_{2})$ are disjoint transpositions. Let $G^{0}_{1}\doteq(\ker(\Theta_{1}:\pi_{1}(\mathbb{C}^{2}-S_{1})\rightarrow S_{n})\bigcap\ker(\emph{deg:\,}\pi_{1}(\mathbb{C}^{2}-S_{1})\rightarrow\mathbb{Z}))/K_{1}\ ,$ where _deg_ is the degree morphism. Then ${\rm Ab}(G^{0}_{1})\simeq(\mathbb{Z}\oplus\mathbb{Z}_{2})^{15}\quad\mbox{and}\ \quad[G^{0}_{1},G^{0}_{1}]\simeq\mathbb{Z}_{2}\oplus\mathbb{Z}_{2}$ ###### Proof. See [4, Thm. 4.6].∎ The group $\pi_{1}(\mathbb{C}^{2}-S_{1})/K_{1}$ is called the _stabilized fundamental group_. Note that the stabilization procedure does not affect $\pi_{1}(\mathbb{C}^{2}-S_{2})=G_{2}$, since $G_{2}\simeq\widetilde{B}_{16}/\mbox{Ker}\tilde{\alpha}$, and $\widetilde{B}_{16}$ is already stabilized. ### 4.3. Comparing the BMT’s In this subsection we prove that the BMF’s of the branch curves $X_{1}$ and $X_{2}$ are not equivalent. We will do this by looking at the stabilized fundamental groups, related to $X_{1}$ and $X_{2}$. We denote by $K_{2}$ the normal subgroup of $\pi_{1}(\mathbb{C}^{2}-S_{2})$ generated by all commutators $[\gamma_{1},\gamma_{2}],\gamma_{1},\gamma_{2}$ \- geometric generators of $\pi_{1}(\mathbb{C}^{2}-S_{2})$, such that $\Theta_{2}(\gamma_{1})$ and $\Theta_{2}(\gamma_{2})$ are disjoint transpositions (here $\Theta_{2}:\pi_{1}(\mathbb{C}^{2}-S_{2})\rightarrow S_{16}$ is the geometric monodromy morphism). As was noted, $\pi_{1}(\mathbb{C}^{2}-S_{2})/K_{2}=\pi_{1}(\mathbb{C}^{2}-S_{2})$. We also note that $K_{1}=K_{2}$; since it is enough to pick one pair of geometric generators (e.g., $x_{2}$ and $(x_{2})_{x_{3}x_{1}}$, when the $x_{i}$’s are geometric generators), and define $K_{1}=K_{2}=\langle[x_{2},(x_{2})_{x_{3}x_{1}}]\rangle$. ###### Theorem 4.3. $G_{1}/K_{1}\not\simeq G_{2}/K_{2}$. ###### Proof. Note that $\widetilde{B}_{16}/\mbox{Ker}\tilde{\alpha}\simeq G_{2}\simeq G_{2}/K_{2}$. Denote $G^{0}_{2}\doteq(\ker(\Theta_{2}:\pi_{1}(\mathbb{C}^{2}-S_{2})\rightarrow S_{n})\bigcap\ker(\mbox{deg:\,}\pi_{1}(\mathbb{C}^{2}-S_{2})\rightarrow\mathbb{Z}))/K_{2}.$ It is known from [18] what is the commutant subgroup of $\tilde{P}_{n,0}=\mbox{ker}(\widetilde{B}_{n}\rightarrow S_{n})\bigcap\mbox{ker}(\mbox{deg}:\widetilde{B}\rightarrow\mathbb{Z});$ Explicitely, $[\tilde{P}_{n,0},\tilde{P}_{n,0}]$ is isomorphic to $\mathbb{Z}_{2}$. Therefore, $[G^{0}_{2},G^{0}_{2}]$ is a subgroup of $\mathbb{Z}_{2}$, whereas $[G^{0}_{1},G^{0}_{1}]\simeq\mathbb{Z}_{2}\oplus\mathbb{Z}_{2}$. But if $G_{1}/K_{1}\simeq G_{2}/K_{2}$ were isomorphic, then these two commutant subgroups would be equal.∎ ###### Remark 4.4. _We believe that an explicit computation of $\pi_{1}(\mathbb{C}^{2}-S_{1})$ (as in [11], [17]) would have shown that $K_{1}=\\{e\\}$._ ### 4.4. Computation for the Galois covers Let $\widetilde{\pi}_{i}:\widetilde{X}_{i}\rightarrow\mathbb{C}^{2}$ be the Galois covering corresponding to $\pi_{1}$ (see [11] for definitions). Recall that $\pi_{1}(\widetilde{X}_{i})=\ker\Theta_{i}/\langle\Gamma^{2}_{i,j}\rangle$ where $\Theta_{i}:\pi_{1}(\mathbb{C}^{2}-S_{i},\star)\rightarrow S_{n},\quad n=\mbox{deg}\pi_{i}\ (i=1\ \mbox{or}\ 2;$ the degree is the same) and $\\{\Gamma_{i,j}\\}$ are the generators of $\pi_{1}(\mathbb{C}^{2}-S_{i})$, for $i=1,2$. In [1] it was proved that $\pi_{1}(\widetilde{X}^{Aff}_{1})=\\{e\\}$. This is also the case for $X_{2}$. We know that the divisibility index of (the embedding of) $X_{2}$ is 1. Since $G_{2}$ is a quotient of $\widetilde{B}_{16}$, we can now use [10, Theorem 4.1] to prove that $\pi_{1}(\widetilde{X}^{Aff}_{2})=\\{e\\}$. The Main Result: Since the stabilized fundamental groups induced from them are not isomorphic, $\varphi_{1}$ is not Hurwitz-equivalent to $\varphi_{2}$. Therefore, $X_{1}$ and $X_{2}$ are not BMT–equivalent. Note that this inequivalence cannot be deduced from the computation of the fundamental groups of the Galois covers, as these groups are isomorphic. ## 5\. Appendix: Invariance rules for the BMF of a 5-point This appendix shows that the BMF of a 5-point is invariant under certain braids. We focus on the BMF $\varphi_{2,3}$, where the invariance rules for the other $\varphi_{i,j}$ ($i=1,2,\,1\leq j\leq 10$) are calculated in the same way. Recall that two factorizations are Hurwitz equivalent if they are obtained from each other by a finite sequence of Hurwitz moves. Definition: A factorized expression invariant under $h$ Let $t=t_{1}\cdot\ldots\cdot t_{m}$ be a factorized expression in a group $G$. We say that $t$ is invariant under $h\in G$ if $(t_{1})_{h}\cdot\ldots\cdot(t_{m})_{h}$ is Hurwitz equivalent to $t_{1}\cdot\ldots\cdot t_{m}$. We recall now a few invariance rules (see [16, section 3]): _Invariance rule_ II: $Z^{2}_{i,j\,j^{\prime}}$ ($Z^{2}_{i\,i^{\prime},j\,j^{\prime}}$)is invariant under $Z^{q}_{j\,j^{\prime}}$ (resp. $Z^{q}_{j\,j^{\prime}}Z^{p}_{i\,i^{\prime}}$). _Invariance rule_ III: $Z^{(3)}_{i,j\,j^{\prime}}$ is invariant under $Z^{q}_{j\,j^{\prime}}$. For our purposes (see the last paragraph in the proof of Lemma (3.2)), it is enough to prove the following ###### Proposition 5.1. $\varphi_{2,3}$ is invariant under $(Z_{1\,1^{\prime}}Z_{3\,3^{\prime}})^{p}(Z_{2\\!1\,2\\!1^{\prime}}Z_{16\,16^{\prime}})^{q}\,\,\forall p,q\in\mathbb{Z}$. ###### Proof. We first look at the factors outside $(F_{3}\cdot(F_{3})_{\vartheta})_{Z_{\alpha_{3}}}$. By the Invariance rule II, the factors $Z^{2}_{3\,3^{\prime},9},Z^{2}_{9^{\prime},16\,16^{\prime}},\tilde{Z}^{2}_{9^{\prime},16\,16^{\prime}},\tilde{Z}^{2}_{3\,3^{\prime},9}$ are invariant under $Z_{3\,3^{\prime}}$ and $Z_{16\,16^{\prime}}$; by ([16, invariance remark (iv)]), these factors are also invariant under $Z_{1\,1^{\prime}}$ and $Z_{2\\!1\,2\\!1^{\prime}}$ (since the paths are disjoint). Again, by the same invariance remark, $\tilde{Z}_{9,9^{\prime}}$ is invariant under $Z_{i\,i^{\prime}}\,\,i=1,3,16,21$. By the Invariance rule III, the factors $Z^{(3)}_{1\,1^{\prime},9}$ and $\bar{Z}^{(3)}_{9`,2\\!1\,2\\!1^{\prime}}$ are invariant under $Z_{1\,1^{\prime}}$ and $Z_{2\\!1\,2\\!1^{\prime}}$ (and also under $Z_{3\,3^{\prime}}$ and $Z_{16\,16^{\prime}}$ by the Invariance remark (iv)). We note that the conjugation by the braid $Z_{\alpha_{3}}$ is actually conjugation by $Z^{2}_{3\,3^{\prime},9}Z^{2}_{1\,1^{\prime},9}$, so it is also invariant under $Z_{i\,i^{\prime}}\,\,i=1,3,16,21$ (by invariance rule II and remark (iv)). When looking at the expression $F_{3}\cdot(F_{3})_{\vartheta}$, we see that this case was already done in [3, invariance property 8.7]; it was proved there that $F_{3}\cdot(F_{3})_{\vartheta}$ is invariant under $(Z_{1\,1^{\prime}}Z_{3\,3^{\prime}})^{p}(Z_{2\\!1\,2\\!1^{\prime}}Z_{16\,16^{\prime}})^{q}$.∎ ## References * [1] M. Amram, C. Ciliberto, R. Miranda, M. Teicher, Braid monodromy factorization for a non-prime $K3$ surface branch curve , Israel Journal of Mathematics, to appear. * [2] M. Amram, M. Friedman, M. Teicher,_The fundamental group of complement of a branch curve of a Hirzebruch surface $F_{2,(2,2)}$_, submitted to Topology. * [3] M. Amram, M. Teicher, _The fundamental group of the complement of the branch curve of the double torus_ , Journal of Mathematics, 40(4), (2003), 587-893. * [4] D. Auroux, S. K. Donaldson, L. Katzarkov, M. Yotov, Fundamental groups of complements of plane curves and symplectic invariants, Topology 43, (2004), 1285-1318. * [5] C. Ciliberto, R. Miranda, On the Gaussian map for canonical curves of low genus, Duke Mathemtical J., 61, No. 2 (1990), 417-442. * [6] C. Ciliberto, R. Miranda, M. Teicher, Pillow degenerations of K3 surfaces, In: “Applications of Algebraic Geometry to Coding Theory, Physics, and Computation”, NATO Science Series II, Vol. 36 (2001), 53-63. * [7] M. Friedman, M. Teicher, On the fundamental group related to the Hirzebruch surface $F_{1}$, submitted to Sci. China ser. A. * [8] V. S. Kulikov and M. Teicher, Braid monodromy factorizations and diffeomorphism types, Izv. Ross. Akad. Nauk Ser. Mat. 64(2), (2000), 89-120 , [Russian]; English transl., Izvestiya Math. 64(2), (2000), 311-341. * [9] E. Liberman, M. Teicher, The Hurwitz equivalence problem is undecidable, math.LO/0511153, preprint. * [10] C. Liedtke, On Fundamental Groups of Galois closures of generic projections, Bonner Mathematische Schriften Nr. 367 (2004). * [11] B. Moishezon, On cuspidal branch curves, J. Algebraic Geometry 2 (1993) no. 2, 309-384. * [12] B. Moishezon, A. Robb and M. Teicher, On Galois covers of Hirzebruch surfaces, Math. Ann. 305, (1996), 493-539. * [13] B. Moishezon, M. Teicher, Simply connected algebraic surfaces of positive index, Invent. Math. 89 (1987), 601-643. * [14] B. Moishezon, M. Teicher, Braid group technique in complex geometry, I, Line arrangements in $\mathbb{CP}^{2}$, Contemp. Math. 78, (1988), 425-555. * [15] B. Moishezon, M. Teicher, Braid group technique in complex geometry, II, From arrangements of lines and conics to cuspidal curves, Algebraic Geometry, Lecture Notes in Math., vol. 1479 (1990), 131-180. * [16] B. Moishezon, M. Teicher, Braid group techniques in complex geometry IV: Braid monodromy of the branch curve $S_{3}$ of $V_{3}\rightarrow\mathbb{CP}^{2}$ and application to $\pi:(\mathbb{CP}^{2}-S_{3},\ast)$, Contemp. Math. 162, (1993), 332-358. * [17] B. Moishezon, M. Teicher, Braid group techniques in complex geometry, V: The fundamental group of complements of a branch curve of Veronese generic projection, Communications in Analysis and Geometry 4, (1996), no. 1, 1-120. * [18] A. Robb, The topology of branch curves of complete intersections, Doctoral Thesis, Columbia University, (1994). * [19] A. Robb, On branch curves of Algebraic Surfaces, Stud. Adv. Math. 5, (1997), 193-221. * [20] E.R. Van Kampen,On the fundamental group of an algebraic curve, Amer. J. Math. 55 (1933), 255-260.

arxiv-papers

2024-09-04T02:48:53.803240

0802.2379

Towards A Realistic Grand Gauge-Higgs Unification Scenario111 Talk given at International Workshop on Grand Unified Theories : Current Status and Future Prospects (GUT07), December 17-19 2007, Kusatsu, Japan. This talk is based on a paper collaboration with C.S. Lim [1]. Nobuhito Maru Department of Physics, Kobe University, Kobe 657-8501, Japan abstract In this talk, we discuss an attempt to construct a realistic model of the grand gauge-Higgs unification. We investigate a 5D $SU(6)$ grand gauge-Higgs unification model compactified on an orbifold $S^{1}/Z_{2}$. Ordinary quarks and leptons, together with right-handed neutrinos, are just accommodated into a minimal set of representations of the gauge group, without introducing any exotic states in the same representations. The proton decay turns out to be forbidden at least at the tree level. We also find a correct electroweak symmetry breaking $SU(2)_{L}\times U(1)_{Y}\to U(1)_{em}$ is easily realized by introducing suitable number of adjoint fermions. The hierarchy problem, especially the problem of how to stabilize the Higgs mass under the quantum correction, has played a key role to motivate the physics beyond the standard model. Almost all possible scenarios to solve the problem invoke to some sort of symmetry in order to protect the Higgs mass at the quantum level. Supersymmetry is the most popular scenario and has been extensively discussed. Recently the gauge-Higgs unification scenario [2, 3, 4] has obtained a revived interest as a possible new avenue to solve the problem [5]. In this scenario the Higgs is regarded as the extra space component of higher dimensional gauge fields and the quantum corrections to the Higgs mass becomes finite by higher dimensional gauge symmetry without relying on the supersymmetry. This fact has been verified [5]-[10] at one loop level and [11] at two loop level in various models by explicit calculations. By the same reasoning, the gauge-Higgs unification predicts the physical observables to be finite thanks to the higher dimensional gauge invariance in spite of its nonrenormalizability. The divergence structure of S and T parameters in the gauge-Higgs unification was investigated and an interesting relation between them, which is only spcific to the gauge-Higgs unification, was predicted to be finte in six dimensional case [12]. The anomalous magnetic moment in the gauge-Higgs unification was found to be finite in arbitrary dimension [13]. Recently, the gluon fusion process [14, 15, 16] and the two photon decay of the Higgs boson [15] crucial for the LHC physics were studied and a very interesting prediction was reported. There have been many works on rich structure of the theory and its phenomenology, see [17]-[35]. Another possible interesting scenario is to regard the Higgs as a pseudo Nambu-Goldstone (PNG) boson due to the breakdown of some global symmetry. As far as the global symmetry is larger than local gauge symmetry, even though some N-G bosons are absorbed to gauge bosons via Higgs mechanism on the spontaneous symmetry breaking, there should remain some physical PNG bosons, which can be identified as the Higgs bosons [36]-[42]. The scenario faces a difficulty at quantum level, once gauge interactions are switched on. Namely, the Higgs mass suffers from a quadratic divergence, essentially because the original global symmetry is partly gauged and therefore the global symmetry is hardly broken by the gauge couplings. However, such difficulty may be avoided, once direct products of identical global symmetries are taken. It is interesting to note that such “dimensional deconstruction” or related “little Higgs” scenarios [43] can be regarded as a kind of gauge-Higgs unification, where the extra space has finite number of lattice points. In fact, it is a recent remarkable progress to have established the relation between a four dimensional theory with global symmetry $G$ and a five dimensional gauge theory with gauge symmetry $G$, through the AdS/CFT correspondence (holographic approach) [44, 45]. The gauge hierarchy problem was originally discussed in the framework of Grand Unified Theory (GUT) as the problem to keep the discrepancy between the GUT scale and the weak scale. So it will be meaningful to test the possible scenarios in the framework of GUT. The PNG boson scenario in the framework of GUT was discussed long time ago [36]. Since the global symmetry needs to be larger than the gauge symmetry $SU(5)$, a minimal model with an $SU(6)$ global symmetry was proposed. As explained above, unfortunately the Higgs boson suffers from a quadratic divergence at quantum level, and SUSY was introduced to eliminate the unwanted divergence. In this work, in view of the recent progress mentioned above, we attempt to construct a GUT model based on the scenario of gauge-Higgs unification, which is refered to as “grand gauge-Higgs unification”. A nice thing in our scenario is that the Higgs mass is automatically stabilized at the quantum level without relying on the SUSY. What we adopt is a minimal grand gauge-Higgs unification model, i.e. 5-dimensional (5D) GUT with an $SU(6)$ gauge symmetry. It is interesting to note that an $SU(6)$ symmetry emerges again, as suggested by the AdS/CFT correspondence. This is because in the gauge-Higgs unification, the gauge symmetry needs to be enlarged from the minimal one $SU(5)$, since Higgs inevitably belongs to the adjoint representation of the gauge group while the Higgs should behave as the fundamental representation of $SU(5)$. In addition to the finite Higgs mass, as a bonus, we find that the sector of fermionic zero-mode of the theory just accommodates three generations of quarks and leptons. Namely, we do not encounter the problem of introducing massless exotic particles in the representations quarks and leptons belong to, which often happens in the gauge-Higgs unification. Another remarkable feature, we will see below, is that the dangerous proton decay due to the exchange of GUT particles turns out to be prohibited at least at the tree level without any symmetry. This is due to the splitting multiplet mechanism. We will also discuss how the desirable gauge symmetry breaking is realized via Hosotani mechanism [4]. We will see the desirable pattern of gauge symmetry breaking is realized without introducing additional scalar matter fields. In ref. [46], an elegant 5D $SU(6)$ grand gauge-Higgs unification model was discussed, but as a SUSY theory. In this context, a non-SUSY $SU(6)$ grand gauge-Higgs unification model has been already studied [21], where the main focus was in the pattern of electroweak symmetry breaking and a viable Higgs mass satisfying the experimental lower limit was obtained. The set-up of our model concerning gauge-Higgs sector just follows the model of [21] and [46]. The 5D space-time we consider has an extra space compactified on an orbifold $S^{1}/Z_{2}$ with a radius $R$, whose coordinate is $y$. On the fixed points $y=0,\pi R$, the different $Z_{2}$ parities are assigned as $\displaystyle P={\rm diag}(+,+,+,+,+,-)~{}{\rm at}~{}y=0,\quad P^{\prime}={\rm diag}(+,+,-,-,-,-)~{}{\rm at}~{}y=\pi R,$ (1) which implies the gauge symmetry breaking pattern $\displaystyle SU(6)\to SU(5)\times U(1)~{}{\rm at}~{}y=0,$ (2) $\displaystyle SU(6)\to SU(2)\times SU(4)\times U(1)~{}{\rm at}~{}y=\pi R.$ (3) in each fixed point. These symmetry breaking patterns are inspired by [21, 36, 46]. It is instructive to see the concrete parity assignments of each component of the 4D gauge field $A_{\mu}$ and 4D scalar field $A_{5}$, $\displaystyle A_{\mu}$ $\displaystyle=$ $\displaystyle\left(\begin{array}[]{cccccc}(+,+)&(+,+)&(+,-)&(+,-)&(+,-)&(-,-)\\\ (+,+)&(+,+)&(+,-)&(+,-)&(+,-)&(-,-)\\\ (+,-)&(+,-)&(+,+)&(+,+)&(+,+)&(-,+)\\\ (+,-)&(+,-)&(+,+)&(+,+)&(+,+)&(-,+)\\\ (+,-)&(+,-)&(+,+)&(+,+)&(+,+)&(-,+)\\\ (-,-)&(-,-)&(-,+)&(-,+)&(-,+)&(+,+)\\\ \end{array}\right),$ (10) $\displaystyle A_{5}$ $\displaystyle=$ $\displaystyle\left(\begin{array}[]{cccccc}(-,-)&(-,-)&(-,+)&(-,+)&(-,+)&(+,+)\\\ (-,-)&(-,-)&(-,+)&(-,+)&(-,+)&(+,+)\\\ (-,+)&(-,+)&(-,-)&(-,-)&(-,-)&(+,-)\\\ (-,+)&(-,+)&(-,-)&(-,-)&(-,-)&(+,-)\\\ (-,+)&(-,+)&(-,-)&(-,-)&(-,-)&(+,-)\\\ (+,+)&(+,+)&(+,-)&(+,-)&(+,-)&(-,-)\\\ \end{array}\right),$ (17) where $(+,-)$ means that the $Z_{2}$ parity is even (odd) at $y=0(y=\pi R)$, for instance. Kaluza-Klein (KK) mode expansion in each type of the parity assignment is given by $\displaystyle\Phi_{(+,+)}(x,y)$ $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{2\pi R}}\left[\phi^{(0)}_{(+,+)}(x)+\sqrt{2}\sum_{n=1}^{\infty}\phi_{(+,+)}^{(n)}(x)\cos\left(\frac{n}{R}y\right)\right],$ (18) $\displaystyle\Phi_{(+,-)}(x,y)$ $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{\pi R}}\sum_{n=0}^{\infty}\phi_{(+,-)}^{(n)}(x)\cos\left(\frac{n+\frac{1}{2}}{R}y\right),$ (19) $\displaystyle\Phi_{(-,+)}(x,y)$ $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{\pi R}}\sum_{n=0}^{\infty}\phi_{(-,+)}^{(n)}(x)\sin\left(\frac{n+\frac{1}{2}}{R}y\right),$ (20) $\displaystyle\Phi_{(-,-)}(x,y)$ $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{\pi R}}\sum_{n=1}^{\infty}\phi_{(-,-)}^{(n)}(x)\sin\left(\frac{n}{R}y\right).$ (21) Noting that the 4D massless KK zero mode appears only in the $(+,+)$ component, the gauge symmetry breaking by orbifolding is found (see $A_{\mu}$ parity assignment) to be $\displaystyle SU(6)\to SU(3)_{C}\times SU(2)_{L}\times U(1)_{Y}\times U(1)_{X}.$ (22) Here the hypercharge $U(1)_{Y}$ is contained in the upper-left $5\times 5$ block of Georgi-Glashow $SU(5)$. Therefore, we obtain $\displaystyle g_{3}=g_{2}=\sqrt{\frac{5}{3}}g_{Y},$ (23) at the unification scale, which will be not far from $1/R$. This means that the Weinberg angle is just the same as the Georgi-Glashow $SU(5)$ GUT, namely $\sin^{2}\theta_{W}=3/8$ ($\theta_{W}$: Weinberg angle) at the classical level. In fact, we can explicitly confirm it for a ${\bf 6^{*}}$ representation given in (28) below, $\displaystyle\sin^{2}\theta_{W}=\frac{{\rm Tr}I_{3}^{2}}{{\rm Tr}Q^{2}}=\frac{(\frac{1}{2})^{2}+(-\frac{1}{2})^{2}}{(\frac{1}{3})^{2}\times 3+(-1)^{2}}=\frac{3}{8}$ (24) where $I_{3}$ is the third component of $SU(2)_{L}$ and $Q$ is an electric charge. In order to compare with the experimental data, the gauge coupling running effects have to be taken into account. Such a study is beyond the scope of this work. Since Higgs belongs to the doublet of $SU(2)_{L}$ and the electroweak gauge symmetry is embedded into ordinary $SU(5)$, the Z boson mass is given as $\displaystyle M_{Z}^{2}=\frac{M_{W}^{2}}{\cos^{2}\theta_{W}}=\sqrt{\frac{8}{5}}M_{W}\simeq 102{\rm GeV}$ (25) at the classical level. On the other hand, concerning $A_{5}$, the zero mode appears only in the doublet component, as we wish. The colored Higgs has a mass at least of the order of $1/R$, and the doublet-triplet splitting is automatically realized [47]. Let us now turn to the non-trivial question of how quarks and leptons can be accommodated into the representations of $SU(6)$, without introducing exotic states in the zero mode sector. Key observation is that the fundamental representation ${\bf 6}$ of $SU(6)$ contains a doublet of $SU(2)_{L}$ and symmetric products of ${\bf 6}$ easily introduces a triplet of $SU(2)_{L}$, which is exotic. We therefore focus on the possibility of totally antisymmetric tensor representations of $SU(6)$. We find that the minimal set to accommodate one generation of quarks and leptons contains two ${\bf 6^{*}}$ and one ${\bf 20}$ representations. Their parity assignments are fixed according to (1), $\displaystyle{\bf 6^{*}}$ $\displaystyle=$ $\displaystyle\left\\{\begin{array}[]{l}{\bf 6^{*}}_{L}=\underbrace{({\bf 3^{*},1})_{(1/3,-1)}^{(+,-)}\oplus l_{L}({\bf 1,2})_{(-1/2,-1)}^{(+,+)}}_{{\bf 5^{*}}}\oplus\underbrace{({\bf 1,1})_{(0,5)}^{(-,-)}}_{{\bf 1}}\\\ {\bf 6^{*}}_{R}=\underbrace{({\bf 3^{*},1})_{(1/3,-1)}^{(-,+)}\oplus({\bf 1,2})_{(-1/2,-1)}^{(-,-)}}_{{\bf 5^{*}}}\oplus\underbrace{\nu_{R}({\bf 1,1})_{(0,5)}^{(+,+)}}_{{\bf 1}}\end{array}\right.$ (28) $\displaystyle{\bf 6^{*}}$ $\displaystyle=$ $\displaystyle\left\\{\begin{array}[]{l}{\bf 6^{*}}_{L}=\underbrace{({\bf 3^{*},1})_{(1/3,-1)}^{(-,-)}\oplus({\bf 1,2})_{(-1/2,-1)}^{(-,+)}}_{{\bf 5^{*}}}\oplus\underbrace{({\bf 1,1})_{(0,5)}^{(+,-)}}_{{\bf 1}}\\\ {\bf 6^{*}}_{R}=\underbrace{d_{R}^{*}({\bf 3^{*},1})_{(1/3,-1)}^{(+,+)}\oplus({\bf 1,2})_{(-1/2,-1)}^{(+,-)}}_{{\bf 5^{*}}}\oplus\underbrace{({\bf 1,1})_{(0,5)}^{(-,+)}}_{{\bf 1}}\\\ \end{array}\right.$ (31) $\displaystyle{\bf 20}$ $\displaystyle=$ $\displaystyle\left\\{\begin{array}[]{l}{\bf 20}_{L}=\underbrace{q_{L}({\bf 3,2})_{(1/6,-3)}^{(+,+)}\oplus({\bf 3^{*},1})_{(-2/3,-3)}^{(+,-)}\oplus({\bf 1,1})_{(1,-3)}^{(+,-)}}_{{\bf 10}}\\\ \hskip 42.67912pt\oplus\underbrace{({\bf 3^{*},2})_{-1/6,3}^{(-,+)}\oplus({\bf 3,1})_{(2/3,-3)}^{(-,-)}\oplus({\bf 1,1})_{(-1,3)}^{(-,-)}}_{{\bf 10^{*}}}\\\ {\bf 20}_{R}=\underbrace{({\bf 3,2})_{(1/6,-3)}^{(-,-)}\oplus({\bf 3^{*},1})_{(-2/3,-3)}^{(-,+)}\oplus({\bf 1,1})_{(1,-3)}^{(-,+)}}_{{\bf 10}}\\\ \hskip 42.67912pt\oplus\underbrace{({\bf 3^{*},2})_{-1/6,3}^{(+,-)}\oplus u_{R}({\bf 3,1})_{(2/3,-3)}^{(+,+)}\oplus e_{R}({\bf 1,1})_{(-1,3)}^{(+,+)}}_{{\bf 10^{*}}}\\\ \end{array}\right.$ (36) where the numbers written by the bold face in the parenthesis are the representations under $SU(3)_{C}\times SU(2)_{L}$. The numbers written in the subscript denote the charges under $U(1)_{Y}\times U(1)_{X}$. $L(R)$ means the left(right)-handed chirality. Note that the difference between the first and the second ${\bf 6^{*}}$ representations lies in the relative sign of the parity at $y=0$. The corresponding representations of $SU(5)$ are also displayed. It is interesting that the charged lepton doublet $l_{L}$ and the right-handed down quark singlet $d_{R}^{*}$ are separately embedded in different ${\bf 5^{*}}$ representations. Similarly, the quark doublet $q_{L}$ and the right-handed up quark $u_{R}$, electron $e_{R}$ are separately embedded in different ${\bf 10}$ representations. A remarkable fact is that one generation of quarks and leptons (including $\nu_{eR}$) is elegantly embedded as the zero modes of the minimal representations, without introducing any exotic particles. Since the zero mode sector is nothing but the matter content of the standard model (including $\nu_{eR}$), we have no 4D gauge anomalies with respect to the standard model gauge group. As the wave functions of zero modes are $y$-independent and the non-zero KK modes are vector-like, there is no anomalies even in the 5D sense. As for the remaining $U(1)_{X}$, we can easily see that the symmetry is anomalous and is broken at the quantum level. Thus its gauge boson should become heavy and is expected to be decoupled from the low energy sector of the theory [19]. Next, let us study whether we can obtain the correct pattern of electroweak symmetry breaking $SU(2)_{L}\times U(1)_{Y}\to U(1)_{em}$. We will see below that for such purpose the minimal set of matter fields is not sufficient and we need to introduce several massless fermions belonging to the adjoint representation of $SU(6)$. One-loop induced Higgs ($A_{5}$) potential due to the matter fields $N_{{\rm ad}}\times{\bf 35}\oplus 3\times(2\times{\bf 6^{*}}\oplus{\bf 20})$ ($N_{{\rm ad}}$ and 3 denote the number of adjoint fermions and 3 generations, respectively) is calculated as $\displaystyle V(\alpha)$ $\displaystyle=$ $\displaystyle C\left[(4N_{{\rm ad}}-3)\sum_{n=1}^{\infty}\frac{1}{n^{5}}\left\\{\cos(2\pi n\alpha)+2\cos(\pi n\alpha)+6(-1)^{n}\cos(\pi n\alpha)\right\\}\right.$ (37) $\displaystyle\left.+48\sum_{n=1}^{\infty}\frac{1+(-1)^{n}}{n^{5}}\cos(\pi n\alpha)\right]$ where $C\equiv\frac{3}{128\pi^{7}R^{5}}$. The dimensionless parameter $\alpha$ is defined by $\langle A_{5}\rangle\equiv\frac{\alpha}{gR}\frac{\lambda_{27}}{2}$ where $\lambda_{27}$ is the twenty seventh generator of $SU(6)$ possessing the values in the (2,6) component of $6\times 6$ matrix. As can be seen in Fig. 1, if we have no adjoint fermion $N_{{\rm ad}}=0$, the potential is minimized at $\alpha=1$ where the desired electroweak symmetry breaking is not realized, namely $SU(2)\times U(1)\to U(1)\times U(1)$. Figure 1: One-loop Higgs potential with no adjoint fermion, massless fermions of $3\times({\bf 6^{*}}+{\bf 6^{*}})$ and $3\times{\bf 20}$. The potential minimum is located at $\alpha=1$. Figure 2: One-loop Higgs potential with massless fermions of several adjoints, $3\times({\bf 6^{*}}+{\bf 6^{*}}+{\bf 20})$ representations. The plots from the left to the right correspond to the cases with one to four adjoint fermions. Their minimum is located at $\alpha=0.417571,0.307592,0.27334,0.256505$, respectively. On adding the adjoint fermions, we can see from Fig. 2 that the nontrivial minimum appears in the range $0<\alpha<1$ where the desired electroweak symmetry breaking $SU(2)_{L}\times U(1)_{Y}\to U(1)_{em}$ is realized. Note that the value of $\alpha$ at the minimum tends to become smaller, as the number of the adjoint fermions is larger. This feature is useful in order to make a Higgs mass heavy. Higgs mass can be obtained from the second derivative of the potential as $\displaystyle m_{H}^{2}$ $\displaystyle=$ $\displaystyle\left.g^{2}R^{2}\frac{d^{2}V(\alpha)}{d\alpha^{2}}\right|_{\alpha=\alpha_{0}}$ (38) $\displaystyle=$ $\displaystyle\left.-\frac{3g_{4}^{2}M_{W}^{2}}{32\pi^{4}\alpha^{2}}\left[(4N_{{\rm ad}}-3)\sum_{n=1}^{\infty}\frac{1}{n^{3}}\left\\{2\cos(2\pi n\alpha)+\cos(\pi n\alpha)+3(-1)^{n}\cos(\pi n\alpha)\right\\}\right.\right.$ $\displaystyle\left.\left.+24\sum_{n=1}^{\infty}\frac{1+(-1)^{n}}{n^{3}}\cos(\pi n\alpha)\right]\right|_{\alpha=\alpha_{0}}$ where $\alpha_{0}$ denotes the value of $\alpha$ at the minimum of the potential. The relation derived from the gauge-Higgs unification $M_{W}=\alpha_{0}/R$ is used in the last expression. The gauge coupling in four dimensions $g_{4}$ is related to the gauge coupling in five dimensions $g$ through $g_{4}^{2}=\frac{g^{2}}{2\pi R}$. The Higgs masses for several choices of $N_{{\rm ad}}$ are numerically calculated and tabulated in the table below. Adj No. | $\alpha_{0}$ | Higgs mass ---|---|--- 20 | 0.216557 | 113.9 $g_{4}$ GeV 21 | 0.216083 | 116.9 $g_{4}$ GeV We can obtain a viable Higgs mass satisfying the experimental lower bound if more than 20 adjoint fermions are introduced ($g_{4}$ is assumed to be ${\cal O}(1)$). Here we note that this result is just an existence proof not a realistic example for getting a relatively heavy Higgs mass. In our results, the compactification scale is a little bit low, $1/R=M_{W}/\alpha_{0}\simeq 370$ GeV, which contradics with the current experimental data. Therefore, we need further investigations for obtaining a viable Higgs mass with more realsitic situation. As one of the possibilities, it would be interesting to analyze the Higgs potential with only three pairs of fermions in the representations $({\bf 6}^{*}+{\bf 6}^{*}+{\bf 20})$ on the warped space since it has been suggested that the Higgs mass on the warped space is enhanced comparing to the case of flat space [27]. The extension to the case of massive fermion is straightforward. We can incorporate a $Z_{2}$-odd bulk mass of the type $M\epsilon(y)\ (\epsilon(y):\mbox{sign function})$ for fermions. The Higgs potential is then given by [48] $\displaystyle V(\alpha)$ $\displaystyle=$ $\displaystyle C\left[(4N_{{\rm ad}}-3)\sum_{n=1}^{\infty}\frac{1}{n^{5}}\left(1+nz+\frac{1}{3}n^{2}z^{2}\right)e^{-nz}\right.$ (39) $\displaystyle\left.\times\left\\{\cos(2\pi n\alpha)+2\cos(\pi n\alpha)+6(-1)^{n}\cos(\pi n\alpha)\right\\}\right.$ $\displaystyle\left.+48\sum_{n=1}^{\infty}\left(1+nz+\frac{1}{3}n^{2}z^{2}\right)e^{-nz}\frac{1+(-1)^{n}}{n^{5}}\cos(\pi n\alpha)\right]$ where $z\equiv 2\pi RM$ and the bulk masses of fermions are taken to be a common value $M$ for simplicity. We will skip all the detail of the analysis by use of this potential, except reporting that there do not appear any drastic qualitative and quantitative change from the case of $M=0$. The relation $M_{W}=\alpha_{0}/R$ tells us that the compactification scale $1/R$ is not so far from the weak scale $M_{W}$, and therefore the GUT scale also cannot be extremely greater than the weak scale $M_{W}$ (Above the compactification scale, a power-law running of gauge couplings is expected [49]). Thus we have to worry about possible too rapid proton decay. Interestingly, such baryon number violating amplitude concerning KK zero modes turns out to be forbidden at least at the tree level. This is essentially because the quarks and leptons are separated into different representations in our model although they are accommodated in the same representation in ordinary $SU(5)$ GUT. From (36) we learn the type of baryon number (and lepton number) changing vertices of $A_{\mu}$ and $A_{5}$ is limited. Namely concerning fermionic zero-modes, only possibility is $u_{R}\leftrightarrow e_{R}$ due to “colored” 4D gauge boson $A_{\mu}$ with ($SU(3)$, $SU(2)$) quantum number $({\bf 3,1})$ and $q_{L}\leftrightarrow e_{R}$ due to “lepto- quark” 4D scalar $A_{5}$ with $({\bf 3,2})$. Let us note these relevant $A_{\mu}$ and $A_{5}$ are “bosonic partners” of colored Higgs and $X,Y$ gauge bosons in ordinary $SU(5)$ GUT. Though these bosons couple to baryon number changing currents, these interactions do not lead to net baryon number violation, since each of gauge or Higgs boson couples to unique baryon number violating current. In the diagrams where these bosons are exchanged, one vertex with $\Delta B=1/3$ and another vertex with $\Delta B=-1/3$ which is the Hermitian conjugate of the other necessarily appear, thus leading to no net baryon number violation. In other words, we can assign definite baryon (and lepton) number to each boson, and in such a sense baryon number is preserved at each vertex. It will be definitely necessary to consider whether such mechanism to preserve net baryon number is also operative at loop diagrams, though it is not discussed here. In summary, we have investigated a 5D $SU(6)$ grand gauge-Higgs unification model compactified on an orbifold $S^{1}/Z_{2}$, with a realistic matter content. Three generation of quarks and leptons with additional right-handed neutrinos are just embedded as the zero modes of the minimal set of representations, $3\times(2\times{\bf 6^{*}}+{\bf 20})$, without introducing any exotic particles in the same representations. As a remarkable feature of the model, we have found the dangerous proton decay is forbidden at the tree level. We have also found that the desired pattern of electroweak symmetry breaking is dynamically realized by introducing suitable number of adjoint fermions. Higgs mass was also calculated and shown to become heavy, if certain number of the adjoint fermions are introduced. Searching for a simpler matter content yielding a reasonable Higgs mass is desirable and a very nontrivial task. This is left for a future work. There are still many issues to be studied. The construction of the realistic hierarchy of Yukawa couplings is a fundamental problem in the gauge-Higgs unification since Yukawa coupling is naively the gauge coupling which is flavor independent. One of the promising proposals to avoid the problem [18] is that the mixing between the bulk massive fermions and the brane localized quarks and leptons generates non-local Yukawa couplings after integrating out the bulk massive fermions. The huge hierarchy is then realized by an order one tuning of the bulk mass. It is very important to study whether this proposal can be incorporated into the present model. To study the energy evolution of gauge couplings and their unification is another important issue. These issues will be discussed elsewhere. Acknowledgments The author would like to thank the organizers of the workshop for providing me with an oppotunity to talk at such a nice workshop. The work of the author was supported in part by the Grant-in-Aid for Scientific Research of the Ministry of Education, Science and Culture, No.18204024. ## References * [1] C.S. Lim and N. Maru, _Phys. Lett._ B653, 320 (2007). * [2] N. S. Manton, _Nucl. Phys._ B158, 141 (1979). * [3] D. B. Fairlie, _Phys. Lett._ B82, 97 (1979); _J. Phys._ G 5, L55 (1979). * [4] Y. Hosotani, _Phys. Lett._ B126, 309 (1983); _Phys. Lett._ B129, 193 (1983); _Annals Phys._ 190, 233 (1989). * [5] H. Hatanaka, T. Inami and C. S. Lim, _Mod. Phys. Lett._ A13, 2601 (1998); * [6] I. Antoniadis, K. Benakli and M. Quiros, _New J. Phys._ 3, 20 (2001). * [7] G. von Gersdorff, N. Irges and M. Quiros, _Nucl. Phys._ B635, 127 (2002). * [8] The first reference in [26]. * [9] K. Hasegawa, C. S. Lim and N. Maru, _Phys. Lett._ B604, 133 (2004). * [10] C. S. Lim, N. Maru and K. Hasegawa, arXiv:hep-th/0605180. * [11] N. Maru and T. Yamashita, _Nucl. Phys._ B754, 127 (2006); Y. Hosotani, N. Maru, K. Takenaga and T. Yamashita, _Prog. Theor. Phys._ 118, 1053 (2007). * [12] C. S. Lim and N. Maru, _Phys. Rev._ D75, 115011 (2007). * [13] Y. Adachi, C. S. Lim and N. Maru, _Phys. Rev._ D76, 075009 (2007) * [14] A. Falkowski, arXiv:0711.0828 [hep-ph]. * [15] N. Maru and N. Okada, arXiv:0711.2589 [hep-ph](To appear in PRD). * [16] I. Low, talk at Eötövs-Cornell 2007 Summer Workshop on Particle Theory, http://www.lns.cornell.edu/~eotcor07/budapest_ianlow.pdf; I. Low and R. Rattazzi, to appear. * [17] M. Kubo, C. S. Lim and H. Yamashita, _Mod. Phys. Lett._ A17, 2249 (2002). * [18] C. Csáki, C. Grojean and H. Murayama, _Phys. Rev._ D67, 085012 (2003). * [19] C. A. Scrucca, M. Serone and L. Silvestrini, _Nucl. Phys._ B669, 128 (2003). * [20] I. Gogoladze, Y. Mimura and S. Nandi, _Phys. Lett._ B560, 204 (2003); _Phys. Rev._ D72, 055006 (2005). * [21] N. Haba, Y. Hosotani, Y. Kawamura and T. Yamashita, _Phys. Rev._ D70, 015010 (2004). * [22] Y. Hosotani, S. Noda and K. Takenaga, _Phys. Rev._ D69, 125014 (2004); _Phys. Lett._ B607, 276 (2005). * [23] C. Biggio and M. Quiros, _Nucl. Phys._ B703, 199 (2004). * [24] G. Martinelli, M. Salvatori, C. A. Scrucca and L. Silvestrini, _JHEP_ 0510, 037 (2005). * [25] K. y. Oda and A. Weiler, _Phys. Lett._ B606, 408 (2005). * [26] R. Contino, Y. Nomura and A. Pomarol, _Nucl. Phys._ B671, 148 (2003); K. Agashe, R. Contino and A. Pomarol, _Nucl. Phys._ B719, 165 (2005); K. Agashe and R. Contino, _Nucl. Phys._ B742, 59 (2006); K. Agashe, R. Contino, L. Da Rold and A. Pomarol, _Phys. Lett._ B641, 62 (2006); R. Contino, L. Da Rold and A. Pomarol, _Phys. Rev._ D75, 055014 (2007). * [27] Y. Hosotani and M. Mabe, _Phys. Lett._ B615, 257 (2005); Y. Hosotani, S. Noda, Y. Sakamura and S. Shimasaki, _Phys. Rev._ D73, 096006 (2006); Y. Sakamura and Y. Hosotani, _Phys. Lett._ B645, 442 (2007); Y. Hosotani and Y. Sakamura, _Prog. Theor. Phys._ 118, 935 (2007); Y. Sakamura, _Phys. Rev._ D76, 065002 (2007). * [28] M. S. Carena, E. Ponton, J. Santiago and C. E. M. Wagner, _Phys. Rev._ D76, 035006 (2007); A. D. Medina, N. R. Shah and C. E. M. Wagner, _Phys. Rev._ D76, 095010 (2007); M. Carena, A. D. Medina, B. Panes, N. R. Shah and C. E. M. Wagner, arXiv:0712.0095 [hep-ph]. * [29] G. Panico and A. Wulzer, _JHEP_ 0705, 060 (2007). * [30] G. Panico and M. Serone, _JHEP_ 0505, 024 (2005); G. Panico, M. Serone and A. Wulzer, _Nucl. Phys._ B739, 186 (2006); _Nucl. Phys._ B762, 189 (2007). * [31] G. Cacciapaglia, C. Csaki and S. C. Park, _JHEP_ 0603, 099 (2006); * [32] N. Haba, S. Matsumoto, N. Okada and T. Yamashita, _JHEP_ 0602, 073 (2006); I. Gogoladze, N. Okada and Q. Shafi, _Phys. Lett._ B655, 257 (2007); _Phys. Lett._ B659, 316 (2008). * [33] N. Maru and K. Takenaga, _Phys. Rev._ 72, 046003 (2005); _Phys. Lett._ B637, 287 (2006); _Phys. Rev._ D74, 015017 (2006). * [34] Y. Hosotani, arXiv:hep-ph/0607064. * [35] M. Sakamoto and K. Takenaga, _Phys. Rev._ D75, 045015 (2007). * [36] K. Inoue, A. Kakuto and H. Takano, _Prog. Theor. Phys._ 75, 664 (1986). * [37] A. A. Anselm and A. A. Johansen, _Phys. Lett._ B200, 331 (1988); A. A. Anselm, _Sov. Phys. JETP_ 67 (1988) 663 [_Zh. Eksp. Teor. Fiz._ 94 (1988) 26]. * [38] Z. G. Berezhiani and G. R. Dvali, _Bull. Lebedev Phys. Inst._ 5 (1989) 55 [_Kratk. Soobshch. Fiz._ 5 (1989) 42]. * [39] R. Barbieri, G. R. Dvali and A. Strumia, _Nucl. Phys._ B391, 487 (1993); R. Barbieri, G. R. Dvali and M. Moretti, _Phys. Lett._ B312, 137 (1993). * [40] R. Barbieri, G. R. Dvali, A. Strumia, Z. Berezhiani and L. J. Hall, _Nucl. Phys._ B432, 49 (1994). * [41] Z. Berezhiani, C. Csaki and L. Randall, _Nucl. Phys._ B444, 61 (1995). * [42] Z. Berezhiani, _Phys. Lett._ B355, 481 (1995). * [43] N. Arkani-Hamed, A.G. Cohen and H. Georgi, _Phys. Lett._ B513, 232 (2001); N. Arkani-Hamed, A.G. Cohen, E. Katz and A.E. Nelson, _JHEP_ 0207, 034 (2002); M. Schmaltz and D. Tucker-Smith, _Ann. Rev. Nucl. Part. Sci._ 55 229 (2005); M. Perelstein, hep-ph/0512128. * [44] J. M. Maldacena, _Adv. Theor. Math. Phys._ 2, 231 (1998) [_Int. J. Theor. Phys._ 38, 1113 (1999)]; S. S. Gubser, I. R. Klebanov and A. M. Polyakov, _Phys. Lett._ B428, 105 (1998); E. Witten, _Adv. Theor. Math. Phys._ 2, 253 (1998). * [45] N. Arkani-Hamed, M. Porrati and L. Randall, _JHEP_ 0108, 017 (2001); R. Rattazzi and A. Zaffaroni, _JHEP_ 0104, 021 (2001). * [46] G. Burdman and Y. Nomura, _Nucl. Phys._ B656, 3 (2003). * [47] Y. Kawamura, _Prog. Theor. Phys._ 105, 999 (2001); _Prog. Theor. Phys._ 105, 691 (2001). * [48] The second reference in [33]. * [49] K. R. Dienes, E. Dudas and T. Gherghetta, _Phys. Lett._ B436, 55 (1998).

arxiv-papers

2024-09-04T02:48:53.811329

0802.2385

# Essential variables and positions in terms Slavcho Shtrakov http://home.swu.bg/shtrakov ###### Abstract. The paper deals with $\Sigma-$composition of terms, which allows us to extend the derivation rules in formal deduction of identities. The concept of essential variables and essential positions of terms with respect to a set of identities is a key step in the simplification of the process of formal deduction. $\Sigma-$composition of terms is defined as replacement between $\Sigma$-equal terms. This composition induces $\Sigma R-$deductively closed sets of identities. In analogy to balanced identities we introduce and investigate $\Sigma-$balanced identities for a given set of identities $\Sigma$. ###### Key words and phrases: Composition of terms, Essential position in a term, Globally invariant congruence, Stable variety. ###### 2000 Mathematics Subject Classification: Primary: 08B05; Secondary: 08A02, 03C05, 08B15 ## 1\. Introduction Let ${\mathcal{F}}$ be any finite set, the elements of which are called $operation\ symbols.$ Let $\tau:{\mathcal{F}}\to N$ be a mapping into the non- negative integers; for $f\in{\mathcal{F}},$ the number $\tau(f)$ will denote the _arity_ of the operation symbol $f.$ The pair $(\mathcal{F},\tau)$ is called a _type_ or _signature_. If it is obvious what the set ${\mathcal{F}}$ is, we will write “$type\ \tau$”. The set of symbols of arity $p$ is denoted by ${\mathcal{F}}_{p}.$ Let $X$ be a finite set of variables, and let $\tau$ be a type with the set of operation symbols ${\mathcal{F}}=\cup_{j\geq 0}{\mathcal{F}}_{j}.$ The set $W_{\tau}(X)$ of _terms of type $\tau$ _ with variables from $X$ is the smallest set such that 1. (i) $X\cup\mathcal{F}_{0}\subseteq W_{\tau}(X)$; 2. (ii) if $f$ is an $n-$ary operation symbol and $t_{1},\ldots,t_{n}$ are terms, then the “string” $f(t_{1}\ldots t_{n})$ is a term. An algebra ${\mathcal{A}}=\langle A;\mathcal{F}^{\mathcal{A}}\rangle$ of type $\tau$ is a pair consisting of a set $A$ and an indexed set $\mathcal{F}^{\mathcal{A}}$ of operations, defined on $A$. If $f\in\mathcal{F}$, then $f^{\mathcal{A}}$ denotes a $\tau(f)$-ary operation on the set $A.$ We denote by ${\mathcal{A}}lg(\tau)$ the class of all algebras of type $\tau$. If $s,t\in W_{\tau}(X)$, then the pair $s\approx t$ is called an identity of type $\tau$. $Id(\tau)$ denotes the set of all identities of type $\tau.$ An identity $t\approx s\in Id(\tau)$ is satisfied in the algebra ${\mathcal{A}}$, if the term operations $t^{\mathcal{A}}$ and $s^{\mathcal{A}}$, induced by the terms $t$ and $s$ on the algebra ${\mathcal{A}}$ are equal, i.e., $t^{\mathcal{A}}=s^{\mathcal{A}}.$ In this case we write ${\mathcal{A}}\models t\approx s$ and if $\Sigma$ is a set of identities of type $\tau$, then ${\mathcal{A}}\models\Sigma$ means that ${\mathcal{A}}\models t\approx s$ for all $t\approx s\in\Sigma$. Let $\Sigma$ be a set of identities. For $t,s\in W_{\tau}(X)$ we write $\Sigma\models t\approx s$ if, given any algebra $\mathcal{A}$, $\mathcal{A}\models\Sigma\ \Rightarrow\ \mathcal{A}\models t\approx s.$ The operators $Id$ and $Mod$ are defined for classes of algebras $K\subseteq\mathcal{A}lg(\tau)$ and for sets of identities $\Sigma\subseteq Id(\tau)$ as follows $\displaystyle Id(K)$ $\displaystyle:=\\{t\approx s\mid{\mathcal{A}}\in K~{}\Rightarrow{\mathcal{A}}\models t\approx s\\},\ \text{ and}$ $\displaystyle Mod(\Sigma)$ $\displaystyle:=\\{{\mathcal{A}}\mid\ t\approx s\in\Sigma~{}\Rightarrow{\mathcal{A}}\models t\approx s\\}.$ The set of fixed points with respect to the closure operators $IdMod$ and $ModId$ form complete lattices ${\mathcal{L}}(\tau)$ and ${\mathcal{E}}(\tau)$ of all varieties of type $\tau$ and of all equational theories (logics) of type $\tau$. In [1] deductive closures of sets of identities are used to describe some elements of these lattices. We will apply the concept of $\Sigma-$compositions of terms to study the lattices ${\mathcal{L}}(\tau)$ and ${\mathcal{E}}(\tau)$. We use the concept of essential variables, as defined in [5] and therefore we consider such variables with respect to a given set of identities, which is a fully invariant congruence. In Section 2 we investigate the concept of $\Sigma-$essential variables and positions. The fictive (non-essential) variables and positions are used to simplify the deductions of identities in equational theories. We introduce $\Sigma-$composition of terms for a given set $\Sigma$ of identities. In Section 3 we describe the closure operator $\Sigma R$ in the set of all identities of a given type, which generate extensions of fully invariant congruences. The varieties which satisfy $\Sigma R-$closed sets are fully invariant congruences and they are called stable. The stable varieties are compared to solid ones [2, 4, 6]. In Section 4 we introduce and study $\Sigma-$balanced identities and prove that $\Sigma-$balanced property is closed under $\Sigma R$-deductions. ## 2\. Compositions of terms If $t$ is a term, then the set $var(t)$ consisting of those elements of $X$ which occur in $t$ is called the set of _input variables (or variables)_ for $t$. If $t=f(t_{1},\ldots,t_{n})$ is a non-variable term, then $f$ is the _root symbol (root)_ of $t$ and we will write $f=root(t).$ For a term $t\in W_{\tau}(X)$ the set $Sub(t)$ of its subterms is defined as follows: if $t\in X\cup\mathcal{F}_{0}$, then $Sub(t)=\\{t\\}$ and if $t=f(t_{1},\ldots,t_{n})$, then $Sub(t)=\\{t\\}\cup Sub(t_{1})\cup\ldots\cup Sub(t_{n}).$ The $depth$ of a term $t$ is defined inductively: if $t\in X\cup\mathcal{F}_{0}$ then $Depth(t)=0;$ and if $t=f(t_{1},\ldots,t_{n})$, then $Depth(t)=max\\{Depth(t_{1}),\ldots,Depth(t_{n})\\}+1.$ ###### Definition 2.1. Let $r,s,t\in W_{\tau}(X)$ be three terms of type $\tau$. By $t(r\leftarrow s)$ we will denote the term, obtained by simultaneous replacement of every occurrence of $r$ as a subterm of $t$ by $s$. This term is called the _inductive composition_ of the terms $t$ and $s$, by $r$. In particular, 1. (i) $t(r\leftarrow s)=t$ if $r\notin Sub(t)$; 2. (ii) $t(r\leftarrow s)=s$ if $t=r$, and 3. (iii) $t(r\leftarrow s)=f(t_{1}(r\leftarrow s),\ldots,t_{n}(r\leftarrow s))$, if $t=f(t_{1},\ldots,t_{n})$ and $r\in Sub(t)$, $r\neq t$. If $r_{i}\notin Sub(r_{j})$ when $i\neq j$, then $t(r_{1}\leftarrow s_{1},\ldots,r_{m}\leftarrow s_{m})$ means the inductive composition of $t,r_{1},\ldots,r_{m}$ by $s_{1},\ldots,s_{m}$. In the particular case when $r_{j}=x_{j}$ for $j=1,\ldots,m$ and $var(t)=\\{x_{1},\ldots,x_{m}\\}$ we will briefly write $t(s_{1},\ldots,s_{m})$ instead of $t(x_{1}\leftarrow s_{1},\ldots,x_{m}\leftarrow s_{m})$. Any term can be regarded as a tree with nodes labelled as the operation symbols and its leaves labelled as variables or nullary operation symbols. Often the tree of a term is presented by a diagram of the corresponding term as it is shown by Figure 1. Let $\tau$ be a type and $\mathcal{F}$ be its set of operation symbols. Denote by $maxar=\max\\{\tau(f)\mid f\in\mathcal{F}\\}$ and $N_{\mathcal{F}}:=\\{m\in N\mid m\leq maxar\\}$. Let $N_{\mathcal{F}}^{*}$ be the set of all finite strings over $N_{\mathcal{F}}.$ The set $N_{\mathcal{F}}^{*}$ is naturally ordered by $p\preceq q\iff p$ is a prefix of $q.$ The Greek letter $\varepsilon$, as usual denotes the empty word (string) over $N_{\mathcal{F}}.$ To distinguish between different occurrences of the same operation symbol in a term $t$ we assign to each operation symbol a position, i.e., an element of a given set. Usually positions are finite sequences (strings) of natural numbers. Each position is assigned to a node of the tree diagram of $t$, starting with the empty sequence $\varepsilon$ for the root and using the integers $j$, $1\leq j\leq n_{i}$ for the $j$-th branch of an $n_{i}$-ary operational symbol $f_{i}$. So, let the position $p=a_{1}a_{2}\ldots a_{s}\in N^{*}_{\mathcal{F}}$ be assigned to a node of $t$ labelled by the $n_{i}$-ary operational symbol $f_{i}$. Then the position assigned to the $j$-th child of this node is $a_{1}a_{2}\ldots a_{s}j$. The set of positions of a term $t$ is denoted by $Pos(t)$ and it is illustrated by Example 2.1. Thus we have $Pos(t)\subseteq N_{\mathcal{F}}^{*}$. Let $t\in W_{\tau}(X)$ be a term of type $\tau$ and let $sub_{t}:Pos(t)\to Sub(t)$ be the function which maps each position in a term $t$ to the subterm of $t$, whose root node occurs at that position. ###### Definition 2.2. Let $t,r\in W_{\tau}(X)$ be two terms of type $\tau$ and $p\in Pos(t)$ be a position in $t.$ The positional composition of $t$ and $r$ on $p$ is the term $s:=t(p;r)$ obtained from $t$ by replacing the term $sub_{t}(p)$ by $r$ on the position $p$, only. ###### Example 2.1. Let $\tau=(2)$, $t=f(f(x_{1},f(f(f(x_{1},x_{2}),x_{2}),x_{3})),x_{4})$ and $u=f(x_{4},x_{1})$. The positions of $t$ and $u$ are written on their nodes in Figure 1. Then the positional composition of $t$ and $u$ on the position $121\in Pos(t)$ is $t(121;u)=f(f(x_{1},f(f(x_{4},x_{1}),x_{3})),x_{4})$ and $sub_{t}(121)=f(f(x_{1},x_{2}),x_{2})$. $12$$12$$121$$121$$\varepsilon$$1211$$1211$$1$$1$$1$$x_{1}$$x_{1}$$11$$11$$122$$122$$1212$$1212$$2$$12112$$x_{3}$$x_{3}$$x_{2}$$x_{1}$$x_{1}$$x_{2}$$\varepsilon$$\varepsilon$$2$$2$$x_{4}$$x_{4}$$12111$$x_{1}$$x_{4}$$x_{4}$$t$$u$$t(121;u)$ Figure 1. Positional composition of terms ###### Remark 2.1. The positional composition has the following properties: 1. 1. If $\langle\langle p_{1},p_{2}\rangle,\langle t_{1},t_{2}\rangle\rangle$ is a pair with $p_{1}\not\prec p_{2}\ \&\ p_{2}\not\prec p_{1}$, then $t(p_{1},p_{2};t_{1},t_{2})=t(p_{1};t_{1})(p_{2};t_{2})=t(p_{2};t_{2})(p_{1};t_{1});$ 2. 2. If $S=\langle p_{1},\ldots,p_{m}\rangle$ and $T=\langle t_{1},\ldots,t_{m}\rangle$ with $(\forall p_{i},p_{j}\in S)\ (i\neq j\Rightarrow p_{i}\not\prec p_{j}\ \&\ p_{j}\not\prec p_{i})$ and $\pi$ is a permutation of the set $\\{1,\ldots,m\\}$, then $t(p_{1},\ldots,p_{m};t_{1},\ldots,t_{m})=t(p_{\pi(1)},\ldots,p_{\pi(m)};t_{\pi(1)},\ldots,t_{\pi(m)}).$ 3. 3. If $t,s,r\in W_{\tau}(X)$, $p\in Pos(t)$ and $q\in Pos(s)$, then $t(p;s(q;r))=t(p;s)(pq;r)$. 4. 4. Let $s,t\in W_{\tau}(X)$ and $r\in Sub(t)$ be terms of type $\tau$. Let $\\{p_{1},\ldots,p_{m}\\}=\\{p\in Pos(t)\mid sub_{t}(p)=r\\}$. Then we have $t(p_{1},\ldots,p_{m};s):=t(p_{1};s)(p_{2};s),\ldots,(p_{m};s)=t(r\leftarrow s),$ which shows that any inductive composition can be represented as a positional one. On the other side there are examples of positional compositions which can not be realized as inductive compositions. ###### Definition 2.3. Let $\Sigma\subseteq Id(\tau)$, $t\in W_{\tau}(X_{n})$ be an $n-$ary term of type $\tau$, $\mathcal{A}=\langle A,\mathcal{F}\rangle$ be an algebra of type $\tau$ and let $x_{i}\in var(t)$ be a variable which occurs in $t.$ (i) [5] The variable $x_{i}$ is called _essential_ for $t$ with respect to the algebra $\mathcal{A}$ if there are $n+1$ elements $a_{1},\ldots,a_{i-1},a,b,a_{i+1},\ldots,a_{n}\in A$ such that $t^{\mathcal{A}}(a_{1},\ldots,a_{i-1},a,a_{i+1},\ldots,a_{n})\neq t^{\mathcal{A}}(a_{1},\ldots,a_{i-1},b,a_{i+1},\ldots,a_{n}).$ The set of all essential variables for $t$ with respect to $\mathcal{A}$ will be denoted by $Ess(t,\mathcal{A})$. $Fic(t,\mathcal{A})$ denotes the set of all variables in $var(t)$, which are not essential with respect to $\mathcal{A}$, called fictive ones. (ii) A variable $x_{i}$ is said to be _ $\Sigma-$essential_ for a term $t$ if there is an algebra $\mathcal{A}$, such that $\mathcal{A}\models\Sigma$ and $x_{i}\in Ess(t,\mathcal{A}).$ The set of all $\Sigma-$essential variables for $t$ will be denoted by $Ess(t,\Sigma).$ If a variable is not $\Sigma-$essential for $t$, then it is called _$\Sigma-$ fictive_ for $t$. $Fic(t,\Sigma)$ denotes the set of all $\Sigma-$fictive variables for $t.$ ###### Proposition 2.1. If $\Sigma_{1}\subseteq\Sigma_{2}\subseteq Id(\tau)$, $t\in W_{\tau}(X)$ and $x_{i}\in Ess(t,\Sigma_{2})$, then $x_{i}\in Ess(t,\Sigma_{1})$. ###### Theorem 2.1. Let $t\in W_{\tau}(X)$ and $\Sigma\subseteq Id(\tau)$. A variable $x_{i}$ is $\Sigma-$essential for $t$ if and only if there is a term $r$ of type $\tau$ such that $r\neq x_{i}$ and $\mathcal{A}\not\models t\approx t(x_{i}\leftarrow r)$ for some algebra $\mathcal{A}\in\mathcal{A}lg(\tau)$ with $\mathcal{A}\models\Sigma.$ ###### Proof. Let $t\in W_{\tau}(X_{n})$ for some $n\in N$ and let $\mathcal{A}\in\mathcal{A}lg(\tau)$ be an algebra for which $\mathcal{A}\models\Sigma$ and $x_{i}\in Ess(t,\mathcal{A})$. Then from Lemma 3.5 of [5] it follows that $\mathcal{A}\not\models t\approx t(x_{i}\leftarrow x_{n+1}).$ Hence $\mathcal{A}\not\models t\approx t(x_{i}\leftarrow r)$ with $r=x_{n+1}$. Conversely, let us assume that there is a term $r$, $r\neq x_{i}$ of type $\tau$ with $\mathcal{A}\not\models t\approx t(x_{i}\leftarrow r)$ for an algebra $\mathcal{A}\in\mathcal{A}lg(\tau)$ with $\mathcal{A}\models\Sigma.$ Let $m\in N$ be a natural number for which $r\in W_{\tau}(X_{m})$. So, there are $m+n$ values $a_{1},\ldots,a_{i-1},a_{i},a_{i+1},\ldots,a_{n},b_{1},\ldots,b_{m}\in A$ such that $r^{\mathcal{A}}(b_{1},\ldots,b_{m})\neq a_{i}$ and $t^{\mathcal{A}}(a_{1},\ldots,a_{i-1},a_{i},a_{i+1},\ldots,a_{n})\neq t^{\mathcal{A}}(a_{1},\ldots,a_{i-1},r^{\mathcal{A}}(b_{1},\ldots,b_{m}),a_{i+1},\ldots,a_{n}).$ The last inequality shows that $x_{i}\in Ess(t,\mathcal{A}).$ Hence $x_{i}$ is $\Sigma-$essential for $t$.∎ ###### Corollary 2.1. If $t\approx s\in\Sigma$ and $x_{i}\in Fic(t,\Sigma)$, then for each term $r\in W_{\tau}(X)$, we have $\Sigma\models t(x_{i}\leftarrow r)\approx s.$ ###### Corollary 2.2. A variable $x_{i}$ is $\Sigma-$essential for $t\in W_{\tau}(X_{n})$ if and only if $x_{i}$ is essential for $t$ with respect to any $Mod(\Sigma)$-free algebra with at least $n+1$ free generators. ###### Corollary 2.3. Let $\Sigma\subseteq Id(\tau)$ be a set of identities of type $\tau$ and $t\approx s\in\Sigma$. If a variable $x_{i}$ is $\Sigma-$fictive for $t$, then it is fictive for $s$ with respect to each algebra $\mathcal{A}\in Mod(\Sigma)$. The concept of $\Sigma-$essential positions is a natural extension of $\Sigma-$essential variables. ###### Definition 2.4. Let $\mathcal{A}=\langle A,\mathcal{F}\rangle$ be an algebra of type $\tau$, $t\in W_{\tau}(X_{n})$, and let $p\in Pos(t)$. (i) If $x_{n+1}\in Ess(t(p;x_{n+1}),\mathcal{A})$, then the position $p\in Pos(t)$ is called _essential_ for $t$ with respect to the algebra $\mathcal{A}$. The set of all essential positions for $t$ with respect to $\mathcal{A}$ is denoted by $PEss(t,\mathcal{A}).$ When a position $p\in Pos(t)$ is not essential for $t$ with respect to $\mathcal{A}$, it is called _fictive_ for $t$ with respect to $\mathcal{A}$. The set of all fictive positions with respect to $\mathcal{A}$ is denoted by $PFic(t,\mathcal{A}).$ (ii) If $x_{n+1}\in Ess(t(p;x_{n+1}),\Sigma)$, then the position $p\in Pos(t)$ is called _$\Sigma-$ essential_ for $t$. The set of $\Sigma-$essential positions for $t$ is denoted by $PEss(t,\Sigma).$ When a position is not $\Sigma-$essential for $t$ it is called _ $\Sigma-$fictive_. $PFic(t,\Sigma)$ denotes the set of all $\Sigma-$fictive positions for $t.$ The set of $\Sigma$-essential subterms of $t$ is defined as follows: $SEss(t,\Sigma):=\\\ \\{sub_{t}(p)\mid$ $\ p\in PEss(t,\Sigma)\\}$. $SFic(t,\Sigma)$ denotes the set $SFic(t,\Sigma):=Sub(t)\setminus SEss(t,\Sigma)$. So, $\Sigma$-essential subterms of a term are subterms which occur at a $\Sigma$-essential position. Since one subterm can occur at more than one position in a term, and can occur in both $\Sigma$-essential and non-$\Sigma$-essential positions, we note that a subterm is $\Sigma$-essential if it occurs at least once in a $\Sigma$-essential position, and $\Sigma$-fictive otherwise. ###### Example 2.2. Let $\tau=(2)$ and let $t=f(f(x_{1},x_{2}),f(f(x_{1},x_{2}),x_{3}))$. Let us consider the variety $RB=Mod(\Sigma)$ of rectangular bands, where $~{}\Sigma=\\{f(x_{1},f(x_{2},x_{3}))\approx f(f(x_{1},x_{2}),x_{3})\approx f(x_{1},x_{3}),\ f(x_{1},x_{1})\approx x_{1}\\}.$ It is not difficult to see that the $\Sigma-$essential positions and subterms of $t$ are $PEss(t,\Sigma)=\\{\varepsilon,1,11,2,22\\},$ --- $SEss(t,\Sigma)=\\{t,f(x_{1},x_{2}),x_{1},f(f(x_{1},x_{2}),x_{3}),x_{3}\\}$ $PFic(t,\Sigma)=Pos(t)\setminus PEss(t,\Sigma)=\\{12,21,211,212\\},$ $SFic(t,\Sigma)=\\{x_{2}\\}.$ The $\Sigma-$essential and $\Sigma-$fictive positions of $t$ are represented by large and small black circles, respectively in Figure 2. Note that $|PFic(t,\Sigma)|>|SFic(t,\Sigma)|$. This is because there is one subterm, $f(x_{1},x_{2})$, which occurs more than once, once each in an essential and non-essential position, so that $|Pos(t)|>|Sub(t)|$. $t$$1$$f$21$f$$x_{3}$211$x_{2}$$x_{2}$$x_{1}$1112$x_{1}$212$f$$f$222$\varepsilon$ Figure 2. $\Sigma-$essential positions for $t$ from Example 2.2. ###### Theorem 2.2. If $p\in PEss(t,\Sigma)$, then each position $q\in Pos(t)$ with $q\preceq p$ is $\Sigma-$essential for $t$. ###### Proof. Let $sub_{t}(q)=s$ and $sub_{t}(p)=r$. Now, $q\preceq p$ implies that $r\in Sub(s)$ and $Sub(r)\subset Sub(s).$ Let $n$ be a natural number such that $t\in W_{\tau}(X_{n}).$ From $p\in PEss(t,\Sigma)$ it follows that there is a term $v\in W_{\tau}(X)$ for which $v\neq x_{n+1}$ and $\Sigma\not\models t\approx t(p;x_{n+1})(x_{n+1}\leftarrow v),\quad\mbox{i.e.,}\quad\Sigma\not\models t\approx t(p;v).$ Consequently, there is an algebra $\mathcal{A}=\langle A,\mathcal{F}\rangle$ of type $\tau$ such that $\mathcal{A}\models\Sigma\quad\mbox{and}\quad t^{\mathcal{A}}\neq t(p;v)^{\mathcal{A}}.$ Let $m$ be a natural number such that $t\in W_{\tau}(X_{m})$ and $v\in W_{\tau}(X_{m})$. Let $(a_{1},\ldots,a_{m})\in A^{m}$ be a tuple such that $t^{\mathcal{A}}(a_{1},\ldots,a_{n})\neq t(p;v)^{\mathcal{A}}(a_{1},\ldots,a_{m}).$ Let $u\in W_{\tau}(X_{m})$ be the term $u=s(q^{\prime};v),$ where $p=q\circ q^{\prime}$ and $q^{\prime}\in Pos(s).$ Hence we have $\Sigma\models t(p;v)\approx t(q;u)$ and $t^{\mathcal{A}}(a_{1},\ldots,a_{n})\neq t(q;u)^{\mathcal{A}}(a_{1},\ldots,a_{m}).$ Consequently $t^{\mathcal{A}}\neq t(q;u)^{\mathcal{A}}$, i.e., $\Sigma\not\models t\approx t(q;x_{n+1})(x_{n+1}\leftarrow u)$ and $q\in PEss(t,\Sigma)$. ∎ ###### Corollary 2.4. If $q\in PFic(t,\Sigma)$, then each position $p\in Pos(t)$ with $q\preceq p$ is $\Sigma-$fictive for $t$. ###### Theorem 2.3. Let $t\in W_{\tau}(X)$ be a term of type $\tau$ and let $\Sigma\subseteq Id(\tau)$ be a set of identities of type $\tau.$ If $p\in PFic(t,\Sigma)$, then $\Sigma\models t\approx t(p;v),$ for each term $v\in W_{\tau}(X).$ ###### Proof. Let $p\in PFic(t,\Sigma)$ and let us suppose that the theorem is false. Then there is a term $v\in W_{\tau}(X)$ with $v\neq sub_{t}(p)$, such that $\Sigma\not\models t\approx t(p;v).$ Let $sub_{t}(p)=r$ and let $n$ be a natural number, such that $v,t\in W_{\tau}(X_{n})$. Then $t(p;v)=t(p;x_{n+1})(x_{n+1}\leftarrow v)\quad\mbox{ and }\quad t=t(p;r)=t(p;x_{n+1})(x_{n+1}\leftarrow r).$ Our supposition shows that $\Sigma\not\models t\approx t(p;v)\ \iff\ \Sigma\not\models t(p;x_{n+1})(x_{n+1}\leftarrow r)\approx t(p;x_{n+1})(x_{n+1}\leftarrow v).$ Hence there is an algebra $\mathcal{A}=\langle A,\mathcal{F}\rangle$ and $n+2$ elements $a_{1},\ldots,a_{n},a,b$ of $A$ such that $(t(p;x_{n+1}))^{\mathcal{A}}(a_{1},\ldots,a_{n},a)\neq(t(p;x_{n+1}))^{\mathcal{A}}(a_{1},\ldots,a_{n},b),$ where $a=r^{\mathcal{A}}(a_{1},\ldots,a_{n})$ and $b=v^{\mathcal{A}}(a_{1},\ldots,a_{n}).$ This means that $x_{n+1}\in Ess(t(p;x_{n+1}),\Sigma)$. Hence $p\in PEss(t,\Sigma)$, which is a contradiction. ∎ ###### Corollary 2.5. If $p\in Pos(t)$ is a $\Sigma-$fictive position for $t$, then $p$ is fictive for $t$ with respect to each algebra $\mathcal{A}$ with $\mathcal{A}\models\Sigma.$ ###### Corollary 2.6. If $p\in PEss(t,\Sigma)$, $t\in W_{\tau}(X_{n})$, then $p$ is essential for $t$ with respect to each $Mod(\Sigma)-$free algebra with at least $n+1$ free generators. If $\Sigma\models t\approx s$ and $s\in Sub(t)$ is a proper subterm of $t$, one might expect that the positions of $t$ which are “outside” of $s$ have to be $\Sigma-$fictive. To see that this is not true, we consider the set of operations $\vee,\wedge$ and $\neg$ with type $\tau:=(2,2,1)$. Let $\Sigma$ be the set of identities satisfied in a Boolean algebra. Then it is easy to prove that if $t=x_{1}\wedge(x_{2}\vee\neg(x_{2}))$, then we have $\Sigma\models t\approx x_{1}$, but $PEss(t,\Sigma)=Pos(t).$ Now, we are going to generalize composition of terms and to describe the corresponding deductive systems. Let $\Sigma$ be a set of identities of type $\tau$. Two terms $t$ and $s$ are called _$\Sigma$ -equivalent_ (briefly, _$\Sigma$ -equal_) if $\Sigma\models t\approx s$. ###### Definition 2.5. Let $t,r,s\in W_{\tau}(X)$ and $\Sigma S_{r}^{t}=\\{v\in Sub(t)\mid\Sigma\models r\approx v\\}$ be the set of all subterms of $t$ which are $\Sigma-$equal to $r$. Term $\Sigma-$composition of $t$ and $r$ by $s$ is defined as follows 1. (i) $t^{\Sigma}(r\leftarrow s)=t$ if $\Sigma S_{r}^{t}=\emptyset$; 2. (ii) $t^{\Sigma}(r\leftarrow s)=s$ if $\Sigma\models t\approx r$, and 3. (iii) $t^{\Sigma}(r\leftarrow s)=f(t_{1}^{\Sigma}(r\leftarrow s),\ldots,t_{n}^{\Sigma}(r\leftarrow s)),$ if $t=f(t_{1},\ldots,t_{n})$. Let $\Sigma P_{r}^{t}=\\{p\in Pos(t)\mid sub_{t}(p)\in\Sigma S_{r}^{t}\\}$ be the set of all positions of subterms of $t$ which are $\Sigma-$equal to $r$. Let $P_{r}^{t}=\\{p_{1},\ldots,p_{m}\\}$ be the set of all the minimal elements in $\Sigma P_{r}^{t}$ with respect to the ordering $\prec$ in the set of positions, i.e., $p\in P_{r}^{t}$ if for each $q\in P_{r}^{t}$ we have $q\not\prec p$. Let $r_{j}=sub_{t}(p_{j})$ for $j=1,\ldots,m$. Clearly, $t^{\Sigma}(r\leftarrow s)=t(P_{r}^{t};s).$ ###### Example 2.3. Let us consider the set $\Sigma$ of identities satisfied in the variety $RB$ of rectangular bands (see Example 2.2). Let $r=f(x_{1},x_{2})$ and let the terms $t$ and $u$ be the same as in Example 2.1. Then we have $\Sigma S_{r}^{t}=\\{f(x_{1},x_{2}),f(f(x_{1},x_{2}),x_{2})\\},\ \Sigma P_{r}^{t}=\\{1211,121\\},P_{r}^{t}=\\{121\\}$ and $t^{\Sigma}(r\leftarrow u)=t(121;u)$ (see Figure 1). So, the term $t^{\Sigma}(r\leftarrow u)$ is the term obtained from $t$ by replacing $r$ by $u$ at any minimal positions whose subterm is $\Sigma$-equal to $r$, where minimality refers to the order $\preceq$ on the set of positions. ###### Proposition 2.2. If $\Sigma\models r\approx v$, and $u,t\in W_{\tau}(X)$ then: 1. (i) $\Sigma\models t^{\Sigma}(u\leftarrow u)\approx t$; 2. (ii) $P^{t}_{r}=P^{t}_{v}$; 3. (iii) $\Sigma\models t^{\Sigma}(r\leftarrow u)\approx t^{\Sigma}(v\leftarrow u)$. ###### Proof. (i) If $t^{\Sigma}(u\leftarrow u)=t$, then the proposition is obvious. Let us assume that $t^{\Sigma}(u\leftarrow u)\neq t$. Hence $\Sigma P^{t}_{u}\setminus P^{t}_{u}=\\{q_{1},\ldots,q_{k}\\}\neq\emptyset$. Let $P^{t}_{u}=\\{p_{1},\ldots,p_{m}\\}$ and $p_{i}\in P^{t}_{u}$. If $p_{i}\not\prec q_{j}$ for all $j\in\\{1,\ldots,k\\}$, then since $\Sigma\models sub_{t}(p_{i})\approx u$ and $D_{5}$ we obtain $\Sigma\models t(p_{i};u)\approx t.$ If $p_{i}\prec q_{j}$, for some $j\in\\{1,\ldots,k\\}$, then we have $\Sigma\models sub_{t}(p_{i})\approx sub_{t}(q_{j})\approx u$. Since $t=t(p_{i};sub_{t}(p_{i}))=t(p_{i};sub_{t}(p_{i}))(q_{j};sub_{t}(q_{j}))$ we obtain $\Sigma\models t(p_{i};u)\approx t.$ Finally, we have $\Sigma\models t^{\Sigma}(u\leftarrow u)\approx t$. (ii) and (iii) are clear. ∎ ###### Corollary 2.7. 1. (i) $P^{t^{\Sigma}(u\leftarrow u)}_{u}=\Sigma P^{t^{\Sigma}(u\leftarrow u)}_{u}$; 2. (ii) $t^{\Sigma}(u\leftarrow u)^{\Sigma}(u\leftarrow v)=t^{\Sigma}(u\leftarrow u)(u\leftarrow v)$ for any term $v\in W_{\tau}(X)$. ###### Proposition 2.3. If $\Sigma\models t\approx s$ and $\Sigma\models r\approx v$, then $P^{t}_{r}\subseteq PFic(t,\Sigma)\quad\iff\quad P^{s}_{v}\subseteq PFic(s,\Sigma).$ Next we consider a deductive system, which is based on the $\Sigma$-compositions of terms. ## 3\. Stable varieties and globally invariant congruences Our next goal is to introduce deductive closures on the subsets of $Id(\tau)$ which generate elements of the lattices ${\mathcal{L}}(\tau)$ and ${\mathcal{E}}(\tau)$. These closures are based on two concepts - satisfaction of an identity by a variety and deduction of an identity. ###### Definition 3.1. [1] A set $\Sigma$ of identities of type $\tau$ is $D-$deductively closed if it satisfies the following axioms (some authors call them “deductive rules”, “derivation rules”, “productions”, etc.): 1. $D_{1}$ (reflexivity) $t\approx t\in\Sigma$ for each term $t\in W_{\tau}(X)$; 2. $D_{2}$ (symmetry) $(t\approx s\in\Sigma)\ \Rightarrow\ s\approx t\in\Sigma$; 3. $D_{3}$ (transitivity) $(t\approx s\in\Sigma)\ \&\ (s\approx r\in\Sigma)\ \Rightarrow\ t\approx r\in\Sigma$; 4. $D_{4}$ (variable inductive substitution) $(t\approx s\in\Sigma)\ \&\ (r\in W_{\tau}(X))\ \Rightarrow\ t(x\leftarrow r)\approx s(x\leftarrow r)\in\Sigma$; 5. $D_{5}$ (term positional replacement) $(t\approx s\in\Sigma)\ \&\ (r\in W_{\tau}(X))\ \&\ (sub_{r}(p)=t)\ \Rightarrow\ r(p;s)\approx r\in\Sigma$. For any set of identities $\Sigma$ the smallest $D-$deductively closed set containing $\Sigma$ is called the $D-$closure of $\Sigma$ and it is denoted by $D(\Sigma).$ Let $\Sigma$ be a set of identities of type $\tau.$ For $t\approx s\in Id(\tau)$ we say $\Sigma\vdash t\approx s$ (“$\Sigma$ $D-$proves $t\approx s$”) if there is a sequence of identities ($D-$deduction) $t_{1}\approx s_{1},\ldots,t_{n}\approx s_{n}$, such that each identity belongs to $\Sigma$ or is a result of applying any of the derivation rules $D_{1}-D_{5}$ to previous identities in the sequence and the last identity $t_{n}\approx s_{n}$ is $t\approx s.$ According to [1], $\Sigma\models t\approx s$ if and only if $\ t\approx s\in D(\Sigma)$ and the closure $D(\Sigma)$ is a fully invariant congruence for each set $\Sigma$ of identities of a given type. It is known that there exists a variety $V\subset\mathcal{A}lg(\tau)$ with $Id(V)=\Sigma$ if and only if $\Sigma$ is a fully invariant congruence (Theorem 14.17 [1]). Using properties of the essential variables and positions we can divide the rules $D_{4}$ and $D_{5}$ into four rules which distinguish between operating with essential or fictive objects in the identities. ###### Proposition 3.1. A set $\Sigma$ is $D-$deductively closed if it satisfies rules $D_{1},D_{2},D_{3}$ and 1. $D_{4}^{\prime}$ (essential variable inductive substitution) $(t\approx s\in\Sigma)\ \&\ (r\in W_{\tau}(X))\ \&\ (x\in Ess(t,\Sigma))\ \Rightarrow\ t(x\leftarrow r)\approx s(x\leftarrow r)\in\Sigma$; 2. $D_{4}^{\prime\prime}$ (fictive variable inductive substitution) $(t\approx s\in\Sigma)\ \&\ (r\in W_{\tau}(X))\ \&\ (x\in Fic(t,\Sigma))\ \Rightarrow\ t(x\leftarrow r)\approx s\in\Sigma$; 3. $D_{5}^{\prime}$ (essential positional term replacement) $(t\approx s\in\Sigma)\ \&\ (sub_{r}(p)=t,\ p\in PEss(r,\Sigma))\ \Rightarrow\ r(p;s)\approx r\in\Sigma;$ 4. $D_{5}^{\prime\prime}$ (fictive positional term replacement) $(t,s,r\in W_{\tau}(X))\ \&\ (sub_{r}(p)=t,\ p\in PFic(r,\Sigma))\ \Rightarrow\ r(p;s)\approx r\in\Sigma$. We will say that a set $\Sigma$ of identities is complete if $D(\Sigma)=Id(\tau)$. It is clear that if $\Sigma$ is a complete set, then $Mod(\Sigma)$ is a trivial variety. For fictive positions in terms and complete sets of identities, we have: $\Sigma\ \text{is complete}\ \iff(\forall t\in W_{\tau}(X))\ \ Pos(t)=PFic(t,\Sigma).$ Let $\Sigma$ be a non-complete set of identities. Then from Theorem 2.1 and Theorem 2.3, it follows that when applying the rules $D_{4}^{\prime\prime}$ and $D_{5}^{\prime\prime}$, we can skip these steps in the deduction process, without any reflection on the resulting identities. Hence, if $\Sigma$ is a non-complete set of identities, then $\Sigma$ is $D-$deductively closed if it satisfies the rules $D_{1},D_{2},D_{3},D_{4}^{\prime},D_{5}^{\prime}.$ In order to obtain new elements in the lattices $\mathcal{L}(\tau)$ and $\mathcal{E}(\tau)$, we have to extend the derivation rules $D_{1}-D_{5}$. ###### Definition 3.2. A set $\Sigma$ of identities is $\Sigma R$-deductively closed if it satisfies the rules $D_{1},D_{2},D_{3},D_{4}$ and 1. $\Sigma R_{1}$ _( $\Sigma$ replacement)_ $(t\approx s,r\approx v,u\approx w\in\Sigma)\&\ (r\in SEss(t,\Sigma))\ \&\ (v\in SEss(s,\Sigma))$ --- $\Rightarrow t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)\in\Sigma.$ For any set of identities $\Sigma$ the smallest $\Sigma R-$deductively closed set containing $\Sigma$ is called $\Sigma R-$closure of $\Sigma$ and it is denoted by $\Sigma R(\Sigma).$ Let $\Sigma$ be a set of identities of type $\tau.$ For $t\approx s\in Id(\tau)$ we say $\Sigma\vdash_{\Sigma R}t\approx s$ ($``\Sigma$ $\Sigma R$-proves $t\approx s"$) if there is a sequence of identities $t_{1}\approx s_{1},\ldots,t_{n}\approx s_{n}$, such that each identity belongs to $\Sigma$ or is a result of applying any of the derivation rules $D_{1},D_{2},D_{3},D_{4}$ or $\Sigma R_{1}$ to previous identities in the sequence and the last identity $t_{n}\approx s_{n}$ is $t\approx s.$ Let $t\approx s$ be an identity and $\mathcal{A}$ be an algebra of type $\tau$. $\mathcal{A}\models_{\Sigma R}t\approx s$ means that $\mathcal{A}\models t^{\Sigma}(r\leftarrow v)\approx s^{\Sigma}(r\leftarrow v)$ for every $r\in SEss(t,\Sigma)\cap SEss(s,\Sigma)$ and $v\in W_{\tau}(X)$. Let $\Sigma$ be a set of identities. For $t,s\in W_{\tau}(X)$ we say $\Sigma\models_{\Sigma R}t\approx s$ (read: “$\Sigma$ $\Sigma R-$yields $t\approx s$”) if, given any algebra $\mathcal{A}$, $\mathcal{A}\models_{\Sigma R}\Sigma\quad\Rightarrow\quad\mathcal{A}\models_{\Sigma R}t\approx s.$ ###### Theorem 3.1. $\Sigma R$ is a closure operator in the set $Id(\tau)$, i.e., 1. (i) $\Sigma\subseteq\Sigma R(\Sigma)$; 2. (ii) $\Sigma_{1}\subseteq\Sigma_{2}\Rightarrow\Sigma_{1}R(\Sigma_{1})\subseteq\Sigma_{2}R(\Sigma_{2})$; 3. (iii) $\Sigma R(\Sigma R(\Sigma))=\Sigma R(\Sigma)$. The following lemma is clear. ###### Lemma 3.1. For each set $\Sigma\subseteq Id(\tau)$ and for each identity $t\approx s\in Id(\tau)$ we have $\Sigma\vdash_{\Sigma R}t\approx s\iff\Sigma R(\Sigma)\vdash t\approx s.$ ###### Theorem 3.2. (The Completeness Theorem for $\Sigma R$-Equational Logic) Let $\Sigma\subseteq Id(\tau)$ be a set of identities and $t\approx s\in Id(\tau)$. Then $\Sigma\models_{\Sigma R}t\approx s\iff\Sigma\vdash_{\Sigma R}t\approx s.$ ###### Proof. The implication $\Sigma\vdash_{\Sigma R}t\approx s\Rightarrow\Sigma\models_{\Sigma R}t\approx s$ follows by $\Sigma\vdash_{\Sigma R}t\approx s\Rightarrow t\approx s\in\Sigma R(\Sigma)$ since we have used only properties under which $\Sigma R(\Sigma)$ is closed, i.e., under $D_{1},D_{2},D_{3},D_{4}$ and $\Sigma R_{1}.$ For the converse of this, let us note that for $t\in W_{\tau}(X)$ we have $\Sigma\vdash_{\Sigma R}t\approx t$ and if $t\approx s\in\Sigma$ then $\Sigma\vdash_{\Sigma R}t\approx s$. If $\Sigma\vdash_{\Sigma R}t\approx s$, then there is a formal $\Sigma R-$deduction $t_{1}\approx s_{1},\ldots,t_{n}\approx s_{n}\quad\mbox{of}\quad t\approx s.$ But then $t_{1}\approx s_{1},\ldots,t_{n}\approx s_{n},\ \ s_{n}\approx t_{n}$ is a $\Sigma R-$deduction of $s\approx t.$ If $\Sigma\vdash_{\Sigma R}t\approx s$ and $\Sigma\vdash_{\Sigma R}s\approx r$ let $t_{1}\approx s_{1},\ldots,t_{n}\approx s_{n}$ be a $\Sigma R-$deduction of $t\approx s$ and let $\overline{s}_{1}\approx r_{1},\ldots,\overline{s}_{k}\approx r_{k}$ be a $\Sigma R-$deduction of $s\approx r.$ Then $t_{1}\approx s_{1},\ldots,t_{n}\approx s_{n},\ \ \overline{s}_{1}\approx r_{1},\ldots,\overline{s}_{k}\approx r_{k},\ \ t_{n}\approx r_{k}$ is a $\Sigma R-$deduction of $t\approx r.$ Hence $\Sigma\vdash_{\Sigma R}t\approx r.$ Let $\Sigma\models_{\Sigma R}t\approx s$, $\Sigma\models_{\Sigma R}r\approx v$ , $\Sigma\models_{\Sigma R}u\approx w$ and $\Sigma\models_{\Sigma R}t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$. Suppose that $\Sigma\vdash_{\Sigma R}t\approx s$, $\Sigma\vdash_{\Sigma R}r\approx v$ and $\Sigma\vdash_{\Sigma R}u\approx w$. Let $t_{1}\approx s_{1},\ldots,t_{n}\approx s_{n}$, $r_{1}\approx v_{1},\ldots,r_{m}\approx v_{m}$ and $u_{1}\approx w_{1},\ldots,u_{k}\approx w_{k}$ be $\Sigma R-$deductions of $t\approx s$, $r\approx v$ and $u\approx w$. Then $\displaystyle t_{1}\approx s_{1},\ldots,t_{n}\approx s_{n},\quad r_{1}\approx v_{1},\ldots,r_{m}\approx v_{m},$ $\displaystyle u_{1}\approx w_{1},\ldots,u_{k}\approx w_{k},\quad t_{n}^{\Sigma}(r_{m}\leftarrow u_{k})\approx s_{n}^{\Sigma}(v_{m}\leftarrow w_{k})$ is a $\Sigma R-$deduction of $t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$. Hence $\Sigma\vdash_{\Sigma R}t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$. ∎ ###### Theorem 3.3. For each set of identities $\Sigma$ the closure $\Sigma R(\Sigma)$ is a fully invariant congruence, but $\Sigma R(\Sigma)$ is not in general equal to $D(\Sigma)$. ###### Proof. Let $\Sigma$ be a $\Sigma R-$deductively closed set of identities. We will prove that $\Sigma$ is a fully invariant congruence. It has to be shown that $\Sigma$ satisfies the rule $D_{5}$, i.e., if $r\in W_{\tau}(X)$, $t\approx s\in\Sigma$ and $p\in Pos(r)$, then $r(p;t)\approx r(p;s)\in\Sigma.$ If $p\notin PEss(r,\Sigma)$, then according to Proposition 3.1 we have $\Sigma\models r(p;v)\approx r(p;w)$ for all terms $v,w\in W_{\tau}(X)$. Let $p\in PEss(r,\Sigma)$ and let $n$ be a natural number such that $r,t,s\in W_{\tau}(X_{n})$ and let us consider the term $u=r(p;x_{n+1})$. Clearly, $u\approx u\in\Sigma$, because of $D_{1}.$ We have $u^{\Sigma}(x_{n+1}\leftarrow v)=u(x_{n+1}\leftarrow v)=u(p;v)$ for each $v\in W_{\tau}(X).$ Now from $\Sigma R_{1}$ we obtain $u^{\Sigma}(x_{n+1}\leftarrow t)\approx u^{\Sigma}(x_{n+1}\leftarrow s)\in\Sigma,\mbox{ i.e., }u(x_{n+1}\leftarrow t)\approx u(x_{n+1}\leftarrow s)\in\Sigma,$ and $r(p;t)\approx r(p;s)\in\Sigma$. Furthermore, we will produce a fully invariant congruence $\Sigma$, which is not $\Sigma R-$ deductively closed. Let us consider the variety $SG=Mod(\Sigma)$ of semigroups, where $\Sigma=\\{f(x_{1},f(x_{2},x_{3}))\approx f(f(x_{1},x_{2}),x_{3})\\}$. From Theorem 14.17 of [1] it follows that if $\Sigma$ is a fully invariant congruence, then $D(\Sigma)=Id(Mod(\Sigma)).$ Hence $Id(SG)=D(\Sigma).$ Let $t=f(f(f(x_{1},x_{2}),x_{1}),x_{2})\quad\mbox{ and }\quad s=f(f(x_{1},x_{2}),f(x_{1},x_{2})).$ It is not difficult to see that $\Sigma\models t\approx s$, i.e., $t\approx s\in D(\Sigma)$. Let us set $r=v=f(x_{1},x_{2})$ and $u=w=x_{1}$. Clearly, for each $z\in W_{\tau}(X)$ we have $PEss(z,\Sigma)=Pos(z)$ and $r\in SEss(t,\Sigma)$, and $v\in SEss(s,\Sigma)$. Since $P_{r}^{t}=\\{11\\}$ and $P_{s}^{v}=\\{1,2\\}$, we obtain $t^{\Sigma}(r\leftarrow u)=f(f(x_{1},,x_{1}),x_{2})\quad\mbox{ and }\quad s^{\Sigma}(v\leftarrow w)=f(x_{1},x_{1}).$ Hence $\Sigma\not\models t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w).$ On the other side we have $t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)\in\Sigma R(\Sigma)$. Consequently, $D(\Sigma)$ is a proper subset (equational theory) of $\Sigma R(\Sigma)$ and $Mod(\Sigma R(\Sigma))$ is a proper subvariety of $SG$, which contains the variety $RB$ of rectangular bands as a subvariety, according to Example 3.1, below. ∎ ###### Lemma 3.2. For each set $\Sigma\subseteq Id(\tau)$ and for each identity $t\approx s\in Id(\tau)$ we have $\Sigma\models_{\Sigma R}t\approx s\iff\Sigma R(\Sigma)\models t\approx s.$ The lemma follows immediately from Lemma 3.1 and Theorem 3.2. ###### Definition 3.3. A set of identities $\Sigma$ is called a globally invariant congruence if it is $\Sigma R-$deductively closed. A variety $V$ of type $\tau$ is called stable if $Id(V)$ is $\Sigma R-$deductively closed, i.e., $Id(V)$ is a globally invariant congruence. Note that when $\Sigma$ is a globally invariant congruence it is possible to apply substitutions or replacements in any place (operation symbol) of terms which explains the word “globally”. ###### Example 3.1. Now, we will produce a fully invariant congruence $\Sigma$, which is a globally invariant congruence. Let us consider the variety $RB=Mod(\Sigma)$ of rectangular bands, where $\Sigma$ is defined as in Example 2.2. The set $\Sigma$ consists of all equations $s\approx t$ such that the first variable (leftmost) of $s$ agrees with the first variable of $t$, i.e., $leftmost(t)=leftmost(s)$ and the last variable (rightmost) of $s$ agrees with the last variable of $t$, i.e., $rightmost(t)=rightmost(s)$. It is well known that $\Sigma$ is a fully invariant congruence and a totally invariant congruence ( see [2, 4]). From Theorem 14.17 of [1] it follows that $Id(RB)=D(\Sigma).$ Let $t,s,r,v,u,w\in W_{\tau}(X)$ be six terms such that $\Sigma\models t\approx s,\ \Sigma\models r\approx v,\ \Sigma\models u\approx w$, $r\in SEss(t,\Sigma)$ and $v\in SEss(s,\Sigma)$. Thus we have $\displaystyle leftmost(t)$ $\displaystyle=leftmost(s),$ $\displaystyle leftmost(r)$ $\displaystyle=leftmost(v),$ $\displaystyle leftmost(u)$ $\displaystyle=leftmost(w),$ $\displaystyle rightmost(t)$ $\displaystyle=rightmost(s),$ $\displaystyle rightmost(r)$ $\displaystyle=rightmost(v),\quad$ $\displaystyle rightmost(u)$ $\displaystyle=rightmost(w).$ From $r\in SEss(t,\Sigma)$ and $v\in SEss(s,\Sigma)$ (see Example 2.2), we obtain $leftmost(t)=leftmost(r),\quad leftmost(s)=leftmost(v)\ \mbox{ or}$ $rightmost(t)=rightmost(r),\quad rightmost(s)=rightmost(v).$ Hence $leftmost(t^{\Sigma}(r\leftarrow u))=leftmost(s^{\Sigma}(v\leftarrow w))\ \mbox{ and }$ $rightmost(t^{\Sigma}(r\leftarrow u))=rightmost(s^{\Sigma}(v\leftarrow w)).$ We are going to compare globally invariant congruences with the totally invariant congruences, defined by hypersubstitutions. In [2, 4] the solid varieties are defined by adding a new derivation rule which uses the concept of hypersubstitutions. Let $\sigma:\mathcal{F}\to W_{\tau}(X)$ be a mapping which assigns to every operation symbol $f\in\mathcal{F}_{n}$ an $n-$ary term. Such mappings are called _hypersubstitutions_ (of type $\tau$). If one replaces every operation symbol $f$ in a given term $t\in W_{\tau}(X)$ by the term $\sigma(f)$, then the resulting term $\hat{\sigma}[t]$ is the image of $t$ under the extension $\hat{\sigma}$ on the set $W_{\tau}(X)$. The monoid of all hypersubstitutions is denoted by $Hyp(\tau).$ Let $\Sigma$ be a set of identities. The hypersubstitution derivation rule is defined as follows: 1. $H_{1}$ _(hypersubstitution)_ $(t\approx s\in\Sigma\ \&\ \sigma\in Hyp(\tau))\ \Rightarrow\ \hat{\sigma}[t]\approx\hat{\sigma}[s]\in\Sigma.$ A set $\Sigma$ is called _$\chi-$ deductively closed (hyperequational theory, or totally invariant congruence)_ if it is closed with respect to the rules $D_{1},D_{2},D_{3},D_{4},D_{5}$ and $H_{1}$. The $\chi-$closure $\chi(\Sigma)$ of a set $\Sigma$ of identities is defined in a natural way and the meaning of $\Sigma\models_{\chi}$ and $\Sigma\vdash_{\chi}$ is clear. It is obvious that $D(\Sigma)\subseteq\chi(\Sigma)$ for each set of identities $\Sigma\subseteq Id(\tau)$. There are examples of $\Sigma$ such that $D(\Sigma)\neq\chi(\Sigma)$, which shows that the corresponding variety $Mod(\chi(\Sigma))$ is a proper subvariety of $Mod(D(\Sigma))$. A variety $V$ for which $Id(V)$ is $\chi-$deductively closed is called _solid_ variety of type $\tau$ [2]. A more complex closure operator on sets of identities is studied in [3]. This operator is based on the concept of coloured terms and multi- hypersubstitutions. The next proposition deals with the relations between the closure operators $\Sigma R$ and $\chi$. ###### Proposition 3.2. There exists a stable variety, which is not a solid variety. ###### Proof. Let us consider the type $\tau=(2)$ and $\Sigma=\\{f(f(x_{1},x_{2}),x_{1})\approx f(x_{1},x_{1})\\}$. We will show that $LA=Mod(\Sigma)$ is a stable variety. So, we have to prove that (1) $\Sigma\models t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w),$ when $\Sigma\models t\approx s$, $\Sigma\models r\approx v$, $\Sigma\models u\approx w$, $r\in SEss(t,\Sigma)$ and $v\in SEss(s,\Sigma)$. We will proceed by induction on $Depth(t)$ \- the depth of the term $t$. The case $Depth(t)=0$ is trivial. Let $Depth(t)=1$. If $t=f(x_{1},x_{2})$ then $\Sigma\models t\approx s$ implies $s=f(x_{1},x_{2})$ and (1) is satisfied in this case. Let us consider the case $t=f(x_{1},x_{1})$. If $s=f(x_{1},x_{1})$ then clearly (1) holds. Let $s=f(f(x_{1},s_{1}),x_{1})$ for some $s_{1}\in W_{\tau}(X)$. Since the positions in $s_{1}$ are $\Sigma$-fictive in $s$ it follows that $v$ can be one of the terms $x_{1}$ or $s$. On the other side we have $SEss(t,\Sigma)=\\{x_{1},t\\}$. Hence (1) is satisfied, again. Our inductive supposition is that if $Depth(t)<k$ then (1) is satisfied for all $s,r,u,v,w\in W_{\tau}(X)$ with $\Sigma\models t\approx s$, $\Sigma\models r\approx v$, $\Sigma\models u\approx w$, $r\in SEss(t,\Sigma)$ and $v\in SEss(s,\Sigma)$. Let $Depth(t)=k>1$ and $Depth(s)\geq k$. Then we have $t=f(t_{1},t_{2})$ and $s=f(s_{1},s_{2})$, such that $t_{1}$ or $t_{2}$ is not a variable. If $\Sigma\models t\approx r$ or $\Sigma\models s\approx v$, then by Definition 2.5 $(ii)$ and the transitivity $D_{3}$, it follows that (1) is $\Sigma\models u\approx w$ and we are done. Next, we assume that $\Sigma\not\models t\approx r$ and $\Sigma\not\models s\approx v$. First, let $t_{1}\in X$. Then $s_{1}=t_{1}$ and $\Sigma\models t_{2}\approx s_{2}$. Thus, from the inductive supposition it follows that (1) is satisfied. Second, let $t_{1}\notin X$. Then we have $s_{1}\notin X$, also. Hence $t=f(f(t_{11},t_{12}),t_{2})$, $s=f(f(s_{11},s_{12}),s_{2})$ and $\Sigma\models t_{2}\approx s_{2}$. Let $\Sigma\models t_{11}\approx t_{2}$ and $\Sigma\models s_{11}\approx s_{2}$. Then we have $\Sigma\models t\approx f(t_{2},t_{2})$ and $\Sigma\models f(t_{2},t_{2})\approx f(s_{2},s_{2})$. On the other side all positions in $t_{12}$ and $s_{12}$ are $\Sigma$-fictive in $t$ and $s$, respectively. Thus we have $\Sigma\models t^{\Sigma}(r\leftarrow u)\approx f(t_{2}^{\Sigma}(r\leftarrow u),t_{2}^{\Sigma}(r\leftarrow u))\ \mbox{ and }$ $\Sigma\models s^{\Sigma}(v\leftarrow w)\approx f(s_{2}^{\Sigma}(v\leftarrow w),s_{2}^{\Sigma}(v\leftarrow w)).$ Hence (1) is satisfied, in this case, again. If $\Sigma\models t_{11}\approx t_{2}$ and $\Sigma\not\models s_{11}\approx s_{2}$ then we have $\Sigma\models f(t_{2},t_{2})\approx f(s_{1},s_{2})$ and $\Sigma\models s_{1}\approx s_{2}\approx t_{2}$. This implies that (1) is satisfied, again. Let $\Sigma\not\models t_{11}\approx t_{2}$ and $\Sigma\models t_{1}\approx t_{2}$. Then we have $\Sigma\models t\approx f(t_{1},t_{2})$. Now, we proceed similarly as in the case $\Sigma\models t_{11}\approx t_{2}$ and $\Sigma\not\models s_{11}\approx s_{2}$. If $\Sigma\not\models t_{11}\approx t_{2}$ and $\Sigma\not\models t_{1}\approx t_{2}$, then we have $\Sigma\not\models s_{1}\approx s_{2}$. Hence $\Sigma\models t_{1}\approx s_{1}$ and $\Sigma\models t_{2}\approx s_{2}$. Again, from the inductive supposition we prove (1). To prove that $LA$ is not a solid variety, let us consider the following terms $y=f(f(x_{1},x_{2}),x_{1})$ and $z=f(x_{1},x_{1})$. Let $\sigma\in Hyp(\tau)$ be the hypersubstitution, defined as follows: $\sigma(f(x_{1},x_{2})):=f(x_{2},x_{1})$. It is clear that $\Sigma\models y\approx z$. On the other side we have $\hat{\sigma}[y]=f(x_{1},f(x_{2},x_{1}))$ and $\hat{\sigma}[z]=f(x_{1},x_{1})$. Thus, we obtain $\Sigma\not\models\hat{\sigma}[y]\approx\hat{\sigma}[z]$. Hence $LA$ is not a solid variety. ∎ ###### Remark 3.1. By analogy, it follows that the variety $RA=Mod(\\{f(x_{1},f(x_{2},x_{1}))\approx f(x_{1},x_{1})\\})$ is stable, but not solid, also. The varieties of left-zero bands $L0=Mod(\\{f(x_{1},x_{2})\approx x_{1}\\})$ and of right-zero bands $R0=Mod(\\{f(x_{1},x_{2})\approx x_{2}\\})$ are other examples of stable varieties, which are not solid ones. We do not know whether there is a non-trivial solid variety which is not stable? ## 4\. $\Sigma-$balanced identities and simplification of deductions Regular identities [1, 4] are identities in which the same variables occur on each side of the identity. Balanced identities are identities in which each variable occurs the same number of times on each side of the identity. In an analogous way we consider the concept of $\Sigma-$balanced identities. Let $t,r\in W_{\tau}(X)$ be two terms of type $\tau$ and $\Sigma\subset Id(\tau)$ be a set of identities. $EP^{t}_{r}$ denotes the set of all $\Sigma$-essential positions from $P^{t}_{r}$, i.e., $EP^{t}_{r}=PEss(t,\Sigma)\cap P^{t}_{r}$. ###### Definition 4.1. Let $\Sigma\subset Id(\tau)$. We will say that an identity $t\approx s$ of type $\tau$ is $\Sigma-$_balanced_ if $|EP^{t}_{q}|=|EP^{s}_{q}|$ for all $q\in W_{\tau}(X)$. ###### Example 4.1. Let $\Sigma$ be the set of identities satisfied in the variety $RB$ of rectangular bands (see Example 2.2). Let us consider the following three terms $t=f(f(x_{1},x_{2}),f(x_{1},x_{3}))$, $s=f(x_{1},f(f(x_{1},x_{2}),x_{3}))$ and $r=f(f(f(x_{1},f(x_{3},x_{2})),x_{3}),f(x_{1},x_{3}))$. Clearly, $\Sigma\models t\approx s$, $\Sigma\models t\approx r$, $SEss(t,\Sigma)=SEss(r,\Sigma)=\\{x_{1},f(x_{1},x_{2}),f(x_{1},x_{3}),x_{3}\\}$ and $SEss(s,\Sigma)=\\{x_{1},f(x_{1},x_{3}),x_{3}\\}$. Thus we have $EP^{t}_{x_{1}}=\\{11\\},$ $EP^{t}_{x_{3}}=\\{22\\},$ $EP^{t}_{f(x_{1},x_{2})}=\\{1\\},$ $EP^{t}_{f(x_{1},x_{3})}=\\{\varepsilon\\},$ $EP^{t}_{t}=\\{\varepsilon\\},$ $EP^{s}_{x_{1}}=\\{1\\},$ $EP^{s}_{x_{3}}=\\{22\\},$ $EP^{s}_{f(x_{1},x_{3})}=\\{\varepsilon\\},$ $EP^{s}_{s}=\\{\varepsilon\\},$ and $EP^{r}_{x_{1}}=\\{111\\},$ $EP^{r}_{x_{3}}=\\{22\\},$ $EP^{r}_{f(x_{1},x_{2})}=\\{11\\},$ $EP^{r}_{f(x_{1},x_{3})}=\\{\varepsilon\\},$ $EP^{r}_{r}=\\{\varepsilon\\}.$ Hence the identity $t\approx r$ is $\Sigma-$balanced, but $t\approx s$ is not $\Sigma-$balanced. ###### Theorem 4.1. Let $\Sigma\subset Id(\tau)$ be a set of $\Sigma-$balanced identities. If there is a $\Sigma R-$deduction of $t\approx s$ with $\Sigma-$balanced identities, then $t\approx s$ is a $\Sigma-$balanced identity of type $\tau.$ ###### Proof. Let $t,s,r\in W_{\tau}(X)$ and let $t\approx s$ and $s\approx r$ be two $\Sigma-$balanced identities of type $\tau$. Then for each term $q\in W_{\tau}(X)$ we have $|EP^{t}_{q}|=|EP^{s}_{q}|\quad\mbox{and}\quad|EP^{s}_{q}|=|EP^{r}_{q}|.$ Hence $|EP^{t}_{q}|=|EP^{r}_{q}|$ which shows that the identity $t\approx r$ is $\Sigma-$balanced, too. Let $t\approx s$ is a $\Sigma-$balanced identity in $\Sigma R(\Sigma)$ and let $r\in W_{\tau}(X)$ be a term with $t\in SEss(r,\Sigma)$ and $sub_{r}(p)=t$. We have $\Sigma\models_{\Sigma R}r(p;s)\approx r$. From Proposition 2.2, we obtain $(EP^{t}_{q}=EP^{s}_{q}\ \&\ t\in SEss(r,\Sigma))\ \Rightarrow EP^{r}_{q}=EP^{r(p;s)}_{q}$ for all $q\in W_{\tau}(X)$. Consequently the identity $r(p;s)\approx r$ is $\Sigma-$balanced, too. Let $t\approx s$, $r\approx v$ and $u\approx w$ be $\Sigma-$balanced identities from $\Sigma R(\Sigma)$. We have to prove that the resulting identity $t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$ is $\Sigma-$balanced. This will be done by induction on the depth (also called “height” by some authors). (i) The basis of induction is $Depth(t)=1$ (the case $Depth(t)=0$ is trivial). Let $t=f(x_{1},\ldots,x_{n})\in W_{\tau}(X_{n})$ and let $s=g(s_{1},\ldots,s_{m}).$ Hence, if $r=x_{i}$ for some $i\in\\{1,\ldots,n\\}$, then $EP^{t}_{r}=\\{i\\}$ and $|EP^{t}_{r}|=|EP^{s}_{v}|=1$. (ia) If $\Sigma\models r\approx t$, then $\Sigma\models v\approx s$ and we have $t^{\Sigma}(r\leftarrow u)=u$ and $s^{\Sigma}(v\leftarrow w)=w$. Hence $t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$ is $\Sigma-$balanced in this case. (ib) Let $\Sigma\not\models r\approx x_{i}$ for each $x_{i}\in X_{n}$, i.e., $r\not\in X_{n}$ and $\Sigma\not\models r\approx t$. If $r\notin X_{n}\cup\\{t\\}$, then $EP^{t}_{r}=EP^{s}_{v}=\emptyset$. Thus we have $t^{\Sigma}(r\leftarrow u)=t\ \mbox{ and}\ s^{\Sigma}(v\leftarrow w)=s$ and the resulting identity $t\approx s$ is $\Sigma-$balanced. (ic) Let $\Sigma\models r\approx x_{i}$ for some $x_{i}\in X_{n}$ and $\Sigma\not\models r\approx t$. We have $t^{\Sigma}(r\leftarrow u)=t^{\Sigma}(x_{i}\leftarrow u)\ \mbox{and}\ s^{\Sigma}(v\leftarrow w)=s^{\Sigma}(x_{i}\leftarrow w).$ Then $EP^{t^{\Sigma}(x_{i}\leftarrow u)}_{q}=EP^{u}_{q}$ and $EP^{s^{\Sigma}(x_{i}\leftarrow w)}_{q}=EP^{w}_{q}$ for each $q\in W_{\tau}(X)$, i.e., the resulting identity $t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$ is $\Sigma-$balanced, again. (ii) Let $Depth(t)>1$, $t=f(t_{1},\ldots,t_{n})$ and $s\in W_{\tau}(X)$, such that the identity $\Sigma\models t\approx s$ is $\Sigma$-balanced. Suppose that for each $t^{\prime}\in SEss(t,\Sigma)$ with $t^{\prime}\neq t$ the following is true: if $\Sigma\models t^{\prime}\approx s^{\prime}$ is $\Sigma$-balanced identity for some $s^{\prime}\in W_{\tau}(X)$, then $t^{\prime\Sigma}(r\leftarrow u)\approx s^{\prime\Sigma}(v\leftarrow w)$ is $\Sigma$-balanced, also. (iia) Let $\Sigma\models r\approx t$. Then we have $EP^{t^{\Sigma}(x_{i}\leftarrow u)}_{q}=EP^{u}_{q}\ \mbox{and}\ EP^{s^{\Sigma}(x_{i}\leftarrow w)}_{q}=EP^{w}_{q}$ for each $q\in W_{\tau}(X)$ and the resulting identity is $\Sigma-$balanced in that case, again. (iib) Let $\Sigma\not\models r\approx t$ and $\Sigma\models r\approx t_{i}$ for some $i=1,\ldots,n.$ As in the case $(ic)$ it can be proved that the identity $t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$ is $\Sigma-$balanced. (iic) Let $\Sigma\not\models r\approx t$, $\Sigma\not\models r\approx t_{i}$ for each $i=1,\ldots,n$ and there is $j\in\\{1,\ldots,n\\}$ with $EP^{t_{j}}_{r}=\\{r_{j1},\ldots,r_{jk_{j}}\\}\neq\emptyset.$ Without loss of generality assume that all such $j$ are the natural numbers from the set $L=\\{1,\ldots,l\\}$ with $l\leq n.$ Let $j\in L$. If $\Sigma\models t\approx t_{j}$ is a $\Sigma-$balanced identity, then $\Sigma\models s\approx t_{j}$ is $\Sigma-$balanced, also and by our assumption, we have that $t_{j}^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$ is $\Sigma-$balanced identity. Hence we have $|EP^{t}_{q}|=|EP^{t_{j}}_{q}|$ for all $q\in W_{\tau}(X).$ This implies that the resulting identity is $\Sigma-$ balanced in this case, also. If $\Sigma\not\models t\approx t_{j}$ for all $j\in L$ then since $t\approx s$ is a $\Sigma-$balanced identity, there are subterms $s_{1},\ldots,s_{l}$ of $s$ such that $\Sigma\models t_{j}\approx s_{j}$. According to our inductive supposition the last identities are $\Sigma-$balanced. Consequently, $|EP^{t_{j}}_{r}|=|EP^{s_{j}}_{r}|,\ EP^{t}_{r}=\cup_{j=1}^{l}EP^{t_{j}}_{r}\ \mbox{ and}\ EP^{s}_{v}=\cup_{j=1}^{l}EP^{s_{j}}_{v}.$ Hence $t^{\Sigma}(r\leftarrow u)\approx s^{\Sigma}(v\leftarrow w)$ is a $\Sigma-$balanced identity. ∎ The complexity of the problem of deduction depends on the complexity of the algorithm for checking when a position of a term is essential or not with respect to a set of identities. The complexity of that algorithm for finite algebras is discussed in [5], but it is based on the full exhaustion of all possible cases. There should be a case or cases, when the process of deduction can be effectively simplified. This is, for instance, when a variable $x$ does not belong to $var(t)$ and $\Sigma\models t\approx s.$ Therefore we obtain $x\notin Ess(t,\Sigma)\cup Ess(s,\Sigma)$(see Theorem 2.1). Then we can skip the rules $D_{4}^{\prime\prime}$ and $D_{5}^{\prime\prime}$, according to Proposition 3.1. Obviously, it is very easy to check if $x\in var(t)$ or not. ## References * [1] S. Burris, H. Sankappanavar, _A Course in Universal Algebra_ , The millennium edition, 2000 * [2] K.Denecke, D.Lau, R.Pöschel, D.Schweigert, _Solidifyable Clones,_ General Algebra 20, Heldermann Verlag , Berlin 1993, pp.41-69. * [3] K. Denecke, J. Koppitz, Sl. Shtrakov, _Multi-Hypersubstitutions and Coloured Solid Varieties_ , J. Algebra and Computation, Volume 16, Number 4, August, 2006, pp.797-815. * [4] E. Graczýnska, _On Normal and Regular Identities and Hyperidentities_ , Universal and Applied Algebra, Turawa, Poland 3 - 7 May 1988, World Scientific (1989), 107-135. * [5] Sl. Shtrakov, K. Denecke, _Essential Variables and Separable Sets in Universal Algebra_ , J. Multi. Val. Logic, 2002, vol. 8(2), pp 165-181. * [6] W. Taylor, _Hyperidentities and Hypervarieties,_ Aequationes Mathematicae, 23(1981), 30-49.

arxiv-papers

2024-09-04T02:48:53.816352

0802.2389

# Ultrahigh electron mobility in suspended graphene K. I. Bolotina K. J. Sikesb Z. Jianga,d M. Klimac G. Fudenberga J. Honec P. Kima H. L. Stormera,b,e,∗ Departments of aPhysics, bApplied Physics, cMechanical Engineering, Columbia University, New York, NY 10027, USA dNational High Magnetic Field Laboratory, Tallahassee, FL 32310, USA eBell Labs, Alcatel-Lucent Technologies, Murray Hill, NJ 07974, USA ###### Abstract We have achieved mobilities in excess of 200,000 cm2V-1s-1 at electron densities of $\sim$2$\times$1011 cm-2 by suspending single layer graphene. Suspension $\sim$150 nm above a Si/SiO2 gate electrode and electrical contacts to the graphene was achieved by a combination of electron beam lithography and etching. The specimens were cleaned in situ by employing current-induced heating, directly resulting in a significant improvement of electrical transport. Concomitant with large mobility enhancement, the widths of the characteristic Dirac peaks are reduced by a factor of 10 compared to traditional, non-suspended devices. This advance should allow for accessing the intrinsic transport properties of graphene. A. Graphene; B. Nanofabrication; D. Electronic transport ###### pacs: 73.50.-h; 73.63.-b; 81.07.-b; 81.16.-c Graphene, the latest addition to the family of two-dimensional (2D) materials, is distinguished from its cousins by its unusual band structure, rendering the quasiparticles in it formally identical to massless, chiral fermions. The experimental realization of graphene thus presents tantalizing opportunities to study phenomena ranging from the topological phase resulting in exotic quantum Hall states novoselov ; yuanbo to the famous Klein paradox – the anomalous tunneling of relativistic particles rise_graphene . However, despite tremendous interest and concerted experimental efforts [1-23], the presence of strong impurity scattering – which limits the electron mean free path to less than a micron – has been a major barrier to progress. At the same time, there is strong evidence that graphene is a nearly perfect crystal free of the structural defects elena ; ishigami that characterize most conductors. As a result, it has been put forth that the scattering of charge carriers stems from extrinsic sources nomura ; dassarma ; electrostatic ; geim_intr . Although the exact nature of the scattering that limits the mobility of graphene devices remains unclear, evidence has mounted that interactions with the underlying substrate are largely responsible. Surface charge traps chencharged ; dassarma ; nomura ; electrostatic , interfacial phonons chen_limits , substrate stabilized ripples suspend_geim ; ishigami ; geim_intr , and fabrication residues on or under the graphene sheet may all contribute. Consequently, improving substrate quality or eliminating the substrate altogether by suspending graphene over a trench seems a promising strategy towards higher quality samples. While devices suspended over the substrate were achieved in the past suspend_geim ; bunch , they lacked multiple electrical contacts thus precluding transport measurements. In this Letter we report the fabrication of electrically contacted suspended graphene and achieve a tenfold improvement in mobility as compared to the best values reported in the literature for traditional devices fabricated on a substrate. Besides opening new avenues for studying the intrinsic physics of Dirac fermions, this improvement demonstrates the dominant role played by extrinsic scattering in limiting the transport properties of unsuspended graphene samples. The fabrication of a suspended graphene device starts with optically locating a single-layer mechanically exfoliated graphene flake on top of a silicon substrate covered with 300 nm of SiO2. Single-layer graphene flakes are identified based on their contrast geim_contrast , and later confirmed via measurements of the half-integer quantum Hall effect yuanbo ; novoselov . We avoid patterning the flakes using oxygen plasma etching melinda ; novoselov , as it may introduce additional defects in the bulk and dangling bonds at the edges of graphene. Instead, we choose natural flakes of approximately rectangular shape suitable for fabrication into Hall bars. Electron beam lithography is employed to pattern the contacts to the flake. The contact material (3 nm Cr followed by 100 nm of Au) is deposited by thermal evaporation followed by a liftoff in warm acetone. The large size and thickness of the electrodes enhances the mechanical rigidity of the device. Suspension of the graphene flake is achieved by dipping the entire device into 1:6 buffered oxide etch (BOE) for 90 seconds, which uniformly removes approximately 150 nm of SiO2 across the substrate, including the area below the flake (SiO2 masked by the gold electrodes remains unetched). Uniform etching of the substrate directly below the flake is crucial for our process as it allows the fabrication of large-area suspended graphene, while maintaining the parallel plate capacitor geometry for our device. To our knowledge, this unexpected etching anisotropy in the presence of graphene was not reported before; it is, however, consistent with the rapid propagation of BOE along the SiO2/graphene interface me_unpublished . Finally, the device is transferred from BOE to ethanol and dried in a critical-point-drying step to avoid the surface-tension-induced collapse of the suspended graphene sheet. Figure 1: (a) SEM image of a typical suspended six-probe graphene device taken at $15^{\circ}$ with respect to the sample plane. (b) AFM image of the suspended device #1 before the measurements. (c) AFM image of the device #1 after the measurements with graphene removed by a short oxygen plasma etch (same z scale). (d) Device schematic, side-view. Degenerately doped silicon gate (blue), partly etched SiO2 (green), suspended single-layer graphene (pink) and Au/Cr electrodes (orange). Figure 1a shows a scanning electron microscope (SEM) image of a finished device taken at $15^{\circ}$ angle with respect to the sample plane. The graphene is apparent as a thin sheet suspended above the surface of the remaining SiO2. The sheet is supported by six gold electrodes attached to SiO2, which have been slightly undercut during the BOE etching step (see Fig. 1d). Atomic force microscopy (AFM) (Figs. 1b,c) demonstrates convincingly the integrity of the graphene sheet, its suspension above the oxide and the flatness of the substrate below it. Fig. 1b clearly indicates a flat graphene surface $\sim$150 nm above the surface of SiO2. The single layer of carbon atoms, which makes up graphene, is remarkably robust and is not damaged by repeated AFM imaging. Fig. 1c show the same device after completion of the electrical measurement and after removal of the suspended graphene via an oxygen plasma etch o2etch . It reveals the previously hidden SiO2 substrate below the graphene. The height variation of the substrate is less than 20 nm, with a slight bowing towards the center of the device. We thus conclude that our fabrication process results in graphene devices suspended $\sim$150 nm above SiO2 substrate (Fig. 1d). Electrical measurements on suspended graphene devices are performed in a sample-in-vacuum cryostat with a pressure of less than $5\times 10^{-5}$ mtorr. A total of one four-probe and two six-probe devices were measured. Before cooling the cryostat to its base temperature of $\sim$5 K the devices are thermally annealed in situ to 400 K, as this has been shown to reduce spurious doping in unsuspended samples ishigami ; schendin . Four-probe measurements are performed using standard low-frequency lock-in techniques with the excitation current less than $I=100$ nA. A typical measurement consists of sending the current between electrodes labeled 1 and 4 in Fig. 1a and recording the voltages $V_{xx}$ ( $V_{xy}$ ) between electrodes 2 and 3 ( 2 and 6 ) respectively. The resistance is calculated as $R_{xx}=V_{xx}/I$ and the Hall resistance as $R_{xy}=V_{xy}/I$. To convert resistance to resistivity we estimate the ratio of sample width to spacing between voltage probes from images such as shown in Fig. 1. Following the general approach for extended voltage probes we use the center-to-center distance along the current path ($L$) as the sample length and the distance between voltage probes perpendicular to the current path as the sample width ($W$). The sheet resistivity $\rho_{xx}$ is then calculated as $\rho_{xx}=R_{xx}(W/L)$. The uncertainty in actual current and voltage distribution within our specimens may place an error on the estimated value of $\rho_{xx}$ of less than 30%. The resistivity is measured as a function of gate voltage $V_{g}$ applied between graphene and the degenerately doped silicon substrate. Special care is taken not to collapse the devices electrostatically, as applying gate voltage $V_{g}$ of either sign leads to an attractive force between the flexible suspended graphene bunch ; electrostatic and the gate. The observation of graphene collapse at $V_{g}=20$ V in similar samples leads us to limit the range of applied gate voltages to $\pm 5$ V throughout our experiments. Following Bunch _et al._ bunch , we estimate the force acting on our typical device #1 at $V_{g}=\pm 5$ V as $F=\frac{\epsilon_{0}\epsilon^{2}LWV_{g}^{2}}{2(d_{0}+d_{1}\epsilon)^{2}}\sim 3\times 10^{-8}$ N, where $d_{0},d_{1}=150$ nm are thicknesses of the remaining and etched SiO2 and $L,W\sim 3$ $\mu$m are the length and the width of the device. Using simple mechanics, we estimate the maximum strain $\varepsilon$ in graphene to be in the range $V_{g}=\pm 5$ V as $\varepsilon\sim 0.5(\frac{F}{EtW})^{2/3}\sim 5\times 10^{-4}$, assuming a Young modulus $E$=1 TPa and a thickness $t=$0.34 nm bunch . We deduce that this strain level does not significantly affect electronic transport in graphene. The blue line of Fig. 2a shows the low temperature resistivity $\rho_{xx}$ of sample #1, measured as a function of the gate voltage $V_{g}$. We observe the Dirac peak, indicated by a maximum in the resistivity, at the gate voltage VD close to zero. The small reproducible fluctuations in $\rho_{xx}(V_{g})$ are consistent with universal conductance fluctuation, typically seen in mesoscopic devices melinda ; ucf . The carrier density $n$ is determined via Hall effect measurements as $n(V_{g})=B/e\rho_{xy}(V_{g},B)$, where $B$ is the applied magnetic field. The gate capacitance of the device is calculated as $C_{g}=n(V_{g})e/(V_{g}-V_{D})\sim 60$ aF$\mu$m-2. novoselov ; yuanbo The measured capacitance is close to the value $C_{g}\sim$ 47$\pm$5 aF$\mu$m-2 expected for graphene suspended 150$\pm$20 nm above 150$\pm$20 nm of residual SiO2, as calculated using the serial capacitor model. This provides an independent verification that the device is suspended during the measurements. Finally, using the above carrier density, we determine the electron mobility $\mu=1/ne\rho_{xx}\sim$ 28,000 cm2V-1s-1 at $n=2\times 10^{11}$ cm-2. This is comparable to the best reported values for unsuspended devices at the same density yuanbo ; novoselov ; ong ; yanwen . Thus, despite removing the substrate, at this stage the scattering in graphene is not significantly reduced, which leads us to the conclusion that it is caused by residual impurities absorbed on the graphene surface. Further mobility enhancement requires removal of the remaining impurities. This is accomplished by sending a large current through the device. For unsuspended samples, this current annealing was demonstrated to heat the graphene sheet locally to an estimated $T\sim 600$ C and to desorb most of the residues remaining on the surface of the device from the fabrication steps. While current annealing has been shown to improved the quality of electrical transport in unsuspended devices, the treatment did not lead to significant mobility enhancement bachtold . Most likely, impurities permanently trapped at the interface between graphene and the substrate are responsible for this lack of improvement. Suspended devices, on the other hand, are not be subject to such limitations, since impurities from both sides of the graphene sheet are free to desorb. Current annealing is implemented by ramping the current across the device up to a predefined setpoint, waiting for several minutes, decreasing the current to zero and remeasuring the electrical transport properties of the specimen. The procedure is applied repeatedly until changes appear in the gate response of the device, which start to occur only at very large current densities of $\sim 2\times 10^{8}$ A/cm2, estimated assuming a graphene thickness 0.34 nm. Figure 2: (a) Measured four-probe resistivity $\rho_{xx}$ as a function of gate voltage $V_{g}$ for the device #1 before (blue) and after (red) current annealing; data from traditional high-mobility device on the substrate (gray dotted line) shown for comparison. The gate voltage is limited to $\pm$5 V range to avoid mechanical collapse. (b) Mobility $\mu=1/en\rho_{xx}$ as a function of carrier density $n$ for the same devices. For every device measured, current annealing leads to a remarkable difference in the transport properties compared to the initial state, which we illustrate using device #1 as an example. Upon current annealing, the resistance of sample #1 decreases by more than a factor of 8 for voltages away from the Dirac point. At the same time the width of the Dirac peak reduces by about a factor of 20, while the maximum resistivity of the device hardly changes (Fig. 2a). These large changes reflect a greatly improved sample quality. We quantify this improvement via three different measures: carrier mobility, width of the Dirac peak and the onset field of Shubnikov deHaas oscillations. Our first measure of sample quality is carrier mobility $\mu$ evaluated at high electron density, where $\mu$ saturates. In unsuspended devices, the mobility ranges between 2,000 and 25,000 cm2V-1s-1 with $\mu\sim 25,000$ cm2V-1s-1 at $n=5\times$1012 cm-2 being the highest value reported in the literature ong ; yanwen ; novoselov . Due to the gate voltage limitation in our devices we measure the mobility at a smaller density $n=2\times$1011 cm-2, where the highest reported $\mu$ is about 30,000 cm2V-1s-1 (Fig. 2b, dotted line). This value is comparable to the mobility of 28,000 cm2V-1s-1 (Fig. 2b, blue line) in the suspended sample #1 before current annealing. Upon current annealing, the resistance decrease in sample #1 translates into an increase of mobility to 230,000 cm2V${}^{-1}s^{-1}$ (Fig. 2b, red line) measured at our highest density of $n=2\times$1011 cm-2. Every suspended device exhibits mobilities higher than 60,000 cm2V-1s-1 after annealing. Our peak mobility of 230,000 cm2V-1s-1 represents an improvement of about a factor of 10 over values reported in the literature so far, and is the central result of this work. In addition to the mobility enhancement, we notice that the Dirac peak of suspended and annealed samples is very narrow compared to both that of suspended devices before annealing and traditional substrate supported devices. We argue that the width of the Dirac peak is related to the charge inhomogeneity inside the sample. As has been demonstrated recently, at small charge densities the graphene breaks into mesoscopic puddles of hole and electrons yacoby . The mechanism causing the formation of puddles is debated dassarma ; nomura ; geim_intr , but it is accepted that the presence of puddles changes transport characteristics, resulting in a broadened Dirac peak. We quantify the changes in sample quality by measuring $\Delta W_{Dirac}$, defined as twice the carrier density at which the resistivity decreases by a factor of two from its maximum value. Such $\Delta W_{Dirac}$ provides an upper bound for the charge inhomogeneity due to puddle formation. In device #1, for example, the Dirac peak narrows to about 2$\times$1010 cm-2 (Fig. 2b, red line), an improvement of more than 10 times compared to the same sample before annealing (Fig. 2b, blue line) and compared to typical high mobility unsuspended devices (Fig. 2b, black dotted line). We remark that the reduced charge inhomogeneity is correlated with enhanced carrier mobility (Fig. 3b). Compared to unsuspended samples (black squares) where a typical charge inhomogeneity is 2-9$\times$1011 cm-2 while the mobility ranges from 2,000-30,000 cm2V-1s-1, the suspended and annealed samples (red circles) exhibit both an order of magnitude higher mobility and an order of magnitude lower charge inhomogeneity, following the trend seen in the unsuspended devices. Finally, we turn to the onset of the Shubnikov deHaas oscillations as a measure of sample quality. In a simple model, these oscillations commence at magnetic field $B_{SdH}$ strong enough for a charge carrier to complete one cyclotron orbit without scattering, which is equivalent to $\omega_{c}\tau\sim 1$, where $\omega_{c}$ is the cyclotron frequency and $\tau$ is the scattering time. In graphene, a semiclassical relation yields $\omega_{c}=ev_{F}B_{SdH}/\hbar(\pi n)^{1/2}$, where $v_{F}=10^{6}$ m/s is the Fermi velocity. This results in an estimate $\tau\sim\hbar(\pi n)^{1/2}/ev_{F}B_{SdH}$. Figure 3a shows the SdH effect in our highest mobility specimen, sample #1. Oscillations are observed as low as $B_{SdH}\sim 250$ mT (Fig 3a, red line), while no SdH oscillations are observed before current annealing (Fig. 3a, blue line). Other suspended devices exhibit $B_{SdH}$ ranging from 250 to 600 mT, and we estimate $\tau\sim 2\times 10^{-13}$ s for the best device at $n=2\times 10^{11}$ cm-2. On the other hand, in unsuspended devices SdH oscillations at the same density are seen at fields larger than $\sim$ 700 mT, corresponding to $\tau\sim 7\times 10^{-14}$ s. Therefore, the early onset of Shubnikov deHaas oscillation in the suspended devices is consistent with reduced electron scattering time and thus is indicative of cleaner samples. While the onset of the SdH oscillations is a qualitative measure for sample quality, we cannot deduce directly a quantum scattering time $\tau_{q}$, since other factors, such as density inhomogeneity, also affect the onset. Figure 3: (a) $\rho_{xx}$ component of Hall resistance as a function of magnetic field for the suspended sample #1 before annealing (blue) and after annealing (red) at $n=2\times 10^{11}$ cm-2 and $T\sim 5$ K. (b) Full width at half maximum of the Dirac peak $\Delta W_{Dirac}$ plotted as a function of device mobility $\mu$ for all three measured suspended devices (red circles) and previously studied devices on the substrate (black squares). (c) Conductivity $\sigma$ as a function of carrier density $n$ for the sample #1 after current annealing. Summarizing the results of our transport measurements on in-situ annealed, suspended graphene samples, we observe a considerable improvement in sample quality measured by the enhanced mobility, reduced sample inhomogeneity and increased scattering time. In particular, we observe about an order of magnitude improvement in carrier mobility and sample homogeneity, while the improvement in the onset field of the SdH oscillations is about factor of 3. Overall, we conclude that our fabrication procedure results in very clean samples containing far fewer scatterers compared to the previously studied substrate supported devices. Interestingly, suspended samples prior to current annealing as well as current annealed but unsuspended samples bachtold do not exhibit the aforementioned quality improvement. This suggests that impurities trapped between the SiO2 and graphene are limiting the mobility of the current generation of unsuspended graphene devices. Finally, we consider the nature of the residual scatterers in our devices. Upon current annealing, the carrier mean free path $l$ in our samples approaches the typical dimensions of the device. Indeed, using a semiclassical relation between the mobility and the mean free path dassarma $\sigma=en\mu=\frac{2e^{2}}{h}(k_{F}l)$, where $k_{F}=(\pi n)^{1/2}$, we estimate $l\sim 1.2~{}\mu$m for the sample #1 at $n=2\times 10^{11}$ cm-2. Therefore, both the edges of the device and the electrodes may contribute considerably to scattering. This is consistent with the observed strongly sublinear dependence of the conductivity $\sigma(n)=1/\rho_{xx}(n)$ as a function of carrier density $n$ (Fig. 3c). Such behavior was argued to result from the short-range scattering dassarma ; yanwen , typically associated with point defects or sample edges. Overall, we speculate that extrinstic sources of scattering may still be the limiting factor in the present geometry and that larger area devices may exhibit even higher mobilities. ## Acknowledgements We acknowledge fruitful discussions with and experimental help from Erik Henriksen, Jeffrey Kysar, Andrea Young, Barbaros Özyilmaz, and Pablo Jarillo- Herrero. This work is supported by the NSF (No. DMR-03-52738), NSEC grant CHE-0641523, NYSTAR, DOE (No. DE-AIO2-04ER46133 and No. DEFG02-05ER46215), ONR (No. N000150610138), FENA MARCO, W. M. Keck Foundation, and the Microsoft Project Q. ## References * (1) K. S. Novoselov _et al._ , Nature 438 (2005) 197. * (2) Y. Zhang, Y. -W. Tan, H. L. Stormer and P. Kim, Nature 438 (2005) 201. * (3) A. K. Geim and K. S. Novoselov, Nature Materials 6 (2007) 183. * (4) E. Stolyarova _et al._ , PNAS 104 (2007) 9209. * (5) M. Ishigami, J. H. Chen, W. G. Cullen, M. S. Fuhrer, E. D. Williams, Nano Lett. 7 (2007) 1643. * (6) J. H. Chen, C. Jang, M. S. Fuhrer, E. D. Williams, M. Ishigami, Nature Physics 4, (2008) 377. * (7) E. H. Hwang, S. Adam, and S. Das Sarma, Phys. Rev. Lett. 98 (2007) 186806. * (8) K. Nomura and A. H. MacDonald, Phys. Rev. Lett. 96 (2006) 256602. * (9) J. Sabio _et al._ , arXiv:cond-mat/0712.2232v2 * (10) S. V. Morozov _et al._ , Phys. Rev. Lett. 100 (2008) 016602. * (11) J. H. Chen, C. Jang, S. Xiao, M. Ishigami, M. S. Fuhrer, Nature Nanotech. 3, (2008) 206. * (12) J. C. Meyer _et al._ , Nature 446 (2007) 60. * (13) J. S. Bunch _et al._ , Science 315 (2007) 490. * (14) K. I. Bolotin _et al._ , unpublished. * (15) We use 6 seconds long oxygen plasma etch, with power at 50 W and oxygen pressure 200 mT. * (16) S. V. Morozov _et al._ , Phys. Rev. Lett. 97 (2006) 016801. * (17) M. Y. Han, B. Oezyilmaz, Y. Zhang, and P. Kim, Phys. Rev. Lett. 98 (2007) 206805. * (18) Y. -W. Tan _et al._ Phys. Rev. Lett. 99 (2007) 246803. * (19) J. G. Checkelsky, L. Li, and N. P. Ong, cond-mat/0708.1959. * (20) J. Moser, A. Barreiro, A. Bachtold, Appl. Phys. Lett. 91 (2007) 163513. * (21) F. Schedin _et al._ , Nature Mater. 6 (2007) 652. * (22) J. Martin _et al._ , Nature Phys. 10.1038/nphys781 (2007). * (23) P. Blake _et al._ , Appl. Phys. Lett. 91 (2007) 063124. * (24) C. Berger, Science 312 (2006) 1191.

arxiv-papers

2024-09-04T02:48:53.822437

0802.2430

# The Modeling of Time-Structured Multiturn Injection into Fermilab Main Injector (Microbunch Injection with Parasitic Longitudinal Painting) Phil S. Yoon, David E. Johnson, and Weiren Chou Fermi National Accelerator Laboratory, Batavia, IL 60510 E-mail: syoon@fnal.gov Work supported by Fermilab Research Alliance (FRA), LLC under contract No. DE-AC02-07-CH11359 with the United States Department of Energy (February 2024) ###### Abstract This paper presents the modeling of time-structured multiturn injection for an upgraded Main Injector with the 8-GeV Superconducting RF proton driver, or an ILC-style linac, or a Project-X linac. The Radio-Frequency mismatch between a linac and the upgraded Main Injector will induce parasitic longitudinal painting in RF-phase direction. Several different scenarios with a choice of different RF parameters for single RF system and double RF system in the presence of longitudinal space charge have been investigated. From the studies of microbunch injection with the aid of ESME (2003) numerical simulations, it is found that the dual RF system with a choice of appropriate RF parameters allows us to overcome the space-charge limitation set by beam intensity during the multiturn-injection process. A double RF system with a harmonic ratio ($R_{H}=H_{2}/H_{1}$) of 2.0 and a voltage ratio ($R_{V}=V_{2}/V_{1}$) of 0.5 are most favored to reduce both longitudinal and transverse effects of space charge in the Main Injector. ## Microbunch Injection into the Main Injector from a Superconducting RF linac into the Main Injector After studying the method of time-structured multiturn injection from the present 400-MeV linac to Fermilab’s Booster [1], we have further explored the method of microbunch injection for applications to the Main Injector: from a future Superconducting RF (SRF) linac to an upgraded Main Injector (FMI-2)111 We will use Main Injector (FMI) and upgraded Main Injector (FMI-2) interchangeably herein. including parasitic longitudinal painting. The future SC linac referred herein can be the 8-GeV SC xsRF linac proton driver [2, 3, 4], or an ILC-style linac, or the Project-X linac [5]. ### Overview of the Main Injector The Main Injector (MI) is a ring with a circumference of about 3.3 (km). The central roles of the MI is to connect to the Tevatron, the Booster, the Antiproton source, switchyard, and the Recycler ring via a number of beam transport lines in the Fermilab accelerator complex. The MI can accelerate and decelerate particles between 8 GeV and 150 GeV, depending on the mode of operation. The harmonic number of the MI is 588 and harmonic RF at injection is 52.8114 (MHz)222 For the sake of brevity and convenience, MI RF will be referred to as 53 MHz hereafter. ### Overview of the 8-GeV Superconducting RF Linac An 8-GeV SC linac has been proposed as a single-stage $H^{-}$ injector into the Main Injector as a replacement for the aging 400-MeV Linac and 8-GeV Booster. This would be the highest energy $H^{-}$ multiturn injection system in the world. Fermilab has been carrying out design studies [4, 6] of the SC linac and injection systems[7] for the last several years. The linac design[8] utilizes a warm-temperature 325-MHz RFQ and rebuncher cavities to bunch the beam at 325 MHz. At $\beta=0.89$ (about $E_{kin}=1.1~{}GeV$), the RF of the SC cavities is 1.3 GHz. The ultimate bunch structure required for injection into the MI will be formed by a 325-MHz fast chopper system[9]. The fast chopper system will be required to remove individual 325-MHz bunches or groups of bunches to be matched to the MI RF structure and provide a beam abort notch: two out of every six microbunches333 The microbunch is referred to as a 325-MHz bunch hereafter. are to be removed. ### Time Structure of the Main Injector In the injection modeling for the MI with the aid of ESME[10], a train of four microbunches, produced by the fast chopper system located at the front end of the 8-GeV SC RF linac, are injected into the upgraded Main Injector. Figure 1 is a schematic illustrating one MI RF bucket that is populated with an initial train of four microbunches. The two chopped microbunches are represented by two consecutive empty 325-MHz RF buckets. Figure 1: Three consecutive MI RF buckets of 53 MHz. A train of four microbunches of 325 MHz are synchronously injected into each standing 53-MHz bucket. Two chopped microbunches are equivalent to 6 ns. More details of the time structure of the MI RF bucket after the first synchronous injection from a SC linac are illustrated in Figure 2. Figure 2: The time structure of a single RF system of an upgraded Main Injector: principal RF harmonic of 52.8 (MHz), sub-harmonic of 325 (MHz), and a beam abort notch of 6 ns Figure 3: Single RF system: the injection of 4 microbunches at the $1^{st}$ turn with RF waveform drawn in the background A total beam notch per MI RF bucket is about 6 ns, which corresponds to two 325-MHz RF buckets. Figure 3 is a phase-space ($\Delta E,\theta$) plot of the very first train of four microbunches with a RF-voltage waveform drawn in the background. The following is a list of longitudinal parameters that can be found in the header of each ESME phase-space plot in this memo. iter (number of turns), $H_{B}$ (bucket height), $S_{B}$ (bucket area), $S_{b}$ (bunch area), $V$ (RF voltage), $E_{s}$ (synchronous energy), $\nu_{s}$ (synchrotron tune), pdot ($dp/dt$), $\eta$ (slip factor), $\tau_{s}$ (revolution period), h (harmonic number), $\Psi$ (synchronous phase), and N (number of macroparticles) The horizontal axis is $\Delta\theta$ ($=\theta-\theta_{s}$) in units of degrees with one full revolution representing 360 (deg), the vertical axis on the left is $\Delta E$ ($=E-E_{s}$) in units of MeV, and the vertical axis on the right is $V_{rf}$ in units of kV for the RF waveform. The azimuthal-density444 The azimuthal density is referred to as the charge density, or the azimuthal profile of macroparticles. and energy-density distribution555 The energy-density distribution is referred to as the profile of macroparticles in energy ($\Delta E$) direction. of one of four microbunches are plotted in Figures 4 and 5, respectively. The root-mean- square (RMS) width of one microbunch is 1.96 $\times 10^{-4}$ (deg), which corresponds to 6.05 (ps). The red area indicates the tail portion of a microbunch. The fractional tail portion is about 1.34 $\%$. To make the tail portion of a microbunch stand out, the density distribution is also plotted on a logarithmic scale. The RMS value of initial energy spread is about 0.26 (MeV) per each microbunch. Regarding the MI to be a 11.1338-$\mu$s, or 360-degree ring, we used the following conversion factor: $0.032335~{}(deg/ns)\vspace{-0.2cm}$ (1) Figure 4: Zero-centered charge density with tail portion; (left) on a linear scale (right) on a logarithmic scale Figure 5: Zero-centered energy density; (left) on a linear scale (right) on a logarithmic scale The principal MI RF is 53 (MHz) and the SC-linac bunching frequency is 325 (MHz). From the MI RF, we can obtain an integer harmonic number of 588. The width of a MI RF bucket is 18.935 (ns), into which trains of microbunches are repeatedly injected over 270 turns, filling in a MI RF bucket in the region of $\pm$ 6 (ns) around the center of a MI RF bucket. The beam notch that is kept free of beams is 3.3 (ns) long, following an earlier injection study with a long bunch[11]. In the following sections, we will explore and discuss several microbunch-injection scenarios, based upon the RF parameters used in the present MI operation. ### Microbunch-Injection Scenarios for the Main Injector The goals of the modeling of the time-structured multiturn injection into FMI-2 with a SC linac are threefold: 1. (1) To find the optimized Fermilab Main Injector RF system with the following aspects: 1. (i) Efficient RF capture methods with minimum particle losses 2. (ii) Main Injector RF system with adequate RF parameters 3. (iii) Minimization of space-charge effects in all degrees of freedom 4. (iv) Optimization of total capture time and total injection time 2. (2) To design a fast beam-chopper system to make nearly loss-free injection attainable 3. (3) To investigate any other limitations on the intensity upgrades from the present Main Injector With the above goals in mind, we modeled four different scenarios of synchronous injection over 270 injection turns, but with different RF harmonic systems and RF parameters. The following is the parameters of the Main Injector that are consistently employed in ESME simulations for all scenarios presented herein. Table 1: Main Injector Parameters for ESME Simulations Mean Radius | 528.297 (m) ---|--- Beam Momentum (at injection) | 8.889 ($GeV/c$) $\gamma_{tr}^{2}$ (transition gamma) | 466.53572 Principal Harmonic No. | 588 $\Phi_{s}$ (synchronous phase) | 0.0 (deg) effective beam radius (at injection) | 0.0050 (m) effective beam-pipe radius (at injection) | 0.051 (m) No. of Space-Charge Bins | 64 No. of FFT Bins | 64 Microbunch Intensity | 2.65 $\times 10^{8}$ Total Intensity (after injection is complete) | 1.54 $\times 10^{14}$ SCENARIO (I) [t]synchronous injection of a train of 4 microbunches into a standing RF bucket single RF Harmonic (53 MHz) $V_{rf}$ = 800 (kV) (fixed RF voltage with no ramping) 270 injection turns ($\sim$ 3 ms) $\rightarrow$ 2,700 turns ($\sim$ 30 ms) longitudinal space charge included longitudinal painting in RF- phase ($\theta$) direction We begin with a single RF system of 53 (MHz) and the fixed RF voltage ($V_{rf}$) of 800 (kV). Referring to Figure 3, a series of four blobs captured in a RF bucket represent the first train of microbunches that are synchronously injected into a standing RF bucket. The dotted sinusoidal curve for RF voltage waveform in the background of the phase-space figure indicates the single RF harmonic and the amplitude of RF voltage. The single turn in the MI in terms of the average machine circumference and the revolution period can be expressed as follows: $360~{}(deg)=11.1338~{}(\mu s)=3319.388~{}(m)$ (2) For one MI RF bucket, the following Eqn. (3) is obtained from Eqn. (2) as illustrated in Figure 2. $52.8114~{}(MHz)=18.9353~{}(ns)=0.6122~{}(deg)=5.645~{}(m)$ (3) Due to the longitudinal mismatch, or RF mismatch between the SC linac and the FMI-2, an inherent phase slip between 325-MHz microbunch structure and the MI stationary buckets could be parasitically induced in the form of longitudinal painting in RF-phase direction. With $f$ and $\lambda$ being radio frequency and RF wavelength, respectively, we computed the RF ratio ($\mathcal{R}_{rf}$). $\displaystyle\mathcal{R}_{rf}$ $\displaystyle=\frac{f_{PD}}{f_{MI}}=\frac{325~{}\text{MHz}}{52.8114~{}\text{MHz}}$ (4) $\displaystyle=\frac{\lambda_{MI}}{\lambda_{PD}}=\frac{5.6453~{}(m)}{0.9173~{}(m)}$ $\displaystyle=6.154$ Referring to Figure 2, Eqn. (4) implies that a total of 6 linac RF buckets of 325 MHz can fit in one MI RF bucket. Since the RF ratio is a non-integer number, a modulus of the RF ratio is computed as for the case of the Booster injection modeling[1]. $\mathcal{R}_{rf}\Bigr{|}_{mod}=0.154$ (5) One can then arrive at the range of phase slip within one MI RF bucket. $0.155\times 0.0995~{}(deg)=0.0154~{}(deg)$ (6) Figure 6 shows trains of microbunches and discrete charge distribution after 3 injection turns with phase slips included. For each successive MI RF bucket, the 325-MHz bunches advance by 0.0154 (deg). Hence, it is estimated that every 6 turns the phase of injected microbunches slips through one 325-MHz bucket. $\vspace{0.3cm}\delta\theta_{rf}=\frac{0.0995~{}(deg)}{0.0154~{}(deg)}=6.46$ (7) Having this amount of phase slip included, we simulated the SC linac-to-MI injection with ESME over 270 injection turns. After the 270 turns, the longitudinal phase space is painted in RF-phase space as illustrated by Figure 7. As in Fiugre 8, one-peaked but continuous distribution of charge density666 Charge density, or azimuthal density is referred to as charge line density, for convenience. at the 270-th turn is shown. The evolution of longitudinal emittance growth for the case of a single RF harmonic is shown in Figure 9. As the number of macroparticles increases, so does the longitudinal emittance up to about 0.08 (eV-s). Then, the emittance flattens out until the end of the simulation run. The induced voltage ($\mathcal{V}_{sc}$) due to space-charge fields ($\mathcal{E}(z)$) is proportional to the g-factor, which is usually defined as $1+2\ln(R_{w}/R_{b})$. Hence, from the charge distribution, space- charge-induced voltage per turn is computed at a specified turn. In particular, we looked at the induced voltage at the last injection turn of 270-th turn as shown in Figure 11. $\mathcal{V}_{sc}\propto\frac{1}{\beta\gamma^{2}}\Bigl{[}1+2\ln(R_{w}/R_{b})\Bigr{]}\frac{d\lambda}{dz}$ (8) where $R_{w}$, and $R_{b}$ denote effective beam pipe radius, and effective beam radius, respectively. In the case of the Main Injector at injection energy of 8.0 GeV, the g-factor is about 5.64. In Figure 12, the peak collective voltage in frequency domain ($\hat{V}_{FD}$) is drawn at each turn. The collective voltage goes up to about 50.0 (keV) at the end of the injection. (a) phase space (b) charge density Figure 6: [Scenario (I)] Phase space and discrete charge density after 3 injection turns (a) 270 turns (b) 2,700 turns Figure 7: [Scenario (I)] Synchronous injection of microbunches with a single RF system (a) after 270 turns (b) after 2,700 turns; Note that LSC in the figure title stands for Longitudinal Space Charge. (a) 270 turns (b) 2,700 turns Figure 8: [Scenario (I)] Distribution of charge density with single RF harmonic at the 270-th turn and at the 2,700-th turn. Figure 9: [Scenario (I)] Evolution of longitudinal emittance with a single RF harmonic over 270 turns Figure 10: [Scenario (I)] Evolution of the number of injected macroparticles with a single RF harmonic over 270 turns Figure 11: [Scenario (I)] Induced $\Delta E$ per turn due to space charge ($\Delta E/turn$) with a single RF harmonic at $270^{th}$ turn Figure 12: [Scenario (I)] Time evolution of collective voltage in frequency domain over 270 turns Figure 13: [Scenario (I)] Time evolution of bunching factor As the bunching factor is usually defined as average current ($\langle~{}I~{}\rangle$) over peak current ($\hat{I}$) for convenience, the value is upper-bounded at unity. Figure 13 shows that the bunching factor converges to about 0.3 for the case of Scenario (I). SCENARIO (II) In Scenario (II), the injection process described in Scenario (I) is followed by ramping up the RF voltage linearly up to 150 $\%$ of the initial RF voltage. Thus, the RF voltage ramping lasts for additional 27 (ms), and the total injection time elapses about 30 (ms). [t]$V_{rf,~{}i}$ = 800 (kV), $V_{rf,~{}f}$ = 1,200 (kV) ramping RF voltage after the injection is complete At the end of RF voltage ramping, the macroparticles are well captured in a RF bucket as shown in Figure 14. Figure 15 shows the variation of RF voltage from 800 (kV) to 1,200 (kV) starting from the 270-th turn through the 2,700-th turn. In comparison to Figure 8 from Scenario (I), the gradient of charge distribution ($d\lambda/dz$) are reduced and spread more out with ramping RF voltage over extended 2,700 turns (cf. Figure 16). However, it still shows a peaked distribution centered around the origin. The longitudinal emittance ($\varepsilon_{l}$) grows up to 0.085 (eV-s) with a fixed RF voltage at 800. (kV). Once the RF voltage starts ramping linearly, the emittance continues to grow gradually up to 0.09 (eV-s) as shown in Figure 17. In an attempt to reduce further the space-charge-induced voltage and to produce a more uniform charge distribution, a dual RF harmonic system is explored in Scenarios (III) and (IV) in the following subsections. Figure 14: [Scenario (II)] Synchronous injection of microbunches with a single RF harmonic and ramping RF voltage Figure 15: [Scenario (II)] RF voltage curve over 2,700 turns Figure 16: [Scenario (II)] Distribution of charge density with a single RF harmonic and ramping RF voltage after 2,700 turns Figure 17: [Scenario (II)] Time evolution of longitudinal emittance ($\epsilon_{l}$) with a single RF system SCENARIO (III) [t]Dual RF Harmonics: $f_{rf,~{}1}$ = 53 MHz and $f_{rf,~{}2}$ = 106 MHz $H_{1}~{}=~{}588$ and $H_{2}~{}=~{}1176$ $R_{H}$ = $H_{2}/H_{1}$ = 2.0 $V_{rf,~{}1}~{}=~{}400.~{}(kV)$ and $V_{rf,~{}2}~{}=~{}300.~{}(kV)$ (fixed RF voltages) $R_{H}$ = $V_{rf,~{}2}/V_{rf,~{}1}$ = 0.75 In scenario (III), a dual RF harmonic system of 400 (kV) on 53 (MHz) and 300 (kV) on 106 (MHz) is explored. The higher-order RF harmonic ($H_{2}=1176$) is twice of the principal RF harmonic ($H_{1}=588$). The secondary RF voltage ($V_{rf,~{}2}$) is 75 $\%$ of the principal RF voltage ($V_{rf,~{}1}$). The RF voltage waveform of the higher harmonic for a dual RF system can be $\mathcal{V}=\mathcal{V}_{rf,~{}1}\sin(H_{1}\phi_{1}+\psi_{1})+\mathcal{V}_{rf,~{}2}\sin(H_{2}\phi_{2}+\psi_{2}),$ (9) where $\mathcal{V}_{rf}$, $H$, $\phi$, and $\psi$ are RF voltage, harmonic number, phase angle of each macroparticle, and phase of RF cavity, respectively. As can be seen in Figure 18, the waveform has a negative slope around the stable phase of 0 (deg). By adding a higher secondary harmonic RF voltage to a principal harmonic RF voltage, we can create a flat-bottom potential energy. As a result, the RF-bucket contour with a dual RF system is also flattened at the top and bottom on a phase-space plot. Hence, the charge distribution follows the flat contour shape. Also, the principal RF voltage can be half as high as the RF voltage used for the case of single RF harmonic. Since the flat-bottom potential energy irons out the peaked charge distribution, the utilization of a dual RF system gives a great advantage to the single RF system. Eventually, the dual RF system will help us lower the beam-current limitations caused by space-charge effects. After 2,700 turns, the formation of localized macroparticle distributions are observed around three local bumps of the dual-harmonic voltage waveform. Figure 20 shows the time evolution of injected macroparticles inside a dual RF bucket. From turn 1 through turn 270, a train of four microbunches are injected with parasitic phase offsets. After the injection process is complete, in order to reach an equilibrium state, injected macroparticles are circulated up to 2,700 turns with no further injection. Note that as time elapses after the completion of injection process, sporadic white gaps in between lumps of macroparticles are gradually disappeared. In Figures from 22 through 24 three plots of longitudinal phase space, charge density, and energy density comprise each row corresponding to turn number under a phase-space plot. With the choice of RF voltages $R_{V}=0.75$, a bi-modal charge distribution is created due to a pair of potential wells around the stable phase of 0 (deg). However, with the help of longitudinal painting, the contour of charge distribution becomes smoother as the turn number increases, but the dual peaks remain in the charge distribution. Figure 18: [Scenario (III)] RF bucket and dual RF waveform in the background (a) after 270 turns (b) after 2,700 turns Figure 19: [Scenario (III)] Distribution of macroparticles with longitudinal painting (a) after 270 turns (b) after 2,700 turns Figure 20: [Scenario (II)] Distribution of charge density with longitudinal painting included (a) $1^{st}$ turn (b) $100^{th}$ turn (c) $200^{th}$ turn (d) $270^{th}turn$ (e) $600^{th}$ turn (f) $900^{th}$ turn (g) $1200^{th}turn$ (h) $1500^{th}$ turn (i) $1800^{th}$ turn (j) $2100^{th}turn$ (k) $2400^{th}$ turn (l) $2700^{th}$ turn Figure 21: [Scenario (III)] Time evolution of phase space with longitudinal painting included, starting from the $1^{st}$ injection turn through 2700 turns (a) $100^{th}$ turn (b) charge density (c) energy density (d) $200^{th}$ turn (e) charge density (f) energy density (g) $270^{th}$ turn (h) charge density (i) energy density (j) $300^{th}$ turn (k) charge density (l) energy density Figure 22: [Scenario (III)] Time evolution of phase space with longitudinal painting starting from $100^{th}$ turn through $300^{th}$ turns (a) $600^{th}$ turn (b) charge density (c) energy density (d) $900^{th}$ turn (e) charge density (f) energy density (g) $1200^{th}$ turn (h) charge density (i) energy density (j) $1500^{th}$ turn (k) charge density (l) energy density Figure 23: [Scenario (III)] Time evolution of phase space with longitudinal painting starting from $600^{th}$ turn through $1500^{th}$ turns (a) $1800^{th}$ turn (b) charge density (c) energy density (d) $2100^{th}$ turn (e) charge density (f) energy density (g) $2400^{th}$ turn (h) charge density (i) energy density (j) $2700^{th}$ turn (k) charge density (l) energy density Figure 24: [Scenario (III)] Time evolution of phase space with longitudinal painting starting from $1800^{th}$ turn through $2700^{th}$ turns SCENARIO (IV) [t]Dual RF Harmonics: $f_{rf,~{}1}~{}=~{}53~{}MHz$ and $f_{rf,~{}2}=106~{}MHz$ $H_{2}=1176$ and $H_{1}=588$ $R_{H}=f_{rf,~{}2}/f_{rf,~{}1}=2.0$ $V_{rf,~{}1}$ = 400 (kV) and $V_{rf,~{}2}$ = 200 (kV) (fixed RF voltages) Scenario (IV) is under the same conditions as in scenario (III), except that the secondary RF voltage is 50 $\%$ of the principal RF voltage. It is more advantageous in that the RF waveform has a nice plateau around the stable phase of 0 (deg) as shown in Figure 25. Through the longitudinal painting at each turn, a total of 1080 (270 $\times$ 4) microbunches that are injected over 270 turns and further circulated up to 30 (ms) gradually transform into a continuous macrobunch, or long bunch spanning in between -0.2 (deg) and +0.2 (deg). Unlike in Figure 20 (a), localization of macroparticles is not observed in Figure 27 (a) after the injection of 270 turns is complete. After further circulation of beams up to 2,700 turns with no further injection, the beam charge distribution becomes smoother in the region of -0.2 (deg) and +0.2 (deg). (see Figures 1.19 and 1.20). As a consequence, the bi-modal distribution of charge density observed from the outcome of Scenario (III) is not observed as shown in Figures 22 through 24. As the number of injected macroparticles increases, longitudinal painting progresses, and time elapses, the fine structure of charge distribution gradually disappears. Eventually, after the 2,700 turns, the contour of charge distribution becomes smoother. Including phase offsets without energy jitter, this effect stands out in charge distribution, rather than energy distribution as observed in Scenario (III). Figure 25: [Scenario (IV)] RF waveform for a dual RF system with the RF voltage ratio of 0.5 (a) at the 270-th turn (b) at the 2,700-th turn Figure 26: [Scenario (IV)] Synchronous injection of microbunches with a dual RF harmonic (a) at $270^{th}$ turns and (b) at $2700^{th}$ turns (a) at the 270-th turn (b) at the 2,700-th turn Figure 27: [Scenario (IV)] Distribution of charge density with a dual RF harmonic (a) at $270^{th}$ turns and (b) at $2700^{th}$ turns (a) $1^{st}$ turn (b) $100^{th}$ turn (c) $200^{th}$ turn (d) $270^{th}turn$ (e) $600^{th}$ turn (f) $900^{th}$ turn (g) $1200^{th}turn$ (h) $1500^{th}$ turn (i) $1800^{th}$ turn (j) $2100^{th}turn$ (k) $2400^{th}$ turn (l) $2700^{th}$ turn Figure 28: [Scenario (IV)] Time evolution of phase space with longitudinal painting included, starting from the $1^{st}$ injection turn through 2700 turns (a) $100^{th}$ turn (b) charge density (c) energy density (d) $200^{th}$ turn (e) charge density (f) energy density (g) $270^{th}$ turn (h) charge density (i) energy density (j) $300^{th}$ turn (k) charge density (l) energy density Figure 29: [Scenario (IV)] Time evolution of phase space with longitudinal painting starting from $100^{th}$ turn through $300^{th}$ turns (a) $600^{th}$ turn (b) charge density (c) energy density (d) $900^{th}$ turn (e) charge density (f) energy density (g) $1200^{th}$ turn (h) charge density (i) energy density (j) $1500^{th}$ turn (k) charge density (l) energy density Figure 30: [Scenario (IV)] Time evolution of phase space with longitudinal painting starting from $600^{th}$ turn through $1500^{th}$ turns (a) $1800^{th}$ turn (b) charge density (c) energy density (d) $2100^{th}$ turn (e) charge density (f) energy density (g) $2400^{th}$ turn (h) charge density (i) energy density (j) $2700^{th}$ turn (k) charge density (l) energy density Figure 31: [Scenario (IV)] Time evolution of phase space with longitudinal painting starting from $1,800^{th}$ turn through $2,700^{th}$ turns As Figure 32 shows, the encouraging result is obtained with the dual RF harmonic system. The longitudinal emittance grows about 50 $\%$ less than in the cases of a single RF harmonic system. The peak induced voltage ($\hat{\mathstrut\mathcal{V}}_{sc}$) due to space charge is around 40 (kV) per turn as in Figure 33. Figure 34 shows the evolution of peak voltage induced by space charge. By using the phase modulation in a controlled fashion, which will allow us to maneuver charge distribution, we will be able to lower space- charge voltage further. The bunching factor calculated at each turn for Scenario (IV) turns out to be close to that of Scenario (I) with a single RF harmonic as in Figure 35. It should be noted that the bunching factor calculations are important in that it serves as an indicator of how large tune spreads will be prior to 3-D space-charge calculations. Figure 32: [Scenario (IV)] The growth of longitudinal emittance with a dual RF system over 2,700 turns Figure 33: [Scenario (IV)] Additional $\Delta E$ induced by longitudinal space charge with a dual RF harmonics after 2,700 turns Figure 34: [Scenario (IV)] Evolution of peak voltage in frequency domain with a dual RF harmonics over 2,700 turns Figure 35: [Scenario (IV)] Evolution of bunching factor with a dual RF system ## Concluding Remarks We have investigated different scenarios of microbunch injection-methods between a super-conducting linac and the MI ring under the influence of longitudinal space charge: from the 8-GeV linac proton driver to the Main Injector The RF mismatch between a linac and a ring can induce phase shifts with trains of microbunches, which serve as longitudinal painting in a parasitic fashion. Hence, it would be rather advantageous to use harmonics of non-integer ratio between a linac and a ring in order to induce longitudinal painting. Besides, subsequent charge redistribution in longitudinal direction can be achieved through additional beam circulation with no further beam injection. Since the roaming of charge in longitudinal phase space can reduce the gradient of charge distribution, induced voltage due to space charge can be reduced accordingly. In addition to painting in phase, future simulations are planned to include both phase and energy jitters due to errors in the SC RF linac. Because of the short bunch length of the linac beam, it is anticipated that the impact of the broad-band impedance may play an important role in the longitudinal dynamics[13]. An optimized dual RF system and longitudinal painting can overcome beam-intensity limitations induced by space charge in high-intensity machines. Four injection scenarios manifest that a double RF system with the harmonic ratio ($\mathcal{R}_{H}=1176/588$) of 2.0 and the voltage ratio ($\mathcal{R}_{V}=200kV/400kV$) of 0.5 are most favored. All of the scenarios for the time-structured multiturn injection including animated simulation results are available on a Fermilab web site [14]. ## Acknowledgement Authors wish to thank J. Maclachlan of the Proton Source Department and J-P. Carneiro of Accelerator Physics Center (APC) at Fermilab. Authors had useful discussions with J. Maclachlan for MI injection modeling with macrobunches at an early stage. J-P. Carneiro provided us with an input file of the 8-GeV SRF linac distribution that were generated by the TRACK code[8]. ## Bibliography * [1] P. S. Yoon, Chapter 7, Ph.D. dissertation, Oct. 2007 * [2] G. W. Foster and J. A. MacLachlan, A multi-mission 8-GeV Injector Linac as a Fermilab Booster Replacement, Proceedings of LINAC 2002, Gyeongju, Korea, p. 826. * [3] G. W. Foster, An 8-GeV Superconducting Injector Linac, Proceedings of PAC 2005, Knoxville, TN * [4] G. W. Foster ed., An 8-GeV SC RF Linac Proton Driver Techinical Design Study, http://protondriver.fnal.gov/$\\#Technical_{D}esign_{L}ink$ version 56-1, Nov. 2005. * [5] Fermilab Steering Group, Fermilab Steering Group Report Fermilab-Pub-07-672-DO, Dec. 2007 * [6] Agenda for the $H^{-}$ Transport and Injection Mini-Workshop, December 9-10, 2004, Fermilab, URL: http://www-bd.fnal.gov/pdriver/H-workshop/ * [7] D. E. Johnson, P. S. Yoon, C-J Liaw, D. Raperia, and J. Beebee-Wang An 8-GeV $H^{-}$ Multi-turn Injection System for the Fermilab Main Injector Proc. of PAC 2007, Alberquque, NM; Fermilab-Conf-07-287-AD * [8] P. N. Ostroumov, Physics Design of the 8-GeV $H^{-}$ Linac, New Journal of Physics vol. 8 No.11 $[November2006]$ p. 281, * [9] R. Madrak, Fast Beam Chopper for MEBT in the High Intensity Neutrino Source Fermilab Beams-Doc 2901, April 2007 * [10] J. A. MacLachlan, Multiparticle Dynacmics in the E-$\Phi$ Tracking Code ESME, Fermilab-Conf-02$/$102, 2002 * [11] J. A. MacLachlan, Presentation at the Proton Driver meeting, 2005 * [12] P. S. Yoon, Presentation at the Proton Driver meeting, 2005 * [13] G. P. Jackson, Impact of Microwave Impedance in Proton Driver Beams during Synchrotron Injection, $34^{th}$ ICFA Advanced Beam Dynamics Workshop on High Power Superconducting Ion, Proton, and Multispecies Ions, Naperville, IL May 2005 * [14] P. S. Yoon, The Injection Modeling for the Main Injector URL: http://www-ap.fnal.gov/~syoon/SC/FMI.html

arxiv-papers

2024-09-04T02:48:53.827397

0802.2469

# Optimal Controlled teleportation via several kinds of three-qubit states Ting Gao1,2, Feng-Li Yan2,3, and You-Cheng Li3 1 College of Mathematics and Information Science, Hebei Normal University, Shijiazhuang 050016, China 2 Max-Planck-Institut für Quantenoptik, Hans-Kopfermann-Str. 1, D-85748 Garching, Germany 3 College of Physics and Information Engineering, Hebei Normal University, Shijiazhuang 050016, China ###### Abstract The probability of successfully controlled teleportating an unknown qubit using a general three-particle state is investigated. We give the analytic expressions of maximal probabilities of successfully controlled teleportating an unknown qubit via several kinds of tripartite states including a tripartite GHZ state and a tripartite W-state. ###### pacs: 03.67.Hk, 89.70.+c ## I Introduction Bennett _et al._ BBCJPW showed that an arbitrary unknown state of a qubit could be teleported from a sender to a spatially distant receiver with the aid of long-range Einstein-Podolsky-Rosen (EPR) correlations and the transmission of two bits of classical information. Since then, quantum teleportation has been developed by many authors FujiiPRA2003 ; BaAnPRA2003 ; SLKPpra2001 ; BTBSRBSL ; JBS ; VaidmanPRA1994 ; BKprl1998 ; BDEFMS ; GRpra ; GYWQIC ; KB ; YCHpra2004 ; PatiPRA2000 ; APpla ; PAjob ; DLLZWpra2005 ; YW due to its important applications in quantum communication YZ-EPJB2004 and quantum computation. In past several years quantum teleportation has been also experimentally demonstrated by several groups BPMEWZ ; NKL . The controlled quantum teleportation scheme was presented by Karlsson and Bourennane KB . In the scheme, an unknown state can be perfectly transported from one place to another place via previously shared GHZ state by means of local operations and classical communications (LOCC) under the permission of the third party. The signal state can not be transmitted unless all three sides agree to cooperate. The controlled quantum teleportation is useful in networked quantum information processing and cryptographic conferencing AT ; BHMpra1996 ; Townsend ; BVKpra2003 , and controlled quantum secure direct communication Gaozna and has other interesting applications, such as in opening account on the agreement of managers in a network. Recently, a number of works on controlled quantum teleportation have also been proposed YCHpra2004 ; PatiPRA2000 ; DLLZWpra2005 ; YW , where they restrict themselves to the special quantum channels, such as GHZ state or W state. If a nonmaximally entangled state is taken as quantum channel, then one can not teleport a qubit with unit probability and unit fidelity. However, it is possible to teleport a qubit with a probability $p<1$, which is called probabilistic quantum teleportation APpla ; PAjob . More recently, the probabilistic scheme has been generalized to teleport $N$ qubits GRpra ; GYWQIC . The entanglement property lies at the very heart of quantum information theory. The reason is that entanglement is the physical resource to perform some of the most important quantum information tasks, such as quantum teleportation, quantum computation etc. In VPCirac-prl(04) , Verstraete, Popp, and Cirac introduced a new concept which they called localizable entanglement (LE). This quantity not only has a very well defined physical meaning that treats entanglement as a truly physical resource, but also establishes a very close connection between entanglement and correlation functions. The LE $E_{ij}$ is defined as the maximum of the average entanglement between the spins $i$ and $j$ over all possible outcomes $E_{ij}=\max_{\varepsilon}\sum_{s}p_{s}E(|\phi_{s}\rangle),$ (1) where $p_{s}$ denotes the probability to obtain the two-spin state $|\phi_{s}\rangle$ after performing the measurement $|s\rangle$ in the rest of the system, $E(|\phi_{s}\rangle)$ is the chosen measure of entanglement of $|\phi_{s}\rangle$. The determination of the LE is a formidable task since it involves optimization over all possible local measurement strategies, and thus can not be determined in general. However, Verstaete, Popp, and Cirac gave tight upper bound and lower pound in case of $E(|\phi_{s}\rangle)$ being the concurrence of $|\phi_{s}\rangle$. We determined the exact value of this kind of LE of the general tripartite state, and obtained the analytic expression of another kind of LE, the maximal successful probability of controlled teleporting a qubit of unknown information from a sender to a remote receiver via the control of a third agent by the use of a general three-qubit state GYL1 . In this paper, we give the exact values of the maximal probabilities of successfully controlled teleportating an unknown qubit via many kinds of tripartite states including a tripartite GHZ state and a tripartite W-state. The paper is outlined as follows. In Section II, we present a scheme for controlled probabilistic quantum teleportation of an arbitrary unknown qubit with a general three-qubit state. Furthermore, the successful probability of this teleportation is also obtained. In Section III, we show how to select measuring basis to reach the maximal successful probability of controlled teleporting the complete information about an arbitrary unknown state of a qubit using some kinds of three-qubit states. A brief summary is given in Section IV. ## II The controlled quantum teleportation using a general three-particle state Acín _et al._ AACJLT gave the minimal decomposition of any pure three-qubit state in terms of orthogonal product states built from local bases——a generalization of the two-quantum-bit Schmidt decomposition. They proved that for any pure three-quantum-bit state the existence of local bases which allow one to build a set of five orthogonal product states in terms of which the state can be written in a unique form. That is, for every pure state of a composite system, 123, there exist orthonormal states $|0\rangle_{1}$, $|1\rangle_{1}$ for system 1, orthonormal states $|0\rangle_{2}$, $|1\rangle_{2}$ for system 2, and orthonormal states $|0\rangle_{3}$, $|1\rangle_{3}$ for system 3 such that $\begin{array}[]{ll}|\Psi\rangle_{123}=&a_{0}|000\rangle_{123}+a_{1}e^{\texttt{i}\mu}|100\rangle_{123}+a_{2}|101\rangle_{123}\\\ &+a_{3}|110\rangle_{123}+a_{4}|111\rangle_{123},\\\ &a_{i}\geq 0,~{}0\leq\mu\leq\pi,~{}\Sigma_{i=0}^{4}a_{i}^{2}=1.\\\ \end{array}$ (2) It is uniquely characterized by the five entanglement parameters. Suppose that Alice is to deliver an unknown state to a distant receiver Bob supervised by the controller Charlie via a quantum channel of a normalized general pure three-qubit state in (2), where particle 1 belongs to Charlie, particle 2 is in Alice’s side, while Bob has particle 3. Let $a_{0}\neq 0$ through out the paper. Since if $a_{0}=0$, then $|\Psi\rangle_{123}$ is a tensor product state of a pure state of particle 1 and a pure state of particles 2 and 3, but not a true tripartite entangled state. Bob can get the qubit of quantum information carried by the unknown state only if he obtains the permission of Charlie (i.e., Charlie is trustworthy and cooperative). After getting the approval of Charlie, Alice and Bob begin their teleportation under the control of Charlie. The controller Charlie measures his particle in the basis $\begin{array}[]{c}|x\rangle=\cos\frac{\theta}{2}|0\rangle+e^{\texttt{i}\varphi}\sin\frac{\theta}{2}|1\rangle,\\\ |x\rangle^{\perp}=\sin\frac{\theta}{2}|0\rangle-e^{\texttt{i}\varphi}\cos\frac{\theta}{2}|1\rangle,\\\ \end{array}$ (3) and broadcasts his measurement result. Here $\theta\in[0,\pi],\varphi\in[0,2\pi]$. The tripartite state $|\Psi\rangle_{123}$ can be reexpressed as $\begin{array}[]{lll}|\Psi\rangle_{123}&=&\sqrt{p_{1}}|x\rangle_{1}|\Phi_{1}\rangle_{23}+\sqrt{p_{2}}|x\rangle_{1}^{\perp}|\Phi_{2}\rangle_{23}.\end{array}$ (4) Here $\displaystyle p_{1}$ $\displaystyle=$ $\displaystyle\sin^{2}\frac{\theta}{2}+a_{0}^{2}\cos\theta+a_{0}a_{1}\cos(\mu-\varphi)\sin\theta,$ (5) $\displaystyle p_{2}$ $\displaystyle=$ $\displaystyle\cos^{2}\frac{\theta}{2}-a_{0}^{2}\cos\theta- a_{0}a_{1}\cos(\mu-\varphi)\sin\theta,$ (6) $\displaystyle|\Phi_{1}\rangle_{23}$ $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{p_{1}}}[(a_{0}\cos\frac{\theta}{2}+a_{1}e^{\texttt{i}(\mu-\varphi)}\sin\frac{\theta}{2})|00\rangle_{23}$ (7) $\displaystyle+a_{2}e^{-\texttt{i}\varphi}\sin\frac{\theta}{2}|01\rangle_{23}+a_{3}e^{-\texttt{i}\varphi}\sin\frac{\theta}{2}|10\rangle_{23}$ $\displaystyle+a_{4}e^{-\texttt{i}\varphi}\sin\frac{\theta}{2}|11\rangle_{23}],$ $\displaystyle|\Phi_{2}\rangle_{23}$ $\displaystyle=$ $\displaystyle\frac{1}{\sqrt{p_{2}}}[(a_{0}\sin\frac{\theta}{2}-a_{1}e^{\texttt{i}(\mu-\varphi)}\cos\frac{\theta}{2})|00\rangle_{23}$ (8) $\displaystyle- a_{2}e^{-\texttt{i}\varphi}\cos\frac{\theta}{2}|01\rangle_{23}-a_{3}e^{-\texttt{i}\varphi}\cos\frac{\theta}{2}|10\rangle_{23}$ $\displaystyle- a_{4}e^{-\texttt{i}\varphi}\cos\frac{\theta}{2}|11\rangle_{23}].$ After Charlie’s measurement, the quantum channel is collapsed to $|\Phi_{1}\rangle_{23}$ and $|\Phi_{2}\rangle_{23}$ with probability $p_{1}$ and $p_{2}$, respectively. By Schmidt decomposition, $\displaystyle|\Phi_{1}\rangle_{23}$ $\displaystyle=$ $\displaystyle\sqrt{\lambda_{10}}|0_{2}^{\prime}0_{3}^{\prime}\rangle+\sqrt{\lambda_{11}}|1^{\prime}_{2}1^{\prime}_{3}\rangle,$ (9) $\displaystyle|\Phi_{2}\rangle_{23}$ $\displaystyle=$ $\displaystyle\sqrt{\lambda_{20}}|\bar{0}_{2}\bar{0}_{3}\rangle+\sqrt{\lambda_{21}}|\bar{1}_{2}\bar{1}_{3}\rangle,$ (10) where $\\{0_{2}^{\prime},1_{2}^{\prime}\\}$ and $\\{\bar{0}_{2},\bar{1}_{2}\\}$ ( $\\{0_{3}^{\prime},1_{3}^{\prime}\\}$, and $\\{\bar{0}_{3},\bar{1}_{3}\\}$ ) are orthonormal bases of system 2 (system 3), and Schmidt coefficients $\begin{array}[]{cc}\lambda_{10}=\frac{1-\sqrt{1-C_{1}^{2}}}{2},&\lambda_{11}=\frac{1+\sqrt{1-C_{1}^{2}}}{2},\\\ \lambda_{20}=\frac{1-\sqrt{1-C_{2}^{2}}}{2},&\lambda_{21}=\frac{1+\sqrt{1-C_{2}^{2}}}{2}.\end{array}$ (11) Here $C_{1}=\frac{|a_{0}a_{4}e^{-\texttt{i}\varphi}\sin\theta+2(a_{1}a_{4}e^{\texttt{i}\mu}-a_{2}a_{3})e^{-2i\varphi}\sin^{2}\frac{\theta}{2}|}{p_{1}}$ and $C_{2}=\frac{|a_{0}a_{4}e^{-\texttt{i}\varphi}\sin\theta-2(a_{1}a_{4}e^{\texttt{i}\mu}-a_{2}a_{3})e^{-2i\varphi}\cos^{2}\frac{\theta}{2}|}{p_{2}}$ are the concurrence of $|\Phi_{1}\rangle_{23}$ and $|\Phi_{2}\rangle_{23}$, respectively. For simplicity, we write (9) and (10) as $\displaystyle|\Phi_{1}\rangle_{23}$ $\displaystyle=$ $\displaystyle\sqrt{\lambda_{10}}|0_{2}0_{3}\rangle+\sqrt{\lambda_{11}}|1_{2}1_{3}\rangle,$ (12) $\displaystyle|\Phi_{2}\rangle_{23}$ $\displaystyle=$ $\displaystyle\sqrt{\lambda_{20}}|0_{2}0_{3}\rangle+\sqrt{\lambda_{21}}|1_{2}1_{3}\rangle.$ (13) Suppose that the unknown quantum state the sender Alice wants to teleport to Bob is $|\psi\rangle_{4}=\alpha|0\rangle_{4}+\beta|1\rangle_{4},|\alpha|^{2}+|\beta|^{2}=1.$ (14) If the measurement outcome of Charlie is $|x\rangle_{1}$, then the collect state of particles 2, 3 and 4 is $\begin{array}[]{ll}&|\psi\rangle_{4}|\Phi_{1}\rangle_{23}\\\ =&\sqrt{\lambda_{10}|\alpha|^{2}+\lambda_{11}|\beta|^{2}}[\frac{1}{\sqrt{2}}(|\phi^{+}\rangle_{24}\frac{\alpha\sqrt{\lambda_{10}}|0\rangle_{3}+\beta\sqrt{\lambda_{11}}|1\rangle_{3}}{\sqrt{\lambda_{10}|\alpha|^{2}+\lambda_{11}|\beta|^{2}}}\\\ &+|\phi^{-}\rangle_{24}\frac{\alpha\sqrt{\lambda_{10}}|0\rangle_{3}-\beta\sqrt{\lambda_{11}}|1\rangle_{3}}{\sqrt{\lambda_{10}|\alpha|^{2}+\lambda_{11}|\beta|^{2}}})]\\\ &+\sqrt{\lambda_{11}|\alpha|^{2}+\lambda_{10}|\beta|^{2}}[\frac{1}{\sqrt{2}}(|\psi^{+}\rangle_{24}\frac{\beta\sqrt{\lambda_{10}}|0\rangle_{3}+\alpha\sqrt{\lambda_{11}}|1\rangle_{3}}{\sqrt{\lambda_{11}|\alpha|^{2}+\lambda_{10}|\beta|^{2}}}\\\ &+|\psi^{-}\rangle_{24}\frac{\beta\sqrt{\lambda_{10}}|0\rangle_{3}-\alpha\sqrt{\lambda_{11}}|1\rangle_{3}}{\sqrt{\lambda_{11}|\alpha|^{2}+\lambda_{10}|\beta|^{2}}})],\end{array}$ (15) where $|\phi^{\pm}\rangle=\frac{1}{\sqrt{2}}(|00\rangle\pm|11\rangle),|\psi^{\pm}\rangle=\frac{1}{\sqrt{2}}(|01\rangle\pm|10\rangle)$. Alice makes a Bell measurement on her particles 2 and 4. She obtains $|\phi^{+}\rangle_{24}$, $|\phi^{-}\rangle_{24}$, $|\psi^{+}\rangle_{24}$, and $|\psi^{-}\rangle_{24}$ with probability $\left(\frac{\sqrt{\lambda_{10}|\alpha|^{2}+\lambda_{11}|\beta|^{2}}}{\sqrt{2}}\right)^{2}$, $\left(\frac{\sqrt{\lambda_{10}|\alpha|^{2}+\lambda_{11}|\beta|^{2}}}{\sqrt{2}}\right)^{2}$, $\left(\frac{\sqrt{\lambda_{11}|\alpha|^{2}+\lambda_{10}|\beta|^{2}}}{\sqrt{2}}\right)^{2}$, and $\left(\frac{\sqrt{\lambda_{11}|\alpha|^{2}+\lambda_{10}|\beta|^{2}}}{\sqrt{2}}\right)^{2}$, respectively. Then she conveys her measurement outcome to Bob over a classical communication channel. In order to achieve teleportation, Bob needs to introduce an auxiliary particle $b$ with the initial state $|0\rangle_{b}$ and performs a collective unitary transformation $U_{3b}=\left(\begin{array}[]{cccc}1&0&0&0\\\ 0&1&0&0\\\ 0&0&\frac{\sqrt{\lambda_{10}}}{\sqrt{\lambda_{11}}}&\sqrt{1-\frac{\lambda_{10}}{\lambda_{11}}}\\\ 0&0&-\sqrt{1-\frac{\lambda_{10}}{\lambda_{11}}}&\frac{\sqrt{\lambda_{10}}}{\sqrt{\lambda_{11}}}\\\ \end{array}\right)$ on the state of particles 3 and $b$. Then the measurement on his auxiliary particle $b$ follows. If his measurement result is $|0\rangle_{b}$, Bob can fix up the state of his particle 3, recovering $|\psi\rangle$, by applying an appropriate local unitary operation. The achievable successful probability of teleporting the unknown state in (14) via $|\Phi_{1}\rangle_{23}$ is $2\left(\frac{\sqrt{\lambda_{10}|\alpha|^{2}+\lambda_{11}|\beta|^{2}}}{\sqrt{2}}\right)^{2}\left(\frac{\sqrt{\lambda_{10}}}{\sqrt{\lambda_{10}|\alpha|^{2}+\lambda_{11}|\beta|^{2}}}\right)^{2}+2\left(\frac{\sqrt{\lambda_{11}|\alpha|^{2}+\lambda_{10}|\beta|^{2}}}{\sqrt{2}}\right)^{2}\left(\frac{\sqrt{\lambda_{10}}}{\sqrt{\lambda_{11}|\alpha|^{2}+\lambda_{10}|\beta|^{2}}}\right)^{2}=2\lambda_{10}$. Similarly, if the measurement result of Charlie is $|x\rangle^{\perp}_{1}$, the achievable successful probability of teleporting the state in (14) via $|\Phi_{2}\rangle_{23}$ is $2\lambda_{20}$. Therefore, probability $p$ of successfully controlled teleporting an unknown qubit (14) using a general three-particle state in (2) is $\begin{array}[]{rl}p=&2p_{1}\lambda_{10}+2p_{2}\lambda_{20}\\\ =&1-\left(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right).\end{array}$ (16) Here $\begin{array}[]{ll}&P(\theta,\varphi)=p_{1}^{2}(1-C_{1}^{2})\\\ =&\frac{1}{4}a_{0}^{2}a_{1}^{2}\cos 2(\varphi-\mu)+3{a_{1}}{a_{2}}{a_{3}}{a_{4}}\cos\mu+\frac{1}{8}(3-4a_{0}^{2}\\\ &+4a_{0}^{4}+2a_{0}^{2}a_{1}^{2}-12a_{2}^{2}a_{3}^{2}-4a_{0}^{2}a_{4}^{2}-12a_{1}^{2}a_{4}^{2})\\\ &+\frac{1}{8}\cos 2\theta(1-4a_{0}^{2}+4a_{0}^{4}-2a_{0}^{2}a_{1}^{2}-2a_{0}^{2}a_{1}^{2}\cos 2(\varphi-\mu)\\\ &-4a_{2}^{2}a_{3}^{2}+8{a_{1}}{a_{2}}{a_{3}}{a_{4}}\cos\mu+4a_{0}^{2}a_{4}^{2}-4a_{1}^{2}a_{4}^{2})\\\ &-\cos\theta(\frac{1}{2}-a_{0}^{2}-2a_{2}^{2}a_{3}^{2}+4{a_{1}}{a_{2}}{a_{3}}{a_{4}}\cos\mu-2a_{1}^{2}a_{4}^{2})\\\ &+{a_{0}}\sin\theta[2{a_{2}}{a_{3}}{a_{4}}\cos\varphi+{a_{1}}(1-2a_{4}^{2})\cos(\varphi-\mu)]\\\ &-\frac{1}{2}{a_{0}}\sin 2\theta[2{a_{2}}{a_{3}}{a_{4}}\cos\varphi+{a_{1}}(1-2a_{0}^{2}-2a_{4}^{2})\cos(\varphi-\mu)],\end{array}$ (17) $\begin{array}[]{ll}&Q(\theta,\varphi)=p_{1}^{2}(1-C_{2}^{2})\\\ =&\frac{1}{4}a_{0}^{2}a_{1}^{2}\cos 2(\varphi-\mu)+3{a_{1}}{a_{2}}{a_{3}}{a_{4}}\cos\mu+\frac{1}{8}(3-4a_{0}^{2}\\\ &+4a_{0}^{4}+2a_{0}^{2}a_{1}^{2}-12a_{2}^{2}a_{3}^{2}-4a_{0}^{2}a_{4}^{2}-12a_{1}^{2}a_{4}^{2})\\\ &+\frac{1}{8}\cos 2\theta(1-4a_{0}^{2}+4a_{0}^{4}-2a_{0}^{2}a_{1}^{2}-2a_{0}^{2}a_{1}^{2}\cos 2(\varphi-\mu)\\\ &-4a_{2}^{2}a_{3}^{2}+8{a_{1}}{a_{2}}{a_{3}}{a_{4}}\cos\mu+4a_{0}^{2}a_{4}^{2}-4a_{1}^{2}a_{4}^{2})\\\ &+\cos\theta(\frac{1}{2}-a_{0}^{2}-2a_{2}^{2}a_{3}^{2}+4{a_{1}}{a_{2}}{a_{3}}{a_{4}}\cos\mu-2a_{1}^{2}a_{4}^{2})\\\ &-{a_{0}}\sin\theta[2{a_{2}}{a_{3}}{a_{4}}\cos\varphi+{a_{1}}(1-2a_{4}^{2})\cos(\varphi-\mu)]\\\ &-\frac{1}{2}{a_{0}}\sin 2\theta(2{a_{2}}{a_{3}}{a_{4}}\cos\varphi+{a_{1}}(1-2a_{0}^{2}-2a_{4}^{2})\cos(\varphi-\mu)).\end{array}$ (18) Obviously, $\begin{array}[]{lllll}Q(\theta,\varphi)&=&P(\pi-\theta,\varphi+\pi)&\texttt{if}&\varphi\in[0,\pi],\\\ Q(\theta,\varphi)&=&P(\pi-\theta,\varphi-\pi)&\texttt{if}&\varphi\in[\pi,2\pi],\end{array}$ (19) $\begin{array}[]{lllll}P(\theta,\varphi)&=&Q(\pi-\theta,\varphi+\pi)&\texttt{if}&\varphi\in[0,\pi],\\\ P(\theta,\varphi)&=&Q(\pi-\theta,\varphi-\pi)&\texttt{if}&\varphi\in[\pi,2\pi],\end{array}$ (20) and $\begin{array}[]{lll}\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)}&=&\sqrt{P(\pi,\varphi)}+\sqrt{Q(\pi,\varphi)},\\\ \sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}&=&\sqrt{P(\theta,2\pi)}+\sqrt{Q(\theta,2\pi)}.\end{array}$ (21) It is clear that the successful probability is the same as the above (16) if the operation order is changed. That is, if Alice makes a Bell state measurement on her particles 2 and 4 first, the third party Charlie’s measurement on his particle 1 follows, after that Bob operates his particle to acquire a qubit of quantum information Alice sends, instead of the above operation order, then they can also achieve the successful probability (16) of controlled teleportation. ## III The maximal successful probability of controlled quantum teleportation using three-particle states with $a_{1}a_{2}a_{3}a_{4}\sin\mu=0$ In GYL1 , we determined the analytic expression of the maximal successful probability of controlled teleportation by using the general tripartite state (2) with $a_{0}a_{1}a_{2}a_{3}a_{4}\sin\mu\neq 0$. In this section, we give the analytic expression of the localizable entanglement (LE), the maximum of probability of successfully controlled teleporting an unknown qubit state (14) via every three-qubit state (2) satisfying $a_{1}a_{2}a_{3}a_{4}\sin\mu=0$ and investigate how to achieve it (that is, Charlie finds optimal measurement basis). Obviously, the maximum of (16) is $\begin{array}[]{rl}p_{\max}=&\max\\{p\\}\\\ =&\max\\{2p_{1}\lambda_{10}+2p_{2}\lambda_{20}\\}\\\ =&1-\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}.\end{array}$ (22) In order to get the maximum $p_{\max}$ of $p$ in (22), we need only obtain the minimum of $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}$ $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}.\\\ \end{array}$ (23) In other words, to reach the maximal probability of exact controlled teleportation through an arbitrary partially entangled quantum channel (2), the supervisor Charlie needs only to choose optimal measurement basis, i.e. he selects $\theta_{0}$ and $\varphi_{0}$ such that $\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}=\sqrt{P(\theta_{0},\varphi_{0})}+\sqrt{Q(\theta_{0},\varphi_{0})}$. Note that the minimum of $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}$ should occur at the points such that $P(\theta,\varphi)=0$, $Q(\theta,\varphi)=0$, $\displaystyle\frac{\partial\left(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right)}{\partial\theta}$ $\displaystyle=$ $\displaystyle 0,$ (24) $\displaystyle\frac{\partial\left(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right)}{\partial\varphi}$ $\displaystyle=$ $\displaystyle 0,$ (25) or, the boundary of $\theta$ and $\varphi$. From Eq.(24) and Eq.(25), there are $\displaystyle\frac{\partial P(\theta,\varphi)}{\partial\theta}\frac{\partial Q(\theta,\varphi)}{\partial\varphi}-\frac{\partial Q(\theta,\varphi)}{\partial\theta}\frac{\partial P(\theta,\varphi)}{\partial\varphi}$ $\displaystyle=$ $\displaystyle 0,$ (26) $\displaystyle P(\theta,\varphi)\left(\frac{\partial Q(\theta,\varphi)}{\partial\varphi}\right)^{2}-Q(\theta,\varphi)\left(\frac{\partial P(\theta,\varphi)}{\partial\varphi}\right)^{2}$ $\displaystyle=$ $\displaystyle 0,$ (27) $\displaystyle P(\theta,\varphi)\left(\frac{\partial Q(\theta,\varphi)}{\partial\theta}\right)^{2}-Q(\theta,\varphi)\left(\frac{\partial P(\theta,\varphi)}{\partial\theta}\right)^{2}$ $\displaystyle=$ $\displaystyle 0.$ (28) Let $y=\cot\theta$, $t=\cot\frac{\theta}{2}$, $\theta\in(0,\pi)$, then $\sin\theta=\frac{1}{\sqrt{y^{2}+1}}=\frac{2t}{1+t^{2}}$,$\cos\theta=\frac{y}{\sqrt{1+y^{2}}}=\frac{t^{2}-1}{t^{2}+1}$,$t=y+\sqrt{y^{2}+1}$, $t\in(0,+\infty)$. These will be useful throughout the paper. Next we give controller’s optimal measurement basis (3) for every kind of given quantum channel (2) satisfying $a_{1}a_{2}a_{3}a_{4}\sin\mu=0$. That is, we determine the two parameters $\theta$ and $\varphi$ in measurement basis (3). To state clearly, we classify the quantum channel (2) into the following cases. Next we examine quantum channels (2) with the following different characterizations, give the maximal probability of exact teleportation via any three-qubit state (2) with $a_{1}a_{2}a_{3}a_{4}\sin\mu=0$, and characterizes the tripartite states that can collapse to an EPR pair after Charlie’s measurement. ### III.1 $a_{1}=a_{2}=a_{3}=0$, and $a_{0}a_{4}\neq 0$ The quantum channel (2) with three coefficients being 0 is the only one satisfying $a_{1}=a_{2}=a_{3}=0$ and $a_{0}a_{4}\neq 0$, since others are biseparable (one party is not entangled with the other two parties) and can not be used as quantum channel of controlled teleportation. It can be seen that $\begin{array}[]{ll}&\min\left\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right\\}\\\ =&\min\big{\\{}\sqrt{\frac{1}{4}[\cos\theta-(1-2a_{0}^{2})]^{2}}\\\ &~{}~{}~{}~{}~{}~{}+\sqrt{\frac{1}{4}[\cos\theta+(1-2a_{0}^{2})]^{2}}\big{\\}}\\\ =&|1-2a_{0}^{2}|,\\\ \end{array}$ (29) for each $\varphi\in[0,2\pi]$, and all $\theta$ satisfying $|\cos\theta|\leq|1-2a_{0}^{2}|$. It implies that $p_{\max}=1-|1-2a_{0}^{2}|.$ (30) Note that if $\cos\theta=1-2a_{0}^{2}\neq 0$ or $\cos\theta=-1+2a_{0}^{2}\neq 0$, then $P(\theta,\varphi)=0$ or $Q(\theta,\varphi)=0$, which means that after Charlie’s measurement the state of Alice’s particles 2 and 3 can be an EPR pair with probability $2a_{0}^{2}(1-a_{0}^{2})$. Moreover, if $\cos\theta=1-2a_{0}^{2}=0$ or $\cos\theta=-1+2a_{0}^{2}=0$, i.e. $\theta=\frac{\pi}{2}$, $a_{0}=a_{4}=\frac{1}{\sqrt{2}}$, then $P(\theta,\varphi)=Q(\theta,\varphi)=0$, that is, after Charlie measures quantum channel in basis $\\{\frac{|0\rangle+e^{\texttt{i}\varphi}|1\rangle}{\sqrt{2}},\frac{|0\rangle-e^{\texttt{i}\varphi}|1\rangle}{\sqrt{2}}\\}$, particles 2 and 3 are collapsed to a Bell state with probability 1. It follows that perfect quantum teleportation can be achieved if $a_{0}=\frac{1}{\sqrt{2}}$ (i.e quantum channel is in GHZ state) and $\theta=\frac{\pi}{2}$. That is, one can send perfect unknown state to another using GHZ state as a quantum channel via controller’s measurement in the basis $\\{\frac{|0\rangle+e^{\texttt{i}\varphi}|1\rangle}{\sqrt{2}},\frac{|0\rangle-e^{\texttt{i}\varphi}|1\rangle}{\sqrt{2}}\\}$, where $\varphi\in[0,2\pi]$. ### III.2 $a_{1}=a_{4}=0$, and $a_{0}a_{2}a_{3}\neq 0$ Here we consider the quantum channel with coefficients satisfying $a_{1}=a_{4}=0$ and $a_{0}a_{2}a_{3}\neq 0$. These states are called tri-Bell states in AACJLT . Note that W-state $|W\rangle=\frac{1}{\sqrt{3}}(|001\rangle+|010\rangle+|100\rangle)$ (31) is contained here, since the quantum channel (2) in case of $a_{1}=a_{4}=0$ and $a_{0}=a_{2}=a_{3}=\frac{1}{\sqrt{3}}$ is LOCC equivalent to W-state. If $P(\theta,\varphi)=0$, then $\cos\theta=\frac{1+2{a_{2}}{a_{3}}}{1-2a_{0}^{2}+2{a_{2}}{a_{3}}}$, or $\cos\theta=\frac{-1+2{a_{2}}{a_{3}}}{-1+2a_{0}^{2}+2{a_{2}}{a_{3}}}$. It is not difficult to prove that $|\frac{1+2{a_{2}}{a_{3}}}{1-2a_{0}^{2}+2{a_{2}}{a_{3}}}|>1$, $|\frac{-1+2{a_{2}}{a_{3}}}{-1+2a_{0}^{2}+2{a_{2}}{a_{3}}}|>1$ if $a_{2}\neq a_{3}$, and $\frac{-1+2{a_{2}}{a_{3}}}{-1+2a_{0}^{2}+2{a_{2}}{a_{3}}}=-1$ if $a_{2}=a_{3}$. Thus, $P(\theta,\varphi)=0$ if and only if $a_{2}=a_{3}$, and $\theta=\pi$. From (19), there is $Q(\theta,\varphi)=0$ if and only if $a_{2}=a_{3}$, and $\theta=0$. Combining (19) and (20), there is $\begin{array}[]{ll}&\left(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right)\left|{}_{P(\theta,\varphi)=0}\right.\\\ =&\left(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right)\left|{}_{Q(\theta,\varphi)=0}\right.\\\ =&a_{0}^{2}.\end{array}$ (32) Therefore, for the quantum channel (2) such that $a_{1}=a_{4}=0$, $a_{2}=a_{3}$ and $a_{0}a_{2}a_{3}\neq 0$, can be collapsed to a Bell state with probability $p_{1}=p_{2}=1-a_{0}^{2}$ by Charlie measuring his particle in the basis $\\{|0\rangle,|1\rangle\\}$. Next we suppose that $P(\theta,\varphi)\neq 0$ and $Q(\theta,\varphi)\neq 0$. From (28), we derive that $a_{0}^{4}a_{2}^{2}a_{3}^{2}\left(-1+2a_{0}^{2}+4a_{2}^{2}a_{3}^{2}\right)\cos\theta\sin^{2}\theta=0$. Thus, $\theta=0,\frac{\pi}{2},\pi$. By checking, $\theta=0,\frac{\pi}{2},\pi$ are roots of (24). Note that if $\theta=0,\pi$, then $a_{2}\neq a_{3}$ by $P(\theta,\varphi)\neq 0$ and $Q(\theta,\varphi)\neq 0$. Obviously, $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}=a_{0}^{2}+|a_{2}^{2}-a_{3}^{2}|>0$ if $\theta=0,\pi$ and $a_{2}\neq a_{3}$; $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}=\sqrt{1-4a_{2}^{2}a_{3}^{2}}>0$ if $\theta=\frac{\pi}{2}$. Therefore, $\begin{array}[]{ll}&\min\left\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right\\}\\\ =&\min\big{\\{}a_{0}^{2}+|a_{2}^{2}-a_{3}^{2}|,\sqrt{1-4a_{2}^{2}a_{3}^{2}}\big{\\}}\\\ >&0.\end{array}$ (33) That is, for quantum channel (2) with $a_{1}=a_{4}=0$, $a_{0}a_{2}a_{3}\neq 0$ and $a_{0}^{2}+|a_{2}^{2}-a_{3}^{2}|<\sqrt{1-4a_{2}^{2}a_{3}^{2}}$, the controller should choose measurement basis $\\{|0\rangle,|1\rangle\\}$; otherwise, he selects measurement basis (3) with $\theta=\frac{\pi}{2}$ (i.e. he measures in the basis $\\{\frac{|0\rangle+e^{\texttt{i}\varphi}|1\rangle}{\sqrt{2}},\frac{|0\rangle-e^{\texttt{i}\varphi}|1\rangle}{\sqrt{2}}\\}$, where $\varphi\in[0,2\pi]$), thus, the controlled teleportation in Section II can achieve maximal successful probability $p_{\max}<1$. ### III.3 one is $a_{1}=a_{2}=0$ and $a_{0}a_{3}a_{4}\neq 0$, the other is $a_{1}=a_{3}=0$ and $a_{0}a_{2}a_{4}\neq 0$ Now we discuss the quantum channel (2) with the characterization $a_{1}=a_{2}=0$ and $a_{0}a_{3}a_{4}\neq 0$. First, $P(\theta,\varphi)\neq 0$ and $Q(\theta,\varphi)\neq 0$. If $P(\theta,\varphi)=0$, then $\cos\theta=\frac{1-2a_{0}^{2}-4a_{0}^{2}{a_{4}}{\sqrt{-a_{3}^{2}}}}{1-4a_{0}^{2}+4a_{0}^{4}+4a_{0}^{2}a_{4}^{2}}$, or $\cos\theta=\frac{1-2a_{0}^{2}+4a_{0}^{2}{a_{4}}{\sqrt{-a_{3}^{2}}}}{1-4a_{0}^{2}+4a_{0}^{4}+4a_{0}^{2}a_{4}^{2}}$, which are impossible. Thus, $P(\theta,\varphi)\neq 0$. By (19), there is also $Q(\theta,\varphi)\neq 0$. The quantum channel (2) with the characterization $a_{1}=a_{2}=0$ and $a_{0}a_{3}a_{4}\neq 0$ can never collapse to an EPR pair after Charlie’s measurement. Second, (28) implies $a_{0}^{4}(1-2a_{0}^{2})a_{4}^{2}a_{3}^{2}\cos\theta\sin^{2}\theta=0$. It follows that $\theta=0,\frac{\pi}{2}$, $\pi$. By checking, $\theta=0,\frac{\pi}{2},\pi$ are the roots of (24). We can get $\begin{array}[]{ll}\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}&=\sqrt{P(\frac{\pi}{2},\varphi)}+\sqrt{Q(\frac{\pi}{2},\varphi)}\\\ &=\sqrt{1-4a_{0}^{2}a_{4}^{2}}>0.\\\ \end{array}$ (34) $p_{\max}=1-\sqrt{1-4a_{0}^{2}a_{4}^{2}}<1,$ (35) where $\theta=\frac{\pi}{2}$. Similarly, when $a_{1}=a_{3}=0$, and $a_{0}a_{2}a_{4}\neq 0$, we have $p_{\max}=1-\sqrt{1-4a_{0}^{2}a_{4}^{2}}<1,$ (36) where $\theta=\frac{\pi}{2}$. That is, the controlled teleportation via quantum channel (2) satisfying $a_{1}=a_{2}=0$ and $a_{0}a_{3}a_{4}\neq 0$, or $a_{1}=a_{3}=0$ and $a_{0}a_{2}a_{4}\neq 0$ can only succeed with optimal probability $p_{\max}=1-\sqrt{1-4a_{0}^{2}a_{4}^{2}}<1$. ### III.4 one is $a_{2}=a_{4}=0$ and $a_{0}a_{1}a_{3}\neq 0$, the other is $a_{3}=a_{4}=0$ and $a_{0}a_{1}a_{2}\neq 0$ For the quantum channel with characterization either $a_{2}=a_{4}=0$ and $a_{0}a_{1}a_{3}\neq 0$, or $a_{3}=a_{4}=0$ and $a_{0}a_{1}a_{2}\neq 0$, since $P(\theta,\varphi)=p_{1}^{2}$, $Q(\theta,\varphi)=p_{2}^{2}$, so $\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}=1$, and $p_{\max}=0$ for both cases. It can also be seen directly from the quantum channel (2) with these two characterizations being biseparable states. ### III.5 $a_{2}=a_{3}=0$ and $a_{0}a_{1}a_{4}\neq 0$ The quantum channel (2) with coefficients satisfying $a_{2}=a_{3}=0$, but $a_{0}a_{1}a_{4}\neq 0$, are extended GHZ states according to the classification in AACJLT . For this kind of quantum channel, we can derive that $\displaystyle P(\theta,\varphi)$ $\displaystyle=$ $\displaystyle\frac{1}{4}[1-2a_{4}^{2}+2a_{0}a_{1}\cos(\varphi-\mu)\sin\theta$ (37) $\displaystyle-(1-2a_{0}^{2}-2a_{4}^{2})\cos\theta]^{2},$ $\displaystyle Q(\theta,\varphi)$ $\displaystyle=$ $\displaystyle\frac{1}{4}[-1+2a_{4}^{2}+2a_{0}a_{1}\cos(\varphi-\mu)\sin\theta$ (38) $\displaystyle-(1-2a_{0}^{2}-2a_{4}^{2})\cos\theta]^{2},$ and $\begin{array}[]{c}p_{\max}=1-\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}=1-|1-2a_{4}^{2}|,\end{array}$ (39) where $\theta,\varphi$ satisfy $|2a_{0}a_{1}\cos(\varphi-\mu)\sin\theta-(1-2a_{0}^{2}-2a_{4}^{2})\cos\theta|\leq|1-2a_{4}^{2}|$. Note that the set $S$ of $(\theta,\varphi)$ such that $|2a_{0}a_{1}\cos(\varphi-\mu)\sin\theta-(1-2a_{0}^{2}-2a_{4}^{2})\cos\theta|\leq|1-2a_{4}^{2}|$ is a region, as we can see easily that $(\frac{\pi}{2},\frac{\pi}{2}+\mu)$, $(\frac{\pi}{2},\frac{3\pi}{2}+\mu)\in S$ in case of $0\leq\mu\leq\frac{\pi}{2}$, and $(\frac{\pi}{2},\frac{\pi}{2}+\mu)$, $(\frac{\pi}{2},-\frac{\pi}{2}+\mu)\in S$ in case of $\frac{\pi}{2}\leq\mu\leq\pi$. From (37) and (38), we have that $P(\theta,\varphi)=Q(\theta,\varphi)=0$ if and only if $a_{4}=\frac{1}{\sqrt{2}}$ and $a_{1}\cos(\varphi-\mu)\sin\theta+a_{0}\cos\theta=0$, which means that as long as Charlie measures the quantum channel $a_{0}|000\rangle+a_{1}e^{\texttt{i}\mu}|100\rangle+a_{4}|111\rangle,~{}~{}a_{0}^{2}+a_{1}^{2}=a_{4}^{2}=\frac{1}{2},$ (40) in the basis (3) satisfying $a_{1}\cos(\varphi-\mu)\sin\theta+a_{0}\cos\theta=0$, where $\theta\in[0,\pi]$, $\varphi\in[0,2\pi]$, Alice and Bob can obtain an EPR pair with certainty (i.e. with probability $p=1$), and Alice teleports her one qubit information to Bob with probability 1 and with unit fidelity. Note that the state $a_{0}|000\rangle+a_{1}|100\rangle+a_{4}|111\rangle,~{}~{}a_{0}^{2}+a_{1}^{2}=a_{4}^{2}=\frac{1}{2},$ (41) is LOCC equivalent to the state in (40). That is, the states in (41) or (40) can be used for perfect controlled quantum teleportation. Clearly, one can achieve perfect teleportation via these states by the controller making a measurement on his particle using the basis (3) satisfying $a_{1}\cos(\varphi-\mu)\sin\theta+a_{0}\cos\theta=0$. More important, this kind of states in (41) or (40) are different from GHZ state according to the classification in AACJLT . ### III.6 $a_{1}=0$ and $a_{0}a_{2}a_{3}a_{4}\neq 0$ In this section, we investigate the quantum channel with coefficients having $a_{1}=0$ and $a_{0}a_{2}a_{3}a_{4}\neq 0$. Evidently, $\begin{array}[]{ll}&\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)}\\\ =&\sqrt{P(\pi,\varphi)}+\sqrt{Q(\pi,\varphi)}\\\ =&a_{0}^{2}+\sqrt{(1-a_{0}^{2})^{2}-4a_{2}^{2}a_{3}^{2}}\\\ =&a_{0}^{2}+\sqrt{[a_{4}^{2}+(a_{2}-a_{3})^{2}][a_{4}^{2}+(a_{2}+a_{3})^{2}]},\end{array}$ (42) and $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$ in case of $\theta=0,\pi$. We first examine the condition of $P(\theta,\varphi)=0$. If $P(\theta,\varphi)=0$, then $\begin{array}[]{ll}\cos\varphi=&[2\sin\theta(1-\cos\theta){a_{0}}{a_{2}}{a_{3}}{a_{4}}]^{-1}[-\frac{1}{4}+a_{2}^{2}a_{3}^{2}\\\ &+a_{0}^{2}a_{4}^{2}+\cos\theta(\frac{1}{2}-a_{0}^{2}-2a_{2}^{2}a_{3}^{2})\\\ &-\frac{1}{4}{\cos^{2}\theta}(1-4a_{0}^{2}+4a_{0}^{4}-4a_{2}^{2}a_{3}^{2}+4a_{0}^{2}a_{4}^{2})].\end{array}$ (43) Note that $\begin{array}[]{rl}z=&[2\sin\theta(1-\cos\theta){a_{0}}{a_{2}}{a_{3}}{a_{4}}]^{-1}[-\frac{1}{4}+a_{2}^{2}a_{3}^{2}\\\ &+a_{0}^{2}a_{4}^{2}+\cos\theta(\frac{1}{2}-a_{0}^{2}-2a_{2}^{2}a_{3}^{2})\\\ &-\frac{1}{4}{\cos^{2}\theta}(1-4a_{0}^{2}+4a_{0}^{4}-4a_{2}^{2}a_{3}^{2}+4a_{0}^{2}a_{4}^{2})]\\\ =&\frac{-a_{0}^{4}t^{4}+(-2a_{0}^{2}+2a_{0}^{4}+4a_{0}^{2}a_{4}^{2})t^{2}-{(1-a_{0}^{2})}^{2}+4a_{2}^{2}a_{3}^{2}}{8t{a_{0}}{a_{2}}{a_{3}}{a_{4}}}\end{array}$ (44) tends to $-\infty$ when $t$ tends to 0 and $+\infty$, and $z=-1$ holds only if $a_{2}=a_{3}$ and $t=\frac{a_{4}}{a_{0}}$. It follows that $z\leq-1$, where the equality occurs if and only if $a_{2}=a_{3}$ and $\theta=\theta_{0}$, where $\cot\frac{\theta_{0}}{2}=\frac{a_{4}}{a_{0}}$ and $0<\theta_{0}<\pi$. Therefore, only if $a_{2}=a_{3}$, $\varphi=\pi$, $\theta=\theta_{0}$, there is $P(\theta,\varphi)=0$. By (19) and (20), we have $Q(\theta,\varphi)=0$ iff $a_{2}=a_{3}$, $\varphi=0$, and $\theta=\pi-\theta_{0}$, and $\begin{array}[]{ll}&\big{(}\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\big{)}\big{|}_{P(\theta,\varphi)=0}\\\ =&\sqrt{Q(\theta_{0},\pi)}|_{a_{2}=a_{3}}\\\ =&\big{(}\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\big{)}\big{|}_{Q(\theta,\varphi)=0}\\\ =&\sqrt{P(\pi-\theta_{0},0)}|_{a_{2}=a_{3}}\\\ =&\frac{\sqrt{a_{0}^{8}+2a_{0}^{4}a_{4}^{2}-2a_{0}^{6}a_{4}^{2}+a_{4}^{4}-2a_{0}^{2}a_{4}^{4}-3a_{0}^{4}a_{4}^{4}+8a_{0}^{2}a_{3}^{2}a_{4}^{4}-4a_{3}^{4}a_{4}^{4}}}{a_{0}^{2}+a_{4}^{2}}\\\ =&\sqrt{1-4a_{3}^{2}+4a_{3}^{4}-4a_{4}^{2}+12a_{3}^{2}a_{4}^{2}+4a_{4}^{4}}.\end{array}$ (45) That is, $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}=\sqrt{1-4a_{3}^{2}+4a_{3}^{4}-4a_{4}^{2}+12a_{3}^{2}a_{4}^{2}+4a_{4}^{4}}$ if $P(\theta,\varphi)=0$ or $Q(\theta,\varphi)=0$. It is shown that an EPR pair can be obtained with probability $p_{1}=p_{2}=\frac{a_{0}^{2}(1-a_{0}^{2}+a_{4}^{2})}{a_{0}^{2}+a_{4}^{2}}$ if $a_{2}=a_{3}$ and Charlie using measurement basis (3) with $(\theta,\varphi)=(\theta_{0},\pi)$, or $(\theta,\varphi)=(\pi-\theta_{0},0)$, where $\cot\frac{\theta_{0}}{2}=\frac{a_{4}}{a_{0}}$ and $0<\theta_{0}<\pi$. Next we assume that $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$. From (26) and (27), there are $\sin\varphi[2a_{2}{a_{3}}{a_{4}}\cos\varphi\sin\theta-2a_{0}^{3}\cos\theta+{{{a}}_{0}}\cos\theta(1-2a_{4}^{2})]=0$ and $\begin{array}[]{rc}\sin\varphi[{a_{0}}(1-2a_{4}^{2})\cos\theta{\sin^{2}\theta}&\\\ +2a_{0}^{3}{\cos^{3}\theta}+2{a_{2}}{a_{3}}{a_{4}}\cos\varphi{\sin^{3}\theta}]&=0,\\\ \end{array}$ respectively. By $2a_{2}{a_{3}}{a_{4}}\cos\varphi\sin\theta-2a_{0}^{3}\cos\theta+{{{a}}_{0}}\cos\theta(1-2a_{4}^{2})=0$, and ${a_{0}}(1-2a_{4}^{2})\cos\theta{\sin^{2}\theta}+2a_{0}^{3}{\cos^{3}\theta}+2{a_{2}}{a_{3}}{a_{4}}\cos\varphi{\sin^{3}\theta}=0$, there are $\theta=\frac{\pi}{2},\varphi=\frac{\pi}{2},\frac{3\pi}{2}$. Evidently, $\begin{array}[]{ll}\sqrt{P(\frac{\pi}{2},\frac{\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{\pi}{2})}&=\sqrt{P(\frac{\pi}{2},\frac{3\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{3\pi}{2})}\\\ &=\sqrt{1-4a_{2}^{2}a_{3}^{2}-4a_{0}^{2}a_{4}^{2}}.\\\ \end{array}$ (46) Next we need only to examine the case $\sin\varphi=0$, i.e. $\varphi=0,\pi,2\pi$, while $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$ and $\theta\in(0,\pi)$. Now let us consider the case $\varphi=0$. From (28), we obtain $\begin{array}[]{ll}&-4a_{0}^{3}\left(2ya_{0}a_{2}a_{3}+a_{4}\right)[4y^{2}a_{0}^{2}a_{2}a_{3}a_{4}^{2}+2ya_{0}(2a_{0}^{4}+\\\ &2a_{0}^{2}a_{4}^{2}-3a_{0}^{2}-4a_{2}^{2}a_{3}^{2}-a_{4}^{2}+1)a_{4}\\\ &+a_{2}a_{3}(2a_{0}^{2}-4a_{0}^{2}a_{4}^{2}+4a_{2}^{2}a_{3}^{2}+2a_{4}^{2}-1)]\\\ =&0,\end{array}$ (47) which implies that $\displaystyle y$ $\displaystyle=$ $\displaystyle\cot\theta_{1}=-\frac{a_{4}}{2a_{0}a_{2}a_{3}},$ (48) $\displaystyle y$ $\displaystyle=$ $\displaystyle\cot\theta_{2}=\left\\{\begin{array}[]{lll}\frac{a_{3}(1-2a_{3}^{2}-2a_{4}^{2})}{2a_{0}a_{2}a_{4}},&\texttt{if}&a_{2}>a_{3},\\\ \frac{a_{2}(1-2a_{2}^{2}-2a_{4}^{2})}{2a_{0}a_{3}a_{4}},&\texttt{if}&a_{2}<a_{3},\\\ \end{array}\right.$ (51) $\displaystyle y$ $\displaystyle=$ $\displaystyle\cot\theta_{3}=\left\\{\begin{array}[]{lll}\frac{a_{2}(1-2a_{2}^{2}-2a_{4}^{2})}{2a_{0}a_{3}a_{4}},&\texttt{if}&a_{2}>a_{3},\\\ \frac{a_{3}(1-2a_{3}^{2}-2a_{4}^{2})}{2a_{0}a_{2}a_{4}},&\texttt{if}&a_{2}<a_{3}.\\\ \end{array}\right.$ (54) Note that when $a_{2}=a_{3}$, then there is $\cot\theta_{2}=\cot\theta_{3}=\cot(\pi-\theta_{0})=\frac{1-2a_{3}^{2}-2a_{4}^{2}}{2a_{0}a_{4}}=\frac{a_{0}^{2}-a_{4}^{2}}{2a_{0}a_{4}}$, which implies that $\theta_{2}=\theta_{3}=\pi-\theta_{0}$ (i.e. $Q(\theta_{2},0)=0$) since $\theta_{2},\theta_{3},\pi-\theta_{0}\in(0,\pi)$. Hence $a_{2}\neq a_{3}$ in the expression $y=\cot\theta_{2}$ and $y=\cot\theta_{3}$ because of the hypothesis $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$. By checking, it can be proved that $\theta_{1},\theta_{2}$ are the roots of $\frac{\partial\left(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\right)}{\partial\theta}=0$. Thus $\begin{array}[]{ll}&\min\left\\{\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\right\\}\\\ =&\left\\{\begin{array}[]{lll}\min\left\\{\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{P(\theta_{2},0)}+\sqrt{Q(\theta_{2},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\right\\},&\texttt{if}&a_{2}\neq a_{3},\\\ \min\left\\{\sqrt{P(\pi-\theta_{0},0)}\big{|}_{a_{2}=a_{3}},\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\right\\},&\texttt{if}&a_{2}=a_{3}.\\\ \end{array}\right.\end{array}$ (55) For the case $\varphi=\pi$ and the case $\varphi=2\pi$, from (19), (20), and (21), we have $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\pi)}+\sqrt{Q(\theta,\pi)}\\}\\\ =&\min\\{\sqrt{P(\pi-\theta,0)}+\sqrt{Q(\pi-\theta,0)}\\}\\\ =&\min\\{\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\\}\\\ =&\min\\{\sqrt{P(\theta,2\pi)}+\sqrt{Q(\theta,2\pi)}\\}.\\\ \end{array}$ (56) Therefore, if $a_{2}\neq a_{3}$, then $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}\\\ =&\min\big{\\{}\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)},\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\\\ &~{}~{}~{}~{}~{}~{}~{}\sqrt{P(\theta_{2},0)}+\sqrt{Q(\theta_{2},0)},\sqrt{P(\frac{\pi}{2},\frac{\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{\pi}{2})}\big{\\}};\\\ \end{array}$ (57) if $a_{2}=a_{3}$, then $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}\\\ =&\min\big{\\{}\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)},\sqrt{P(\pi-\theta_{0},0)}\big{|}_{a_{2}=a_{3}},\\\ &~{}~{}~{}~{}~{}~{}~{}\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{P(\frac{\pi}{2},\frac{\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{\pi}{2})}\big{\\}}.\\\ \end{array}$ (58) ### III.7 $a_{4}=0$ and $a_{0}a_{1}a_{2}a_{3}\neq 0$ In this section, we consider the quantum channel (2) with coefficients satisfying $a_{4}=0$ and $a_{0}a_{1}a_{2}a_{3}\neq 0$. Note that $\begin{array}[]{ll}&\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)}\\\ =&\sqrt{P(\pi,\varphi)}+\sqrt{Q(\pi,\varphi)}\\\ =&a_{0}^{2}+\sqrt{(1-a_{0}^{2})^{2}-4a_{2}^{2}a_{3}^{2}}\\\ =&a_{0}^{2}+\sqrt{[a_{1}^{2}+(a_{2}-a_{3})^{2}][a_{1}^{2}+(a_{2}+a_{3})^{2}]},\end{array}$ and $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$ in case of $\theta=0,\pi$. Next we suppose that $\sin\theta\neq 0$, that is $\theta\in(0,\pi)$. We begin with the discussion of the condition such that $P(\theta,\varphi)=0$. We can see that $\begin{array}[]{ll}&P(\theta,\varphi)\\\ =&\frac{1}{4}[1+2{{{a}}_{0}}{a_{1}}\cos(\varphi-\mu)\sin\theta-2{a_{2}}{a_{3}}\\\ &-\cos\theta(1-2a_{0}^{2}-2{a_{2}}{a_{3}})][1+2{{{a}}_{0}}{a_{1}}\cos(\varphi-\mu)\sin\theta\\\ &+2{a_{2}}{a_{3}}-\cos\theta(1-2a_{0}^{2}+2{a_{2}}{a_{3}})]\\\ =&\frac{1}{(1+t^{2})^{2}}[1+(t^{2}-1)a_{0}^{2}+2t{a_{0}}{a_{1}}\cos(\varphi-\mu)-2{a_{2}}{a_{3}}]\\\ &\times[1+(t^{2}-1)a_{0}^{2}+2t{a_{0}}{a_{1}}\cos(\varphi-\mu)+2{a_{2}}{a_{3}}]\\\ =&0\end{array}$ (59) implies that $\begin{array}[]{c}\cos(\varphi-\mu)=z_{1}=\frac{-1+a_{0}^{2}-t^{2}a_{0}^{2}-2{a_{2}}{a_{3}}}{2t{a_{0}}{a_{1}}},\\\ \cos(\varphi-\mu)=z_{2}=\frac{-1+a_{0}^{2}-t^{2}a_{0}^{2}+2{a_{2}}{a_{3}}}{2t{a_{0}}{a_{1}}}.\\\ \end{array}$ (60) Note that both $z_{1}=\frac{-1+a_{0}^{2}-t^{2}a_{0}^{2}-2{a_{2}}\,{a_{3}}}{2\,t\,{a_{0}}\,{a_{1}}}$ and $z_{2}=\frac{-1+a_{0}^{2}-t^{2}a_{0}^{2}+2\,{a_{2}}\,{a_{3}}}{2\,t\,{a_{0}}\,{a_{1}}}$ tend to $-\infty$ when $t\rightarrow 0$ and $t\rightarrow+\infty$. Since $z_{1}\leq-2\sqrt{\frac{1-a_{0}^{2}+2\,{a_{2}}\,{a_{3}}}{2\,t\,{a_{0}}\,{a_{1}}}}\sqrt{\frac{t^{2}a_{0}^{2}}{2\,t\,{a_{0}}\,{a_{1}}}}=-\frac{\sqrt{a_{1}^{2}+(a_{2}+a_{3})^{2}}}{a_{1}}<-1$, there is no $\varphi$ such that $\cos(\varphi-\mu)=z_{1}=\frac{-1+a_{0}^{2}-t^{2}a_{0}^{2}-2\,{a_{2}}\,{a_{3}}}{2\,t\,{a_{0}}\,{a_{1}}}$. Note that $z_{2}\leq-2\sqrt{\frac{1-a_{0}^{2}-2\,{a_{2}}\,{a_{3}}}{2\,t\,{a_{0}}\,{a_{1}}}}\sqrt{\frac{t^{2}a_{0}^{2}}{2\,t\,{a_{0}}\,{a_{1}}}}=-\frac{\sqrt{a_{1}^{2}+(a_{2}-a_{3})^{2}}}{a_{1}}\leq-1$, where the equality $z_{2}=-1$ holds iff $t=\frac{a_{1}}{a_{0}}$ and $a_{2}=a_{3}$. It follows that $\cos(\varphi-\mu)=z_{2}=\frac{-1+a_{0}^{2}-t^{2}a_{0}^{2}+2\,{a_{2}}\,{a_{3}}}{2\,t\,{a_{0}}\,{a_{1}}}$ iff $a_{2}=a_{3}$, $\varphi=\pi+\mu$, and $\theta=\theta_{0}$, where $\cot\frac{\theta_{0}}{2}=\frac{a_{1}}{a_{0}}$. Thus, $P(\theta,\varphi)=0$ if and only if $a_{2}=a_{3}$, $\varphi=\pi+\mu$, and $\theta=\theta_{0}$. Therefore, $\begin{array}[]{ll}&\big{(}\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\big{)}\big{|}_{P(\theta,\varphi)=0}\\\ =&\sqrt{Q(\theta_{0},\mu+\pi)}\big{|}_{a_{2}=a_{3}}\\\ =&\frac{\sqrt{\left(a_{0}^{4}+a_{1}^{2}+a_{0}^{2}\,a_{1}^{2}-2\,a_{1}^{2}\,a_{3}^{2}\right)\,\left(a_{0}^{4}+a_{1}^{2}+a_{0}^{2}\,a_{1}^{2}+2\,a_{1}^{2}\,a_{3}^{2}\right)}}{a_{0}^{2}+a_{1}^{2}}.\\\ \end{array}$ (61) By (19), we know that $Q(\theta,\varphi)=0$ if and only if $a_{2}=a_{3}$, $\varphi=\mu$, and $\theta=\pi-\theta_{0}$ and $\begin{array}[]{rl}&\big{(}\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\big{)}\big{|}_{Q(\theta,\varphi)=0}\\\ =&\sqrt{P(\pi-\theta_{0},\mu)}\big{|}_{a_{2}=a_{3}}\\\ =&\big{(}\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\big{)}\big{|}_{P(\theta,\varphi)=0}.\\\ \end{array}$ (62) From above, we can see that the quantum channel with $a_{4}=0$, $a_{2}=a_{3}$, and $a_{0}a_{1}a_{2}a_{3}\neq 0$ can be collapsed to an EPR pair with probability $p_{1}=p_{2}=\frac{a_{0}^{2}(1-a_{0}^{2}-a_{1}^{2})}{a_{0}^{2}+a_{1}^{2}}$ via Charlie’s appropriate measurement (Charlie measures his particle in measurement basis (3) with $(\theta,\varphi)=(\theta_{0},\mu+\pi)$ or $(\theta,\varphi)=(\pi-\theta_{0},\mu)$). In the following we suppose that $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$. From (26) and (27), we get $\begin{array}[]{rl}\sin(\varphi-\mu)[{a_{0}}\cos\theta+{{{a}}_{1}}\cos(\varphi-\mu)\sin\theta]&=0,\\\ \sin(\varphi-\mu)[{a_{0}}\cos\theta+{{{a}}_{1}}\cos(\varphi-\mu)\sin\theta]&\\\ \times[{\sin^{2}\theta}+2a_{0}^{2}{\cos^{2}\theta}+2{a_{0}}{a_{1}}\cos\theta\cos(\varphi-\mu)\sin\theta]&=0.\end{array}$ (63) It follows that the minimum of $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}$ must occur at either the hyperplane $\sin(\varphi-\mu)=0$ or the hyperplane ${{{a}}_{0}}\cos\theta+{{{a}}_{1}}\cos(\varphi-\mu)\sin\theta=0$ when $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$ and $\theta\in(0,\pi)$. Now we consider the two hyperplanes. If ${{{a}}_{0}}\cos\theta+{{{a}}_{1}}\cos(\varphi-\mu)\sin\theta=0$, then $\cos(\varphi-\mu)=-\frac{a_{0}\cot\theta}{a_{1}}$, $\begin{array}[]{ll}&\big{(}\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\big{)}\big{|}_{\cos(\varphi-\mu)=-\frac{a_{0}\cot\theta}{a_{1}}}\\\ =&\sqrt{1-4a_{2}^{2}a_{3}^{2}}.\\\ \end{array}$ (64) For $\sin(\varphi-\mu)=0$, i.e. $\varphi=\mu,\mu+\pi$ in case of $\mu\in(0,\pi]$, or $\varphi=0,\pi,2\pi$ in case of $\mu=0$, we first investigate the case $\varphi=\mu$. From (28), we have $\begin{array}[]{ll}({a_{0}}\cos\theta+{a_{1}}\sin\theta)[2a_{0}{a_{1}}({a}_{0}^{2}-a_{1}^{2})\sin 2\theta-2a_{0}^{2}({\sin^{2}\theta}&\\\ +2a_{1}^{2}\cos 2\theta)+{\sin^{2}\theta}(1-2a_{1}^{2}-4a_{2}^{2}a_{3}^{2})]=0,&\end{array}$ (65) that is, ${a_{0}}\cos\theta+{a_{1}}\sin\theta=0$, or $2a_{0}{a_{1}}({a}_{0}^{2}-a_{1}^{2})\sin 2\theta-2a_{0}^{2}({\sin^{2}\theta}+2{{a}}_{1}^{2}\cos 2\theta)+{\sin^{2}\theta}(1-2a_{1}^{2}-4a_{2}^{2}a_{3}^{2})=0$. From ${a_{0}}\cos\theta+{a_{1}}\sin\theta=0$, there is $\cot\theta_{1}=-\frac{a_{1}}{a_{0}}.$ Clearly, $\theta_{1}$ is a root of (24). From $\begin{array}[]{rl}2a_{0}{a_{1}}({a}_{0}^{2}-a_{1}^{2})\sin 2\theta-2a_{0}^{2}({\sin^{2}\theta}+2{{a}}_{1}^{2}\cos 2\theta)&\\\ +{\sin^{2}\theta}(1-2a_{1}^{2}-4a_{2}^{2}a_{3}^{2})&=0,\\\ \end{array}$ (66) there is $\displaystyle\cot\theta_{2}$ $\displaystyle=$ $\displaystyle\frac{a_{0}^{2}-a_{1}^{2}-|a_{2}^{2}-a_{3}^{2}|}{2a_{0}a_{1}},$ $\displaystyle\cot\theta_{3}$ $\displaystyle=$ $\displaystyle\frac{a_{0}^{2}-a_{1}^{2}+|a_{2}^{2}-a_{3}^{2}|}{2a_{0}a_{1}},$ where $a_{2}\neq a_{3}$. (Since when $a_{2}=a_{3}$ there is $\theta_{2}=\theta_{3}=\pi-\theta_{0}$, i.e. $Q(\pi-\theta_{0},\mu)=0$, which contradict with the hypothesis $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$). By checking, we see $\theta_{3}$ is a root of $\frac{\partial\big{(}\sqrt{P(\theta,\mu)}+\sqrt{Q(\theta,\mu)}\big{)}}{\partial\theta}=0$, while $\theta_{2}$ is not. Hence, $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\mu)}+\sqrt{Q(\theta,\mu)}\\}\\\ =&\left\\{\begin{array}[]{lll}\min\left\\{\sqrt{P(\theta_{1},\mu)}+\sqrt{Q(\theta_{1},\mu)},\sqrt{P(\theta_{3},\mu)}+\sqrt{Q(\theta_{3},\mu)},\sqrt{P(0,\mu)}+\sqrt{Q(0,\mu)}\right\\},&\texttt{if}&a_{2}\neq a_{3},\\\ \min\left\\{\sqrt{P(\theta_{1},\mu)}+\sqrt{Q(\theta_{1},\mu)},\sqrt{P(\pi-\theta_{0},\mu)}|_{a_{2}=a_{3}},\sqrt{P(0,\mu)}+\sqrt{Q(0,\mu)}\right\\},&\texttt{if}&a_{2}=a_{3}.\\\ \end{array}\right.\end{array}$ (67) For $\varphi=\mu+\pi,\mu+2\pi$, there is $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\mu+\pi)}+\sqrt{Q(\theta,\mu+\pi)}\\}\\\ =&\min\\{\sqrt{P(\pi-\theta,\mu)}+\sqrt{Q(\pi-\theta,\mu)}\\}\\\ =&\min\\{\sqrt{P(\theta,\mu)}+\sqrt{Q(\theta,\mu)}\\}\\\ =&\min\\{\sqrt{P(\theta,\mu+2\pi)}+\sqrt{Q(\theta,\mu+2\pi)}\\}.\end{array}$ (68) from (20) and (19). Thus, if $a_{2}=a_{3}$, then $\begin{array}[]{ll}&\min\big{\\{}\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\big{\\}}\\\ =&\min\big{\\{}a_{0}^{2}+\sqrt{(1-a_{0}^{2})^{2}-4a_{2}^{2}a_{3}^{2}},\sqrt{1-4a_{2}^{2}a_{3}^{2}},\\\ &~{}~{}~{}~{}~{}~{}~{}\sqrt{P(\theta_{1},\mu)}+\sqrt{Q(\theta_{1},\mu)},\sqrt{Q(\theta_{0},\pi+\mu)}\big{|}_{a_{2}=a_{3}}\big{\\}};\\\ \end{array}$ (69) if $a_{2}\neq a_{3}$, then $\begin{array}[]{ll}&\min\big{\\{}\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\big{\\}}\\\ =&\min\big{\\{}a_{0}^{2}+\sqrt{(1-a_{0}^{2})^{2}-4a_{2}^{2}a_{3}^{2}},\sqrt{1-4a_{2}^{2}a_{3}^{2}},\\\ &~{}~{}~{}~{}~{}~{}~{}\sqrt{P(\theta_{1},\mu)}+\sqrt{Q(\theta_{1},\mu)},\sqrt{P(\theta_{3},\mu)}+\sqrt{Q(\theta_{3},\mu)}\big{\\}}.\\\ \end{array}$ (70) Here $\cot\frac{\theta_{0}}{2}=\frac{a_{1}}{a_{0}}$, $\cot\theta_{1}=-\frac{a_{1}}{a_{0}}$, $\cot\theta_{3}=\frac{a_{0}^{2}-a_{1}^{2}+|a_{2}^{2}-a_{3}^{2}|}{2a_{0}a_{1}}$. ### III.8 One is $a_{2}=0$ and $a_{0}a_{1}a_{3}a_{4}\neq 0$, the other is $a_{3}=0$ and $a_{0}a_{1}a_{2}a_{4}\neq 0$ We now discuss the quantum channel (2), the coefficients of which satisfying $a_{2}=0$, but $a_{0}a_{1}a_{3}a_{4}\neq 0$. When $\sin\theta=0$, i.e. $\theta=0,\pi$, then $\begin{array}[]{ll}&\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)}\\\ =&\sqrt{P(\pi,\varphi)}+\sqrt{Q(\pi,\varphi)}\\\ =&a_{0}^{2}+\sqrt{(1-a_{0}^{2})^{2}-4a_{1}^{2}a_{4}^{2}}\\\ =&a_{0}^{2}+\sqrt{[a_{3}^{2}+(a_{1}-a_{4})^{2}][a_{3}^{2}+(a_{1}+a_{4})^{2}]},\\\ \end{array}$ (71) and $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$ in case of $\theta=0,\pi$. Next we suppose that $\sin\theta\neq 0$. First, we prove that $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$. If $P(\theta,\varphi)=0$, then $\displaystyle\cos(\varphi-\mu)$ $\displaystyle=$ $\displaystyle\frac{1}{2a_{0}^{2}a_{1}^{2}\sin^{2}\theta}\big{[}-a_{0}^{3}a_{1}\sin 2\theta+8\sqrt{-a_{0}^{2}a_{1}^{2}a_{3}^{2}a_{4}^{2}\cos^{2}\frac{\theta}{2}\sin^{6}\frac{\theta}{2}}$ $\displaystyle-4a_{0}a_{1}(1-2a_{4}^{2})\cos\frac{\theta}{2}\sin^{3}\frac{\theta}{2}\big{]},$ $\displaystyle\cos(\varphi-\mu)$ $\displaystyle=$ $\displaystyle-\frac{1}{2a_{0}^{2}a_{1}^{2}\sin^{2}\theta}\big{[}a_{0}^{3}a_{1}\sin 2\theta+8\sqrt{-a_{0}^{2}a_{1}^{2}a_{3}^{2}a_{4}^{2}\cos^{2}\frac{\theta}{2}\sin^{6}\frac{\theta}{2}}$ $\displaystyle+4a_{0}a_{1}(1-2a_{4}^{2})\cos\frac{\theta}{2}\sin^{3}\frac{\theta}{2}\big{]}.$ But $8\sqrt{-a_{0}^{2}a_{1}^{2}a_{3}^{2}a_{4}^{2}\cos^{2}(\frac{\theta}{2})\sin^{6}(\frac{\theta}{2})}$ is an imaginary number, so $P(\theta,\varphi)\neq 0$. From (19), there is also $Q(\theta,\varphi)\neq 0$. That is, no matter what kind of measurement basis Charlie choose, Alice and Bob can never share an EPR pair after his measurement. Second, the minimum of $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}$ should occur at the point satisfying (24) and (25), or $\theta=0$. From (26) and (27), there are $\sin(\varphi-\mu)[a_{0}\cos\theta+a_{1}\cos(\varphi-\mu)\sin\theta]=0,$ (72) and $\begin{array}[]{ll}&\sin(\varphi-\mu)[a_{0}\cos\theta+a_{1}\cos(\varphi-\mu)\sin\theta][2a_{0}^{2}\cos^{2}\theta\\\ &+a_{0}a_{1}\cos(\varphi-\mu)\sin 2\theta+(1-2a_{4}^{2})\sin^{2}\theta]=0.\end{array}$ (73) It follows that the minimum of $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}$ should occur at the point such that $\sin(\varphi-\mu)=0$, or $a_{0}\cos\theta+a_{1}\cos(\varphi-\mu)\sin\theta=0$, or $\theta=0$. From $a_{0}\cos\theta+a_{1}\cos(\varphi-\mu)\sin\theta=0$, we know that $\cos(\varphi-\mu)=-\frac{a_{0}\cot\theta}{a_{1}}$, and $\begin{array}[]{cl}&\left(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right)\big{|}_{\\{\theta\in(0,\pi),\cos(\varphi-\mu)=-\frac{a_{0}\cot\theta}{a_{1}}\\}}\\\ =&\sqrt{1-4a_{0}^{2}a_{4}^{2}-4a_{1}^{2}a_{4}^{2}}.\end{array}$ (74) For the case $\sin(\varphi-\mu)=0$, since equality (68) holds, we need only to consider the case $\varphi=\mu$. From (28),we derive $\begin{array}[]{cl}&\sin 3\theta\left(2\,a_{0}^{2}\,{a_{1}}-a_{1}+4\,a_{0}^{4}\,{a_{1}}+2\,a_{1}^{3}-12\,a_{0}^{2}\,a_{1}^{3}\right)\\\ &+\sin\theta\left(3\,{a_{1}}-6\,a_{0}^{2}\,{a_{1}}+4\,a_{0}^{4}\,{a_{1}}-6\,a_{1}^{3}+4\,a_{0}^{2}\,a_{1}^{3}\right)\\\ &+\cos\theta\left({a_{0}}-2\,a_{0}^{3}-2\,{a_{0}}\,a_{1}^{2}-4\,a_{0}^{3}\,a_{1}^{2}-4\,{a_{0}}\,a_{1}^{4}\right)\\\ &+\cos 3\theta\left(2\,a_{0}^{3}-{a_{0}}+2\,{a_{0}}\,a_{1}^{2}-12\,a_{0}^{3}\,a_{1}^{2}+4\,{a_{0}}\,a_{1}^{4}\right)\\\ =&\\{-16\cot^{3}\theta\,a_{0}^{3}\,a_{1}^{2}+16\cot^{2}\theta\,a_{0}^{2}\,{a_{1}}\,\left(a_{0}^{2}-2\,a_{1}^{2}\right)\\\ &+4\cot\theta{a_{0}}\,\left[1-2\,a_{1}^{2}-4\,a_{1}^{4}+2a_{0}^{2}(4\,a_{1}^{2}-1)\right]\\\ &+4\,\left(1-2\,a_{0}^{2}\right){a_{1}}\left(1-2\,a_{1}^{2}\right)\\}\sin^{3}\theta\\\ =&0,\end{array}$ (75) which follows that $\displaystyle\cot\theta=\cot\theta_{1}$ $\displaystyle=$ $\displaystyle-\frac{a_{1}}{a_{0}},$ $\displaystyle\cot\theta=\cot\theta_{2}$ $\displaystyle=$ $\displaystyle\frac{1-2a_{1}^{2}}{2a_{0}a_{1}},$ $\displaystyle\cot\theta=\cot\theta_{3}$ $\displaystyle=$ $\displaystyle\frac{-1+2a_{0}^{2}}{2a_{0}a_{1}},$ where $\theta_{1}$ and $\theta_{2}$ are the roots of equation $\frac{\partial(\sqrt{P(\theta,\mu)}+\sqrt{Q(\theta,\mu)})}{\partial\theta}=0$, while $\theta_{3}$ is not. It is not difficult to obtain $\begin{array}[]{cl}&\min\big{\\{}\sqrt{P(\theta,\mu)}+\sqrt{Q(\theta,\mu)}\big{\\}}\\\ =&\min\big{\\{}\sqrt{P(\theta_{1},\mu)}+\sqrt{Q(\theta_{1},\mu)},\sqrt{P(\theta_{2},\mu)}+\sqrt{Q(\theta_{2},\mu)},\\\ &~{}~{}~{}~{}~{}~{}~{}~{}\sqrt{P(0,\mu)}+\sqrt{Q(0,\mu)}\big{\\}}.\end{array}$ (76) Therefore, for quantum channel (2) with $a_{2}=0$ and $a_{0}a_{1}a_{3}a_{4}\neq 0$, there is $\begin{array}[]{cl}&\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}\\\ =&\min\Big{\\{}\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)},\sqrt{1-4a_{0}^{2}a_{4}^{2}-4a_{1}^{2}a_{4}^{2}},\\\ &~{}~{}~{}~{}~{}~{}~{}\sqrt{P(\theta_{1},\mu)}+\sqrt{Q(\theta_{1},\mu)},\sqrt{P(\theta_{2},\mu)}+\sqrt{Q(\theta_{2},\mu)}\Big{\\}}.\end{array}$ (77) Similarly, for quantum channel (2) with $a_{3}=0$ and $a_{0}a_{1}a_{2}a_{4}\neq 0$ we also obtain $\begin{array}[]{cl}&\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}\\\ =&\min\Big{\\{}a_{0}^{2}+\sqrt{(1-a_{0}^{2})^{2}-4a_{1}^{2}a_{4}^{2}},\sqrt{1-4a_{0}^{2}a_{4}^{2}-4a_{1}^{2}a_{4}^{2}},\\\ &~{}~{}~{}~{}~{}~{}\sqrt{P(\theta_{1},\mu)}+\sqrt{Q(\theta_{1},\mu)},\sqrt{P(\theta_{2},\mu)}+\sqrt{Q(\theta_{2},\mu)}\Big{\\}}.\end{array}$ (78) Here $\cot\theta_{1}=-\frac{a_{1}}{a_{0}}$. ### III.9 $\mu=0$ and $a_{0}a_{1}a_{2}a_{3}a_{4}\neq 0$ In this section, we investigate the quantum channel (2) with $\mu=0$ and $a_{0}a_{1}a_{2}a_{3}a_{4}\neq 0$. Evidently, $\begin{array}[]{ll}&\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)}\\\ &=\sqrt{P(\pi,\varphi)}+\sqrt{Q(\pi,\varphi)}\\\ &=a_{0}^{2}+\sqrt{(1-a_{0}^{2})^{2}-4(a_{2}a_{3}-a_{1}a_{4})^{2}}\\\ &=a_{0}^{2}+\sqrt{[(a_{2}-a_{3})^{2}+(a_{1}+a_{4})^{2}][(a_{2}+a_{3})^{2}+(a_{1}-a_{4})^{2}]}\\\ \end{array}$ (79) and $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$ in case of $\theta=0,\pi$. In the following, we suppose that $\theta\neq 0,\pi$, i.e. $\theta\in(0,\pi)$. We first consider the case $P(\theta,\varphi)=0$. If $P(\theta,\varphi)=0$, then $\displaystyle\cos\varphi$ $\displaystyle=$ $\displaystyle z_{1}$ $\displaystyle=$ $\displaystyle\frac{1}{2\,t\,{a_{0}}\,a_{1}^{2}}\\{-t^{2}\,a_{0}^{2}\,{a_{1}}-(1-a_{0}^{2}){a_{1}}$ $\displaystyle-2(a_{2}a_{3}-a_{1}a_{4})a_{4}-2[t^{2}\,a_{0}^{2}\,{a_{1}}\,{a_{2}}\,{a_{3}}\,{a_{4}}+$ $\displaystyle({a_{2}}\,{a_{3}}-{a_{1}}\,{a_{4}})(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})]^{\frac{1}{2}}\\},$ $\displaystyle\cos\varphi$ $\displaystyle=$ $\displaystyle z_{2}$ $\displaystyle=$ $\displaystyle\frac{1}{2\,t\,{a_{0}}\,a_{1}^{2}}\\{-t^{2}\,a_{0}^{2}\,{a_{1}}-(1-a_{0}^{2}){a_{1}}$ $\displaystyle-2(a_{2}a_{3}-a_{1}a_{4})a_{4}+2[t^{2}\,a_{0}^{2}\,{a_{1}}\,{a_{2}}\,{a_{3}}\,{a_{4}}+$ $\displaystyle({a_{2}}\,{a_{3}}-{a_{1}}\,{a_{4}})(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})]^{\frac{1}{2}}\\}.$ Note that if ${a_{1}}\,{a_{4}}-{a_{2}}\,{a_{3}}>0$, then $t\in[t_{0},+\infty)$; if ${a_{1}}\,{a_{4}}-{a_{2}}\,{a_{3}}\leq 0$, then $t\in(0,+\infty)$. Here, $t_{0}=\sqrt{\frac{\left({a_{1}}\,{a_{4}}-{a_{2}}\,{a_{3}}\right)(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})}{a_{0}^{2}\,{a_{1}}\,{a_{2}}\,{a_{3}}\,{a_{4}}}}$. Obviously, both $z_{1}$ and $z_{2}$ go to $-\infty$ when $t\rightarrow+\infty$. We can prove that $z=z_{1}=\frac{1}{2\,t\,{a_{0}}\,a_{1}^{2}}\\{-t^{2}\,a_{0}^{2}\,{a_{1}}-(1-a_{0}^{2}){a_{1}}-2(a_{2}a_{3}-a_{1}a_{4})a_{4}-2[t^{2}\,a_{0}^{2}\,{a_{1}}\,{a_{2}}\,{a_{3}}\,{a_{4}}+\left({a_{2}}\,{a_{3}}-{a_{1}}\,{a_{4}}\right)(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})]^{\frac{1}{2}}\\}$ has no intersection point with the straight line $z=-1$. Since $z_{1}$ is a continuous function of $t$ and tends to $-\infty$ when $t$ tends to $+\infty$, so $z_{1}<-1$. Thus, there is no $\varphi$ satisfying $\cos\varphi=z_{1}$. Let us look at $z=z_{2}$. $z_{2}=-1$ implies that $\displaystyle[-t^{2}\,a_{0}^{2}\,{a_{1}}-(1-a_{0}^{2}){a_{1}}-2(a_{2}a_{3}-a_{1}a_{4})a_{4}+2ta_{0}a_{1}^{2}]^{2}$ (80) $\displaystyle-4\\{t^{2}\,a_{0}^{2}\,{a_{1}}\,{a_{2}}\,{a_{3}}\,{a_{4}}+\left({a_{2}}\,{a_{3}}-{a_{1}}\,{a_{4}}\right)$ $\displaystyle\times(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})\\}$ $\displaystyle=$ $\displaystyle a_{1}^{2}\,\left[-t^{2}\,a_{0}^{2}+2\,t\,{a_{0}}({a_{1}}+{a_{4}})-1+a_{0}^{2}+2\,{a_{2}}\,{a_{3}}-2\,{a_{1}}\,{a_{4}}\right]$ $\displaystyle\times\left[-t^{2}\,a_{0}^{2}+2\,t\,{a_{0}}({a_{1}}-{a_{4}})-1+a_{0}^{2}-2\,{a_{2}}\,{a_{3}}+2\,{a_{1}}\,{a_{4}}\right]$ $\displaystyle=$ $\displaystyle 0,$ which follows that $\displaystyle t=t_{1}=\frac{a_{1}-a_{4}-\sqrt{-(a_{2}+a_{3})^{2}}}{a_{0}},$ $\displaystyle t=t_{2}=\frac{a_{1}-a_{4}+\sqrt{-(a_{2}+a_{3})^{2}}}{a_{0}},$ $\displaystyle t=t_{3}=\frac{a_{1}+a_{4}-\sqrt{-(a_{2}-a_{3})^{2}}}{a_{0}},$ $\displaystyle t=t_{4}=\frac{a_{1}+a_{4}+\sqrt{-(a_{2}-a_{3})^{2}}}{a_{0}}.$ Here $t_{1}$ and $t_{2}$ are imaginary numbers, while $t_{3}$ and $t_{4}$ are real numbers only if $a_{2}=a_{3}$. By checking, when $a_{2}=a_{3}$, $t_{3}=t_{4}=\frac{a_{1}+a_{4}}{a_{0}}$ is a root of equation $z_{2}=-1$, and the maximum point of function $z=z_{2}$. Since $z=z_{2}$ is a continuous function of $t$, tends to $-\infty$ when $t$ tends to $+\infty$, and has only one intersection point with straight line $z=-1$, there must be $z_{2}\leq-1$, where the equality holds iff $a_{2}=a_{3}$ and $t=t_{3}=t_{4}=\frac{a_{1}+a_{4}}{a_{0}}$. It means that $P(\theta,\varphi)=0$ iff $a_{2}=a_{3}$, $\varphi=\pi$, and $\theta=\theta_{0}$, where $\cot\frac{\theta_{0}}{2}=\frac{a_{1}+a_{4}}{a_{0}}$. From (19), we can show that $Q(\theta,\varphi)=0$ iff $a_{2}=a_{3}$, $\varphi=0$, and $\theta=\pi-\theta_{0}$. Therefore $\displaystyle(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)})\big{|}_{P(\theta,\varphi)=0}$ $\displaystyle=$ $\displaystyle\sqrt{Q(\theta_{0},\pi)}\big{|}_{a_{2}=a_{3}}$ $\displaystyle=$ $\displaystyle\sqrt{P(\pi-\theta_{0},0)}\big{|}_{a_{2}=a_{3}}$ $\displaystyle=$ $\displaystyle(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)})\big{|}_{Q(\theta,\varphi)=0}$ $\displaystyle=$ $\displaystyle\sqrt{(1-2a_{4}^{2})^{2}+4a_{3}^{2}(-1+a_{1}^{2}+a_{3}^{2}+2a_{1}a_{4}+3a_{4}^{2})}.$ It is shown that the quantum channel (2) with $\mu=0$, $a_{0}a_{1}a_{2}a_{3}a_{4}\neq 0$, and $a_{2}=a_{3}$, collapses to an EPR pair with probability $p_{1}=p_{2}=\frac{a_{0}^{2}(1-a_{0}^{2}+3a_{1}^{2}+4a_{1}a_{4}+a_{4}^{2})}{a_{0}^{2}+(a_{1}+a_{4})^{2}}$ after Charlie’s measurement in the basis (3) with either $\varphi=\pi$ and $\theta=\theta_{0}$, or $\varphi=0$ and $\theta=\pi-\theta_{0}$ in case of $a_{2}=a_{3}$. In the following we suppose that $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$. By (26), there is $\begin{array}[]{cl}&\sin\varphi\\{2\cos\varphi(a_{2}a_{3}-a_{1}a_{4})(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})\\\ &-ya_{0}[a_{2}a_{3}a_{4}(1+2a_{1}^{2}-2a_{2}^{2}-2a_{3}^{2})\\\ &-2a_{1}(a_{2}^{2}a_{3}^{2}-a_{2}^{2}a_{4}^{2}-a_{3}^{2}a_{4}^{2})]\\}\\\ =&0.\end{array}$ (81) Note that $a_{2}a_{3}a_{4}(1+2a_{1}^{2}-2a_{2}^{2}-2a_{3}^{2})-2a_{1}(a_{2}^{2}a_{3}^{2}-a_{2}^{2}a_{4}^{2}-a_{3}^{2}a_{4}^{2})=a_{1}a_{4}^{2}\neq 0$ if $a_{2}a_{3}=a_{1}a_{4}$. Therefore, $a_{2}a_{3}a_{4}(1+2a_{1}^{2}-2a_{2}^{2}-2a_{3}^{2})-2a_{1}(a_{2}^{2}a_{3}^{2}-a_{2}^{2}a_{4}^{2}-a_{3}^{2}a_{4}^{2})=0$ implies that $\cos\varphi=0$. Thus, (81) implies that $\sin\varphi=0$, $\cos\varphi=0$, or $y=\frac{2\,\cos\varphi({a_{2}}{a_{3}}-{a_{1}}{a_{4}})[(a_{1}^{2}+a_{4}^{2}){a_{2}}{a_{3}}+a_{1}(a_{2}^{2}+a_{3}^{2})a_{4}]}{a_{0}[a_{2}a_{3}a_{4}(1+2a_{1}^{2}-2a_{2}^{2}-2a_{3}^{2})-2a_{1}(a_{2}^{2}a_{3}^{2}-a_{2}^{2}a_{4}^{2}-a_{3}^{2}a_{4}^{2})]},$ (82) in case of $\sin 2\varphi\neq 0$. From (27), there is $\begin{array}[]{l}4y^{3}a_{0}^{3}(a_{2}a_{3}-a_{1}a_{4})(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})\\\ +8y^{2}a_{0}^{2}a_{1}\cos\varphi(a_{2}a_{3}-a_{1}a_{4})(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})\\\ +y{a_{0}}\\{4a_{1}^{2}{\cos^{2}\varphi}(a_{2}a_{3}-a_{1}a_{4})(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})\\\ +({a_{1}}+2{a_{2}}{a_{3}}{a_{4}}-2{a_{1}}a_{4}^{2})[a_{2}a_{3}(-2a_{4}^{3}-4a_{1}^{2}a_{4}\\\ +a_{4}+2a_{1}a_{2}a_{3})-2a_{1}a_{4}^{2}(a_{2}^{2}+a_{3}^{2})]\\}\\\ +2\cos\varphi({a_{1}}+2{a_{2}}{a_{3}}{a_{4}}-2{a_{1}}a_{4}^{2})[a_{2}^{2}a_{3}^{2}(a_{1}^{2}+a_{4}^{2})\\\ +a_{1}a_{2}a_{3}a_{4}(1-2a_{1}^{2}-2a_{4}^{2})-a_{1}^{2}a_{4}^{2}(a_{2}^{2}+a_{3}^{2})]\\\ =0.\end{array}$ (83) Combining (82) and (83), we obtain that $\begin{array}[]{ll}&\cos\varphi(a_{1}+2a_{2}a_{3}a_{4}-2a_{1}a_{4}^{2})\\{a_{0}^{2}[2a_{1}(a_{2}^{2}a_{3}^{2}-a_{2}^{2}a_{4}^{2}\\\ &-a_{3}^{2}a_{4}^{2})-a_{2}a_{3}a_{4}(1+2a_{1}^{2}-2a_{2}^{2}-2a_{3}^{2})]^{2}\\\ &+4\cos^{2}\varphi({a_{2}}{a_{3}}-{a_{1}}{a_{4}})^{2}(a_{1}a_{3}+a_{2}a_{4})^{2}(a_{1}a_{2}+a_{3}a_{4})^{2}\\}\\\ &=0,\end{array}$ (84) which implies that $\cos\varphi=0$. Next we only need to examine the two cases $\sin\varphi=0$, and $\cos\varphi=0$, that is, $\varphi=0,\frac{\pi}{2},\pi,\frac{3\pi}{2},2\pi$. Now let look at the hyperplane $\varphi=0$. From (28), we have $\begin{array}[]{ll}&[2y{a_{0}}({a_{2}}{a_{3}}-{a_{1}}{a_{4}})+2{a_{1}}{a_{2}}{a_{3}}+{a_{4}}-2a_{1}^{2}{a_{4}}]\\\ &\times\\{{a_{2}}{a_{3}}-2a_{0}^{2}{a_{2}}{a_{3}}-2a_{1}^{2}{a_{2}}{a_{3}}+4a_{0}^{2}a_{1}^{2}{a_{2}}{a_{3}}\\\ &-4a_{2}^{3}a_{3}^{3}-2{a_{1}}{a_{4}}+6a_{0}^{2}{a_{1}}{a_{4}}-4a_{0}^{4}{a_{1}}{a_{4}}+2a_{1}^{3}{a_{4}}\\\ &-4a_{0}^{2}a_{1}^{3}{a_{4}}+8{a_{1}}a_{2}^{2}a_{3}^{2}{a_{4}}-2{a_{2}}{a_{3}}a_{4}^{2}+4a_{0}^{2}{a_{2}}{a_{3}}a_{4}^{2}\\\ &-4a_{1}^{2}{a_{2}}{a_{3}}a_{4}^{2}+2{a_{1}}a_{4}^{3}-4a_{0}^{2}{a_{1}}a_{4}^{3}\\\ &-4y^{2}a_{0}^{2}(a_{1}a_{3}+a_{2}a_{4})(a_{1}a_{2}+a_{3}a_{4})\\\ &-2y{a_{0}}(-2a_{0}^{2}{a_{1}}{a_{2}}{a_{3}}+2a_{1}^{3}{a_{2}}{a_{3}}+{a_{4}}-3a_{0}^{2}{a_{4}}\\\ &+2a_{0}^{4}{a_{4}}+a_{1}^{2}{a_{4}}-2a_{1}^{4}{a_{4}}-4a_{2}^{2}a_{3}^{2}{a_{4}}+6{a_{1}}{a_{2}}{a_{3}}a_{4}^{2}\\\ &-a_{4}^{3}+2a_{0}^{2}a_{4}^{3}-2a_{1}^{2}a_{4}^{3})\\}\\\ =&0.\end{array}$ (85) First, we consider the case $a_{1}a_{4}-a_{2}a_{3}\neq 0$. In this case, (85) has the following solution $\displaystyle y=\cot\theta_{1}$ $\displaystyle=$ $\displaystyle\frac{2a_{1}a_{2}a_{3}+a_{4}-2a_{1}^{2}a_{4}}{2a_{0}(a_{1}a_{4}-a_{2}a_{3})},$ $\displaystyle\texttt{where if}~{}a_{2}=a_{3},\texttt{then}~{}a_{1}a_{4}-a_{2}a_{3}\neq\frac{1}{2};$ $\displaystyle y=\cot\theta_{2}$ $\displaystyle=$ $\displaystyle\frac{-2a_{1}a_{2}a_{4}+a_{3}(1-2a_{1}^{2}-2a_{3}^{2}-2a_{4}^{2})}{2a_{0}(a_{1}a_{3}+a_{2}a_{4})},$ $\displaystyle\texttt{where}~{}a_{2}\neq a_{3};$ $\displaystyle y=\cot\theta_{3}$ $\displaystyle=$ $\displaystyle\frac{-2a_{1}a_{3}a_{4}+a_{2}(1-2a_{1}^{2}-2a_{2}^{2}-2a_{4}^{2})}{2a_{0}(a_{1}a_{2}+a_{3}a_{4})},$ $\displaystyle\texttt{where}~{}a_{2}\neq a_{3}.$ By checking, we show that $\theta_{1}$ is a root of $\frac{\partial(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)})}{\partial\theta}\Big{|}_{a_{1}a_{4}-a_{2}a_{3}\neq 0}=0$, while $\theta_{2}$ is a root only if $a_{2}>a_{3}$, and $\theta_{3}$ is a root only if $a_{2}<a_{3}$. Second, we consider the case $a_{1}a_{4}-a_{2}a_{3}=0$. From (85), we get $\displaystyle y_{4}=\cot\theta_{4}=$ $\displaystyle\frac{(1-2a_{0}^{2})(-1+a_{0}^{2}-a_{1}^{2}+a_{4}^{2}+|a_{2}^{2}-a_{3}^{2}|)}{4a_{0}(1-a_{0}^{2})a_{1}},$ $\displaystyle\texttt{where}~{}a_{2}\neq a_{3};$ $\displaystyle y_{5}=\cot\theta_{5}=$ $\displaystyle\frac{(1-2a_{0}^{2})(-1+a_{0}^{2}-a_{1}^{2}+a_{4}^{2}-|a_{2}^{2}-a_{3}^{2}|)}{4a_{0}(1-a_{0}^{2})a_{1}},$ $\displaystyle\texttt{where}~{}a_{2}\neq a_{3}.$ By checking, we know that only $\theta_{4}$ is a root of $\frac{\partial(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)})}{\partial\theta}\left|{}_{a_{1}a_{4}-a_{2}a_{3}=0}\right.=0$. Therefore, $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\\}\\\ =&\left\\{\begin{array}[]{lllll}\min\left\\{\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{P(\theta_{2},0)}+\sqrt{Q(\theta_{2},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\right\\},&\texttt{if}&a_{2}>a_{3}&\texttt{and}&a_{1}a_{4}-a_{2}a_{3}\neq 0,\\\ \min\left\\{\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{P(\theta_{3},0)}+\sqrt{Q(\theta_{3},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\right\\},&\texttt{if}&a_{2}<a_{3}&\texttt{and}&a_{1}a_{4}-a_{2}a_{3}\neq 0,\\\ \min\left\\{\sqrt{P(\theta_{4},0)}+\sqrt{Q(\theta_{4},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\right\\},&\texttt{if}&a_{2}\neq a_{3}&\texttt{and}&a_{1}a_{4}-a_{2}a_{3}=0,\\\ \min\left\\{\sqrt{P(\pi-\theta_{0},0)}|_{a_{2}=a_{3}},\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\right\\},&\texttt{if}&a_{2}=a_{3}.\end{array}\right.\end{array}$ (86) For $\varphi=\pi,2\pi$, by (56), there is $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\pi)}+\sqrt{Q(\theta,\pi)}\\}\\\ =&\min\\{\sqrt{P(\theta,2\pi)}+\sqrt{Q(\theta,2\pi)}\\}\\\ =&\min\\{\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\\}.\end{array}$ (87) . Now we investigate the hyperplanes $\varphi=\frac{\pi}{2}$ and $\varphi=\frac{3\pi}{2}$. From $\frac{\texttt{d}(\sqrt{P(\theta,\frac{\pi}{2})}+\sqrt{Q(\theta,\frac{\pi}{2})})}{\texttt{d}\theta}=0$, there is $\cos\theta=0$, i.e. $\theta=\frac{\pi}{2}$. Similarly, by $\frac{\texttt{d}(\sqrt{P(\theta,\frac{3\pi}{2})}+\sqrt{Q(\theta,\frac{3\pi}{2})})}{\texttt{d}\theta}=0$, we also get $\theta=\frac{\pi}{2}$. It is direct from (19) and (20) that $\begin{array}[]{ll}&\sqrt{P(\frac{\pi}{2},\frac{\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{\pi}{2})}\\\ =&\sqrt{P(\frac{\pi}{2},\frac{3\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{3\pi}{2})}\\\ =&\sqrt{1-4a_{2}^{2}a_{3}^{2}+8a_{1}a_{2}a_{3}a_{4}-4a_{0}^{2}a_{4}^{2}-4a_{1}^{2}a_{4}^{2}}.\end{array}$ (88) Therefore, $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}\\\ =&\min\\{\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)},\sqrt{P(\frac{\pi}{2},\frac{\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{\pi}{2})},\\\ &~{}~{}~{}~{}~{}~{}\min\\{\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\\}\\}.\\\ \end{array}$ (89) ### III.10 $\mu=\pi$ and $a_{0}a_{1}a_{2}a_{3}a_{4}\neq 0$ In this section, we examine the quantum channel (2) with coefficients satisfying $\mu=\pi$ and $a_{0}a_{1}a_{2}a_{3}a_{4}\neq 0$. Let us start with the special case $\sin\theta=0$. In this case, there is $\begin{array}[]{ll}&\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)}\\\ &=\sqrt{P(\pi,\varphi)}+\sqrt{Q(\pi,\varphi)}\\\ &=a_{0}^{2}+\sqrt{(1-a_{0}^{2})^{2}-4(a_{1}a_{4}+a_{2}a_{3})^{2}}\\\ &=a_{0}^{2}+\sqrt{[(a_{2}-a_{3})^{2}+(a_{1}-a_{4})^{2}][(a_{2}+a_{3})^{2}+(a_{1}+a_{4})^{2}]},\\\ \end{array}$ (90) Evidently, $P(0,\varphi)=Q(\pi,\varphi)=a_{0}^{2}\neq 0$, $P(\pi,\varphi)=0$ if and only if $a_{2}=a_{3}$ and $a_{1}=a_{4}$, and $Q(0,\varphi)=0$ if and only if $a_{2}=a_{3}$ and $a_{1}=a_{4}$. Next we suppose that $\sin\theta\neq 0$, i.e. $\theta\in(0,\pi)$. First, we look for the condition of $P(\theta,\varphi)=0$, $Q(\theta,\varphi)=0$. If $P(\theta,\varphi)=0$, then $\begin{array}[]{ll}&\cos\varphi=z_{1}\\\ =&\frac{1}{2a_{0}a_{1}^{2}t}[t^{2}a_{0}^{2}a_{1}+(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})-2\sqrt{-t^{2}a_{0}^{2}a_{1}a_{2}a_{3}a_{4}+(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}],\\\ &\cos\varphi=z_{2}\\\ =&\frac{1}{2a_{0}a_{1}^{2}t}[t^{2}a_{0}^{2}a_{1}+(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})+2\sqrt{-t^{2}a_{0}^{2}a_{1}a_{2}a_{3}a_{4}+(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}].\\\ \end{array}$ (91) Here $(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})>0$, and $t\in(0,\sqrt{\frac{(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}{a_{0}^{2}a_{1}a_{2}a_{3}a_{4}}}]$, since $(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})\leq 0$ implies that $P(\theta,\varphi)\neq 0$. Directly, $(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})>0$ if and only if $a_{2}^{2}+a_{3}^{2}<\frac{a_{2}a_{3}(a_{1}^{2}+a_{4}^{2})}{a_{1}a_{4}}$. Note that $\begin{array}[]{ll}&(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})\\\ =&a_{1}(a_{1}^{2}-a_{4}^{2})+a_{1}(a_{2}^{2}+a_{3}^{2})-2a_{2}a_{3}a_{4}\\\ >&0\end{array}$ in case of $a_{1}>a_{4}$, $\begin{array}[]{ll}&(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})\\\ =&a_{1}(a_{1}^{2}-a_{4}^{2})+a_{1}(a_{2}^{2}+a_{3}^{2})-2a_{2}a_{3}a_{4}\\\ <&\frac{(a_{1}^{2}-a_{4}^{2})(a_{2}a_{3}+a_{1}a_{4})}{a_{4}}\\\ <&0\end{array}$ in case of $a_{1}<a_{4}$ and $(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})>0$, and $\begin{array}[]{ll}&[(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})]^{2}\\\ &-\\{2\sqrt{(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}\\}^{2}\\\ =&a_{1}^{2}[(a_{1}-a_{4})^{2}+(a_{2}-a_{3})^{2}][(a_{1}+a_{4})^{2}+(a_{2}+a_{3})^{2}]\\\ >&0.\\\ \end{array}$ (92) It follows that $\begin{array}[]{ll}&(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})\\\ &+2\sqrt{(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}\\\ =&(a_{1}-a_{4})(a_{1}^{2}+2a_{3}^{2}+a_{1}a_{4})+2a_{3}\sqrt{(a_{1}-a_{4})^{2}(a_{3}^{2}+a_{1}a_{4})}\\\ >&0\\\ \end{array}$ (93) and $\begin{array}[]{ll}&(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})\\\ &-2\sqrt{(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}\\\ =&(a_{1}-a_{4})(a_{1}^{2}+2a_{3}^{2}+a_{1}a_{4})-2a_{3}\sqrt{(a_{1}-a_{4})^{2}(a_{3}^{2}+a_{1}a_{4})}\\\ >&0\\\ \end{array}$ (94) in case of $a_{1}>a_{4}$; $\begin{array}[]{ll}&(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})\\\ &+2\sqrt{(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}\\\ =&(a_{1}-a_{4})(a_{1}^{2}+2a_{3}^{2}+a_{1}a_{4})+2a_{3}\sqrt{(a_{1}-a_{4})^{2}(a_{3}^{2}+a_{1}a_{4})}\\\ <&0\\\ \end{array}$ (95) and $\begin{array}[]{ll}&(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})\\\ &-2\sqrt{(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}\\\ =&(a_{1}-a_{4})(a_{1}^{2}+2a_{3}^{2}+a_{1}a_{4})-2a_{3}\sqrt{(a_{1}-a_{4})^{2}(a_{3}^{2}+a_{1}a_{4})}\\\ <&0\\\ \end{array}$ (96) in case of $a_{1}<a_{4}$. Thus, if $a_{1}>a_{4}$, then both $z_{1}$ and $z_{2}$ go to $+\infty$ when $t\rightarrow 0$; if $a_{1}<a_{4}$, then both $z_{1}$ and $z_{2}$ go to $-\infty$ when $t\rightarrow 0$. Now suppose that $a_{1}>a_{4}$. From $z_{1}=1$, there is $\begin{array}[]{rc}[t^{2}a_{0}^{2}a_{1}-2ta_{0}a_{1}^{2}+(1-a_{0}^{2})a_{1}-2a_{4}(a_{2}a_{3}+a_{1}a_{4})]^{2}&\\\ -[2\sqrt{-t^{2}a_{0}^{2}a_{1}a_{2}a_{3}a_{4}+(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}]^{2}&=0,\end{array}$ (97) the solutions of which are $\begin{array}[]{ll}t=&t_{11}=\frac{a_{1}+a_{4}-\sqrt{-(a_{2}+a_{3})^{2}}}{a_{0}},\\\ t=&t_{12}=\frac{a_{1}+a_{4}+\sqrt{-(a_{2}+a_{3})^{2}}}{a_{0}},\\\ t=&t_{13}=\frac{a_{1}-a_{4}-\sqrt{-(a_{2}-a_{3})^{2}}}{a_{0}},\\\ t=&t_{14}=\frac{a_{1}-a_{4}+\sqrt{-(a_{2}-a_{3})^{2}}}{a_{0}}.\\\ \end{array}$ (98) Here, $t_{11}$ and $t_{12}$ are imaginary numbers, and $t_{13}$ and $t_{14}$ are positive real numbers only if $a_{2}=a_{3}$ and $a_{1}>a_{4}$. It is not difficult to check that $\theta_{0}$ is a root of $z_{1}=1$ and the minimum point of $z_{1}$ in case of $a_{1}>a_{4}$ and $a_{2}=a_{3}$. Here $t=\cot\frac{\theta_{0}}{2}=\frac{a_{1}-a_{4}}{a_{0}}$. $z_{1}\geq 1$ comes directly since $z=z_{1}$ is a continuous function of $t$, goes to $+\infty$ when $t\rightarrow 0$, and has only one intersection point with straight line $z=1$, where the equality $z_{1}=1$ holds iff $a_{2}=a_{3}$,$a_{1}>a_{4}$ and $\cot\frac{\theta_{0}}{2}=\frac{a_{1}-a_{4}}{a_{0}}$. We can show that $z_{2}>1$ in case of $a_{1}>a_{4}$ in the same way. Similarly, we can prove that if $a_{1}<a_{4}$, then $z_{1}<-1$, and $z_{2}\leq-1$, where $z_{2}=-1$ iff $a_{2}=a_{3}$, $a_{1}<a_{4}$, and $t=\cot\frac{\overline{\theta}_{0}}{2}=\frac{-a_{1}+a_{4}}{a_{0}}$. Therefore, $P(\theta,\varphi)=0$ iff $\varphi=0$ and $\theta=\theta_{0}$ in case of $a_{2}=a_{3}$ and $a_{1}>a_{4}$, or $\varphi=\pi$ and $\theta=\overline{\theta}_{0}$ in case of $a_{2}=a_{3}$ and $a_{1}<a_{4}$. It can be derived that $\begin{array}[]{ll}&\left(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\right)\big{|}_{P(\theta,\varphi)=0,~{}\theta\in(0,\pi)}\\\ =&\sqrt{Q(\theta_{0},0)}|_{\\{a_{2}=a_{3},a_{1}>a_{4}\\}}\\\ =&\sqrt{Q(\overline{\theta}_{0},\pi)}|_{\\{a_{2}=a_{3},a_{1}<a_{4}\\}}\\\ =&\sqrt{(1-2a_{4}^{2})^{2}+4a_{3}^{2}(-1+a_{1}^{2}+a_{3}^{2}-2a_{1}a_{4}+3a_{4}^{2})}.\\\ \end{array}$ (99) Immediately, from (19) and (20), $Q(\theta,\varphi)=0$ iff $\varphi=\pi$ and $\theta=\pi-\theta_{0}$ in case of $a_{2}=a_{3}$ and $a_{1}>a_{4}$, or $\varphi=0$ and $\theta=\pi-\overline{\theta}_{0}$ in case of $a_{2}=a_{3}$ and $a_{1}<a_{4}$, and $\begin{array}[]{ll}&(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)})|_{Q(\theta,\varphi)=0}\\\ =&(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)})|_{P(\theta,\varphi)=0}.\\\ \end{array}$ (100) From above, it can be seen that the quantum channel (2), the coefficients of which satisfy $\mu=\pi$, $a_{0}a_{1}a_{2}a_{3}a_{4}\neq 0$ and $a_{2}=a_{3}$, can be collapsed to a Bell state with probability $p_{1}=p_{2}=\frac{a_{0}^{2}(1-a_{0}^{2}+3a_{1}^{2}-4a_{1}a_{4}+a_{4}^{2})}{a_{0}^{2}+(a_{1}-a_{4})^{2}}$ by Charlie’s measurement in the basis (3) with $(\theta,\varphi)=(\theta_{0},0)$ or $(\theta,\varphi)=(\pi-\theta_{0},\pi)$ in case of $a_{1}>a_{4}$, or $(\theta,\varphi)=(\overline{\theta}_{0},\pi)$ or $(\theta,\varphi)=(\pi-\overline{\theta}_{0},0)$ in case of $a_{1}<a_{4}$. In particular, this quantum channel can also be purified to an EPR pair with probability $2a_{1}^{2}+2a_{2}^{2}$ via controller’s measurement in the basis $|0\rangle,|1\rangle$ in case of $a_{1}=a_{4}$. In the following, we suppose that $P(\theta,\varphi)Q(\theta,\varphi)\neq 0$. In order to obtain the minimum of $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}$, we need to find the points such that equations (26) and (27) hold. From (26), we get $\sin\varphi=0$, or $\begin{array}[]{ll}&ya_{0}[2a_{1}a_{2}^{2}a_{3}^{2}-a_{2}a_{3}a_{4}(-2a_{0}^{2}-4a_{1}^{2}-2a_{4}^{2}+1)\\\ &-2a_{1}a_{4}^{2}(a_{2}^{2}+a_{3}^{2})]\\\ &-2\cos\varphi(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})\\\ =&0.\\\ \end{array}$ (101) By (27), we have $\sin\varphi=0$, or $\begin{array}[]{ll}&4y^{3}a_{0}^{3}(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})\\\ &-8y^{2}a_{0}^{2}a_{1}\cos\varphi(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})\\\ &+ya_{0}[(a_{1}-2a_{2}a_{3}a_{4}-2a_{1}a_{4}^{2})(2{a_{1}}a_{2}^{2}a_{3}^{2}-{a_{2}}{a_{3}}{a_{4}}\\\ &+4a_{1}^{2}{a_{2}}{a_{3}}{a_{4}}-2{a_{1}}a_{2}^{2}a_{4}^{2}-2{a_{1}}a_{3}^{2}a_{4}^{2}+2{a_{2}}{a_{3}}a_{4}^{3})\\\ &+4a_{1}^{2}\cos^{2}\varphi(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})]\\\ &+2\cos\varphi(-a_{1}+2a_{2}a_{3}a_{4}+2a_{1}a_{4}^{2})[a_{2}^{2}a_{3}^{2}(a_{1}^{2}+a_{4}^{2})\\\ &-a_{1}a_{2}a_{3}a_{4}(-2a_{1}^{2}-2a_{4}^{2}+1)-a_{1}^{2}(a_{2}^{2}+a_{3}^{2})a_{4}^{2}]\\\ =&0.\end{array}$ (102) If $(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})=0$, then (101) and (102) become $y=0,$ (103) and $y[a_{4}^{2}-(1-2a_{0}^{2})a_{1}^{2}]-2a_{0}a_{1}\cos\varphi(a_{1}^{2}+a_{4}^{2})=0,$ (104) respectively. However, if $(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})\neq 0$, then (101) and (102) become $\begin{array}[]{cl}&\cos\varphi\\\ =&\frac{ya_{0}\\{-2{a_{1}}a_{4}^{2}(a_{3}^{2}+a_{2}^{2})+{a_{2}}{a_{3}}[2{a_{1}}{a_{2}}{a_{3}}+{a_{4}}+2(a_{1}^{2}-a_{2}^{2}-a_{3}^{2}){a_{4}}]\\}}{2(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}\\\ \end{array}$ (105) and $\frac{y(1+y^{2})a_{0}^{3}a_{2}^{2}a_{3}^{2}a_{4}^{2}(2a_{2}a_{3}a_{4}-a_{1}(1-2a_{4}^{2}))^{2}}{(a_{2}a_{3}+a_{1}a_{4})(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})}=0,$ (106) respectively. That is, $y=0$, $\cos\varphi=0$. For getting the minimum point of $\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}$, it is enough for us to consider the hyperplanes $\sin\varphi=0$, and $\cos\varphi=0$. For hyperplane $\sin\varphi=0$, we need only to consider the case $\varphi=0$ by (19), (20), and (21). For hyperplane $\cos\varphi=0$, we need only to consider the case $\varphi=\frac{\pi}{2}$ by (19) and (20). Now, let us consider the case $\varphi=0$. In this case (28) becomes $\begin{array}[]{ll}&[-2a_{1}a_{2}a_{3}+a_{4}-2a_{1}^{2}a_{4}+2ya_{0}(a_{2}a_{3}+a_{1}a_{4})]\\\ &\times[4\,y^{2}\,a_{0}^{2}({a_{1}}\,{a_{3}}-{a_{2}}\,{a_{4}})({a_{1}}\,{a_{2}}-{a_{3}}\,{a_{4}})\\\ &+2\,y\,{a_{0}}(2\,{a_{1}}\,{a_{2}}\,{a_{3}}-4\,a_{1}^{3}\,{a_{2}}\,{a_{3}}-2\,{a_{1}}\,a_{2}^{3}\,{a_{3}}\\\ &-2\,{a_{1}}\,{a_{2}}\,a_{3}^{3}-a_{2}^{2}\,{a_{4}}+4\,a_{1}^{2}\,a_{2}^{2}\,{a_{4}}+2\,a_{2}^{4}\,{a_{4}}-a_{3}^{2}\,{a_{4}}\\\ &+4\,a_{1}^{2}\,a_{3}^{2}\,{a_{4}}+2\,a_{3}^{4}\,{a_{4}}-8\,{a_{1}}\,{a_{2}}\,{a_{3}}\,a_{4}^{2}+2\,a_{2}^{2}\,a_{4}^{3}+2\,a_{3}^{2}a_{4}^{3})\\\ &+({a_{2}}-2\,a_{1}^{2}\,{a_{2}}-2\,a_{2}^{3}+2{a_{1}}{a_{3}}{a_{4}}-2{a_{2}}\,a_{4}^{2})\\\ &\times({a_{3}}-2\,a_{1}^{2}{a_{3}}-2\,a_{3}^{3}+2\,{a_{1}}\,{a_{2}}{a_{4}}-2{a_{3}}a_{4}^{2})]\\\ =&0.\\\ \end{array}$ (107) If $(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})\neq 0$, then $\begin{array}[]{ll}y=&\cot\theta_{1}=\frac{2\,{a_{1}}\,{a_{2}}\,{a_{3}}-{a_{4}}+2a_{1}^{2}\,{a_{4}}}{2\,{a_{0}}\,\left({a_{2}}\,{a_{3}}+{a_{1}}\,{a_{4}}\right)},\\\ y=&\cot\theta_{2}=\frac{-{a_{2}}+2a_{1}^{2}\,{a_{2}}+2a_{2}^{3}-2\,{a_{1}}\,{a_{3}}\,{a_{4}}+2\,{a_{2}}a_{4}^{2}}{2\,{a_{0}}\,\left({a_{1}}\,{a_{2}}-{a_{3}}\,{a_{4}}\right)},\\\ y=&\cot\theta_{3}=\frac{-{a_{3}}+2a_{1}^{2}\,{a_{3}}+2a_{3}^{3}-2\,{a_{1}}\,{a_{2}}\,{a_{4}}+2\,{a_{3}}a_{4}^{2}}{2\,{a_{0}}\,\left({a_{1}}\,{a_{3}}-{a_{2}}\,{a_{4}}\right)},\\\ \end{array}$ (108) where $\theta_{1}$ is a root of equation (24), while $\theta_{2}$ is a root of equation (24) only in case of $a_{2}<a_{3}$, and $\theta_{3}$ is a root of equation (24) only in case of $a_{2}>a_{3}$. If $({a_{1}}\,{a_{2}}-{a_{3}}\,{a_{4}})=0$, and $a_{1}\neq a_{4}$, then $a_{2}\neq a_{3}$, and $\begin{array}[]{rl}y&=\cot\theta_{4}=\frac{-a_{1}+2a_{1}^{3}+2a_{1}a_{3}^{2}}{2a_{0}(a_{1}^{2}+a_{3}^{2})},\\\ y&=\cot\theta_{5}=\frac{-a_{1}+2a_{1}^{3}+2a_{1}a_{3}^{2}}{2a_{0}(a_{1}^{2}-a_{4}^{2})},\\\ \end{array}$ (109) where $\theta_{4}$ is a root of equation (24), while $\theta_{5}$ is a root of equation (24) only if $a_{2}>a_{3}$. If $({a_{1}}\,{a_{3}}-{a_{2}}\,{a_{4}})=0$, and $a_{1}\neq a_{4}$, then $a_{2}\neq a_{3}$, and $\begin{array}[]{rl}y&=\cot\theta_{6}=\frac{-a_{1}+2a_{1}^{3}+2a_{1}a_{2}^{2}}{2a_{0}(a_{1}^{2}+a_{2}^{2})},\\\ y&=\cot\theta_{7}=\frac{-a_{1}+2a_{1}^{3}+2a_{1}a_{2}^{2}}{2a_{0}(a_{1}^{2}-a_{4}^{2})},\\\ \end{array}$ (110) where $\theta_{6}$ is a root of equation (24), while $\theta_{7}$ is a root only if $a_{2}<a_{3}$. If $a_{1}=a_{4}$ and $a_{2}=a_{3}$, then $y=\cot\theta_{8}=-\frac{a_{0}a_{1}}{2(a_{1}^{2}+a_{2}^{2})},$ (111) where $\theta_{8}$ is a root of equation (24). Therefore, $\begin{array}[]{ll}&\min\big{\\{}\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\big{\\}}\\\ =&\left\\{\begin{array}[]{lll}\min\\{\left(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\right)|_{a_{2}>a_{3}}\\},&\texttt{if}&a_{2}>a_{3},\\\ \min\\{\left(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\right)|_{a_{2}<a_{3}}\\},&\texttt{if}&a_{2}<a_{3},\\\ \min\\{\left(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\right)|_{a_{2}=a_{3}}\\},&\texttt{if}&a_{2}=a_{3}.\\\ \end{array}\right.\end{array}$ (112) Here, $\begin{array}[]{ll}&\min\\{\left(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\right)|_{a_{2}>a_{3}}~{}\\}\\\ =&\left\\{\begin{array}[]{lll}\min\\{\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{P(\theta_{3},0)}+\sqrt{Q(\theta_{3},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\\},&\texttt{if}&(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})\neq 0,\\\ \min\\{\sqrt{P(\theta_{4},0)}+\sqrt{Q(\theta_{4},0)},\sqrt{P(\theta_{5},0)}+\sqrt{Q(\theta_{5},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\\},&\texttt{if}&a_{1}a_{2}-a_{3}a_{4}=0~{}\texttt{and}~{}a_{1}\neq a_{4},\\\ \min\\{\sqrt{P(\theta_{6},0)}+\sqrt{Q(\theta_{6},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\\},&\texttt{if}&a_{1}a_{3}-a_{2}a_{4}=0~{}\texttt{and}~{}a_{1}\neq a_{4};\\\ \end{array}\right.\end{array}$ (113) $\begin{array}[]{ll}&\min\\{\left(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\right)|_{a_{2}<a_{3}}~{}\\}\\\ =&\left\\{\begin{array}[]{lll}\min\\{\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{P(\theta_{2},0)}+\sqrt{Q(\theta_{2},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\\},&\texttt{if}&(a_{1}a_{3}-a_{2}a_{4})(a_{1}a_{2}-a_{3}a_{4})\neq 0,\\\ \min\\{\sqrt{P(\theta_{4},0)}+\sqrt{Q(\theta_{4},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\\},&\texttt{if}&a_{1}a_{2}-a_{3}a_{4}=0~{}\texttt{and}~{}a_{1}\neq a_{4},\\\ \min\\{\sqrt{P(\theta_{6},0)}+\sqrt{Q(\theta_{6},0)},\sqrt{P(\theta_{7},0)}+\sqrt{Q(\theta_{7},0)},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\\},&\texttt{if}&a_{1}a_{3}-a_{2}a_{4}=0~{}\texttt{and}~{}a_{1}\neq a_{4};\\\ \end{array}\right.\end{array}$ (114) $\begin{array}[]{ll}&\min\\{\left(\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\right)|_{a_{2}=a_{3}}~{}\\}\\\ =&\left\\{\begin{array}[]{lll}\min\\{\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{Q(\theta_{0},0)}|_{a_{2}=a_{3},a_{1}>a_{4}},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\\},&\texttt{if}&a_{1}>a_{4},\\\ \min\\{\sqrt{P(\theta_{1},0)}+\sqrt{Q(\theta_{1},0)},\sqrt{Q(\overline{\theta}_{0},\pi)}|_{a_{2}=a_{3},a_{1}<a_{4}},\sqrt{P(0,0)}+\sqrt{Q(0,0)}\\},&\texttt{if}&a_{1}<a_{4},\\\ \min\\{\sqrt{P(\theta_{8},0)}+\sqrt{Q(\theta_{8},0)},a_{0}^{2}\\},&\texttt{if}&a_{1}=a_{4}.\\\ \end{array}\right.\end{array}$ (115) By (19), (20) and (21), we have $\begin{array}[]{ll}\min\\{\sqrt{P(\theta,\pi)}+\sqrt{Q(\theta,\pi)}\\}=&\min\\{\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\\}=\min\\{\sqrt{P(\theta,2\pi)}+\sqrt{Q(\theta,2\pi)}\\}\\\ \end{array}$ (116) Now let us examine the case $\varphi=\frac{\pi}{2}$. In this case, (28) goes to $\displaystyle 4ya_{0}^{4}[1-2a_{0}^{2}-4(a_{2}a_{3}+a_{1}a_{4})^{2}][a_{2}^{2}a_{3}^{2}$ (117) $\displaystyle+2a_{1}a_{2}a_{3}a_{4}-a_{4}^{2}(a_{2}^{2}+a_{3}^{2})](1+y^{2})^{-\frac{3}{2}}=$ $\displaystyle 0,$ which implies that $\theta=\frac{\pi}{2}$. It is shown that $\theta=\frac{\pi}{2}$ is the only possible extreme point of both $\sqrt{P(\theta,\frac{\pi}{2})}+\sqrt{Q(\theta,\frac{\pi}{2})}$ and $\sqrt{P(\theta,\frac{3\pi}{2})}+\sqrt{Q(\theta,\frac{3\pi}{2})}$, and $\begin{array}[]{ll}&\sqrt{P(\frac{\pi}{2},\frac{\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{\pi}{2})}\\\ =&\sqrt{P(\frac{\pi}{2},\frac{3\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{3\pi}{2})}\\\ =&\sqrt{1-4(a_{2}a_{3}+a_{1}a_{4})^{2}-4a_{0}^{2}a_{4}^{2}}.\\\ \end{array}$ (118) To sum up, $\begin{array}[]{ll}&\min\\{\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)}\\}\\\ =&\min\Big{\\{}\sqrt{P(0,\varphi)}+\sqrt{Q(0,\varphi)},\sqrt{P(\frac{\pi}{2},\frac{\pi}{2})}+\sqrt{Q(\frac{\pi}{2},\frac{\pi}{2})},\\\ &~{}~{}~{}~{}~{}~{}~{}\min\\{\sqrt{P(\theta,0)}+\sqrt{Q(\theta,0)}\\},\\\ &~{}~{}~{}~{}~{}~{}~{}(\sqrt{P(\theta,\varphi)}+\sqrt{Q(\theta,\varphi)})|_{P(\theta,\varphi)=0}\Big{\\}}.\\\ \end{array}$ (119) ## IV Conclusion In brief, we give the analytic expression of the localizable entanglement (LE), the maximum of probability of successfully controlled teleporting an unknown qubit state (14) via every three-qubit state (2) satisfying $a_{1}a_{2}a_{3}a_{4}\sin\mu=0$ and investigate how to achieve it (that is, Charlie finds optimal measurement basis, performs optimal measurements, and communicates the results). The authors thank Prof. J. I. Cirac for his fruitful discussions and for his hospitality during their stay at Max-Planck-Institut für Quantenoptik. This work was supported by the National Natural Science Foundation of China under Grant No: 10671054, Hebei Natural Science Foundation of China under Grant Nos: A2005000140, 07M006, and the Key Project of Science and Technology Research of Education Ministry of China under Grant No: 207011. ## References * (1) C. H. Bennett, G. Brassard, C. Crepeau, R. Jozsa, A. Peres, and W. K. Wooters, Phys. Rev. Lett. 70, 1895 (1993). * (2) M. Fujii, Phys. Rev. A 68, 050302 (2003). * (3) N. Ba An, Phys. Rev. A 68, 022321 (2003). * (4) W. P. Bowen, N. Treps, B. C. Buchler, R. Schnabel, T. C. Ralph, Hans-A. Bachor, T. Symul, and P. K. Lam, Phys. Rev. A 67, 032302 (2003). * (5) T. J. Johnson, S. D. Bartlett, and B. C. Sanders, Phys. Rev. A 66, 042326 (2002). * (6) L. Vaidman, Phys. Rev. A 49, 1473 (1994). * (7) S. L. Braunstein and H. J. Kimble, Phys. Rev. Lett. 80, 869 (1998). * (8) W. Son, J. Lee, M. S. Kim, and Y.-J. Park, Phys. Rev. A 64, 064304 (2001). * (9) D. Bruß, D. P. DiVincenzo, A. Ekert, C. A. Fuchs, C. Macchiavello, and J. A. Smolin, Phys. Rev. A, 57, 2368 (1998). * (10) G. Gordon and G. Rigolin, Phys. Rev. A 73, 042309 (2006). * (11) T. Gao, F. L. Yan, and Z. X. Wang, Quantum Information and Computation, 4, 186 (2004). * (12) A. Karlsson and M. Bourennane, Phys. Rev. A 58, 4394 (1998). * (13) C. P. Yang, S. I. Chu, and S. Han, Phys. Rev. A 70, 022329 (2004). * (14) A. K. Pati, Phys. Rev. A 61, 022308 (2000). * (15) P. Agrawal and A. K. Pati, Phys. Lett. A 305, 12 (2002). * (16) A. K. Pati and P. Agrawal, J. Opt. B: Quantum Semiclassical Opt. 6, S844 (2004). * (17) F. L. Yan and D. Wang, Phys. Lett. A 316, 297 (2003). * (18) F. G. Deng, C. Y. Li, Y. S. Li, H. Y. Zhou, and Y. Wang, Phys. Rev. A 72, 022338 (2005). * (19) F. L. Yan, and X. Q. Zhang, The European Physical Journal B, 41 75 (2004). * (20) D. Bouwmeester, J. W. Pan, K. Mattle, M. Eibl, H. Weinfurter, and A. Zeilinger, Nature (London) 390, 575 (1997). * (21) A. Furusawa, J. L. S rensen, S. L. Braunstein, C. A. Fuchs, H. J. Kimble, and E. S. Polzik, Science 282, 706 (1998). * (22) M. A. Nielsen, E. Knill, and R. Laflamme, Nature (London) 396, 52 (1998). * (23) M. Hillery, V. Bužek, and A. Berthiaume, Phys. Rev. A 59, 1829 (1999). * (24) B. Aoun, M.Tarifi, e-print, quantum-ph/0401076. * (25) E. Biham, B. Huttner, and T. Mor, Phys. Rev. A 54, 2651 (1996). * (26) P. D. Townsend, Nature 385, 47 (1997). * (27) S. Bose, V. Vedral, and P. L. Knight, Phys. Rev. A 57, 822 (1998). * (28) T. Gao, Z. Naturforsch. 59a, 597 (2004). * (29) F. Verstraete, M. Popp, and J. I. Cirac, Phys. Rev. Lett. 92, 027901 (2004). * (30) T. Gao, F. L. Yan, and Y. C. Li, e-print, quantum-ph/0710.1055. * (31) A. Acín, A. Andrianov, L. Costa, E. Jané, J. I. Latorre, and R. Tarrach, Phys. Rev. Lett. 85, 1560 (2000).

arxiv-papers

2024-09-04T02:48:53.834055

0802.2520

# Binary Black Hole Encounters, Gravitational Bursts and Maximum Final Spin Matthew C. Washik James Healy Frank Herrmann Ian Hinder Deirdre M. Shoemaker Pablo Laguna Center for Gravitational Wave Physics The Pennsylvania State University, University Park, PA 16802 Richard A. Matzner Center for Relativity and Department of Physics The University of Texas at Austin, Austin, TX 78712 ###### Abstract The spin of the final black hole in the coalescence of nonspinning black holes is determined by the “residual” orbital angular momentum of the binary. This residual momentum consists of the orbital angular momentum that the binary is not able to shed in the process of merging. We study the angular momentum radiated, the spin of the final black hole and the gravitational bursts in a sequence of equal mass encounters. The initial orbital configurations range from those producing an almost direct infall to others leading to numerous orbits before infall, with multiple bursts of radiation while merging. Our sequence consists of orbits with fixed impact parameter. What varies is the initial linear, or equivalently angular, momentum of the black holes. For this sequence, the final black hole of mass $M_{h}$ gets a maximum spin parameter $a/M_{h}\approx 0.823\,$, with this maximum occurring for initial orbital angular momentum $L/M^{2}_{h}\approx 1.176\,$. ###### pacs: 04.60.Kz,04.60.Pp,98.80.Qc A few years ago, after a decades-long period of development, breakthroughs were made in computational modeling of strong gravitational fields that now allow numerical relativists to successfully simulate binary black holes (BBH) from inspiral through merger. In general terms, there are now two computational recipes to follow. One of them is based on a generalized harmonic formulation of the Einstein equations Pretorius (2006, 2005) and uses _excision_ Brandt et al. (2000); Shoemaker et al. (2003) of the black hole (BH) singularities. The other recipe, called the moving puncture recipe, involves a BSSN Shibata et al. (1992); Baumgarte and Shapiro (1999) formulation, _punctures_ to model BH singularities and a gauge condition for these punctures to move throughout the computational domain Campanelli et al. (2006a); Baker et al. (2006). Using these recipes, many studies involving interacting BHs and their generated gravitational radiation have been carried out, including gravitational recoil Baker et al. (2007a); Gonzalez et al. (2007); Herrmann et al. (2007a); Baker et al. (2006); Campanelli et al. (2007a); Pollney et al. (2007); Koppitz et al. (2007), spin hang-up Campanelli et al. (2006b) and matches to post-Newtonian (PN) approximations Baker et al. (2007b); Hannam et al. (2007). Most center on astrophysical implications and connection to future gravitational wave observations. BBH simulations also enable studies of strong non-linear phenomena regardless of traditional gravitational astrophysics consequences. A recent example is the work in Ref. Pretorius and Khurana (2007) on the self-similar behavior found in the approach to the merger/flyby threshold of BBHs. Similar merger thresholds in BBH encounters or scatterings form the context for our work. We consider orbits in which the BHs initially fly past one another, but then fall back to orbit and merge. We focus on the gravitational waveform and the angular momentum radiated from such encounters. Serendipitously, we find significant astrophysical implications, both the existence of a maximum in the final BH spin and of multiple encounter orbits with associated multiple bursts of gravitational radiation. Ref. Pretorius and Khurana (2007) considered only the first close encounter or “whirl,” and the study did not extend the evolutions to find possible fall-back orbits such as those here considered. The work in Ref. Pretorius and Khurana (2007) and our work here have to date been the only studies considering these highly eccentric orbits; while there have been high-order PN studies of inspiral, cases studied so far have described relatively smooth inspirals Königsdörffer and Gopakumar (2006). All our orbits are parabolic or hyperbolic encounters. Depending on the merger, the fraction of angular momentum radiated varies significantly ($0.05\lower 2.15277pt\hbox{$\;\buildrel<\over{\sim}\;$}J_{rad}/L\lower 2.15277pt\hbox{$\;\buildrel<\over{\sim}\;$}0.55$ with $L$ the initial orbital angular momentum of the binary). This emission of angular momentum sets an upper limit of $a/M_{h}\approx 0.823\,$for the spin parameter of the final BH; this maximum occurs when $L/M^{2}_{h}\approx 1.176\,$, with $M_{h}$ the mass of the final merged BH. As in our previous BBH studies Herrmann et al. (2007a, b); Herrmann et al. (2007), we use a code based on the BSSN formulation and the moving puncture recipe. The results here were obtained with a $634^{3}\,M$ computational domain consisting of 10 refinement levels, with finest resolution of $M/52$. We set up nonspinning equal mass BHs using Bowen-York initial data Bowen and York (1980). The mass of each BH is $M/2$, computed from $\sqrt{A_{ah}/16\pi}$ with $A_{ah}$ their apparent horizon area. The data have the BHs on the $x$-axis: BH± is located at $\pm 5\,M$ and has linear momentum $\vec{P}_{\pm}=(\mp P\,\cos{\theta},\pm P\,\sin{\theta},0)$. We keep the angle constant at $\theta=26.565^{\circ}=\tan^{-1}(1/2)$; thus the impact parameter is $\sim 4.47\,M$. The total initial orbital angular momentum is given by $\vec{L}/M^{2}=10\,(P/M)\,\sin{\theta}\,\hat{z}$. We obtain a one-parameter family of initial data by varying the magnitude of the initial momentum in the range $0.1145\leq P/M\leq 0.3093$. At the lower limit of the momenta, merger occurs within less than half an orbit of inspiral. We then consider successively higher initial momentum until we find solutions that will clearly require a very long, “infinite” time to merge. The results are summarized in Figs. 1 and 2. The top panel in Fig. 1 shows the spin $a/M_{h}$ of the final BH as a function of the initial orbital angular momentum $L/M_{h}^{2}$. The spin and mass of the final BH were computed using the apparent horizon formula Campanelli et al. (2007b); Herrmann et al. (2007b). The bottom panel of Fig. 1 displays the fraction of angular momentum radiated ($J_{rad}/L=1-a\,M_{h}/L$). Figure 2 shows, as a function of $L/M_{h}^{2}$, in the top panel the final mass $M_{h}/M_{adm}$ relative to the total ADM mass and in the bottom panel the fraction of energy radiated $E_{rad}/M_{adm}=1-M_{h}/M_{adm}$. The vertical lines in Figs. 1 and 2 denote the value of $L/M_{h}^{2}$ where $a/M_{h}$ is maximum. We have also calculated both the radiated angular momentum and energy via the Weyl tensor. The results are consistent with those in Figs. 1 and 2. However, the values obtained form $J_{rad}/L=1-a\,M_{h}/L$ and $E_{rad}/M_{adm}=1-M_{h}/M_{adm}$ are more accurate because they are not as susceptible to resolution effects as those derived from wave extraction. We have carried out simulations at resolutions of $M/45$, $M/48$, $M/52$ and $M/64$ for ten representative cases in Figs. 1 and 2 to check convergence and make error estimates. We found that the results are consistent with the 4th-order accuracy of our code and that the errors in the quantities displayed in these figures are not larger than 3%. We have selected six encounters that are representative of the different behaviors in our series. These six cases are $L/M^{2}=$ 0.512, 1.208, 1.352, 1.376, 1.382 and 1.387 or equivalently $L/M_{h}^{2}=$ 0.521, 1.282, 1.522, 1.480, 1.498 and 1.554. We will refer to them as encounters Ea, Eb, Ec, Ed, Ee and Ef, respectively. Cases Ed, Ee and Ef correspond to the last three points in Figs. 1 and 2. Figure 1: Top panel, spin of the final BH $a/M_{h}$ and, bottom panel, angular momentum radiated $J_{rad}/L$ vs the initial orbital angular momentum $L/M_{h}^{2}$. Figure 2: Top panel, mass of the final BH $M_{h}/M_{adm}$ and, bottom panel, energy radiated $E_{rad}/M_{adm}$ vs the initial orbital angular momentum $L/M_{h}^{2}$. Figure 3: BH tracks of the encounters. The coordinate dimensions of the top four panels are $12\,M\times 12\,M$ and $16\,M\times 16\,M$ for the 2 bottom panels. Figure 4: Waveforms for the Ea, Eb and Ec encounters. For $L/M_{h}^{2}\lower 2.15277pt\hbox{$\;\buildrel<\over{\sim}\;$}0.8$, the radiated angular momentum is $J_{rad}/L\lower 2.15277pt\hbox{$\;\buildrel<\over{\sim}\;$}0.15$, so the final BH has $a/M_{h}$ close to $L/M_{h}^{2}$. The evolution is rather simple in these cases: immediate merger, with minimal inspiral. For instance, in case Ea (Fig. 3), $L/M_{h}^{2}=0.521$, and $J_{rad}/L=0.05$; thus most of the angular momentum goes into the final BH, $a/M_{h}=0.496$. Fig. 4.Ea shows the corresponding radiated gravitational wave ($M\,r\,$Re$\Psi_{4}^{2,2}$). All waveforms were extracted at radius $50\,M$. As the initial angular momentum increases, the radiated angular momentum also increases, suppressing and limiting the spin of the final BH. Eventually for large enough initial angular momentum, so much angular momentum is radiated that, as seen in Fig. 1, the final spin reaches a maximum of $a/M_{h}\approx 0.823\,$at $L/M^{2}_{h}\approx 1.176\,$. Fig. 3.Eb shows the tracks of the BHs in the neighborhood of this maximum. Fig. 4.Eb shows the corresponding radiated waveform. For even larger initial angular momentum, the spin of the final BH actually decreases for increasing $L/M_{h}^{2}$. The reason is that the merger is not only preceded by several hang-up orbits Campanelli et al. (2006b); Pretorius and Khurana (2007), but also the merger yields a highly distorted BH that radiates copiously as it settles down. Case Ec with $a/M_{h}\approx 0.68$ and $L/M_{h}^{2}\approx 1.522$ represents this situation in which almost 50% of the initial angular momentum is radiated (see path in Fig. 3.Ec and radiated waveform in Fig. 4.Ec). A persistent feature of the mergers with $L/M_{h}^{2}\lower 2.15277pt\hbox{$\;\buildrel<\over{\sim}\;$}1.3$ is that the separation between the BHs (the coordinate distance between the punctures) decreases monotonically with time (monotonic inspiral). Comparing cases Ea, Eb and Ec in Fig. 4, we see general qualitative agreement: inspiral-generated gravitational waves with frequency and amplitude increasing in time, followed by essentially fixed-frequency ringdown waves. There is, however, a hint of disappearance of the monotonic spiral in case Ec. The amplitude of the gravitational radiation has a “shoulder” at about time $\sim 110\,M$. For a period of time equal to two wave oscillations, the decline of the amplitude ceases and then recommences. The relative orbital separation as a function of time (Fig. 6.Ec) clearly shows there is a plateau in the separation centered at time $\sim 50\,M$, which is absent for cases Ea or Eb. For a brief period of time there is a closely circular phase in which the BHs “want” to fly apart, but just manage to stay at roughly constant separation. The last three points in Figs. 1 and 2 are the cases labeled Ed, Ee and Ef. They describe orbits without immediate merger but “escape” and recapture; they all show initial approaches followed by increasing mid-evolution separations of $14\,M$, $25\,M$ and $42\,M$ before the final merger (see Figs. 3 and 6). Because the interaction involves two close approaches, there are two bursts of gravitational radiation, one from the first flyby Yunes et al. (2007) and the other from the merger (see Fig. 5). We are currently investigating astrophysical implications of detections of these multiple gravitational bursts and hangups in globular clusters Kocsis et al. (2006). For the Ef case, there is an approximate hangup with separation $\sim 4-5\,M$ around time $\sim 950\,M$ similar to the shoulder seen in Fig. 6.Ec around time $\sim 50\,M$. This structure shows up in the waveform for this Ef case; we actually see a (lower amplitude) precursor to the radiation burst associated with the merger, a hint that orbits with many repeated bounces are possible. For even slightly ($0.1\%$) greater initial angular momentum than case Ef, the BHs complete approximately one loop and then escape. This is a possible indication of chaotic behavior (exponential dependence on initial conditions, c.f. Ref. Pretorius and Khurana (2007)). Repeated bounce orbits would have to be found with initial angular momentum very slightly above that which resulted in Fig. 3.Ef. As with all critical phenomena, the problem becomes one of careful tuning of the parameters. Note that these interactions of nonspinning BHs produce chaotic orbital dynamics, in contrast to the chaos found in spin evolutions Levin (2000); Hartl and Buonanno (2005). One of the main conclusions of our work is that there is an upper limit on the Kerr parameter for the final merged BH from nonspinning BH merger. For our sequence this maximum is $a/M_{h}\approx 0.823\,$. We can understand this observation by examining the timing of the formation of the final BH and the radiation from the merger. It appears that the merger occurs through an intermediate excited state which is essentially a highly distorted BH. We say “essentially” because a substantial amount of angular momentum is also radiated in the plunge immediately before the apparent horizon forms. This is consistent with close limit BBH calculations Price and Pullin (1994) that show merging BHs behaving like a perturbed BH, even before a common apparent horizon forms, so long as the merging BHs are inside the peak of the effective potential of what will be the final BH. This intermediate state emits the largest part of the radiated energy and angular momentum. Because this mechanism is universal (excitation of such a state is inevitable, and it will inevitably radiate), it suggests that no merger of equal mass (or presumably, roughly equal mass) BHs can lead to a final BH with maximal spin parameter $a/M_{h}\approx 1$. This result does not directly affect spin up by accretion since mass accretion will not excite the low $l$ modes that strongly radiate angular momentum. Thus typical gas accretion can in principle lead to final spins much closer to the limit $a/M_{h}=1$. Figure 5: Waveforms for the Ed, Ee and Ef encounters. Figure 6: BBH coordinate separation for the Ec, Ed, Ee and Ef encounters. Work was supported by NSF grants PHY-0653443 to DS, PHY-0653303, PHY-0555436 and PHY-0114375 to PL and PHY-0354842 and NASA grant NNG 04GL37G to RAM. Computations under allocation TG-PHY060013N, and at the Texas Advanced Computation Center, University of Texas at Austin. We thank M. Ansorg, T. Bode, A. Knapp and E. Schnetter for contributions to our computational infrastructure. ## References * Pretorius (2006) F. Pretorius, Class. Quant. Grav. 23, S529 (2006), eprint gr-qc/0602115. * Pretorius (2005) F. Pretorius, Phys. Rev. Lett. 95, 121101 (2005). * Brandt et al. (2000) S. Brandt, R. Correll, R. Gómez, M. F. Huq, P. Laguna, L. Lehner, P. Marronetti, R. A. Matzner, D. Neilsen, J. Pullin, et al., Phys. Rev. Lett. 85, 5496 (2000). * Shoemaker et al. (2003) D. Shoemaker, K. L. Smith, U. Sperhake, P. Laguna, E. Schnetter, and D. Fiske, Class. Quantum Grav. 20, 3729 (2003), gr-qc/0301111. * Shibata et al. (1992) M. Shibata, T. Nakamura, and K. Oohara, Prog. Theor. Phys. 88, 1079 (1992). * Baumgarte and Shapiro (1999) T. W. Baumgarte and S. L. Shapiro, Astrophys. J, to appear pp. 4849–4857 (1999), eprint astro-ph/9801294. * Campanelli et al. (2006a) M. Campanelli, C. O. Lousto, P. Marronetti, and Y. Zlochower, Phys. Rev. Lett. 96, 111101 (2006a). * Baker et al. (2006) J. G. Baker, J. Centrella, D.-I. Choi, M. Koppitz, and J. van Meter, Phys. Rev. Lett. 96, 111102 (2006). * Baker et al. (2007a) J. G. Baker et al., Astrophys. J. 668, 1140 (2007a), astro-ph/0702390. * Gonzalez et al. (2007) J. A. Gonzalez, U. Sperhake, B. Bruegmann, M. Hannam, and S. Husa, Phys. Rev. Lett. 98, 091101 (2007). * Herrmann et al. (2007a) F. Herrmann, I. Hinder, D. Shoemaker, P. Laguna, and R. A. Matzner (2007a), eprint gr-qc/0701143. * Baker et al. (2006) J. G. Baker, J. Centrella, D.-I. Choi, M. Koppitz, J. R. van Meter, and M. C. Miller, Ap. J. Lett. 653, L93 (2006). * Campanelli et al. (2007a) M. Campanelli, C. O. Lousto, Y. Zlochower, and D. Merritt, Phys. Rev. Lett. 98, 231102 (2007a), gr-qc/0702133. * Pollney et al. (2007) D. Pollney et al., Phys. Rev. D76, 124002 (2007), arXiv:0707.2559 [gr-qc]. * Koppitz et al. (2007) M. Koppitz et al. (2007), gr-qc/0701163. * Campanelli et al. (2006b) M. Campanelli, C. O. Lousto, and Y. Zlochower, Phys. Rev. D 74, 041501 (2006b). * Baker et al. (2007b) J. G. Baker, J. R. van Meter, S. T. McWilliams, J. Centrella, and B. J. Kelly, Phys. Rev. Lett. 99, 181101 (2007b), gr-qc/0612024. * Hannam et al. (2007) M. Hannam, S. Husa, U. Sperhake, B. Brugmann, and J. A. Gonzalez, preprint (arXiv:0706.1305) (2007). * Pretorius and Khurana (2007) F. Pretorius and D. Khurana, Class. Quant. Grav. 24, S83 (2007), eprint gr-qc/0702084. * Königsdörffer and Gopakumar (2006) C. Königsdörffer and A. Gopakumar, Phys. Rev. D 73, 124012 (2006). * Herrmann et al. (2007b) F. Herrmann, I. Hinder, D. M. Shoemaker, P. Laguna, and R. A. Matzner, Phys. Rev. D76, 084032 (2007b), eprint arXiv:0706.2541 [gr-qc]. * Herrmann et al. (2007) F. Herrmann, I. Hinder, D. Shoemaker, and P. Laguna, Class. Quant. Grav. 24, S33 (2007). * Bowen and York (1980) J. M. Bowen and J. W. York, Phys. Rev. D 21, 2047 (1980). * Campanelli et al. (2007b) M. Campanelli, C. O. Lousto, Y. Zlochower, B. Krishnan, and D. Merritt, Phys. Rev. D75, 064030 (2007b), eprint gr-qc/0612076. * Yunes et al. (2007) N. Yunes, C. F. Sopuerta, L. J. Rubbo, and K. Holley-Bockelmann (2007), eprint arXiv:0704.2612 [astro-ph]. * Kocsis et al. (2006) B. Kocsis, M. E. Gaspar, and S. Marka, Astrophys. J. 648, 411 (2006), eprint astro-ph/0603441. * Levin (2000) J. Levin, Physical Review Letters 84, 3515 (2000), eprint arXiv:gr-qc/9910040. * Hartl and Buonanno (2005) M. D. Hartl and A. Buonanno, Phys. Rev. D 71, 024027 (2005), eprint arXiv:gr-qc/0407091. * Price and Pullin (1994) R. H. Price and J. Pullin, Phys. Rev. Lett. 72, 3297 (1994), eprint gr-qc/9402039.

arxiv-papers

2024-09-04T02:48:53.843136

0802.2556

# The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud Gopal Narayanan11affiliation: Dept. of Astronomy, Univ. of Massachusetts, Amherst MA 01003 , Mark H. Heyer11affiliation: Dept. of Astronomy, Univ. of Massachusetts, Amherst MA 01003 , Christopher Brunt11affiliation: Dept. of Astronomy, Univ. of Massachusetts, Amherst MA 01003 22affiliation: University of Exeter , Paul F. Goldsmith33affiliation: Jet Propulsion Laboratory, California Institute of Technology , Ronald Snell11affiliation: Dept. of Astronomy, Univ. of Massachusetts, Amherst MA 01003 , Di Li33affiliation: Jet Propulsion Laboratory, California Institute of Technology gopal@astro.umass.edu ###### Abstract The FCRAO Survey of the Taurus Molecular Cloud observed the 12CO and 13CO J=1-0 emission from 98 deg2 of this important, nearby star forming region. This set of data with 45″ resolution comprises the highest spatial dynamic range image of an individual molecular cloud constructed to date, and provides valuable insights to the molecular gas distribution, kinematics, and the star formation process. In this contribution, we describe the observations, calibration, data processing, and characteristics of the noise and line emission of the survey. The angular distribution of 12CO and 13CO emission over 1 km s-1 velocity intervals and the full velocity extent of the cloud are presented. These reveal a complex, dynamic medium of cold, molecular gas. ISM: clouds — ISM: general — ISM: molecules — ISM: kinematics and dynamics — surveys ## 1 Introduction The Taurus Molecular Cloud has long been a valued target for studies of star formation and the interstellar medium. Its proximity (140 pc) and displacement from the Galactic Plane (b$\sim$-19∘) afford high spatial resolution views of a star forming region with little or no confusion from background stars and gas. However, the large angular extent of the Taurus cloud ($>$100 deg2) on the sky has limited most previous investigations to small, targeted areas or full coverage with coarse angular resolution. Given the complexity of the ISM and the star formation process, such limited views may miss fine, structural features within the cloud or fail to recognize large scale patterns in the gas. Yet such structure provides critical clues to the prevailing physical processes that govern the formation of stars. The sheer volume of the data that have been obtained and the number of analyses that have been carried out preclude giving a complete listing of the references to Taurus, so we will have to be selective rather than comprehensive, recognizing that we may have omitted many valuable contributions. The most complete inventory of the molecular gas content within the Taurus cloud is provided by Ungerechts & Thaddeus (1987), who observed 12CO J=1-0 emission from 750 deg2 of the Taurus-Auriga-Perseus regions. They estimate the molecular mass resident within the Taurus-Auriga cloud to be 3.5$\times$104 M⊙. However, the 30′ angular resolution of this survey precludes an examination of the small scale structure of the cloud. Targeted studies with higher angular resolution of 13CO and C18O emission from individual sub-clouds of Taurus reveal some of the relationships between the molecular gas, magnetic fields, and star formation but offer little insight to the coupling of these structures to larger scales and features (Schloerb & Snell, 1984; Heyer et al. , 1987; Mizuno et al. , 1995; Onishi et al. , 1996). Other studies have utilized other molecular tracers and even higher angular resolution to probe gas having different characteristics in some limited small regions of Taurus. Some examples include Langer et al. (1995) employing CCS, Onishi et al. (1996) and Onishi et al. (1998) using C18O, Onishi et al. (2002) using H13CO+, and Tatematsu et al. (2004) employing N2H+. Many individual cores have been observed in NH3, a tracer in which they appear well-defined, as indicated by the compilation of Jijina, Myers, & Adams (1999). The Leiden/Dwingeloo 21 cm study (Burton & Hartmann, 1994) traced the atomic hydrogen towards Taurus, but with an angular resolution of 35′. One investigation (Shuter et al., 1987) used the Arecibo radio telescope having an angular resolution of 4′, but included only $\sim$1300 positions to probe the self–absorption seen in the 21 cm HI line. This cold atomic hydrogen appears to be associated with molecular gas (Li & Goldsmith, 2003; Goldsmith & Li, 2005), but the limited sampling of Shuter et al. does not reveal much about its morphology. The far–infrared emission from Taurus has been studied by Abergel et al. (1995), who also compared it to moderate resolution maps of 13CO J = 1$\rightarrow$0 emission. The dust column density distribution has been examined from 2MASS extinction by Padoan et al. (2002) and does bear a quite close resemblance to the integrated intensity of 13CO, and thus to the column density of gas. With the deployment of heterodyne focal plane arrays at millimeter and submillimeter wavelengths mounted on large or moderate-sized telescopes, it is possible to construct high spatial dynamic range, spectroscopic images of molecular line emission from interstellar clouds (Heyer, 2000). In this contribution, we present the data from the spectroscopic imaging campaign of 12CO and 13CO J=1-0 emission obtained with the FCRAO 14m telescope and the 32 element SEQUOIA focal plane array. The combined dataset of 12CO and 13CO allows us to make a much more accurate inventory of gas column density and mass for the Taurus molecular cloud than hitherto attempted. At high gas column densities, because of its greater optical depth, 12CO is insensitive to variations in underlying column density. However, because of its higher abundance, it is effective in tracing much of the lower column density regions in the molecular cloud. The 13CO transition, with its lower optical depth, allows us to probe deeper into the envelopes and trace the underlying structure of the molecular gas. Using both 12CO and 13CO we can derive the structure and column density of the molecular gas in different regimes of column density. The data from this survey constitute the most detailed view constructed to date of a molecular cloud and provides critical information for detailed studies of gas dynamics within the molecular interstellar medium. In this paper, we summarize the instrumentation, and data collection and analysis methods, and present global characteristics of noise and line emission from this survey. Detailed scientific motivation and results can be found in Goldsmith et al. (2008). ## 2 Observations ### 2.1 Instrumentation All observations presented in this paper were taken with the 14 meter diameter millimeter-wavelength telescope and the 32 pixel focal plane array SEQUOIA (Erickson et al. , 1999) of the Five College Radio Astronomy Observatory. The FWHM beam sizes of the telescope at the observed frequencies are 45″ (115.271202 GHz) and 47″ (110.201353 GHz). The main beam efficiencies at these frequencies are 0.45 and 0.50 respectively as determined from measurements of Jupiter. Previous measurements of the extended error beam of the telescope and radome structure by measuring the disks of the sun and moon indicate that there can be $\sim 25$% net contribution from extended emission outside the main beam from a region $\sim 0.5\arcdeg$ in diameter. The shape of this error beam is approximately circular, but the amount of contribution of emission at any given point from this error beam pattern depends on details of the distribution of the emission from the source. All data presented here are in T${}_{A}^{*}$ (K), uncorrected for telescope beam efficiencies. The backends were comprised of a system of 64 autocorrelation spectrometers each configured with 25 MHz bandwidth and 1024 spectral channels. No smoothing was applied to the autocorrelation function so the spectral resolution was 29.5 kHz per channel corresponding to 0.076 km s-1 (12CO) and 0.080 km s-1(13CO) at the observed frequencies. The total coverage in velocity is 65 km s-1 (12CO) and 68 km s-1 (13CO) respectively. The spectrometers were centered at a vLSR of 6 km s-1. ### 2.2 Data Collection The Taurus Molecular Cloud was observed over two observing seasons starting in November 2003 and ending in May 2005. The 12CO and 13CO lines were observed simultaneously enabling excellent positional registration and calibration. System temperatures ranged from 350-500 K for the 12CO line and 150-300 K for the 13CO line. The fiducial center position of the map was $\alpha$(2000) = $04^{h}32^{m}44^{s}.6$, $\delta$(2000) = $24\arcdeg 25\arcmin 13\arcsec.08$. The $\sim 98$ deg2 selected region of the cloud was divided into 356 submaps, each $30\arcmin\times 30\arcmin$ in size. Each submap was observed using the On-The-Fly (OTF) mapping technique in which the telescope is continuously scanned back and forth along the Right Ascension axis while rapidly reading the spectrometers. The telescope was scanned at a rate such that 2 samples of data were collected while traversing the FWHM beam width at 115 GHz (45″). An additional ramping offset of 1′ at the beginning and end of each scan was used to ensure stable motion of the antenna. OTF scans were spaced by 34″ in declination. Since the SEQUOIA dewar is fixed with respect to the Azimuth- Elevation coordinate system, the scanning rate and declination offset between scan rows assured that the target field was densely sampled by the array. A given position on the sky was observed by most of the pixels in the array. Such redundant measurements increase the effective integration time at that position and also dilute the effects that arise from the small variations of noise and gain between the pixels of the array. With a focal plane array and ramping offsets, there are data points that lie outside of the target 30′$\times$30′ area. Upon convolution into a regularly spaced grid (see §2.3), these peripheral points contribute to the integration times of points located within but near the edge of the target area. To increase our mapping efficiency the submaps were spaced by 33′, which enabled the peripheral spectra from contiguous submaps to be coadded with sufficient integration time to achieve the desired sensitivity. The target T${}_{A}^{*}$ sensitivities, $\sigma$, for each convolved (see §2.3) 12CO and 13CO spectrum were 0.65 K and 0.25 K respectively as gauged by the root-mean square of antenna temperature values within signal-free backend channels. The sensitivity of a submap was evaluated by the median rms value of $\sigma$, from the set of constituent spectra. Submaps with rms values in excess of 0.75 K were re-observed until the median value of $\sigma$ of the co-added set of data was less than this maximum allowed value. Position-switched spectral data requires a clean OFF position, which is free of line emission over the entire footprint of the SEQUOIA array. We used 5 different OFF positions spread around the Taurus molecular cloud. The right ascension and declinations of the OFF positions used in J2000 co-ordinates are ($04^{h}20^{m}00^{s}$, $23\arcdeg 00\arcmin 00\arcsec$), ($04^{h}42^{m}00^{s}.9$, $23\arcdeg 25\arcmin 41\arcsec$), ($04^{h}54^{m}22^{s}.7$, $24\arcdeg 43\arcmin 28\arcsec$), ($04^{h}01^{m}15^{s}$, $27\arcdeg 32\arcmin 00\arcsec$), and ($04^{h}20^{m}05^{s}.01$, $32\arcdeg 24\arcmin 59\arcsec.2$). For any given submap observed, we typically chose the closest OFF position. The first two positions were known to be free of CO emission from previous studies. The other positions were derived from constructing small maps in 12CO and 13CO at each position (a guess at an initial OFF was derived from the IRAS 100$\mu$m map). These maps were typically 10$\arcmin\times$ 10$\arcmin$ in size, and it was verified that 12CO was not detectable to $\sim 0.25$ K level in the map. In addition to the OTF maps, for each day of observing, pointing and focus measurements were made using a SiO maser source, IKTau ($03^{h}53^{m}28^{s}.81$, $11\arcdeg 24\arcmin 22\arcsec.60$) every few hours. The pointing accuracy of the FCRAO 14 m telescope is typically better than 5″. After each pointing observation, a single 80 second long position-switched observation in 12CO and 13CO was performed at the fiducial center position to keep track of the relative calibration accuracy of the whole survey (see §3.1). For the observing campaign, a dynamic website with a relational database backend was created. From this website, the observing team could monitor the status of observations made to date, plan new observations, generate observing scripts, immediately reduce the OTF map after it was completed using a simple data reduction pipeline, view the integrated intensities of the spectra at the fiducial center position, and produce summary statistics of the map. The website also listed the sensitivities obtained in each submap, and produced prioritized list of submaps left to observe. ### 2.3 Data Processing On-the-Fly mapping generates a set of data that is densely but irregularly sampled on the sky. To construct regularly sampled spectroscopic data cubes and to coadd spatially redundant measurements, the data were convolved into an output grid using a kernel that accounts for the edge taper of the 14m antenna to minimize noise aliasing, and retains the full angular resolution of the telescope. In addition to this spatial weighting, a spectrum was further weighted by the factor $\sigma^{-2}$, where $\sigma$ is the measured rms noise. The spatial pixel separation in the regridded maps is $20\arcsec$. Following this initial processing of submaps into data cubes, a systematic trend was present in the spectral baselines owing to gain drifts in the receiver. Each OTF scan begins with the measurement of OFF position, followed by two raster scanned rows of ON position observations, finally ending in an OFF position observation. The final spectra were computed using both OFF position observations weighting them equally. Spectra taken close in time to either of the OFF positions typically show a concave curvature in the baseline. While spectra taken approximately midway between the two OFFs typically show a convex curvature in baseline. These curvatures were small with respect to the noise of any single spectrum but were evident when averaging spectra with comparable time displacements from the off measurement or when integrating individual spectra over velocity intervals comparable to the half period of the baseline curvature. To rectify these baseline curvatures, we subtracted a baseline polynomial that depended on the spectral and sky coordinates of spectra within a given submap. We found that a second order polynomial along the spectral axis was sufficient and variations along the spatial axes could be adequately described by a first order polynomial. Accordingly, for each submap, we fit the parameterized baseline polynomial, $T_{base}(x,y,v)=\sum_{i=0}^{2}(v-v_{\circ})^{i}f_{i}(x-x_{\circ})g_{i}(y-y_{\circ})$ $None$ where $x_{\circ},y_{\circ},v_{\circ}$ are the central coordinates of the submap, and $f_{i}=a_{i}+b_{i}(x-x_{\circ})$ and $g_{i}=c_{i}+d_{i}(y-y_{\circ})$ parameterize the variation of the baseline curvature over the spatial axes. In the fitting procedure, we exclude the window $-5<V_{LSR}<15$ km s-1 that contains the spectral line emission from the Taurus Molecular Cloud. This procedure has the advantage that only 12 parameters are required to specify the baselines for an entire submap, rather than the three per spectrum that would be needed for a standard baseline subtraction. All of the parameters are recorded so that the unbaselined data could be recovered if necessary. The fitting of a smooth baseline function to an entire submap also avoids the inadvertent removal of a broad, low-level, localized emission line component that could be produced by outflows and other energetic processes. The maximum amplitudes of the baseline curvature were typically greater in the 13CO data, where the most severe cases were still limited to levels less than 25 mK ($\sigma/10$). To demonstrate the need and effect of this baselining procedure, we show in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Clouda 13CO spectra averaged over 10′$\times$10′ boxes from a submap representative of the worse cases of baseline curvature. We emphasize that the curvature is not readily evident in a single spectrum. The thick solid lines show the fitted parameterized baselines averaged over the same boxes. The spatial variation in curvature is most significant along the scanning ($x$) direction, as noted above, with a slow drift in the $y$ direction as the ON-OFF elevation difference varies during the 2-hour submap observation. Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloudb shows the same average spectra after removal of the fitted baselines, with thick lines at zero intensity for reference. The 356 regridded submaps, each of which have a size of $\sim 30\arcmin\times 30\arcmin$, were further consolidated into a set of larger data cubes. There are 88 such consolidated cubes, forming a grid of $11\times 8$ data cubes, dubbed hard–edge cubes (as they do not have overlapping regions between two contiguous cubes). Each ’hard-edge’ cube is assembled from a set of input regridded $30\arcmin\times 30\arcmin$ cubes, after removing the spectral and spatially derived baselines described above from each cube, and subsequently averaging the data together, weighting them by $\sigma^{-2}$, where $\sigma$ is the rms in the derived baseline. In angular offsets, the full extent of the combined hard–edge cubes are (5.75°, $-5.75$°) in RA offsets and ($-2.75$°, 5.75°) in Dec offsets from the fiducial center of the map. Thus for the full $11.5\arcdeg\times 8.5\arcdeg$ region spaced at 20″, there are 3,167,100 spectra in each isotopologue in the combined set of hard–edge cubes. Most of the hard–edge cubes have a spatial size of 1 square degree, except for the cubes that lie on the four edges of the region covered, which measure 1.25 square degrees. The hard–edge cubes at the four corners of the Taurus map have a size of 1.5625 square degrees ($1.25\arcdeg\times 1.25\arcdeg$). ## 3 Results ### 3.1 Calibration Uncertainty in Data In order to track the relative calibration of the survey that was taken over the course of two observing seasons, we made position-switched (PS) measurements on the central (0,0) position every few hours. Since SEQUOIA does not rotate with the sky, only the (0,0) position was uniformly repeated. These (0,0) spectra were baseline subtracted and integrated over the velocity intervals 2-10 km s-1 and 3-9 km s-1 for 12CO and 13CO respectively to produce an integrated line intensity. Both the line profile shape and integrated intensity offer a cross check on the pointing and focus state of the telescope system. The set of these measurements also provide a quantitative measure of the calibration uncertainty of the data. Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud shows the full set of measurements taken over the course of the survey. The mean and standard deviation, weighted by the statistical error for each spectrum, of the integrated intensity for both isotopologues are $12.5\pm 0.93$ K km s-1 and $4.6\pm 0.43$ K km s-1. After subtracting the statistical uncertainties in quadrature, the estimated calibration uncertainties for 12CO and 13CO are $0.87$ and $0.43$ K km s-1 respectively, corresponding to a relative calibration uncertainty of 7% for 12CO and 9.3% for 13CO. Submaps that followed or preceded position switched measurements that significantly deviated from these mean values were reobserved. ## 4 Noise Properties Quantitative measures of cloud structure rely on the sensitivity of the data, its uniformity across the target field, and the underlying noise characteristics of the data. For individual submaps, On-the-Fly Mapping with a focal plane array produces near uniform sensitivity over most of the field. Toward the edges of the submap, the noise necessarily increases as there is less integration time accumulated at these positions. However, our placement of submaps ensured that there are sufficient overlaps of contiguous fields so that these spectra accumulated additional integration time. Indeed, upon the construction of the 1 deg2 cubes, the sensitivity of these “edge” spectra is comparable or better than the spectra from the central parts of any submap. For a given spectrum, the statistical error of the antenna temperature at any channel, $\sigma$, is conventionally estimated from the standard deviation of antenna temperatures within intervals in which no signal is present. This measure includes noise contributions from the instrument and atmosphere. For the Taurus Survey, we have calculated $\sigma$ for all the co-added convolved and resampled 12CO and 13CO spectra excluding values within the velocity range of the Taurus cloud (-5 $<V_{LSR}<+15$ km s-1). The cumulative distributions of $\sigma(^{12}CO)$ and $\sigma(^{13}CO$) are shown in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud. The steepness of these distributions provides an approximate measure of noise uniformity. The first, second, and third quartile values are 0.53 K, 0.58 K, and 0.63 K for 12CO and 0.23 K, 0.26 K, and 0.28 K for 13CO. The antenna temperature distribution of all voxels within the data cube offers another measure of the noise properties. In Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud, this distribution is shown for the composite 12CO and 13CO data cubes of the survey. For both data cubes, statistical noise contributes to the peak component centered on T${}_{A}^{*}$=0.0 K. Signal from the Taurus cloud is responsible for the excess positive emission. In the ideal case in which all spectra have gaussian distributed fluctuations with constant rms value, $\sigma$, the dispersion of the voxel distribution about T${}_{A}^{*}$=0 K would be equivalent to $\sigma$. More realistically, the noise is not uniform and not necessarily gaussian. For example, the use of a common reference position for many source spectra introduces correlated noise. With OTF mapping using a focal plane array, where a given position on the sky is sampled by many pixels in the array, this effect is present but not as severe as discrete mapping methods that share reference positions (Heyer et al. , 1998; Jackson et al. , 2006). As shown in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud, the noise of this survey is not uniform. To assess departures from the gaussian character of the distribution of noise in the 12CO and 13CO data, we have generated equivalently sized data cubes filled with gaussian noise values that follow the same distribution of $\sigma$ values shown in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud. The respective distributions of these purely gaussian noise spectra, shown as the dotted lines in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud, provide an excellent match to the zero-centered noise components. We conclude that the noise measured in the 12CO and 13CO data cubes can not be readily distinguished from pure gaussian noise. It is dominated by statistical fluctuations from the receiver and the sky rather than by systematic contributions. ## 5 Images Two dimensional images are generated from the spectroscopic data cubes by reducing the information of each spectrum to a single scalar value. These reductions include maps of maximum value, the integration of spectra over a set of velocity intervals (zero moment), and higher moments such as the centroid velocity or line width. To convey a fraction of the information resident within these data cubes, we show some of these reductions in images of the Taurus Cloud. Zero moment images derived from the 12CO and 13CO data cubes over the VLSR range $0$ to $12$ km s-1 in 1 km s-1 intervals are presented in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud. The two images show quite distinct distributions between the two isotopologues. The 12CO J=1-0 integrated emission is mostly diffuse, even towards the three known primary sub clouds of Taurus (Heiles’ Cloud 2, Barnard 18, L1495). In the most faint regions, there are weak “streaks” or striations of 12CO emission that are typically aligned with the local magnetic field direction (Goldsmith et al. , 2008; Heyer et al. , 2008). The 13CO J=1-0 emission is mostly distributed within high contrast filaments. However, even within the diffuse regions, the weak 13CO emission exhibits a striated pattern. These differences between the 12CO and 13CO distributions can be readily attributed to the higher opacity of the 12CO emission that likely limits its probed volume to the low column density envelope of the cloud. Images of peak 12CO and 13CO intensity are shown in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud. For the optically thick 12CO line, the peak intensity is a valuable measure of the excitation temperature. In Figures The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud-The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud, we show the distribution of 12CO and 13CO J=1-0 emission averaged over 1 km s-1 velocity intervals centered at VLSR 0.5, 1.5, 2.5 … 12.5 km s-1. As noted by many previous studies (Brunt & Mac Low, 2004; Lazarian & Pogosyan, 2000), more of the emission is found at the higher spatial frequencies when integrating over small velocity intervals. The structure that is measured arises from variations in the velocity field rather than those of density or column density (Brunt & Mac Low, 2004). These narrow velocity integrated images reveal stunning textural patterns and individual features that occur over a broad range of scales. There are regions of faint, low surface brightness emission, most notably evident in the 12CO data. Often, this emission component exhibits low amplitude striations that are similar to the wind swept structures observed within terrestrial cirrus clouds. Owing to lower optical depth, the 13CO emission shows more high contrast emission originating from higher column density regions located deeper within the cloud. The feature located within the southwest corner with VLSR $\sim$ 10 km s-1 in Figure 16 may not be associated with the Taurus cloud. ## 6 Box Averaged Spectra To convey the coarse velocity field of the cloud and to further emphasize the quality of the data, we have constructed average 12CO and 13CO J=1-0 spectra from each 1, 1.25, or 1.56 deg2 cube. These are shown in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud overlayed upon images of integrated intensity. Each spectrum is an average of 32,400, 40,500, or 50,625 individual spectra with typical rms values of 0.014 K for 12CO and 0.006 K for 13CO (T${}_{A}^{*}$). These rms values are greater than one would expect averaging this number of spectra each with the respective median values shown in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud. This discrepancy arises from spatially correlated noise imposed on spectra sharing a common reference measurement in the OTF data collection scheme. The average spectra demonstrate excellent baseline fidelity with no evidence for any systematic noise contribution. The average spectra do reveal several significant properties of the Taurus velocity field. The previously established large scale velocity gradient across the Taurus Molecular Cloud is apparent in the systemic shift of the spectra from blueshifted velocities on the eastern side to redshifted velocities on the western portion of the cloud. In addition, many of the average spectra exhibit asymmetries and multiple velocity components that attest to the complex structure along the line of sight. ## 7 Emission Statistics The 12CO and 13CO data cubes of Taurus provide panoramic views of the structure of a 104 M⊙ molecular cloud. The resident information is sufficiently vast to require statistical descriptions of the data that may offer insight to the prevailing conditions or state of the cloud. The cumulative probability density function is defined to be $P(f_{\circ})={{\int\int dxdyF(x,y,f_{\circ})}\over{\int\int dxdyF(x,y,f_{max})}}$ $None$ where $F(x,y,f_{\circ})=\left\\{\begin{array}[]{ll}0&\mbox{for $f(x,y)>f_{\circ}$}\\\ 1&\mbox{for $f(x,y)\leq f_{\circ}$},\\\ \end{array}\right.$ with $f(x,y)$ the 2 dimensional distribution of some measured parameter, $f_{\circ}$ a moving threshold of that parameter, and $f_{max}$ the maximum value of the measured parameter in the whole map. The weighted cumulative PDF biases each bin of $f(x,y)$ by the total signal, $\int T(x,y,v)dv$, of all contributing pixels, $P_{w}(f_{\circ})={{\int\int dxdyF(x,y,f_{\circ})\int T(x,y,v)dv}\over{\int\int dxdy\int T(x,y,v)dv}}$ $None$ Figures The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud and The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud show the cumulative PDF and weighted cumulative PDF for integrated intensity (denoted as W(12CO) and W(13CO) respectively), and peak temperature distributions (denoted as Tmax(12CO) and Tmax(13CO) respectively). The weighting function used is the corresponding integrated intensity. The unweighted cumulative PDF shows the cumulative fraction of projected area as a function of integrated intensity (Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud) and peak temperature (Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud). The weighted cumulative PDF shows the cumulative fraction of integrated intensity as a function of integrated intensity (Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud) and peak temperature (Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud). The median values for the weighted PDFs of integrated intensity are 6.9 and 1.9 K km s-1 for 12CO and 13CO respectively. Similarly, for peak temperature, the weighted median values are 4.1 K and 1.9 K. These figures demonstrate that much of the 12CO and 13CO signal originates from lines of sight within the low surface brightness regime of the Taurus cloud. For example, the weighted cumulative PDFs demonstrate that half of the 12CO flux of a cloud is emitted within the low surface brightness portion of the Taurus molecular cloud having 12CO integrated intensity less than 6.9 K km s-1 and peak temperature $\leq 4.1$ K. In terms of projected area, only $\sim 20$% of the cloud’s area has integrated intensity $\geq 6.9$ K km s-1 or peak temperature $\geq 4.1$ K in 12CO. ## 8 Conclusions The FCRAO Survey of the Taurus Molecular Cloud is a powerful set of data to investigate interstellar gas dynamics and the star formation process with high spatial resolution and spatial dynamic range. It offers a valuable complement to observations at other wavelengths that probe the dust component and the population of young stellar objects. In this paper, we summarized the instrumentation, data collection and processing procedures used in the survey. We also characterized the noise and signal distributions of the survey. The overall structure of the cloud, its column density distribution and mass, and relationship with the magnetic field are discussed in Goldsmith et al. (2008). This work was supported by NSF grant AST 05-40852 to the Five College Radio Astronomy Observatory, NSF grant AST-0407019 to Cornell University, and by the Jet Propulsion Laboratory, California Institute of Technology. We thank Yvonne Tang and Marko Krco for for assistance with observations. ## References * Abergel et al. (1995) Abergel, A., Boulanger, F., Fukui, Y., & Mizuno, A. 1995, A&A, 111, 483 * Brunt & Mac Low (2004) Brunt, C.M. & Mac Low, M. 2004, ApJ, 604, 196 * Burton & Hartmann (1994) Burton, W.B. & Hartmann, D. 1994, in Unveiling Large–Scale Structures Behind the Milky Way, ASP Conf. Serives, Vol. 67, C. Balkowski & R.C. Kraan-Kortweg eds. (San Francisco: Astronomical Society of the Pacific), 31 * Erickson et al. (1999) Erickson, N.R.. Grosslein, R.M., Erickson, R.B., & Weinreb, S. 1999, IEEE Trans. Microwave Theory Tech., 47(12), 2212 * Goldsmith & Li (2005) Goldsmith, P.F. & Li, D. 2005, ApJ, 622, 938 * Goldsmith et al. (2008) Goldsmith, P. F., Heyer, M. H., Narayanan, G., Snell, R. L., Li, D., & Brunt, C., 2008, to be submitted to ApJ * Heyer et al. (1987) Heyer, M.H., Vrba, F., Snell, R.L., Schloerb, F.P., Strom, S.E., Goldsmith, P.F., & Strom, K.M. 1987, ApJ, 321, 855 * Heyer et al. (1998) Heyer, M.H., Brunt, C., Snell, R.L., Howe, J.E., Schloerb, F.P., Carpenter, J.M. 1998, ApJS, 115, 241 * Heyer (2000) Heyer, M. H. 2000, Imaging at Radio through Submillimeter Wavelengths, 217, 213 * Heyer et al. (2008) Heyer, M.H., Gong, H., Ostriker, E., & Brunt 2008, submitted to ApJ * Jackson et al. (2006) Jackson, J. M., Rathborne, J. M., Shah, R. Y., Simon, R., Bania, T. M., Clemens, D. P., Chambers, E. T., Johnson, A. M., Dormody, M., Lavoie, R., Heyer, M. H. 2006, ApJS, 163, 145 * Jijina, Myers, & Adams (1999) Jijina, J., Myers, P.C., & Adams, F.C. 1999, ApJS, 125, 161 * Langer et al. (1995) Langer, W.D., Velusamy, T., Kuiper, T.B.H., Levin, S., Olsen, E., & Migenes, V. 1995, ApJ, 453, 293 * Lazarian & Pogosyan (2000) Lazarian, A., & Pogosyan, D. 2000, ApJ, 537, 720 * Li & Goldsmith (2003) Li, D. & Goldsmith, P.F. 2003, ApJ, 585, 823 * Mizuno et al. (1995) Mizuno, A., Onishi, T., Yonekura, Y., Nagahama, T., Ogawa, H., & Fukui, Y. 1995, ApJ, 445, L161 * Onishi et al. (1996) Onishi, T., Mizuno, A., Kawamura, A., Ogawa, H., Fukui, Y. 1996, ApJ, 465, 815 * Onishi et al. (1998) Onishi, T., Mizuno, A., Kawamura, A., Ogawa, H., & Fukui, Y. 1998, ApJ, 502, 296 * Onishi et al. (2002) Onishi, T., Mizuno, A., Kawamura, A., Tachihara, K., & Fukui, Y. 2002, ApJ, 575, 950 * Padoan et al. (2002) Padoan, P., Cambrésy, L., & Langer, W.D. 1992, ApJ, 580, L57 * Schloerb & Snell (1984) Schloerb, F.P. & Snell, R.L. 1984, ApJ, 283, 129 * Shuter et al. (1987) Shuter, W.L.H., Dickman, R.L., & Klatt, C. 1987, ApJ, 322, L103 * Tatematsu et al. (2004) Tatematsu, K., Umemoto, T., Kandori, R., & Sekimoto, Y. 2004, ApJ, 606, 333 * Ungerechts & Thaddeus (1987) Ungerechts, H. & Thaddeus, P. 1987, ApJ, 322, 706 (a) 13CO spectra in a $30\arcmin\times 30\arcmin$ region averaged over $10\arcmin\times 10\arcmin$ and fitted parameterized baselines (thick lines). The rightmost column of spectra corresponds to observations taken close in time to an OFF measurement, while the leftmost column corresponds to spectra taken farthest in time from an OFF observation. The middle column represents a time in between these extremes. See the text for details. (b) The same box–averaged spectra after baseline removal, with thick lines at zero intensity for reference. Integrated intensity in the position-switched spectra for the (0,0) position for both polarizations and both isotopologues over the entire observing period. The 12CO data have integrated intensity $\simeq$ 12.5 K km s-1and the 13CO data have integrated intensity $\simeq$ 4.6 K km s-1. The error bars denote $1\sigma$ statistical errors in these values. The 12CO data are denoted by crosses and plus symbols and the 13CO data by squares and diamonds, for the two polarizations. The horizontal solid lines show the mean integrated intensities derived for each data set and the dashed lines indicate $\pm$ standard deviation about these mean values. The gap in coverage in the middle of year 2004 is due to the normal summer shutdown period of the FCRAO 14 m telescope. Cumulative distributions of rms values, $\sigma(^{12}$CO) and $\sigma(^{13}$CO) derived from signal–free channels within the Taurus 12CO and 13CO data cubes. The distribution of voxel values within the composite 12CO and 13CO data cubes (solid lines). The dotted lines show the distribution of values obtained from data cubes of equal size containing only gaussian noise with values of $\sigma$ that follow the same distribution shown in Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud. These provide an excellent fit to the data and demonstrate the near gaussian character of the noise of individual spectra. Integrated Intensity Images for 12CO (top) and 13CO (bottom). The images are obtained over $-5$ to $20$ km s-1 and $3$ to $9$ km s-1 for 12CO and 13CO respectively. The colorbar on the right shows the integrated intensity scale in K.km s-1. In the 13CO figure, we also overlay outlines of a few well-known regions in Taurus as designated by Onishi et al. (1996). Images of peak intensity, Tmax, for 12CO (top) and 13CO (bottom). The peak temperature in the velocity range of 0 to 12 km s-1 is used in both cases. The colorbar on the right shows the Tmax scale in antenna temperature units (K). Images of 12CO (top) and 13CO (bottom) emission integrated between VLSR 0 to 1 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 1 to 2 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 2 to 3 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 3 to 4 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 4 to 5 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 5 to 6 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 6 to 7 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 7 to 8 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 8 to 9 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 9 to 10 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 10 to 11 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 11 to 12 km s-1. Same as Figure The Five College Radio Astronomy Observatory CO Mapping Survey of the Taurus Molecular Cloud for VLSR 12 to 13 km s-1. Mosaic of box-averaged spectra over the 88 hard-edged sub-cubes for 12CO (top) and 13CO (bottom) overlayed on images of the integrated intensity of each isotopologue. The velocity scale on the x-axis is $0$ to $15$ km s-1, and the temperature scale ranges from -0.1 to 4.5 K for 12CO and -0.005 to 2.0 K for 13CO. The vertical dotted line in each spectrum denotes VLSR=7.0 km s-1. Weighted (solid line) and unweighted (dotted line) cumulative probability density functions as a function of 12CO and 13CO integrated intensity (denoted as W(12CO) and W(13CO) respectively). The unweighted and weighted cumulative PDFs describe the fractional contribution by projected area and integrated intensity emission respectively. Weighted (solid line) and unweighted (dotted line) cumulative probability density functions as a function of 12CO and 13CO peak intensity.

arxiv-papers

2024-09-04T02:48:53.847468

0802.2564

with # Spectral Measure of Robustness in Complex Networks Jun Wu wujunpla@hotmail.com College of Information Systems and Management, National University of Defense Technology, Changsha 410073, P. R. China. Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100080, P. R. China Institute for Mathematical Sciences, Imperial College London, London SW7 2PG, United Kingdom Yue-Jin Tan Hong-Zhong Deng Yong Li Bin Liu Xin Lv College of Information Systems and Management, National University of Defense Technology, Changsha 410073, P. R. China. ###### Abstract We introduce the concept of natural connectivity as a robustness measure of complex networks. The natural connectivity has a clear physical meaning and a simple mathematical formulation. It characterizes the redundancy of alternative paths by quantifying the weighted number of closed walks of all lengths. We show that the natural connectivity can be derived mathematically from the graph spectrum as an average eigenvalue and that it increases strictly monotonically with the addition of edges. We test the natural connectivity and compare it with other robustness measures within a scenario of edge elimination. We demonstrate that the natural connectivity has an acute discrimination which agrees with our intuition. ###### pacs: 89.75.Hc, 89.75.Fb, 02.10.Ox We are surrounded by networks. Networks with complex topology describe a wide range of systems in nature and society. The study of complex networks has become an important area of multidisciplinary research involving physics, mathematics, biology, social sciences, informatics, and other theoretical and applied sciences Newman (2003); Boccaletti et al. (2006). Complex networks rely for their function and performance on their robustness, i.e., the ability to endure threats and survive accidental events. For example, modern society is dependent on its critical infrastructure networks: communication, electrical power, rail, and fuel distribution networks. Failure of any of these critical infrastructure networks can bring the ordinary activities of work and recreation to a standstill. Other examples of robustness arise in biological and social systems, including questions such as the stability of social organizations in the face of famine, war, or even changes in social policy. Because of its broad application, robustness has become a central topic in all complex networks and receives growing attention. As a basic concept of graph theory, the connectivity of a graph is an important and probably the earliest measure of robustness of a network Frank and Frisch (1970). Vertex (edge) connectivity, defined as the size of the smallest vertex (edge) cut, determines in a certain sense the robustness of a graph to the deletion of vertices (edges). However, the vertex or edge connectivity only partly reflects the ability of graphs to retain certain degrees of connectedness after deletion. Other improved measures were introduced and studied, including super connectivity Bauer et al. (1981), conditional connectivity Harary (1983), restricted connectivity Esfahanian and Hakimi (1988), fault diameter Krishnamoorthy and Krishnamirthy (1987), toughness Chvátal (1973), scattering number Jung (1978), tenacity Cozzen et al. (1995), expansion parameter Alon (1986) and isoperimetric number Mohar (1989). In contrast to vertex (edge) connectivity, these new measures consider both the cost to damage a network and how badly the network is damaged. However, from an algorithmic point of view, it is unfortunate that the problem of calculating these measures for general graphs is NP-complete. This implies that these measures are of no great use within the context of complex networks. Another remarkable measure to unfold the robustness of a network is the second smallest eigenvalue of the Laplacian matrix, also known as the algebraic connectivity. Fiedler Fiedler (1973) showed that the magnitude of the algebraic connectivity reflects how well connected the overall graph is, i.e., the larger the algebraic connectivity is, the more difficult it is to cut a graph into independent components. Hence, there is a vast literature on the algebraic connectivity (see Merris (1994) for a survey). However, the algebraic connectivity is equal to zero for all disconnected networks. Therefore, it is too coarse a measure for complex networks. The prime study regarding network robustness within the context of complex networks came from random graph theory Bollobás (1985) and was stimulated by the work of Albert et al. Albert et al. (2000). Instead of a strict extremal property, they proposed a statistical measure, i.e., the critical removal fraction of vertices (edges) for the disintegration of a network, to characterize the robustness of complex networks. The disintegration of networks can be observed from the decrease of network performance. The most common performance measurements include the diameter, the size of largest component, the average path length, the efficiency Latora and Marchiori (2001); Crucitti et al. (2003) and the number of reachable vertex pairs Palmer et al. (2001); Siganos et al. (2006). As the fraction of removed vertices or edges increases, the network will eventually collapse at a critical fraction. It is suggested that scale-free networks display an exceptional robustness against random failure, but show poor performance against intentional attack Albert et al. (2000). As an expansion of the work by Albert et al., Wu et al. Wu et al. (2007a, b) studied the robustness of complex networks under incomplete information, i.e., one can only obtain the information of partial vertices. Cohen et al. Cohen et al. (2000, 2001) developed the first analytical approach to calculating the critical removal fraction of a network under random failure or intentional attack. Callaway et al. Callaway et al. (2000) put forward an alternative and more general approach using a generalization of the generating function formalism. If we consider a source vertex and a termination vertex, there may be several alternative paths between them. When one path fails, the two vertices can still communicate through other alternative paths. It is intuitive that the more alternative paths, the more robust the connectedness between the two vertices. This observation leads us to consider the redundancy of alternative paths as the root of the robustness of networks, which ensures that the connection between vertices still remains possible in spite of damage to the network. Although it would be ideal to define this redundancy as the number of alternative paths of different lengths for all pairs of vertices, this measure is very difficult to calculate. Note, however, that the number of closed walks in a network is a good index for the number of alternative paths. In this paper, we propose a new robustness measure of complex networks based on the number of closed walks. A complex network can be viewed as a simple undirected graph $G(V,E)$, where $V$ is the set of vertices, and $E\subseteq V\times V$ is the set of edges. Let $N=\left|V\right|$ and $M=\left|E\right|$ be the number of vertices and the number of edges, respectively. Let $d_{i}$ be the degree of vertex $v_{i}$. Let $d_{\min}$ be the minimum degree and $d_{\max}$ be the maximum degree of $G$. Let $A(G)=(a_{ij})_{N\times N}$ be the adjacency matrix of $G$, where $a_{ij}=a_{ji}=1$ if vertex $v_{i}$ and $v_{j}$ are adjacent, and $a_{ij}=a_{ji}=0$ otherwise. A walk of length $k$ in a graph $G$ is an alternating sequence of vertices and edges $v_{0}e_{1}v_{1}e_{2}...e_{k}v_{k}$, where $v_{i}\in V$ and $e_{i}=(v_{i-1},v_{i})\in E$. A walk is closed if $v_{0}=v_{k}$. Closed walks are directly related to the subgraphs of the graph. For instance, a closed walk of length $k=2$ corresponds to an edge and a closed walk of length $k=3$ represents a triangle. Note that a closed walk can be trivial, i.e., containing repeated vertices, leading to the length of a closed walk being infinite. The number of closed walks is an important index for complex networks. For example, Estrada et al. have measured vertex centrality Estrada and Rodríguez-Velázquez (2005) and network bipartivity Estrada and Rodríguez- Velázquez (2005) based on the number of closed walks. Here we define the redundancy of alternative paths as the number of closed walks of all lengths. Considering that shorter closed walks have more influence on the redundancy of alternative paths than longer closed walks and to avoid the number of closed walks of all lengths to diverge, we scale the contribution of closed walks to the redundancy of alternative paths by dividing them by the factorial of the length $k$. That is, we define a weighted sum of numbers of closed walks $S=\sum\nolimits_{k=0}^{\infty}{\left({n_{k}/k!}\right)}$, where $n_{k}$ is the number of closed walks of length $k$. Using matrix theory, we know that $n_{k}=\sum\limits_{i_{1},i_{2},...i_{k}}{a_{i_{1}i_{2}}}a_{i_{2}i_{3}}...a_{i_{k}i_{1}}=trace(A^{k})=\sum\limits_{i=1}^{N}{\lambda_{i}^{k}},$ (1) where $\lambda_{i}$ is the $i$th largest eigenvalue of $A(G)$. Specifically, $n_{2}=\sum\nolimits_{i}{d_{i}}=2M$. Using Eq. (1), we obtain $S=\sum\limits_{k=0}^{\infty}{\frac{{n_{k}}}{{k!}}}=\sum\limits_{k=0}^{\infty}{\sum\limits_{i=1}^{N}{\frac{{\lambda_{i}^{k}}}{{k!}}}}=\sum\limits_{i=1}^{N}{\sum\limits_{k=0}^{\infty}{\frac{{\lambda_{i}^{k}}}{{k!}}}}=\sum\limits_{i=1}^{N}{e^{\lambda_{i}}}.$ (2) Eq. (2) shows that the weighted sum of closed walks of all lengths can be derived from the graph spectrum. Noting that $S$ will be a large number for large $N$, we scale $S$ and denote it by $\bar{\lambda}$ $\bar{\lambda}=\ln\left({\frac{{S}}{N}}\right)=\ln\left({\frac{{\sum\limits_{i=1}^{N}{e^{\lambda_{i}}}}}{N}}\right),$ (3) which corresponds to an ’average eigenvalue’. We propose to call it _natural connectivity_ or _natural eigenvalue_. A desired property of natural connectivity is that it changes monotonically when edges are added or deleted. Let $G+e$ be the graph obtained by adding an edge $e$ to $G$ and let $\hat{n}_{k}=\hat{n}^{\prime}_{k}+\hat{n}^{\prime\prime}_{k}$ be the number of closed walks of length $k$ in $G+e$, where $\hat{n}^{\prime}_{k}$ is the number of closed walks of length $k$ containing $e$ and $\hat{n}^{\prime\prime}_{k}$ is the number of closed walks of length $k$ containing no $e$. Note that $\hat{n}^{\prime}_{k}=n_{k}$ and $\hat{n}^{\prime\prime}_{k}\geq 0$, thus $\hat{n}_{k}\geq n_{k}$. It is easy to show that $\hat{n}_{k}>n_{k}$ for some $k$, e.g., $\hat{n}_{2}=n_{2}+2$. Consequently, $\bar{\lambda}(G+e)>\bar{\lambda}(G)$, indicating that the natural connectivity increases strictly monotonically as edges are added. In Fig. 1, we illustrate two simple graphs with six vertices, where graph (b) is obtained by adding an edge to graph (a). Our intuition suggests that graph (b) should be more robust than graph (a), which agrees with our measure. The natural connectivity of graph (a) and (b) are 1.0878 and 1.3508, respectively. However, some robustness measures mentioned above can not distinguish the two graphs. For example, both graphs have identical edge connectivity 2 and identical algebraic connectivity 0.7369. Figure 1: Graph (b) is obtained from graph (a) by adding an edge. Both graphs have identical edge connectivity and identical algebraic connectivity, but are distinguished by our proposed natural connectivity. It is evident from Eq. (3) that $\lambda_{1}\geq\bar{\lambda}\geq\lambda_{N}$. Moreover, for a given number of vertices $N$ and following the discussion on monotonicity above, the empty graph consisting of isolated vertices has the minimum natural connectivity and the complete graph, whose vertices are pairwise adjacent, has the maximum natural connectivity. It is known that $\lambda_{1}=\lambda_{2}=...=\lambda_{N}=0$ for the empty graph, and $\lambda_{1}=N-1,\lambda_{2}=\lambda_{3}=...=\lambda_{N}=-1$ for the complete graph Cvetković et al. (1979). Hence we obtain the following bound for the natural connectivity $0\leq\bar{\lambda}\leq\ln((N-1)e^{-1}+e^{N-1})-\ln N\approx N-\ln N.$ (4) To explore in depth the natural connectivity measure and compare it with other robustness measures, we consider a scenario of edge elimination. As edges are deleted, we expect the decrease of the robustness measure, and we also expect different behavior for different edge elimination strategies. We generate initial networks with a power-law degree distribution using the BA model Barabási and Albert (1999). We remark that the type of network has no effect on the analysis and conclusions. We consider four edge elimination strategies: (i) deleting the edges randomly (random strategy); (ii) deleting the edges connecting high-degree vertices and high-degree vertices in the descending order of $d_{i}\cdot d_{j}$, where $d_{i}$ and $d_{j}$ are the degrees of the end vertices of an edge (rich-rich strategy); (iii) deleting the edges connecting low-degree vertices and low-degree vertices in the ascending order of $d_{i}\cdot d_{j}$ (poor-poor strategy); (iv) deleting the edges connecting high-degree vertices and low-degree vertices in the descending order of $\left|{d_{i}-d_{j}}\right|$ (rich-poor strategy). Along with the natural connectivity, we investigate three other robustness measures: edge connectivity $\kappa_{\rm{E}}(G)$, algebraic connectivity $a(G)$ and critical removal fraction of vertices under random failure $f_{c}^{R}$. To find the critical removal fraction of vertices, we choose $\kappa\equiv<k^{2}>/<k>\leq 2$ as the criterion for the disintegration of networks Cohen et al. (2000). The results are shown in Fig. 2. Each measure is an average over 100 realizations of a BA network. Figure 2: The robustness measured by edge connectivity (a), algebraic connectivity (b), critical removal fraction of vertices (c) and natural connectivity (d) as a function of number of deleted edges for four edge elimination strategies: random strategy (circles), rich-rich strategy (squares), poor-poor strategy (diamonds) and rich-poor strategy (triangles). The initial network is generated using the BA model, where $N=1000$ and $<k>\approx 6$. Each quantity is an average over 100 realizations. From Fig. 2(a) and Fig. 2(b), we find similar behaviour between $\kappa_{\rm{E}}(G)$ and $a(G)$. The first observation is that deleting a small quantity of edges connecting high-degree vertices and high-degree vertices has no obvious effect on the robustness measured by the edge or algebraic connectivity. On the other hands, the robustness drops rapidly under the poor-poor strategy, i.e., when small quantity of edges connecting low- degree vertices and low-degree vertices are deleted. It is generally believed that the edges between high-degree vertices are important, and the edges between low-degree vertices are inessential for the global network robustness. For example, in the Internet, the failure of the links between core routers will bring a disaster, but there is no effect on the network robustness if we disconnect two terminal computers. Clearly, robustness measures based on edge or algebraic connectivity do not agree with our intuition. These unexpected features can be explained by the bound $a(G)\leq\kappa(G)\leq\kappa_{\rm{E}}(G)\leq d_{\min}$, also known as Fiedler’s inequality Fiedler (1973), where $\kappa(G)$ is the vertex connectivity. In fact, we find that the probability of $\kappa{}_{\rm{E}}(G)=d_{\min}$ almost approaches to 1. After a few edges connecting low-degree vertices and low-degree vertices are deleted, $d_{\min}$ decreases to zero rapidly, but $d_{\min}$ is preserved under rich-rich strategy. Moreover, we find that, for all four strategies, the edge or algebraic connectivity is equal to zero after particular edges are deleted, even in the case where only very few vertices are separated from the largest cluster. This means that both the edge connectivity and the algebraic connectivity lose discrimination when the network is disconnected. Figure 2(c) shows the critical removal fraction of vertices $f_{c}^{R}$ as a function of the number of deleted edges. Contrary to the result of edge or algebraic connectivity and in agreement with our intuition, we observe that the rich-rich strategy is the most effective edge elimination strategy and the poor-poor strategy is the worst one. Nevertheless, we find that there are irregular fluctuations in the curves. This shows that the critical removal fraction is not a sensitive measure of robustness, especially for small sized networks. In Fig. 2(d), we display the results of the natural connectivity according to Eq. (3). We find a clear variation of the measure with distinct differences between the four edge elimination strategies, showing a clear ranking for the four edge elimination strategies: rich-rich strategy $\succ$ rich-poor strategy $\succ$ random strategy $\succ$ poor-poor strategy, which agrees with our intuition. For the random strategy, we observe a linear decrease of the natural connectivity. For the rich-rich strategy or rich-poor strategy, the natural connectivity decreases rapidly with the edge elimination. For poor- poor strategy, deleting a small quantity of edges connecting low-degree vertices and low-degree has weak effect on the robustness. Moreover, we find that the curves for natural connectivity are surprisingly smooth, which indicates that the natural connectivity can measure the robustness of complex networks stably even for very small sized networks. In fact, we have found that the curves for natural connectivity are also smooth without averaging over 100 realizations, viz. for one individual network. However, in the case of individual networks, we find stepped curves for the edge or algebraic connectivity and large fluctuations for the critical removal fraction. In summary, we have proposed the concept of natural connectivity as a spectral measure of robustness in complex networks. The natural connectivity is rooted in the inherent structural properties of a network. The theoretical motivation of our measure arises from the fact that the robustness of a network comes from the redundancy of alternative paths. The natural connectivity is expressed in mathematical form as an average eigenvalue and allows a precise quantitative analysis of the network robustness. Our measure works both in connected and disconnected networks. We have shown that it changes strictly monotonically with the addition or deletion of edges. To test our natural connectivity measure and compare it with other measures, we have designed a scenario of edge elimination, in which four different edge elimination strategies are considered. We have demonstrated that the natural connectivity has an acute discrimination in measuring the robustness of complex networks and can detect small variations of robustness stably. Rich information about the topology and dynamical processes can be extracted from the spectral analysis of the networks. The natural connectivity sets up a bridge between graph spectra and the robustness of complex networks. It is of great theoretical and practical significance in network design and optimization to link the robustness to other network structural or dynamical properties (e.g., efficiency, synchronization, diffusion, searchability). ###### Acknowledgements. We would like to thank Mauricio Barahona and Quoqing Zhang for useful discussions and comments. This work is in part supported by the National Science Foundation of China under Grant No. 70501032, No. 70771111 and No. 60673168. This work is also partly supported by the Hi-Tech Research and Development Program of China under Grant No. 2006AA01Z207. ## References * Newman (2003) M. E. J. Newman, SIAM Review 45, 167 (2003). * Boccaletti et al. (2006) S. Boccaletti, V. Latora, Y. Moreno, M. Chavez, and D. U. Hwanga, Physics Reports 424, 175 (2006). * Frank and Frisch (1970) H. Frank and I. T. Frisch, IEEE Trans. Commun. Technol. COM-18, 567 (1970). * Bauer et al. (1981) D. Bauer, F. Boesch, C. Suffel, and R. Tindell, in _The Theory and Application of Graphs_ (Wiley, New York, 1981), pp. 89–98. * Harary (1983) F. Harary, Networks 13, 346 (1983). * Esfahanian and Hakimi (1988) A. H. Esfahanian and S. L. Hakimi, Information processing Letters 27, 195 (1988). * Krishnamoorthy and Krishnamirthy (1987) M. S. Krishnamoorthy and B. Krishnamirthy, Computers and Mathematics with Applications 13, 577 (1987). * Chvátal (1973) V. Chvátal, Discr. Math. 5, 215 (1973). * Jung (1978) H. A. Jung, J. Combin. Theory B 24, 125 (1978). * Cozzen et al. (1995) M. Cozzen, D. Moazzami, and S. Stueckle, in _Seventh International Conference on the Theory and Applications of Graphs_ (Wiley, New York, 1995), pp. 1111–1122. * Alon (1986) N. Alon, Combinatorica 6, 83 (1986). * Mohar (1989) B. Mohar, J. Comb. Theory Ser. B 47, 274 (1989). * Fiedler (1973) M. Fiedler, Czech. Math. J. 23, 298 (1973). * Merris (1994) R. Merris, Linear A lgebra and Its App lications 197, 143 (1994). * Bollobás (1985) B. Bollobás, _Random Graphs_ (Academic Press, New York, 1985). * Albert et al. (2000) R. Albert, H. Jeong, and A. L. Barabási, Nature 406, 378 (2000). * Latora and Marchiori (2001) V. Latora and M. Marchiori, Phys. Rev. Lett. 87, 198701 (2001). * Crucitti et al. (2003) P. Crucitti, V. Latora, M. Marchiori, and A. Rapisarda, Physica A. 320, 622 (2003). * Palmer et al. (2001) C. Palmer, G. Siganos, M. Faloutsos, C. Faloutsos, and P. Gibbons, in _Workshop on Network-Related Data Management_ (Santa Barbara, 2001). * Siganos et al. (2006) G. Siganos, S. L. Tauro, and M. Faloutsos, Journal of Communications and Networks 8, 339 (2006). * Wu et al. (2007a) J. Wu, H.-Z. Deng, Y.-J. Tan, and Y. Li, Chinese Physics Letter 24, 2138 (2007a). * Wu et al. (2007b) J. Wu, Y.-J. Tan, H.-z. Deng, and D. Zhu, Journal of Physics A 40, 2665 (2007b). * Cohen et al. (2000) R. Cohen, K. Erez, D. ben-Avraham, and S. Havlin, Phys. Rev. Lett. 85, 4626 (2000). * Cohen et al. (2001) R. Cohen, K. Erez, D. ben-Avraham, and S. Havlin, Phys. Rev. Lett. 86, 3682 (2001). * Callaway et al. (2000) D. S. Callaway, M. E. J. Newman, S. H. Strogatz, and D. J. Watts, Phys. Rev. Lett. 85, 5468 (2000). * Estrada and Rodríguez-Velázquez (2005) E. Estrada and J. A. Rodríguez-Velázquez, Phys. Rev. E 71, 056103 (2005). * Estrada and Rodríguez-Velázquez (2005) E. Estrada and J. A. Rodríguez-Velázquez, Phys. Rev. E 72, 046105 (2005). * Cvetković et al. (1979) D. M. Cvetković, M. Doob, and H. Sachs, _Spectra of Graphs_ (Academic Press, New York, 1979). * Barabási and Albert (1999) A. L. Barabási and R. Albert, Science 286, 509 (1999).

arxiv-papers

2024-09-04T02:48:53.852564

0802.2579

# Centrality Dependence of $\mathrm{{p}_{T}}$ Spectra for Identified Hadrons in Au+Au and Cu+Cu Collisions at $\sqrt{s_{NN}}=200$ GeV††thanks: Supported by the National Natural Science Foundation of China (10775089, 10475049) and the science fund of Qufu Normal University. Wang Yun-Fei1 Shao Feng-Lan1;1) Song Jun1 Wei De-Ming1 Xie Qu-Bing2 shaofl@mail.sdu.edu.cn 1 (Department of Physics, Qufu Normal University, Qufu 273165, China) 2 (Department of Physics, Shandong University, Jinan 250100, China) ###### Abstract The centrality dependence of transverse momentum spectra for identified hadrons at midrapidity in Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV is systematically studied in a quark combination model. The $\mathrm{{p}_{T}}$ spectra of $\pi^{\pm}$, $K^{\pm}$, $p(\bar{p})$ and $\Lambda(\bar{\Lambda})$ in different centrality bins and the nuclear modification factors ($R_{CP}$) for these hadrons are calculated. The centrality dependence of the average collective transverse velocity $\langle\beta\,(r)\rangle$ for the hot and dense quark matter is obtained in Au+Au collisions, and it is applied to a relative smaller Cu+Cu collision system. The centrality dependence of $\mathrm{{p}_{T}}$ spectra and the $R_{CP}$ for $\pi^{0}$, $K_{s}^{0}$ and $\Lambda$ in Cu+Cu collisions at $\sqrt{s_{NN}}=200$ GeV are well described. The results show that $\langle\beta\,(r)\rangle$ is only a function of the number of participants $N_{part}$ and it is independent of the collision system. ###### keywords: relativistic high energy nucleus-nucleus collisions, nuclear modification factor, transverse collective flow, quark combination model ###### pacs: 2 5.75.-q, 25.75.Ld, 25.75.Nq 00footnotetext: 0 — LABEL:LastPage ## 1 Introduction The hadronic collectivity is one of the important properties in ultra- relativistic heavy ion collisions. It provides a lot of information on the initial spatial anisotropy of the reaction zone[2, 3, 4, 5], the degree of thermalization [6, 7] and the hadronization mechanism of the hot and dense medium produced in nucleus-nucleus collisions[8, 9]. It can also help to understand the broadening of jetlike particle correlations[10] and high $\mathrm{{p}_{T}}$ jet-quenching[11, 12]. Strong partonic multiple scatterings in nucleus-nucleus collisions would generate the collectivity of quarks, which then result in the observed collectivity of final hadrons. The quark number scaling of the hadronic elliptic flow $\upsilon_{2}$ is a piece of evidence for this original quark collectivity [13, 14, 15]. Being a key hadronization mechanism, quark combination picture has successfully described many features of multi-particle production in high energy heavy ion collisions, e.g. the high $p/\pi$ ratio in intermediate transverse momentum region[16, 17, 18], the quark number scaling behavior of hadron elliptic flow[9, 19, 20] and its fine structure at small $\mathrm{{p}_{T}}$[21], hadron longitudinal and transverse momentum distributions[8, 22, 23, 24], the yields and multiplicity ratios[25]. Therefore, one can extract the collectivity of the hot and dense quark matter from the data of hadrons through quark combination mechanism. In general, the nucleus-nucleus collisions in different centralities will produce different sizes of the hot and dense quark matter. This would cause the transverse collective flow for the hot and dense quark matter varying with collision centralities. This variance of collective flow in quark level would be embodied in the transverse momentum spectra of final hadrons[26, 27] and particularly in their nuclear modification factors $R_{CP}$. The thermal and hydrodynamic models have described the centrality dependence of the transverse momentum distributions for final hadrons in low $\mathrm{{p}_{T}}$ region with a statistical hadronization method[28, 29, 30, 31, 32, 33]. Using the quark recombination at intermediate $\mathrm{{p}_{T}}$ and parton fragmentation at high $\mathrm{{p}_{T}}$, the Duke group in Ref. [9] has explained the strong suppression of hadron $R_{CP}$ at high transverse momenta, and the baryon- meson difference of hadron $R_{CP}$ in intermediate $\mathrm{{p}_{T}}$ region. In the present paper, with quark combination at all $\mathrm{{p}_{T}}$, we use our quark combination model to study the hadron $\mathrm{{p}_{T}}$ spectra from central to peripheral collisions. We investigate the $\mathrm{{p}_{T}}$ spectra of identified hadrons at midrapidity in different centralities in Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV to obtain the centrality dependence of transverse collective flow for the hot and dense quark matter. Furthermore, we apply it to relative smaller Cu+Cu collision systems, calculate the transverse momentum spectra of final hadrons and nuclear modification factors $R_{CP}$ in Cu+Cu collisions at $\sqrt{s_{NN}}=200$ GeV, and compare them with the experimental data from STAR and PHENIX Collaborations. In the next two sections, we briefly introduce the quark combination model and the transverse momentum spectra of quarks just before hadronization. The results and discussions are in Sect. 4. Summary is given in Sect. 5. ## 2 The quark combination model Within the same quark combination mechanism, all kinds of combination-like models, such as recombination model[16, 22, 34], and coalescence model[17, 20], have their own features. Our quark combination model was first proposed for high energy $e^{+}e^{-}$ and $pp$ collisions[35, 36, 37, 38, 39, 40]. Recently we have extended the model to ultra-relativistic heavy ion collisions[21, 23, 25, 41]. The model describes the production of initially produced ground state mesons ($36-plets$) and baryons ($56-plets$). In principle, it can also be applied to the production of excited states [37] and exotic states[25]. These hadrons through combination of constituent quarks are then allowed to decay into the final state hadrons. We take into account the decay contributions of all resonances of $56-plet$ baryons and $36-plet$ mesons, and cover all available decay channels by using the decay program of PYTHIA 6.1 [42]. The main idea is to line up quarks and anti-quarks in a one- dimensional order in phase space, e.g. in rapidity, and let them combine into initial hadrons one by one following a combination rule. See the second section of Ref. [25] for the short description of such a rule. Of course, we also take into account the near correlation in transverse momentum by limiting the maximum transverse momentum difference for quarks and antiquarks as they combine into hadrons. The flavor SU(3) symmetry with strangeness suppression in the yields of initially produced hadrons is fulfilled in the model [35, 37]. ## 3 $\mathrm{{p}_{T}}$ spectra of the constituent quarks at hadronization It is known that the measured hadron $\mathrm{{p}_{T}}$ spectra in relativistic heavy ion collisions exhibit a two-component behavior. The spectra take an exponential form at low $\mathrm{{p}_{T}}$ and a power-law form at high $\mathrm{{p}_{T}}$. Based on parton-hadron duality, the transverse momentum spectra of constituent quarks just before hadronization should also have the same property. In principle, the quarks just before hadronization come from two parts, i.e. the thermal quarks from the hot medium produced in collisions and the minijet quarks from initial hard collisions. The final hadrons are the total contribution of the two parts. But, just as shown in the second figure in Ref. [9], the minijet quarks dominate the large transverse momenta where thermal quarks take a very small proportion, and thermal quarks dominate the low transverse momentum region where the minijet quarks have a small contribution. Therefore, we can neglect the two small contributions, similar to the treatment in Ref.[17], and adopt a piecewise function to describe approximately the transverse momentum distribution of constituent quarks: $\displaystyle\dfrac{dN_{q}}{{2\pi\hskip 2.84526pt\mathrm{{p}_{T}}d\mathrm{\mathrm{{p}_{T}}}}}$ $\displaystyle=\theta(\mathrm{{p}_{0}}-\mathrm{{p}_{T}})N_{th}f_{th}(\mathrm{{p}_{T}})$ (1) $\displaystyle+\theta(\mathrm{{p}_{T}}-\mathrm{{p}_{0}})N_{jet}f_{jet}(\mathrm{{p}_{T}}),$ where $\theta(x)$ is the step function, $N_{th}$ is the number of thermal quarks and $N_{jet}$ is the number of minijet quarks. $\mathrm{{p}_{0}}$ is the transition point from thermal distribution to power-law distribution, which is determined by the spectra continuity. In fact, the interaction between the thermal quarks and the minijet quarks leads to a smooth spectrum around $\mathrm{{p}_{0}}$, and we neglect this effect in this paper. The hot and dense quark matter produced in nucleus-nucleus collisions at RHIC energies shows a significant collective character[13, 14, 43]. The $\mathrm{{p}_{T}}$ spectra of thermal quarks at hadronization can be described by a thermal phenomenological model incorporating the transverse flow of thermal medium[44]. The quarks and antiquarks transversely boost with a flow velocity profile $\beta_{r}(r)$ as a function of transverse radial position $r$. $\beta_{r}(r)$ is parameterized by the surface velocity $\beta_{s}$: $\beta_{r}(r)=\beta_{s}\,\xi^{\,n}$, where $\xi=r/R_{max}$, and $R_{max}$ is the thermal source maximum radius ($0<\xi<1$). The transverse flow of thermal medium can be equivalently described by a superposition of a set of thermal sources, each boosted with transverse rapidity $\rho=tanh^{-1}\beta_{r}$ [44]: $\displaystyle f_{th}(\mathrm{{p}_{T}})=\dfrac{dn_{th}}{{2\pi\hskip 2.84526pt\mathrm{{p}_{T}}d\mathrm{{p}_{T}}}}=$ $\displaystyle A\int_{0}^{1}\xi\,d\xi\,m_{T}\,$ (2) $\displaystyle\times{}I_{0}\bigg{(}\dfrac{\mathrm{{p}_{T}}\,sinh\,\rho}{T}\bigg{)}$ $\displaystyle K_{1}\bigg{(}\dfrac{m_{T}\,cosh\,\rho}{T}\bigg{)},$ where $A$ is the normalization constant in the region $\mathrm{{p}_{T}}\in[0,\mathrm{{p}_{0}}]$. $I_{0}$ and $K_{1}$ are the modified Bessel functions. $m_{T}=\surd{\overline{{\mathrm{{p}_{T}}}^{2}+m^{2}}}$ is the transverse mass of the constituent quark. $T$ is the hadronization temperature. The average transverse velocity can be written as $\langle\beta_{r}\rangle=\dfrac{\int\beta_{s}\,\xi^{\,n}\xi\,d\xi}{\int\xi\,d\xi}=\dfrac{2}{n+2}\beta_{s}.$ (3) With fixed hadronization temperature $T=170$ MeV and parameter $n=0.5$, the average transverse velocity $\langle\beta_{r}\rangle$ is able to characterize the transverse collective flow of the hot and dense quark matter. The quarks and antiquarks with high transverse momenta are mainly from the minijets created in initial hard collisions among nucleons. Here the so-called minijet quarks are those just before hadronization. They are the parton remnants after the revolution of the initial hard partons by gluon radiation and split, and are different from those in the fragmentation model. The $\mathrm{{p}_{T}}$ spectra of minijet quarks at hadronization can be parameterized as follows: $f_{jet}(\mathrm{{p}_{T}})=\dfrac{dn_{jet}}{{2\pi\hskip 2.84526pt\mathrm{{p}_{T}}d\mathrm{\mathrm{{p}_{T}}}}}=B\,\bigg{(}1+\dfrac{\mathrm{{p}_{T}}}{\mathrm{{p}_{0}}}\bigg{)}^{\,-\alpha},\\\ $ (4) where $B$ is the normalization constant in the region $\mathrm{{p}_{T}}\in(\mathrm{{p}_{0}},\infty)$. There are four independent parameters in Eq. (1) to determine the transverse momentum distributions of quarks: $N_{th}$, $N_{jet}$, $\langle\beta_{r}\rangle$ and $\alpha$. Here, we extract the values of these parameters for the light and strange quark $\mathrm{{p}_{T}}$ spectra at midrapidity from the data of $\pi^{0}$ and $K_{s}^{0}$[45, 46], respectively. In our quark combination model, removing the resonance decay contributions from the measured $\pi^{0}$ and $K_{s}^{0}$ transverse momentum distributions, we get the initially produced $\pi^{0}$ and $K_{s}^{0}$ transverse momentum spectra. The values of parameters are inversely extracted from these initial spectra. We obtain four groups of results corresponding to the centrality bins $0-10\%$, $20-40\%$, $40-60\%$, and $60-80\%$. They are shown in Fig. $1$. The lines in the figure are the parameterized results, from which we can get the quark $\mathrm{{p}_{T}}$ spectra in any collision centrality. In Fig. 2, we also show the quark $\mathrm{{p}_{T}}$ spectra in four collision centralities mentioned-above. The values of the parameters $N_{th}$ (a), $N_{jet}$ (b), $\langle\beta_{r}\rangle$ (c) and $\alpha$ (d) for the $\mathrm{{p}_{T}}$ spectra of light quarks(filled circles) and strange quarks(open circles) at midrapidity in four different centrality bins. The corresponding centrality bins are $0-10\%$, $20-40\%$, $40-60\%$, and $60-80\%$. The solid and dashed lines are the parameterized results for light and strange quarks respectively. The $\mathrm{{p}_{T}}$ spectra of the light quarks (a) and strange quarks (b) at midrapidity in $Au+Au$ collisions at $\sqrt{s_{NN}}=200$ GeV. One can see from Fig. 1 that the centrality dependence of the parameter $N_{th}$ is different from that of $N_{jet}$. As we know, the thermal quarks which dominate the low $\mathrm{{p}_{T}}$ region carry most of the collision energy. The available energy used to produce hadrons in different centrality is proportional to the number of nucleon participants $N_{part}$ at the fixed collision energy. Therefore, the number of thermal quarks is approximately proportional to $N_{part}$. The minijet quarks with high $\mathrm{{p}_{T}}$ are mainly from the hard-jet created in initial hard collisions among nucleons, and their quantity is mainly determined by the number of binary collisions $N_{coll}$, which is obviously different from $N_{part}$. Generally, the higher the collision centrality is, the bigger the bulk volume for hot medium, the stronger the transverse collective flow $\beta$. The bigger the bulk volume for hot medium is, the more energy loss for minijet quarks, thus the steeper the minijet-quark spectra and the bigger the parameter $\alpha$. It is further observed that the strange quarks are different from light quarks not only in the quantity (due to strangeness suppression) but also in the momentum distribution. It is easy to understand that the spectrum of strange quarks at high $\mathrm{{p}_{T}}$, due to its heavier effective mass, is steeper than that of light quarks, i.e. $\alpha^{(s)}>\alpha^{(u,d)}$. In the low $\mathrm{{p}_{T}}$ range, however, the spectrum of strange quarks is flatter than the light quarks because $\beta^{(s)}>\beta^{(u,d)}$. By analyzing the data of multi-strange hadrons $\phi$, $\Xi$ and $\Omega$, Ref. [47] also draws the same conclusion. Furthermore, the similar property is also obtained in longitudinal orientation [48]. As we know, the expansion evolution of the plasma in the partonic phase is also a process of obtaining the effective mass for partons. Due to the heavier effective mass, the strange quarks may undergo a stronger hydrodynamic expansion in the partonic phase than the light quarks. As is known to all, though the nucleus-nucleus collisions at top RHIC energy exhibit a high degree of transparency, there are still a few net-quarks which stopped in the midrapidity region. The $p_{T}$ spectra of both $\pi^{0}$ and $K_{s}^{0}$ can not reflect the information of net quarks[41]. We obtain the number of net quarks at midrapidity in different centrality bins by fitting the rapidity densities of net-proton[27]. Note that the ratios of $\pi^{-}/\pi^{+}$ and $\bar{p}/p$ measured by STAR and PHENIX Collaborations reveal weak dependence of centrality and transverse momentum[27, 49]. The transverse momentum distribution of net quarks is taken to be the same as that of the newborn light quarks in the model. With the input, we can give the transverse momentum distributions of various hadrons in different collision centralities and the nuclear modification factors $R_{CP}$ for these hadrons. Just as mentioned in the above section, we consider the decay contributions from all available decay channels of all resonances by using the decay program of PYTHIA 6.1 [42]. Therefore, we can directly compare our calculated results with the experimental data. ## 4 Results and discussions ### 4.1 The $\mathrm{{p}_{T}}$ spectra of hadrons in Au+Au collisions We firstly calculate the transverse momentum spectra of $\pi^{\pm}$ and $p\,(\bar{p})$ in different centrality bins in Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV. The results are shown in Fig. $2$. Here, the pion spectra are corrected to remove the feed-down contributions from $K_{s}^{0}$ and $\Lambda(\bar{\Lambda})$. The $\mathrm{{p}_{T}}$ spectra of hadrons in low $\mathrm{{p}_{T}}$ region are specially shown in the inserted plots. One can see that the calculated results agree well with the data from the STAR Collaboration. The strange hadron production can better reflect the property of the hot and dense quark matter produced in collisions. We also compute the transverse momentum distributions of strange hadrons $K^{\pm}$, $\Lambda(\bar{\Lambda})$ in different centrality bins. The results are shown in Fig. $3$ and compared with the data. The transverse momentum spectra of $\pi^{\pm}$ and $p\,(\bar{p})$ at midrapidity in different centrality bins in Au+Au collisions at $\sqrt{s_{NN}}=200$GeV. The data are taken from STAR Collaboration [49]. The transverse momentum spectra of strange hadrons in different centrality bins in Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV. The data are taken from PHENIX and STAR Collaborations [27, 50]. As shown above, the good agreement between the model predictions and the experimental data confirms the validity of our model from the central to the peripheral collisions. ### 4.2 The nuclear modification factors $R_{CP}$ for hadrons in Au+Au collisions The nuclear modification factors $R_{CP}$ can reflect more precisely the variation of hadron $\mathrm{{p}_{T}}$ spectra in different centrality bins. It is quantified as[49]: $R_{CP}(\mathrm{{p}_{T}})=\frac{[(d^{2}N/(2\pi\mathrm{{p}_{T}}d\mathrm{{p}_{T}}dy))/N_{bin}]^{Central}}{[(d^{2}N/(2\pi\mathrm{{p}_{T}}d\mathrm{{p}_{T}}dy))/N_{bin}]^{Peripheral}}.$ (5) In Fig. $4$, we give the computed results of $R_{CP}$ for $\pi^{+}+\pi^{-}$ and $p+\bar{p}$ ($0-10\%/60-80\%$), $K^{\pm}$ and $\Lambda+\bar{\Lambda}$ (0-5%/60-80%), and compare them with the experimental data. The nuclear modification factors $R_{CP}$ for identified hadrons in Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV. The data are taken from STAR Collaboration [49, 50, 51]. The hadrons in low $\mathrm{{p}_{T}}$ region are mainly from the thermal quark combination. As shown in the first figure, the transverse collective flow of the hot and dense quark matter, denoted by $\langle\beta_{r}\rangle$, decreases with the falling centrality. This leads to the $\mathrm{{p}_{T}}$ spectra of thermal quarks becoming softer in peripheral collisions, and then results in an increasing trend of $R_{CP}$ for hadrons in low $\mathrm{{p}_{T}}$ region. The hadrons with intermediate $\mathrm{{p}_{T}}$ are mainly produced by the combination of thermal quarks with minijet quarks. As we all know, with the decreasing collision centrality, the volume size of the hot and dense quark matter becomes small and the energy loss of minijet quarks correspondingly becomes small as they traverse the hot and dense medium. It leads to the $\mathrm{{p}_{T}}$ spectra of minijet quarks becoming harder in peripheral collisions, which is embodied in the decreasing value of parameter $\alpha$ shown in the first figure with the falling centrality. With the increasing of $\mathrm{{p}_{T}}$, the hadron $R_{CP}$ goes over from the rise caused by the combination of thermal quarks to fall caused by the combination of minijet quarks. One can see that the quark combination model well describes the behavior of $R_{CP}$ for final hadrons in the whole $\mathrm{{p}_{T}}$ region considering the variation of the transverse collective flow of the hot and dense quark matter. In addition, the data show that the $R_{CP}$ for baryons clearly exhibits less suppression compared with that of mesons in intermediate transverse momentum region. This type dependence of $R_{CP}$, which is dependent upon the number of constituent quarks rather than hadronic mass, has been qualitatively discussed earlier in Refs. [9, 51] as an experimental support of the quark combination picture. Our results further manifest this baryon-meson difference of $R_{CP}$ in intermediate $\mathrm{{p}_{T}}$ region. The experimental data show that the baryon-meson difference of $R_{CP}$ disappears at higher $\mathrm{{p}_{T}}$. Using the fragmentation mechanism, the Duke group in Ref. [9] has explained this common degree of suppression for both baryons and mesons at high $\mathrm{{p}_{T}}$. Our results from the quark combination can also describe this behavior of $R_{CP}$ at high $\mathrm{{p}_{T}}$. These two different hadronization mechanisms produce similar results. It suggests that the disappearance of baryon-meson difference of $R_{CP}$ at high $\mathrm{{p}_{T}}$ is not caused by hadronization mechanism. The hard scatterings which take place near the surface of the collisions produce the back-to-back dijets. One-side jets escape almost without energy loss, while the away-side jets lose significant energy as they traverse the hot and dense matter. The hadrons with high $\mathrm{{p}_{T}}$ in all collision centralities are mostly from these jets without energy loss. The transverse momentum distributions of these minijet quarks with high $\mathrm{{p}_{T}}$ in central collisions, except the quantity, are almost the same with those in peripheral collisions. Therefore, no matter what the hadronization mechanism is, there exists a similar suppression of the $R_{CP}$ for baryons and mesons at high $\mathrm{{p}_{T}}$. It is also observed that the calculated $R_{CP}$ deviates from the data to a certain degree. The reason may be that some effects, such as the production of excited-state hadrons and final-state rescattering, are not considered currently in the model. As we know, a small quantity of the excited-state hadrons are produced in relativistic heavy ion collisions, and the yields and momentum distributions of final hadrons are influenced by the decay contribution of these excited-state hadrons to a certain extent. As the decay branch ratios of many excited-state hadrons are incompletely measured, the contributions of excited-state hadrons are neglected in the current model. On the other hand, the perfect quark-number scaling of hadron elliptic flow $v_{2}$ suggests that the influence of final-state rescattering on hadron distribution is finite and small[13], so we also neglect it in the work. These two effects would affect the fine observable $R_{CP}$. ### 4.3 The $\mathrm{{p}_{T}}$ spectra and nuclear modification factors $R_{CP}$ for hadrons in Cu+Cu collisions Now, we apply the centrality(participants) dependence of parameters in Au+Au collisions which determine the $\mathrm{{p}_{T}}$ spectra of quarks in different centralities, to the relative smaller Cu+Cu collision system at $\sqrt{s_{NN}}=200$ GeV. In Fig. $5$, we show the calculation results for the transverse momentum spectra of $\pi^{0},K_{s}^{0}$ and hyperon $\Lambda$ in different centrality bins and $R_{CP}$ for these hadrons in Cu+Cu collisions at $\sqrt{s_{NN}}=200$ GeV. The $\mathrm{{p}_{T}}$ spectra of hadrons in low $\mathrm{{p}_{T}}$ region are specially shown in the inserted plots. We find that the results agree well with the data in the low transverse momentum region in all centrality bins. This implies that the transverse collective flow of the hot and dense quark matter, i.e. $\langle\beta_{r}\rangle$, is only a function of $N_{part}$ and irrelevant to the collision system at the same collision energy. The good agreement of our results with the $\pi^{0}$ data at high transverse momenta also suggests that the energy loss of minijet quarks in Cu+Cu collisions is almost the same with that in Au+Au collisions with the same participants $N_{part}$ at $\sqrt{s_{NN}}=200$ GeV. It is consistent with the recent measurement of STAR Collaboration[53, 54]. The above results suggest that the hot and dense quark matter produced in Au+Au and Cu+Cu collisions at the same $N_{part}$ and collision energy has similar strong-interacting character. Of course, even at the same $N_{part}$ and collision energy, the initial spatial eccentricity of the overlap collision geometry in Au+Au collisions is obviously different from that in Cu+Cu collision systems. The difference is clearly reflected by some important observations, e.g. the elliptic flow of final hadrons and the global polarization of hyperon[13, 51, 55, 56, 57]. The transverse momentum spectra of identified hadrons in different centrality bins (a) and their nuclear modification factors $R_{CP}$ (b) in Cu+Cu collisions at $\sqrt{s_{NN}}=200$ GeV. The data are taken from PHENIX and STAR Collaborations [45, 52]. ## 5 Summary Using the quark combination model, we study systematically the centrality dependence of the transverse momentum distributions for the identified hadrons at midrapidity in Au+Au and Cu+Cu collisions at $\sqrt{s_{NN}}=200$ GeV. The centrality dependence of the parameters for quark $\mathrm{{p}_{T}}$ spectra is extracted from the data of $\pi^{0}$ and $K_{s}^{0}$ in Au+Au collisions. We calculate the transverse momentum distributions of $\pi^{\pm}$, $p\,(\bar{p})$, $K^{\pm}$ and $\Lambda(\bar{\Lambda})$ in five centrality bins. The good agreement between our results and the data indicates that the quark combination hadronization mechanism is applicable to all collision centralities. The nuclear modification factors $R_{CP}$ for $\pi^{+}+\pi^{-}$ and $p+\bar{p}$ ($0-10\%/60-80\%$), $K^{\pm}$ and $\Lambda+\bar{\Lambda}$(0-5%/60-80%) are calculated and compared with the data. The quark combination model well describes the behavior of $R_{CP}$ for final hadrons in the whole $\mathrm{{p}_{T}}$ region considering the decrease of the transverse collective flow of the hot and dense quark matter from the central collisions to the peripheral collisions. The disappearance of the baryon-meson difference of $R_{CP}$ at higher $\mathrm{{p}_{T}}$ is derived from the same transverse momentum distribution of minijet quarks between the central and the peripheral collisions rather than the hadronization mechanism. Furthermore, we apply the $N_{part}$ dependence of parameters to the relative smaller Cu+Cu collision system at the same collision energy. We calculate the transverse momentum spectra of $\pi^{0}$, $K_{s}^{0}$, and $\Lambda$ at midrapidity in different centrality bins and $R_{CP}$ of these hadrons in Cu+Cu collisions at $\sqrt{s_{NN}}=200$ GeV. The results agree well with the data in low transverse momentum region in all centrality bins. It suggests that the transverse collective flow of the hot and dense quark matter is only the function of $N_{part}$ and independent of collision system. The calculated $\pi^{0}$ spectrum at high $\mathrm{{p}_{T}}$ is also in good agreement with the data. These results suggest that the hot and dense quark matter produced in Au+Au and Cu+Cu collisions at the same $N_{part}$ and collision energy has similar strong-interacting character. ###### Acknowledgements. We are grateful to Wang Q., Liang Z. T., Yao T. and Han W. for helpful discussions. ## References * [1] * [2] Reisdorf W. and Ritter H. G., Annu. Rev. Nucl. Part. Sci., 1997, 47: 663—709 * [3] Sorge H., Phys. Rev. Lett., 1999, 82 : 2048—2051 * [4] Kolb P., Sollfrank J., and Heinz U., Phys. Lett., 1999, B459: 667—672 * [5] Teaney D. and Shuryak E. V., Phys. Rev. Lett., 1999, 83: 4951—4954 * [6] Kolb P., Sollfrank J., and Heinz U., Phys. Rev., 2000, C62 : 054909 * [7] Voloshin S. A. and Poskanzer A. M., Phys. Lett., 2000, B474 : 27—32 * [8] Greco V., Ko C. M. and Levai P., Phys. Rev., 2003, C68 : 034904 * [9] Fries R. J., Müller B., Nonaka C., et al.Phys. Rev., 2003, C68: 044902 * [10] Armesto N., Salgado C. A., and Wiedemann U. A., Phys. Rev. Lett., 2004, 93: 242301 * [11] Wang X. N., and Gyulassy M., Phys. Rev. Lett., 1992, C68: 1480—1483 * [12] Armesto N., Salgado C. A., and Wiedemann U. A., Phys. Rev., 2005, C72: 064910 * [13] Adare A., et al., (PHENIX Collaboration), Phys. Rev. Lett., 2007, 98: 162301 * [14] Adare A., et al., (PHENIX Collaboration), Phys. Rev. Lett., 2003, 91: 182301 * [15] Enokizono A., for the PHENIX Collaboration, Nucl. Phys., 2007, A787: 37c—43c * [16] Fries R. J., Müller B., Nonaka C. et al., Phys. Rev. Lett., 2003, 90: 202303 * [17] Greco V., Ko C. M. and Levai P., Phys. Rev. Lett., 2003, 90: 202302 * [18] Hwa R. C. and Yang C. B., Phys. Rev., 2003, C67: 034902 * [19] Voloshin S. A., Nucl. Phys., 2003, A715: 379c—388c * [20] Molnar D. and Voloshin S. A., Phys. Rev. Lett., 2003, 91: 092301 * [21] Yao T., Xie Q. B. and Shao F. L., Chin. Phys., 2008, C32: 356—362. * [22] Hwa R. C. and Yang C. B., Phys. Rev., 2004, C70: 024905 * [23] Shao F. L., Yao T., and Xie Q. B., Phys. Rev., 2007, C75: 034904 * [24] Hwa R. C. and Tian Z., Phys. Rev., 2005, C72: 024908 * [25] Shao F. L., Xie Q. B. and Wang Q., Phys. Rev., 2005, C71: 044903 * [26] Arsene I., et al., (BRAHMS Collaboration), Phys. Rev., 2005, C72: 014908 * [27] Adler S. S., et al., (PHENIX Collaboration), Phys. Rev., 2004, C69: 034909 * [28] Schnedermann E. and Heinz U., Phys. Rev. Lett., 1992, 69: 2908—2911 * [29] Kolb P. F. and Rapp R., Phys. Rev., 2003, C67: 044903. * [30] Nonaka C. and Bass S. A., Phys. Rev., 2007, C75: 014902 * [31] Heinz U. and Kolb P., Nucl. Phys., 2002, A702: 269—280 * [32] Huovinen P., Nucl. Phys., 2005, A761: 296—312 * [33] Prorok D., Phys. Rev., 2006, C73: 064901 * [34] Pratt S. and Pal S., Phys. Rev., 2005, C71: 014905 * [35] Xie Q. B. and Liu X. M., Phys. Rev., 1988, D38: 2169—2177 * [36] Liang Z. T. and Xie Q. B., Phys. Rev., 1991, D43: 751—759 * [37] Wang Q. and Xie Q. B., J. Phys., 1995, G21: 897—904 * [38] Zhao J. Q., Wang Q. and Xie Q. B., Sci. Sin., 1995, A38: 1474—1487 * [39] Wang Q., Si Z. G. and Xie Q. B., Int. J. Mod. Phys., 1996, A11: 5203—5210 * [40] Si Z. G., Xie Q. B. and Wang Q., Commun. Theor. Phys., 1997, 28: 85—94 * [41] Song J., Shao F. L., Xie Q. B., et al., The influence of net-quarks on the yields and rapidity spectra of identified hadrons, arXiv: hep-ph/0801.0918 * [42] Sjostrand T., Eden P., Friberg C., et al. , Comput. Phys. Commun., 2001, 135: 238—259 * [43] Abelev B. I., et al., (STAR Collaboration), Phys. Rev. Lett., 2007, 99: 112301 * [44] Schnedermann E., Sollfrank J., and Heinz U., Phys. Rev., 1993, C48: 2462—2475 * [45] Sakaguchi T., for the PHENIX Collaboration, Contributions for Quark Matter 2006 Poster Session, Shanghai, China, 14th-20th November, 2006, arXiv: nucl-ex/0703027 * [46] Adams J., et al., STAR, STAR-RICH Collaborations, nucl-ex/0601042 * [47] Chen J. H., Jin F., Gangadharan D., et al., Parton distribution at hadronization from bulk dense matter produced at RHIC, arXiv: 0801.2265[nucl-th] * [48] Song J., Shao F. L., Xie Q. B., et al., Rapidity dependence of hadron production in central Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV, arXiv: nucl-th/0703095 * [49] Abelev B. I., et al., (STAR Collaboration), Phys. Rev. Lett., 2006, 97: 152301 * [50] Admas J., et al., (STAR Collaboration), Phys. Rev. Lett, 2006, 98: 062301 * [51] Adams J., et al., (STAR Collaboration), Phys. Rev. Lett., 2003, 92: 052302 * [52] Timmins A. R., for the STAR Collaboration, Quark Matter 2006 Proceedings, Shanghai, China, 14th-20th November, 2006, arXiv: nucl-ex/0708.3290 * [53] Hollis R. S., for the STAR Collaboration, proceedings for the Lake Louise Winter Institute 2007, Alberta, Canada, 19th-24th February 2007, arXiv: nucl-ex/0705.0686 * [54] Bekele S., for the BRAHMS Collaboration, arXiv: nucl-ex/0601011 * [55] Liang Z. T. and Wang X. N., Phys. Rev. Lett., 2005, 94: 102301 * [56] Abelev B. I., et al., (STAR Collaboration), Phys. Rev., 2007, C76: 024915 * [57] Back B. B., et al., (PHOBOS Collaboration), Phys. Rev. Lett., 2002, 89: 222301

arxiv-papers

2024-09-04T02:48:53.856760

0802.2598

# Centrality, system size and energy dependences of charged-particle pseudo- rapidity distribution De-ming Wei 1 Feng-lan Shao 2 Jun Song 2 and Yun-fei Wang2 1 Shandong Province Linqu Experimental Middle School, Shandong 262600, People’s Republic of China; dmwqf2008@hotmail.com 2 Department of Physics, Qufu Normal University, Shandong 273165, People’s Republic of China ###### Abstract Utilizing the three-fireball picture within the quark combination model, we study systematically the charged particle pseudorapidity distributions in both Au+Au and Cu+Cu collision systems as a function of collision centrality and energy, $\sqrt{s_{NN}}=$ 19.6, 62.4, 130 and 200 GeV, in full pseudorapidity range. We find that: (i)the contribution from leading particles to $dN_{ch}/d\eta$ distributions increases with the decrease of the collision centrality and energy respectively; (ii)the number of the leading particles is almost independent of the collision energy, but it does depend on the nucleon participants $N_{part}$; (iii)if Cu+Cu and Au+Au collisions at the same collision energy are selected to have the same $N_{part}$, the resulting of charged particle $dN/d\eta$ distributions are nearly identical, both in the mid-rapidity particle density and the width of the distribution. This is true for both 62.4 GeV and 200 GeV data. (iv)the limiting fragmentation phenomenon is reproduced. (iiv) we predict the total multiplicity and pseudorapidity distribution for the charged particles in Pb+Pb collisions at $\sqrt{s_{NN}}=5.5$ TeV. Finally, we give a qualitative analysis of the $N_{ch}/<N_{part}/2>$ and $dN_{ch}/d\eta/<N_{part}/2>|_{\eta\approx 0}$ as function of $\sqrt{s_{NN}}$ and $N_{part}$ from RHIC to LHC. ###### keywords: Relativistic Heavy-Ion Collider; Quark Combination Model(QCM), Pseudorapidity Distribution. PACS numbers: 25.75.-q,25.75.Ag ## 1 Introduction In relativistic heavy ion collisions at RHIC energies, the charged particles are produced copiously in vacuum. The number of charged particles per unit pseudo-rapidity $dN_{ch}/d\eta$, and in particular its dependence on some variables, such as rapidity, collision centrality and energy, are the important observables, from which a lot of information about the hot and dense matter created in collisions can be extracted[1][8]. From the pseudo-rapidity density and the transverse energy per particle, one can determine via Bjorken method the initial energy density of the fireball which can provide one piece of evidence for the deconfinement phase transition. In the fragmentation region, the charged particle production, in general, is thought to be distinct from that at mid-rapidity, although there is no obvious evidence for two separate regions at any of the RHIC energies. The pseudo-rapidity density $dN_{ch}/d\eta$ in forward rapidity region carries some information of leading particles produced in collisions [9]. The experimental data about the charged- particle pseudo-rapidity density in both Au+Au and Cu+Cu collision systems have been presented by the PHOBOS collaboration [10][13], the PHENIX collaboration [14], and the BRAHMS Collaboration [15, 16]. The data for the scaled and shifted pseudo-rapidity distribution $dN_{ch}/d\eta^{\prime}/\langle N_{part}/2\rangle$, exhibit the limiting fragmentation phenomenon in both Au+Au and Cu+Cu collisions at different energies and centralities[17, 18]. Recombination of partons [19][24], Partonic coalescence [25][28] and QCM [29][31] have been made to described many observations. In our previous work [29], using a Gaussian-like shape rapidity distribution for constituent quarks as a result of the Landau hydrodynamic evolution[32, 33], we have presented the pseudo-rapidity distributions of charged particles in Au+Au collisions as a function of collision centrality and energy. The calculation results are in good agrement with the data in central collisions. In peripheral collisions, our predictions are slightly lower than data in high rapidity range. The reason may be that we have not considered the contribution of leading particles. In present work, taking into account the leading particle influence, we apply a three-fireball picture [9, 34, 35] to describe the evolution of the hot and dense quark matter produced in collisions, and obtain the rapidity distribution of the constituent quarks just before hadronization. Then let these constituent quarks combine into initial hadrons according to a quark combination rule, and allow the resonances in the initial hadrons to further decay into final hadrons with the help of the event generator PYTHIA 6.1 [36]. ## 2 Three-fireball picture and quark combination model In this section, we introduce the three-fireball picture, which is used to describe the rapidity distribution of quark and antiquarks just before hadronization. In addition, we briefly introduce the QCM which describes the hadronization of these quarks and antiquarks produced in collisions. ### 2.1 Three-fireball picture It is known that the nucleus-nucleus collisions at RHIC energies are neither fully stopped nor fully penetrated. As the incident nuclei penetrate through the target nuclei, the most of the collision energy is deposited in collision region to form a big central fireball, and the penetrating quark matter forms two small fireballs, i.e. target and projectile fireballs, in forward rapidity region. The charged hadron pseudorapidity distribution is the total contributions from the three fireballs. The big central fireball which contains the main part of collision energy, controls the rough shape of charged particle pseudorapidity distribution (width and height). Relativistic hydrodynamics has successfully described the evolution of system before hadronization. Here we use a Gaussian-type rapidity distribution for constituent quarks as a result of the Landau hydrodynamic evolution [37][40], $f(y)=\frac{1}{\sqrt{2\pi\sigma^{2}}}\exp\big{(}-\frac{y^{2}}{2\sigma^{2}}\big{)},$ (1) where $\sigma^{2}\approx\frac{2c_{s}^{2}}{1-c_{s}^{4}}\ln\big{(}\frac{E\sqrt{s_{NN}}}{2\,m_{p}\,\epsilon_{c}}\big{)}.$ (2) Here, $E$ is the effective energy offered by per participant pair, and it is used to produce the central fireball. $m_{p}$ is proton mass and $c_{s}$ is the sound velocity. $\epsilon_{c}$ is the energy in the volume of a free hadron at hadronization. All quarks and anti-quarks in the central fireball are within the rapidity range $[-y_{max},y_{max}]$, $y_{max}=\frac{c_{s}}{1+c_{s}^{2}}\ln{\frac{E\sqrt{s_{NN}}}{2\,m_{p}\,\epsilon_{c}}}.$ (3) The average constituent quark number in the big central fireball can be obtained from a simple quark production model [29, 30] $\langle{N_{q}}\rangle=2[(\alpha^{2}+\beta E)^{1/2}-\alpha]\langle{N_{\rm part}}/2\rangle,$ (4) where the parameter $\beta\approx$ $3.6$ GeV, and the parameter $\alpha=\beta m-\frac{1}{4}$, $m$ is averaged quark mass and it is taken to be 0.36 GeV, they are the same with Ref. Feng-Lan Shao:2007a. The two penetrating fireballs mainly consist of leading light quarks. We also adopt a Gaussian type rapidity distribution for leading quarks $f^{\prime}(y)=\frac{1}{\langle N_{q(T/P)}\rangle}\frac{dN_{q(T/P)}}{dy}=\frac{\exp(-\frac{(y+y_{0})^{2}}{2\sigma^{\prime 2}})}{\sqrt{2\pi\sigma^{\prime 2}}},$ (5) where $N_{q(T/P)}$ is the total quark number in the penetrating target and projectile fireballs. $y_{0}=\pm\frac{y_{beam}+y_{max}}{2}$ is the rapidity center for target and projectile fireballs respectively. Rapidity distribution range of quarks in the center of mass frame is $y\in[-y_{beam},-y_{max}]$ for target fireball and $y\in[y_{max},y_{beam}]$ for projectile fireball, respectively. In this work, the spectrum width of penetrating fireballs is taken to be $\sigma^{\prime}=0.18$. The total energy of the three fireballs in nucleus-nucleus collisions is $\sqrt{s_{NN}}\,\langle N_{part}/2\rangle$, $E_{(T+P)}=(\sqrt{s_{NN}}-E)\langle N_{part}/2\rangle,$ (6) where $E_{(T+P)}$ is total energy of the two penetrating fireballs. The average number of quarks in penetrating projectile and target fireballs $\langle N_{q(T+P)}\rangle$ is determined by $E_{(T+P)}$: $\langle N_{q(T+P)}\rangle=\frac{E_{(T+P)}}{\langle E_{q}\rangle}.$ (7) $\langle E_{q}\rangle$ is the average energy of each quark in the penetrating projective/target fireballs, and it can be written as: $\langle E_{q}\rangle=\int\limits_{y_{max}}^{y_{beam}}{m_{T}\cosh(y)f^{\prime}(y)dy},$ (8) where $m_{T}=\sqrt{m^{2}+p^{2}_{T}}$ is the transverse mass of leading quarks. The transverse momentum $p_{T}$ of leading quarks is approximately taken to be 0.25 GeV, one third of the value of net-proton at forward rapidity $y\approx 3$[41]. ## 3 The quark combination model The QCM was first proposed for high energy $e^{+}e^{-}$ and $pp$ collisions and recently it was extended to ultra-relativistic heavy ion collisions [31, 42]. The model describes the production of initially produced ground state mesons ($36-plets$) and baryons ($56-plets$). In principle the model can also be applied to the production of excited states [43]. These hadrons through combination of constituent quarks are then allowed to decay into the final state hadrons. We take into account the decay contributions of all resonances of $56-plets$ baryons and $36-plets$ mesons, and cover all available decay channels by using the decay program of PYTHIA 6.1 [36]. The main idea is to line up $N_{q}$ quarks and anti-quarks in a one-dimensional order in phase space, e.g. in rapidity, and let them combine into initial hadrons one by one following a combination rule (see section 2 of Ref. Shao:2004 for a short description of such a rule). We note that it is very straightforward to define the combination in one dimensional phase space, but it is highly complicated to do it in two or three dimensional phase space [44]. The flavor SU(3) symmetry with strangeness suppression in the yields of initially produced hadrons is fulfilled in the model [30, 43]. ## 4 Centrality, system size and energy dependence of charged-particle pseudo-rapidity distribution In this section, we will study the system size, energy and centrality dependence of pseudorapidity distribution of charged particles, and multiplicity distribution in mid and forward rapidity range respectively in relativistic heavy ion collisions. Moreover, if the particle production mechanisms in A+A collisions at RICH and LHC are the same, we predict the energy and centrality dependences of $\frac{N_{ch}}{<N_{part}/2>}$ and $\frac{dN_{ch}}{d\eta<N_{part}/2>}|_{\eta\approx 0}$. From Eq. (6), we can see that there is only one free variable, i.e. $E$ or $E_{(T+P)}$, which should be determined from the experimental data. In the present work, we determine the effective energy $E$ for central fireball by fitting the pseudorapidity density $\frac{dN_{ch}}{d\eta}|_{\eta\approx 0}$ in Au+Au collisions at $\sqrt{s}=130$ GeV. Then the $E_{(T+P)}$ and leading quark number $\langle N_{q(T+P)}\rangle$ can be naturally obtained. Using this method, we get the leading quark number in different collision centralities and parameterize it as the function of nucleon participants $N_{part}$ $\displaystyle N_{q(T+P)}=-84.44+35.82*N_{part}^{0.4}.$ (9) At other energies, basing on the relation between the number of leading quarks and the centrality, we can get $E$, and the pseudorapidity distribution in different centralities within a quark combination model. file=fig1.eps,width=10cm Figure 1: Charged hadron pseudo-rapidity distributions for different centralities in Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV. The lines are our results and the points are data taken from PHOBOS. Applying Eq. (9) to other RHIC energies, we firstly calculate the charged particle pseudorapidity distributions for different centralities in Au+Au collisions at $\sqrt{s_{NN}}=200$ GeV. The data are taken from PHOBOS [11, 12]. The results are shown in Fig. 1. The dashed lines are the contribution from the central fireballs. The dotted lines in the forward pseudorapidity range show the contributions of penetrating target and projectile fireballs respectively. The solid lines are the total contribution of the three fireballs. One can see that our results are in good agreement with the data. In addition, we find that the contribution from leading particles to $dN_{ch}/d\eta$ distributions increases with the decrease of the collision centralities. We also give the results of $dN_{ch}/d\eta$ distribution at $\sqrt{s_{NN}}$=19.6, 62.4 GeV, and they are shown in Fig. 2. The data are taken from PHOBOS [11, 12]. The agreement of calculated results with the data is also satisfactory. The Fig. 3 shows the charged-particle pseudorapidity distributions in most central Au+Au collisions at $\sqrt{s_{NN}}$=19.6, 62.4, 130 and 200 GeV. The results indicate that the contribution from leading particles to $dN_{ch}/d\eta$ distributions increases with the decrease of collision energy. file=fig2.eps,width=12cm Figure 2: Charged hadron pseudo-rapidity distributions for different centralities in Au+Au collisions at $\sqrt{s_{NN}}$=19.6,62.4 GeV. The lines are our results and the points are data taken from PHOBOS. file=fig3.eps,width=10cm Figure 3: The energy dependence of charged particles pseudo-rapidity distributions in most central Au+Au collisions. The lines are our results. The data are taken from PHOBOS. Recently, PHOBOS Collaboration have presented the data on charged-particle pseudorapidity distributions in Cu+Cu collisions at $\sqrt{s_{NN}}$=62.4, 200 GeV [13]. The other goal of this paper is to investigate the systematic dependence of particle production in nuclear collision at RHIC energies, in terms of overall $dN/d\eta$ distributions. We apply Eq. (9) to Cu+Cu collisions, and give the charged-particle pseudorapidity distributions as a function of centrality at $\sqrt{s_{NN}}$=62.4, 200 GeV. The results are shown in Fig. 4, and compared with the data. As we can see, our results are roughly consistent with the experimental data, but are slightly lower than the data in the mid-rapidity range, especially for the central Cu+Cu collisions. The reason may be that we did not consider the difference between the collision geometry in Cu+Cu and Au+Au, even at the same $N_{part}$ in Eq. (9), especially for the central Cu+Cu collisions. The two system have same shape, which indicate a similarity particle production mechanism between Au+Au and Cu+Cu, the results is the same as Ref. Bialas and A. Bzdak. Following that, we think the Au+Au and Pb+Pb collisions have the same particle production mechanism, so we predict the pseudorapidity distribution and charged particle multiplicity in Pb+Pb collisions at $\sqrt{s_{NN}}$=5.5 TeV in the followings. file=fig4.eps,width=13cm Figure 4: Particles pseudo-rapidity distributions for different centralities in Cu+Cu collisions at $\sqrt{s_{NN}}$=62.4, 200 GeV. The lines are our results, and the data are taken from PHOBOS. file=fig5.eps,width=10cm Figure 5: Charged particles pseudorapidity distributions calculated by QCM in most central Pb+Pb collisions at $\sqrt{s}=5.5$ TeV. file=fig6.eps,width=10cm Figure 6: The scaled, shifted pseudorapidity density at $\sqrt{s_{NN}}$=19.6,62.4,130,200 GeV in most central Au+Au,Cu+Cu,and Pb+Pb collisions at $\sqrt{s}$=5.5TeV. The lines are our results, the symbols are data taken from PHOBOS. file=fig7.eps,width=13cm Figure 7: The total number of charged particles (a) and mid-rapidity density (b) per participant pair shown as a function of $N_{part}$ for $\sqrt{s}$ =19.6, 62.4, 130 and 200 GeV in Au+Au collisions, and $\sqrt{s}$ =5500 GeV in Pb+Pb collisions. The solid symbols are data taken from PHOBOS, and the open symbols are our results in (a) and (b). The lines are our prediction at LHC in (a) and (b). The Large Hadron Collider (LHC) is scheduled to begin operation in May 2008. The most pressing issue for the early days at the LHC is to establish the global features of heavy ion collisions. This involves the estimation of the inclusive charged-particle yield and the charged pseudorapidity distribution and so on. In this work, extending Eq. (9) to LHC energy, we can predict the charged particle pseudorapidity distribution as a function of centrality in Pb+Pb collisions. As an example, we calculate the most central Pb+Pb collisions at $\sqrt{s_{NN}}=5500$ GeV in Fig. 5. The total charged-particle multiplicity is about $18170$, and pseudorapidity density $dN_{ch}/d\eta\mid_{\mid\eta\mid<0.5}$ is about 1630. Note that the sound velocity is taken be 1/3. To separate the trivial kinematic broadening of the $dN_{ch}/d\eta$ distribution from the more interesting dynamics, we also study the scaled, shifted pseudorapidity distribution $dN_{ch}/d\eta^{\prime}/\langle N_{part}/2\rangle$ , where $\eta^{\prime}=\eta-y_{beam}$ , in most central Au+Au, Cu+Cu and Pb+Pb collisions at different energies. The calculation results are shown in Fig. 6. We observe that the data at various energies and systems fall on a common limiting curve. The centrality dependence of $N_{ch}/<N_{part}/2>$ and $dN_{ch}/d\eta/<N_{part}/2>|_{\eta\approx 0}$ at $\sqrt{s_{NN}}$=19.6, 62.4, 130, 200 GeV in Au+Au collisions (open symbol) compared with the data (solid symbol) taken from PHOBOS [13] and the prediction at 5500 GeV in Pb+Pb collisions (the solid line) are shown in Fig. 7. Fig. 8 shows the c.m. energy dependence of $N_{ch}/<N_{part}/2>$ and $dN_{ch}/d\eta/<N_{part}/2>|_{\eta\approx 0}$ from RHIC to LHC predicted by the QCM (the lines). It shows that the $N_{ch}/<N_{part}/2>$ and $dN_{ch}/d\eta/<N_{part}/2>|_{\eta\approx 0}$ from RHIC to LHC can grow at logarithmically with $\sqrt{s_{NN}}$. file=fig8.eps,width=12cm Figure 8: The charged particles mean multiplicity $<N_{ch}>$ and pseudorapidity density of charged particle per participant pair ($N_{part}/2$ ) as a function of the c.m. energy of collision. The lines are our results and the symbols are data taken from PHOBOS. ## 5 summery Within a combination model, we study the charged particle pseudo-rapidity distributions in Au+Au and Cu+Cu collision systems as a function of collision centrality and energy ($\sqrt{s_{NN}}=19.6,62.4,130$ and $200$ GeV), in full pseudo-rapidity range. We use a toy model, i.e. three fireballs, to describe the evolution of the hot and dense quark matter produced in collisions. The big central fireball which carries the main part of collision energy controls the rough shape of the charged particle pseudo-rapidity distribution. We apply the Landau relativistic hydrodynamic model to describe the the evolution of highly excited and possibly deconfined quark matter created in the big central fireball. As a result, we obtain a Gaussian-type rapidity spectra of constituent quarks before hadronization. The other two small fireballs in foreword rapidity carry the information of the leading particles. We also use a Gaussian-type rapidity spectra of constituent quarks before hadronization. Then we use our combination model to describe the hadronization of initially produced hadrons including resonances, whose decays are dealt with by the event generator PYTHIA 6.3 [36]. Firstly, by studying the contribution of leading particles to charged-particle pseudo-rapidity distribution in Au + Au collisions for different centralities at $130$ GeV, we extract the centrality dependence of the average number of leading quarks from the data. Then we extend it to other RHIC energies. We calculate the charged particle pseudo- rapidity distributions in both Au+Au and Cu+Cu collision systems as a function of collision centrality, at $\sqrt{s_{NN}}=$ 19.6, 62.4 and 200 GeV, in full pseudo-rapidity range. The calculation results are in good agreement with data. To separate the trivial kinematic broadening of the distributions of the pseudo-rapidity density from more interesting dynamics, we compute the scaled and shifted pseudo-rapidity density distributions $dN_{ch}/d\eta^{\prime}/\langle N_{part}/2\rangle$ with $\eta^{\prime}=\eta- y_{beam}$ at collision energies 19.6, 62.4, 130 and 200 GeV. The good agreement with data is found. Furthermore, we predict the total multiplicity and pseudo-rapidity distribution for the charged particles in most central Pb+Pb collisions at $\sqrt{s_{NN}}=5.5$ TeV. Through investigating detailed $dN_{ch}/d\eta$ distributions, we find that: (i)The contribution from leading particles to $dN_{ch}/d\eta$ distributions increases with the decrease of the collision centrality and energy respectively; (ii)The number of leading particles is, independent of collision energy, only a function of nucleon participants $N_{part}$for the same system; (iii)If Cu+Cu and Au+Au collisions at the same collision energy are selected to have the same $N_{part}$, the resulting of charged particle $dN/d\eta$ distributions are nearly identical, both in the mid-rapidity particle density and the width of the distribution. This is true for both 62.4 GeV and 200 GeV data. (iv)The limiting fragmentation phenomenon is reproduced. Furthermore, we predict the total multiplicity and pseudorapidity distribution for the charged particles in Pb+Pb collisions at $\sqrt{s_{NN}}=5.5$ TeV, and find the $N_{ch}/<N_{part}/2>$ and $dN_{ch}/d\eta/<N_{part}/2>|_{\eta\approx 0}$ from RHIC to LHC can grow at logarithmically with $\sqrt{s_{NN}}$. ## Acknowledgments The authors thank Qu-bing Xie, Wei-han and Tao-yao for helpful discussions. The work is supported in part by the National Natural Science Foundation of China under the grant 10775089 and 10475049. ## References * [1] Xin-Nian Wang and Gyulassy Miklos, Phys. Rev. Lett. 86: 3496 (2001). * [2] Shi-Yuan Li and Xin-Nian Wang, Phys. Lett. B 527, p. 85 (2002). * [3] D. Kharzeev and M. Nardi, Phys. Lett. B 507, 121 (2001). * [4] D. Kharzeev and E. Levin, Phys. Lett. B 523, 79 (2001). * [5] A. Capella and D. Sousa, Phys. Lett. B 511, p. 185 (2001). * [6] J. Dias de Deus and R. Ugoccioni, Phys. Lett. B 494, 53 (2000). * [7] K. JEskola, K. Kajantie and P. V. Ruuskanen et al., Nucl. Phys. B 570, p. 379 (2000). * [8] K. J. Eskola, K. Kajantie and P. V. Ruuskanen et al., Phys. Lett. B 543, p. 208 (2002). * [9] Fu-Hu Liu, Phys. Rev. C 66, 047902 (2002). * [10] PHOBOS Collab.( B. B. Backet al. ), Phys. Rev. Lett. 85, 3100 (2000). * [11] PHOBOS Collaboration. (B. B. Back et al.), Phys. Rev. Lett. 91, 052303 (2003). * [12] PHOBOS Collaboration. (B. B. Back et al.), Phys. Rev. Lett. 87, 102303 (2001). * [13] Roland G et al. Nucl. Phys. A, 2006, 774: 113-128. * [14] PHENIX Collaboration. (K.Adcox et al.), Phys. Rev. Lett. 86, 3500 (2001). * [15] BRAHMS Collaboration. ( I. G. Beardenet al.), Phys. Lett. B 523, 227 (2001). * [16] BRAHMS Collaboration. ( I. G. Beardenet al.), Phys. Rev. Lett., 88, 202301 (2002). * [17] PHOBOS Collaboration. ( R. Nouiceret al.), arXiv:nucl-ex/0601026. * [18] PHOBOS Collaboration. ( R. Nouiceret al.), arXiv:nucl-th/0608038. * [19] R. J. Fries, B. M$\ddot{u}$ller, C. Nonaka and S. A. Bass, Phys. Rev. Lett. 90 , 202303 (2003). * [20] R. J. Fries, B. M$\ddot{u}$ller, C. Nonaka and S. A. Bass Phys. Rev. C 68, 44902 (2003). * [21] Rudolph C. Hwa and C. B. Yang, Phys. Rev. C 67,034902 (2003). * [22] Rudolph C, Tan Zhi-guang. Phys. Rev. C 72, 024908 (2005). * [23] Hwa Rudolph C. Hwa and C. B. Yang, Phys. Rev. C 70, 024905 (2004). * [24] R. L. Thews arXiv:hep-ph/0609121. * [25] Zi-wei Lin and D nes Moln r. Phys. Rev. C 68, 044901 (2003). * [26] V. Greco and C. M. Ko, Phys. Rev. C 68, 034904 (2003). * [27] V. Greco, C. M. Ko, Phys. Rev. C 71, 041901(R)(2005). * [28] Peter Kolb Phys. Rev. C 69, 051901(2004). * [29] Feng-lan Shao, Tao Yao and Qu-bing Xie,Phys. Rev. C 75, 034904 (2007). * [30] Q. B. Xie and X. M.Liu, Phys. Rev. D 38, 2169 (1988). * [31] T. Yao, Q. b. Xie and F. l. Shao, arXiv:nucl-th/0606033. * [32] K. G. Sarkisyn Edward and S. Sakharov Alexander, arXiv:hep-ph/0410324. * [33] K. G. Sarkisyn Edward and S. Sakharov Alexander, AIP Conf.Proceedings. 828, p. 35 (2006). * [34] Qu-bing Xie and Xi-ming Liu, HIGH ENERGY PHYSICS AND NUCLEAR PHYSICS 11, 2(1987)(in Chinese). * [35] Lian-shou Liu and Ta-chung meng, Phys. Rev. D 27, 2640 (1983). * [36] T. Sjostrand, P. Eden and C. Friberg et al. Comput.Phys.Commun. 135, 238 (2001). * [37] L. D. Landau and Akad. Izv, Nauk SSSR Ser. Fiz. 17: 51 (1953). * [38] S. Z. Belen’kij and L. D. Landau, Usp. Fiz. Nauk. 56, 309 (1955). * [39] B. Mohanty and J. Alam, Phys. Rev. C 68: 064903 (2003). * [40] Fred Cooper and Graham Frye Phys. Rev. D 11, 192 (1975) * [41] BRAHMS Collaboration, ( I. G. Bearden et al.), Phys. Rev. Lett. 94, 162301 (2005). * [42] F. l. Shao, Q. b. Xie and Q. Wang, Phys. Rev. C 71, 044903 (2005). * [43] Q. Wang and Q. B. Xie, J. Phys. G, 21, 897 (1995). * [44] M. Hofmann, M. Bleicher and S. Scherer et al., Phys. Lett. B 478, 161 (2000). * [45] A. Bialas and A. Bzdak, arXiv:hep-ph/0407.3720.

arxiv-papers

2024-09-04T02:48:53.861324

0802.2637

# Topological Black Holes in Lovelock-Born-Infeld Gravity M. H. Dehghani1,2111email address: mhd@shirazu.ac.ir, N. Alinejadi1 and S. H. Hendi3222email address: hendi@mail.yu.ac.ir 1Physics Department and Biruni Observatory, College of Sciences, Shiraz University, Shiraz 71454, Iran 2Research Institute for Astrophysics and Astronomy of Maragha (RIAAM), Maragha, Iran 3 Physics Department, College of Sciences, Yasouj University, Yasouj 75914, Iran ###### Abstract In this paper, we present topological black holes of third order Lovelock gravity in the presence of cosmological constant and nonlinear electromagnetic Born-Infeld field. Depending on the metric parameters, these solutions may be interpreted as black hole solutions with inner and outer event horizons, an extreme black hole or naked singularity. We investigate the thermodynamics of asymptotically flat solutions and show that the thermodynamic and conserved quantities of these black holes satisfy the first law of thermodynamic. We also endow the Ricci flat solutions with a global rotation and calculate the finite action and conserved quantities of these class of solutions by using the counterterm method. We compute the entropy through the use of the Gibbs- Duhem relation and find that the entropy obeys the area law. We obtain a Smarr-type formula for the mass as a function of the entropy, the angular momenta, and the charge, and compute temperature, angular velocities, and electric potential and show that these thermodynamic quantities coincide with their values which are computed through the use of geometry. Finally, we perform a stability analysis for this class of solutions in both the canonical and the grand-canonical ensemble and show that the presence of a nonlinear electromagnetic field and higher curvature terms has no effect on the stability of the black branes, and they are stable in the whole phase space. ###### pacs: 04.40.Nr, 04.20.Jb, 04.70.Bw, 04.70.Dy ## I Introduction Over the last few years, several extra-dimensional models have been introduced in an attempt to deal with the hierarchy problem. These models can lead to rather unique and spectacular signatures at Terascale colliders such as the LHC and ILC. In higher dimensions, it is known that the Einstein-Hilbert (EH) Lagrangian, $R$, can only be regarded as the first order term in an effective action, so one may on general grounds expect that as one probes energies approaching the fundamental scale, significant deviations from EH expectations are likely to appear. This motivates one to consider the more general class of gravitational action: $I_{G}=\frac{1}{16\pi}\int d^{n+1}x\sqrt{-g}\mathcal{F}(R,R_{\mu\nu}R^{\mu\nu},R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma}).$ The presence of higher curvature terms can also be seen in the renormalization of quantum field theory in curved spacetime Birrel , or in the construction of low energy effective action of string theory Green . Among the higher curvature gravity theories, the so-called Lovelock gravity is quite special, whose Lagrangian consist of the dimensionally extended Euler densities. This Lagrangian is obtained by Lovelock as he tried to calculate the most general tensor that satisfies properties of Einstein’s tensor in higher dimensions Lovelock . Since the Lovelock tensor contains derivatives of metrics of order not higher than two, the quantization of linearized Lovelock theory is free of ghosts Boulware . Thus, it is natural to study the effects of higher curvature terms on the properties and thermodynamics of black holes. Accepting the nonlinear terms of the invariants constructed by Riemann tensor on the gravity side of the action, it seems natural to add the nonlinear terms in the matter action too. Thus, in the presence of an electromagnetic field, it is worthwhile to apply the action of Born-Infeld BI instead of the Maxwell action. In this paper, we generalize static and rotating black hole solutions of third order Lovelock gravity in the presence of Maxwell field DSh ; DM1 to the case of these solutions in the presence of nonlinear electromagnetic fields. Indeed, it is interesting to explore new black hole solutions in higher curvature gravity and investigate which properties of black holes are peculiar to Einstein gravity, and which are robust features of all generally covariant theories of gravity. The first aim to relate the nonlinear electrodynamics and gravity has been done by Hoffmann Hoffmann . He obtained a solution of the Einstein equations for a pointlike Born-Infeld charge, which is devoid of the divergence of the metric at the origin that characterizes the Reissner-Nordström solution. However, a conical singularity remained there, as it was later objected by Einstein and Rosen. The spherically symmetric solutions in Einstein-Born-Infeld gravity with or without a cosmological constant have been considered by many authors EBI ; Dey , while the rotating solutions of this theory is investigated in Rastegar . Also, these kinds of solutions in the presence of a dilaton field have been introduced in DHSR . The static black hole solutions of Gauss-Bonnet-Born-Infeld gravity have been constructed in Ref. Wil2 , and the rotating solution of this theory has been considered in DH . The out line of our paper is as follows. We present the topological black holes of third order Lovelock gravity in the presence of Born-Infeld field in Sec. II. In Sec. III, we calculate the thermodynamic quantities of asymptotically flat solutions and investigate the first law of thermodynamics. In Sec. IV we introduce the rotating solutions with flat horizon and compute the thermodynamic and conserved quantities of them. We also perform a stability analysis of the solutions both in canonical and grand canonical ensemble. We finish our paper with some concluding remarks. ## II Topological Black Holes The action of third order Lovelock gravity in the presence of nonlinear Born- Infeld electromagnetic field is $I_{G}=\frac{1}{16\pi}\int d^{n+1}x\sqrt{-g}\left(-2\Lambda+\mathcal{L}_{1}+\alpha_{2}\mathcal{L}_{2}+\alpha_{3}\mathcal{L}_{3}+L(F)\right),$ (1) where $\Lambda$ is the cosmological constant, $\alpha_{2}$ and $\alpha_{3}$ are the second and third order Lovelock coefficients, $\mathcal{L}_{1}=R$ is just the Einstein-Hilbert Lagrangian, $\mathcal{L}_{2}=R_{\mu\nu\gamma\delta}R^{\mu\nu\gamma\delta}-4R_{\mu\nu}R^{\mu\nu}+R^{2}$ is the Gauss-Bonnet Lagrangian, $\displaystyle\mathcal{L}_{3}$ $\displaystyle=$ $\displaystyle 2R^{\mu\nu\sigma\kappa}R_{\sigma\kappa\rho\tau}R_{\phantom{\rho\tau}{\mu\nu}}^{\rho\tau}+8R_{\phantom{\mu\nu}{\sigma\rho}}^{\mu\nu}R_{\phantom{\sigma\kappa}{\nu\tau}}^{\sigma\kappa}R_{\phantom{\rho\tau}{\mu\kappa}}^{\rho\tau}+24R^{\mu\nu\sigma\kappa}R_{\sigma\kappa\nu\rho}R_{\phantom{\rho}{\mu}}^{\rho}$ (2) $\displaystyle+3RR^{\mu\nu\sigma\kappa}R_{\sigma\kappa\mu\nu}+24R^{\mu\nu\sigma\kappa}R_{\sigma\mu}R_{\kappa\nu}+16R^{\mu\nu}R_{\nu\sigma}R_{\phantom{\sigma}{\mu}}^{\sigma}-12RR^{\mu\nu}R_{\mu\nu}+R^{3}$ is the third order Lovelock Lagrangian, and $L(F)$ is the Born-Infeld Lagrangian given as $L(F)=4\beta^{2}\left(1-\sqrt{1+\frac{F^{2}}{2\beta^{2}}}\right).$ (3) In the limit $\beta\rightarrow\infty$, $L(F)$ reduces to the standard Maxwell form $L(F)=-F^{2}$, where $F_{\mu\nu}=\partial_{\mu}A_{\nu}-\partial_{\nu}A_{\mu}$. Varying the action (1) with respect to the metric tensor $g_{\mu\nu}$ and electromagnetic vector field $A_{\mu}$ the equations of gravitation and electromagnetic fields are obtained as: $G_{\mu\nu}^{(1)}+\Lambda g_{\mu\nu}+\alpha_{2}G_{\mu\nu}^{(2)}+\alpha_{3}G_{\mu\nu}^{(3)}=\frac{1}{2}g_{\mu\nu}L(F)+\frac{2F_{\mu\lambda}F_{{\nu}}^{\lambda}}{\sqrt{1+\frac{F^{2}}{2\beta^{2}}}},$ (4) $\partial_{\mu}\left(\frac{\sqrt{-g}F^{\mu\nu}}{\sqrt{1+\frac{F^{2}}{2\beta^{2}}}}\right)=0,$ (5) where $G_{\mu\nu}^{(1)}$ is the Einstein tensor, and $G_{\mu\nu}^{(2)}$ and $G_{\mu\nu}^{(3)}$ are the second and third order Lovelock tensors given as Muller : $G_{\mu\nu}^{(2)}=2(R_{\mu\sigma\kappa\tau}R_{\nu}^{\phantom{\nu}\sigma\kappa\tau}-2R_{\mu\rho\nu\sigma}R^{\rho\sigma}-2R_{\mu\sigma}R_{\phantom{\sigma}\nu}^{\sigma}+RR_{\mu\nu})-\frac{1}{2}\mathcal{L}_{2}g_{\mu\nu},$ (6) $\displaystyle G_{\mu\nu}^{(3)}$ $\displaystyle=$ $\displaystyle-3(4R^{\tau\rho\sigma\kappa}R_{\sigma\kappa\lambda\rho}R_{\phantom{\lambda}{\nu\tau\mu}}^{\lambda}-8R_{\phantom{\tau\rho}{\lambda\sigma}}^{\tau\rho}R_{\phantom{\sigma\kappa}{\tau\mu}}^{\sigma\kappa}R_{\phantom{\lambda}{\nu\rho\kappa}}^{\lambda}+2R_{\nu}^{\phantom{\nu}{\tau\sigma\kappa}}R_{\sigma\kappa\lambda\rho}R_{\phantom{\lambda\rho}{\tau\mu}}^{\lambda\rho}$ (7) $\displaystyle-R^{\tau\rho\sigma\kappa}R_{\sigma\kappa\tau\rho}R_{\nu\mu}+8R_{\phantom{\tau}{\nu\sigma\rho}}^{\tau}R_{\phantom{\sigma\kappa}{\tau\mu}}^{\sigma\kappa}R_{\phantom{\rho}\kappa}^{\rho}+8R_{\phantom{\sigma}{\nu\tau\kappa}}^{\sigma}R_{\phantom{\tau\rho}{\sigma\mu}}^{\tau\rho}R_{\phantom{\kappa}{\rho}}^{\kappa}$ $\displaystyle+4R_{\nu}^{\phantom{\nu}{\tau\sigma\kappa}}R_{\sigma\kappa\mu\rho}R_{\phantom{\rho}{\tau}}^{\rho}-4R_{\nu}^{\phantom{\nu}{\tau\sigma\kappa}}R_{\sigma\kappa\tau\rho}R_{\phantom{\rho}{\mu}}^{\rho}+4R^{\tau\rho\sigma\kappa}R_{\sigma\kappa\tau\mu}R_{\nu\rho}+2RR_{\nu}^{\phantom{\nu}{\kappa\tau\rho}}R_{\tau\rho\kappa\mu}$ $\displaystyle+8R_{\phantom{\tau}{\nu\mu\rho}}^{\tau}R_{\phantom{\rho}{\sigma}}^{\rho}R_{\phantom{\sigma}{\tau}}^{\sigma}-8R_{\phantom{\sigma}{\nu\tau\rho}}^{\sigma}R_{\phantom{\tau}{\sigma}}^{\tau}R_{\mu}^{\rho}-8R_{\phantom{\tau}{\sigma\mu}}^{\tau\rho}R_{\phantom{\sigma}{\tau}}^{\sigma}R_{\nu\rho}-4RR_{\phantom{\tau}{\nu\mu\rho}}^{\tau}R_{\phantom{\rho}\tau}^{\rho}$ $\displaystyle+4R^{\tau\rho}R_{\rho\tau}R_{\nu\mu}-8R_{\phantom{\tau}{\nu}}^{\tau}R_{\tau\rho}R_{\phantom{\rho}{\mu}}^{\rho}+4RR_{\nu\rho}R_{\phantom{\rho}{\mu}}^{\rho}-R^{2}R_{\nu\mu})-\frac{1}{2}\mathcal{L}_{3}g_{\mu\nu}.$ Here we want to obtain the $(n+1)$-dimensional static solutions of Eqs. (4) and (5). We assume that the metric has the following form: $ds^{2}=-f(r)dt^{2}+\frac{dr^{2}}{f(r)}+r^{2}d\Omega^{2},$ (8) where $d\Omega^{2}=\left\\{\begin{array}[]{cc}d\theta_{1}^{2}+\sum\limits_{i=2}^{n-1}\prod\limits_{j=1}^{i-1}\sin^{2}\theta_{j}d\theta_{i}^{2}&k=1\\\ d\theta_{1}^{2}+\sinh^{2}\theta_{1}d\theta_{2}^{2}+\sinh^{2}\theta_{1}\sum\limits_{i=3}^{n-1}\prod\limits_{j=2}^{i-1}\sin^{2}\theta_{j}d\theta_{i}^{2}&k=-1\\\ \sum\limits_{i=1}^{n-1}d\phi_{i}^{2}&k=0\end{array}\right\\}$ represents the line element of an $(n-1)$-dimensional hypersurface with constant curvature $(n-1)(n-2)k$ and volume $V_{n-1}$. Using Eq. (5), one can show that the vector potential can be written as $A_{\mu}=-\sqrt{\frac{(n-1)}{2n-4}}\frac{q}{r^{n-2}}\digamma(\eta)\delta_{\mu}^{0},$ (9) where $q$ is an integration constant which is related to the charge parameter and $\eta=\frac{{(n-1)(n-2)q^{2}}}{2\beta^{2}r^{2n-2}}.$ In Eq. (9) and throughout the paper, we use the following abbreviation for the hypergeometric function ${{}_{2}F_{1}}\left({\left[\frac{1}{2},\frac{n-2}{2n-2}\right],\left[\frac{3n-4}{2n-2}\right],-z}\right)=\digamma(z).$ (10) The hypergeometric function $\digamma(\eta){\rightarrow 1}$ as $\eta\rightarrow 0$ ($\beta\rightarrow\infty$) and therefore $A_{\mu}$ of Eq. (9) reduces to the gauge potential of Maxwell field. One may show that the metric function $f(r)=k+\frac{r^{2}}{\alpha}\left(1-g(r)^{1/3}\right),$ (11) $g(r)=1+\frac{3\alpha m}{r^{n}}-\frac{12\alpha\beta^{2}}{n(n-1)}\left[1-\sqrt{1+\eta}-\frac{\Lambda}{2\beta^{2}}+\frac{(n-1)\eta}{(n-2)}\digamma(\eta)\right]$ (12) satisfies the field equations (4) in the special case $\displaystyle\alpha_{2}$ $\displaystyle=$ $\displaystyle\frac{\alpha}{(n-2)(n-3)},$ $\displaystyle\alpha_{3}$ $\displaystyle=$ $\displaystyle\frac{\alpha^{2}}{72(_{\phantom{-}{4}}^{n-2})},$ where $m$ is the mass parameter. Solutions of Gauss-Bonnet gravity are not real in the whole range $0\leq r<\infty$ and one needs a transformation to make them real DH ; DehBord . But, here the metric function $f(r)$ is real in the whole range $0\leq r<\infty$. In order to consider the asymptotic behavior of the solution, we put $m=q=0$ where the metric function reduces to $f(r)=k+\frac{r^{2}}{\alpha}\left[1-\left(1+\frac{6\Lambda\alpha}{n(n-1)}\right)^{1/3}\right].$ (13) Equation (13) shows that the asymptotic behavior of the solution is AdS or dS provided $\Lambda<0$ or $\Lambda>0$. The case of asymptotic flat solutions ($\Lambda=0$) is permitted only for $k=1$. As in the case of black holes of Gauss-Bonnet-Born-Infeld gravity Wil2 ; DH , the above metric given by Eqs. (8), (11) and (12) has an essential timelike singularity at $r=0$. Seeking possible black hole solutions, we turn to looking for the existence of horizons. The event horizon(s), if there exists any, is (are) located at the root(s) of $g^{rr}=f(r)=0$. Denoting the largest real root of $f(r)$ by $r_{+}$, we consider first the case that $f(r)$ has only one real root. In this case $f(r)$ is minimum at $r_{+}$ and therefore $f^{\prime}(r_{+})=0$. That is, $(n-1)k\left[3(n-2)r_{+}^{4}+3(n-4)k\alpha r_{+}^{2}+(n-6)k^{2}\alpha^{2}\right]+12r_{+}^{6}\beta^{2}\left(1-\sqrt{1+\eta_{+}}\right)-6{\Lambda}r_{+}^{6}=0.$ (14) One can find the extremal value of mass, $m_{\mathrm{ext}}$, in terms of parameters of metric function by finding $r_{+}$ from Eq. (14) and inserting it into equation $f(r_{+})=0$. Then, the metric of Eqs. (8), (11) and (12) presents a black hole solution with inner and outer event horizons provided $m>m_{\mathrm{ext}}$, an extreme black hole for $m=m_{\mathrm{ext}}$ [temperature is zero since it is proportional to $f^{\prime}(r_{+})$] and a naked singularity otherwise. It is a matter of calculation to show that $m_{\mathrm{ext}}$ for $k=0$ becomes $m_{\mathrm{ext}}=\frac{2(n-1)q_{\mathrm{ext}}^{2}}{n}\left(\frac{\Lambda(\Lambda-4\beta^{2})}{2(n-1)(n-2)\beta^{2}q_{\mathrm{ext}}^{2}}\right)^{(n-2)/(2n-2)}\digamma(\frac{\Lambda(\Lambda-4\beta^{2})}{4\beta^{4}}).$ (15) The Hawking temperature of the black holes can be easily obtained by requiring the absence of conical singularity at the horizon in the Euclidean sector of the black hole solutions. One obtains ${T}_{+}=\frac{f^{\prime}(r_{+})}{4\pi}=\frac{(n-1)k\left[3(n-2)r_{+}^{4}+3(n-4)k\alpha r_{+}^{2}+(n-6)k^{2}\alpha^{2}\right]+12r_{+}^{6}\beta^{2}\left(1-\sqrt{1+\eta_{+}}\right)-6{\Lambda}r_{+}^{6}}{12\pi(n-1)r_{+}(r_{+}^{2}+k\alpha)^{2}}.$ (16) It is worthwhile to note that $T_{+}$ is zero for $m=m_{\mathrm{ext}}$. ## III Thermodynamics of Asymptotically Flat Black Holes for $k=1$ In this section, we consider the thermodynamics of spherically symmetric black holes which are asymptotically flat. This is due to the fact that only the entropy of asymptotically black holes of Lovelock gravity is well known Myer . Usually entropy of black holes satisfies the so-called area law of entropy which states that the black hole entropy equals one-quarter of the horizon area Bek . One of the surprising and impressive features of this area law of entropy is its universality. It applies to all kinds of black holes and black strings of Einstein gravity Haw . However, in higher derivative gravity the area law of entropy is not satisfied in general fails . It is known that the entropy of asymptotically flat black holes of Lovelock gravity is Myer $S=\frac{1}{4}\sum_{k=1}^{[(n-2)/2]}k\alpha_{k}\int d^{n-1}x\sqrt{\tilde{g}}\tilde{\mathcal{L}}_{k-1},$ (17) where the integration is done on the $(n-1)$-dimensional spacelike hypersurface of the Killing horizon, $\tilde{g}_{\mu\nu}$ is the induced metric on it, $\tilde{g}$ is the determinant of $\tilde{g}_{\mu\nu}$, and $\tilde{\mathcal{L}}_{k}$ is the $k$th order Lovelock Lagrangian of $\tilde{g}_{\mu\nu}$. Thus, the entropy for asymptotically flat black holes in third order Lovelock gravity is $S=\frac{1}{4}\int d^{n-1}x\sqrt{\tilde{g}}\left(1+2\alpha_{2}\tilde{R}+3\alpha_{3}(\tilde{R}_{\mu\nu\sigma\kappa}\tilde{R}^{\mu\nu\sigma\kappa}-4\tilde{R}_{\mu\nu}\tilde{R}^{\mu\nu}+\tilde{R}^{2})\right),$ (18) where $\tilde{R}_{\mu\nu\rho\sigma}$ and $\tilde{R}_{\mu\nu}$ are Riemann and Ricci tensors and $\tilde{R}$ is the Ricci scalar for the induced metric $\tilde{g}_{ab}$ on the $(n-1)$-dimensional horizon. It is a matter of calculation to show that the entropy of black holes is $S=\frac{V_{n-1}}{4}\left(r_{+}^{4}+\frac{2(n-1)}{n-3}\alpha r_{+}^{2}+\frac{n-1}{n-5}\alpha^{2}\right)r_{+}^{n-5}.$ (19) The charge of the black hole can be found by calculating the flux of the electric field at infinity, yielding $Q=\frac{V_{n-1}}{4\pi}\sqrt{\frac{(n-1)(n-2)}{2}}q.$ (20) The electric potential $\Phi$, measured at infinity with respect to the horizon, is defined by $\Phi=A_{\mu}\chi^{\mu}\left|{}_{r\rightarrow\infty}-A_{\mu}\chi^{\mu}\right|_{r=r_{+}},$ (21) where $\chi=\partial/\partial t$ is the null generator of the horizon. One finds $\Phi=\sqrt{\frac{(n-1)}{2(n-2)}}\frac{q}{r_{+}^{n-2}}\digamma(\eta_{+}).$ (22) The ADM (Arnowitt-Deser-Misner) mass of black hole can be obtained by using the behavior of the metric at large $r$. It is easy to show that the mass of the black hole is $M=\frac{V_{n-1}}{16\pi}\left(n-1\right)m.$ (23) We now investigate the first law of thermodynamics. Using the expression for the entropy, the charge, and the mass given in Eqs. (19), (20) and (23), and the fact that $f(r_{+})=0$, one obtains $\displaystyle M(S,Q)$ $\displaystyle=$ $\displaystyle\frac{\left(n-1\right)}{16\pi}\left\\{\frac{2r_{+}^{n}}{n(n-1)}\left(2\beta^{2}\left[1-\sqrt{1+\Im}+\frac{(n-1)\Im}{(n-2)}\digamma(\Im)\right]-\Lambda\right)\right.$ (24) $\displaystyle\left.-r_{+}^{n-2}+\alpha r_{+}^{n-4}-\frac{\alpha^{2}r_{+}^{n-6}}{3}\right\\},$ where $\Im=\frac{16\pi^{2}Q^{2}}{\beta^{2}r_{+}^{2n-2}}.$ In Eq. (24) $r_{+}$ is the real root of Eq. (19) which is a function of $S$. One may then regard the parameters $S$ and $Q$ as a complete set of extensive parameters for the mass $M(S,Q)$ and define the intensive parameters conjugate to them. These quantities are the temperature and the electric potential $T=\left(\frac{\partial M}{\partial S}\right)_{Q},\ \ \ \ \Phi=\left(\frac{\partial M}{\partial Q}\right)_{S}.$ (25) Computing $\partial M/\partial r_{+}$ and $\partial S/\partial r_{+}$ and using the chain rule, it is easy to show that the intensive quantities calculated by Eq. (25) coincide with Eqs. (16) and (22), respectively. Thus, the thermodynamic quantities calculated in Eqs. (16) and (22) satisfy the first law of thermodynamics, $dM=TdS+\Phi dQ.$ (26) ## IV Thermodynamics of Asymptotically AdS Rotating Black Branes with Flat Horizon Now, we want to endow our spacetime solution (8) for $k=0$ with a global rotation. These kinds of rotating solutions in Einstein gravity have been introduced in Lem . In order to add angular momentum to the spacetime, we perform the following rotation boost in the $t-\phi_{i}$ planes $t\mapsto\Xi t-a_{i}\phi_{i},\hskip 14.22636pt\phi_{i}\mapsto\Xi\phi_{i}-\frac{a_{i}}{l^{2}}t$ (27) for $i=1...[n/2]$, where $[x]$ is the integer part of $x$. The maximum number of rotation parameters is due to the fact that the rotation group in $n+1$ dimensions is $SO(n)$ and therefore the number of independent rotation parameters is $[n/2]$. Thus the metric of an asymptotically AdS rotating solution with $p\leq[n/2]$ rotation parameters for flat horizon can be written as $\displaystyle ds^{2}$ $\displaystyle=$ $\displaystyle-f(r)\left(\Xi dt-{{\sum_{i=1}^{p}}}a_{i}d\phi_{i}\right)^{2}+\frac{r^{2}}{l^{4}}{{\sum_{i=1}^{p}}}\left(a_{i}dt-\Xi l^{2}d\phi_{i}\right)^{2}$ (28) $\displaystyle\ \text{ }+\frac{dr^{2}}{f(r)}-\frac{r^{2}}{l^{2}}{\sum_{i<j}^{p}}(a_{i}d\phi_{j}-a_{j}d\phi_{i})^{2}+r^{2}{{\sum_{i=p+1}^{n-1}}}d\phi_{i},$ where $\Xi=\sqrt{1+\sum_{i}^{k}a_{i}^{2}/l^{2}}$. Using Eq. (5), one can show that the vector potential can be written as $A_{\mu}=-\sqrt{\frac{(n-1)}{2n-4}}\frac{q}{r^{n-2}}\digamma(\eta)\left(\Xi\delta_{\mu}^{0}-\delta_{\mu}^{i}a_{i}\right)\text{(no sum on }i\text{)}.$ (29) One can obtain the temperature and angular momentum of the event horizon by analytic continuation of the metric. One obtains ${T}_{+}{=}\frac{f^{\prime}(r_{+})}{4\pi\Xi}={{\frac{r_{+}}{2(n-1)\pi\Xi}}}\left(2\beta^{2}(1-\sqrt{1+\eta_{+}})-{\Lambda}\right),$ (30) $\Omega_{i}=\frac{a_{i}}{\Xi l^{2}},$ (31) where $\eta_{+}=\eta(r=r_{+})$. Next, we calculate the electric charge and potential of the solutions. The electric charge per unit volume $V_{n-1}$ can be found by calculating the flux of the electric field at infinity, yielding $Q=\frac{1}{4\pi}\sqrt{\frac{(n-1)(n-2)}{2}}\,\,\Xi q.$ (32) Using Eq. (21) and the fact that $\chi=\partial_{t}+{\sum_{i}^{k}}\Omega_{i}\partial_{\phi_{i}}$ is the null generator of the horizon, the electric potential $\Phi$ is obtained as $\Phi=\sqrt{\frac{(n-1)}{2(n-2)}}\frac{q}{\Xi r_{+}^{n-2}}\digamma(\eta_{+}).$ (33) ### IV.1 Conserved quantities of the solutions Here, we calculate the action and conserved quantities of the black brane solutions. In general the action and conserved quantities of the spacetime are divergent when evaluated on the solutions. A systematic method of dealing with this divergence for asymptotically AdS solutions of Einstein gravity is through the use of the counterterms method inspired by the anti-de Sitter conformal field theory (AdS/CFT) correspondence Mal . For asymptotically AdS solutions of Lovelock gravity with flat boundary, $\widehat{R}_{abcd}(\gamma)=0$, the finite action is DBS ; DM1 $I=I_{G}+\frac{1}{8\pi}\int_{\partial\mathcal{M}}d^{n}x\sqrt{-\gamma}\left\\{L_{1b}+\alpha_{2}L_{2b}+\alpha_{3}L_{3b}\right\\}+\frac{1}{8\pi}\int_{\partial\mathcal{M}}d^{n}x\sqrt{-\gamma}\left(\frac{n-1}{L}\right),$ (34) where $L$ is $\displaystyle L$ $\displaystyle=$ $\displaystyle\frac{15l^{2}\sqrt{\alpha(1-\lambda)}}{5l^{2}+9\alpha-l^{2}\lambda^{2}-4l^{2}\lambda},$ (35) $\displaystyle\lambda$ $\displaystyle=$ $\displaystyle(1-\frac{3\alpha}{l^{2}})^{1/3}.$ (36) One may note that $L$ reduces to $l$ as $\alpha$ goes to zero. The first integral in Eq. (34) is a boundary term which is chosen such that the variational principle is well defined. In this integral $L_{1b}=K$, $L_{2b}=2(J-2\widehat{G}_{ab}^{(1)}K^{ab})$ and $L_{3b}=3(P-2\widehat{G}_{ab}^{(2)}K^{ab}-12\widehat{R}_{ab}J^{ab}+2\widehat{R}J-4K\widehat{R}_{abcd}K^{ac}K^{bd}-8\widehat{R}_{abcd}K^{ac}K_{e}^{b}K^{ed}),$ where $\gamma_{\mu\nu}$ and $K$ are induced metric and trace of extrinsic curvature of boundary, $\widehat{G}_{ab}^{(1)}$ and $\widehat{G}_{ab}^{(2)}$ are the $n$-dimensional Einstein and second order Lovelock tensors (Eq. (6)) of the metric $\gamma_{ab}$ and $J$ and $P$ are the trace of $J_{ab}=\frac{1}{3}(2KK_{ac}K_{b}^{c}+K_{cd}K^{cd}K_{ab}-2K_{ac}K^{cd}K_{db}-K^{2}K_{ab}),$ (37) and $\displaystyle P_{ab}$ $\displaystyle=$ $\displaystyle\frac{1}{5}\\{[K^{4}-6K^{2}K^{cd}K_{cd}+8KK_{cd}K_{e}^{d}K^{ec}-6K_{cd}K^{de}K_{ef}K^{fc}+3(K_{cd}K^{cd})^{2}]K_{ab}$ (38) $\displaystyle-(4K^{3}-12KK_{ed}K^{ed}+8K_{de}K_{f}^{e}K^{fd})K_{ac}K_{b}^{c}-24KK_{ac}K^{cd}K_{de}K_{b}^{e}$ $\displaystyle+(12K^{2}-12K_{ef}K^{ef})K_{ac}K^{cd}K_{db}+24K_{ac}K^{cd}K_{de}K^{ef}K_{bf}\\}.$ Using Eqs. (1) and (34), the finite action per unit volume $V_{n-1}$ can be calculated as $I=-\frac{1}{T_{+}}\left\\{\frac{r_{+}^{n}}{16\pi l^{2}}-\frac{r_{+}^{n}\beta^{2}(\sqrt{1+\eta_{+}}-1)}{4n(n-1)\pi}+\frac{(n-1)q^{2}}{8n\pi r_{+}^{(n-2)}}\digamma(\eta_{+})\right\\}.$ (39) Using the Brown-York method Brown , the finite energy-momentum tensor is $T^{ab}=\frac{1}{8\pi}\\{(K^{ab}-K\gamma^{ab})+2\alpha_{2}(3J^{ab}-J\gamma^{ab})+3\alpha_{3}(5P^{ab}-P\gamma^{ab})+\frac{n-1}{L}\gamma^{ab}\ \\},$ (40) and the conserved quantities associated with the Killing vectors $\partial/\partial t$ and $\partial/\partial\phi^{i}$ are $\displaystyle M$ $\displaystyle=$ $\displaystyle\frac{1}{16\pi}m\left(n\Xi^{2}-1\right),$ (41) $\displaystyle J_{i}$ $\displaystyle=$ $\displaystyle\frac{1}{16\pi}n\Xi ma_{i},$ (42) which are the mass and angular momentum of the solution. Now using Gibbs-Duhem relation $S=\frac{1}{T}(M-Q\Phi-{{\sum_{i=1}^{k}}}\Omega_{i}J_{i})-I,$ (43) and Eqs. (33), (39) and (41)-(42) one obtains $S=\frac{\Xi}{4}r_{+}^{n-1}$ (44) for the entropy per unit volume $V_{n-1}$. This shows that the entropy obeys the area law for our case where the horizon is flat. ### IV.2 Stability of the solutions Calculating all the thermodynamic and conserved quantities of the black brane solutions, we now check the first law of thermodynamics for our solutions with flat horizon. We obtain the mass as a function of the extensive quantities $S$, $\mathbf{J}$, and $Q$. Using the expression for charge mass, angular momenta and entropy given in Eqs. (32), (41), (42), (44) and the fact that $f(r_{+})=0$, one can obtain a Smarr-type formula as $M(S,\mathbf{J},Q)=\frac{(nZ-1)J}{nl\sqrt{Z(Z-1)}},$ (45) where $J=\left|\mathbf{J}\right|=\sqrt{\sum_{i}^{k}J_{i}^{2}}$ and $Z=\Xi^{2}$ is the positive real root of the following equation: $\frac{Z^{1/2(n-1)}}{\sqrt{Z-1}}=\frac{\left[4l^{2}\beta^{2}\left(1-\sqrt{1+\eta_{+}}\right)+n(n-1)\right]S^{n/(n-1)}}{(n-1)\pi lJ}+\frac{4\pi lQ^{2}\digamma(\frac{\pi^{2}Q^{2}}{\beta^{2}S^{2}})}{(n-2)J(4S)^{(n-2)/(n-1)}}.$ (46) One may then regard the parameters $S$, $J_{i}$’s, and $Q$ as a complete set of extensive parameters for the mass $M(S,\mathbf{J},Q)$ and define the intensive parameters conjugate to them. These quantities are the temperature, the angular velocities, and the electric potential $T=\left(\frac{\partial M}{\partial S}\right)_{J,Q},\ \ \Omega_{i}=\left(\frac{\partial M}{\partial J_{i}}\right)_{S,Q},\ \ \Phi=\left(\frac{\partial M}{\partial Q}\right)_{S,J}.$ (47) Straightforward calculations show that the intensive quantities calculated by Eq. (47) coincide with Eqs. (30), (31) and (33). Thus, these quantities satisfy the first law of thermodynamics: $dM=TdS+{{{\sum_{i=1}^{k}}}}\Omega_{i}dJ_{i}+\Phi dQ.$ Finally, we investigate the local stability of charged rotating black brane solutions of third order Lovelock gravity in the presence of nonlinear electrodynamic Born-Infeld field in the canonical and grand canonical ensembles. In the canonical ensemble, the positivity of the heat capacity $C_{\mathbf{J},Q}=T_{+}/(\partial^{2}M/\partial S^{2})_{\mathbf{J},Q}$ and therefore the positivity of $(\partial^{2}M/\partial S^{2})_{\mathbf{J},Q}$ is sufficient to ensure the local stability. Using the fact that ${{}_{2}F_{1}}\left(\left[{\frac{3}{2},\frac{3{n-4}}{{2n-2}}}\right],\left[{\frac{5{n-6}}{{2n-2}}}\right],-z\right)=\frac{(3n-4)}{(n-1)z}\left\\{\digamma(\frac{\pi^{2}Q^{2}}{\beta^{2}S^{2}})-\frac{1}{\sqrt{1+z}}\right\\},$ it is easy to show that $\displaystyle\frac{\partial^{2}M}{\partial S^{2}}=\frac{2[(n-1)(n-2)^{2}q^{2}-\Lambda r_{+}^{(2n-2)}\sqrt{1+\eta_{+}}+2\beta^{2}r_{+}^{(2n-2)}\left(\sqrt{1+\eta_{+}}-1\right)]}{(n-1)^{2}\pi\Xi^{2}r_{+}^{(3n-4)}\sqrt{1+\eta_{+}}}-$ $\displaystyle\frac{8(\Xi^{2}-1)\Lambda\left(-r_{+}^{2n-2}\sqrt{1+\eta_{+}}(\frac{n(n-1)}{2}+2\beta^{2}l^{2})+(n-1)(n-2)\beta l^{2}q^{2}+2\beta^{3}l^{2}r_{+}^{(2n-2)}\right)^{2}}{\pi ml^{2}\beta^{2}\Xi^{2}(n-1)^{4}(4\Xi^{2}+1)(1+\eta_{+})r^{(4n-6)}}.$ (48) Both of the two terms of Eq. (48) are positive, and therefore the condition for thermal equilibrium in the canonical ensemble is satisfied. In the grand canonical ensemble, the positivity of the determinant of the Hessian matrix of $M(S,Q,\mathbf{J})$ with respect to its extensive variables $X_{i}$, $\mathbf{H}_{X_{i}X_{j}}^{M}=\left(\partial^{2}M/\partial X_{i}\partial X_{j}\right)$, is sufficient to ensure the local stability. It is a matter of calculation to show that the determinant of $\mathbf{H}_{S,Q,\mathbf{J}}^{M}$ is: $\left|\mathbf{H}_{SJQ}^{M}\right|=\frac{64\pi\left(2(n-2)\beta^{2}\eta_{+}-\Lambda\sqrt{1+\eta_{+}}+2\beta^{2}\left(\sqrt{1+\eta_{+}}-1\right)\right)}{(n-2)(n-1)^{3}ml^{2}\Xi^{6}r_{+}^{2(n-2)}\left[(n-2)\Xi^{2}+1\right]\sqrt{1+\eta_{+}}}\digamma({\eta}_{+})+\frac{8(n-1)\Xi}{r_{+}}T_{+}.$ (49) Equation (49) shows that the determinant of the Hessian matrix is positive, and therefore the solution is stable in the grand canonical ensemble too. The stability analysis given here shows that the higher curvature and nonlinear Maxwell terms in the action have no effect on the stability of black holes with flat horizon, and these kinds of black holes are thermodynamically stable as in the case of toroidal black holes of Einstein-Maxwell gravity Lem2 . This phase behavior is also commensurate with the fact that there is no Hawking- Page transition for a black object whose horizon is diffeomorphic to $\mathbb{R}^{p}$ and therefore the system is always in the high temperature phase Wit2 . ## V CLOSING REMARKS In this paper we considered both the nonlinear scalar terms constructing from the curvature tensor and electromagnetic field tensor in gravitational action, which are on similar footing with regard to the string corrections on gravity and electrodynamic sides. We presented static topological black hole solutions of third order Lovelock gravity in the presence of Born-Infeld gravity, which are asymptotically AdS for negative cosmological constant, dS for positive $\Lambda$. For the case of solutions with positive curvature horizon ($k=1$), one can also have asymptotically flat solutions provided $\Lambda=0$. The topological solutions obtained in this paper may be interpreted as black holes with two inner and outer event horizons for $m>m_{\mathrm{ext}}$ , extreme black holes for $m=m_{\mathrm{ext}}$ or naked singularity otherwise. We found that these solutions reduce to the solutions of Einstein-Born-Infeld gravity as the Lovelock coefficients vanish, and reduce to the solutions of third order Lovelock gravity in the presence of Maxwell field as $\beta$ goes to infinity DSh . We consider thermodynamics of asymptotically flat solutions and found that the first law of thermodynamics is satisfied by the conserved and thermodynamic quantities of the black hole. We also consider the rotating solution with flat horizon and computed the action and conserved quantities of it through the use of counterterm method. We found that the entropy obeys the area law for black branes with flat horizon. We obtained a Smarr-type formula for the mass of the black brane as a function of the entropy, the charge, and the angular momenta, and found that the conserved and thermodynamics quantities satisfy the first law of thermodynamics. We also studied the phase behavior of the $(n+1)$-dimensional rotating black branes in third order Lovelock gravity and showed that there is no Hawking-Page phase transition in spite of the angular momenta of the branes and the presence of a nonlinear electromagnetic field. Indeed, we calculated the heat capacity and the determinant of the Hessian matrix of the mass with respect to $S$, $\mathbf{J}$, and $Q$ of the black branes and found that they are positive for all the phase space, which means that the brane is locally stable for all the allowed values of the metric parameters. ###### Acknowledgements. This work has been supported by Research Institute for Astrophysics and Astronomy of Maragha. ## References * (1) N. D. Birrell and P. C. W. Davies, Quantum Fields in Curved Space, (Cambridge University Press, Cambridge, England, 1982). * (2) M. B. Greens, J. H. Schwarz and E. Witten, Superstring Theory, (Cambridge University Press, Cambridge, England, 1987); D. Lust and S. Theusen, Lectures on String Theory, (Springer, Berlin, 1989); J. Polchinski, String Theory, (Cambridge University Press, Cambridge, England, (1998). * (3) D. Lovelock, J. Math. Phys. 12, 498 (1971); N. Deruelle and L. Farina-Busto, Phys. Rev. D 41, 3696 (1990); G. A. MenaMarugan, ibid. 46, 4320 (1992); 4340 (1992). * (4) D. G. Boulware and S. Deser, Phys. Rev. Lett. 55, 2656 (1985). * (5) M. Born and L. Infeld, Proc. Roy. Soc. Lond. A 144 (1934) 425. * (6) M. H. Dehghani and M. Shamirzaie, Phys. Rev. D 72, 124015 (2005). * (7) M. H. Dehghani and R. B. Mann, Phys. Rev. D 73, 104003 (2006). * (8) B. Hoffmann, Phys. Rev. 47, 877 (1935). * (9) M. Demianski, Found. Phys. 16, 187 (1986); H. P. de Oliveira, Class. Quantum Grav. 11, 1469 (1994); G. W. Gibbons and D. A. Rasheed, Nucl. Phys. B 454, 185 (1995). * (10) R. G. Cai, D. W. Pang and A. Wang, Phys. Rev. D 70, 124034 (2004); I. P. Neupane, ibid. 67, 061501 (2003); I. P. Neupane, ibid. 69, 084011 (2004); T. K. Dey, Phys. Lett. B 595, 484 (2004). * (11) M. H. Dehghani, H. R. Rastegar Sedehi, Phys. Rev. D 74, 124018 (2006). * (12) M. H. Dehghani, S. H. Hendi, A. Sheykhi and H. R. Rastegar, J. Cos. Astrop. Phys. 0702, 020 (2007). * (13) D. L. Wiltshire, Phys. Lett. B 169, 36 (1986); D. L. Wiltshire, Phys. Rev. D 38, 2445 (1988); M. Aiello, R. Ferraro and G. Giribet, ibid. 70, 104014 (2004). * (14) M. H. Dehghani and S. H. Hendi, Int. J. Mod. Phys. D 16, 1829 (2007). * (15) F. Muller-Hoissen, Phys. Lett. B 163, 106 (1985). * (16) M. H. Dehghani, G. H. Bordbar and M. Shamirzaie, Phys. Rev. D 74, 064023 (2006). * (17) T. Jacobson and R. C. Myers, Phys. Rev. Lett. 70, 3684 (1993); R. M. Wald, Phys. Rev. D 48, R3427, (1993); M. Visser, ibid. 48, 5697 (1993); T. Jacobson, G. Kang and R. C. Myers, ibid. 49, 6587,(1994); V. Iyer and R. M. Wald, ibid. 50, 846 (1994). * (18) J. D. Bekenstein, Phys. Rev. D 7, 2333 (1973); S. W. Hawking, Nature (London) 248, 30 (1974); G. W. Gibbons and S. W. Hawking, Phys. Rev. D 15, 2738 (1977). * (19) C. J. Hunter, Phys. Rev. D 59, 024009 (1998); S. W. Hawking, C. J. Hunter and D. N. Page, ibid. 59, 044033 (1999); R. B. Mann, ibid. 60, 104047 (1999); ibid. 61, 084013 (2000); * (20) M. Lu and M. B. Wise, Phys. Rev. D 47, R3095,(1993); M. Visser, ibid. 48, 583 (1993). * (21) J. P. S. Lemos and V. T. Zanchin, Phys Rev. D 54, 3840 (1996); J. P. S. Lemos, Phys. Lett. B 353, 46 (1995). * (22) J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231. * (23) M. H. Dehghani, N. Bostani and A. Sheykhi, Phys. Rev. D 73, 104013 (2006). * (24) J. D. Brown and J. W. York, Phys. Rev. D 47, 1407 (1993). * (25) C. S. Peca and J. P. S. Lemos, J. Math. Phys. 41, 4783 (2000). * (26) E. Witten, Adv. Theor. Math. Phys. 2, 505 (1998).

arxiv-papers

2024-09-04T02:48:53.866025

0802.2766

Development of bakelite based Resistive Plate Chambers S. Biswasa, S. Bhattacharyab, S. Boseb, S. Chattopadhyaya, S. Sahab, M.K. Sharanb, Y.P. Viyogic aVariable Energy Cyclotron Centre, 1/AF Bidhan Nagar, Kolkata-700 064, India bSaha Institute of Nuclear Physics, 1/AF Bidhan Nagar, Kolkata-700 064, India cInstitute of Physics, Sachivalaya Marg, Bhubaneswar, Orissa-751 005, India Abstract A Comparative study has been performed on Resistive Plate Chambers made of different grades of bakelite paper laminates, produced and commercially available in India. The chambers, operated in the streamer mode using argon : tetrafluroethane : isobutane in 34:59:7 mixing ratio, are tested with cosmic rays for the efficiency and the stability with cosmic rays. A particular grade of bakelite (P-120, NEMA LI-1989 Grade XXX), used for high voltage insulation in humid conditions, was found to give satisfactory performance with stable efficiency of $>$ 96% continuously for more than 110 days. A silicone treatment of the inner surfaces of the bakelite RPC is found to be necessary for operation of the detector. Key words: RPC; Streamer mode; Bakelite; Cosmic rays PACS: 29.40.Cs 1\. Introduction The Resistive Plate Chambers (RPCs), first developed by Santonico et al. [1] using bakelite and by Yu. N. Pestov et al. [2, 3] and subsequently by others [4], using silicate glass are used extensively in high energy physics experiments. The RPCs are being considered for the following reasons a) relatively low cost of materials used in making RPCs, b) robust fabrication procedure and handling and c) excellent time and position resolution. Primarily used for generating faster trigger for muon detection [5], time of flight (TOF) [6, 7] measurement, and tracking capabilities in multi layer configurations, they are successfully used in BELLE [8], BaBar [9], BESIII [10], and several upcoming LHC experiments (ATLAS,CMS etc.) [11, 12]. RPCs are used in neutrino experiments like OPERA where its excellent time resolution and tracking capabilities are exploited [13]. The RPCs are also being explored for use in PET imaging with TOF-PET [14], detection of $\gamma$-rays [15] and neutrons [16, 17] over a large area. The RPCs are made up of high resistive plates (e.g. glass, bakelite, ceramics etc.) as electrodes, which help to contain the discharge created by the passage of a charged particle or an ionizing radiation in a gas volume, and pick-up strips are used to collect the resulting signals. Typical time resolution for a single gap RPC is $\sim$ 1-2 ns. By reducing the gaps between the electrodes or by using multi-gap configuration, time resolution in such a detector can be reduced to $<$ 100 ps [18, 19]. The RPCs are operated in two modes, viz., the proportional mode and the streamer mode [20]. Over the years, one of the main concerns with the use of RPCs is their long term stability. In the proportional mode, a small amount of charge is produced in the gas, which allows the RPC to recover in a relatively shorter time to handle high counting rates ($\sim$ 1 KHz/cm2). Ageing effects caused by the accumulated charge is also relatively less in this mode. In the streamer mode, the amount of charge produced is considerably larger creating induced signals of larger magnitude. But, the recovery time is larger and the irreversible damage caused by the accumulated charge reduces the life of the RPC. However, several remedial measures can be taken to prolong its life under streamer mode of operation. Careful choice of materials, smoothness of the surfaces to avoid localization of excess charges, surface treatment to reduce the surface resistivity or providing alternate leakage path for post-streamer recovery are adopted in the major high energy physics experiments. Prolonged stable operation in streamer mode of the BELLE RPCs, though made of glass, is a testimony to many serious efforts taken for the above cause [8]. The glass-based RPCs are found to be more stable mainly for low rate applications, even though some erosion effects are found for such cases, particularly when these are operated in the streamer mode [20]. This has been attributed to the corroding of the glass surface due to the large charge build up in the streamer mode of operation. However, in the proportional mode of operation, the detectors can be operated for longer period. At the end of nineties, it was found that the RPCs based on bakelites show serious ageing effects reducing the efficiency drastically [21]. Detailed investigations revealed that the use of linseed oil for the surface treatment in such cases was the main reason for this ageing effect [22, 23]. Efforts were subsequently made to look for alternatives to linseed oil treatment, or to develop bakelite sheets which can be used without the application of linseed oil [24]. It has, however, been found that for several ongoing and future applications (e.g. CMS detectors), bakelite based RPCs are chosen as preferred options mainly due to cheaper cost of fabrication. In the proposed India-based Neutrino Observatory (INO), the RPCs have been chosen as the prime active detector for muon detection in an Iron Calorimeter (ICAL) [25]. As proposed presently, ICAL is a sampling calorimeter consisting of 140 layers of magnetized iron, each of 60 mm thickness, using RPCs of 2m $\times$ 2m area as active media sandwiched between them. A 50 Kton ICAL is expected to consist of about 27000 RPC modules. For ICAL RPCs, main design criteria are (a) good position resolution, (b) good timing resolution (c) ease of fabrication in large scale with modular structure and most importantly (d) low cost. Detailed R & D are being performed on glass RPCs for this application. In this article, we report a parallel effort on building and testing of the RPC modules using the bakelite obtained from the local industries in India. The aim of the study is to achieve stable performance of the RPC detector for prolonged operation. The paper is organized as follows. In the next section, we describe the method of assembly of the RPC modules. The section 3 contains measurements of the bulk resistivity of the bakelite sheets used in this application while the cosmic ray set-up used in our experiment are discussed in section 4. The results of the study are reported in section 5. 2\. Construction of the RPC modules Figure 1: Schematic diagram of a resistive plate chamber. A schematic view of the assembled RPC modules is shown in the Fig. 1\. Two 300 mm $\times$ 300 mm $\times$ 2 mm bakelite sheets are used as electrodes. The inner surfaces of the two sheets are separated by a 2 mm gap. Uniform separation of the electrodes are ensured by using five button spacers of 10 mm diameter and 2 mm thickness, and edge spacers of 300 mm $\times$ 8 mm $\times$ 2 mm dimension, both being made of polycarbonate. Two nozzles for gas inlet and outlet, also made of polycarbonate, are placed as part of the edge spacers. All the spacers and nozzles are glued to the bakelite sheets using Araldite® epoxy adhesive (grade: AY 103∗∗MP, made by Huntsman A.M., (Europe)). The 2 mm thick active gas gap of the RPC modules are leak-checked using argon and helium sniffer probes. The edges of the bakelite sheets are sealed by applying a layer of the epoxy adhesive to prevent permeation of moisture. After proper cleaning, a graphite coating is made on the outer surfaces of bakelite sheets to distribute the applied voltage uniformly over the entire RPC. A gap of 10 mm from the edges to the graphite layer is maintained to avoid external sparking. The surface resistivity varies from 500 K$\Omega$/$\Box$ to 1 M$\Omega$/$\Box$ for different samples. The graphite coating, applied by using a spray gun, however, results in a non-uniformity (less than 20%) for a particular coated surface. Two small copper foils $\sim$ 20 $\mu$m thick are pasted by kapton tape on both the outer surfaces for the application of high voltage. The HV connectors are soldered on these copper strips. Equal HVs are applied on both the surfaces. In order to collect the accumulated induced charges, pick-up strips are placed above the graphite coated surfaces with minimum air-gap. The pick-up strips are made of copper (20 $\mu$m thick), pasted on one side of 10 mm thick foam. The area of each strip is 300 mm $\times$ 30 mm with a separation of 2 mm between two adjacent strips. The pick-up strips are covered with 100 $\mu$m thick kapton foils to insulate them from the graphite layers. The ground plane, made of aluminium, is pasted on the other side of the foam. The signals from different strips are sent through a ribbon cable, followed by RG-174/U coaxial cables using proper impedance matching. The gases used in the RPC are mixtures of Argon, Isobutane and Tetrafluroethane (R-134a) in varying proportion. The gases are pre-mixed, stored in a stainless steel container and sent to the detector using stainless steel tubes. A typical flow rate of 0.4 ml per minute resulting in $\sim$ 3 changes of gap volume per day is maintained by the gas delivery system. A systematic analysis was made for R-134a and Isobutane before use in the system by a Prisma Quadstar 422 Residual Gas Analyzer. The composition was found to be 98.83% for R-134a with 0.75% O2 and 0.41% N2 and 98.93% for Isobutane with 1.07% H2. 3\. Measurement of bulk resistivity of bakelite The bakelite sheets are phenolic resin bonded paper laminates. In the present work, three types of bakelite sheets have been used to build as many modules. They are (a) mechanical grade bakelite (P-1001), (b) Superhylam grade and (c) electrical grade (P-120). The P-1001 and P-120 grade bakelites are manufactured by Bakelite Hylam, India and the Superhylam grade is obtained from the other manufacturer Super Hylam, India. The sheets of P-1001 and P-120 are matt finished whereas superhylam is glossy finished. The P-1001 has good mechanical properties whereas the P-120 has good mechanical and electrical properties under humid conditions prevalent in India. The bulk resistivity of the electrode plates of the RPC is an important parameter [26]. The high resistivity helps in controlling the time resolution, singles counting rate and also prevents the discharge from spreading through the whole gas. We have measured the bulk resistivities of the bakelite sheets via the measurement of the leakage current. Figure 2: The volume resistivity ($\rho$) as a function of the applied voltage for three grades of bakelites. Table 1: Mechanical and electrical properties of different grades of bakelite. Trade | NEMA | BS-2572 | Density | Electrical | Surface | Bulk ---|---|---|---|---|---|--- Name | LI-1989 | Grade | (g/cc) | strength | finish | resistivity | Grade | | | (kV/mm) | | ($\Omega$-cm) P-1001 | X | P1 | 1.38 | 3.5 | Matt | 6.13 $\times$ 1010 Superhylam | - | P2 | 1.72 | 9.5 | Glossy | 1.25 $\times$ 1011 P-120 | XXX | P3 | 1.22 | 9.5 | Matt | 3.67 $\times$ 1012 This measurement is performed at the same place and the same environment where the RPCs have been tested. The test set up is kept in a temperature and humidity controlled room. These two parameters have been monitored during the experiment and are nearly the same around that time. The bulk resistivities of different grade materials at 4 kV are tabulated in Table 1. The volume resistivity($\rho$) vs. voltage(V) characteristics of different grade materials are shown in Fig. 2. It is clear from the figure that the bulk resistivity is considerably higher for the P-120 grade bakelite. For the P-1001 grade, resistivity is much lower and it cannot sustain high voltage above 4 kV. Therefore, the P-1001 grade is not considered further for building up the RPC. The superhylam grade, though having lower bulk resistivity than P-120, would stand high voltages up to 6 kV. This is also considered for the fabrication of RPC detector. 4\. Cosmic ray test setup Figure 3: Schematic representation of the cosmic ray test setup. Fig. 3 shows the schematic of the setup for testing the RPC modules using cosmic rays. Three scintillators, two placed above the RPC plane and one placed below are used for obtaining the trigger from the incidence of the cosmic rays. The coincidence between scintillator I (350 mm $\times$ 250 mm size), scintillator II (350 mm $\times$ 250 mm size) and the finger scintillator(III) (200 mm $\times$ 40 mm size) is taken as the Master trigger. Finally, the signal obtained from the pick-up strip of the chamber is put in coincidence with the master trigger obtained above. This is referred to as the coincidence trigger of the RPC. The width of the finger scintillator is made smaller than the total width of the two adjacent readout strips, thereby needing a correction for dead zones in between two readout strips. The high voltage to the RPC, are applied at the ramping rate of 5 V/s on both the electrodes. The streamer pulses are obtained starting from the high voltage of 5 kV across the RPC. The high voltage is applied to some of the RPCs by using the CAEN Mod.N470 unit and to the others using the CAEN Mod.N471A unit. The leakage current as recorded by the high voltage system is studied. The signals from two consecutive strips covered by the scintillator III are ORed to form the final signal, for the next part of the pulse processing. The Philips Scientific leading edge discriminators (Model 708) are used for the scintillators and the RPC pulses. Various thresholds are used on the discriminators to reduce the noise. For our final results, a threshold of 40 mV is used on the RPC signal. We have used a CAMAC-based data acquisition system LAMPS, developed by Electronics Division, Bhabha Atomic Research Centre, Trombay, India. Counts accumulated in a CAEN (Model 257) scalar over a fixed time period are recorded at regular intervals, and saved in a periodic log database. The temperature and the humidity are monitored at the time of measurement. 5\. Results An important and obvious goal of any RPC detector development is to study the long term stability with high efficiency. In that spirit, the following studies are performed in the cosmic ray test bench of the RPC detectors. The efficiency of the RPC detector, taken as the ratio between the coincidence trigger rates of the RPC and the master trigger rates of the 3-element plastic scintillator telescope as mentioned in sec.4, is first studied by varying the applied HV for each detector. The rates are calculated from data taken over 30 minutes duration for each HV setting. The temperature and humidity during these measurement are recorded to be about 22-25∘C and 63-65% respectively. The average master trigger rate is $\simeq$ 0.005 Hz/cm2. The variation of efficiency with applied HV is shown in the Fig. 4 and that of the singles counting rates with the HV is shown in the Fig. 5. It is seen that for both the bakelite grades, the efficiency has increased from 20% to 75% as HV is ramped up from 6.5 kV to 6.8 kV. The efficiency for the superhylam grade gradually increases and reaches the plateau at $\sim$ 96% from 7.5 kV, while that of the P-120 grade reaches a maximum of $\sim$ 79% at 7.2 kV and then decreases steadily up to $\sim$ 35% as the HV is increased to 9 kV. The singles counting rates in both the cases, however, have increased more or less exponentially with sudden jumps around 6.5-7.0 kV (see Fig. 5), i.e. near the points where the efficiency becomes uniform (in case of superhylam) or starts to decrease (in case of P-120). This possibly indicates the onset of a breakdown regime that recovers in a reasonable time for the superhylam grade RPC but works the other way for the P-120 grade bakelite RPC. It should, however, be noted that the singles counting rate and the leakage current of the superhylam RPC are both larger than those of the P-120 RPC, which are expected on the basis of smaller bulk resistivity of the superhylam grade bakelite. Figure 4: The efficiency as a function of high voltage for two RPCs (silicone coated & uncoated P-120 and Superhylam grade bakelite) obtained with a gas mixture of Argon (34%) + Isobutane (7%) + R-134a (59%). The threshold for the RPCs are set at 40 mV for P-120 and 50mV for superhylam. Figure 5: Singles counting rate as a function of high voltage. In order to investigate the reason for the above phenomena, and taking cue from the fact that superhylam surfaces are glossy finished while the P-120 surfaces are matt finished, we have dismantled the detectors and made surface profile scans over a 5 mm span of the surfaces using DekTak 117 Profilometer. These scans, done also for the P-1001 grade, are shown in Fig. 6. It is clearly seen that the three surfaces have a short range variation (typically $\sim$ 0.1 $\mu$m length scale) and a long range variation (typically $\sim$ 1 $\mu$m length scale). The long range surface fluctuation, which is a measure of non - uniformity, averaged over several scans are: 0.84 $\pm$ 0.12 $\mu$m (P-120), 0.49 $\pm$ 0.17 $\mu$m (superhylam) and 0.88 $\pm$ 0.09 $\mu$m (P-1001). Thus the long range fluctuations,within the limits of experimental uncertainties, are nearly the same. On the other hand the short range fluctuations, a measure of surface roughness, are: 0.64 $\pm$ 0.06 $\mu$m (P-120), 0.17 $\pm$ 0.02 $\mu$m (superhylam), and 0.63 $\pm$ 0.13 $\mu$m (P-1001), and thus indicate a superior surface quality of the superhylam grade. Figure 6: Linear surface profile scans of the three grades of bakelite sheets. To explore a remedial measure, we have applied a thin layer of viscous silicone fluid (chemical formula : [R2SiO]n, where R = organic groups such as methyl, ethyl, or phenyl) [coefficient of viscosity = 5500 cP, manufactured by Metroark Limited, Kolkata, India] on the inner surfaces of the P-120 bakelite sheets. About 1 gm of the fluid is applied over 300 mm $\times$ 300 mm area. Based on the specific gravity (1.02 at 23∘C) of the fluid, the estimated coating thickness would be $\sim$ 10 $\mu$m. This material is chosen for the following reasons: a) very low chemical reactivity with the gases used; b) good thermal stability over a wide temperature range (from -100 to 250 ∘C); c) very good electrical insulator; d) excellent adhesion to most of the solid materials, and e) low vapour pressure, which is essential for stable operation over a reasonable time period. The silicone treated surfaces are kept under infrared lamp for 24 hours to allow the viscous fluid to fill all the micro- crevices on the surface. The reassembled detector is tested at the same set- up. The results of efficiency and singles count rate measurements, shown in the Figs. 4 and 5, indicate a remarkable improvement in the performance of the P-120 detector. The efficiency reaches from 20% to 75% as the HV is increased from 5.7 kV to 6.2 kV, while the singles count rate, as a whole has decreased by a factor of 5\. This indicates quenching of micro-discharge after silicone treatment, which is very much desirable for functioning of the detector. The efficiency in this case reaches $>$ 95% plateau at 7 kV. It is worth noting that surface treatment with insulating / non-polar liquid as a remedial measure was first demonstrated for the BaBar RPCs. However, formation of stalagmites by polymerisation of uncured linseed oil droplets had created conducting paths through the gap, thereby causing irreversible damage to the bakelite plates [22]. The process of linseed oil treatment was later changed by increasing the proportion of eptane as a thinner to produce a thinner coating (10-30 $\mu$m) on the inner surface [27]. Our observation that silicone coating of the inner surfaces aides the proper functioning of our P-120 bakelite RPC detector once again confirms the importance of smooth surface finish of the inner surfaces. To judge the improvement in the overall performance of the RPC detector, we have measured the leakage current through the RPC detector with and without silicone coating and the plot of these as a function of the applied HV as shown in the Fig. 7. Both the plots show a common feature that the current- voltage curves have two distinctly different slopes. While the gas gap behaves as an insulator in the lower range of applied voltage and hence the slope over this span scales as the conductance of the polycarbonate spacers, at higher range of voltage, the gas behaves as a conducting medium due to the formation of the streamers. Therefore, the slope over this range scales as the conductance of the gas gap. It is seen that the slope in the higher range of voltage is much steeper for the RPC without silicone coating and hence it points to the fact that some sort of uncontrolled streamers are being formed in the gas gap causing a degradation of the efficiency. This possibly does not happen in the RPC detector with silicone coating. Figure 7: Current as a function of the applied voltage for RPC made by P-120 grade bakelite. We have also examined the effect of discriminator threshold setting on the efficiency curves of the RPC with silicone treated surfaces. These are plotted in the Fig. 8. It is clear that the efficiency curves do not depend much on the threshold setting from 20 mV to 80 mV, except that the efficiency plateau is marginally higher at the lowest threshold setting of 20 mV. Figure 8: Efficiency versus high voltage for different thresholds for silicone coated P-120 grade bakelite RPC with silicone coating. The effect of humidity on the efficiency curves has also been studied. This measurement has been done at relative humidities of 58% and 67% of the laboratory environment and at the same room temperature of $\sim$ 23∘C. These curves, plotted in the Fig. 9, indicate no effect of humidity on the efficiency. However, the leakage currents, measured simultaneously and plotted in the Fig. 10, are a bit larger at higher humidity. This observation indicates that charge leakage through the exterior surfaces may be contributing more at higher humidity. Figure 9: Efficiency versus high voltage for different humidities for silicone coated P-120 grade bakelite RPC. Figure 10: Current versus high voltage for different humidities for silicone coated P-120 grade bakelite RPC. The long term stability of the bakelite RPCs has been studied using the same cosmic ray test set-up. The coincidence trigger counts of the RPCs and the master trigger counts, accumulated over every 2 hours, have been recorded continuously for more than 6 months. The room temperature and the relative humidity have been controlled to keep them less than 24∘C and 80%, respectively. When the humidity was larger, particularly during the monsoon season, the test set-up was shut down till it came down to 80% and below. The singles count rates of the RPCs have also been recorded simultaneously. The Figs. 11 and 12 depict the variation of efficiency and singles count rates over the above mentioned period for both the grades of RPCs. The superhylam grade RPC has worked with an efficiency of $>$ 95% which remained steady for 25 days, but beyond that, it deteriorated gradually to $\sim$ 86% efficiency within next 13 days. The singles count rate, however, has increased from day one from 1 Hz/cm2 to 10 Hz/cm2 within 10 days, and then it increased slowly over the next 28 days. After that period, the singles count rate shot up to $>$ 30 Hz/cm2. The leakage current gradually increased from 3-4 $\mu$A to $>$ 10 $\mu$A within that period. This RPC was discontinued after 38 days and the silicone coated P-120 grade RPC was then mounted. The efficiency measured was $\sim$ 96% and above and has remained steady for more than 110 days. The singles count rate also has remained steady around 0.1 Hz/cm2. The leakage current was found to be marginally dependent on temperature and humidity, though it has remained steady at $\sim$ 400 nA during the operation. Figure 11: Efficiency as a function of time for two RPC prototypes. Figure 12: The single counting rate as a function of time for the two RPC prototypes. The superhylam grade RPC has also been tested again after a gap of a few months. It has shown the same higher leakage current ($>$ 10 $\mu$A) and lower efficiency ($\sim$ 86%) indicating that some intrinsic breakdown of the bulk material may have taken place. 6\. Conclusion We have made a comparative study of bakelite RPCs made from two different grades of bakelite. The RPC, made of superhylam grade bakelite with melamine coated glossy finished surface is found to have a shorter life. On the other hand, the RPC made from P-120 grade bakelite with matt finished surfaces, which are coated with a thin layer of viscous silicone fluid, are found to work steadily for more than 110 days showing a constant efficiency of $>$ 96% without any degradation. The detector is found to be less immune to variation in humidity which makes it a viable alternative to semiconductive glass based RPC. 7\. Acknowledgement We are thankful to Prof. Naba Kumar Mandal of TIFR, India and Prof. Kazuo Abe of KEK, Japan for their encouragement and many useful suggestions in course of this work. We are also grateful to Dr. C. Bhattacharya, Mr. G.S.N. Murthy, Mr. M.R. Dutta Majumdar, Mr. S.K. Thakur and Mr. S.K. Bose of VECC for their help in the work. We acknowledge the service rendered by Mr. Avijit Das of SINP for surface profile scans of the bakelite sheets used by us. We would like to thank the SINP workshop for making the components of the detectors, and Mr. Ganesh Das of VECC for meticulously fabricating the detectors. Finally we acknowledge the help received from the scientific staff of Electronics Workshop Facility of SINP for building the gas flow control and delivery system of the gas mixing unit used in this study. ## References * [1] R. Santonico, R. Cardarelli, Nucl. Inst. and Meth. 187 (1981) 377. * [2] Yu. N. Pestov, Nucl. Inst. and Meth. 196 (1982) 45. * [3] W. B. Atwood et al., Nucl. Inst. and Meth. 206 (1983) 99. * [4] M. Anelli et al., Nucl. Inst. and Meth. A 300 (1991) 572. * [5] Gy. L. Bencze et al., Nucl. Inst. and Meth. A 340 (1994) 466. * [6] R. Cardarelli et al., Nucl. Inst. and Meth. A 263 (1988) 20. * [7] A. Blanco et al., Nucl. Inst. and Meth. A 513 (2003) 8. * [8] A. Abashian et al.,Nucl. Inst. and Meth. A 479 (2002) 117. * [9] BaBar Technical Design Report, BaBar Collaboration, SLAC Report SLAC-R-95-457, March 1995. * [10] The BESIII Detector, IHEP-BEPCII-SB-13, IHEP, Beijing. * [11] ATLAS Technical Design Report, Muon Spectrometer, CERN/LHCC/97-22, Geneva, 1997. * [12] CMS - Technical Proposal, CERN/LHCC/94-38, December 1994. * [13] M. Guler et al., OPERA, an appearance experiment to search for $\nu_{\mu}$-$\nu_{\tau}$ oscillations in the CNGS beam, CERN/SPSC 2000-028. * [14] M. Couceiro et al., Nucl. Inst. and Meth. A 580 (2007) 915. * [15] P. Camarri, Nucl. Inst. and Meth. A 572 (2007) 476. * [16] M. Abbrescia et al., Nucl. Phys. B (Proc. Suppl.) 125 (2003) 43\. * [17] M. Abbrescia et al., Nucl. Inst. and Meth. A 506 (2003) 101. * [18] P. Fonte et al., Nucl. Inst. and Meth. A 449 (2000) 295. * [19] A. Blanco et al., Nucl. Inst. and Meth. A 535 (2004) 272. * [20] G. Bruno, Eur Phys. J. C 33 (2004) s1032. * [21] J. Va’vra, Nucl. Inst. and Meth. A 515 (2003) 354. * [22] F.Anulli et al., Nucl. Inst. and Meth. A 508 (2003) 128. * [23] F. Anulli et al., Nucl. Inst. and Meth. A 515 (2003) 322. * [24] J. Zhang et al., Nucl. Inst. and Meth. A 540 (2005) 102. * [25] INO Project Report, INO/2006/01, June 2006, $\langle$http://www.imsc.res.in/$\sim$ino/$\rangle$. * [26] G. Bencivenni et al., Nucl. Inst. and Meth. A 332 (1993) 368. * [27] F.Anulli et al., Nucl. Inst. and Meth. A 539 (2005) 155.

arxiv-papers

2024-09-04T02:48:53.874597

0802.2782

OU-HET 599 February 2008 Fiber Bundles and Matrix Models Takaaki Ishii*** e-mail address : ishii@het.phys.sci.osaka-u.ac.jp, Goro Ishiki††† e-mail address : ishiki@het.phys.sci.osaka-u.ac.jp, Shinji Shimasaki‡‡‡ e-mail address : shinji@het.phys.sci.osaka-u.ac.jp and Asato Tsuchiya§§§ e-mail address : tsuchiya@het.phys.sci.osaka-u.ac.jp, address after April 2008 : Department of Physics, Shizuoka University, 836 Ohya, Suruga-ku, Shizuoka 422-8529, Japan Department of Physics, Graduate School of Science Osaka University, Toyonaka, Osaka 560-0043, Japan We investigate relationship between a gauge theory on a principal bundle and that on its base space. In the case where the principal bundle is itself a group manifold, we also study relations of those gauge theories with a matrix model obtained by dimensionally reducing them to zero dimensions. First, we develop the dimensional reduction of Yang-Mills (YM) on the total space to YM- higgs on the base space for a general principal bundle. Second, we show a relationship that YM on an $SU(2)$ bundle is equivalent to the theory around a certain background of YM-higgs on its base space. This is an extension of our previous work [29], in which the same relationship concerning a $U(1)$ bundle is shown. We apply these results to the case of $SU(n+1)$ as the total space. By dimensionally reducing YM on $SU(n+1)$, we obtain YM-higgs on $SU(n+1)/SU(n)\simeq S^{2n+1}$ and on $SU(n+1)/(SU(n)\times U(1))\simeq CP^{n}$ and a matrix model. We show that the theory around each monopole vacuum of YM-higgs on $CP^{n}$ is equivalent to the theory around a certain vacuum of the matrix model in the commutative limit. By combining this with the relationship concerning a $U(1)$ bundle, we realize YM-higgs on $SU(n+1)/SU(n)\simeq S^{2n+1}$ in the matrix model. We see that the relationship concerning a $U(1)$ bundle can be interpreted as Buscher’s T-duality. ###### Contents 1. 1 Introduction and conclusion 2. 2 Typical relationships 3. 3 Dimensional reduction on a principal bundle 4. 4 Extension of the matrix T-duality 1. 4.1 Nontrivial vacua and transformation between patches 2. 4.2 $G=U(1),SU(2)$ 3. 4.3 Example: $S^{7}\rightarrow S^{4}$ 5. 5 Gauge theories on $SU(n+1)(/H)$ and matrix model 1. 5.1 Dimensional reduction of YM theory on a group manifold 2. 5.2 Dimensional reduction of YM theory on $SU(n+1)$ 3. 5.3 Relations among gauge theories on $SU(n+1)/H$ 6. 6 Interpretation as Buscher’s T-duality 7. A Spherical harmonics 8. B Derivation of (4.22) 9. C Group manifold and coset space 10. D Metrics of $SU(3)$, $S^{5}$ and $CP^{2}$ 11. E Fuzzy $CP^{n}$ ## 1 Introduction and conclusion Emergence of space-time is one of the key concepts in matrix models as nonperturbative definition of superstring [1, 2, 3]. This phenomenon was first observed in the relationship between a gauge theory and a matrix model. This is the so-called large N reduction [4]. It states that a large $N$ planar gauge theory is equivalent to the matrix model that is its dimensional reduction to zero dimensions unless the $U(1)^{D}$ symmetry is broken, where $D$ denotes the dimensionality of the original gauge theory. However, the $U(1)^{D}$ symmetry is in general spontaneously broken for $D>2$. There are two improved versions of the large $N$ reduced model that preserve the $U(1)^{D}$ symmetry. One is the quenched reduced model [5, 6, 7, 8]. The other is the twisted reduced model [9], which was later rediscovered in the context of the noncommutative field theories [10]. The T-duality for D-brane effective theories [11], which we call the matrix T-duality in this paper, share the same idea with the large N reduced model. The statement of the matrix T-duality is that $U(N)$ Yang-Mills (YM) on $R^{p}\times S^{1}$ is equivalent to $U(N\times\infty)$ YM-higgs on $R^{p}$ which is a dimensional reduction of $U(N\times\infty)$ YM on $R^{p}\times S^{1}$ if a periodicity (orbifolding) condition is imposed. Also, deconstruction [12] and supersymmetric lattice gauge theories inspired by it [13] are analogs of the matrix T-duality. The above developments are all concerning gauge theories on flat space-time. It is important to understand how gauge theories on curved space-time are realized in matrix models or gauge theories in lower dimensions, because it would lead us to gain some insights into how curved space-time is realized in matrix models as nonperturbative definition of superstring. Note that an interesting approach to the description of curved spacetime by matrices was proposed in [14]. In [15], Takayama and three of the present authors found relationships among the $SU(2|4)$ symmetric theories. Here the $SU(2|4)$ symmetric theories include ${\cal N}=4$ super Yang Mills (SYM) on $R\times S^{3}/Z_{k}$, 2+1 SYM on $R\times S^{2}$ [16] and the plane wave matrix model (PWMM) [17]. These theories are related by dimensional reductions and possess common features: mass gap, discrete spectrum and many discrete vacua. From the gravity duals of those vacua proposed in [18], the following relations between these theories are suggested: the theory around each vacuum of 2+1 SYM on $R\times S^{2}$ is equivalent to the theory around a certain vacuum of PWMM, and the theory around each vacuum of ${\cal N}=4$ SYM on $R\times S^{3}/Z_{k}$ is equivalent to the theory around a certain vacuum of 2+1 SYM on $R\times S^{2}$ with the periodicity imposed. Combining these two equivalences, we can say that the theory around each vacuum of ${\cal N}=4$ SYM on $R\times S^{3}/Z_{k}$ is realized in PWMM. In [15], these equivalences were shown directly on the gauge theory side. The results in [15] not only serve as a nontrivial check of the gauge/gravity correspondence for the $SU(2|4)$ theories, but they are also interesting in the following aspects. Much work has been already done on the realization of the gauge theories on the fuzzy sphere [19, 20, 21, 22] by matrix models [23] and on the monopoles on the fuzzy sphere [24, 25, 26, 28, 27]. Note that the realization of the fuzzy sphere by matrix models can be viewed as an extension of the twisted reduced model to curved space. Here in the relation between 2+1 SYM on $R\times S^{2}$ and PWMM, it was manifestly shown that the continuum limit of concentric fuzzy spheres correspond to multi monopoles. The relation between ${\cal N}=4$ SYM on $R\times S^{3}/Z_{k}$ and 2+1 SYM on $R\times S^{2}$ can be regarded as an extension of the matrix T-duality to that on a nontrivial $U(1)$ bundle, $S^{3}/Z_{k}$, whose base space is $S^{2}$. Furthermore, in [29], we generalized the matrix T-duality to that on an arbitrary $U(1)$ bundle. As an application of these results, in [30], Ohta and the present authors investigated relationships among Chern- Simons theory on a $U(1)$ bundle over a Riemann surface, BF theory with a mass term on the Riemann surface, which is equivalent to two-dimensional Yang-Mills on the Riemann surface, and a matrix model. It was discussed that the former two (topological) field theories associated with topological strings can be realized in the matrix model. The results in [15] also suggests an interesting possibility of a nonperturbative formulation of ${\cal N}=4$ SYM on $R\times S^{3}$ by PWMM, which would lead to a nonperturbative test of the AdS/CFT correspondence. This paper is aimed at further investigation of the above developments concerning the large $N$ reduction and the matrix T-duality on curved space. First, we develop a dimensional reduction of YM on the total space to YM-higgs on the base space for a general principal bundle. This also enables us to dimensionally reduce YM on a group manifold to a matrix model. Second, as an extension of the work [29], in the case in which the fiber is $SU(2)$, we show that YM on the total space is equivalent to a certain vacuum111Throughout this paper, we consider gauge theories on manifolds with the Euclidean signature. Here ‘vacuum’ represents a configuration that gives the global minimum of the classical action. of YM-higgs on the base space with the periodicity imposed. This enables us to realize YM on an $SU(2)^{k}\times U(1)^{l}$ bundle in YM- higgs on its base space. We apply the above results to the case of $SU(n+1)$ as the total space. $SU(n+1)$ is viewed as $SU(n)$ bundle over $SU(n+1)/SU(n)\simeq S^{2n+1}$ or $SU(n)\times U(1)$ bundle over $SU(n+1)/(SU(n)\times U(1))\simeq CP^{n}$, and $SU(n+1)/SU(n)\simeq S^{2n+1}$ is viewed as $U(1)$ bundle over $CP^{n}$. By the dimensional reduction, we obtain YM-higgs on $S^{2n+1}$ and $CP^{n}$ and a matrix model. We find the commutative (continuum) limit of gauge theory on fuzzy $CP^{n}$ [28, 31, 32, 33, 34, 35] realized in the matrix model coincides with YM-higgs on $CP^{n}$. Namely, we show that the theory around each monopole vacuum of YM-higgs on $CP^{n}$ is equivalent to the theory around a certain vacuum of the matrix model. By combing this with the extended matrix T-duality, we realize YM-higgs on $SU(n+1)/SU(n)\simeq S^{2n+1}$ in the matrix model. We also show that the extended matrix T-duality of the $U(1)$ case developed in [29] can be interpreted as Buscher’s T-duality [36]. In the remainder of this section, we describe the organization of the present paper, providing our results in detail, and finally describe some outlook. From the same reasoning as the case of the $SU(2|4)$ symmetric theories, the following relationships among YM on $S^{3}$, YM-higgs on $S^{2}$ and a matrix model hold. These theories are related to each other by dimensional reductions. The theory around each vacuum of YM-higgs on $S^{2}$ is equivalent to the theory around a certain vacuum of the matrix model. YM on $S^{3}$ is equivalent to the theory around a certain vacuum of YM-higgs on $S^{2}$ with the periodicity imposed. Eventually, YM on $S^{3}$ is realized in the matrix model. It can be said that our results in this paper are extension of these relationships. In section 2, we show these relationships in order to illustrate our basic ideas. In section 3, we develop a dimensional reduction on a general principal fiber bundle. We start with YM on the total space, dimensionally reduce the fiber directions and obtain a YM-higgs on the base space. In section 4, we examine a relationship between YM on the total space and YM- higgs on the base space obtained in section 3. In section 4.1, we first examine the transformations of the fields from a local patch to another local patch in YM-higgs on the base space. In section 4.2, using the observation in section 4.1, we show that when the fiber is $U(1)$ or $SU(2)$, YM on the total space is equivalent to the theory around a certain vacuum of YM-higgs on the base space with the periodicity imposed. This vacuum is given by multimonopole configuration on the base space. We already found the $U(1)$ case of this equivalence in [29]. In the $SU(2)$ case, we also use the result in section 2 that YM on $S^{3}$ is realized in the matrix model. As a generalization, we realize YM on $SU(2)^{k}\times U(1)^{l}$ bundle in YM-higgs on its base space. In section 4.3, as an example, we consider $S^{7}$ which is an $SU(2)$ bundle over $S^{4}$. In Fig. 1, we summarize our results in sections 3 and 4. Figure 1: Matrix T-duality for $G=U(1)$, $SU(2)$ In section 5, we examine a series of $SU(n+1)$ symmetric theories. Fig. 2 summarizes our findings in section 5 and their relation to other sections. The case of $n=1$ is nothing but the example discussed in section 2. In this case, YM on $SU(2)$ is the same as YM-higgs on $S^{3}$ because $SU(2)\simeq S^{3}$. In section 5.1, as a special case of section 3, we consider a dimensional reduction of YM on a group manifold $\tilde{G}$ to a coset space $\tilde{G}/H$ where $H$ is a subgroup of $\tilde{G}$. Namely, we view $\tilde{G}$ as an $H$ bundle over $\tilde{G}/H$. By dimensionally reducing the Killing vectors on $\tilde{G}$ to those on $\tilde{G}/H$, we obtain a theory on $\tilde{G}/H$ expressed in terms of the Killing vectors. Then, we show that this theory on $\tilde{G}/H$ is rewritten into YM-higgs on $\tilde{G}/H$ obtained in section 3. In section 5.2, we apply the results in section 5.1 to the case of $\tilde{G}=SU(n+1)$ and obtain a series of theories in Fig. 2 which possess $SU(n+1)$ symmetry. If we take $SU(n)$ as $H$, we obtain YM-higgs on $S^{2n+1}$. Note that the isometry of this $S^{2n+1}$ is not $SO(2n+2)$ but $SU(n+1)$. For $n\geq 2$, it is different from the ordinary $S^{2n+1}$ but homeomorphic to the ordinary one, and is called a squashed $S^{2n+1}$. If we take $SU(n)\times U(1)$ as $H$, we obtain YM-higgs on $CP^{n}$. Finally if we take $SU(n+1)$ itself as $H$, we obtain a matrix model whose action is shown in Fig. 2, where $f_{ABC}$ is the structure constant of the $SU(n+1)$ Lie algebra. As indicated in Fig. 2, these dimensional reductions can also be performed step by step: we obtain YM-higgs on $CP^{n}$ from YM-higgs on $S^{2n+1}$ and the matrix model from YM-higgs on $CP^{n}$. In the case of $n=2$, as an application of the result in section 4.2, we see that YM on $SU(3)$ is equivalent to the theory around a vacuum of YM-higgs on $S^{5}$ with the periodicity imposed ((i) in Fig. 2). Since $S^{2n+1}$ can be viewed as a $U(1)$ bundle over $CP^{n}$, in section 5.3, we show as an application of the results in section 4.2 that the theory around each vacuum of YM-higgs on $S^{2n+1}$ is equivalent to the theory around a vacuum of YM-higgs on $CP^{n}$ with the periodicity imposed ((ii) in Fig. 2). In section 5.3, we show that the theory around each abelian monopole vacuum of YM-higgs on $CP^{n}$ is equivalent to a certain vacuum of the matrix model ((iii) in Fig. 2). Combining these results, we also show that the theory around the trivial vacuum of YM-higgs on $S^{2n+1}$ is realized in the matrix model ((iv) in Fig. 2). YM on $SU(3)$ is realized in YM-higgs on $CP^{2}$ ((v) in Fig. 2). Finally, we make a comment: it follows from the result in section 4 that YM on $SU(n+1)$ is realized in YM-higgs on $SU(n+1)/(SU(2)^{k}\times U(1)^{l})$. Figure 2: A series of theories studied in section 5. In section 6, we discuss how the extended matrix T-dulaity found in [29] and reviewed in section 4.2 is interpreted as Buscher’s T-duality. In appendices A-D, we describe some details. It is an open problem whether YM on $SU(n+1)$ with $n\geq 2$ is realized in the matrix model. Presumably, we need to construct noncommutative counterparts of non-Abelian monopoles of YM-higgs on $S^{2n+1}$ or $CP^{n}$ in the matrix model. Realization of YM on $SU(n+1)$ in the matrix model should enable us to extend the matrix T-duality to the case of $G=SU(n+1)$. Of course, the matrix T-duality for a general $G$ should still be investigated. It is important to see whether the matrix T-duality in the $SU(2)$ case is associated with the nonabelian T-duality discussed within the nonlinear sigma models [37]. It is also relevant to identify the commutative limit of the matrix model consisting of the square of the commutators and the generalized Myers term with the $SU(n+1)$ structure constant which has been examined in [33, 38] and find its higher-dimensional origin. Analysis in this paper is classical. Whether the relationships among the gauge theories we found hold quantum mechanically is a nontrivial and important problem. It should be noted that in the quantum correspondence no orbifolding condition is needed in the matrix T-duality as far as the planar limit is concerned. This is nothing but the large $N$ reduction and enables us to make the size of matrices become finite and play a role of the ultraviolet cutoff. In particular, we expect to give a nonperturbative definition of ${\cal N}=4$ SYM on $R\times S^{3}$ in the planar limit in terms of PWMM [39]. ## 2 Typical relationships In this section, to illustrate our ideas, we describe relationships among YM on $S^{3}$, YM-higgs on $S^{2}$ and a matrix model. These relationships are essentially the same as those among the $SU(2|4)$ symmetric theories found in [15]. We consider $S^{3}$ with radius $2/\mu$ and regard it as the $U(1)$ ($S^{1}$) Hopf bundle on $S^{2}$ with radius $1/\mu$. $S^{3}$ with radius $2/\mu$ is defined by $\displaystyle\\{(w_{1},w_{2})\in C^{2}\>|\>|w_{1}|^{2}+|w_{2}|^{2}=4/\mu^{2}\\}.$ (2.1) The Hopf map $\pi:\>S^{3}\rightarrow CP^{1}\;(S^{2})$ is defined by $\displaystyle(w_{1},w_{2})\rightarrow[(w_{1},w_{2})]\equiv\\{\lambda(w_{1},w_{2})|\lambda\in C\backslash\\{0\\}\\}.$ (2.2) Two patches are introduced on $CP^{1}$: the patch I $(w_{1}\neq 0)$ and the patch II $(w_{2}\neq 0)$. On the patch I the local trivialization is given by $\displaystyle(w_{1},w_{2})\rightarrow\left(\frac{w_{2}}{w_{1}},\frac{w_{1}}{|w_{1}|}\right)\in\mbox{(patch I)}\times U(1),$ (2.3) while on the patch II the local trivialization is given by $\displaystyle(w_{1},w_{2})\rightarrow\left(\frac{w_{1}}{w_{2}},\frac{w_{2}}{|w_{2}|}\right)\in\mbox{(patch II)}\times U(1).$ (2.4) The equation (2.1) is solved as $\displaystyle w_{1}=\frac{2}{\mu}\cos\frac{\theta}{2}\>e^{i\sigma_{1}},\;\;\;w_{2}=\frac{2}{\mu}\sin\frac{\theta}{2}\>e^{i\sigma_{2}},$ (2.5) where $0\leq\theta\leq\pi$ and $0\leq\sigma_{1},\>\sigma_{2}<2\pi$. We put $\displaystyle\varphi=\sigma_{1}-\sigma_{2},\;\;\;\psi=\sigma_{1}+\sigma_{2},$ (2.6) and can change the ranges of $\varphi$ and $\psi$ to $0\leq\varphi<2\pi$ and $0\leq\psi<4\pi$. The periodicity is expressed as $\displaystyle(\theta,\varphi,\psi)\sim(\theta,\varphi+2\pi,\psi+2\pi)\sim(\theta,\varphi,\psi+4\pi).$ (2.7) From the local trivializations (2.3) and (2.4), one can see that $\theta$ and $\varphi$ are regarded as the angular coordinates of the base space $S^{2}$ through the stereographic projection. The patch I corresponds to $0\leq\theta<\pi$, while the patch II corresponds to $0<\theta\leq\pi$. The metric of $S^{3}$ is given as follows: $\displaystyle ds_{S^{3}}^{2}$ $\displaystyle=|dw_{1}|^{2}+|dw_{2}|^{2}$ $\displaystyle=\frac{1}{\mu^{2}}(d\theta^{2}+\sin^{2}\theta d\varphi^{2}+(d\psi+\cos\theta d\varphi)^{2}).$ (2.8) In the remainder of this section, the upper sign is taken in the patch I and the lower sign in the patch II. From (2.3), (2.4) and (2.8), one sees that the fiber $S^{1}$ is parameterized by $y=\frac{1}{\mu}(\psi\pm\varphi)$ and its radius is given by $2/\mu$. The connection 1-form is given by $\displaystyle\omega=\frac{\mu}{2}\left(dy+\frac{1}{\mu}(\cos\theta\mp 1)d\varphi\right).$ (2.9) The connection 1-form provides the vertical-horizontal decomposition by determining the inverse of the dreibein $E_{A}^{M}$ through $\omega(E_{\alpha}^{M})=0$, $E_{3}^{\mu}=0$ and $E_{3}^{y}=1$, where $A=1,2,3$, $\alpha=1,2$, $M=\theta,\varphi,y$ and $\mu=\theta,\varphi$. The inverse of the dreibein is determined as $\displaystyle E_{1}^{\theta}=\mu,\;\;\;E_{2}^{\varphi}=\frac{\mu}{\sin\theta},$ $\displaystyle E_{2}^{y}=\mu\frac{\cos\theta\mp 1}{\sin\theta},\;\;\;E_{3}^{y}=1,$ $\displaystyle\mbox{others}=0.$ (2.10) The dreibein are given by $\displaystyle E^{1}_{\theta}=e^{1}_{\theta}=\frac{1}{\mu},\;\;\;E^{2}_{\varphi}=e^{2}_{\varphi}=\frac{1}{\mu}\sin\theta,$ $\displaystyle E^{3}_{\varphi}=\frac{1}{\mu}(\cos\theta\mp 1),\;\;\;E^{3}_{y}=1,$ $\displaystyle\mbox{others}=0,$ (2.11) where $e_{\mu}^{\alpha}$ are the zweibein of $S^{2}$. We start with YM on $S^{3}$ $\displaystyle S_{S^{3}}=\frac{1}{4g_{S^{3}}^{2}}\int\frac{d\Omega_{3}}{(\mu/2)^{3}}\mbox{tr}(F_{AB}F_{AB}).$ (2.12) The vertical-horizontal decomposition tells us how to relate the gauge field on $S^{3}$ to the gauge field and the higgs field on $S^{2}$: $\displaystyle A_{\alpha}=a_{\alpha},$ $\displaystyle A_{3}=\phi.$ (2.13) Or equivalently $\displaystyle A_{\theta}=a_{\theta},$ $\displaystyle A_{\varphi}=a_{\varphi}+\frac{1}{\mu}(\cos\theta\mp 1)\phi,$ $\displaystyle A_{y}=\phi.$ (2.14) In (2.13) and (2.14), in order to make a dimensional reduction, we assume that the both sides are independent of $y$. Then, substituting (2.13) into (2.12) yields a YM-higgs on $S^{2}$, $\displaystyle S_{S^{2}}=\frac{1}{g_{S^{2}}^{2}}\int\frac{d\Omega_{2}}{\mu^{2}}{\rm tr}\left(\frac{1}{2}(f_{12}+\mu\phi)^{2}+\frac{1}{2}(D_{\alpha}\phi)^{2}\right),$ (2.15) where $g_{S^{2}}^{2}=\frac{\mu}{4\pi}g_{S^{3}}^{2}$. It is convenient for us to rewrite (2.15) using the three-dimensional flat space notation. We define a three-dimensional vector field in terms of $a_{\alpha}$ and $\phi$ [16]: $\displaystyle\vec{X}=\phi\vec{e}_{r}+a_{1}\vec{e}_{\varphi}-a_{2}\vec{e}_{\theta},$ (2.16) where $\vec{e}_{r}=(\sin\theta\cos\varphi,\sin\theta\sin\varphi,\cos\theta)$ and $\vec{e}_{\theta}=\frac{\partial\vec{e}_{r}}{\partial\theta},\;\;\vec{e}_{\varphi}=\frac{1}{\sin\theta}\frac{\partial\vec{e}_{r}}{\partial\varphi}$. We also introduce the angular momentum operator in three-dimensional flat space, $\displaystyle\vec{L}^{(0)}=-i\vec{e}_{\phi}\partial_{\theta}+i\frac{1}{\sin\theta}\vec{e}_{\theta}\partial_{\phi}.$ (2.17) Then, (2.15) is rewritten as $\displaystyle S_{S^{2}}=\frac{1}{g_{S^{2}}^{2}}\int\frac{d\Omega_{2}}{\mu^{2}}\frac{1}{2}\mbox{tr}\left(\mu X_{A}+i\mu\epsilon_{ABC}L^{(0)}_{B}X_{C}+\frac{i}{2}\epsilon_{ABC}[X_{B},X_{C}]\right)^{2}.$ (2.18) By dropping all the derivatives, we dimensionally reduce (2.18) to zero dimensions to obtain a matrix model: $\displaystyle S_{mm}=\frac{1}{g_{mm}^{2}}\frac{1}{2}\mbox{tr}\left(\mu X_{A}+\frac{i}{2}\epsilon_{ABC}[X_{B},X_{C}]\right)^{2},$ (2.19) where $g_{mm}^{2}=\frac{\mu^{2}}{4\pi}g_{S^{2}}^{2}$. The cross term in the above action is nothing but the Myers term [40]. It was first found in [41] that (2.19) is obtained from (2.12) through the dimensional reduction. We can obtain (2.18) and (2.19) directly from (2.12) in the following way. We parameterize the gauge field on $S^{3}$ as $A=X_{A}E^{A}$ [18], where $E^{A}$ is the right invariant 1-form defined in appendix A. Then, by using the Maurer-Cartan equation (A.4), we evaluate the curvature 2-form as $\displaystyle F$ $\displaystyle=dA+iA\wedge A$ $\displaystyle=\frac{1}{2}\epsilon_{ABC}\left(i\mu\epsilon_{CDE}{\cal L}_{D}X_{E}+\mu X_{C}+i\epsilon_{CDE}X_{D}X_{E}\right)E^{A}\wedge E^{B},$ (2.20) where ${\cal L}_{A}$ are the Killing vector dual to $E^{A}$, the explicit form of which is given in (A.7). Noting that ${\cal L}_{A}$ reduces to $L^{(0)}_{A}$ when $X_{A}$ is independent of $y$, one can easily see that (2.12) is dimensionally reduced to (2.18). Moreover, if we assume that $X_{A}$ is independent of all coordinates, we obtain the matrix model (2.19) directly from (2.12). The theories (2.15) and (2.19) possess many nontrivial vacua. Let us see how those vacua are described. First, the vacuum configurations of (2.15) with the gauge group $U(M)$ are determined by $\displaystyle f_{12}+\mu\phi=0,$ $\displaystyle D_{\alpha}\phi=0.$ (2.21) In the gauge in which $\phi$ is diagonal, (2.21) is solved as $\displaystyle\hat{a}_{1}=0,$ $\displaystyle\hat{a}_{2}=\frac{\cos\theta\mp 1}{\sin\theta}\hat{\phi},$ $\displaystyle\hat{\phi}=\frac{\mu}{2}\mbox{diag}(\cdots,\underbrace{n_{s-1},\cdots,n_{s-1}}_{N_{s-1}},\underbrace{n_{s},\cdots,n_{s}}_{N_{s}},\underbrace{n_{s+1},\cdots,n_{s+1}}_{N_{s+1}},\cdots),$ (2.22) where the gauge field takes the configurations of Dirac’s monopoles, so that $n_{s}$ must be integers due to Dirac’s quantization condition. Note also that $\sum_{s}N_{s}=M$. Thus the vacua of YM-higgs on $S^{2}$ are classified by the monopole charges $n_{s}/2$ and their degeneracies $N_{s}$. Next, the vacuum configurations of (2.19) with the gauge group $U(\hat{M})$ are determined by222There is a solution to the equations of motion of the matrix model (2.19), $X_{A}=\frac{\mu}{2}L_{A}$, which does not satisfy (2.23). It turns out that the theory around this solution is unstable. $\displaystyle[X_{A},X_{B}]=i\mu\epsilon_{ABC}X_{C}.$ (2.23) (2.23) is solved as $\displaystyle\hat{X}_{A}=\mu L_{A},$ (2.24) where $L_{A}$ are the representation matrices of the $SU(2)$ generators which are in general reducible, and are decomposed into irreducible representations: $\displaystyle L_{A}=\begin{pmatrix}\rotatebox[origin={tl}]{-35.0}{$\cdots\;\;\;\overbrace{\rotatebox[origin={c}]{35.0}{$L_{A}^{[j_{s-1}]}$}\;\cdots\;\rotatebox[origin={c}]{35.0}{$L_{A}^{[j_{s-1}]}$}}^{\rotatebox{35.0}{$N_{s-1}$}}\;\;\;\overbrace{\rotatebox[origin={c}]{35.0}{$L_{A}^{[j_{s}]}$}\;\cdots\;\rotatebox[origin={c}]{35.0}{$L_{A}^{[j_{s}]}$}}^{\rotatebox{35.0}{$N_{s}$}}\;\;\;\overbrace{\rotatebox[origin={c}]{35.0}{$L_{A}^{[j_{s+1}]}$}\;\cdots\;\rotatebox[origin={c}]{35.0}{$L_{A}^{[j_{s+1}]}$}}^{\rotatebox{35.0}{$N_{s+1}$}}\;\;\;\cdots$}\end{pmatrix},$ (2.25) where $L_{A}^{[j]}$ are the spin $j$ representation matrices of $SU(2)$ and $\sum_{s}N_{s}(2j_{s}+1)=\hat{M}$. The vacua of the matrix model are classified by the $SU(2)$ representations $[j_{s}]$ and their degeneracies $N_{s}$. (2.25) represents concentric fuzzy spheres with different radii. In the remainder of this section, we show relationships among the theories (2.12), (2.15) and (2.19). First, we show that the theory around the vacuum (2.22) of YM-higgs on $S^{2}$ is equivalent to the theory around the vacuum (2.24) of the matrix model if one puts $2j_{s}+1=N_{0}+n_{s}$ and takes the $N_{0}\rightarrow\infty$ limit with $g_{mm}^{2}/N_{0}$ fixed to $g_{S^{2}}^{2}\mu^{2}/4\pi$. We decompose the fields into the background corresponding (2.22) and the fluctuation as $X_{A}^{(s,t)}\rightarrow\hat{X}_{A}^{(s,t)}+X_{A}^{(s,t)}$, where $(s,t)$ label the (off-diagonal) blocks. Then, (2.18) is expanded around (2.22) as $\displaystyle S_{S^{2}}$ $\displaystyle=\frac{1}{g_{S^{2}}^{2}}\int\frac{d\Omega_{2}}{\mu^{2}}\frac{1}{2}\sum_{s,t}\mbox{tr}\left[\left(\mu X_{A}^{(s,t)}+i\mu\epsilon_{ABC}L^{(q_{st})}_{B}X_{C}^{(s,t)}+\frac{i}{2}\epsilon_{ABC}[X_{B},X_{C}]^{(s,t)}\right)\right.$ $\displaystyle\left.\qquad\qquad\qquad\qquad\qquad\times\left(\mu X_{A}^{(t,s)}+i\mu\epsilon_{ADE}L^{(q_{ts})}_{D}X_{E}^{(t,s)}+\frac{i}{2}\epsilon_{ADE}[X_{D},X_{E}]^{(t,s)}\right)\right],$ (2.26) where $q_{st}=(n_{s}-n_{t})/2$. $\vec{L}^{(q)}$ is the angular momentum operator in the presence of a monopole with the magnetic charge $q$ at the origin, which takes the form [42] $\displaystyle\vec{L}^{(q)}=\vec{L}^{(0)}-q\frac{\cos\theta\mp 1}{\sin\theta}\vec{e}_{\theta}-q\vec{e}_{r}.$ (2.27) We make a harmonic expansion of (2.26) by expanding the fluctuation in terms of the monopole vector spherical harmonics $\tilde{Y}_{JmqA}^{\rho}$ defined in appendix A as $\displaystyle X_{A}^{(s,t)}=\sum_{\rho=0,\pm 1}\sum_{\tilde{Q}\geq|q_{st}|}\sum_{m=-Q}^{Q}X_{Jm\rho}^{(s,t)}\tilde{Y}_{JmqA}^{\rho},$ (2.28) where $Q=J+\frac{(1+\rho)\rho}{2}$ and $\tilde{Q}=J-\frac{(1-\rho)\rho}{2}$. Substituting (2.28) into (2.26) yields $\displaystyle S_{S^{2}}$ $\displaystyle=\frac{4\pi}{g_{S^{2}}^{2}\mu^{2}}\mbox{tr}\left[\frac{\mu^{2}}{2}\sum_{s,t}\rho^{2}(J+1)^{2}X_{Jm\rho}^{(s,t)\dagger}X_{Jm\rho}^{(s,t)}\right.$ $\displaystyle+i\mu\sum_{s,t,u}\rho_{1}(J_{1}+1){\cal E}_{J_{1}m_{1}q_{st}\rho_{1}\;J_{2}m_{2}q_{tu}\rho_{2}\;J_{3}m_{3}q_{us}\rho_{3}}X_{J_{1}m_{1}\rho_{1}}^{(s,t)}X_{J_{2}m_{2}\rho_{2}}^{(t,u)}X_{J_{3}m_{3}\rho_{3}}^{(u,s)}$ $\displaystyle-\frac{1}{2}\sum_{s,t,u,v}(-1)^{m-q_{su}+1}{\cal E}_{J-mq_{us}\rho\;J_{1}m_{1}q_{st}\rho_{1}\;J_{2}m_{2}q_{tu}\rho_{2}}{\cal E}_{Jmq_{su}\rho\;J_{3}m_{3}q_{uv}\rho_{3}\;J_{4}m_{4}q_{vs}\rho_{4}}$ $\displaystyle\left.\qquad\qquad\qquad\times X_{J_{1}m_{1}\rho_{1}}^{(s,t)}X_{J_{2}m_{2}\rho_{2}}^{(t,u)}X_{J_{3}m_{3}\rho_{3}}^{(u,v)}X_{J_{4}m_{4}\rho_{4}}^{(v,s)}\right],$ (2.29) where ${\cal E}_{J_{1}m_{1}q_{st}\rho_{1}\;J_{2}m_{2}q_{tu}\rho_{2}\;J_{3}m_{3}q_{us}\rho_{3}}$ is defined in (A.42) and we have used (A.39). Similarly we decompose the matrices into the background given by (2.24) and the fluctuation as $X_{i}\rightarrow\hat{X}_{i}+X_{i}$ and obtain the theory around (2.24): $\displaystyle S_{mm}$ $\displaystyle=\frac{1}{g_{mm}^{2}}\frac{1}{2}\sum_{s,t}\mbox{tr}\left[\left(\mu X_{A}^{(s,t)}+i\mu\epsilon_{ABC}L_{B}\circ X_{C}^{(s,t)}+\frac{i}{2}\epsilon_{ABC}[X_{B},X_{C}]^{(s,t)}\right)\right.$ $\displaystyle\left.\qquad\qquad\qquad\qquad\times\left(\mu X_{A}^{(t,s)}+i\mu\epsilon_{ADE}L_{D}\circ X_{E}^{(t,s)}+\frac{i}{2}\epsilon_{ADE}[X_{D},X_{E}]^{(t,s)}\right)\right],$ (2.30) where $L_{A}\circ$ is defined by $\displaystyle L_{A}\circ X_{B}^{(s,t)}=L_{A}^{[j_{s}]}X_{B}^{(s,t)}-X_{B}^{(s,t)}L_{A}^{[j_{t}]}.$ (2.31) We make a harmonic expansion for (2.30) by expanding the fluctuation in terms of the fuzzy vector spherical harmonics $\hat{Y}_{Jm(j_{s}j_{t})A}^{\rho}$ defined in appendix A as $\displaystyle X_{A}^{(s,t)}=\sum_{\rho=0,\pm 1}\sum_{\tilde{Q}\geq|j_{s}-j_{t}|}^{j_{s}+j_{t}}\sum_{m=-Q}^{Q}X_{Jm\rho}^{(s,t)}\otimes\hat{Y}_{Jm(j_{s}j_{t})A}^{\rho}.$ (2.32) Since $j_{s}+j_{t}=N_{0}+\frac{n_{s}+n_{t}}{2}-1$, $N_{0}$ plays a role of the ultraviolet cutoff. Note also that $j_{s}-j_{t}=(n_{s}-n_{t})/2=q_{st}$. Substituting (2.32) into (2.30) yields $\displaystyle S_{mm}$ $\displaystyle=\frac{N_{0}}{g_{mm}^{2}}\mbox{tr}\left[\frac{\mu^{2}}{2}\sum_{s,t}\rho^{2}(J+1)^{2}X_{Jm\rho}^{(s,t)\dagger}X_{Jm\rho}^{(s,t)}\right.$ $\displaystyle+i\mu\sum_{s,t,u}\rho_{1}(J_{1}+1)\hat{{\cal E}}_{J_{1}m_{1}(j_{s}j_{t})\rho_{1}\;J_{2}m_{2}(j_{t}j_{u})\rho_{2}\;J_{3}m_{3}(j_{u}j_{s})\rho_{3}}X_{J_{1}m_{1}\rho_{1}}^{(s,t)}X_{J_{2}m_{2}\rho_{2}}^{(t,u)}X_{J_{3}m_{3}\rho_{3}}^{(u,s)}$ $\displaystyle-\frac{1}{2}\sum_{s,t,u,v}(-1)^{m-q_{su}+1}\hat{{\cal E}}_{J-m(j_{u}j_{s})\rho\;J_{1}m_{1}(j_{s}j_{t})\rho_{1}\;J_{2}m_{2}(j_{t}j_{u})\rho_{2}}\hat{{\cal E}}_{Jm(j_{s}j_{u})\rho\;J_{3}m_{3}(j_{u}j_{v})\rho_{3}\;J_{4}m_{4}(j_{v}j_{s})\rho_{4}}$ $\displaystyle\left.\qquad\qquad\qquad\times X_{J_{1}m_{1}\rho_{1}}^{(s,t)}X_{J_{2}m_{2}\rho_{2}}^{(t,u)}X_{J_{3}m_{3}\rho_{3}}^{(u,v)}X_{J_{4}m_{4}\rho_{4}}^{(v,s)}\right],$ (2.33) where $\hat{{\cal E}}_{J_{1}m_{1}(j_{s}j_{t})\rho_{1}\;J_{2}m_{2}(j_{t}j_{u})\rho_{2}\;J_{3}m_{3}(j_{u}j_{s})\rho_{3}}$ is defined in (A.42) and we have used (A.39). In the $N_{0}\rightarrow\infty$ limit, the ultraviolet cutoff goes to infinity and $\hat{{\cal E}}_{J_{1}m_{1}(j_{s}j_{t})\rho_{1}\;J_{2}m_{2}(j_{t}j_{u})\rho_{2}\;J_{3}m_{3}(j_{u}j_{s})\rho_{3}}$ reduces to ${\cal E}_{J_{1}m_{1}q_{st}\rho_{1}\;J_{2}m_{2}q_{tu}\rho_{2}\;J_{3}m_{3}q_{us}\rho_{3}}$ as shown in appendix A. Namely, this limit corresponds to the commutative (continuum) limit of the fuzzy spheres. Hence, in the limit in which $N_{0}\rightarrow\infty$ and $g_{mm}\rightarrow\infty$ such that $g_{mm}^{2}/N_{0}=g_{S^{2}}^{2}\mu^{2}/4\pi$, (2.33) agrees with (2.29). We have proven our statement. Next, we show that the theory around a certain vacuum of $U(M=N\times\infty)$ YM-higgs on $S^{2}$ with a periodicity condition imposed is equivalent to $U(N)$ YM on $S^{3}$. This is an extension of the matrix T-duality to a nontrivial fiber bundle. The vacuum of YM-higgs on $S^{2}$ we take is given by (2.22) with $s$ running from $-\infty$ to $\infty$, $n_{s}=s$ and $N_{s}=N$. $4\pi g_{S^{2}}^{2}/\mu$ is identified with the coupling constant on $S^{3}$, $g_{S^{3}}^{2}$. We decompose the fields on $S^{2}$ into the background and the fluctuation, $\displaystyle a_{\alpha}\rightarrow\hat{a}_{\alpha}+a_{\alpha},$ $\displaystyle\phi\rightarrow\hat{\phi}+\phi,$ (2.34) and impose the periodicity (orbifolding) condition on the fluctuation, $\displaystyle a_{\alpha}^{(s+1,t+1)}=a_{\alpha}^{(s,t)}\equiv a_{\alpha}^{(s-t)},$ $\displaystyle\phi^{(s+1,t+1)}=\phi^{(s,t)}\equiv\phi^{(s-t)}.$ (2.35) The fluctuations are gauge-transformed from the patch I to the patch II as [29] $\displaystyle{a^{\prime}}_{\\!\alpha}^{(s-t)}=e^{-i(s-t)\phi}a_{\alpha}^{(s-t)},$ $\displaystyle{\phi^{\prime}}^{(s-t)}=e^{-i(s-t)\phi}\phi^{(s-t)}.$ (2.36) We make the Fourier transformation for the fluctuations on each patch to construct the gauge field on the total space from the fields on the base space: $\displaystyle A_{\alpha}(\theta,\varphi,\psi)$ $\displaystyle=\sum_{w}a_{\alpha}^{(w)}(\theta,\varphi)e^{-i\frac{\mu}{2}wy},$ $\displaystyle A_{y}(\theta,\varphi,\psi)$ $\displaystyle=\sum_{w}\phi^{(w)}(\theta,\varphi)e^{-i\frac{\mu}{2}wy}.$ (2.37) We see from (2.36) that the lefthand sides of (2.37) are indeed independent of the patches. We substitute (2.37) into (2.26) and divide an overall factor $\sum_{s}$ to extract a single period. Then, we obtain $U(N)$ YM on $S^{3}$. The details of this calculation are given as a special case of (4.13) and (4.14). Finally, combining the above two statements, we see that the theory around (2.24) of the matrix model where $s$ runs from $-\infty$ to $\infty$, $2j_{s}+1=N_{0}+s$ is equivalent to $U(N)$ YM on $S^{3}$ if the $N_{0}\rightarrow\infty$ limit is taken with $g_{mm}^{2}/N_{0}$ fixed to $\frac{g_{S^{3}}^{2}\mu^{3}}{16\pi^{2}}$, the periodicity condition is imposed on the fluctuation on $S^{2}$ and the overall factor $\Sigma_{s}$ is divided. In this way, $S^{3}$ is realized in terms of the three matrices $X_{1},\;X_{2},\;X_{3}$. In sections 3-5, we generalize the results in this section. We set $\mu=1$ and set all other dimensionful parameters to a certain constant value. ## 3 Dimensional reduction on a principal bundle In this section, we provide the dimensional reduction of YM on a principal $G$ bundle to its base space. The case of principal $U(1)$ bundles was already given in [29]. Here we consider the case where $G$ is nonabelian. First, we give a metric and a vielbein of a fiber bundle on which pure YM is defined. We consider a principal $G$-bundle $P$ on a manifold $M$. The base space $M$ has a covering $\cal S$, and the total space has a covering $\\{\pi^{-1}(U)|U\in{\cal S}\\}$. $\pi^{-1}(U)$ is diffeomorphic to $U\times G$ by the local trivialization. Thus it is parameterized by $z^{M}=(x^{\mu},y^{m})\;(\mu=1,\cdots,\dim M\;;\;m=1,\cdots,\dim G)$, where $x^{\mu}$ parameterize the local patch $U$ and $y^{m}$ parameterize an element of $G$. We assume that the connection of $P$ is expressed as $\displaystyle\omega=g^{-1}(y)b(x)g(y)-i\>g^{-1}(y)dg(y).$ (3.1) where $g(y)\in G$, $b(x)=b_{\mu}^{a}(x)T^{a}dx^{\mu}$ and $T^{a}$ are the generators of the Lie group $G$. The transition functions of a principal bundle act on fibers by left multiplication. If there is overlap between $U$ and $U^{\prime}$, the relation between fiber coodinates, $g(y)$ on $U$ and $g(y^{\prime})$ on $U^{\prime}$, is given by $\displaystyle g(y^{\prime})=k(x)\>g(y)$ (3.2) where $k(x)\in G$. In the overlapping region $U\cap U^{\prime}$, $b(x)$ must transform as $\displaystyle b^{\prime}(x^{\prime})=k(x)\>b(x)\>k^{-1}(x)+i\>dk(x)k^{-1}(x).$ (3.3) Indeed, by using (3.3), we can show $\displaystyle\omega=g(y)^{-1}\>b(x)\>g(y)-i\>g(y)^{-1}dg(y)=g(y^{\prime})^{-1}\>b^{\prime}(x^{\prime})\>g(y^{\prime})-i\>g(y^{\prime})^{-1}dg(y^{\prime}).$ (3.4) We assume that the total space is endowed with a metric that has the fibered structure determined by the connection (3.1) and the isometry. As shown in [43], such metric can be locally expressed as333Throughout of this paper, we use the following normalizations for the traces: $\mbox{Tr}(T^{a}T^{b})=\frac{1}{2}\delta_{ab}$ for the structure group of the fiber bundle and $\mbox{tr}(T^{a}T^{b})=\delta_{ab}$ for the gauge group. $\displaystyle ds^{2}$ $\displaystyle=G_{MN}dz^{M}dz^{N}$ $\displaystyle=g_{\mu\nu}(x)dx^{\mu}dx^{\nu}+2{\rm Tr}\omega^{2}$ $\displaystyle=g_{\mu\nu}(x)dx^{\mu}dx^{\nu}+\\{e_{m}^{a}(y)dy^{m}-b^{a}_{\mu}(x)dx^{\mu}\\}^{2}.$ (3.5) Here $g_{\mu\nu}$ is a metric on the base space and $e_{m}^{a}(y)\;(a=\dim M+1,\cdots,\dim P)$ are the components of the right invariant Maurer-Cartan 1-form of $G$, which is defined by $\displaystyle dg(y)g(y)^{-1}=-ie^{a}_{m}(y)T^{a}dy^{m}.$ (3.6) We have assumed that the coefficient of the second term in (3.5) is just $\delta_{ab}$ so that the resultant dimensionally reduced theory is simple, although it is allowed to take $y$ independent function $\xi_{ab}(x)$. The Maurer-Cartan 1-form satisfies the Maurer-Cartan equation $\displaystyle de^{a}-\frac{1}{2}f^{abc}e^{b}\wedge e^{c}=0,$ (3.7) where $f^{abc}$ is the structure constant of the Lie algebra of $G$, and is regarded as the vielbein of the Cartan-Killing metric on $G$ defined by $\displaystyle h_{mn}(y)dy^{m}dy^{n}$ $\displaystyle\equiv-2{\rm Tr}\left(dgg^{-1}\right)^{2}$ $\displaystyle=e^{a}_{m}(y)e^{a}_{n}(y)dy^{m}dy^{n}.$ (3.8) Note that $e^{a}_{m}(y)$ and $b(x)$ in the metric (3.5) are defined locally on $U$ and must be transformed from $U$ to $U^{\prime}$: the transformation of $e^{a}_{m}(y)$ is determined by (3.2) and an equality $\displaystyle dg(y^{\prime})g(y^{\prime})^{-1}=-ie^{a}_{m}(y^{\prime})T^{a}dy^{\prime m},$ (3.9) while the transformation of $b(x)$ is given in (3.3). By introducing a vielbein on the base space, $e^{\alpha}_{\mu}(x)\;(\alpha=1,\cdots,\dim M)$, one can write a vielbein and its inverse on the total space as follows: $\displaystyle E^{A}_{\;\;M}(z)=\begin{pmatrix}e^{\alpha}_{\mu}(x)&0\\\ -b^{a}_{\mu}(x)&e^{a}_{m}(y)\end{pmatrix},\quad E^{M}_{\;\;A}(z)=\begin{pmatrix}e^{\mu}_{\alpha}(x)&0\\\ e^{m}_{a}(y)b^{a}_{\alpha}(x)&e^{m}_{a}(y)\end{pmatrix},$ (3.10) where $e^{\mu}_{\alpha}$ and $e^{m}_{a}$ are the inverse of $e^{\alpha}_{\mu}$ and $e^{a}_{m}$, respectively, and $b^{a}_{\alpha}(x)\equiv e^{\mu}_{\alpha}(x)b^{a}_{\mu}(x)$. The local Lorentz frame defined by (3.10) gives the vertical-horizontal decomposition of vectors and 1-forms on the total space. Namely, $\alpha=1,\cdots,\dim M$ correspond to the directions to those of the base space and $a=\dim M+1,\cdots,\dim P$ correspond of the fiber space. Again, we remark that these expressions are defined locally on $U$. From (3.3) and (3.9), we can obtain relationships of the vielbeins between on $U$ and on $U^{\prime}$ as $\displaystyle E^{\prime\alpha}=E^{\alpha},$ $\displaystyle E^{\prime a}={\rm Ad}(k)^{ab}E^{b}.$ (3.11) where ${\rm Ad}(k)$ is the adjoint representation of $k(x)$. (2.8) is a counterpart of (3.5), and (2.10) and (2.11) are a counterpart of (3.10). We next consider a gauge theory on the total space and make a dimensional reduction of the fiber direction to obtain a gauge theory on the base space. We start with $U(N)$ YM on the total space: $\displaystyle S_{P}=\frac{1}{g^{2}_{P}}\int d^{D}z\sqrt{G}\>{\rm tr}\left(\frac{1}{4}F_{MN}F^{MN}\right).$ (3.12) where $D=\dim P$ and $F_{MN}=\partial_{M}A_{N}-\partial_{N}A_{M}+i[A_{M},A_{N}]$. In order to make the reduction, we perform the vertical-horizontal decompostion for the gauge field $A_{M}(z)$ and the derivatives $\partial_{M}$ according to (3.10). The gauge field is decomposed as $\displaystyle A_{M}(z)=A_{\alpha}(z)E^{\alpha}_{M}(x)+A_{a}(z)E^{a}_{M}(z).$ (3.13) After the reduction, horizontal components $A_{\alpha}$ and vertical components $A_{a}$ of the gauge field will be naturally identified with the gauge field and the higgs fields on the base space, respectively. The field strength in the local Lorentz frame is rewritten as follows: $\displaystyle F_{\alpha\beta}=\nabla^{(M)}_{\alpha}A_{\beta}-\nabla^{(M)}_{\beta}A_{\alpha}+i[A_{\alpha},A_{\beta}]-b^{a}_{\alpha\beta}A_{a}+ib^{a}_{\alpha}{\cal L}_{a}A_{\beta}-ib^{a}_{\beta}{\cal L}_{a}A_{\alpha},$ $\displaystyle F_{\alpha a}=e^{\mu}_{\alpha}\partial_{\mu}A_{a}+i\>[A_{\alpha},A_{b}]-f^{abc}b^{b}_{\alpha}A_{c}-i{\cal L}_{a}A_{\alpha}+ib^{b}_{\alpha}{\cal L}_{b}A_{a},$ $\displaystyle F_{ab}=f^{abc}A_{c}+i\>[A_{a},A_{b}]+i{\cal L}_{a}A_{b}-i{\cal L}_{b}A_{a}.$ (3.14) Here we have defined the following quantities: $\displaystyle b^{a}_{\alpha\beta}$ $\displaystyle\equiv e^{\mu}_{\alpha}e^{\nu}_{\beta}\\{\partial_{\mu}b^{a}_{\nu}-\partial_{\nu}b^{a}_{\mu}-f^{abc}b^{b}_{\mu}b^{c}_{\nu}\\},$ $\displaystyle\nabla^{(M)}_{\alpha}A_{\beta}$ $\displaystyle\equiv e^{\mu}_{\alpha}\left(\partial_{\mu}A_{\beta}+\omega_{\mu\>\beta}^{\quad\gamma}A_{\gamma}\right),$ $\displaystyle{\cal L}_{a}$ $\displaystyle\equiv-ie^{m}_{a}\partial_{m},$ (3.15) where $\omega$ is the spin connection on the base space defined by $e^{\alpha}_{\mu}$ and ${\cal L}_{a}$ are the right invariant Killing vectors on the total space, which represent the isometry. Note that our calculations have been performed on $U$ so far. When it is performed on $U^{\prime}$, the quantities on $U^{\prime}$ must be used. The transformation of $b^{a}_{\alpha}(x)$ between on $U$ and on $U^{\prime}$ is given by (3.3), so that that of $b^{a}_{\alpha\beta}(x)$ is given by $\displaystyle b^{\prime a}_{\alpha\beta}(x^{\prime})={\rm Ad}(k)^{ab}b^{b}_{\alpha\beta}(x).$ (3.16) The gauge field with the local Lorentz index must be transformed according to (3.11) as $\displaystyle A^{\prime}_{\alpha}=A_{\alpha},$ $\displaystyle A^{\prime}_{a}={\rm Ad}(k)^{ab}A_{b}.$ (3.17) In order to make the dimensional reduction, we relate the fields on the total space to those on the base space as $\displaystyle A_{\alpha}$ $\displaystyle=a_{\alpha},$ $\displaystyle A_{a}$ $\displaystyle=\phi_{a},$ (3.18) where $a_{\alpha}$ are the gauge field in the local Lorentz frame and $\phi_{a}$ are higgs fields on the base space. We assume the both sides in (3.18) are independent of $y^{m}$. Using subscript of curved space, we can write (3.18) equivalently as $\displaystyle A_{\mu}$ $\displaystyle=a_{\mu}-b^{a}_{\mu}\phi_{a},$ $\displaystyle A_{m}$ $\displaystyle=e^{a}_{m}\phi_{a}.$ (3.19) Here (3.18) and (3.19) are a generalization of (2.13) and (2.14), respectively. Substituting (3.14) and (3.18) into (3.12) and using $\sqrt{G}=\sqrt{g}\sqrt{h}$, we obtain YM-higgs on the base space: $\displaystyle S_{M}$ $\displaystyle=\frac{1}{g_{M}^{2}}\int d^{d}x\sqrt{g}\>{\rm tr}\bigg{\\{}\frac{1}{4}\left(f_{\alpha\beta}-b^{a}_{\alpha\beta}\phi_{a}\right)^{2}+\frac{1}{2}\left(\nabla^{(M)}_{\alpha}\phi_{a}+i\>[a_{\alpha},\phi_{a}]-f^{abc}b^{b}_{\alpha}\phi_{c}\right)^{2}$ $\displaystyle\qquad+\frac{1}{4}\left(f^{abc}\phi_{c}+i\>[\phi_{a},\phi_{b}]\right)^{2}\bigg{\\}},$ (3.20) where $g_{M}^{2}=(\int dy\sqrt{h})^{-1}g_{P}^{2}=\frac{1}{\text{Vol}(G)}g_{P}^{2}$, $d=\dim M$ and $f_{\alpha\beta}=\nabla^{(M)}_{\alpha}a_{\beta}-\nabla^{(M)}_{\beta}a_{\alpha}+i[a_{\alpha},a_{\beta}]$. Note that the connection in the fiber bundle can generate nontrivial mass terms of the higgs fields. This is reminiscent of the flux compactification in string theory. ## 4 Extension of the matrix T-duality In this section, we extend the matrix T-duality on nontrivial $U(1)$ bundles developed in [29] to that on nontrivial $SU(2)$ bundles. ### 4.1 Nontrivial vacua and transformation between patches As in the example in section 2, the theory on the base space (3.20) has monopolelike vacua, which are in general patch-dependent if the principal bundle we consider is nontrivial. Here we describe the vacua and their patch- dependence. We also consider how the fields of the theory are transformed from a patch to another. We examine, in particular, the transformation properties of fluctuations around the vacua. It is seen from (3.20) that the condition for the vacua is given by $\displaystyle f_{\alpha\beta}-b^{a}_{\alpha\beta}\phi_{a}=0,$ $\displaystyle\nabla^{(M)}_{\alpha}\phi_{a}+i\>[a_{\alpha},\phi_{a}]-f^{abc}b^{b}_{\alpha}\phi_{c}=0,$ $\displaystyle f^{abc}\phi_{c}+i\>[\phi_{a},\phi_{b}]=0.$ (4.1) They are satisfied by the following configurations: $\displaystyle\hat{a}_{\alpha}(x)=b^{a}_{\alpha}(x)\hat{\phi}_{a}=b^{a}_{\alpha}(x)L_{a},$ $\displaystyle\hat{\phi}_{a}=L_{a},$ (4.2) where $L_{a}$ are the generators of the Lie algebra of $G$ satisfying $[L_{a},L_{b}]=if^{abc}L^{c}$ and generally reducible. Note that as mentioned in section 3.1, $b^{a}(x)$ are generally patch-dependent quantities. The vacua are, therefore, also patch-dependent. From (3.3) and (4.2), we can read off the transformation properties for the vacua between patches: $\displaystyle\hat{a}^{\prime}(x)=K(x)\hat{a}(x)K(x)^{-1}+i\>dK(x)K(x)^{-1},$ $\displaystyle\hat{\phi}^{\prime}_{a}={\rm Ad}(k(x))_{ab}K(x)\hat{\phi}_{b}K(x)^{-1}=\hat{\phi}_{a},$ (4.3) where $K(x)$ is obtained by replacing $T^{a}$ in $k(x)$ in (3.2) by $\hat{\phi}_{a}=L_{a}$. Note that this is the gauge transformation by $K(x)$ except for the rotation of $\hat{\phi}_{a}$ by ${\rm Ad}(k(x))$, which comes from (3.17). Let us consider the theory around the vacua (4.2) and decompose the fields into the backgrounds and fluctuations: $\displaystyle a_{\alpha}(x)=\hat{a}_{\alpha}(x)+\tilde{a}_{\alpha}(x),$ $\displaystyle\phi_{a}(x)=\hat{\phi}_{a}+\tilde{\phi}_{a}(x).$ (4.4) The fluctuations are transformed between patches as $\displaystyle\tilde{a}^{\prime}_{\alpha}(x)=K(x)\tilde{a}_{\alpha}(x)K(x)^{-1},$ $\displaystyle\tilde{\phi}^{\prime}_{a}(x)={\rm Ad}(k(x))_{ab}K(x)\tilde{\phi}_{b}(x)K(x)^{-1}.$ (4.5) One can easily see that the action (3.20) is indeed invariant under the transformation (4.3) and (4.5). ### 4.2 $G=U(1),SU(2)$ In this subsection, we consider the case in which the fiber is $U(1)$ or $SU(2)$. In the case of $G=U(1)$, the matrix T-duality indeed works as shown in [29] and its typical example was given in section 2. We extend the matrix T-duality to the case of $G=SU(2)$ by applying the fact described in section 2 that YM on $S^{3}$ is realized in the matrix model. First, we review the matrix T-duality in the case of $G=U(1)$, which is a generalization of the relationship between YM on $S^{3}$ and YM-higgs on $S^{2}$ in section 2. In this case, the metric (3.5) reduces to the following form: $\displaystyle ds^{2}=g_{\mu\nu}(x)dx^{\mu}dx^{\nu}+(dy- b_{\mu}(x)dx^{\mu})^{2},$ (4.6) where $y$ represents the fiber direction and $0\leq y<2\pi$. We put $\dim M=d$. (2.8) indeed takes the form of (4.6). YM-higgs on the base space obtained from YM on the total space is given as the $U(1)$ case of (3.20): $\displaystyle S_{M}$ $\displaystyle=\frac{1}{g_{M}^{2}}\int d^{d}x\sqrt{g}\>{\rm tr}\bigg{\\{}\frac{1}{4}\left(f_{\alpha\beta}-b_{\alpha\beta}\phi\right)^{2}+\frac{1}{2}\left(\nabla^{(M)}_{\alpha}\phi+i\>[a_{\alpha},\phi]\right)^{2}\bigg{\\}}.$ (4.7) (2.15) is a special case of (4.7). We show that we obtain the $U(N)$ YM on the total space from the $U(N\times\infty)$ YM-higgs on the base space through the following procedure: we choose a certain background of the $U(N\times\infty)$ YM-higgs on the base space, expand the theory around the background and impose a periodicity condition. Note, first, that a general background of (4.7) is given by $\displaystyle\hat{a}_{\alpha}=b_{\alpha}\hat{\phi},$ $\displaystyle\hat{\phi}=-\>\mbox{diag}(\cdots,\underbrace{n_{s-1},\cdots,n_{s-1}}_{N_{s-1}},\underbrace{n_{s},\cdots,n_{s}}_{N_{s}},\underbrace{n_{s+1},\cdots,n_{s+1}}_{N_{s+1}},\cdots),$ (4.8) which is a counterpart of (2.22). We decompose the fields into the backgrounds and the fluctuations as $\displaystyle a_{\alpha}$ $\displaystyle\rightarrow\hat{a}_{\alpha}+a_{\alpha},$ $\displaystyle\phi$ $\displaystyle\rightarrow\hat{\phi}+\phi.$ (4.9) In particular, we take the following background: $s$ running from $-\infty$ to $\infty$, $n_{s}=s$ and $N_{s}=N$. We label the (off-diagonal) blocks by $(s,t)$ and impose the periodicity (orbifolding) condition on the fluctuations as in (2.35): $\displaystyle a_{\alpha}^{(s+1,t+1)}=a_{\alpha}^{(s,t)}\equiv a^{(s-t)}_{\alpha},$ $\displaystyle\phi^{(s+1,t+1)}=\phi^{(s,t)}\equiv\phi^{(s-t)}.$ (4.10) The fluctuations are gauge-transformed from $U$ to $U^{\prime}$ as $\displaystyle{a^{\prime}}_{\alpha}^{(s-t)}$ $\displaystyle=e^{-i(s-t)v(x)}a^{(s-t)}_{\alpha},$ $\displaystyle\phi^{\prime(s-t)}$ $\displaystyle=e^{-i(s-t)v(x)}\phi^{(s-t)},$ (4.11) where $e^{-iv}$ is a transition function; $e^{-iy^{\prime}}=e^{-iv(x)}e^{-iy}$. (2.36) is a special case of (4.11). We make the Fourier transformation for the fluctuations on each patch to construct the gauge field on the total space: $\displaystyle A_{\alpha}(x,y)$ $\displaystyle=\sum_{w}a^{(w)}_{\alpha}(x)e^{-iwy},$ $\displaystyle A_{d+1}(x,y)$ $\displaystyle=\sum_{w}\phi^{(w)}(x)e^{-iwy}.$ (4.12) We can see from (4.11) that the lefthand sides in the above equations are indeed invariant under the transformation between patches. Using (4.9) and (4.12), we can rewrite each term in (4.7) as $\displaystyle\left(f_{\alpha\beta}-b_{\alpha\beta}\phi\right)^{(s,t)}$ $\displaystyle\rightarrow\left(\nabla^{(M)}_{\alpha}a_{\beta}-\nabla^{(M)}_{\beta}a_{\alpha}+i[\hat{a}_{\alpha},a_{\beta}]+i[a_{\alpha},\hat{a}_{\beta}]+i[a_{\alpha},a_{\beta}]-b_{\alpha\beta}\phi\right)^{(s,t)}$ $\displaystyle=\left(\nabla^{(M)}_{\alpha}a_{\beta}^{(s-t)}-\nabla^{(M)}_{\beta}a_{\alpha}^{(s-t)}+i[a_{\alpha},a_{\beta}]^{(s-t)}-i(s-t)b_{\alpha}a_{\beta}^{(s-t)}+i(s-t)b_{\beta}a_{\alpha}^{(s-t)}-b_{\alpha\beta}\phi^{(s-t)}\right)$ $\displaystyle=\frac{1}{2\pi}\int dy\left(\nabla^{(M)}_{\alpha}A_{\beta}-\nabla^{(M)}_{\beta}A_{\alpha}+i[A_{\alpha},A_{\beta}]-b_{\alpha\beta}A_{d+1}+b_{\alpha}\partial_{y}A_{\beta}-b_{\beta}\partial_{y}A_{\alpha}\right)\>e^{i(s-t)y}$ $\displaystyle=\frac{1}{2\pi}\int dyF_{\alpha\beta}\>e^{i(s-t)y},$ $\displaystyle\left(\nabla^{(M)}_{\alpha}\phi+i\>[a_{\alpha},\phi]\right)^{(s,t)}$ $\displaystyle\rightarrow\left(\nabla^{(M)}_{\alpha}\phi+i\>[\hat{a}_{\alpha},\phi]+i\>[a_{\alpha},\hat{\phi}]+i\>[a_{\alpha},\phi]\right)^{(s,t)}$ $\displaystyle=\nabla^{(M)}_{\alpha}\phi^{(s-t)}+i\>[a_{\alpha},\phi]^{(s-t)}-i\>(s-t)b_{\alpha}\phi^{(s-t)}+i\>(s-t)a_{\alpha}^{(s-t)}$ $\displaystyle=\frac{1}{2\pi}\int dy\left(\nabla^{(M)}_{\alpha}A_{d+1}+i\>[A_{\alpha},A_{d+1}]-\partial_{y}A_{\alpha}+b_{\alpha}\partial_{y}A_{d+1}\right)\>e^{i(s-t)y}$ $\displaystyle=\frac{1}{2\pi}\int dyF_{\alpha(d+1)}\>e^{i(s-t)y}.$ (4.13) Then (4.7) becomes $\displaystyle S_{M}$ $\displaystyle=\frac{1}{g_{M}^{2}}\int d^{d}x\sqrt{g}\>{\rm tr}\biggl{\\{}\frac{1}{4}\left(f_{\alpha\beta}-b_{\alpha\beta}\phi\right)^{2}+\frac{1}{2}\left(\nabla^{(M)}_{\alpha}\phi_{a}+i\>[a_{\alpha},\phi]\right)^{2}\biggr{\\}}$ $\displaystyle=\frac{1}{g_{M}^{2}}\int d^{d}x\sqrt{g}\>{\rm tr}\Biggl{[}\sum_{s,t}\biggl{\\{}\frac{1}{4}\left(f_{\alpha\beta}-b_{\alpha\beta}\phi\right)^{(s,t)}\left(f_{\alpha\beta}-b_{\alpha\beta}\phi\right)^{(t,s)}$ $\displaystyle\hskip 142.26378pt+\frac{1}{2}\left(\nabla^{(M)}_{\alpha}\phi+i\>[a_{\alpha},\phi]\right)^{(s,t)}\left(\nabla^{(M)}_{\alpha}\phi+i\>[a_{\alpha},\phi]\right)^{(t,s)}\biggr{\\}}\Biggr{]}$ $\displaystyle\rightarrow\frac{1}{g_{M}^{2}}\frac{1}{2\pi}\sum_{w}\int d^{D}z\sqrt{G}\>\frac{1}{4}{\rm tr}\left(F_{AB}F_{AB}\right).$ (4.14) By dividing an overall factor $\sum_{w}$ in the last line in (4.14) to extract a single period, we obtain Yang-Mills theory on the total space. Next we consider the case where fiber is $SU(2)$. In this case, YM-higgs on the base space takes the form $\displaystyle S_{M}$ $\displaystyle=\frac{1}{g_{M}^{2}}\int d^{d}x\sqrt{g}\>{\rm tr}\bigg{\\{}\frac{1}{4}\left(f_{\alpha\beta}-b^{a}_{\alpha\beta}\phi_{a}\right)^{2}+\frac{1}{2}\left(\nabla^{(M)}_{\alpha}\phi_{a}+i\>[a_{\alpha},\phi_{a}]-\epsilon^{abc}b^{b}_{\alpha}\phi_{c}\right)^{2}$ $\displaystyle\qquad\qquad+\frac{1}{4}\left(\epsilon^{abc}\phi_{c}+i\>[\phi_{a},\phi_{b}]\right)^{2}\bigg{\\}}.$ (4.15) We show that we can obtain the $U(N)$ YM on the total space of a nontrivial $SU(2)$-bundle from the YM with three higgs on its base space in a way similar to the case of $G=U(1)$. The vacuum of YM-higgs is given by (4.2) with $L_{a}$ satisfying the $SU(2)$ algebra, $[L_{a},L_{b}]=i\>\epsilon_{abc}L_{c}$, and $L_{a}$ generically take a reducible representation (2.25). We expand the fields around this background, $\displaystyle a_{\alpha}(x)$ $\displaystyle\rightarrow\hat{a}_{\alpha}(x)+a_{\alpha}(x),$ $\displaystyle\phi_{a}(x)$ $\displaystyle\rightarrow\hat{\phi}_{a}+\phi_{a}(x).$ (4.16) We label the (off-diagonal) blocks of the fluctuations by $(s,t)$, which is $(N_{s}(2j_{s}+1))\times(N_{t}(2j_{t}+1))$ matrix, and expand them by the fuzzy spherical harmonics: $\displaystyle a_{\alpha}^{(s,t)}(x)$ $\displaystyle=\sum_{J=|j_{s}-j_{t}|}^{j_{s}+j_{t}}\sum_{m=-J}^{J}a_{\alpha,Jm}^{(s,t)}(x)\otimes\hat{Y}_{Jm(j_{s}j_{t})},$ $\displaystyle\phi_{a}^{(s,t)}(x)$ $\displaystyle=\sum_{J=|j_{s}-j_{t}|}^{j_{s}+j_{t}}\sum_{m=-J}^{J}\phi_{a,Jm}^{(s,t)}(x)\otimes\hat{Y}_{Jm(j_{s}j_{t})}.$ (4.17) We verify from (4.5), (4.17) and (A.26) that the modes are gauge-transformed from $U$ to $U^{\prime}$ as $\displaystyle a^{\prime(s,t)}_{\alpha,Jm}(x)=\sum_{m^{\prime}}\langle Jm|k^{[J]}|Jm^{\prime}\rangle a_{\alpha,Jm^{\prime}}^{(s,t)}(x),$ $\displaystyle\phi^{\prime(s,t)}_{a,Jm}(x)=\sum_{m^{\prime}}{\rm Ad}(k)_{ab}\langle Jm|k^{[J]}|Jm^{\prime}\rangle\phi_{b,Jm^{\prime}}^{(s,t)}(x),$ (4.18) where $k^{[J]}$ is the spin $J$ representation of $SU(2)$ for $k(x)$. In what follows, we assume that as a background we set $2j_{s}+1=N_{0}+s$ with $s$ running from $-T$ to $T$ in (2.25) and take the limit of $N_{0}\rightarrow\infty$ and $T\rightarrow\infty$ in order. For the modes, we impose the periodicity condition: $\displaystyle a_{\alpha,Jm}^{(s+1,t+1)}=a_{\alpha,Jm}^{(s,t)}\equiv a_{\alpha,Jm}^{(q_{st})},$ $\displaystyle\phi_{a,Jm}^{(s+1,t+1)}=\phi_{a,Jm}^{(s,t)}\equiv\phi_{a,Jm}^{(q_{st})},$ (4.19) where $q_{st}\equiv\frac{s-t}{2}$. By using these modes and the spherical harmonics on $S^{3}$, we make Fourier transformation on each patch to construct the gauge field on the total space: $\displaystyle A_{\alpha}(z)=\sum_{Jm\tilde{m}}a_{\alpha,Jm}^{(\tilde{m})}(x)Y_{Jm\tilde{m}}(y),$ $\displaystyle A_{a}(z)=\sum_{Jm\tilde{m}}\phi_{a,Jm}^{(\tilde{m})}(x)Y_{Jm\tilde{m}}(y).$ (4.20) Its inverse is $\displaystyle a_{\alpha,Jm}^{(\tilde{m})}(x)=\int\frac{d\Omega_{3}}{2\pi^{2}}A_{\alpha}(z)Y_{Jm\tilde{m}}^{\dagger}(y),$ $\displaystyle\phi_{a,Jm}^{(\tilde{m})}(x)=\int\frac{d\Omega_{3}}{2\pi^{2}}A_{a}(z)Y_{Jm\tilde{m}}^{\dagger}(y).$ (4.21) From (4.18) and (A.8), it is verified that the lefthand sides in (4.20) are indeed transformed between patches as the gauge field on the total space (3.17). Using (4.17) and (4.21), we can obtain the following equalities: $\displaystyle[L_{a},a_{\alpha}(x)]^{(s,t)}$ $\displaystyle=\int\frac{d\Omega_{3}}{2\pi^{2}}\left({\cal L}_{a}A_{\alpha}(z)\right)Y_{Jpq_{st}}^{\dagger}\otimes\hat{Y}_{Jp(j_{s}j_{t})},$ $\displaystyle[\phi_{a},\phi_{b}]^{(s,t)}$ $\displaystyle=\int\frac{d\Omega_{3}}{2\pi^{2}}[A_{a}(z),A_{b}(z)]Y_{Jmq_{st}}^{\dagger}(y)\otimes\hat{Y}_{Jm(j_{s}j_{t})}.$ (4.22) The derivation of the above equalities is given in appendix B. Substituting these into (4.15), we obtain $\displaystyle S_{M}$ $\displaystyle=\frac{1}{g_{M}^{2}}\int d^{d}x\sqrt{g}\>{\rm tr}\bigg{\\{}\frac{1}{4}\left(f_{\alpha\beta}-b^{a}_{\alpha\beta}\phi_{a}\right)^{2}+\frac{1}{2}\left(\nabla^{(M)}_{\alpha}\phi_{a}+i\>[a_{\alpha},\phi_{a}]-\epsilon^{abc}b^{b}_{\alpha}\phi_{c}\right)^{2}$ $\displaystyle\qquad\qquad+\frac{1}{4}\left(\epsilon^{abc}\phi_{c}+i\>[\phi_{a},\phi_{b}]\right)^{2}\bigg{\\}}$ $\displaystyle=\frac{1}{g_{M}^{2}}\int d^{d}x\sqrt{g}\>{\rm tr}\Biggl{[}\sum_{s,t}\biggl{\\{}\frac{1}{4}\left(f_{\alpha\beta}-b^{a}_{\alpha\beta}\phi_{a}\right)^{(s,t)}\left(f_{\alpha\beta}-b^{a}_{\alpha\beta}\phi_{a}\right)^{(t,s)}$ $\displaystyle\qquad\qquad+\frac{1}{2}\left(\nabla^{(M)}_{\alpha}\phi_{a}+i\>[a_{\alpha},\phi_{a}]-\epsilon^{abc}b^{b}_{\alpha}\phi_{c}\right)^{(s,t)}\left(\nabla^{(M)}_{\alpha}\phi_{a}+i\>[a_{\alpha},\phi_{a}]-\epsilon^{abc}b^{b}_{\alpha}\phi_{c}\right)^{(t,s)}$ $\displaystyle\qquad\qquad+\frac{1}{4}\left(\epsilon^{abc}\phi_{c}+i\>[\phi_{a},\phi_{b}]\right)^{(s,t)}\left(\epsilon^{abc}\phi_{c}+i\>[\phi_{a},\phi_{b}]\right)^{(t,s)}\biggr{\\}}\Biggr{]}$ $\displaystyle\rightarrow\frac{1}{g_{M}^{2}}\frac{N_{0}}{2\pi^{2}}\sum_{w}\int d^{D}z\sqrt{G}$ $\displaystyle\qquad\qquad\times{\rm tr}\bigg{\\{}\frac{1}{4}\left(\nabla^{(M)}_{\alpha}A_{\beta}-\nabla^{(M)}_{\beta}A_{\alpha}+i[A_{\alpha},A_{\beta}]-b^{a}_{\alpha\beta}A_{a}+ib^{a}_{\alpha}{\cal L}_{a}A_{\beta}-ib^{a}_{\beta}{\cal L}_{a}A_{\alpha}\right)^{2}$ $\displaystyle\qquad\qquad+\frac{1}{2}\left(\nabla^{(M)}_{\alpha}A_{a}+i\>[A_{\alpha},A_{b}]-f^{abc}b^{b}_{\alpha}A_{c}-i{\cal L}_{a}A_{\alpha}+ib^{b}_{\alpha}{\cal L}_{b}A_{a}\right)^{2}$ $\displaystyle\qquad\qquad+\frac{1}{4}\left(f^{abc}A_{c}+i\>[A_{a},A_{b}]+i{\cal L}_{a}A_{b}-i{\cal L}_{b}A_{a}\right)^{2}\bigg{\\}}$ $\displaystyle=\frac{1}{g_{M}^{2}}\frac{N_{0}}{2\pi^{2}}\sum_{w}\int d^{d+1}z\sqrt{G}\mbox{tr}\left(\frac{1}{4}F_{AB}F_{AB}\right)$ (4.23) By dividing an overall factor $\sum_{w}$ in the last line in (4.23) to extract a single period, we obtain Yang-Mills theory on the total space. We can easily extend the above matrix T-duality to the case in which the fiber is $SU(2)^{k}\times U(1)^{l}$. As an example, we consider an $SU(2)\times U(1)$ bundle, $P$. Let $a,b,c$ in (3.20) run $0,1,2,3$ such that ‘$0$’ corresponds to the $U(1)$ direction and ‘$1,2,3$’ correspond to the $SU(2)$ direction. We assign $i,j,k$ to the $SU(2)$ direction. We can consider YM- higgs on the $U(1)$ bundle on $M$, $M^{\prime}$, which is obtained by making the dimensional reduction of the $SU(2)$ fiber direction for YM on the $SU(2)\times U(1)$ bundle. We realize the theory around an $SU(2)$ multimonopole background of YM-higgs on $M^{\prime}$ by taking the following background in YM-higgs on $M$ (3.20) and imposing the periodicity condition to the fluctuations: $\displaystyle\hat{\phi}_{0}=-\frac{1}{R}\mbox{diag}(\cdots,t-1,t,t+1,\cdots)\otimes 1_{\hat{M}}+b_{0}^{i}\hat{\phi}_{i},$ $\displaystyle\hat{\phi}_{i}=1_{\infty}\otimes(L_{i}\;\mbox{in (\ref{matrix background})}),$ $\displaystyle\hat{a}_{\alpha}=b_{\alpha}^{a}\hat{\phi}_{a},$ (4.24) where $b_{\alpha}^{0}$ represents the $U(1)$ monopole and $b_{\alpha}^{i}$ represents the SU(2) monopole. $R$ is a certain constant depending on the fiber structure. By setting $2j_{s}+1=N_{0}+s$ with $s$ running from $-T$ to $T$, taking the limit of $N_{0}\rightarrow\infty$ and $T\rightarrow\infty$ in order and imposing the periodicity condition to the fluctuations again, we realize YM on $P$ in YM-higgs on $M$. In a similar way, we can realize YM on an $SU(2)^{k}\times U(1)^{l}$ in YM-higgs on its base space. ### 4.3 Example: $S^{7}\rightarrow S^{4}$ We present an example of our findings in the previous subsection: we consider $S^{7}$ with radius $2$ and regard it as $SU(2)\cong S^{3}$ Hopf bundle on $S^{4}$ with radius $1$. In order to describe $S^{7}$ as $SU(2)$ bundle on $S^{4}$, it is convenient to introduce the quaternion $H$ (see for example [44, 45, 46]). The quaternion algebra is defined by $\displaystyle\bm{i}^{2}=\bm{j}^{2}=\bm{k}^{2}=-1,\quad\bm{i}\bm{j}=-\bm{j}\bm{i}=\bm{k},$ (4.25) $\displaystyle\bm{j}\bm{k}=-\bm{k}\bm{j}=\bm{i},\quad\bm{k}\bm{i}=-\bm{i}\bm{k}=\bm{j}.$ (4.26) An arbitrary element of $H$ is written as $\displaystyle q=a+b\bm{i}+c\bm{j}+d\bm{k}.$ (4.27) where $a,b,c,d\in R$. Its conjugation $q^{*}$ is defined by $\displaystyle q^{*}\equiv a-b\bm{i}-c\bm{j}-d\bm{k}.$ (4.28) The absolute value is given by $\displaystyle|q|\equiv\sqrt{q^{*}q}=\sqrt{a^{2}+b^{2}+c^{2}+d^{2}}\geq 0.$ (4.29) $S^{7}$ with radius $2$ is expressed by using quaternions as follows: $\displaystyle\\{(q_{1},q_{2})\in H^{2}||q_{1}|^{2}+|q_{2}|^{2}=4\\}.$ (4.30) The Hopf map $\pi:S^{7}\rightarrow S^{4}$ is defined by $\displaystyle\pi:(q_{1},q_{2})\rightarrow[(q_{1},q_{2})]\equiv\\{(q_{1},q_{2})q|q\in H\backslash\\{0\\}\\}.$ (4.31) In order to introduce local coordinates one needs to divide $S^{4}$ in two patches: $U_{1}$ ($q_{1}\neq 0$) and $U_{2}$ ($q_{2}\neq 0$). The local trivialization is given on each patch by $\displaystyle\pi^{-1}(U_{1})\ni(q_{1},q_{2})$ $\displaystyle\rightarrow(q_{2}q_{1}^{-1},q_{1}|q_{1}|^{-1})\in U_{1}\times SU(2),$ $\displaystyle\pi^{-1}(U_{2})\ni(q_{1},q_{2})$ $\displaystyle\rightarrow(q_{1}q_{2}^{-1},q_{2}|q_{2}|^{-1})\in U_{2}\times SU(2).$ (4.32) We parameterize $(q_{1},q_{2})$ by using a matrix representation of quaternions as $\displaystyle q_{1}$ $\displaystyle=2\cos\frac{\chi}{2}\;\lambda,$ $\displaystyle q_{2}$ $\displaystyle=2\sin\frac{\chi}{2}\;\kappa\lambda.$ (4.33) where $\kappa,\>\lambda\in SU(2)$ are defined by using Pauli matrices $\sigma^{a}\>(a=1,2,3)$ as $\displaystyle\kappa$ $\displaystyle=e^{i\eta\frac{\sigma^{3}}{2}}e^{i\xi\frac{\sigma^{2}}{2}}e^{i\zeta\frac{\sigma^{3}}{2}},$ $\displaystyle\lambda$ $\displaystyle=e^{-i\psi\frac{\sigma^{3}}{2}}e^{-i\theta\frac{\sigma^{2}}{2}}e^{-i\phi\frac{\sigma^{3}}{2}}.$ (4.34) The ranges of variables in the above equations are $\displaystyle 0\leq\chi\leq\pi,$ $\displaystyle 0\leq\xi\leq\pi,\;0\leq\eta<2\pi,\;0\leq\zeta<4\pi,$ $\displaystyle 0\leq\theta\leq\pi,\;0\leq\phi<2\pi,\;0\leq\psi<4\pi.$ (4.35) In particular, $|\lambda|^{2}=\det\lambda=1$ and $|\kappa|^{2}=\det\kappa=1$ hold. One can easily see from (4.32) and (4.33) that on $U_{1}$ the fiber space $SU(2)$ is described by $\lambda$ while on $U_{2}$ that is described by $\lambda^{\prime}\equiv\kappa\lambda$. In the following, we restrict ourselves to the region $U_{1}$. We denote sets of coordinates as $x^{\mu}=(\chi,\xi,\eta,\zeta)=(\chi,x^{\bar{\mu}})$ and $y^{m}=(\theta,\phi,\psi)$. $x^{\mu}$ are coordinates of $S^{4}$, $x^{\bar{\mu}}$ are those of $S^{3}$ inside of $S^{4}$ and $y^{m}$ are those of $SU(2)$ of fiber. In order to describe a metric of $S^{7}$ explicitly, we introduce the Maurer-Cartan 1-forms for $\kappa$ and $\lambda$ $\displaystyle\kappa(\bar{x})^{\dagger}d\kappa(\bar{x})$ $\displaystyle=i\bar{e}^{a}_{\bar{\mu}}(\bar{x})\frac{\sigma^{a}}{2}dx^{\bar{\mu}},$ $\displaystyle d\lambda(y)\lambda(y)^{\dagger}$ $\displaystyle=-i\>e^{a}_{m}(y)\frac{\sigma^{a}}{2}dy^{m},$ (4.36) where $\bar{x}$ represents the set of $\\{x^{\bar{\mu}}\\}$. Then we define the metric of $S^{7}$ as $\displaystyle ds_{S^{7}}^{2}$ $\displaystyle=\det(dq_{1})+\det(dq_{2}),$ (4.37) which is evaluated as $\displaystyle ds_{S^{7}}^{2}=\Bigl{(}d\chi^{2}+\frac{1}{4}\sin^{2}\chi\;\bar{e}^{a}_{\bar{\mu}}(\bar{x})\bar{e}^{a}_{\bar{\nu}}(\bar{x})dx^{\bar{\mu}}dx^{\bar{\nu}}\Bigr{)}+\Bigl{(}e^{a}_{m}(y)dy^{m}-\sin^{2}\frac{\chi}{2}\>\bar{e}^{a}_{\bar{\mu}}(\bar{x})dx^{\bar{\mu}}\Bigr{)}^{2}.$ (4.38) In the above expression, the first term represents the metric of the base space $S^{4}$ and the second one represents that of the fiber space $SU(2)$ locally. Note that $\frac{1}{4}\bar{e}^{a}_{\bar{\mu}}\bar{e}^{a}_{\bar{\nu}}$ and $\frac{1}{4}e^{a}_{m}e^{a}_{n}$ are a metric of $S^{3}$ with radius $1$. From (4.38) one can read off the vielbein on $S^{4}$ and the local connections of the fiber bundle as $\displaystyle e^{\alpha}_{\;\;\mu}(x)$ $\displaystyle=\begin{pmatrix}1&0\\\ 0&\frac{1}{2}\sin\chi\>\bar{e}^{a}_{\;\;\bar{\mu}}(\bar{x})\end{pmatrix},\qquad e^{\mu}_{\;\;\alpha}(x)=\begin{pmatrix}1&0\\\ 0&\frac{2}{\sin\chi}\>\bar{e}^{\bar{\mu}}_{\;\;a}(\bar{x})\end{pmatrix},$ $\displaystyle b^{a}_{\chi}(x)=0,\quad b^{a}_{\bar{\mu}}(x)=\tan\frac{\chi}{2}\>e^{a}_{\bar{\mu}}(x),$ $\displaystyle b^{a}_{\chi\bar{\nu}}(x)=e^{a}_{\bar{\nu}}(x),\quad b^{a}_{\bar{\mu}\bar{\nu}}(x)=f^{abc}e^{b}_{\bar{\mu}}(x)e^{c}_{\bar{\nu}}(x).$ (4.39) As noted before, when we move to the other region, $U_{2}$, we must change $\lambda$ to $\lambda^{\prime}\equiv\kappa\lambda$. Then, one can easily find that the local connections change to $\displaystyle{b^{\prime}}^{a}_{\chi}(x)=0,\quad{b^{\prime}}^{a}_{\bar{\mu}}(x)=-\cot\frac{\chi}{2}\>{\rm Ad}(\kappa)^{ab}e^{b}_{\bar{\mu}}(x),$ $\displaystyle b^{a}_{\chi\bar{\nu}}(x)={\rm Ad}(\kappa)^{ab}b^{b}_{\chi\bar{\nu}}(x),\quad b^{a}_{\bar{\mu}\bar{\nu}}(x)={\rm Ad}(\kappa)^{ab}b^{b}_{\bar{\mu}\bar{\nu}}(x).$ (4.40) This transformation property is consistent with (3.3). The vacua of (4.15) are given by (4.2), (2.25), (4.39) and (4.40) on each patch. $b^{a}_{\mu}$ and ${b^{\prime}}^{a}_{\mu}$ are known as the gauge field of the Yang monopole [47]. By applying the arguments in the previous subsection, we can show that YM on $S^{7}$ is equivalent to the theory around the multi Yang monopole background of YM-higgs on $S^{4}$ with the periodicity imposed. ## 5 Gauge theories on $SU(n+1)(/H)$ and matrix model In this section, we reveal various relations among gauge theories on $SU(n+1)$ and $SU(n+1)/H$, where $H$ is $SU(n)$ or $SU(n)\times U(1)$ or $SU(n+1)$ which is a subgroup of $SU(n+1)$. Note that $SU(n+1)/SU(n)\simeq S^{2n+1}$ and $SU(n+1)/(SU(n)\times U(1))\simeq CP^{n}$ and for $H=SU(n+1)$ the corresponding gauge theory reduces to a matrix model. First, we develop a general formalism of a dimensional reduction by which one can obtain YM-higgs on $\tilde{G}/H$ from YM on $\tilde{G}$, where $\tilde{G}$ is an arbitrary group manifold. Applying this formalism to the case of $\tilde{G}=SU(n+1)$, we obtain YM-higgs on $S^{2n+1}$ and on $CP^{n}$ and the matrix model. Next, by using the facts explained in appendix E, we show that the YM-higgs on $CP^{n}$ in the most general $U(1)$ monopole background is obtained by taking the commutative limit of the theory around a certain background of the matrix model. We have found the correct form of the YM-higgs type action of such theory on $CP^{n}$. Third, by using the extended matrix T-duality of the $U(1)$ case reviewed in section 4, we show that YM-higgs on $S^{2n+1}$ is equivalent to the theory around a certain background of YM-higgs on $CP^{n}$ with the orbifolding condition imposed. Combining these two facts, we also show that YM-higgs on $S^{2n+1}$ is realized as the theory around an appropriate background of the matrix model with the orbifolding condition imposed. Finally, by using the results in section 4, we show that YM on $SU(n+1)$ is realized in YM-higgs on $SU(N+1)/(SU(2)^{k}\times U(1)^{l})$. In particular, it follows that YM on $SU(3)$ is realized in YM-higgs on $S^{5}$ and on $CP^{2}$. ### 5.1 Dimensional reduction of YM theory on a group manifold In this subsection, we restrict ourselves to the case in which the total space $P$ is itself a group manifold $\tilde{G}$. In this case, we can take the Maurer-Cartan basis and rewrite the YM action on $\tilde{G}$ in such a way that the relation between YM on the total space and YM-higgs on the base space becomes more manifest. In terms of this expression of the YM action, we can easily perform the dimensional reduction to obtain the YM-higgs theory on a coset space $\tilde{G}/H$, where $H$ is a subgroup of $\tilde{G}$. Some conventions on the group manifold $\tilde{G}$ and the coset space $\tilde{G}/H$ are summarized in appendix C. Let us consider pure YM on $\tilde{G}$. In the Maurer-Cartan basis, the gauge potential is written as $A=X_{A}E^{A}$ where $E^{A}$ are the right invariant 1-forms on $\tilde{G}$ which are defined in (C.2). In this basis, the field strength is written as $\displaystyle F$ $\displaystyle=$ $\displaystyle dA+iA\wedge A$ (5.1) $\displaystyle=$ $\displaystyle\frac{1}{2}\left(f_{ABC}X_{C}+i{\cal L}_{A}X_{B}-i{\cal L}_{B}X_{A}+i[X_{A},X_{B}]\right)E^{A}\wedge E^{B},$ where we have used the Maurer-Cartan equation (C.3) and ${\cal L_{A}}$ are the right invariant Killing vectors on $\tilde{G}$ which are defined in (C.13). This is a counterpart of (2.20). Then, the original YM action on $\tilde{G}$ is rewritten as follows: $\displaystyle\frac{1}{g_{\tilde{G}}^{2}}\int{\rm tr}\left(\frac{1}{2}F\wedge*F\right)=\frac{1}{g_{\tilde{G}}^{2}}\int d^{D}z\sqrt{G}\;{\rm tr}\left\\{\frac{1}{4}\left(f_{ABC}X_{C}+i{\cal L}_{A}X_{B}-i{\cal L}_{B}X_{A}+i[X_{A},X_{B}]\right)^{2}\right\\},$ (5.2) where $D={\rm dim}(\tilde{G})$, $G=\det G_{MN}$ and $G_{MN}$ is the metric on $\tilde{G}$. Note that the gauge transformation in this basis is given by $\displaystyle X_{A}\rightarrow UX_{A}U^{-1}-{\cal L}_{A}U\>U^{-1}.$ (5.3) As explained in appendix C, if one drops the derivatives along the fiber direction in ${\cal L}_{A}$, these operators are reduced to the $L_{A}$ which are the Killing vectors on $\tilde{G}/H$ defined in (C.16). By dropping the derivatives along the fiber direction in ${\cal L}_{A}$ in (5.2), therefore, we can obtain the theory on $\tilde{G}/H$, $\displaystyle\frac{1}{g^{2}_{\tilde{G}}}\int{\rm tr}\left(\frac{1}{2}F\wedge*F\right)\rightarrow\frac{1}{g^{2}_{\tilde{G}/H}}\int d^{d}x\sqrt{g}\;{\rm tr}\left\\{\frac{1}{4}\left(f_{ABC}X_{C}+iL_{A}X_{B}-iL_{B}X_{A}+i[X_{A},X_{B}]\right)^{2}\right\\},$ (5.4) where $g^{2}_{\tilde{G}/H}=g^{2}_{\tilde{G}}/{\rm Vol}(H)$, $d=\dim{\tilde{G}/H}$, $g=\det g_{\mu\nu}$ and $g_{\mu\nu}$ is the metric on $\tilde{G}/H$. This is a counterpart of (2.18). The action (5.4) is also rewritten into the YM-higgs form which was obtained in section 3. The relation between the fields $X_{A}$ and the gauge and higgs fields on $\tilde{G}/H$ is given as follows. We introduce the orthogonal vectors to $L_{A}$ as $N_{A}^{a}={\rm Ad}(L(x))_{A}{}^{a},$ (5.5) where $L(x)$ is a representative element of $\tilde{G}/H$ which is defined in (C.1), and ${\rm Ad}$ represents the adjoint action: $gT^{A}g^{-1}=T^{B}{\rm Ad}(g)_{BA}$. One can show the orthonormality conditions, $L_{A}^{\mu}L_{A}^{\nu}=-g^{\mu\nu},\;\;N_{A}^{a}N_{A}^{b}=\delta^{ab},\;\;L_{A}^{\mu}N_{A}^{a}=0,$ (5.6) where $g^{\mu\nu}$ is the inverse of the metric on $\tilde{G}/H$. Furthermore, the following equalities hold: $\displaystyle L_{A}^{\mu}\partial_{\mu}N_{B}^{a}-L_{B}^{\mu}\partial_{\mu}N_{A}^{a}=-2iL_{A}^{\mu}L_{B}^{\nu}b_{\mu\nu}^{a}-f_{abc}(L_{A}^{\mu}N_{B}^{b}-L_{B}^{\mu}N_{A}^{b})b_{\mu}^{c},$ $\displaystyle f_{ABC}N_{C}^{a}-f_{abc}N_{A}^{b}N_{B}^{c}+L_{A}^{\mu}L_{B}^{\nu}b_{\mu\nu}^{a}=0.$ (5.7) We decompose $X_{A}$ into the gauge and higgs fields in terms of $L_{A}^{\mu}$ and $N_{A}^{a}$ as follows [33]: $X_{A}=iL_{A}^{\mu}a_{\mu}+N_{A}^{a}\phi_{a}.$ (5.8) This is a generalization of (2.16). Then, each term in the action (5.4) is rewritten as $\displaystyle f_{ABC}X_{C}$ $\displaystyle=$ $\displaystyle if_{ABC}L_{C}^{\mu}a_{\mu}-L_{A}^{\mu}L_{B}^{\nu}b_{\mu\nu}^{a}\phi_{a}+f_{abc}N_{A}^{b}N_{B}^{c}\phi_{a},$ $\displaystyle iL_{A}X_{B}-iL_{B}X_{A}$ $\displaystyle=$ $\displaystyle- if_{ABC}L_{C}^{\mu}a_{\mu}-L_{A}^{\mu}L_{B}^{\nu}(\partial_{\mu}a_{\nu}-\partial_{\nu}a_{\mu}-2b_{\mu\nu}^{a}\phi_{a})$ $\displaystyle+i(L_{A}^{\mu}N_{B}^{a}-L_{B}^{\mu}N_{A}^{a})(\partial_{\mu}\phi_{a}-f_{abc}b_{\mu}^{b}\phi_{c}),$ $\displaystyle i[X_{A},X_{B}]$ $\displaystyle=$ $\displaystyle- iL_{A}^{\mu}L_{B}^{\nu}[a_{\mu},a_{\nu}]-(L_{A}^{\mu}N_{B}^{a}-L_{B}^{\mu}N_{A}^{a})[a_{\mu},\phi_{a}]+iN_{A}^{a}N_{B}^{b}[\phi_{b},\phi_{a}],$ (5.9) where we have used (5.7). By substituting these equations into the action (5.4) and using (5.6), we indeed obtain the YM-higgs type action (3.20), $\displaystyle S_{\tilde{G}/H}=\frac{1}{g^{2}_{\tilde{G}/H}}\int d^{d}x\sqrt{g}\;{\rm tr}\Biggl{\\{}\frac{1}{4}\left(f_{\mu\nu}-b_{\mu\nu}^{a}\phi_{a}\right)^{2}+\frac{1}{2}\left(D_{\mu}\phi_{a}-f_{abc}b_{\mu}^{b}\phi_{c}\right)^{2}$ $\displaystyle+\frac{1}{4}\left(f_{abc}\phi_{c}+i[\phi_{a},\phi_{b}]\right)^{2}\Biggr{\\}}.$ (5.10) Finally, we consider the case in which $P=\tilde{G}$ and the base manifold is just a point. This is the special case of the above dimensional reduction in which $H$ equals $\tilde{G}$ itself. In this case, the theory on the base space is given by a zero-dimensional matrix model. Dropping all the derivatives in (5.2), we can easily make a dimensional reduction to the matrix model: $\displaystyle\frac{1}{g^{2}_{\tilde{G}}}\int{\rm tr}\left(\frac{1}{2}F\wedge*F\right)\rightarrow\frac{1}{g^{2}_{mm}}\;{\rm tr}\left\\{\frac{1}{4}\left(f_{ABC}X_{C}+i[X_{A},X_{B}]\right)^{2}\right\\},$ (5.11) where $g^{2}_{mm}=g^{2}_{\tilde{G}}/{\rm Vol}(\tilde{G})$. This is a counterpart of (2.19). Of course, we can obtain the matrix model (5.11) also from the theory (5.4) on $\tilde{G}/H$ by dropping the derivatives $L_{A}$. If we regard the original YM on $\tilde{G}$ as YM on a principal $\tilde{G}$ bundle over a point, we obtain (5.11) as a special case of (3.20). ### 5.2 Dimensional reduction of YM theory on $SU(n+1)$ In this subsection, we derive the YM-higgs on $S^{2n+1}$ and on $CP^{n}$ by applying the dimensional reduction discussed in the previous subsection. We also derive the 0-dimensional matrix model in which the YM-higgs on $S^{2n+1}$ and on $CP^{n}$ will be realized. Let us consider the group manifold $SU(n+1)$. We can apply the dimensional reduction developed in section 5.1 to the case of $P=\tilde{G}=SU(n+1)$ and obtain a theory on a coset space $SU(n+1)/H$, where $H$ is a subgroup of $SU(n+1)$. We begin with pure YM on the group manifold $SU(n+1)$ in the Maurer-Cartan basis, $S_{SU(n+1)}=\frac{1}{g^{2}_{SU(n+1)}}\int d^{n(n+2)}z\sqrt{G}\;{\rm tr}\left\\{\frac{1}{4}\left(f_{ABC}X_{C}+i{\cal L}_{A}X_{B}-i{\cal L}_{B}X_{A}+i[X_{A},X_{B}]\right)^{2}\right\\},$ (5.12) where $f_{ABC}$ is the structure constant of $SU(n+1)$, $G=\det G_{MN}$ and $G_{MN}$ is the Cartan-Killing metric on $\tilde{G}$ which is defined in (C.10). Let us consider the dimensional reduction of the above theory to a theory on $\tilde{G}/H$. If we take $H$ to be $SU(n)$, the coset space is given by $SU(n+1)/SU(n)\simeq S^{2n+1}$. By applying the dimensional reduction (5.4) to YM on $SU(n+1)$, therefore, we obtain the YM-higgs theory on $S^{2n+1}$, $S_{S^{2n+1}}=\frac{1}{g^{2}_{S^{2n+1}}}\int d^{2n+1}\tilde{x}\sqrt{\tilde{g}}\;{\rm tr}\left\\{\frac{1}{4}\left(f_{ABC}X_{C}+i\tilde{L}_{A}X_{B}-i\tilde{L}_{B}X_{A}+i[X_{A},X_{B}]\right)^{2}\right\\},$ (5.13) where $\tilde{g}$ represents the determinant of the metric on $S^{2n+1}$, and $\tilde{L}_{A}$’s are the Killing vectors on $S^{2n+1}$. Note that $S^{2n+1}$ that we consider here possesses only $SU(n+1)$ isometry which is smaller than $SO(2n+2)$. In fact, this is not the ordinary round sphere but a squashed sphere. In the case of $n=2$, the metric of this squashed $S^{5}$ is explicitly given in appendix D. Next, we consider the case of $H=SU(n)\times U(1)$. In this case, the coset space is $SU(n+1)/(SU(n)\times U(1))\simeq CP^{n}$. Then, we can obtain the theory on $CP^{n}$ from YM on $SU(n+1)$ through the dimensional reduction, $S_{CP^{n}}=\frac{1}{g^{2}_{CP^{n}}}\int d^{2n}x\sqrt{g}\;{\rm tr}\left\\{\frac{1}{4}\left(f_{ABC}X_{C}+iL_{A}X_{B}-iL_{B}X_{A}+i[X_{A},X_{B}]\right)^{2}\right\\}.$ (5.14) As in the case of $S^{2n+1}$, $g=\det g_{\mu\nu}$, $g_{\mu\nu}$ and $L_{A}$ represent the metric and the Killing vectors on $CP^{n}$ respectively. The theory (5.14) can be obtained also from the theory (5.13) by dropping the derivative along the extra $U(1)$ fiber direction. We can also rewrite (5.13) and (5.14) into the YM-higgs type actions as in (5.10) by using the relation (5.8). For example, (5.14) is rewritten into (5.10) with $\mu,\nu=1,\cdots,2n$ and $a,b,c=0,\cdots,n^{2}-1$. Here, $a,b,c$ are indices of $SU(n)\times U(1)$ and $a=0$ corresponds to the $U(1)$ direction. Finally, we consider the case in which $H$ is $SU(n+1)$ itself. In this case, the coset space is just a point. Then, we obtain the following matrix model by using (5.11): $S_{mm}=\frac{1}{g^{2}_{mm}}\;{\rm tr}\left\\{\frac{1}{4}\left(f_{ABC}X_{C}+i[X_{A},X_{B}]\right)^{2}\right\\}.$ (5.15) This theory is used to realize the theories (5.13) and (5.14) in the next subsection. For $n=1$, the dimensional reductions in this subsection are equivalent to those in section 2. ### 5.3 Relations among gauge theories on $SU(n+1)/H$ In this subsection, we show that the theory (5.14) in a monopole background can be realized by taking the commutative limit of the theory around a nontrivial background of (5.15). Combining this construction and the matrix T-duality, we also show that the theory (5.13) on $S^{2n+1}$ can be realized as the theory around a certain background of the matrix model with the orbifolding condition imposed. Furthermore, we apply the extended matrix T-duality developed in section 4 to YM-higgs on $SU(n+1)/(SU(2)^{k}\times U(1)^{l})$ and show that YM on $SU(n+1)$ is equivalent to the theory around a certain vacuum of YM-higgs on $SU(n+1)/(SU(2)^{k}\times U(1)^{l})$ with the periodicity condition imposed. For $n=2$, we obtain YM on $SU(3)$ from YM- higgs on $S^{5}$ and on $CP^{2}$ through the extended matrix T-duality. First, we review nontrivial backgrounds of the theory (5.14) on $CP^{n}$ and the matrix model (5.15). The theory on $CP^{n}$ has many nontrivial monopole vacua. In particular, we focus on the $U(1)$ monopole background. Recall that we have $n^{2}$ higgs fields $\phi_{a}$. In the $U(1)$ monopole background, only the higgs field along the $U(1)$ direction $\phi_{0}$ acquires its nonzero vacuum expectation value. In the gauge where $\phi_{0}$ is diagonal, the vacuum configurations of the $U(1)$ monopole with the gauge group $U(M)$ are given by $\displaystyle\hat{a}_{\mu}$ $\displaystyle=b_{\mu}^{0}\hat{\phi}_{0},$ $\displaystyle\hat{\phi}_{0}$ $\displaystyle=-\frac{1}{\sqrt{2n(n+1)}}{\rm diag}(\cdots,\underbrace{n_{s-1},\cdots,n_{s-1}}_{N_{s-1}},\underbrace{n_{s},\cdots,n_{s}}_{N_{s}},\underbrace{n_{s+1},\cdots,n_{s+1}}_{N_{s+1}},\cdots),$ $\displaystyle\hat{\phi}_{a}$ $\displaystyle=0,\;\;({\rm for}\;\;a\neq 0).$ (5.16) Here, $\sum_{s}N_{s}=M$ and $n_{s}$ must be integers due to Dirac’s quantization condition. Because of (5.8), the vacuum configurations of $X_{A}$ are equivalently given by $\hat{X}_{A}=-\frac{iL_{A}^{\mu}b_{\mu}^{0}+N_{A}^{0}}{\sqrt{2n(n+1)}}\;{\rm diag}(\cdots,\underbrace{n_{s-1},\cdots,n_{s-1}}_{N_{s-1}},\underbrace{n_{s},\cdots,n_{s}}_{N_{s}},\underbrace{n_{s+1},\cdots,n_{s+1}}_{N_{s+1}},\cdots).$ (5.17) The theory around the background (5.17) is obtained by expanding each block of the fields in (5.14) as $X_{A}^{(s,t)}\rightarrow\hat{X}_{A}^{(s,t)}+X_{A}^{(s,t)}$. Then, the following action is obtained, $\displaystyle\frac{1}{g^{2}_{CP^{n}}}\int d^{2n}x\sqrt{g}\sum_{s,t}\;{\rm tr}\Big{\\{}\frac{1}{4}$ $\displaystyle\left(f_{ABC}X^{(s,t)}_{C}+iL_{A}^{(q_{st})}X_{B}^{(s,t)}-iL_{B}^{(q_{st})}X_{A}^{(s,t)}+i[X_{A},X_{B}]^{(s,t)}\right)$ $\displaystyle\times$ $\displaystyle\left(f_{ABD}X^{(t,s)}_{D}+iL_{A}^{(q_{ts})}X_{B}^{(t,s)}-iL_{B}^{(q_{ts})}X_{A}^{(t,s)}+i[X_{A},X_{B}]^{(t,s)}\right)\Big{\\}},$ (5.18) where $q_{st}=\frac{n_{s}-n_{t}}{2}$ and $L_{A}^{(q)}$ are the angular momentum operators in the presence of a monopole with the magnetic charge $q$, which take the form $L_{A}^{(q)}=L_{A}+\frac{2q}{\sqrt{2n(n+1)}}(iL_{A}^{\mu}b_{\mu}^{0}+N_{A}^{0}).$ (5.19) These operators are the generalization of (2.27) in the case of $S^{2}$. The vacua of the theory (5.15) are determined by $[X_{A},X_{B}]=if_{ABC}X_{C}.$ (5.20) In addition to the trivial solution $X_{A}=0$, there are nontrivial solutions which are given by the representation matrices of the $SU(n+1)$ generators, $\hat{X}_{A}=\hat{L}_{A}.$ (5.21) $\hat{L}_{A}$ are generally in a reducible representation. In order to construct a theory on $CP^{n}$ in a $U(1)$ monopole background, we consider the following representation: $\displaystyle\hat{L}_{A}=\begin{pmatrix}\rotatebox[origin={tl}]{-35.0}{$\cdots\;\;\;\overbrace{\rotatebox[origin={c}]{35.0}{$\hat{L}_{A}^{(s-1)}$}\;\cdots\;\rotatebox[origin={c}]{35.0}{$\hat{L}_{A}^{(s-1)}$}}^{\rotatebox{35.0}{$N_{s-1}$}}\;\;\;\overbrace{\rotatebox[origin={c}]{35.0}{$\hat{L}_{A}^{(s)}$}\;\cdots\;\rotatebox[origin={c}]{35.0}{$\hat{L}_{A}^{(s)}$}}^{\rotatebox{35.0}{$N_{s}$}}\;\;\;\overbrace{\rotatebox[origin={c}]{35.0}{$\hat{L}_{A}^{(s+1)}$}\;\cdots\;\rotatebox[origin={c}]{35.0}{$\hat{L}_{A}^{(s+1)}$}}^{\rotatebox{35.0}{$N_{s+1}$}}\;\;\;\cdots$}\end{pmatrix}.$ (5.22) Here $\hat{L}_{A}^{(s)}$ are the abbreviations of $\hat{L}_{A}^{[\Lambda_{s},0,\cdots,0]}$ which are the generators of $SU(n+1)$ in the irreducible representation specified by the Dynkin index of $SU(n+1)$, $[\Lambda_{s},0,\cdots,0]$. We consider the matrix model (5.15) around the background (5.22) by expanding the each block of the fields around the background :$X_{A}^{(s,t)}\rightarrow\hat{X}_{A}^{(s,t)}+X_{A}^{(s,t)}$. Then, the action takes the following form: $\displaystyle S_{mm}=\frac{1}{g^{2}_{mm}}\sum_{s,t}\;{\rm tr}\Big{\\{}\frac{1}{4}$ $\displaystyle\left(f_{ABC}X_{C}^{(s,t)}+i\hat{L}_{A}\circ X_{B}^{(s,t)}-i\hat{L}_{B}\circ X_{A}^{(s,t)}+i[X_{A},X_{B}]^{(s,t)}\right)$ $\displaystyle\times$ $\displaystyle\left(f_{ABD}X_{D}^{(t,s)}+i\hat{L}_{A}\circ X_{B}^{(t,s)}-i\hat{L}_{B}\circ X_{A}^{(t,s)}+i[X_{A},X_{B}]^{(t,s)}\right)\Big{\\}}.$ (5.23) $\hat{L}_{A}\circ$ are defined as $\hat{L}_{A}\circ X_{B}^{(s,t)}\equiv\hat{L}_{A}^{(s)}X_{B}^{(s,t)}-X_{B}^{(s,t)}\hat{L}_{A}^{(t)}.$ (5.24) We show in the following that the theory (5.23) is equivalent to the theory (5.18) if we put $\Lambda_{s}=N_{0}+n_{s}$ and take $N_{0}\rightarrow\infty$ limit. In order to show this equivalence, we make a harmonic expansion [31, 32, 35]. As explained in appendix E, the $(s,t)$ blocks $X_{A}^{(s,t)}$ in the matrix model are expanded by the basis of rectangular matrices (E.21) as $X_{A}^{(s,t)}=\sum_{J=|q_{st}|}^{(\Lambda_{s}+\Lambda_{t})/2}X_{A}^{(s,t)}{}^{{\bm{\beta}}_{J+q_{st}}}{}_{{\bm{\alpha}}_{J-q_{st}}}\otimes\hat{Y}^{(q_{st})}{}_{{\bm{\beta}}_{J+q_{st}}}{}^{{\bm{\alpha}}_{J-q_{st}}}.$ (5.25) Then, the diagonal coherent map allows us to map the $(s,t)$ blocks to local sections of the monopole bundle on $CP^{n}$ with the charge $q_{st}$, $\displaystyle X_{A}^{(s,t)}$ $\displaystyle=\sum_{J=q_{st}}^{\infty}X_{A}^{(q_{st})}{}^{{\bm{\beta}}_{J+q_{st}}}{}_{{\bm{\alpha}}_{J-q_{st}}}\otimes\hat{Y}^{(q_{st})}{}_{{\bm{\beta}}_{J+q_{st}}}{}^{{\bm{\alpha}}_{J-q_{st}}}$ $\displaystyle\rightarrow\sum_{J=q_{st}}^{\infty}X_{A}^{(q_{st})}{}^{{\bm{\beta}}_{J+q_{st}}}{}_{{\bm{\alpha}}_{J-q_{st}}}\tilde{Y}^{(q_{st})}{}_{{\bm{\beta}}_{J+q_{st}}}{}^{{\bm{\alpha}}_{J-q_{st}}}(w,\bar{w})=X_{A}^{CP(s,t)}(w,\bar{w}),$ (5.26) where we have taken the commutative limit $N_{0}\rightarrow\infty$ and $\tilde{Y}^{(q_{st})}{}_{{\bm{\beta}}_{J+q_{st}}}{}^{{\bm{\alpha}}_{J-q_{st}}}$ are the basis of local sections of the $U(1)$ monopole bundle on $CP^{n}$ which are defined in (E.26). Note that we have put the superscript $CP$ on the quantity in the righthand side of the above equation in order to emphasis that the $X_{A}^{CP(s,t)}$ are the fields on $CP^{n}$ appearing in (5.18). Similarly, $\hat{L}_{A}\circ$ is mapped to $L_{A}^{(q)}$ as shown in (E.31)444In [32], (5.26) and (5.27) are proven to the quadratic order in the fields for all $q$ and to all order for $q=0$. In this paper, we assume that these are also valid to all order for all $q$.: $\hat{L}_{A}\circ X^{(s,t)}_{B}\rightarrow L_{A}^{(q)}X^{CP(s,t)}_{B}(w,\bar{w}).$ (5.27) Using (5.26) and (5.27), we find that the matrix model (5.23) is equivalent to the theory (5.18) on $CP^{n}$ in the commutative limit $N_{0}\rightarrow\infty$. Next, we show that the theory around a certain vacuum of $U(M=N\times\infty)$ YM-higgs on $CP^{n}$ with a periodicity condition imposed is equivalent to $U(N)$ YM-higgs on $S^{2n+1}$. This statement is nothing but the matrix T-duality. As explained in section 4, therefore, we consider the appropriate vacuum which is given by (5.17) (or equivalently (5.16)) with $s$ running from $-\infty$ to $\infty$, $n_{s}=s$ and $N_{s}=N$. We expand the fields on $CP^{n}$ around the background as $\displaystyle X_{A}\rightarrow\hat{X}_{A}+X_{A},$ (5.28) and impose the periodicity (orbifolding) condition on the fluctuation, $\displaystyle X_{A}^{(s+1,t+1)}=X_{A}^{(s,t)}\equiv X_{A}^{(s-t)}.$ (5.29) Then, we define the gauge and higgs fields on $S^{2n+1}$ by the Fourier transforms of the fluctuations on each local coordinate patch: $\displaystyle X_{A}^{S}=\sum_{u}X_{A}^{CP(u)}e^{-iuy},$ (5.30) where $y$ is a coordinate which parameterizes the fiber ($U(1)$) direction and satisfies $0\leq\tau\leq 2\pi$. Here, the superscripts $S$ and $CP$ indicate that $X_{A}^{S}$ and $X_{A}^{CP(w)}$ are the fields on $S^{2n+1}$ and $CP^{n}$ respectively. We substitute (5.30) into (5.18) and divide an overall factor $\sum_{s}$ to extract a single period. Then, we obtain $U(N)$ YM-higgs on $S^{2n+1}$ written in the basis of $X_{A}$ (5.13). Combining the above matrix T-duality and the construction of (5.18) in terms of the matrix model, we find that the theory around (5.22) of the matrix model, where $s$ runs from $-\infty$ to $\infty$ and $\Lambda_{s}=N_{0}+s$, is equivalent to $U(N)$ YM-higgs on $S^{2n+1}$ if we take the limit $N_{0}\rightarrow\infty$, impose the periodicity condition on the fluctuations, and finally divide the overall factor $\sum_{s}$. Finally, it is straightforward to apply the extended matrix T-duality to $SU(2)^{k}\times U(1)^{l}$ bundle on $SU(n+1)/(SU(2)^{k}\times U(1)^{l})$ and show that YM on $SU(n+1)$ is equivalent to the theory around a certain vacuum of YM-higgs on $SU(n+1)/(SU(2)^{k}\times U(1)^{l})$ with the periodicity condition imposed. ## 6 Interpretation as Buscher’s T-duality In this section, let us see that the extended matrix T-duality of the $U(1)$ case, which was obtained in [29] and reviewed in section 4.2, is actually interpreted as the T-duality in Buscher’s sense. We put $\mbox{dim}M=p$. For $G=U(1)$, as in (4.6), the metric of the total space is given by $\displaystyle ds^{2}=G_{MN}dz^{M}dz^{N}=g_{\mu\nu}dx^{\mu}dx^{\nu}+(dy- b_{\mu}dx^{\mu})^{2},$ (6.1) where $M,N=1,\cdots,p+1$ and $\mu,\nu=1,\cdots,p$. We assume that the other fields such as the antisymmetric fields and the dilaton field are trivial. Then, YM on the total space is viewed as the low energy effective theory for the D$p$-branes wrapped on the total space555Here we ignore the transverse directions.. We make the T-duality transformation for the fiber direction to obtain a new geometry [36]: $\displaystyle{ds^{\prime}}^{2}=g_{\mu\nu}dx^{\mu}dx^{\nu}+dy^{2},$ $\displaystyle{B^{\prime}}_{\mu\nu}=0,\;\;\;{B^{\prime}}_{\mu y}=-b_{\mu}.$ (6.2) The D$p$-branes should be transformed to the D$(p-1)$-branes wrapped on the base space. The D$(p-1)$-brane effective action on the new geometry (6.2) is given by $\displaystyle S_{p-1}=\tau_{p-1}\int d^{p}\sigma e^{-\Phi}\sqrt{\det(\tilde{G}_{ab}+\tilde{B}_{ab}+2\pi\alpha^{\prime}F_{ab})},$ (6.3) where $\sigma^{a}\;(a=1,\cdots,p)$ parameterize the world volume of the D$(p-1)$-brane, and $\tilde{G}_{ab}$ and $\tilde{B}_{ab}$ are the pullback of (6.2) on the world volume which is defined through the embedding of world volume $z^{M}(\sigma)$ as $\displaystyle\tilde{G}_{ab}=\frac{\partial z^{M}}{\partial\sigma^{a}}\frac{\partial z^{N}}{\partial\sigma^{b}}{G^{\prime}}_{MN},$ $\displaystyle\tilde{B}_{ab}=\frac{\partial z^{M}}{\partial\sigma^{a}}\frac{\partial z^{N}}{\partial\sigma^{b}}{B^{\prime}}_{MN}.$ (6.4) In the static gauge $x^{\mu}(\sigma)=\sigma^{\mu}$ and $z^{y}(\sigma)=2\pi\alpha^{\prime}\phi$, (6.3) reduces to $\displaystyle S_{p-1}=\tau_{p-1}\int d^{p}x\sqrt{\det(g_{\mu\nu}+(2\pi\alpha^{\prime})^{2}\partial_{\mu}\phi\partial_{\nu}\phi+2\pi\alpha^{\prime}(F_{\mu\nu}+\partial_{\mu}\phi b_{\nu}-\partial_{\nu}\phi b_{\mu}))}.$ (6.5) Up to ${\cal O}({\alpha^{\prime}}^{3})$, this equals $\displaystyle\frac{1}{g_{YM}^{2}}\int d^{p}x\sqrt{g}\left(\frac{1}{4}(F_{\alpha\beta}+\nabla_{\alpha}\phi b_{\beta}-\nabla_{\beta}\phi b_{\alpha})^{2}+\frac{1}{2}(\nabla_{\alpha}\phi)^{2}\right),$ (6.6) where $g_{YM}^{2}=\frac{1}{4\pi\alpha^{2}\tau_{p-1}}$. If we redefine the gauge field as $a_{\alpha}\rightarrow a_{\alpha}+b_{\alpha}\phi$ and make non- abelianization, we obtain from (6.6) $\displaystyle\frac{1}{g_{YM}^{2}}\int d^{p}x\sqrt{g}\mbox{tr}\left(\frac{1}{4}(F_{\alpha\beta}-b_{\alpha\beta}\phi)^{2}+\frac{1}{2}(D_{\alpha}\phi)^{2}\right),$ (6.7) which indeed agrees with (3.20) with $G=U(1)$. ## Acknowledgements We would like to thank T. Higashi for discussions. The work of G.I. is supported in part by the JSPS Research Fellowship for Young Scientists. The work of A.T. is supported in part by Grant-in-Aid for Scientific Research (No. 19540294) from the Ministry of Education, Culture, Sports, Science and Technology. ## Appendix A Spherical harmonics In this appendix, we review the spherical harmonics on $S^{3}$, the monopole harmonics on $S^{2}$ [42] and the fuzzy spherical harmonics [24, 49, 15]. For more details, see [48, 29] and references therein. ### A.1 Spherical harmonics on $S^{3}$ We regard $S^{3}$ as the $SU(2)$ group manifold. We parameterize an element of $SU(2)$ in terms of the Euler angles as $g=e^{-i\varphi J_{3}}e^{-i\theta J_{2}}e^{-i\psi J_{3}},$ (A.1) where $J_{A}$ satisfy $[J_{A},J_{B}]=i\epsilon_{ABC}J_{C}$ and $0\leq\theta\leq\pi$, $0\leq\varphi<2\pi$, $0\leq\psi<4\pi$. The isometry of $S^{3}$ is $SO(4)=SU(2)\times SU(2)$, and these two $SU(2)$’s act on $g$ from left and right, respectively. We construct the right invariant 1-forms, $dgg^{-1}=-i\mu E^{A}J_{A},$ (A.2) where the radius of $S^{3}$ is $2/\mu$. They are explicitly given by $\displaystyle E^{1}=\frac{1}{\mu}(-\sin\varphi d\theta+\sin\theta\cos\varphi d\psi),$ $\displaystyle E^{2}=\frac{1}{\mu}(\cos\varphi d\theta+\sin\theta\sin\varphi d\psi),$ $\displaystyle E^{3}=\frac{1}{\mu}(d\varphi+\cos\theta d\psi),$ (A.3) and satisfy the Maure-Cartan equation $dE^{A}-\frac{\mu}{2}\epsilon_{ABC}E^{B}\wedge E^{C}=0.$ (A.4) The metric is constructed from $E^{A}$ as $ds^{2}=E^{A}E^{A}=\frac{1}{\mu^{2}}\left(d\theta^{2}+\sin^{2}\theta d\varphi^{2}+(d\psi+\cos d\varphi)^{2}\right).$ (A.5) The Killing vectors dual to $E^{A}$ are given by ${\cal{L}}_{A}=-\frac{i}{\mu}E^{M}_{A}\partial_{M},$ (A.6) where $E^{M}_{A}$ are inverse of $E^{A}_{M}$. The explicit form of the Killing vectors are $\displaystyle{\cal{L}}_{1}=-i\left(-\sin\varphi\partial_{\theta}-\cot\theta\cos\varphi\partial_{\varphi}+\frac{\cos\varphi}{\sin\theta}\partial_{\psi}\right),$ $\displaystyle{\cal{L}}_{2}=-i\left(\cos\varphi\partial_{\theta}-\cot\theta\sin\varphi\partial_{\varphi}+\frac{\sin\varphi}{\sin\theta}\partial_{\psi}\right),$ $\displaystyle{\cal{L}}_{3}=-i\partial_{\varphi}.$ (A.7) Because of the Maure-Cartan equation (A.4), the Killing vectors satisfy the SU(2) algebra, $[{\cal{L}}_{A},{\cal{L}}_{B}]=i\epsilon_{ABC}{\cal{L}}_{C}$. The scalar spherical harmonics on $S^{3}$ are given by $Y_{Jm\tilde{m}}(\Omega_{3})=(-1)^{J-\tilde{m}}\sqrt{2J+1}\langle J-\tilde{m}|g^{-1}|Jm\rangle.$ (A.8) These spherical harmonics form the basis of SU(2) algebra generated by ${\cal{L}}_{A}$’s. $\displaystyle{\cal{L}}^{2}Y_{Jm\tilde{m}}$ $\displaystyle=$ $\displaystyle J(J+1)Y_{Jm\tilde{m}},$ $\displaystyle{\cal{L}}_{\pm}Y_{Jm\tilde{m}}$ $\displaystyle=$ $\displaystyle\sqrt{(J\mp m)(J\pm m+1)}Y_{Jm\pm 1\tilde{m}},$ $\displaystyle{\cal{L}}_{3}Y_{Jm\tilde{m}}$ $\displaystyle=$ $\displaystyle mY_{Jm\tilde{m}}.$ (A.9) The complex conjugates of the spherical harmonics are evaluated as $\left(Y_{Jm\tilde{m}}\right)^{*}=(-1)^{m-\tilde{m}}Y_{J-m-\tilde{m}}.$ (A.10) The spherical harmonics also satisfy the orthonormality condition $\int\frac{d\Omega_{3}}{2\pi^{2}}\left(Y_{Jm\tilde{m}}\right)^{*}Y_{J^{\prime}m^{\prime}\tilde{m}^{\prime}}=\delta_{JJ^{\prime}}\delta_{mm^{\prime}}\delta_{\tilde{m}\tilde{m}^{\prime}}.$ (A.11) The integral of the product of three spherical harmonics is given as follows: $\displaystyle{\cal{C}}^{J_{1}m_{1}\tilde{m}_{1}}_{J_{2}m_{2}\tilde{m}_{2}J_{3}m_{3}\tilde{m}_{3}}$ $\displaystyle\equiv$ $\displaystyle\int\frac{d\Omega_{3}}{2\pi^{2}}\left(Y_{J_{1}m_{1}\tilde{m}_{1}}\right)^{*}Y_{J_{2}m_{2}\tilde{m}_{2}}Y_{J_{3}m_{3}\tilde{m}_{3}}$ (A.12) $\displaystyle=$ $\displaystyle\sqrt{\frac{(2J_{2}+1)(2J_{3}+1)}{2J_{1}+1}}C^{J_{1}m_{1}}_{J_{2}m_{2}J_{3}m_{3}}C^{J_{1}\tilde{m}_{1}}_{J_{2}\tilde{m}_{2}J_{3}\tilde{m}_{3}},$ where $C^{J_{1}m_{1}}_{J_{2}m_{2}J_{3}m_{3}}$ is the Clebsch-Gordan coefficient of $SU(2)$. Finally, the spherical harmonics satisfy the completeness condition, $\displaystyle\sum_{Jm\tilde{m}}\left(Y_{Jm\tilde{m}}\right)^{*}(\Omega_{3})Y_{Jm\tilde{m}}(\Omega_{3}^{\prime})=2\pi^{2}\delta(\Omega_{3}-\Omega_{3}^{\prime}),$ (A.13) where $\delta(\Omega_{3})=\frac{8}{\sin\theta}\delta(\theta)\delta(\varphi)\delta(\psi).$ (A.14) ### A.2 Monopole spherical harmonics on $S^{2}$ We adopt the following metric for $S^{2}$: $ds^{2}=\frac{1}{\mu^{2}}(d\theta^{2}+\sin^{2}\theta d\varphi^{2}).$ (A.15) We define two local patches on $S^{2}$ to describe nontrivial $U(1)$ bundles over $S^{2}$: the patch I is specified by $0\leq\theta<\pi$ and the patch II is specified by $0<\theta\leq\pi$. In the following expressions, the upper sign is taken in the patch I and the lower sign in the patch II. The angular momentum operator in the presence of a monopole with magnetic charge $q$ at the origin takes the form $\displaystyle L_{1}^{(q)}=i(\sin\varphi\partial_{\theta}+\cot\theta\cos\varphi\partial_{\varphi})-q\frac{1\mp\cos\theta}{\sin\theta}\cos\varphi,$ $\displaystyle L_{2}^{(q)}=i(-\cos\varphi\partial_{\theta}+\cot\theta\sin\varphi\partial_{\varphi})-q\frac{1\mp\cos\theta}{\sin\theta}\sin\varphi,$ $\displaystyle L_{3}^{(q)}=-i\partial_{\varphi}\mp q,$ (A.16) where $q$ is quantized as $q=0,\pm\frac{1}{2},\pm 1,\pm\frac{3}{2},\cdots$. These operators act on the local sections on $S^{2}$ and satisfy the $SU(2)$ algebra $[L_{A}^{(q)},L_{B}^{(q)}]=i\epsilon_{ABC}L_{C}^{(q)}$. Note that when $q=0$, these operators are reduced to the ordinary angular momentum operators on $S^{2}$ (or $R^{3}$). if we regard $S^{3}$ as a $U(1)$ bundle over $S^{2}$, and parameterize the fiber direction by $y=\psi\pm\varphi$, the above expression (A.16) can be obtained by making a replacement in (A.7): $\partial_{y}\rightarrow-iq$. The monopole spherical harmonics are the basis of local sections on $S^{2}$ and also form the basis of the $SU(2)$ algebra generated by $L_{A}^{(q)}$. The monopole scalar spherical harmonics are given by $\tilde{Y}_{Jmq}(\Omega_{2})=(-1)^{J-q}\sqrt{2J+1}\langle J-q|e^{i\theta J_{2}}|Jm\rangle e^{i(\pm q+m)\varphi}.$ (A.17) Here $J=|q|,|q|+1,|q+2|,\cdots$, $m=-J,-J+1,\cdots,J-1,J$. The existence of the lower bound of the angular momentum $J$ is due to the fact that the magnetic field produced by the monopole also has nonzero angular momentum. Note that the monopole harmonics with $q=0$ do not transform on the overlap of two patches. They correspond to global sections (functions) on $S^{2}$ which are expressed by the ordinary spherical harmonics on $S^{2}$. The action of $L^{(q)}_{A}$ on the monopole spherical harmonics is given by $\displaystyle L^{(q)2}\tilde{Y}_{Jmq}$ $\displaystyle=$ $\displaystyle J(J+1)\tilde{Y}_{Jmq},$ $\displaystyle L^{(q)}_{\pm}\tilde{Y}_{Jmq}$ $\displaystyle=$ $\displaystyle\sqrt{(J\mp m)(J\pm m+1)}\tilde{Y}_{Jm\pm 1q},$ $\displaystyle L^{(q)}_{3}\tilde{Y}_{Jmq}$ $\displaystyle=$ $\displaystyle m\tilde{Y}_{Jmq}.$ (A.18) The complex conjugates of the monopole spherical harmonics are evaluated as $\left(\tilde{Y}_{Jmq}\right)^{*}=(-1)^{m-q}\tilde{Y}_{J-m-q}.$ (A.19) The monopole spherical harmonics are orthonormal to each other, $\int\frac{d\Omega_{2}}{4\pi}\left(\tilde{Y}_{Jmq}\right)^{*}\tilde{Y}_{J^{\prime}m^{\prime}q}=\delta_{JJ^{\prime}}\delta_{mm^{\prime}}.$ (A.20) The integral of three monopole spherical harmonics is equal to the corresponding integral (A.12) on $S^{3}$ with the identification $\tilde{m}=q$, $\displaystyle\int\frac{d\Omega_{2}}{4\pi}\left(\tilde{Y}_{J_{1}m_{1}q_{1}}\right)^{*}\tilde{Y}_{J_{2}m_{2}q_{2}}\tilde{Y}_{J_{3}m_{3}q_{3}}$ $\displaystyle=$ $\displaystyle\sqrt{\frac{(2J_{2}+1)(2J_{3}+1)}{2J_{1}+1}}C^{J_{1}m_{1}}_{J_{2}m_{2}J_{3}m_{3}}C^{J_{1}q_{1}}_{J_{2}q_{2}J_{3}q_{3}}$ (A.21) $\displaystyle=$ $\displaystyle{\cal{C}}^{J_{1}m_{1}q_{1}}_{J_{2}m_{2}q_{2}J_{3}m_{3}q_{3}},$ where the monopole charges must be conserved in the lefthand side of the above equation as $q_{1}+q_{2}+q_{3}=0$. Note that the monopole spherical harmonics are expressed in terms of the spherical harmonics on $S^{3}$: $\displaystyle\tilde{Y}_{Jmq}(\Omega_{2})$ $\displaystyle=$ $\displaystyle e^{iq(\psi\pm\varphi)}Y_{Jmq}(\Omega_{3}),$ $\displaystyle L^{(q)}_{A}\tilde{Y}_{Jmq}(\Omega_{2})$ $\displaystyle=$ $\displaystyle e^{iq(\psi\pm\varphi)}{\cal L}_{A}Y_{Jmq}(\Omega_{3}).$ (A.22) (A.21) and (A.22) represent a map between the local sections on $S^{2}$ and the Kaluza-Klein modes on $S^{3}$. ### A.3 Fuzzy spherical harmonics Let us consider $(2j+1)\times(2j^{\prime}+1)$ rectangular complex matrices. Such matrices are generally expressed as $M=\sum_{r,r^{\prime}}M_{rr^{\prime}}|jr\rangle\langle j^{\prime}r^{\prime}|.$ (A.23) We can define linear maps $\hat{L}_{A}\circ$, which map the set of $(2j+1)\times(2j^{\prime}+1)$ rectangular complex matrices to itself, by their operation on the basis: $\hat{L}_{A}\circ|jr\rangle\langle j^{\prime}r^{\prime}|\equiv\hat{L}_{A}^{[j]}|jr\rangle\langle j^{\prime}r^{\prime}|-|jr\rangle\langle j^{\prime}r^{\prime}|\hat{L}_{A}^{[j^{\prime}]},$ (A.24) where $\hat{L}_{A}^{[j]}$ are the spin $j$ representation matrices of the $SU(2)$ generators. $\hat{L}_{A}\circ$ satisfy the $SU(2)$ algebra $[\hat{L}_{A}\circ,\hat{L}_{B}\circ]=i\epsilon_{ABC}\hat{L}_{C}\circ$. We make a change of a basis of the rectangular matrices from the above basis $\\{|jr\rangle\langle j^{\prime}r^{\prime}|\\}$ to the new basis which is called the fuzzy spherical harmonics: $\hat{Y}_{Jm(jj^{\prime})}=\sqrt{N_{0}}\sum_{r,r^{\prime}}(-1)^{-j+r^{\prime}}C^{Jm}_{jr\;j^{\prime}-r^{\prime}}|jr\rangle\langle j^{\prime}r^{\prime}|,$ (A.25) where $N_{0}$ is a positive integer which will be specified below. For a fixed $J$ the fuzzy spherical harmonics also form a basis of the spin $J$ irreducible representation of $SU(2)$ which is generated by $\hat{L}_{A}\circ$, $\displaystyle(\hat{L}_{A}\circ)^{2}\hat{Y}_{Jm(jj^{\prime})}$ $\displaystyle=$ $\displaystyle J(J+1)\hat{Y}_{Jm(jj^{\prime})},$ $\displaystyle\hat{L}_{\pm}\circ\hat{Y}_{Jm(jj^{\prime})}$ $\displaystyle=$ $\displaystyle\sqrt{(J\mp m)(J\pm m+1)}\hat{Y}_{Jm\pm 1(jj^{\prime})},$ $\displaystyle\hat{L}_{3}\circ\hat{Y}_{Jm(jj^{\prime})}$ $\displaystyle=$ $\displaystyle m\hat{Y}_{Jm(jj^{\prime})}.$ (A.26) The hermitian conjugates of the fuzzy spherical harmonics are evaluated as $\left(\hat{Y}_{Jm(jj^{\prime})}\right)^{\dagger}=(-1)^{m-(j-j^{\prime})}\hat{Y}_{J-m(j^{\prime}j)}.$ (A.27) The fuzzy spherical harmonics satisfy the orthonormality condition under the following normalized trace: $\frac{1}{N_{0}}{\rm tr}\left\\{\left(\hat{Y}_{Jm(jj^{\prime})}\right)^{\dagger}\hat{Y}_{J^{\prime}m^{\prime}(jj^{\prime})}\right\\}=\delta_{JJ^{\prime}}\delta_{mm^{\prime}},$ (A.28) where ${\rm tr}$ stands for the trace over $(2j^{\prime}+1)\times(2j^{\prime}+1)$ matrices. The trace of three fuzzy spherical harmonics is given by $\displaystyle\hat{C}^{J_{1}m_{1}(jj^{\prime\prime})}_{J_{2}m_{2}(jj^{\prime})J_{3}m_{3}(j^{\prime}j^{\prime\prime})}$ $\displaystyle\equiv\frac{1}{N_{0}}{\rm tr}\left\\{\left(\hat{Y}_{J_{1}m_{1}(jj^{\prime\prime})}\right)^{\dagger}\hat{Y}_{J_{2}m_{2}(jj^{\prime})}\hat{Y}_{J_{3}m_{3}(j^{\prime}j^{\prime\prime})}\right\\}$ $\displaystyle=(-1)^{J_{1}+j+j^{\prime\prime}}\sqrt{N_{0}(2J_{2}+1)(2J_{3}+1)}C^{J_{1}m_{1}}_{J_{2}m_{2}J_{3}m_{3}}\left\\{\begin{array}[]{ccc}J_{1}&J_{2}&J_{3}\\\ j^{\prime}&j^{\prime\prime}&j\end{array}\right\\},$ (A.31) where the last factor of the above equation is the $6-j$ symbol. In order to reveal relationships among the fuzzy spherical harmonics, the monopole harmonics on $S^{2}$ and the spherical harmonics on $S^{3}$, we introduce the following parameterization for $j$, $j^{\prime}$ and $j^{\prime\prime}$, $2j+1=N_{0}+\zeta,\;\;2j^{\prime}+1=N_{0}+\zeta^{\prime},\;\;2j^{\prime\prime}+1=N_{0}+\zeta^{\prime\prime}.$ (A.32) $\zeta$, $\zeta^{\prime}$ and $\zeta^{\prime\prime}$ are integers which are grater than $-N_{0}$. Then, in the limit $N_{0}\rightarrow\infty$, one can show that $\hat{C}^{J_{1}m_{1}(jj^{\prime\prime})}_{J_{2}m_{2}(jj^{\prime})J_{3}m_{3}(j^{\prime}j^{\prime\prime})}\rightarrow{\cal{C}}^{J_{1}m_{1}q_{1}}_{J_{2}m_{2}q_{2}J_{3}m_{3}q_{3}}$ (A.33) with the identification $j-j^{\prime\prime}=q_{1}$, $j-j^{\prime}=q_{2}$ and $j^{\prime}-j^{\prime\prime}=q_{3}$. This relation can be proved by using the following asymptotic form of the $6-j$ symbols. If $R\gg 1$, one obtains [50] $\left\\{\begin{array}[]{ccc}a&b&c\\\ d+R&e+R&f+R\\\ \end{array}\right\\}\approx\frac{(-1)^{a+b+c+2(d+e+f+R)}}{\sqrt{2R}}\left(\begin{array}[]{ccc}a&b&c\\\ e-f&f-d&d-e\\\ \end{array}\right),$ (A.34) where the $3-j$ symbol is related to the Clebsch-Gordan coefficient as $\left(\begin{array}[]{ccc}J_{1}&J_{2}&J_{3}\\\ m_{1}&m_{2}&m_{3}\\\ \end{array}\right)=(-1)^{J_{3}+m_{3}+2J_{1}}\frac{1}{\sqrt{2J_{3}+1}}C^{J_{3}m_{3}}_{J_{1}-m_{1}J_{2}-m_{2}}.$ (A.35) The relation (A.33) implies that the fuzzy spherical harmonics $\hat{Y}_{Jm(jj^{\prime})}$ give a matrix regularization of the monopole harmonics $\tilde{Y}_{Jmq}$ through the following correspondence: $\displaystyle j-j^{\prime}$ $\displaystyle\leftrightarrow$ $\displaystyle q,$ $\displaystyle\hat{L}_{A}\circ$ $\displaystyle\leftrightarrow$ $\displaystyle L_{A}^{(q)},$ $\displaystyle\frac{1}{N_{0}}{\rm tr}$ $\displaystyle\leftrightarrow$ $\displaystyle\int\frac{d\Omega_{2}}{4\pi}.$ (A.36) Furthermore, combining the above correspondence and the relations (A.21) and (A.22), we can also map the fuzzy spherical harmonics to the spherical harmonics on $S^{3}$. ### A.4 Vector spherical harmonics We introduce vector spherical harmonics for three different types of the spherical harmonics that we have defined above. The vector spherical harmonics are given by $\displaystyle Y^{\rho}_{Jm\tilde{m}A}(\Omega_{3})$ $\displaystyle=i^{\rho}\sum_{n,p}U_{An}C^{Qm}_{\tilde{Q}p\;1n}Y_{\tilde{Q}p\tilde{m}}(\Omega_{3}),$ $\displaystyle\tilde{Y}^{\rho}_{JmqA}(\Omega_{2})$ $\displaystyle=i^{\rho}\sum_{n,p}U_{An}C^{Qm}_{\tilde{Q}p\;1n}\tilde{Y}_{\tilde{Q}pq}(\Omega_{2}),$ $\displaystyle\hat{Y}^{\rho}_{Jm(jj^{\prime})A}$ $\displaystyle=i^{\rho}\sum_{n,p}U_{An}C^{Qm}_{\tilde{Q}p\;1n}\hat{Y}_{\tilde{Q}p(jj^{\prime})},$ (A.37) where $\rho=-1,0,1$ and $Q=J+\delta_{\rho 1}$, $\tilde{Q}=J+\delta_{\rho\;-1}$. These spherical harmonics transform as the vector representations under $SU(2)$ rotation. The unitary matrix $U$ is given by $U=\left(\begin{array}[]{ccc}-1&0&1\\\ -i&0&-i\\\ 0&\sqrt{2}&0\\\ \end{array}\right).$ (A.38) The vector spherical harmonics satisfy $\displaystyle\frac{1}{\mu}\epsilon_{ABC}\nabla_{B}Y^{\rho}_{Jm\tilde{m}C}=i\epsilon_{ABC}{\cal L}_{B}Y^{\rho}_{Jm\tilde{m}C}+Y^{\rho}_{Jm\tilde{m}A}=\rho(J+1)Y^{\rho}_{Jm\tilde{m}A},$ $\displaystyle i\epsilon_{ABC}L_{B}^{(q)}\tilde{Y}^{\rho}_{JmqC}+\tilde{Y}^{\rho}_{JmqA}=\rho(J+1)\tilde{Y}^{\rho}_{JmqA},$ $\displaystyle i\epsilon_{ABC}\hat{L}_{B}\circ\hat{Y}^{\rho}_{Jm(jj^{\prime})C}+\hat{Y}^{\rho}_{Jm(jj^{\prime})A}=\rho(J+1)\hat{Y}^{\rho}_{Jm(jj^{\prime})A}.$ (A.39) The complex (hermitian) conjugates of these vector harmonics are evaluated as $\displaystyle(Y^{\rho}_{Jm\tilde{m}A})^{*}$ $\displaystyle=(-1)^{m-\tilde{m}+1}Y^{\rho}_{J-m-\tilde{m}A},$ $\displaystyle(\tilde{Y}^{\rho}_{JmqA})^{*}$ $\displaystyle=(-1)^{m-q+1}\tilde{Y}^{\rho}_{J-m-qA},$ $\displaystyle(\hat{Y}^{\rho}_{Jm(jj^{\prime})A})^{\dagger}$ $\displaystyle=(-1)^{m-(j-j^{\prime})+1}\hat{Y}^{\rho}_{J-m(j^{\prime}j)A}.$ (A.40) The orthonormal relations are $\displaystyle\int\frac{d\Omega_{3}}{2\pi^{2}}(Y^{\rho}_{Jm\tilde{m}A})^{*}Y^{\rho^{\prime}}_{J^{\prime}m^{\prime}\tilde{m}^{\prime}A}$ $\displaystyle=\delta_{JJ^{\prime}}\delta_{mm^{\prime}}\delta_{\tilde{m}\tilde{m}^{\prime}}\delta_{\rho\rho^{\prime}},$ $\displaystyle\int\frac{d\Omega_{2}}{4\pi}(\tilde{Y}^{\rho}_{JmqA})^{*}\tilde{Y}^{\rho^{\prime}}_{J^{\prime}m^{\prime}qA}$ $\displaystyle=\delta_{JJ^{\prime}}\delta_{mm^{\prime}}\delta_{\rho\rho^{\prime}},$ $\displaystyle\frac{1}{N_{0}}{\rm tr}\left((\hat{Y}^{\rho}_{Jm(jj^{\prime})A})^{\dagger}\hat{Y}^{\rho^{\prime}}_{J^{\prime}m^{\prime}(j^{\prime}j)A}\right)$ $\displaystyle=\delta_{JJ^{\prime}}\delta_{mm^{\prime}}\delta_{\rho\rho^{\prime}}.$ (A.41) Finally, the integrals (or trace) of three vector harmonics are given by $\displaystyle\int\frac{d\Omega_{3}}{2\pi^{2}}\epsilon_{ABC}Y^{\rho_{1}}_{J_{1}m_{1}\tilde{m}_{1}A}Y^{\rho_{2}}_{J_{2}m_{2}\tilde{m}_{2}B}Y^{\rho_{3}}_{J_{3}m_{3}\tilde{m}_{3}C}$ $\displaystyle={\cal E}_{J_{1}m_{1}\tilde{m}_{1}\rho_{1}J_{2}m_{2}\tilde{m}_{2}\rho_{2}J_{3}m_{3}\tilde{m}_{3}\rho_{3}},$ $\displaystyle\int\frac{d\Omega_{2}}{4\pi}\epsilon_{ABC}\tilde{Y}^{\rho_{1}}_{J_{1}m_{1}q_{1}A}\tilde{Y}^{\rho_{2}}_{J_{2}m_{2}q_{2}B}\tilde{Y}^{\rho_{3}}_{J_{3}m_{3}q_{3}C}$ $\displaystyle={\cal E}_{J_{1}m_{1}q_{1}\rho_{1}J_{2}m_{2}q_{2}\rho_{2}J_{3}m_{3}q_{3}\rho_{3}},$ $\displaystyle\epsilon_{ABC}\frac{1}{N_{0}}{\rm tr}\left(\hat{Y}^{\rho_{1}}_{J_{1}m_{1}(jj^{\prime})A}\hat{Y}^{\rho_{2}}_{J_{2}m_{2}(j^{\prime}j^{\prime\prime})B}\hat{Y}^{\rho_{3}}_{J_{3}m_{3}(j^{\prime\prime}j)C}\right)$ $\displaystyle=\hat{{\cal E}}_{J_{1}m_{1}(jj^{\prime})\rho_{1}J_{2}m_{2}(j^{\prime}j^{\prime\prime})\rho_{2}J_{3}m_{3}(j^{\prime\prime}j)\rho_{3}},$ (A.42) where the monopole charges must be conserved in the lefthand side of the second equality as $q_{1}+q_{2}+q_{3}=0$ and ${\cal E}$, $\hat{\cal E}$ are given by $\displaystyle{\cal E}_{J_{1}m_{1}\tilde{m}_{1}\rho_{1}J_{2}m_{2}\tilde{m}_{2}\rho_{2}J_{3}m_{3}\tilde{m}_{3}\rho_{3}}$ $\displaystyle=\sqrt{6(2J_{1}+1)(2J_{1}+2\rho_{1}^{2}+1)(2J_{2}+1)(2J_{2}+2\rho_{2}^{2}+1)(2J_{3}+1)(2J_{3}+2\rho_{3}^{2}+1)}$ $\displaystyle\;\;\;\times(-1)^{-\frac{\rho_{1}+\rho_{2}+\rho_{3}+1}{2}}\left\\{\begin{array}[]{ccc}Q_{1}&\tilde{Q}_{1}&1\\\ Q_{2}&\tilde{Q}_{2}&1\\\ Q_{3}&\tilde{Q}_{3}&1\end{array}\right\\}\left(\begin{array}[]{ccc}Q_{1}&Q_{2}&Q_{3}\\\ m_{1}&m_{2}&m_{3}\end{array}\right)\left(\begin{array}[]{ccc}\tilde{Q}_{1}&\tilde{Q}_{2}&\tilde{Q}_{3}\\\ \tilde{m}_{1}&\tilde{m}_{2}&\tilde{m}_{3}\end{array}\right),$ (A.50) $\displaystyle\hat{{\cal E}}_{J_{1}m_{1}(jj^{\prime})\rho_{1}J_{2}m_{2}(j^{\prime}j^{\prime\prime})\rho_{2}J_{3}m_{3}(j^{\prime\prime}j)\rho_{3}}$ $\displaystyle=\sqrt{6N_{0}(2J_{1}+1)(2J_{1}+2\rho_{1}^{2}+1)(2J_{2}+1)(2J_{2}+2\rho_{2}^{2}+1)(2J_{3}+1)(2J_{3}+2\rho_{3}^{2}+1)}$ $\displaystyle\;\;\;\times(-1)^{-\frac{\rho_{1}+\rho_{2}+\rho_{3}+1}{2}-\tilde{Q}_{1}-\tilde{Q}_{2}-\tilde{Q}_{3}+2j+2j^{\prime}+2j^{\prime\prime}}\left\\{\begin{array}[]{ccc}Q_{1}&\tilde{Q}_{1}&1\\\ Q_{2}&\tilde{Q}_{2}&1\\\ Q_{3}&\tilde{Q}_{3}&1\end{array}\right\\}\left(\begin{array}[]{ccc}Q_{1}&Q_{2}&Q_{3}\\\ m_{1}&m_{2}&m_{3}\end{array}\right)\left\\{\begin{array}[]{ccc}\tilde{Q}_{1}&\tilde{Q}_{2}&\tilde{Q}_{3}\\\ j^{\prime\prime}&j&j^{\prime}\end{array}\right\\}.$ (A.58) As in (A.33), we can show $\hat{{\cal E}}_{J_{1}m_{1}(jj^{\prime})\rho_{1}J_{2}m_{2}(j^{\prime}j^{\prime\prime})\rho_{2}J_{3}m_{3}(j^{\prime\prime}j)\rho_{3}}\rightarrow{\cal E}_{J_{1}m_{1}\tilde{m}_{1}\rho_{1}J_{2}m_{2}\tilde{m}_{2}\rho_{2}J_{3}m_{3}\tilde{m}_{3}\rho_{3}},$ (A.59) in the limit $N_{0}\rightarrow\infty$ with $j-j^{\prime}=q_{1}$, $j^{\prime}-j^{\prime\prime}=q_{2}$ and $j^{\prime\prime}-j=q_{3}$ fixed. ## Appendix B Derivation of (4.22) In this appendix, we give the derivation of (4.22) in some detail. $\displaystyle[L_{a},a_{\alpha}(x)]^{(s,t)}$ $\displaystyle=a_{\alpha,Jm}^{(q_{st})}(x)\otimes L_{a}\circ\hat{Y}_{Jm(j_{s}j_{t})}$ $\displaystyle=\int\frac{d\Omega_{3}}{2\pi^{2}}A_{\alpha}(z)Y_{Jmq_{st}}^{\dagger}(y)\otimes L_{a}\circ\hat{Y}_{Jm(j_{s}j_{t})}$ $\displaystyle=\int\frac{d\Omega_{3}}{2\pi^{2}}A_{\alpha}(z){\cal L}_{a}Y_{Jmq_{st}}^{\dagger}\otimes\hat{Y}_{Jm(j_{s}j_{t})}$ $\displaystyle=\int\frac{d\Omega_{3}}{2\pi^{2}}\left({\cal L}_{a}A_{\alpha}(z)\right)Y_{Jmq_{st}}^{\dagger}\otimes\hat{Y}_{Jm(j_{s}j_{t})},$ where we have used (A.9) and (A.26). $\displaystyle[\phi_{a},\phi_{b}]^{(s,t)}$ $\displaystyle=\sum_{u}\left(\phi_{a,Jm}^{(q_{su})}\phi_{b,J^{\prime}m^{\prime}}^{(q_{ut})}-\phi_{b,Jm}^{(q_{su})}\phi_{a,J^{\prime}m^{\prime}}^{(q_{ut})}\right)\otimes\hat{Y}_{Jm(j_{s}j_{u})}\hat{Y}_{J^{\prime}m^{\prime}(j_{u}j_{t})}$ $\displaystyle=\sum_{u}\left(\phi_{a,Jm}^{(q_{su})}\phi_{b,J^{\prime}m^{\prime}}^{(q_{ut})}-\phi_{b,Jm}^{(q_{su})}\phi_{a,J^{\prime}m^{\prime}}^{(q_{ut})}\right)\otimes\hat{{\cal C}}^{J^{\prime\prime}m^{\prime\prime}(j_{s}j_{t})}_{Jm(j_{s}j_{u})\;J^{\prime}m^{\prime}(j_{u}j_{t})}\hat{Y}_{J^{\prime\prime}m^{\prime\prime}(j_{s}j_{t})}$ $\displaystyle=\sum_{u,v}\int\frac{d\Omega_{3}}{2\pi^{2}}\frac{d\Omega_{3}^{\prime}}{2\pi^{2}}\left\\{A_{a}(z)A_{b}(z^{\prime})-A_{b}(z)A_{a}(z^{\prime})\right\\}Y_{Jmq_{su}}^{\dagger}(y)Y_{J^{\prime}m^{\prime}q_{vt}}^{\dagger}(y^{\prime})$ $\displaystyle\qquad\qquad\times\int\frac{d\Omega_{3}^{\prime\prime}}{2\pi^{2}}Y_{J^{\prime\prime}m^{\prime\prime}q_{st}}^{\dagger}(y^{\prime\prime})Y_{Jmq_{su}}(y^{\prime\prime})Y_{J^{\prime}m^{\prime}q_{vt}}(y^{\prime\prime})\otimes\hat{Y}_{J^{\prime\prime}m^{\prime\prime}(j_{s}j_{t})}$ $\displaystyle=\int\frac{d\Omega_{3}}{2\pi^{2}}[A_{a}(z),A_{b}(z)]Y_{Jmq_{st}}^{\dagger}(y)\otimes\hat{Y}_{Jm(j_{s}j_{t})}.$ (B.2) In the third and fourth lines of the righthand side, we have used (A.33), the charge conservation $\tilde{m}^{\prime\prime}=\tilde{m}+\tilde{m}^{\prime}$ of ${\cal C}^{J^{\prime\prime}m^{\prime\prime}\tilde{m}^{\prime\prime}}_{Jm\tilde{m}\;J^{\prime}m^{\prime}\tilde{m}^{\prime}}$ and (A.12), so that we have added the new summation over $v$ additionally and replaced $q_{ut}$ by $q_{vt}$. Then, we can regard the summation $\sum_{u,v}$ as $\sum_{q_{su},q_{vt}}$, and the last equality holds due to (A.13). ## Appendix C Group manifold and coset space In this appendix, we describe some conventions on the group manifold $\tilde{G}$ and the coset space $\tilde{G}/H$ which we follow in this paper. We parameterize an element of $\tilde{G}$ as $g(z)=L(x)h(y),$ (C.1) where $L(x)\in\tilde{G}/H$, $h(y)\in H$, the coordinates $z^{M}$, $x^{\mu}$ and $y^{m}$ parameterize $\tilde{G}$, $\tilde{G}/H$ and $H$ respectively and $z^{M}$ are decomposed into $(x^{\mu},y^{i})$. We can construct the right and left invariant 1-forms on $\tilde{G}$ as $dgg^{-1}=-iE^{A}_{R}T^{A},\;\;\;g^{-1}dg=iE_{L}^{A}T^{A},$ (C.2) where $A=1,\cdots,{\rm dim}\tilde{G}$ and $T^{A}$ represent the generators of $\tilde{G}$ which satisfy the Lie algebra of $\tilde{G}$, $[T^{A},T^{B}]=if_{ABC}T^{C}$. We decompose $T^{A}$ into $(T^{\alpha},T^{a})$ where $\alpha=1,\cdots,{\rm dim}\tilde{G}/H,\;a={\rm dim}\tilde{G}/H+1,\cdots,{\rm dim}\tilde{G}$, and we assume that $T^{a}$ satisfy the Lie algebra of $H$ which is a subalgebra of $\tilde{G}$, $[T^{a},T^{b}]=if_{abc}T^{c}$. The both of $E^{A}_{R}$ and $E^{A}_{L}$ satisfy the Maurer-Cartan equation, $dE^{A}_{R}-\frac{1}{2}f_{ABC}E^{B}_{R}\wedge E^{C}_{R}=0,\;\;\;dE^{A}_{L}-\frac{1}{2}f_{ABC}E^{B}_{L}\wedge E^{C}_{L}=0,$ (C.3) We also introduce the right and left invariant 1-form for $L(x)\in\tilde{G}/H$ and $h(y)\in H$ as follows: $\displaystyle dLL^{-1}=-i(e_{R})^{A}_{\mu}(x)T^{A}dx^{\mu},\;\;\;L^{-1}dL=i(e_{L})^{A}_{\mu}(x)T^{A}dx^{\mu},$ $\displaystyle dhh^{-1}=-i(\tilde{e}_{R})^{a}_{m}(y)T^{a}dy^{m},\;\;\;h^{-1}dh=i(\tilde{e}_{L})^{a}_{m}(y)T^{a}dy^{m}.$ (C.4) Then, we can write down the components of $E^{A}_{R}$ and $E_{L}^{A}$ explicitly: $\displaystyle(E_{R})^{A}{}_{M}=\left(\begin{array}[]{cc}(e_{R})^{\alpha}_{\mu}&{\rm Ad}(L)^{\alpha}{}_{b}(\tilde{e}_{R})^{b}_{m}\\\ (e_{R})^{a}_{\mu}&{\rm Ad}(L)^{a}{}_{b}(\tilde{e}_{R})^{b}_{m}\end{array}\right),\;\;\;(E_{L})^{A}{}_{M}=\left(\begin{array}[]{cc}(e_{L})^{\beta}_{\mu}{\rm Ad}(h)_{\beta\alpha}&0\\\ (e_{L})^{\beta}_{\mu}{\rm Ad}(h)_{\beta a}&(\tilde{e}_{L})^{a}_{m}\end{array}\right),$ (C.9) where ${\rm Ad}$ is defined as the adjoint action $gT^{A}g^{-1}=T^{B}{\rm Ad}(g)_{BA}$. The Cartan-Killing metric on $\tilde{G}$ is defined as $ds^{2}=G_{MN}dz^{M}dz^{N}=-2{\rm Tr}(dgg^{-1}dgg^{-1}).$ (C.10) In terms of the components (C.9), the above metric is written as $ds^{2}=(e_{L})^{\alpha}_{\mu}(e_{L})^{\alpha}_{\nu}dx^{\mu}dx^{\nu}+\\{(\tilde{e}_{R})^{a}_{m}dy^{m}-(e_{L})^{a}_{\mu}dx^{\mu}\\}^{2}.$ (C.11) We can regard the group manifold $\tilde{G}$ as the principal $H$ bundle on $\tilde{G}/H$. By comparing (C.11) and (3.5), therefore, we can make the following identifications: $(e_{L})^{\alpha}_{\mu}(e_{L})^{\alpha}_{\nu}=g_{\mu\nu},\;(e_{L})^{a}_{\mu}=b_{\mu}^{a},\;(\tilde{e}_{R})^{a}_{m}(\tilde{e}_{R})^{a}_{n}=h_{mn},$ (C.12) where $g_{\mu\nu}$ and $h_{mn}$ are the metrics on $\tilde{G}/H$ and $H$, respectively, and $b_{\mu}^{a}$ are the local connection 1-forms of the principal $H$ bundle. Namely, we can regard $(e_{L})^{\alpha}_{\mu}$ and $(\tilde{e}_{R})^{a}_{m}$ as the vielbein on $\tilde{G}/H$ and $H$, respectively. The metric (C.10) is invariant under the right and left actions of $\tilde{G}$. The corresponding right and left invariant Killing vectors on $\tilde{G}$ are defined in terms of the inverse of $E^{A}$ as ${\cal L}^{R}_{A}=-i(E_{R})_{A}^{M}\partial_{M},\;\;\;{\cal L}^{L}_{A}=-i(E_{L})_{A}^{M}\partial_{M}.$ (C.13) By using (C.3), we can show that ${\cal L}_{A}^{R}$ and ${\cal L}_{A}^{L}$ satisfy the Lie algebra of $\tilde{G}\times\tilde{G}$, $[{\cal L}_{A}^{R},{\cal L}_{B}^{R}]=if_{ABC}{\cal L}_{C}^{R},\;\;\;[{\cal L}_{A}^{L},{\cal L}_{B}^{L}]=if_{ABC}{\cal L}_{C}^{L},\;\;\;[{\cal L}_{A}^{R},{\cal L}_{B}^{L}]=0,$ (C.14) and they also satisfy the Killing vector equations, $\nabla_{M}{\cal L}_{AN}+\nabla_{N}{\cal L}_{AM}=0,$ (C.15) where $\nabla_{M}$ are the covariant derivative on $\tilde{G}$ and ${\cal L}_{AM}=G_{MN}{\cal L}_{A}^{N}$. We also define the following operators: $L_{A}=-i(E_{R})_{A}^{\mu}\partial_{\mu}.$ (C.16) One can show that $L_{A}$ do not depend on $y^{m}$ and they satisfy $[L_{A},L_{B}]=if_{ABC}L_{C}$ by using (C.14). Furthermore, we can show that $\nabla^{(\tilde{G}/H)}_{\mu}L_{A\nu}+\nabla^{(\tilde{G}/H)}_{\nu}L_{A\mu}=0,$ (C.17) where $L_{A\mu}=g_{\mu\nu}L_{A}^{\nu}$. Namely, $L_{A}$ are the Killing vectors on the coset space $\tilde{G}/H$. ## Appendix D Metrics of $SU(3)$, $S^{5}$ and $CP^{2}$ In this appendix, for concreteness, we give an explicit form of the metrics of $SU(3)$, $SU(3)/SU(2)\simeq S^{5}$ and $SU(3)/(SU(2)\times U(1))\simeq CP^{2}$ [51]. We parameterize an element $g$ of $SU(3)$ as $\displaystyle g=L(\chi,\theta,\varphi,\psi)Z(\tau)V(a,b,c),$ (D.1) where $\displaystyle L(\chi,\theta,\varphi,\psi)=e^{i\varphi\lambda_{3}}e^{i\theta\lambda_{2}}e^{i\psi\lambda_{3}}e^{2i\chi\lambda_{5}},$ $\displaystyle Z(\tau)=e^{-i\sqrt{3}(\tau-2\pi)\lambda_{8}},$ $\displaystyle V(a,b,c)=e^{-ia\lambda_{3}}e^{-ib\lambda_{2}}e^{-ic\lambda_{3}},$ (D.2) and $0\leq\chi\leq\frac{\pi}{2}$, $0\leq\theta\leq\pi$, $0\leq\varphi<2\pi$, $0\leq\psi<4\pi$, $0\leq\tau<2\pi$, $0\leq a<2\pi$, $0\leq b\leq\pi$ and $0\leq c<4\pi$. $\lambda_{1},\cdots,\lambda_{8}$ are the Gell-Mann matrices and satisfy $\mbox{Tr}(\lambda_{a}\lambda_{b})=\frac{1}{2}\delta_{ab}$. The metric of $SU(3)$ is given by $\displaystyle dS_{SU(3)}^{2}=$ $\displaystyle-\frac{1}{2}\mbox{Tr}(dgg^{-1}dgg^{-1})$ $\displaystyle=$ $\displaystyle d\chi^{2}+\frac{1}{4}\sin^{2}\chi\left\\{d\theta^{2}+\sin^{2}\theta d\varphi^{2}+\cos^{2}\chi(d\psi+\cos\theta d\varphi)^{2}\right\\}$ $\displaystyle+\frac{3}{4}\left\\{d\tau+\frac{1}{2}\sin^{2}\chi(d\psi+\cos\theta d\varphi)\right\\}^{2}$ $\displaystyle+\frac{1}{4}\left\\{e^{1}+\cos\chi(\sin\psi d\theta-\sin\theta\cos\psi d\varphi)\right\\}^{2}$ $\displaystyle+\frac{1}{4}\left\\{e^{2}-\cos\chi(\cos\psi d\theta+\sin\theta\sin\psi d\varphi)\right\\}^{2}$ $\displaystyle+\frac{1}{4}\left\\{e^{3}-\frac{1}{2}(1+\cos^{2}\chi)(d\psi+\cos\theta d\varphi)\right\\}^{2},$ (D.3) where $\displaystyle e^{1}=-\sin adb+\cos a\sin bdc,$ $\displaystyle e^{2}=\cos adb+\sin a\sin bdc,$ $\displaystyle e^{3}=da+\cos bdc,$ (D.4) which are the right invariant 1-form of $SU(2)$. $SU(3)$ is an $SU(2)\times U(1)$ bundle over $CP^{2}$. The second line in the righthand side of (D.3) is the Fubini-Study metric of $CP^{2}$. The third line represents the $U(1)$ fiber structure while the fourth, fifth and sixth lines represent the $SU(2)$ fiber structure. $SU(3)$ is also viewed as an $SU(2)$ bundle over $S^{5}\simeq SU(3)/SU(2)$. The second and third lines together correspond to the metric of $S^{5}\simeq SU(3)/SU(2)$. $S^{5}\simeq SU(3)/SU(2)$ is viewed as a $U(1)$ bundle over $CP^{2}$. The metric of the ordinary unit $S^{5}$ is given by the sum of the second and third lines with the factor $3/4$ in the third line replaced by $1/4$. ## Appendix E Fuzzy $CP^{n}$ In this appendix, we give a brief review of a construction of fuzzy $CP^{n}$ [28, 31, 32, 33, 34, 35]. ### E.1 Functions on fuzzy $CP^{n}$ Fuzzy $CP^{n}$ is a well-known example of noncommutative space which is given by the quantization of coadjoint orbit of $SU(n+1)$ in terms of a certain matrix algebra acting on an appropriate representation space $V$. We can determine this matrix algebra and the representation space $V$ by matching the spectrum of functions on $CP^{n}$ and that on fuzzy $CP^{n}$. In order to consider the spectrum of functions on $CP^{n}$, We regard $CP^{n}$ as a coadjoint orbit in the Lie algebra of $SU(n+1)$. $CP^{n}=\\{\;gtg^{-1}\;|\;g\in SU(n+1)\\}\simeq SU(n+1)/(SU(n)\times U(1)),$ (E.1) where $t$ is an element of the $SU(n+1)$ Lie algebra such that the stabilizer of $t$ is given by $SU(n)\times U(1)$. For example, for the case of $CP^{2}$, we can take $t$ to be $\lambda_{8}$ which is invariant under $SU(2)\times U(1)$ adjoint action generated by $\lambda_{1},\lambda_{2},\lambda_{3}$ and $\lambda_{8}$. Functions on $CP^{n}$ should be invariant under the action of $SU(n)\times U(1)$. Then, the space of functions on $CP^{n}$ is given by a direct sum of the representation spaces of $SU(n+1)$ which contain $SU(n)\times U(1)$ invariant states: $C^{\infty}(CP^{n})=\bigoplus_{J=0}^{\infty}V_{[J,0,\cdots,0,J]}$ (E.2) where we denote $[J,0,\cdots,0,J]$ as the Dynkin index of $SU(n+1)$, and $V_{[J,0,\cdots,0,J]}$ represents the corresponding irreducible representation space of the $SU(n+1)$ Lie algebra. One can show that $V_{[J,0,\cdots,0,J]}$ are the only spaces which contain the $SU(n)\times U(1)$ singlets. The space of functions on fuzzy $CP^{n}$ is obtained by introducing a cutoff $\Lambda$ in (E.2) as $\bigoplus_{J=0}^{\Lambda}V_{[J,0,\cdots,0,J]}=V_{[\Lambda,0,\cdots,0]}\otimes V_{[\Lambda,0,\cdots,0]}^{*}.$ (E.3) By definition, it is obvious that the above spectrum on fuzzy $CP^{n}$ tends to the spectrum (E.2) on $CP^{n}$ in the commutative limit $\Lambda\rightarrow\infty$. Note that the righthand side of the above equation can be viewed as a space of matrices. From this viewpoint, we make an identification $V=V_{[\Lambda,0,\cdots,0]}$ and regard functions on fuzzy $CP^{n}$ as matrices acting on the vector space $V$. In particular, the coordinates on fuzzy $CP^{n}$ are identified with $\hat{\xi}_{A}=\hat{L}_{A}^{[\Lambda,0,\cdots,0]},$ (E.4) which are the generators of $SU(n+1)$ in the irreducible representation specified by the Dynkin index $[\Lambda,0,\cdots,0]$. These coordinates on fuzzy $CP^{n}$ are actually reduced to the coordinates on $CP^{n}$ in the commutative limit through a map which will be defined in the last part of this section. ### E.2 Derivatives on fuzzy $CP^{n}$ In order to construct differential operators on fuzzy $CP^{n}$, let us recall the simplest case of fuzzy $CP^{1}\simeq S^{2}$. In this case, we established the differential operators on fuzzy $S^{2}$ in appendix A. As shown in (A.36), the adjoint action of the $SU(2)$ generators is reduced to the action of the Killing vectors on $S^{2}$ in the commutative limit. We can generalize this fact into the case of fuzzy $CP^{n}$ with $n\geq 2$. The adjoint action of the $SU(n+1)$ generators on the space of square matrices (E.3), $[\hat{L}_{A}^{[\Lambda,0,\cdots,0]},\;\cdot\;]$, is mapped into the action of the Killing vectors on the space of functions on $CP^{n}$ in the commutative limit. ### E.3 $U(1)$ monopoles on fuzzy $CP^{n}$ Topologically nontrivial field configurations including $U(1)$ monopoles can be realized on fuzzy $CP^{n}$. If we consider rectangular matrices in addition to the square matrices (E.3), the concept of fiber bundles naturally arises. Let us again consider the case of fuzzy $CP^{1}$. We have shown in appendix A that the basis of $(2j+1)\times(2j^{\prime}+1)$ rectangular matrices, $\hat{Y}_{Jm(jj^{\prime})}$, are mapped into local sections of the $U(1)$ fiber bundle on $S^{2}$. In this correspondence, The difference $j-j^{\prime}$ is identified with the monopole charge $q$ of the $U(1)$ bundle. This fact is also generalized into the case of $CP^{n}$ with $n\geq 2$. For the case of $CP^{n}$, we consider a space of rectangular matrices, $V_{[\Lambda+q,0,\cdots,0,]}\otimes V_{[\Lambda-q,0,\cdots,0,]}^{*}.$ (E.5) Here, the charge $q$ is a half integer and we take $\Lambda\pm q$ to be integers. When $q=0$, $\Lambda$ is an integer and this is the case of square matrices (E.3). We can show that elements of (E.5) are mapped into local sections of $U(1)$ fiber bundle on $CP^{n}$ with the monopole charge $q$. Furthermore, we can extend the action of the differential operators $[\hat{L}_{A}^{[\Lambda,0,\cdots,0]},\;\cdot\;]$ discussed above to the action on rectangular matrices as follows. $\hat{L}_{A}\circ\hat{M}_{q}=\hat{L}_{A}^{[\Lambda+q,0,\cdots,0]}\hat{M}_{q}-\hat{M}_{q}\hat{L}_{A}^{[\Lambda-q,0,\cdots,0]},$ (E.6) where $\hat{M}_{q}$ is an element of (E.5). When $q=0$, $\hat{M}_{0}$ is just a square matrix and $\hat{L}_{A}\circ$ are nothing but the commutators $[\hat{L}_{A}^{[\Lambda,0,\cdots,0]},\;\cdot\;]$. The operators $\hat{L}_{A}\circ$ map the space (E.5) to itself and they are reduced to the angular momentum operators in the presence of a $U(1)$ monopole with the magnetic charge $q$ in the commutative limit. We will show these facts in the following subsections. ### E.4 Fock space representation In order to construct a map between matrices and functions on $CP^{n}$, we introduce the Fock space representation developed in [35]. Let $a^{\dagger}_{\alpha}$, $\alpha=1,2,\cdots,n+1$ be a set of creation operators and $a^{\alpha}$ be a set of annihilation operators which annihilate the Fock vacuum $|0\rangle$. They satisfy the Heisenberg commutation relations. $[a^{\alpha},a^{\beta}]=[a^{\dagger}_{\alpha},a^{\dagger}_{\beta}]=0,\;\;\;[a^{\alpha},a^{\dagger}_{\beta}]=\delta_{\beta}^{\alpha}.$ (E.7) By acting the creation operators on the vacuum state $|0\rangle$, we can construct the entire Fock space ${\cal F}$ which is spanned by $|p_{1},p_{2},\cdots,p_{n+1}\rangle=\frac{1}{\sqrt{p_{1}!p_{2}!\cdots p_{n+1}!}}(a_{1}^{\dagger})^{p_{1}}(a_{2}^{\dagger})^{p_{2}}\cdots(a_{n+1}^{\dagger})^{p_{n+1}}|0\rangle.$ (E.8) In terms of the operators (E.7), we can construct elements of the Lie algebra of $SU(n+1)$, $\hat{L}_{A}=a^{\dagger}_{\alpha}(T_{A})^{\alpha}{}_{\beta}a^{\beta},$ (E.9) where $T_{A}=\hat{L}_{A}^{[1,0,\cdots,0]}$ represent the generators of $SU(n+1)$ in the fundamental representation. We also define the number operator which commutes with all the operators in (E.9). $\hat{N}=a^{\dagger}_{\alpha}a^{\alpha}.$ (E.10) The operators (E.9) and (E.10) act on the Fock space ${\cal F}$, and satisfy $[\hat{L}_{A},\hat{L}_{B}]=if_{ABC}\hat{L}_{C},\;\;\;[\hat{L}_{A},\hat{N}]=0.$ (E.11) We can decompose the Fock space ${\cal F}$ into the eigenspaces of $\hat{N}$ as ${\cal F}=\bigoplus_{p=0}^{\infty}V_{[p,0,\cdots,0]},$ (E.12) where $p$ represent an eigenvalue of $\hat{N}$. The basis of each eigenspace $V_{[p,0,\cdots,0]}$ is formed by $|{\bm{\alpha}}_{p}\rangle=|\alpha_{1},\alpha_{2},\cdots,\alpha_{p}\rangle=\frac{1}{\sqrt{p!}}a_{\alpha_{1}}^{\dagger}a_{\alpha_{2}}^{\dagger}\cdots a_{\alpha_{p}}^{\dagger}|0\rangle,$ (E.13) where ${\bm{\alpha}}_{p}$ is an abbreviation of a set of $p$ indices, $(\alpha_{1},\alpha_{2},\cdots,\alpha_{p})$. Let us consider square matrices which are elements of (E.3). These matrices are generally written as $\hat{M}=\hat{M}^{{\bm{\alpha}}_{\Lambda}}{}_{{\bm{\beta}}_{\Lambda}}|{\bm{\alpha}}_{\Lambda}\rangle\langle{\bm{\beta}}_{\Lambda}|.$ (E.14) We define a new basis of these matrices to see the correspondence with the spectrum of functions on $CP^{n}$, (E.2): $\hat{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}=N_{\Lambda J}^{n}{\cal P}_{{\bm{\beta}}_{J},{\bm{\tau}}_{J}}{}^{{\bm{\alpha}}_{J},{\bm{\sigma}}_{J}}|{\bm{\sigma}}_{J},{\bm{\gamma}}_{\Lambda-J}\rangle\langle{\bm{\tau}}_{J},{\bm{\gamma}}_{\Lambda-J}|,$ (E.15) where $\Lambda-J$ indices ${\bm{\gamma}}_{\Lambda-J}$ are contracted and ${\cal P}_{{\bm{\beta}}_{J},{\bm{\tau}}_{J}}{}^{{\bm{\alpha}}_{J},{\bm{\sigma}}_{J}}$ is the projection operator onto the representation space $V_{[J,0,\cdots,0,J]}$ which appeared in the decomposition (E.3), that is, it removes all traces between ${\bm{\alpha}}_{J}$ and ${\bm{\beta}}_{J}$. For example, $\displaystyle\hat{Y}=N_{\Lambda 0}^{n}{\bm{1}},\;\;\;\hat{Y}_{\beta}{}^{\alpha}=N_{\Lambda 1}^{n}\left(|\beta,{\bm{\gamma}}_{\Lambda-1}\rangle\langle\alpha,{\bm{\gamma}}_{\Lambda-1}|-\frac{1}{2}\delta_{\beta}^{\alpha}{\bm{1}}\right).$ (E.16) Hence, $\hat{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}$ belong to the representation $V_{[J,0,\cdots,0,J]}$ and they are mapped to the corresponding spherical harmonics on $CP^{n}$ in the commutative limit which are elements of (E.2). $N_{\Lambda J}^{n}$ is an appropriate normalization constant which is determined by the following orthonormality of the basis, ${\rm tr}\left((\hat{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}})^{\dagger}\hat{Y}_{{\bm{\tau}}_{J^{\prime}}}{}^{{\bm{\sigma}}_{J^{\prime}}}\right)=\delta_{JJ^{\prime}}{\cal P}_{{\bm{\alpha}}_{J},{\bm{\tau}}_{J}}{}^{{\bm{\beta}}_{J},{\bm{\sigma}}_{J}}.$ (E.17) In the case of $n=1$, $\hat{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}$ are essentially the same as the fuzzy spherical harmonics which are defined in (A.25). The action of differential operators on fuzzy $CP^{n}$ is given by the adjoint action of operators in (E.9). Then, one can evaluate the eigenvalues of the Laplacian as follows: $[\hat{L}_{A},[\hat{L}_{A},\hat{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}]]=J(J+n)\hat{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}.$ (E.18) The above spectrum completely matches the spectrum of functions on $CP^{n}$ up to the cutoff $\Lambda$. In terms of the Fock space representation, we can also express rectangular matrices which are elements of (E.5). Those rectangular matrices are generally expressed as $\hat{M}_{q}=(\hat{M}_{q})^{{\bm{\alpha}}_{\Lambda+q}}{}_{{\bm{\beta}}_{\Lambda-q}}|{\bm{\alpha}}_{\Lambda+q}\rangle\langle{\bm{\beta}}_{\Lambda-q}|.$ (E.19) These matrices are expanded by a similar basis to (E.15). Note that the direct product representation $(\ref{rectangular matrices})$ is decomposed as $\bigoplus_{J=|q|}^{\Lambda}V_{[J+q,0,\cdots,0,J-q]}.$ (E.20) For each representation space in (E.20) with fixed $J$, we can use the following basis: $\hat{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}}=N_{\Lambda Jq}^{n}{\cal P}_{{\bm{\beta}}_{J+q},{\bm{\tau}}_{J-q}}{}^{{\bm{\alpha}}_{J-q},{\bm{\sigma}}_{J+q}}|{\bm{\sigma}}_{J+q},{\bm{\gamma}}_{\Lambda-J}\rangle\langle{\bm{\tau}}_{J-q},{\bm{\gamma}}_{\Lambda-J}|.$ (E.21) As in the case of square matrices, ${\cal P}_{{\bm{\beta}}_{J+q},{\bm{\tau}}_{J-q}}{}^{{\bm{\alpha}}_{J-q},{\bm{\sigma}}_{J+q}}$ is a projection operator onto the space (E.20) with fixed $J$ and $N_{\Lambda Jq}^{n}$ is a normalization constant which is determined by ${\rm tr}\left((\hat{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}})^{\dagger}\hat{Y}^{(q)}{}_{{\bm{\tau}}_{J^{\prime}+q}}{}^{{\bm{\sigma}}_{J^{\prime}-q}}\right)=\delta_{JJ^{\prime}}{\cal P}_{{\bm{\tau}}_{J+q},{\bm{\alpha}}_{J-q}}{}^{{\bm{\sigma}}_{J-q},{\bm{\beta}}_{J+q}}.$ (E.22) When $q=0$, $\hat{Y}^{(0)}{}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}$ are identical with the square matrices (E.15). The action of differential operators on $\hat{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}}$ is given by (E.6). We can evaluate the eigenvalues of the Laplacian as follows: $(\hat{L}_{A}\circ)^{2}\hat{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}}=\left(J(J+1)+\frac{n-1}{n+1}q^{2}\right)\hat{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}}.$ (E.23) The above spectrum is the same as the spectrum of local sections of $U(1)$ bundle on $CP^{n}$ up to the cutoff. We show in the following that the rectangular matrices $\hat{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}}$ are indeed mapped to the local sections on $CP^{n}$. ### E.5 Relation between matrices and sections Let us recall the spherical harmonics on $CP^{n}$. In a spinorial basis, they are given by $\tilde{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}=N_{J}^{n}{\cal P}_{{\bm{\beta}}_{J},{\bm{\tau}}_{J}}{}^{{\bm{\alpha}}_{J},{\bm{\sigma}}_{J}}\bar{w}_{\sigma_{1}}\cdots\bar{w}_{\sigma_{J}}w^{\tau_{1}}\cdots w^{\tau_{J}},$ (E.24) In the above expression, $w^{\alpha}$ are the coordinates of $S^{2n+1}\simeq SU(n+1)/SU(n)$ which satisfy $\sum_{\alpha}|w^{\alpha}|^{2}=1$ and the normalization constant $N_{J}^{n}$ is determined by $\int_{CP^{n}}\omega^{n}\;(\tilde{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}})^{*}\tilde{Y}_{{\bm{\tau}}_{J^{\prime}}}{}^{{\bm{\sigma}}_{J^{\prime}}}=\delta_{JJ^{\prime}}{\cal P}_{{\bm{\alpha}}_{J},{\bm{\tau}}_{J}}{}^{{\bm{\beta}}_{J},{\bm{\sigma}}_{J}},$ (E.25) where $\omega^{n}$ is the volume form on $CP^{n}$. The functions (E.24) are invariant under the $U(1)$ phase rotation so that they can be regarded as global sections (functions) on $CP^{n}$. We can generalize (E.24) to a basis of local sections of the $U(1)$ monopole bundle on $CP^{n}$. The local sections of the monopole bundle with the magnetic charge $q$ can be expanded by $\tilde{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}}=N_{Jq}^{n}{\cal P}_{{\bm{\beta}}_{J+q},{\bm{\tau}}_{J-q}}{}^{{\bm{\alpha}}_{J-q},{\bm{\sigma}}_{J+q}}\bar{w}_{\sigma_{1}}\cdots\bar{w}_{\sigma_{J+q}}w^{\tau_{1}}\cdots w^{\tau_{J-q}},$ (E.26) which are normalized as $\int_{CP^{n}}\omega^{n}\;(\tilde{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}})^{*}\tilde{Y}^{(q)}{}_{{\bm{\tau}}_{J^{\prime}+q}}{}^{{\bm{\sigma}}_{J^{\prime}-q}}=\delta_{JJ^{\prime}}{\cal P}_{{\bm{\alpha}}_{J-q},{\bm{\tau}}_{J+q}}{}^{{\bm{\beta}}_{J+q},{\bm{\sigma}}_{J-q}}.$ (E.27) $\tilde{Y}^{(q)}{}_{{\bm{\beta}}_{J+q}}{}^{{\bm{\alpha}}_{J-q}}$ are not invariant under the $U(1)$ phase rotation, so that they transform as the local sections of the monopole bundle on $CP^{n}$ with the magnetic charge $q$. Note that $\tilde{Y}^{(0)}{}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}$ are nothing but the global sections, $\tilde{Y}_{{\bm{\beta}}_{J}}{}^{{\bm{\alpha}}_{J}}$. The relation between matrices and sections on $CP^{n}$ is given by the diagonal coherent state map [35]. Let us consider a matrix $\hat{M}_{q}$ which is an element of (E.5) and expanded as in (E.19). $\hat{M}_{q}$ corresponds to a section of the monopole bundle on $CP^{n}$ through the map. In particular, when $q=0$, $\hat{M}_{0}$ is just a square matrix and corresponds to a global section on $CP^{n}$. The map to the sections is given by $\tilde{M}_{q}(w,\bar{w})=\langle w,\Lambda+q|\hat{M}_{q}|w,\Lambda-q\rangle,$ (E.28) where $|w,p\rangle=\frac{1}{\sqrt{p!}}(w^{\alpha}a_{\alpha}^{\dagger})^{p}|0\rangle.$ (E.29) The map (E.28) is equivalent to the following replacement up to an over all constant factor, $\displaystyle(a^{\dagger}_{\alpha})^{L}\rightarrow\bar{w}_{\alpha},\;\;\;\;(a^{\alpha})^{L}\rightarrow\frac{\partial}{\partial\bar{w}_{\alpha}},$ $\displaystyle(a^{\alpha})^{R}\rightarrow w^{\alpha},\;\;\;\;(a^{\dagger}_{\alpha})^{R}\rightarrow\frac{\partial}{\partial w^{\alpha}},$ (E.30) where the superscripts $L$ and $R$ express that the operators act on matrices from the left and right, respectively. Through this correspondence, (E.15) and (E.21) are mapped to (E.24) and (E.26) respectively. Furthermore, the differential operators $\hat{L}_{A}\circ$ on fuzzy $CP^{n}$ are mapped to $\hat{L}_{A}\circ\rightarrow L_{A}^{(q)}=\bar{w}_{\alpha}(T_{A})^{\alpha}{}_{\beta}\frac{\partial}{\partial\bar{w}_{\beta}}-w^{\alpha}(T_{A}^{*})_{\alpha}{}^{\beta}\frac{\partial}{\partial w^{\beta}}.$ (E.31) When $q=0$, these operators act on the functions (E.24) and they can be identified with the Killing vectors on $CP^{n}$. In the case $q\neq 0$, however, they act on the local sections (E.26) so that the derivative along the $U(1)$ fiber direction does not vanish and yields additional terms which are proportional to the charge $q$. In this case, the operators (E.31) can be interpreted as the angular momentum operators on $CP^{n}$ in the presence of a monopole with the charge $q$. ## References * [1] T. Banks, W. Fischler, S. H. Shenker and L. Susskind, Phys. Rev. D 55 (1997) 5112 [arXiv:hep-th/9610043]. * [2] N. Ishibashi, H. Kawai, Y. Kitazawa and A. Tsuchiya, Nucl. Phys. B 498 (1997) 467 [arXiv:hep-th/9612115]. * [3] R. Dijkgraaf, E. P. Verlinde and H. L. Verlinde, Nucl. Phys. B 500 (1997) 43 [arXiv:hep-th/9703030]. * [4] T. Eguchi and H. Kawai, Phys. Rev. Lett. 48 (1982) 1063. * [5] G. Bhanot, U. M. Heller and H. Neuberger, Phys. Lett. B 113, 47 (1982). * [6] G. Parisi, Phys. Lett. B 112, 463 (1982). * [7] D. J. Gross and Y. Kitazawa, Nucl. Phys. B 206, 440 (1982). * [8] S. R. Das and S. R. Wadia, Phys. Lett. B 117 (1982) 228 [Erratum-ibid. B 121 (1983) 456]. * [9] A. Gonzalez-Arroyo and M. Okawa, Phys. Rev. D 27 (1983) 2397. * [10] H. Aoki, N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa and T. Tada, Nucl. Phys. B 565 (2000) 176 [arXiv:hep-th/9908141]. * [11] W. I. Taylor, Phys. Lett. B 394 (1997) 283 [arXiv:hep-th/9611042]. * [12] N. Arkani-Hamed, A. G. Cohen and H. Georgi, Phys. Lett. B 513, 232 (2001) [arXiv:hep-ph/0105239]. * [13] D. B. Kaplan, E. Katz and M. Unsal, JHEP 0305, 037 (2003) [arXiv:hep-lat/0206019]. * [14] M. Hanada, H. Kawai and Y. Kimura, Prog. Theor. Phys. 114 (2006) 1295 [arXiv:hep-th/0508211]. * [15] G. Ishiki, S. Shimasaki, Y. Takayama and A. Tsuchiya, JHEP 0611 (2006) 089 [arXiv:hep-th/0610038]. * [16] J. Maldacena, M. M. Sheikh-Jabbari and M. Van Raamsdonk, JHEP 0301 (2003) 038 [arXiv:hep-th/0211139]. * [17] D. Berenstein, J. M. Maldacena and H. Nastase, JHEP 0204 (2002) 013 [arXiv:hep-th/0202021]. * [18] H. Lin and J. M. Maldacena, Phys. Rev. D 74, 084014 (2006) [arXiv:hep-th/0509235]. * [19] J. Madore, Class. Quant. Grav. 9 (1992) 69. * [20] H. Grosse and J. Madore, Phys. Lett. B 283 (1992) 218. * [21] H. Grosse, C. Klimcik and P. Presnajder, Int. J. Theor. Phys. 35 (1996) 231 [arXiv:hep-th/9505175]. * [22] U. Carow-Watamura and S. Watamura, Commun. Math. Phys. 212 (2000) 395 [arXiv:hep-th/9801195]. * [23] S. Iso, Y. Kimura, K. Tanaka and K. Wakatsuki, Nucl. Phys. B 604 (2001) 121 [arXiv:hep-th/0101102]. * [24] H. Grosse, C. Klimcik and P. Presnajder, Commun. Math. Phys. 178 (1996) 507 [arXiv:hep-th/9510083]. * [25] S. Baez, A. P. Balachandran, B. Ydri and S. Vaidya, Commun. Math. Phys. 208 (2000) 787 [arXiv:hep-th/9811169]. * [26] G. Landi, J.Geom.Phys. 37 (2001) 47. * [27] H. Aoki, S. Iso and K. Nagao, Nucl. Phys. B 684 (2004) 162 [arXiv:hep-th/0312199]. * [28] U. Carow-Watamura, H. Steinacker and S. Watamura, J. Geom. Phys. 54 (2005) 373 [arXiv:hep-th/0404130]. * [29] T. Ishii, G. Ishiki, S. Shimasaki and A. Tsuchiya, JHEP 0705 (2007) 014 [arXiv:hep-th/0703021]. * [30] T. Ishii, G. Ishiki, K. Ohta, S. Shimasaki and A. Tsuchiya, arXiv:0711.4235 [hep-th]. * [31] G. Alexanian, A. P. Balachandran, G. Immirzi and B. Ydri, J. Geom. Phys. 42, 28 (2002) [arXiv:hep-th/0103023]. * [32] A. P. Balachandran, B. P. Dolan, J. H. Lee, X. Martin and D. O’Connor, J. Geom. Phys. 43, 184 (2002) [arXiv:hep-th/0107099]. * [33] Y. Kitazawa, Nucl. Phys. B 642, 210 (2002) [arXiv:hep-th/0207115]. * [34] H. Grosse and H. Steinacker, Nucl. Phys. B 707, 145 (2005) [arXiv:hep-th/0407089]. * [35] B. P. Dolan, I. Huet, S. Murray and D. O’Connor, JHEP 0707, 007 (2007) [arXiv:hep-th/0611209]. * [36] T. H. Buscher, Phys. Lett. B 194, 59 (1987). * [37] X. C. de la Ossa and F. Quevedo, Nucl. Phys. B 403 (1993) 377 [arXiv:hep-th/9210021]. * [38] T. Azuma, S. Bal, K. Nagao and J. Nishimura, JHEP 0605 (2006) 061 [arXiv:hep-th/0405277]. * [39] T. Ishii, G. Ishiki, S. Shimasaki and A. Tsuchiya, to appear. * [40] R. C. Myers, JHEP 9912 (1999) 022 [arXiv:hep-th/9910053]. * [41] N. w. Kim, T. Klose and J. Plefka, Nucl. Phys. B 671 (2003) 359 [arXiv:hep-th/0306054]. * [42] T. T. Wu and C. N. Yang, Nucl. Phys. B 107 (1976) 365. * [43] Y. Choquet-Bruhat, Analysis, Manifolds and Physics, Part II (North Holland, 2000). * [44] M. Nakahara, “Geometry, topology and physics,” Boca Raton, USA: Taylor $\&$ Francis (2003) 573 p. * [45] G. L. Naber, “Topology, geometry, and gauge fields: Foundations,” New York, USA: Springer (1997) 396 p. (Texts in applied mathematics. 25). * [46] G. L. Naber, “Topology, geometry, and gauge fields: Interactions,” New York, USA: Springer (2000) 443 p. * [47] C. N. Yang, J. Math. Phys. 19, 320 (1978). * [48] G. Ishiki, Y. Takayama and A. Tsuchiya, JHEP 0610 (2006) 007 [arXiv:hep-th/0605163]. * [49] K. Dasgupta, M. M. Sheikh-Jabbari and M. Van Raamsdonk, “Matrix perturbation theory for M-theory on a PP-wave,” JHEP 0205 (2002) 056 [arXiv:hep-th/0205185]. * [50] D. Varshalovich, A. Moskalev and V. Khersonskii, Quantum Theory of Angular Momentum (World Scientific, Singapore, 1988). * [51] V. Gerdt, R. Horan, A. Khvedelidze, M. Lavelle, D. McMullan and Yu. Palii, J. Math. Phys. 47, 112902 (2006) [arXiv:hep-th/0511245].

arxiv-papers

2024-09-04T02:48:53.882210

0802.2873

# Environmentally induced corrections to the geometric phase in a two-level system Fernando C. Lombardo and Paula I. Villar Departamento de Física Juan José Giambiagi, FCEyN UBA, Facultad de Ciencias Exactas y Naturales, Ciudad Universitaria, Pabellón I, 1428 Buenos Aires, Argentina ###### Abstract We calculate the geometric phase for different open systems (spin-boson and spin-spin models). We study not only how they are corrected by the presence of the different type of environments but also discuss the appearence of decoherence effects. These should be taken into account when planning experimental setups to study the geometric phase in the nonunitary regime. We propose a model with slow decoherence rate in which the geometric phase is still modified and might be measured. Since the work of Berry Berry , the notion of geometric phases has been shown to have important consequences for quantum systems. Berry demonstrated that closed quantum systems could acquire phases that are geometric in nature. He showed that, besides the usual dynamical phase, an additional phase related to the geometry of the space state is generated during an adiabatic evolution. The existence of such a phase is also true for open quantum systems. In particular, when a static potential is exerted on the main system, the wave function of this system acquires a phase and hence the interference term appears multiplied by a phase factor $e^{i\varphi}$. In an interference experiment, its effect on the pattern of the system is related to the phase’s statistical character, particularly, in situations where the potential is not static. Yet more importantly, any source of stochastic noise would create a decaying coefficient, usually called decoherence factor $F$. For a general case, the phase $\varphi$ is described by means of a distribution function LomMazziVillar ; Sanders . No matter how weak the coupling that prevents the system from being isolated, the evolution of an open quantum system is plagued by nonunitary features like decoherence and dissipation. Decoherence, in particular, is a quantum effect whereby the system loses its ability to exhibit coherent behaviour and appears as soon as the partial waves of the main system shift the environment into states orthogonal to each other Visibility . Nowadays, decoherence stands as a serious obstacle in quantum information processing. The geometric phase (GP) for a mixed state under nonunitary evolution has been defined by Tong et.al.Tong as $\displaystyle\Phi$ $\displaystyle=$ $\displaystyle{\rm arg}\\{\sum_{k}\sqrt{\varepsilon_{k}(0)\varepsilon_{k}(\tau)}\langle\Psi_{k}(0)|\Psi_{k}(\tau)\rangle$ (1) $\displaystyle\times$ $\displaystyle e^{-\int_{0}^{\tau}dt\langle\Psi_{k}|\frac{\partial}{\partial t}|{\Psi_{k}}\rangle}\\},$ where $\varepsilon_{k}(t)$ are the eigenvalues and $|\Psi_{k}\rangle$ the eigenstates of the reduced density matrix $\rho_{\rm r}$ (obtained after tracing over the reservoir degrees of freedom). In the last definition, $\tau$ denotes a time after the total system completes a cyclic evolution when it is isolated from the environment. Taking the effect of the environment into account, the system no longer undergoes a cyclic evolution. However, we will consider a quasicyclic path ${\cal P}:t~{}\epsilon~{}[0,\tau]$ with $\tau=2\pi/\Omega$ ($\Omega$ the system’s frequency). When the system is open, the original GP, i.e. the one that would have been obtained if the system had been closed $\Phi^{U}$, is modified. That means, in a general case, the phase is $\Phi=\Phi^{U}+\delta\Phi$, where $\delta\Phi$ depends on the kind of environment coupled to the main systembarnerjee . It is expected that GPs can be only observed in interference experiments carried out in a time scale slow enough to ignore nonadiabatic corrections, but rapid enough to avoid the destruction of the interference pattern by decoherence Gefen . So far, there has been no experimental observation of GPs for mixed states under nonunitary evolutions. The purpose of this short article is to study how GPs are affected by decoherence in different physical scenarios. The decoherence time results very important when trying to measure the GPs since for times longer than the former the GPs, literally, disappear. In this framework, we shall compute the GP for different models using the kinematical approach to the GP given by Eq.(1), and compare the results therein obtained. We shall start by reviewing some of our previuos resultsPRA , and then we shall present further results concerning the environmentally induced corrections to the GP ($\delta\Phi$) in realistic (even experimentally feasible) models. Purely Decohering Solvable Spin-Boson Model. In this section, we shall review the basic results for an open quantum system by presenting a model which is simple enough to be solved analyticallyPRA . In spite of its simplicity, this model captures many of the elements of decoherence theories and sheds some insight into the modification of the GPs due to the presence of the environment. This model has been used by many authors to model decoherence in quantum computersEkert and, in particular, it is extremely relevant to the proposal for observing GPs in a superconducting nanocircuit Falci . The Hamiltonian that describes the complete evolution of the two-state system interacting with the external environment is: $H_{\rm SB}=\frac{1}{2}\hbar\Omega\sigma_{z}+\frac{1}{2}\sigma_{z}\sum_{k}\lambda_{k}(a^{\dagger}_{k}+a_{k})+\sum_{k}\hbar\omega_{k}a_{k}^{{\dagger}}a_{k},$ (2) where the environment is described as a set of harmonic oscillators with a linear coupling in the oscillator coordinate. The interaction between the two- state system and the environment is entirely represented by a Hamiltonian in which the coupling is only through $\sigma_{z}$. In this particular case, $[\sigma_{z},H_{\rm int}]=0$ and the corresponding master equation is much simplified, with no frequency renormalization and dissipation effects. In other words, the model describes a purely decohering mechanism, solely containing the diffusion term ${\cal D}(t)$ whose master equation, after tracing out the environmental degrees of freedom, is given by (with $\hbar=1$) $\dot{\rho_{\rm r}}=-i\Omega[\sigma_{z},\rho_{\rm r}]-{\cal D}(t)[\sigma_{z},[\sigma_{z},\rho_{\rm r}]],$ (3) where ${\cal D}(s)=\int_{0}^{s}ds^{\prime}\int_{0}^{\infty}d\omega I(\omega)\coth\bigg{(}\frac{\omega}{2k_{B}T}\bigg{)}\cos(\omega(s-s^{\prime})),$ and $I(\omega)$ is the spectral density of the environment, usually, $I(\omega)\sim\omega^{n}$ up to some frequency $\Lambda$ that may be large compared to $\Omega$. In particular, the case with $n=1$ is the “ohmic” environment. Then, it is easy to check that $\rho_{\rm r_{01}}(t)=e^{-i\Omega t-{\cal A}(t)}\rho_{\rm r_{01}}(0)$ is the solution for the off-diagonal terms (while the populations remain constant), where ${\cal A}(t)=\int_{0}^{t}ds{\cal D}(t)$. In the following, we shall call $F=\exp(-{\cal A}(t))$ the decoherence factor. Hence, the GP for an initial pure state of the form $|\Psi(0)\rangle=\cos\theta_{0}/2|e\rangle+\sin\theta_{0}/2|g\rangle$, related to a quasicyclic path ${\cal P}:t~{}\epsilon~{}[0,\tau]$ up to first order in the dissipative constant ($\gamma_{0}\propto\lambda_{k}^{2}$) isPRA $\displaystyle\Phi_{\rm SB}$ $\displaystyle\approx$ $\displaystyle\pi(1-\cos\theta_{0})-\frac{\gamma_{0}}{2}\Omega\sin^{2}\theta_{0}\int_{0}^{\tau}~{}dt\bigg{[}\frac{\partial F(t)}{\partial\gamma_{0}}\bigg{]}\bigg{|}_{\gamma_{0}=0}$ (4) $\displaystyle+$ $\displaystyle{\cal O}(\gamma_{0}^{2}).$ In the right side of last expression, we have performed a serial expansion in terms of $\gamma_{0}$. The first term corresponds to the unitary phase $\Phi^{U}$. Consequently, we see that the unitary GP is corrected by a term which depends directly on the kind of environment present PRA . For example, for an ohmic environment in the limit of high temperature $\delta\Phi_{\rm SB}^{\rm HT}=\pi^{2}(\gamma_{0}/\Omega)\pi k_{B}T\sin^{2}\theta_{0}$, while the same environment at zero temperature modifies the unitary phase as $\delta\Phi_{\rm SB}^{\rm T=0}=\frac{\pi}{2}\gamma_{0}(-1+\log(2\pi\Lambda/\Omega))\sin^{2}\theta_{0}$. These results can be compared with those in Sanders ; TongPRA . In those cases, the correction due to the environment is also proportional to $(\gamma_{0}/\Omega)\sin^{2}(\theta_{0})$ (mainly due to the simplified decoherence factor $F=\exp(-\gamma_{0}t)$). However, in our model, these corrections enclose the main characteristic of the model of bath we are taking into account, which allows to evaluate the decoherence time scale properly. In the case of having a bosonic environment, composed by an infinite set of harmonic oscillators, it is not difficult to evaluate the decoherence time scale. This scale should be compared with the time $\tau=2\pi/\Omega$ at which one expect to measure the GP. In the case of an ohmic bath in the high temperature limit, the decoherece time is $t_{D}=1/(\gamma_{0}\pi k_{B}T)$, which is really a very short time scale compared with $\tau$. In the zero temperature case, the decoherence time scales as $t_{D}\sim e^{1/\gamma_{0}}/\Lambda$ which, indeed, can be very large in the case of underdamped environments. In conclusion, one could expect that the GP can be only detected at very low temperature when the atom is mainly coupled to a bosonic fieldPRA . Spin-Spin Model. We shall study another simple solvable model in which the size of the environment has a relevant role. Consider a two-level system coupled to $n$ other two-level systemsZurek . Our main subsystem (one qubit) interacts with the rest of the environmental spins by a bilinear interaction described by the interaction hamiltonian ${\cal H}_{\rm SS}=\frac{\pi}{2}\sum_{k=2}^{N}J_{1k}\sigma_{z}^{1}\sigma_{z}^{k},$ (5) where the system qubit is denoted by the superscript “1”. This coupling is also a purely phase damping mechanism, as in the spin-boson model mentioned above. Given a factorizable initial state of the form $|\Phi(0)\rangle_{1}=[a|0\rangle_{1}+b|1\rangle_{1}]\prod_{k=2}^{n}(\alpha_{k}|0\rangle_{k}+\beta_{k}|1\rangle_{k})$, the interacction entangles the state of the system with the environment. This means that after the interaction, both system and environment states are not longer factorizable. Similarly to the spin-boson model, the density matrix will have constant populations (since $[\sigma_{z},{\cal H}_{tot}]=0$) and the off-diagonal terms will be multiplied by a decoherence factor, as $\rho_{01}^{s}=ab^{*}z(t)$ where $z(t)=\prod_{k=2}^{N}[\cos(\pi J_{1k}t)+i\phi_{+}\phi_{-}\sin(\pi J_{1k}t)],$ (6) where $\phi_{\pm}=|\alpha_{k}|\pm|\beta_{k}|$. Note that $z(t)$ depends on the initial conditions of the environment only through the probabilities of finding the system in the eigenstates of the interaction Hamiltonian $|\alpha_{k}|,|\beta_{k}|$ Zurek . In this case, $z(t)$ plays the role of the decoherence factor $F$ since contains the information related to the tracing out of the spin environment degrees of freedom. In particular, the magnitude of $z(t)$ determines the damping of the phase information originally contained in $\rho_{01}(0)$. In particular, when $|z(t)|\rightarrow 0$, the nonunitary evolution and the irreversibility of the process are evident. However, information can be in principle recoverable for a finite system since $|z(t)|$ is at worst quasiperiodicZurek . The effectiveness of the decoherence mechanism is determined by the dimension of the environment. However, in any case, if $z(t)$ is a complex function, it implies a phase shift and an attenuation of the interference fringes, i.e. a dephasing or decoherent process. In principle, the correction induced on the GP is the same as in Eq.(4), just replacing $F(t)$ by $z(t)$. Let’s take for example the particular case when the environment is composed of only one spin ($k=2$ in Eq.(5)). For the same initial state mentioned above, and considering $|\alpha_{k}|=|\beta_{k}|$, we obtain $z(t)=\cos(\pi Jt)$ (where we set $J\equiv J_{12}$). In this case, $z(t)$ is real and then, its only contribution is to the phase shift of the system, while one spin environment is not effective inducing decoherence on the system. Nevertheless, we will show that this factor induces a correction to the GP which is quadratic in the coupling strength with the environment. In such a case, if one performs a serial expansion in powers of the coupling constant $J$, one obtains that the modification to the unitary phase is at second order. Thus, the correction to the unitary GP is given by $\delta\Phi_{\rm SS}^{\rm z}\approx\frac{4\pi^{4}}{3\Omega^{2}}J^{2}\sin^{2}\theta_{0}.$ (7) This simple result shows that correction to the unitary GP induced by the presence of this environment can be, in principle, detected in an interference experiment, without the constraint imposed by the decoherence time scale. At zero-order, the unitary GP is the same as in Eq.(4) $\Phi^{U}=\pi(1-\cos\theta_{0})$. Hierarchical Qubit-Qubit Decoherence Model. Herein, we shall compute the GP’s correction for a model very similar to the above described spin-spin one. This scenario has the particular feature that it can be implemented to simulate quantum decoherence Teklemariam . In this case, the environment is also limited to only one spin (qubit). However, through the strategy of randomly redressing the phase of the environment qubits during the interaction with the system, it is possible to simulate a much larger environment. Therefore, the result must be averaged over many realizations of this evolution. The dimension of the Hilbert space can not be larger than $N^{2}$, where $N$ is the dimension of the local main system. To remove the information from the finite quantum environment, a classical stochastic field is included. Basically, the technique consists of applying classical kicks to the environment qubits, and then averaging over the realizations of this stochastic noise. This has the effect of scrambling the system information after it has been stored in the quantum environment through the coupling interaction. We shall consider the evolution of this system subject to a sequence of kicks that only affect the environment qubit. Every kick is generated by a transverse magnetic field whose effect is to rotate the environment qubit around the $y$ axis by an angle $\epsilon$ included randomly in the interval $(-\alpha,\alpha)$. In this case, the reduced density matrix is similar to the above models, but for a different decoherence factor $F$. The off-diagonal terms are $\rho_{{\rm r}_{ij}}=ab^{*}f_{ij}$, where $f_{ij}$ carries all the information about the effect of the environment qubit on the system qubit. It is obtained after tracing out the environment degree of freedom and averaging over the many realizations of the external magnetic field Teklemariam . In the case that there are no kicks, i.e. $\alpha=0$, and $f_{12}=\cos(\pi Jt)-ip_{z}\sin(\pi Jt)$, which agrees with the spin-spin model described above ($p_{z}$ is the initial polarization of the environment qubit). In this case there is no decoherence and the GP-correction is given by Eq.(7). The decoherence factor is independent of the kicking rate (no kicks in this limit), and the system qubit rotates independently of the environment qubit. If one allows a complete ramdomization, i.e. the kick angles $\epsilon_{j}$ may vary over the entire interval between $0$ and $2\pi$, the decoherence factor can be approximated, in the limit of faster kicks, byTeklemariam $f_{01}(\Gamma,t)\approx e^{-\frac{\pi^{2}J^{2}t}{2\Gamma}}-ip_{z}\sin(\frac{\pi J}{\Gamma})e^{-\frac{\pi^{2}J^{2}t}{2\Gamma}}$ , where $\Gamma$ is the kick rate. Using this expression, one can evaluate the correction induced on the GP ($\delta\Phi_{\rm SS}^{\rm cr}$) (for the particular case $p_{z}=0$) as $\delta\Phi_{\rm SS}^{\rm cr}\approx\frac{\pi^{4}}{2\Gamma\Omega}J^{2}\sin^{2}\theta_{0}.$ (8) In this situation, the decoherence time is given by $t_{D}=2\Gamma/(J^{2})$ which is larger than $\tau$, making the decoherence process negligible if trying to measure these corrections to the GP. Finally, we shall consider the case of small angles, since it is the regime used by simulations and also for decoherence experiments (usually with $\alpha=\pi/20$). In such a case, it is possible to estimate the decoherence factor as $f_{12}=e^{-\Gamma t\epsilon}(1+\epsilon/2)[\cos(\pi Jt)-ip_{z}\sin(\pi Jt)]$, where $\epsilon=2/3\alpha^{2}$ is a small number ($\epsilon\approx 0.016$ for the given experimentally accesible value of $\alpha$ mentioned above). This decoherence factor determines a very large dephasing scale: $t_{D}=1/(\Gamma\epsilon)$. In this case, we can also evaluate the environmentally induced correction to the GP (up to sencond order in the coupling with the environment and also for small $\epsilon$, and $p_{z}=0$) $\delta\Phi_{\rm SS}^{\rm sa}$ as $\delta\Phi_{\rm SS}^{\rm sa}\approx\frac{\pi}{\Omega}\sin^{2}\theta_{0}\left[\left(\pi\Gamma-\frac{\Omega}{2}\right)\epsilon+\frac{2}{3}\frac{\pi^{4}}{\Omega}J^{2}\right].$ (9) This correction to the GP has a term independent of the coupling constant with the environment $J$, which in this limit is linear with $\epsilon$, the small angle that is rotated due to the kicks. It is worthly noticing that in the limit of $\epsilon\rightarrow 0$, Eq.(9) coincides with the result given by the Zurek’s model. Even though this is a very simple quantum open system model, it is of great interest due to the fact that this scheme enables simulation of the quantum decoherence that usually appears for larger environments. As we have mentioned, one qubit as environment is not enough to produce decoherence on the system qubit in the Zurek’s model. However, in the present case, the phase damping is induced by a sequence of kicks that affect only the environment qubit, generated by a magnetic field that rotates the environment spin by an angle $\epsilon$. We believe that this practical implementation could be suitable for measuring of the complete GP in the case of a nonunitary evolution. This work was supported by UBA, CONICET, and ANPCyT, Argentina. ## References * (1) M.V. Berry, Proc. R. Soc. Lond. A 392, 45 (1984). * (2) F.C. Lombardo, F.D. Mazzitelli, and P.I. Villar, Phys. Rev. A 72, 042111 (2005); F.C. Lombardo and P.I. Villar,J.Phys. A 39, 6509 (2006). * (3) K.P. Marzlin, S. Ghose, and B.C. Sanders, Phys. Rev. Lett. 93, 260402 (2004). * (4) P.I. Villar and F.C. Lombardo, Journal of Physics: Conf. Ser. 67, 012041 (2007); Int. J. Mod. Phys. B 21, 4659 (2007). * (5) D. M. Tong, E. Sjöqvist, L. C. Kwek, and C. H. Oh, Phys. Rev. Lett. 93, 080405 (2004). * (6) S. Banerjee and R. Srikanth Eur. Phys. J. D 46, 335 (2008); A.T. Rezakhani and P. Zanardi, Phys. Rev. A 73, 012107 (2006). * (7) R.S. Whitney, Y. Makhlin, A. Shnirman, and Y. Gefen, Phys. Rev. Lett. 94, 070407 (2005). * (8) F.C. Lombardo and P.I. Villar, Phys. Rev. A 74, 042311 (2006). * (9) G.M. Palma, K. Suominen, and A. Ekert, Proc. R. Soc. London, Ser. A 452, 567 (1996); L. Viola and S. Lloyd, Phys. Rev. A 58, 2733 (1998). * (10) G. Falci, R. Fazio, G.M. Palma, J. Siewert, and V. Vedral, Nature (London) 407, 355 (2000). * (11) A. Carollo, I. Fuentes-Guridi, M. Franca Santos, and V. Vedral, Phys. Rev. Lett. 90, 160402 (2003); Phys. Rev. Lett. 92, 020402 (2004); X.X. Yi, D. M. Tong, L. C. Wang, L. C. Kwek, and C. H. Oh, Phys. Rev. A 73, 052103 (2006). * (12) W.H. Zurek, Phys. Rev. D 26, 1862 (1982). * (13) G. Teklemariam, E.M. Fortunato, C.C. López, J. Emerson, J.P. Paz, T.F. Havel and D.G. Cory, Phys. Rev. A 67, 062316 (2003). * (14) A. Ekert et al., J. Mod. Opt. 47, 2501 (2000).

arxiv-papers

2024-09-04T02:48:53.892549

0802.2889

# Via Aristotle, Leibniz, Berkeley & Mach to necessarily fractal large-scale structure in the Universe D. F. Roscoe ###### Abstract The claim that the large scale structure of the Universe is heirarchical has a very long history going back at least to Charlier’s papers of the early 20th century. In recent years, the debate has centered largely on the works of Sylos Labini, Joyce, Pietronero and others, who have made the quantative claim that the large scale structure of the Universe is quasi-fractal with fractal dimension $D\approx 2$. There is now a concensus that this is the case on medium scales, with the main debate revolving around what happens on the scales of the largest available modern surveys. Apart from the (essentially sociological) problem that their thesis is in absolute conflict with any concept of a Universe with an age of $\approx 14$ billion years or, indeed, of any finite age, the major generic difficulty faced by the proponents of the heirarchical hypothesis is that, beyond hypothesizing the case (eg: Nottale’s _Scale Gravity_), there is no obvious mechanism which would lead to large scale structure being non-trivially fractal. This paper, which is a realization of a worldview that has its origins in the ideas of Aristotle, Leibniz, Berkeley and Mach, provides a surprising resolution to this problem: in its essence, the paper begins with a statement of the primitive self-evident relationship which states that, in the universe of our experience, the amount of material, $M$, in a sphere of redshift radius $R_{z}$ is a monotonic increasing function of $R_{z}$. However, because the precise relationship between any Earth-bound calibration of radial distance and $R_{z}$ is unknowable then fundamental theories cannot be constructed in terms of $R_{z}$, but only in terms of a radial measure, $R$ say, calibrated against known physics. The only certainty is that, for any realistic calibration, there will exist a monotonic increasing relationship between $R_{z}$ and $R$ so that we have $M=f(R)$ for a monotonic increasing function $f$. But the monotonicity implies $R=f^{-1}(M)\equiv g(M)$ which, _in the absence of any prior calibration_ _of_ $R$, can be interpreted as the _definition_ of the radius of an astrophysical sphere in terms of the amount of mass it contains - which is the point of contact with the ideas of Aristotle, Leibniz, Berkeley and Mach. The development of this idea, and the resulting implications for the geometrical structure of physical space, leads necessarily to the final result, which is that large scale structure in the Universe of our experience is fractal of dimension $D=2.$ ## 1 Introduction: The analysis of this paper shows how, as a necessary consequence of a world- view that has its origins in the ideas of Aristotle, Leibniz, Berkeley and Mach, mass in an isotropic Universe is distributed according to $M=m_{0}\left(\frac{R}{R_{0}}\right)^{2}+2\sqrt{\frac{m_{0}m_{1}}{d_{0}}}\left(\frac{R}{R_{0}}\right),$ for undertermined constants $d_{0},m_{0}$ and $m_{1}$ so that, on very large scales, $M\sim R^{2}$. Furthermore, it is shown that, in the limit of very large scales, the Universe approaches a state of perfect dynamical (not static) equilibrium. We begin with a brief review of the phenomenological arguments which led to the basic hypothesis that large-scale structure in the Universe is non- trivially fractal, and comment briefly on the ramifications to canonical cosmology _if_ this hypothesis proves to be correct. We then give a brief account comparing and contrasting the two major world- views which have dominated thinking about the nature of the cosmos for over two thousand years before, finally, providing the quantitative development which leads to the main result of this paper. ### 1.1 The heirarchical Universe A basic assumption of the Standard Model of modern cosmology is that, on some scale, the universe is homogeneous; however, in early responses to suspicions that the accruing data was more consistent with Charlier’s conceptions [1, 2, 3] of an hierarchical universe than with the requirements of the Standard Model, de Vaucouleurs [4] showed that, within wide limits, the available data satisfied a mass distribution law $M\approx r^{1.3}$, whilst Peebles [5] found $M\approx r^{1.23}$. The situation, from the point of view of the Standard Model, continued to deteriorate with the growth of the data-base to the point that, Baryshev et al [6] state > _…the scale of the largest inhomogeneities (discovered to date) is > comparable with the extent of the surveys, so that the largest known > structures are limited by the boundaries of the survey in which they are > detected._ For example, several redshift surveys of the late 20th century, such as those performed by Huchra et al [7], Giovanelli and Haynes [8], De Lapparent et al [9], Broadhurst et al [10], Da Costa et al [11] and Vettolani et al [12] etc discovered massive structures such as sheets, filaments, superclusters and voids, and showed that large structures are common features of the observable universe; the most significant conclusion drawn from all of these surveys was that the scale of the largest inhomogeneities observed in the samples was comparable with the spatial extent of those surveys themselves. In the closing years of the century, several quantitative analyses of both pencil-beam and wide-angle surveys of galaxy distributions were performed: three examples are given by Joyce, Montuori & Labini [13] who analysed the CfA2-South catalogue to find fractal behaviour with $D\,$=$\,1.9\pm 0.1$; Labini & Montuori [14] analysed the APM-Stromlo survey to find fractal behaviour with $D\,$=$\,2.1\pm 0.1$, whilst Labini, Montuori & Pietronero [15] analysed the Perseus-Pisces survey to find fractal behaviour with $D\,$=$\,2.0\pm 0.1$. There are many other papers of this nature, and of the same period, in the literature all supporting the view that, out to $30-40h^{-1}Mpc$ at least, galaxy distributions appeared to be fractal with $D\,$$\approx$$\,2$. This latter view became widely accepted (for example, see Wu, Lahav & Rees [16]), and the open question became whether or not there was transition to homogeneity on some sufficiently large scale. For example, Scaramella et al [17] analyse the ESO Slice Project redshift survey, whilst Martinez et al [18] analyse the Perseus-Pisces, the APM-Stromlo and the 1.2-Jy IRAS redshift surveys, with both groups claiming to find evidence for a cross-over to homogeneity at large scales. At around about this time, the argument reduced to a question of statistics (Sylos Labini & Gabrielli [19], Gabrielli & Sylos Labini [20], Pietronero & Sylos Labini [21]): basically, the proponents of the fractal view began to argue that the statistical tools (eg correlation function methods) widely used to analyse galaxy distributions by the proponents of the opposite view are deeply rooted in classical ideas of statistics and implicitly assume that the distributions from which samples are drawn are homogeneous in the first place. Recently, Hogg et al [22], having accepted these arguments, applied the techniques argued for by the pro-fractal community (which use the _conditional density_ as an appropriate statistic) to a sample drawn from Release Four of the Sloan Digital Sky Survey. They claim that the application of these methods does show a turnover to homogeneity at the largest scales thereby closing, as they see it, the argument. In response, Labini et al [23] have criticized their paper on the basis that the strength of the conclusions drawn is unwarrented given the deficencies of the sample \- in effect, that it is not big enough. To summarize, the proponents of non-trivially fractal large-scale structure have won the argument out to medium distances and the controversy now revolves around the largest scales encompassed by the SDSS. ### 1.2 Theoretical implications The notion of non-trivially fractal large-scale structure in the Universe is highly problematic for proponents of any form of big-bang cosmology, for two reasons: 1. 1. any form of big-bang cosmology implies a finite age for the Universe (approximately 14 billion years in the canonical picture). By contrast, a non- trivially fractal Universe implies the existence of vast individual structures which cannot possibly form on any determinate time scale. So there is an obvious conflict; 2. 2. if there is non-trivially fractal structure in the Universe then, ideally, the mechanism for the formation of such structure should be open to a theoretical understanding. But the equations of General Relativity are hyperbolic so that universal structure at any epoch is always going to be determined primarily by initial conditions - in other words, large-scale fractal structure in the Universe cannot be explained within the confines of General Relativity, but only in terms of conditions which are external to it. It follows that the hypothesis that large-scale structure in the Universe is non-trivially fractal and the hypothesis that General Relativity is the fundamentally correct theory by which gravitational processes in the Universe can be understood, are in direct opposition to each other. This implies that, if we are to take the fractal hypothesis seriously (as the phenomenology suggests we should), then only a radical review of gravitational theory can hope to provide any hope of understanding how such structure can occur. ## 2 A brief history of ideas of space and time The conception of space as the container of material objects is generally considered to have originated with Democritus and, for him, it provided the stage upon which material things play out their existence - _emptiness_ exists and is that which is devoid of the attribute of _extendedness_ (although, interestingly, this latter conception seems to contain elements of the opposite view upon which we shall comment later). For Newton [24], an extension of the Democritian conception was basic to his mechanics and, for him: > _… absolute space, by its own nature and irrespective of anything external, > always remains immovable and similar to itself._ Thus, the absolute space of Newton was, like that of Democritus, the stage upon which material things play out their existence - it had an _objective existence_ for Newton and was primary to the order of things. In a similar way, time - _universal time,_ an absolute time which is the same everywhere - was also considered to possess an objective existence, independently of space and independently of all the things contained within space. The fusion of these two conceptions provided Newton with the reference system - _spatial coordinates_ defined at a _particular time_ \- by means of which, as Newton saw it, all motions could be quantified in a way which was completely independent of the objects concerned. It is in this latter sense that the Newtonian conception seems to depart fundamentally from that of Democritus - if _emptiness_ exists and is devoid of the attribute of _extendedness_ then, in modern terms, the _emptiness_ of Democritus can have no _metric_ associated with it. But it is precisely Newton’s belief in _absolute space & time_ (with the implied virtual clocks and rods) that makes the Newtonian conception a direct antecedent of Minkowski spacetime - that is, of an empty space and time within which it is possible to have an internally consistent discussion of the notion of _metric._ The contrary view is generally considered to have originated with Aristotle [25, 26] for whom there was no such thing as a _void_ \- there was only the _plenum_ within which the concept of the _empty place_ was meaningless and, in this, Aristotle and Leibniz [27] were at one. It fell to Leibniz, however, to take a crucial step beyond the Aristotelian conception: in the debate of Clarke-Leibniz (1715$\sim$1716) [28] in which Clarke argued for Newton’s conception, Leibniz made three arguments of which the second was: > _Motion and position are real and detectable only in relation to other > objects … therefore empty space, a void, and so space itself is an > unnecessary hypothesis._ That is, Leibniz introduced a _relational_ concept into the Aristotelian world view - what we call _space_ is a projection of _relationships_ between material bodies into the perceived world whilst what we call _time_ is the projection of ordered _change_ into the perceived world. Of the three arguments, this latter was the only one to which Clarke had a good objection - essentially that _accelerated motion,_ unlike uniform motion, can be perceived _without_ reference to external bodies and is therefore, he argued, necessarily perceived with respect to the _absolute space_ of Newton. It is of interest to note, however, that in rebutting this particular argument of Leibniz, Clarke, in the last letter of the correspondence, put his finger directly upon one of the crucial consequences of a relational theory which Leibniz had apparently not realized (but which Mach much later would) stating as absurd that: > … _the parts of a circulating body (suppose the sun) would lose the_ vis > centrifuga _arising from their circular motion if all the extrinsic matter > around them were annihilated._ This letter was sent on October 29th 1716 and Leibniz died on November 14th 1716 so that we were never to know what Leibniz’s response might have been. Notwithstanding Leibniz’s arguments against the Newtonian conception, nor Berkeley’s contemporary criticisms [29], which were very similar to those of Leibniz and are the direct antecedents of Mach’s, the practical success of the Newtonian prescription subdued any serious interest in the matter for the next 150 years or so until Mach himself picked up the torch. In effect, he answered Clarke’s response to Leibniz’s second argument by suggesting that the _inertia_ of bodies is somehow induced within them by the large-scale distribution of material in the universe: > _… I have remained to the present day the only one who insists upon > referring the law of inertia to the earth and, in the case of motions of > great spatial and temporal extent, to the fixed stars …_ [30] thereby generalizing Leibniz’s conception of a relational universe. Mach was equally clear in expressing his views about the nature of time: in effect, he viewed _time_ (specifically Newton’s _absolute time_) as a meaningless abstraction. All that we can ever do, he argued in [30], is to measure _change_ within one system against _change_ in a second system which has been defined as the standard (eg it takes half of one complete rotation of the earth about its own axis to walk thirty miles). Whilst Mach was clear about the origins of inertia (in the fixed stars), he did not hypothesize any mechanism by which this conviction might be realized and it fell to others to make the attempt - a typical (although incomplete) list might include the names of Einstein [31], Sciama [32], Hoyle & Narlikar and Sachs [33, 34] for approaches based on canonical ideas of spacetime, and the names of Ghosh [35] and Assis [36] for approaches based on quasi-Newtonian ideas. It is perhaps one of the great ironies of 20thC science that Einstein, having coined the name _Mach’s Principle_ for Mach’s original suggestion and setting out to find a theory which satisfied the newly named Principle, should end up with a theory which, whilst albeit enormously successful, is more an heir to the ideas of Democritus and Newton than to the ideas of Aristotle and Leibniz. One only has to consider the special case solution of Minkowski spacetime, which is empty but metrical, to appreciate this fact. ## 3 From Leibniz to inertia as a relational property In this paper we take the general position of Leibniz about the relational nature of space to be self-evident and considered the question of _spatial metric_ within this general conceptualization - that is, how is the notion of invariant _spatial_ distance to be defined in the Leibnizian particle universe? To answer this question, we begin by considering the universe of our actual experience and show how it is possible to define an invariant measure for the radius of a statistically defined astrophysical sphere purely in terms of the mass it contains (to within a calibration exercise); we then show how the arguments deployed can be extended to define an invariant measure for an _arbitrary_ spatial displacement within the statistically defined astrophysical sphere. In this way, we arrive at a theory within which a metrical three-space (generally non-Euclidean) is projected as a secondary construct out of the primary distribution of universal material. The question of how _time_ arises within this formalism is particularly interesting: the simple requirement that _time_ should be defined in such a way that Newton’s Third Law is automatically satisfied has the direct consequence that _time_ becomes an explicit measure of change within the system, very much as anticipated by Mach. The overall result is a quasi- classical (one-clock) theory of relational gravitation within which: * • conditions of global dynamical equilibrium (that is, everywhere inertial) hold in the limiting case of a particular parameter tending to zero in which case the primary material is distributed fractally, $D=2$; * • point-source perturbations of the limiting case $D=2$ distribution recover the usual Newtonian prescriptions for gravitational effects. The first of these two points refers to the universe that Leibniz was effectively considering in his debate with Clarke of 1715$\sim$1716 \- one within which inertial effects play no part. The second refers to the universe that Clarke used to refute Liebniz’s second argument, and the one that Mach had in mind - the universe of rotations and accelerations. Thus, given our Leibnizian worldview, we see that inertial effects themselves have their fundamental source in changed material relationships - they, too, are _relational_ in nature. ### 3.1 The general argument Following in the tradition of Aristotle, Leibniz, Berkeley and Mach we argue that no consistent cosmology should admit the possibility of an internally consistent discussion of _empty_ metrical space & time - unlike, for example, General Relativity which has the empty spacetime of Minkowski as a particular solution. Recognizing that the most simple space & time to visualize is one which is everywhere inertial, then our worldview is distilled into the elemental question: > _Is it possible to conceive a globally inertial space & time which is > irreducibly associated with a non-trivial global mass distribution and, if > so, what are the properties of this distribution?_ In pursuit of this question, we shall assume an idealized universe: * • _which consists of an infinity of identical,_ _but labelled_ _, discrete ’galaxies’ which possess an ordering property which allows us to say that galaxy $\mathcal{G}_{0}$ is nearer/further than galaxy $\mathcal{G}_{1}$. The redshift properties of galaxies in the real universe are an example of such an ordering property;_ * • _within which ’time’ is to be understood, in a qualitative way, as a measure of_ _process_ _or_ _ordered change_ _in the model universe;_ * • _within which there is at least one origin about which the distribution of material particles is statistically isotropic - meaning that the results of sampling along arbitrary lines of sight over sufficiently long characteristic ’times’ are independent of the directions of lines of sight._ #### 3.1.1 Astrophysical spheres and a mass-calibrated metric for radial displacements It is useful to discuss, briefly, the notion of spherical volumes defined on large astrophysical scales in the universe of our experience: whilst we can certainly give various precise operational definitions of spherical volumes on small scales, the process of giving such definitions on large scales is decidedly ambiguous. In effect, we have to suppose that redshift measurements are (statistically) isotropic when taken from an arbitrary point within the universe and that they vary monotonically with distance on the large scales we are concerned with. With these assumptions, spherical volumes can be defined (statistically) in terms of redshift measurements - however, their radial calibration in terms of ordinary units (such as metres) becomes increasingly uncertain (and even unknown) on very large redshift scales. With these ideas in mind, the primary step taken in answer to the elemental question above is the recognition that, on large enough scales in the universe of our experience (say $>30\,Mpc$), the amount of matter in a given redshift- defined spherical volume in a given epoch can be considered as a well-defined (monotonic) function of the sphere’s (redshift) radius. > _It follows immediately that a redshift calibration of the radius of an > astrophysical sphere at any given epoch has an equivalent mass-defined > calibration._ That is, the redshift-radius, $R_{z}$ say, of any spherical volume at any given time (whatever we might mean by _time_), calibrated in terms of mass, can be considered given by: $R_{z}=f(m,t)\rightarrow\delta R_{z}\approx f(m+\delta m,t)-f(m,t)$ (1) where $m$ is the mass concerned and $f$ is an arbitrary monotonic increasing function of $m$. Thus, for any given $f$, we have immediately defined an invariant radial measurement such that it becomes undefined in the absence of matter - in effect, we have, in principle, a metric which follows Leibniz in the required sense for any displacement which is _purely radial._ In the following, for the sake of simplicity only so that the basic ideas are most easily revealed, we make the assumption that all epochs _are identical_ \- that is, that the universe is in a steady-state \- so that the $t$-dependence in (1) can be dropped.Thus, in place of (1) we have $R_{z}=f(m)\rightarrow\delta R_{z}\approx f(m+\delta m)-f(m)$ (2) It transpires that this step has a significant consequence: * • the theory being developed becomes automatically a one-clock quasi-classical theory. If we do not assume that all epochs are identical, we obtain a two- clock theory of the _frame-time/particle proper time_ type. At this point, we can note that there will be some monotonic relation between a redshift-defined radial displacement $R_{z}$ and a ‘conventional’ radial measure, $R$ say. Thus, the first of (2) above can be written as $R_{z}\equiv g(R)=f(m)\rightarrow R=G(m)$ (3) for undefined monotonic increasing functions, $g$ and $G$. #### 3.1.2 A mass-calibrated metric for arbitrary spatial displacements The foregoing provides a way of giving an invariant measure, defined in terms of mass, for displacements which are purely _radial_ from the chosen origin _._ However, if, for example, a displacement is _transverse_ to a radial vector, then the methodology fails. Thus, we must look for ways of generalizing the above ideas so that we can assign a mass-calibrated metric to _arbitrary_ spatial displacements within the universe - that is, we must look for a way of assigning a _metric_ to this universe. To this end, we consider the second of (3) to be primary and then invert it to give a mass model, $Mass\equiv m=M(R)\equiv M(x^{1},x^{2},x^{3}),$ for our rudimentary universe. Note that we make no assumptions about the relation of the spatial coordinates, $(x^{1},x^{2},x^{3})$, to the redshift- defined radial displacement, $R$. Now consider the normal gradient vector $n_{a}=\nabla_{a}M$ (which does not require any metric stucture for its definition): the change in this arising from a displacement $dx^{k}$ can be _formally_ expressed as $dn_{a}=\nabla_{i}\left(\nabla_{a}M\right)\,dx^{i}\,,$ (4) where we assume that the geometrical connections required to give this latter expression an unambiguous meaning are the usual metrical connections - _except of course, the metric tensor $g_{ab}$ of our curvilinear three-space is not yet defined._ Now, since $g_{ab}$ is not yet defined, then the covariant counterpart of $dx^{a}$, given by $dx_{a}=g_{ai}dx^{i}$, is also not yet defined. However, we note that since $\nabla_{a}\nabla_{b}M$ is nonsingular, then (4) provides a 1:1 mapping between the contravariant vector $dx^{a}$ and the covariant vector $dn_{a}$ so that, in the absence of any other definition, we can _define_ $dn_{a}$ to be the covariant form of $dx^{a}$. In this latter case the metric tensor of our curvilinear three-space automatically becomes $g_{ab}=\nabla_{a}\nabla_{b}M$ (5) which, through the implied metrical connections, is a highly non-linear equation defining $g_{ab}$ to within the specification of $M$. The scalar product $dS^{2}\equiv dn_{i}dx^{i}\equiv g_{ij}dx^{i}dx^{j}$ then provides an invariant measure for the magnitude of the infinitesimal three-space displacement, $dx^{a}$. The units of $dS^{2}$ are easily seen to be those of $mass$ only and so, in order to make them those of $length^{2}$ \- as dimensional consistency requires - we define the working invariant as $ds^{2}\equiv(2r_{0}^{2}/m_{0})dS^{2}$, where $r_{0}$ and $m_{0}$ are scaling constants for the distance and mass scales respectively and the numerical factor has been introduced for later convenience. Finally, if we want $ds^{2}\equiv\left(\frac{r_{0}^{2}}{2m_{0}}\right)dn_{i}dx^{i}\equiv\left(\frac{r_{0}^{2}}{2m_{0}}\right)g_{ij}dx^{i}dx^{j}$ (6) to behave sensibly in the sense that $ds^{2}>0$ whenever $\left|d\textbf{r}\right|>0$ and $ds^{2}=0$ only when $\left|d\textbf{r}\right|=0$, then we must replace the condition of non- singularity of $g_{ab}$ by the condition that it is strictly positive definite; in the physical context of the present problem, this will be considered to be a self-evident requirement. ### 3.2 Analogous interpretation of this definition of $g_{ab}$ It is useful to have an analogous interpretation of the foregoing definition of $g_{ab}$ (and hence invariant distance) based upon how we, as primitive human beings, form intuitive assessments of “distance traversed” in our everyday lives without recourse to formal instruments. In effect, as we travel through a physical environment, we use our changing perspective of the observed scene in a given elapsed time to provide a qualitative assessment of “distance traversed” in that elapsed time. So, briefly, when walking across a tree-dotted landscape, the changing _angular_ relationships between ouselves and the trees provides the information required to assess “distance traversed”, measured in units of human-to-tree displacements, within that landscape. If we remove the perspective information - by, for example, obliterating the scene with dense fog - then all sense of “distance traversed” is destroyed. In the above definition, the part of the tree-dotted landscape is played by the mass-function, $M$, whilst the instantaneous “perspective” on this “landscape” is quantified by the normal vector $n_{a}$ and the change in perspective arising from a coordinate displacement, $dx^{a}$, is quantified by the change in this normal vector, $dn_{a}$. The invariant measure defined at (6) can then be considered to be based on a comparison between $dx^{a}$ (for which an invariant magnitude is required) and $dn_{a}$. ### 3.3 The connection coefficients We have assumed that the geometrical connection coefficients can be defined in some sensible way. To do this, we simply note that, in order to define conservation laws (ie to do physics) in a Riemannian space, it is necessary to be have a generalized form of Gausses’ divergence theorem in the space. This is certainly possible when the connections are defined to be the metrical connections, but it is by no means clear that it is ever possible otherwise. Consequently, the connections are assumed to be metrical and so $g_{ab}$, given at (5), can be written explicitly as $g_{ab}\equiv\nabla_{a}\nabla_{b}M\equiv\frac{\partial^{2}M}{\partial x^{a}\partial x^{b}}-\Gamma_{ab}^{k}\frac{\partial M}{\partial x^{k}},$ (7) where $\Gamma_{ab}^{k}$ are the Christoffel symbols, and given by $\Gamma_{ab}^{k}\leavevmode\nobreak\ =\leavevmode\nobreak\ \frac{1}{2}g^{kj}\left(\frac{\partial g_{bj}}{\partial x^{a}}+\frac{\partial g_{ja}}{\partial x^{b}}-\frac{\partial g_{ab}}{\partial x^{j}}\right).$ ## 4 The metric tensor given in terms of the mass model We have so far made no assumptions about the nature of the coordinate system $\left(x^{1},x^{2},x^{3}\right)$: so, we suppose that each of $\left(x^{1},x^{2},x^{3}\right)$ is calibrated in the same units as $R$ and that we have the usual Pythagorean relationship, $R^{2}=\left(x^{1}\right)^{2}+\left(x^{2}\right)^{2}+\left(x^{3}\right)^{2}$. With this understanding, it is shown, in appendix A, how, for an arbitrarily defined mass model, $M(R)$, (7) can be exactly resolved to give an explicit form for $g_{ab}$ in terms of such a general $M(R)$: defining the notation $\mathbf{R}\equiv(x^{1},x^{2},x^{3}),\leavevmode\nobreak\ \leavevmode\nobreak\ \Phi\equiv\frac{1}{2}\left(\mathbf{R\cdot R}\right)=\frac{1}{2}R^{2}\leavevmode\nobreak\ \leavevmode\nobreak\ {\rm and}\leavevmode\nobreak\ \leavevmode\nobreak\ M^{\prime}\equiv\frac{dM}{d\Phi},$ this explicit form of $g_{ab}$ is given as $g_{ab}=A\delta_{ab}+Bx^{i}x^{j}\delta_{ia}\delta_{jb},$ (8) where $A\equiv\frac{d_{0}M+m_{1}}{\Phi},\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ B\equiv-\frac{A}{2\Phi}+\frac{d_{0}M^{\prime}M^{\prime}}{2A\Phi},$ for arbitrary constants $d_{0}$ and $m_{1}$ where, as inspection of the structure of these expressions for $A$ and $B$ shows, $d_{0}$ is a dimensionless scaling factor and $m_{1}$ has dimensions of mass. Noting now that $M$ always occurs in the form $d_{0}M+m_{1}$, it is convenient to write $\mathcal{M}\equiv d_{0}M+m_{1}$, and to write $A$ and $B$ as $A\equiv\frac{\mathcal{M}}{\Phi},\leavevmode\nobreak\ \leavevmode\nobreak\ B\equiv-\left(\frac{\mathcal{M}}{2\Phi^{2}}-\frac{\mathcal{M}^{\prime}\mathcal{M}^{\prime}}{2d_{0}\mathcal{M}}\right).$ (9) ## 5 A direct physical calibration of the radial scale So far, we have assumed an arbitrary calibration for the radial scale; that is, we have assumed only that $R=f(M)$ where $f$ is an arbitrary monotonic increasing function. In the following, we show that the _geodesic radial scale_ provides a unique definition for $f$. ### 5.1 The geodesic radial scale Using (8) and (9) in (6), and using the identity $x^{i}dx^{j}\delta_{ij}\equiv RdR$ then, with the notation $\Phi\equiv R^{2}/2$, we find, for an arbitrary displacement $d{\bf x}$, the invariant measure: $ds^{2}\equiv g_{ij}dx^{i}dx^{j}=\left(\frac{R_{0}^{2}}{2m_{0}}\right)\left\\{\frac{\mathcal{M}}{\Phi}dx^{i}dx^{j}\delta_{ij}-\Phi\left(\frac{\mathcal{M}}{\Phi^{2}}-\frac{\mathcal{M}^{\prime}\mathcal{M}^{\prime}}{d_{0}\mathcal{M}}\right)dR^{2}\right\\},$ which is valid for the arbitrary calibration $R=f(M)$. If the displacement $d{\bf x}$ is now constrained to be purely radial, then we find $ds^{2}=\left(\frac{R_{0}^{2}}{2m_{0}}\right)\left\\{\Phi\left(\frac{\mathcal{M}^{\prime}\mathcal{M}^{\prime}}{d_{0}\mathcal{M}}\right)dR^{2}\right\\}.$ Use of $\mathcal{M}^{\prime}\equiv d\mathcal{M}/d\Phi$ and $\Phi\equiv R^{2}/2$ reduces this latter relationship to $\displaystyle ds^{2}$ $\displaystyle=$ $\displaystyle\frac{R_{0}^{2}}{d_{0}m_{0}}\left(d\sqrt{\mathcal{M}}\right)^{2}\leavevmode\nobreak\ \leavevmode\nobreak\ \rightarrow\leavevmode\nobreak\ \leavevmode\nobreak\ ds=\frac{R_{0}}{\sqrt{d_{0}m_{0}}}d\sqrt{\mathcal{M}}\leavevmode\nobreak\ \leavevmode\nobreak\ \rightarrow$ $\displaystyle s$ $\displaystyle=$ $\displaystyle\frac{R_{0}}{\sqrt{d_{0}m_{0}}}\left(\sqrt{\mathcal{M}}-\sqrt{\mathcal{M}_{0}}\right),\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ {\rm where}\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ \mathcal{M}_{0}\equiv\mathcal{M}(s=0)$ which defines the invariant magnitude of an arbitrary radial displacement from the origin purely in terms of the mass-model representation $\mathcal{M}\equiv d_{0}M+m_{1}$. By definition, this $s$ the invariant (or _physical_) measure associated with an finite radial displacement, $R$, for which an arbitrary calibration has so far been assumed. We can now remove the arbitrariness associated with this calibration by enforcing the identity $R\equiv s$. To summarize, with the latter identification, the natural physical calibration for the radial scale is given by $R=\frac{R_{0}}{\sqrt{d_{0}m_{0}}}\left(\sqrt{\mathcal{M}}-\sqrt{\mathcal{M}_{0}}\right),$ (10) where $\mathcal{M}_{0}$ is the value of $\mathcal{M}$ at $R=0$. ### 5.2 The Euclidean metric and a fractal $D=2$ matter distribution Remembering $\mathcal{M}\equiv d_{0}M+m_{1}$ (see §4) and noting that $M(R=0)=0$ necessarily, then $\mathcal{M}_{0}=m_{1}$ and so (10) can be equivalently arranged as $\mathcal{M}=\left[\frac{\sqrt{d_{0}m_{0}}}{R_{0}}R+\sqrt{m_{1}}\right]^{2}.$ (11) Using $\mathcal{M}\equiv d_{0}M+m_{1}$ again, then the mass-distribution function can be expressed in terms of the invariant radial displacement as $M=m_{0}\left(\frac{R}{R_{0}}\right)^{2}+2\sqrt{\frac{m_{0}m_{1}}{d_{0}}}\left(\frac{R}{R_{0}}\right)$ (12) which, for the particular case $m_{1}=0$ becomes $M\,=\,m_{0}(R/R_{0})^{2}$. Reference to (8) shows that, with this mass distribution and $d_{0}=1$, then $g_{ab}=\delta_{ab}$ so that the three-space space becomes ordinary Euclidean space. Thus, whilst we have yet to show that a globally inertial space can be associated with a non-trivial global matter distribution (since no temporal dimension, and hence no dynamics has been introduced), we have shown that a globally Euclidean space can be associated with a non-trivial matter distribution, and that this distribution is necessarily fractal with $D\,=\,2$. Note also that, on a large enough scale and for arbitrary values of $m_{1}$, (12) shows that radial distance varies as the square-root of mass from the chosen origin - or, equivalently, the mass varies as $R^{2}$. Consequently, on sufficiently large scales Euclidean space is irreducibly related to a quasi- fractal, $D\,=\,2$, matter distributions. Since $M/R^{2}\,\approx\,m_{0}/R_{0}^{2}$ on a large enough scale then, for the remainder of this paper, the notation $g_{0}\equiv m_{0}/R_{0}^{2}$ is employed. ## 6 The temporal dimension So far, the concept of ‘time’ has only entered the discussion in a qualitative way in §2 \- it has not entered in any quantitative way and, until it does, there can be no discussion of dynamical processes. Since, in its most general definition, time is a parameter which orders change within a system, then a necessary pre-requisite for its quantitative definition is a notion of change within the universe. The most simple notion of change which can be defined in the universe is that of changing relative spatial displacements of the objects within it. Since our model universe is populated solely by primitive galaxies which possess only the property of enumerability (and hence quantification in terms of the amount of material present) then, in effect, all change is gravitational change. This fact is incorporated into the cosmology to be derived by constraining all particle displacements to satisfy the Weak Equivalence Principle. We are then led to a Lagrangian description of particle motions in which the Lagrange density is degree zero in its temporal-ordering parameter. From this, it follows that the corresponding Euler-Lagrange equations form an incomplete set. The origin of this problem traces back to the fact that, because the Lagrangian density is degree zero in the temporal ordering parameter, it is then invariant with respect to any transformation of this parameter which preserves the ordering. This implies that, in general, temporal ordering parameters cannot be identified directly with physical time \- they merely share one essential characteristic. This situation is identical to that encountered in the Lagrangian formulation of General Relativity; there, the situation is resolved by defining the concept of ‘particle proper time’. In the present case, this is not an option because the notion of particle proper time involves the prior definition of a system of observer’s clocks - so that some notion of clock-time is factored into the prior assumptions upon which General Relativity is built. In the present case, it turns out that the isotropies already imposed on the system conspire to provide an automatic resolution of the problem which is consistent with the already assumed interpretation of ‘time’ as a measure of ordered change in the model universe. To be specific, it turns out that the elapsed time associated with any given particle displacement is proportional, via a scalar field, to the invariant spatial measure attached to that displacement. Thus, physical time is defined directly in terms of the invariant measures of process within the model universe and, furthermore, local conditions affect clock rates. ## 7 Dynamical constraints in the model universe Firstly, and as already noted, we are assuming that our notional universe is populated exclusively by primitive objects which possess solely the property of enumeration, and hence quantification. Consequently, all motions in this universe are effectively gravitational, and we model this circumstance by constraining all such motions to satisfy the Weak Equivalence Principle by which we mean that the trajectory of a body is independent of its internal constitution. This constraint can be expressed as: > _C1_ _Particle trajectories are independent of the specific mass values of > the particles concerned;_ Secondly, given the isotropy conditions imposed on the model universe from the chosen origin, symmetry arguments lead to the conclusion that the net action of the whole universe of particles acting on any given single particle is such that any net acceleration of the particle must always appear to be directed through the coordinate origin. Note that this conclusion is independent of any notions of retarded or instantaneous action. This constraint can then be stated as: > _C2_ _Any acceleration of any given material particle must necessarily be > along the line connecting the particular particle to the coordinate origin._ ## 8 Gravitational trajectories Suppose $p$ and $q$ are two arbitrarily chosen point coordinates on the trajectory of the chosen particle, and suppose that (6) is integrated between these points to give the scalar invariant $I(p,q)=\int_{p}^{q}\left(\frac{1}{\sqrt{2g_{0}}}\right)\sqrt{dn_{i}dx^{i}}\equiv\int_{p}^{q}\left(\frac{1}{\sqrt{2g_{0}}}\right)\sqrt{g_{ij}dx^{i}dx^{j}}.$ (13) Then, in accordance with the foregoing interpretation, $I(p,q)$ gives a scalar record of how the particle has moved between $p$ and $q$ defined with respect to the particle’s continually changing relationship with the mass model, $M(R)$. Now suppose $I(p,q)$ is minimized with respect to choice of the trajectory connecting $p$ and $q$, then this minimizing trajectory can be interpreted as a geodesic in the Riemannian space which has $g_{ab}$ as its metric tensor. Given that $g_{ab}$ is defined in terms of the mass model $M(R)$ \- the existence of which is independent of any notion of ‘inertial mass’, then the existence of the metric space, and of geodesic curves within it, is likewise explicitly independent of any concept of inertial-mass. It follows that the identification of the particle trajectory $\mathbf{R}$ with these geodesics means that particle trajectories are similarly independent of any concept of inertial mass, and can be considered as the modelling step defining that general subclass of trajectories which conform to that characteristic phenomenology of gravitation defined by condition C1 of §7. ## 9 The equations of motion Whilst the mass distribution, represented by ${\cal M}$, has been explicitly determined in terms of the geodesic distance at (11), it is convenient to develop the theory in terms of unspecified ${\cal M}$. The geodesic equations in the space with the metric tensor (8) can be obtained, in the usual way, by defining the Lagrangian density ${\cal L}\equiv\left(\frac{1}{\sqrt{2g_{0}}}\right)\sqrt{g_{ij}\dot{x}^{i}\dot{x}^{j}}=\left(\frac{1}{\sqrt{2g_{0}}}\right)\left(A\left(\mathbf{\dot{R}\cdot\dot{R}}\right)+B\dot{\Phi}^{2}\right)^{1/2},$ (14) where $\dot{x}^{i}\equiv dx^{i}/dt$, etc., and writing down the Euler-Lagrange equations $\displaystyle 2A\mathbf{\ddot{R}}$ $\displaystyle+$ $\displaystyle\left(2A^{\prime}\dot{\Phi}-2\frac{\dot{{\cal L}}}{{\cal L}}A\right)\mathbf{\dot{R}}+\left(B^{\prime}\dot{\Phi}^{2}+2B\ddot{\Phi}-A^{\prime}\left(\mathbf{\dot{R}\cdot\dot{R}}\right)-2\frac{\dot{{\cal L}}}{{\cal L}}B\dot{\Phi}\right)\mathbf{R}$ (15) $\displaystyle=$ $\displaystyle 0,$ where $\mathbf{\dot{R}}\equiv d\mathbf{R}/dt$ and $A^{\prime}\equiv dA/d\Phi$, etc. By identifying particle trajectories with geodesic curves, this equation is now interpreted as the equation of motion, referred to the chosen origin, of a single particle satisfying condition C1 of §7. However, noting that the variational principle, (13), is of order zero in its temporal ordering parameter, we can conclude that the principle is invariant with respect to arbitrary transformations of this parameter; in turn, this means that the temporal ordering parameter cannot be identified with physical time. This problem manifests itself formally in the statement that the equations of motion (15) do not form a complete set, so that it becomes necessary to specify some extra condition to close the system. A similar circumstance arises in General Relativity when the equations of motion are derived from an action integral which is formally identical to (13). In that case, the system is closed by specifying the arbitrary time parameter to be the ‘proper time’, so that $d\tau={\cal L}(x^{j},dx^{j})\leavevmode\nobreak\ \leavevmode\nobreak\ \rightarrow\leavevmode\nobreak\ \leavevmode\nobreak\ {\cal L}(x^{j},\frac{dx^{j}}{d\tau})=1,$ (16) which is then considered as the necessary extra condition required to close the system. In the present circumstance, we are rescued by the, as yet, unused condition C2. ## 10 Physical time ### 10.1 Completion of equations of motion Consider C2, which states that any particle accelerations must necessarily be directed through the coordinate origin. This latter condition simply means that the equations of motion must have the general structure $\mathbf{\ddot{R}}=G(t,\mathbf{R},\mathbf{\dot{R}})\mathbf{R},$ for scalar function $G(t,\mathbf{R},\mathbf{\dot{R}})$. In other words, (15) satisfies condition C2 if the coefficient of $\mathbf{\dot{R}}$ is zero, so that $\left(2A^{\prime}\dot{\Phi}-2\frac{\dot{{\cal L}}}{{\cal L}}A\right)=0\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ \rightarrow\frac{A^{\prime}}{A}\dot{\Phi}=\frac{\dot{{\cal L}}}{{\cal L}}\leavevmode\nobreak\ \leavevmode\nobreak\ \rightarrow\leavevmode\nobreak\ \leavevmode\nobreak\ {\cal L}=k_{0}A,$ (17) for arbitrary constant $k_{0}$ which is necessarily positive since $A>0$ and ${\cal L}>0$. The condition (17), which guarantees (C2), can be considered as the condition required to close the incomplete set (15) and is directly analogous to (16), the condition which defines ‘proper time’ in General Relativity. ### 10.2 Physical time defined as process Equation (17) can be considered as that equation which removes the pre- existing arbitrariness in the ‘time’ parameter by defining physical time:- from (17) and (14) we have $\displaystyle{\cal L}^{2}$ $\displaystyle=$ $\displaystyle k_{0}^{2}A^{2}\leavevmode\nobreak\ \rightarrow\leavevmode\nobreak\ A\left(\mathbf{\dot{R}\cdot\dot{R}}\right)+B\dot{\Phi}^{2}=2g_{0}k_{0}^{2}A^{2}\leavevmode\nobreak\ \rightarrow\leavevmode\nobreak\ $ $\displaystyle g_{ij}\dot{x}^{i}\dot{x}^{j}$ $\displaystyle=$ $\displaystyle 2g_{0}k_{0}^{2}A^{2}$ (18) so that, in explicit terms, physical time is defined by the relation $dt^{2}=\left(\frac{1}{2g_{0}k_{0}^{2}A^{2}}\right)g_{ij}dx^{i}dx^{j},\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ {\rm where}\leavevmode\nobreak\ \leavevmode\nobreak\ A\equiv{\frac{{\cal M}}{\Phi}}.$ (19) In short, the elapsing of time is given a direct physical interpretation in terms of the process of displacement in the model universe. Finally, noting that, by (19), the dimensions of $k_{0}^{2}$ are those of $L^{6}/[T^{2}\times M^{2}]$, then the fact that $g_{0}\equiv m_{0}/R_{0}^{2}$ (cf §5) suggests the change of notation $k_{0}^{2}\propto v_{0}^{2}/g_{0}^{2}$, where $v_{0}$ is a constant having the dimensions (but not the interpretation) of ‘velocity’. So, as a means of making the dimensions which appear in the development more transparent, it is found convenient to use the particular replacement $k_{0}^{2}\equiv v_{0}^{2}/(4d_{0}^{2}g_{0}^{2})$. With this replacement, the definition of physical time, given at (19), becomes $dt^{2}=\left(\frac{4d_{0}^{2}g_{0}}{v_{0}^{2}A^{2}}\right)g_{ij}dx^{i}dx^{j}.$ (20) Since, as is easily seen from the definition of $g_{ab}$ given in §4, $g_{ij}dx^{i}dx^{j}$ is necessarily finite and non-zero for a non-trivial displacement $d\mathbf{R}$ ### 10.3 The necessity of $v_{0}^{2}\,\neq\,0$ Equation (20) provides a definition of physical time in terms of basic process (displacement) in the model universe. Since the parameter $v_{0}^{2}$ occurs nowhere else, except in its explicit position in (20), then it is clear that setting $v_{0}^{2}=0$ is equivalent to physical time becoming undefined. Therefore, of necessity, $v_{0}^{2}\neq 0$ and all non-zero finite displacements are associated with a non-zero finite elapsed physical time. ## 11 The cosmological potential The model is most conveniently interpreted when expressed in potential terms and so, in the following, it is shown how this is done. ### 11.1 The equations of motion: potential form From §10, when (17) is used in (15) there results $2A\mathbf{\ddot{R}}+\left(B^{\prime}\dot{\Phi}^{2}+2B\ddot{\Phi}-A^{\prime}\left(\mathbf{\dot{R}\cdot\dot{R}}\right)-2\frac{A^{\prime}}{A}B\dot{\Phi}^{2}\right)\mathbf{R}=0.$ (21) Suppose we define a function $V$ according to $V\equiv C_{0}-\left(\mathbf{\dot{R}\cdot\dot{R}}\right)/2$, for some arbitrary constant $C_{0}$; then, by (18) $V\equiv C_{0}-\frac{1}{2}\left(\mathbf{\dot{R}\cdot\dot{R}}\right)=C_{0}-\frac{v_{0}^{2}}{4d_{0}^{2}g_{0}}A+\frac{B}{2A}\dot{\Phi}^{2},$ (22) where $A$ and $B$ are defined at (9). With unit vector, $\mathbf{\hat{R}}$, then appendix B shows how this function can be used to express (21) in the potential form $\mathbf{\ddot{R}}=-\frac{dV}{dr}\mathbf{\hat{R}}$ (23) so that $V$ is a potential function, and $C_{0}$ is the arbitrary constant usually associated with a potential function. ### 11.2 The potential function, $V$, as a function of $R$ From (22), we have $2C_{0}\,-\,2V\,=\,\dot{R}^{2}+R^{2}\dot{\theta}^{2}=\frac{v_{0}^{2}}{2d_{0}^{2}g_{0}}A-\frac{B}{A}R^{2}\dot{R}^{2}$ so that $V$ is effectively given in terms of $R$ and $\dot{R}$. In order to clarify things further, we now eliminate the explicit appearance of $\dot{R}$. Since all forces are central, then angular momentum is conserved; consequently, after using conserved angular momentum, $h$, and the definitions of $A$, $B$ and ${\cal M}$ given in §4, the foregoing equations can be written as $\displaystyle 2C_{0}$ $\displaystyle-$ $\displaystyle 2V\leavevmode\nobreak\ \leavevmode\nobreak\ \,=\,$ $\displaystyle\dot{R}^{2}$ $\displaystyle+$ $\displaystyle R^{2}\dot{\theta}^{2}\,=\,v_{0}^{2}+\frac{4v_{0}^{2}}{R}\sqrt{\frac{m_{1}}{d_{0}g_{0}}}+\frac{d_{0}-1}{R^{2}}\left(\frac{6m_{1}v_{0}^{2}}{d_{0}^{2}\,g_{0}}-h^{2}\right)$ (24) $\displaystyle+$ $\displaystyle\frac{2}{R^{3}}\sqrt{\frac{d_{0}m_{1}}{g_{0}}}\left(\frac{2m_{1}v_{0}^{2}}{d_{0}^{2}\,g_{0}}-h^{2}\right)+\frac{1}{R^{4}}\frac{m_{1}}{g_{0}}\left(\frac{m_{1}v_{0}^{2}}{d_{0}^{2}\,g_{0}}-h^{2}\right)$ so that $V(R)$ is effectively given by the right-hand side of (24). ## 12 A discussion of the potential function It is clear from (24) that $m_{1}$ plays the role of the mass of the central source which generates the potential, $V$. A relatively detailed description of the behaviour of $V$ is given in appendix C, where we find that there are two distinct classes of solution depending on the free parameters of the system. These classes can be described as: * • A constant potential universe within which all points are dynamically indistinguishable; this corresponds to an inertial material universe, and arises in the case $m_{1}=0,\,\,d_{0}=1$; * • All other possibilities give rise to a ‘distinguished origin’ universe in which either: * – there is a singularity at the centre, $R=0$; * – or there is no singularity at $R=0$ and, instead, the origin is the centre of a non-trivial sphere of radius $R_{min}>0$ which acts as an impervious boundary between the exterior universe and the potential source. In effect, this sphere provides the source with a non-trivial spatial extension so that the classical notion of the massive point-source is avoided. Of these possibilities, the constant potential universe is the one which provides positive answers to our originally posed questions, and it is this which is discussed in detail in the following sections. However, of the two cases in the distinguished origin universe, the no- singularity case offers the interesting possibility of being able to model the gravitational effects created by a central massive source, but without the non-physical singularity at the origin. This case is mentioned here for future reference. ## 13 The fractal $D\,$=$\,2$ inertial universe Reference to (24) shows that the parameter choice $m_{1}=0$ and $d_{0}=1$ makes the potential function constant everywhere, whilst (12) shows how, for this case, universal matter in an equilibrium universe is necessarily distributed as an exact fractal with $D=2$. Thus, the fractal $D=2$ material universe is necessarily a globally inertial equilibrium universe, and the questions originally posed in §3.1 are finally answered. ### 13.1 Implications for theories of gravitation Given that gravitational phenomena are usually considered to arise as mass- driven perturbations of flat inertial backgrounds, then the foregoing result - to the effect that the inertial background is necessarily associated with a non-trivial fractal matter distribution - must necessarily give rise to completely new perspectives about the nature and properties of gravitational phenomena. However, as we show in §13.2, the kinematics in this inertial universe is unusual, and suggests that the inertial material distribution is more properly interpreted as a quasi-photon fractal gas out of which (presumably) we can consider ordinary material to condense in some fashion. ### 13.2 The quasi-photon fractal gas For the case $m_{1}=0$, $d_{0}=1$, the definition $M$ at (12) together with the definitions of $A$ and $B$ in §4 give $A\,=\,\frac{2m_{0}}{R_{0}^{2}},\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ B\,=\,0$ so that, by (22) (remembering that $g_{0}\equiv m_{0}/R_{0}^{2}$) we have $\left(\mathbf{\dot{R}\cdot\dot{R}}\right)\,=\,v_{0}^{2}$ (25) for all displacements in the model universe. It is (almost) natural to assume that the constant $v_{0}^{2}$ in (25) simply refers to the constant velocity of any given particle, and likewise to assume that this can differ between particles. However, each of these assumptions would be wrong since - as we now show - $v_{0}^{2}$ is, firstly, more properly interpreted as a conversion factor from spatial to temporal units and, secondly, is a global constant which applies equally to all particles. To understand these points, we begin by noting that (25) is a special case of (18) and so, by (19), is more accurately written as $dt^{2}\,=\,\frac{1}{v_{0}^{2}}\left(d\mathbf{R\cdot}d\mathbf{R}\right)$ (26) which, by the considerations of §10.2, we recognize as the definition of the elapsed time experienced by any particle undergoing a spatial displacement $d\mathbf{R}$ in the model inertial universe. Since this universe is isotropic about all points, then there is nothing which can distinguish between two separated particles (other than their separateness) undergoing displacements of equal magnitudes; consequently, each must be considered to have experienced equal elapsed times. It follows from this that $v_{0}$ is not to be considered as a locally defined particle speed, but is a globally defined constant which has the effect of converting between spatial and temporal units of measurement - that is, it is more accurately thought of as being a conversion factor from length scales to time scales - very much like Bondi’s interpretation of $c$, the light speed. Thus, one can think of the matter distribution in the equilibrium universe as similar to a _photon gas_ with the Bondian complexity that this ‘gas’ arbitrates between length scales and time scales - within the context of the theory, this is a direct consequence of the fact that _time_ arises automatically as a measure of _change_ within the particle distribution very much as Mach conceived it. ## 14 Summary and Conclusions The idea that universal material might be distributed fractally, with fractal dimension 2, is a difficult one to comprehend since it appears contrary to all common-sensical notions of how the Universe “should be”. As an example of how- deeply entrenched such ideas are, we need only cite the hidden assumptions of homogeneity which underly statistical tools such as two-point correlation function methods, widely used until very recently to determine the nature of material distributions in the Universe. However, this paper argues that the problem goes very much deeper than that, and has as its source the notion of “space” as an objective reality which functions as a container of material - a notion which can be traced from Democritus, through Newton and even to Einstein (in practice). The opposing notion, associated with the names of Aristotle, Leibniz, Berkeley and Mach - and the one developed here - is that, in effect, what we call “space” is a secondary construct projected out of relationships between material objects, which are primary. This paper has shown how these ideas can be given quantitative expression and leads directly to the conclusion that the relationship between the secondary construct of “space” and the primary objects which _appear_ to be contained within “space” is given by $M=m_{0}\left(\frac{R}{R_{0}}\right)^{2}+2\sqrt{\frac{m_{0}m_{1}}{d_{0}}}\left(\frac{R}{R_{0}}\right),$ where $m_{0},m_{1}$ and $d_{0}$ are parameters. In other words, on sufficiently large scales, the relationship between “space” and the material which it “appears” to contain can be naturally interpreted as the statement that _space is irreducibly associated with a fractal, $D=2$, distribution of material within the Universe._ The discussion which led to the foregoing conclusion also unavoidably entailed a corresponding discussion concerning the nature of “time” \- as Mach himself pointed out, where “time” is concerned, the most we can ever do is to define the “time required for process $A$ to occur” in terms of the “time required for process $B$ to occur”. For example, “I can walk 100 miles between one sunrise and the next.” From this viewpoint, it is arguable, for example, that the individual proton - never observed to decay, and therefore a stranger to change -__ exists _out of time_. But the internal distribution of an assembly of (labelled) protons, on the other hand, does change and a sequence of snapshots of such a changing assembly could be considered as an evolutionary sequence _defining_ _of itself the passage of time._ In effect, “time” is defined as a metaphor for “ordered change within a physical system”; this is the definition of “time” which arises automatically from the considerations which lead to the fractal Universe above - there is a set of particles which possess only the property of _enumerability_ from which the concept of “time” arises as a metaphor for “ordered change”. Finally, on sufficiently large scales and according to the foregoing analysis, space & time is necessarily inertial (to any required degree of approximation), and this has fundamental consequences for gravitation theory: specifically, since gravitational phenomena are conventionally considered to arise as mass-driven perturbations of a flat inertial background, then the phenomenology predicted by the analysis - that a flat inertial background is irreducibly associated with a non-trivial fractal distribution of material - must necessarily lead to novel insights into the nature and causes of gravitational phenomena. ## Appendix A A Resolution of the Metric Tensor The general system is given by $g_{ab}=\frac{\partial^{2}M}{\partial x^{a}\partial x^{b}}-\Gamma_{ab}^{k}\frac{\partial M}{\partial x^{k}},$ $\Gamma_{ab}^{k}\leavevmode\nobreak\ \equiv\leavevmode\nobreak\ \frac{1}{2}g^{kj}\left(\frac{\partial g_{bj}}{\partial x^{a}}+\frac{\partial g_{ja}}{\partial x^{b}}-\frac{\partial g_{ab}}{\partial x^{j}}\right),$ and the first major problem is to express $g_{ab}$ in terms of the reference scalar, $M$. The key to this is to note the relationship $\frac{\partial^{2}M}{\partial x^{a}\partial x^{b}}=M^{\prime}\delta_{ab}+M^{\prime\prime}x^{a}x^{b},$ where $M^{\prime}\equiv dM/d\Phi$, $M^{\prime\prime}\equiv d^{2}M/d\Phi^{2}$, $\Phi\equiv R^{2}/2$ and $R^{2}\equiv\left(x^{1}\right)^{2}+\left(x^{2}\right)^{2}+\left(x^{3}\right)^{2}$, since this immediately suggests the general structure $g_{ab}=A\delta_{ab}+Bx^{a}x^{b},$ (27) for unknown functions, $A$ and $B$. It is easily found that $g^{ab}={\frac{1}{A}}\left[\delta_{ab}-\left({\frac{B}{A+2B\Phi}}\right)x^{a}x^{b}\right]$ so that, with some effort, $\Gamma_{ab}^{k}={\frac{1}{2A}}H_{1}-\left({\frac{B}{2A(A+2B\Phi)}}\right)H_{2}$ where $\displaystyle H_{1}$ $\displaystyle=$ $\displaystyle A^{\prime}(x^{a}\delta_{bk}+x^{b}\delta_{ak}-x^{k}\delta_{ab})$ $\displaystyle+$ $\displaystyle B^{\prime}x^{a}x^{b}x^{k}+2B\delta_{ab}x^{k}$ and $\displaystyle H_{2}$ $\displaystyle=$ $\displaystyle A^{\prime}(2x^{a}x^{b}x^{k}-2\Phi x^{k}\delta_{ab})$ $\displaystyle+$ $\displaystyle 2\Phi B^{\prime}x^{a}x^{b}x^{k}+4\Phi Bx^{k}\delta_{ab}.$ Consequently, $\displaystyle g_{ab}$ $\displaystyle=$ $\displaystyle{\frac{\partial^{2}M}{\partial x^{a}\partial x^{b}}}-\Gamma_{ab}^{k}{\frac{\partial M}{\partial x^{k}}}\equiv\delta_{ab}M^{\prime}\left({\frac{A+A^{\prime}\Phi}{A+2B\Phi}}\right)$ $\displaystyle+$ $\displaystyle x^{a}x^{b}\left(M^{\prime\prime}-M^{\prime}\left({\frac{A^{\prime}+B^{\prime}\Phi}{A+2B\Phi}}\right)\right).$ Comparison with (27) now leads directly to $\displaystyle A$ $\displaystyle=$ $\displaystyle M^{\prime}\left({\frac{A+A^{\prime}\Phi}{A+2B\Phi}}\right)=M^{\prime}\left(\frac{(A\Phi)^{\prime}}{A+2B\Phi}\right),$ $\displaystyle B$ $\displaystyle=$ $\displaystyle M^{\prime\prime}-M^{\prime}\left({\frac{A^{\prime}+B^{\prime}\Phi}{A+2B\Phi}}\right).$ The first of these can be rearranged as $B={\frac{M^{\prime}}{2\Phi}}\left(\frac{(A\Phi)^{\prime}}{A}\right)-{\frac{A}{2\Phi}}$ or as $\left(\frac{M^{\prime}}{A+2B\Phi}\right)={\frac{A}{(A\Phi)^{\prime}}},$ and these expressions can be used to eliminate $B$ in the second equation. After some minor rearrangement, the resulting equation is easily integrated to give, finally, $A\equiv{\frac{d_{0}M+m_{1}}{\Phi}},\leavevmode\nobreak\ \leavevmode\nobreak\ \leavevmode\nobreak\ B\equiv-{\frac{A}{2\Phi}}+{\frac{d_{0}M^{\prime}M^{\prime}}{2A\Phi}}.$ ## Appendix B Conservative Form of Equations of Motion From (22), we have $V\equiv-\frac{1}{2}\left(\mathbf{\dot{R}\cdot\dot{R}}\right)=-\frac{k_{0}^{2}A}{2}+\frac{B}{2A}\dot{\Phi}^{2},$ (28) from which we easily find $\frac{dV}{dR}\equiv\frac{\partial V}{\partial R}+\frac{\partial V}{\partial\dot{R}}\frac{\ddot{R}}{\dot{R}}$ $=\frac{-k_{0}^{2}A^{\prime}}{2}R+\frac{\dot{\Phi}^{2}R}{2A}\left(B^{\prime}-\frac{A^{\prime}B}{A}\right)+\frac{B}{A}\left(R\dot{R}^{2}+R^{2}\ddot{R}\right).$ Since $\dot{R}^{2}+R\ddot{R}=\ddot{\Phi}$, then the above expression leads to $\frac{dV}{dR}\mathbf{\hat{R}}=\left(\frac{-k_{0}^{2}A^{\prime}}{2}+\frac{B^{\prime}}{2A}\dot{\Phi}^{2}-\frac{A^{\prime}B}{2A^{2}}\dot{\Phi}^{2}+\frac{B}{A}\ddot{\Phi}\right)\mathbf{R}.$ Writing (23) as $2A\mathbf{\ddot{R}}+2A\frac{dV}{dR}\mathbf{\hat{R}}=0,$ and using the above expression, we get the equation of motion as $2A\mathbf{\ddot{R}}+\left(-k_{0}^{2}AA^{\prime}+B^{\prime}\dot{\Phi}^{2}-\frac{A^{\prime}B}{A}\dot{\Phi}^{2}+2B\ddot{\Phi}\right)\mathbf{R}=0.$ (29) Finally, from (28), we have $k_{0}^{2}A=\frac{B}{A}\dot{\Phi}^{2}+\left(\mathbf{\dot{R}\cdot\dot{R}}\right),$ which, when substituted into (29), gives (21). ## Appendix C Outline analysis of the potential function It is quite plain from (24) that, for any $m_{1}\neq 0$, then the model universe has a preferred centre and that the parameter $m_{1}$ (which has dimensions of mass) plays a role in the potential $V$ which is analogous to the source mass in a Newtonian spherical potential - that is, the parameter $m_{1}$ can be identified as the mass of the potential source in the model universe. However, setting $m_{1}=0$ is not sufficient to guarantee a constant potential field since any $d_{0}\neq 1$ also provides the model universe with a preferred centre. The role of $d_{0}$ is most simply discussed in the limiting case of $m_{1}=0$: in this case, the second equation of (24) becomes $\dot{R}^{2}+R^{2}\dot{\theta}^{2}\,=\,v_{0}^{2}-(d_{0}-1)\frac{h^{2}}{R^{2}}.$ (30) If $d_{0}<1$ then $|\mathbf{\dot{R}}|\rightarrow\infty$ as $R\rightarrow 0$ so that a singularity exists. Conversely, remembering that $v_{0}^{2}>0$ (cf §10.3) then, if $d_{0}>1$, equation (30) restricts real events to the exterior of the sphere defined by $R^{2}=(d_{0}-1)h^{2}/v_{0}^{2}$. In this case, the singularity is avoided and the central ‘massless particle’ is given the physical property of ‘finite extension’. In the more realistic case for which $m_{1}>0$, reference to (24) shows that the $R=0$ singularity is completely avoided whenever $h^{2}>m_{1}v_{0}^{2}/d_{0}^{2}g_{0}$ since then a ‘finite extension’ property for the central massive particle always exists. Conversely, a singularity will necessarily exist whenever $h^{2}\leq m_{1}v_{0}^{2}/d_{0}^{2}g_{0}$. In other words, the model universe has a preferred centre when either $m_{1}>0$, in which case the source of the potential is a massive central particle having various properties depending on the value of $d_{0}$, or when $m_{1}=0$ and $d_{1}\neq 0$. ## References * [1] Charlier, C.V.L., 1908, Astronomi och Fysik 4, 1 (Stockholm) * [2] Charlier, C.V.L., 1922, Ark. Mat. Astron. Physik 16, 1 (Stockholm) * [3] Charlier, C.V.L., 1924, Proc. Astron. Soc. Pac. 37, 177 * [4] De Vaucouleurs, G., 1970, Sci 167, 1203 * [5] Peebles, P.J.E., 1980, The Large Scale Structure of the Universe, Princeton University Press, Princeton, NJ. * [6] Baryshev, Yu V., Sylos Labini, F., Montuori, M., Pietronero, L. 1995 Vistas in Astronomy 38, 419 * [7] Huchra, J., Davis, M., Latham, D.,Tonry, J., 1983, ApJS 52, 89 * [8] Giovanelli, R., Haynes, M.P., Chincarini, G.L., 1986, ApJ 300, 77 * [9] De Lapparent, V., Geller,M.J., Huchra, J.P., 1988, ApJ 332, 44 * [10] Broadhurst, T.J., Ellis, R.S., Koo, D.C., Szalay, A.S., 1990, Nat 343, 726 * [11] Da Costa, L.N., Geller, M.J., Pellegrini, P.S., Latham, D.W., Fairall, A.P., Marzke, R.O., Willmer, C.N.A., Huchra, J.P., Calderon, J.H., Ramella, M., Kurtz, M.J., 1994, ApJ 424, L1 * [12] Vettolani, G., et al., in: Proc. of Schloss Rindberg Workshop: Studying the Universe With Clusters of Galaxies * [13] Joyce, M., Montuori, M., Sylos Labini, F., 1999, Astrophys. J. 514, L5 * [14] Sylos Labini, F., Montuori, M., 1998, Astron. & Astrophys., 331, 809 * [15] Sylos Labini, F., Montuori, M., Pietronero, L., 1998, Phys. Lett., 293, 62 * [16] Wu, K.K.S., Lahav, O., Rees, M.J., 1999, Nature 397, 225 * [17] Scaramella, R., Guzzo, L., Zamorani, G., Zucca, E., Balkowski, C., Blanchard, A., Cappi, A., Cayatte, V., Chincarini, G., Collins, C., Fiorani, A., Maccagni, D., MacGillivray, H., Maurogordato, S., Merighi, R., Mignoli, M., Proust, D., Ramella, M., Stirpe, G.M., Vettolani, G. 1998 A&A 334, 404 * [18] Martinez, V.J., PonsBorderia, M.J., Moyeed, R.A., Graham, M.J. 1998 MNRAS 298, 1212 * [19] Labini, F.S., Gabrielli, A., 2000, _Scaling and fluctuations in galaxy distribution: two tests to probe large scale structures_ , astro-ph0008047 * [20] Gabrielli, A., Sylos Labini, F., 2001, Europhys. Lett. 54 (3), 286 * [21] Pietronero, L., Sylos Labini, F., 2000, Physica A, (280), 125 * [22] Hogg, D.W., Eistenstein, D.J., Blanton M.R., Bahcall N.A, Brinkmann, J., Gunn J.E., Schneider D.P. 2005 ApJ, 624, 54 * [23] Sylos Labini, F., Vasilyev, N.L., Baryshev, Y.V., Archiv.Astro.ph/0610938 * [24] Newton, I., _Principia: Mathematical Principles of Natural Philosophy,_ translated by I.B. Cohen & A. Whitman, University of California Press, 1999 * [25] Aristotle, _Categories,_ in R. McKeon, _The Basic Works of Aristotle,_ Random House, New York, 1941. * [26] Aristotle, _Physics,_ translated by P.H. Wicksteed & F.M. Cornford, Leob Classical Library, Harvard University Press, Cambridge, 1929 * [27] Leibniz, G.W., _Philosophical Essays_ , Edited and translated by R. Ariew & D.Garber, Hackett Publishing Co, Indianapolis, 1989 * [28] Alexander, H.G., _The Leibniz-Clarke Correspondence,_ Manchester University Press, 1984. * [29] Berkeley, G., _Of motion, or the principle and nature of motion and the cause of the communication of motions (_ Latin: _De Motu)_ in M.R.Ayers (ed) _George Berkeley’s Philosophical Works,_ Everyman, London, 1992. * [30] Mach, E., _The Science of Mechanics_ \- _a Critical and Historical Account of its Development_ Open Court, La Salle, 1960 * [31] Einstein, A., _The Foundations of the General Theory of Relativity,_ in A. Einstein, H.A.Lorentz, H.Weyl & H.Minkowski, _The Principle of Relativity,_ Dover, New York, 1952 * [32] Sciama, D.W., 1953, _On the Origin of Inertia,_ Mon. Not. Roy. Astron. Soc, 113, 34 * [33] Sachs, M., _General Relativity and Matter,_ Reidal, Dordrecht, 1982 * [34] Sachs, M., _Quantum Mechanics from General Relativity,_ Reidal, Dordrecht 1986 * [35] Ghosh, A., _Origin of Inertia,_ Apeiron, Montreal, 2000 * [36] Assis, A.K.T., _Relational Mechanics,_ Apeiron, Montreal, 1999

arxiv-papers

2024-09-04T02:48:53.897680

0802.3030

# A MOLECULAR LINE OBSERVATION TOWARD MASSIVE CLUMPS ASSOCIATED WITH INFRARED DARK CLOUDS Takeshi Sakai11affiliation: Nobeyama Radio Observatory, Minamimaki, Minamisaku, Nagano 384-1305, Japan. , Nami Sakai22affiliation: Department of Physics, Graduate School of Science, The University of Tokyo, Tokyo 113-0033, Japan. , Kazuhisa Kamegai33affiliation: Institute of Astronomy, The University of Tokyo, Osawa, Mitaka, Tokyo 181-0015, Japan , Tomoya Hirota44affiliation: National Astronomical Observatory of Japan, Osawa, Mitaka, Tokyo 181-8588, Japan. , Nobuyuki Yamaguchi11affiliation: Nobeyama Radio Observatory, Minamimaki, Minamisaku, Nagano 384-1305, Japan. , Shoichi Shiba22affiliation: Department of Physics, Graduate School of Science, The University of Tokyo, Tokyo 113-0033, Japan. and Satoshi Yamamoto22affiliation: Department of Physics, Graduate School of Science, The University of Tokyo, Tokyo 113-0033, Japan. ###### Abstract We have surveyed the N2H+ $J$=1–0, HC3N $J$=5–4, CCS $J_{N}$=$4_{3}$–$3_{2}$, NH3 ($J$, $K$) = (1, 1), (2, 2), (3, 3), and CH3OH $J$=7–6 lines toward the 55 massive clumps associated with infrared dark clouds by using the Nobeyama Radio Observatory 45 m telescope and the Atacama Submillimeter Telescope Experiment 10 m telescope. The N2H+, HC3N, and NH3 lines are detected toward most of the objects. On the other hand, the CCS emission is detected toward none of the objects. The [CCS]/[N2H+] ratios are found to be mostly lower than unity even in the Spitzer 24 $\mu$m dark objects. This suggests that most of the massive clumps are chemically more evolved than the low-mass starless cores. The CH3OH emission is detected toward 18 out of 55 objects. All the CH3OH-detected objects are associated with the Spitzer 24 $\mu$m sources, suggesting that star formation has already started in all the CH3OH-detected objects. The velocity widths of the CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ and $7_{-1}$–$6_{-1}$ $E$ lines are broader than those of N2H+ $J$=1–0. The CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ and $7_{-1}$–$6_{-1}$ $E$ lines tend to have broader linewidth in the MSX dark objects than in the others, the former being younger or less luminous than the latter. The origin of the broad emission is discussed in terms of the interaction between an outflow and an ambient cloud. ISM: clouds — ISM: molecule — star: formation ## 1 Introduction Infrared dark clouds (IRDCs) are extinction features against the background mid-IR emission (e.g. Simon et al. 2006), which were recognized by the observations with the mid-infrared satellites such as ISO and MSX (Pérault et al. 1996; Egan et al. 1998). Mapping observations of the mm and sub-mm continuum emissions (Carey et al. 1998; Beuther et al. 2002; Rathborne et al. 2006) revealed that massive clumps are associated with several IRDCs. Although such massive clumps have a similar mass to the clumps with ongoing high-mass star formation, they are less active. The gas kinetic temperature of IRDCs is as low as 10–20 K (Carey et al. 1998, Teyssier et al. 2002, Pillai et al. 2006b). Therefore, massive clumps in IRDCs are thought to be starless or at least in a very early stage of high-mass star formation. Recently, evidences of high-mass star formation were found in several massive clumps associated with IRDCs. Rathborne et al. (2005) found luminous sources (9,000$\sim$32,000 $L_{\odot}$) in massive clumps of IRDC G034.43+00.24, which are recognized as high-mass protostars ($\sim$ 10 $M_{\odot}$). Rathborne et al. (2007) also reported that a high-mass protostar ($>$ 8 $M_{\odot}$) is associated with a massive clump in IRDC G024.33+00.11 on the basis of their high angular resolution mm-wave observation. Beuther et al. (2005b) and Beuther & Steinacker (2007) found a protostar in IRDC 18223-3 using the Spitzer and the Plateau de Bure data, which would grow up to a high-mass star at the end of its formation process. In addition to these, indirect evidences of high-mass star formation, such as the CH3OH masers, are found in some IRDCs (Ellingsen 2006; Purcell et al. 2006). As mentioned above, IRDCs are good targets for studies on high-mass star formation. But little attention has been paid for their chemical compositions, although it would be useful to understand the evolutionary stages of IRDCs. Since binary chemical reactions, such as neutral-neutral and ion-neutral reactions, are responsible for chemical processes in the deep inside of molecular clumps, the chemical composition changes slowly toward chemical equilibrium with a time scale of $\sim 10^{6}$ yr. This time scale is comparable with the dynamical time scale of molecular cloud clumps. Therefore, chemical composition would reflect evolutionary stages of clumps. For example, carbon-chain molecules, such as CCS and HC3N, are abundant in the early stage of chemical evolution, while N2H+ and NH3 are abundant in the late stage (e.g. Lee et al. 1996). Furthermore, carbon-chain molecules suffer the great effect from depletion, whereas N2H+ and NH3 are less affected by depletion. When CO, a major destroyer of N2H+, is depleted, N2H+ can be abundant in dense parts. Since NH3 is produced from N2, NH3 can also be abundant (Aikawa et al. 2001, 2005; Bergin & Tafalla 2007). Thus, we can investigate the evolutionary stages of clumps from the ratios between carbon-chain molecules and N2H+ or NH3. This method has been established in low-mass star forming regions. Suzuki et al. (1992) and Benson et al. (1998) surveyed the carbon-chain molecule toward several low-mass cores, and they found that carbon-chain molecules are more abundant in starless cores than star-forming cores, whereas NH3 or N2H+ is more abundant in star-forming cores. In addition, Hirota et al. (2002, 2004) and Hirota & Yamamoto (2006) found that carbon-chain molecules are very abundant in several low-mass starless cores. Recently, Sakai et al. (2007) suggested that this method can also be applied to high-mass star forming regions. Sakai et al. (2006, 2007) observed the W 3 giant molecular cloud (GMC) with the CCS $J_{N}$=$4_{3}$–$3_{2}$ and N2H+ $J$=1–0 lines, and found that the [CCS]/[N2H+] ratio is higher toward a few massive starless cores in the AFGL 333 cloud than toward the other massive star forming cores in the W 3 GMC. Sakai et al. (2007) proposed that the massive starless cores in the AFGL 333 cloud are novel candidates which will form high- or intermediate-mass stars, because the CCS emission has hardly been detected in high-mass star- forming regions (Lai & Crutcher, 2000). Note that definitions of ”core” and ”clump” are described in Sakai et al. (2007). Chemical composition of clumps would be changed by various activities of newly born stars. Saturated organic molecules like CH3OH become abundant due to evaporation of grain mantles. The high excitation lines of CH3OH are therefore employed as a tracer of hot cores and shocked regions (e.g. Bachiller & Pérez Gutiérrez 1997). In this paper, we report a systematic survey of the N2H+ $J$=1–0, CCS $J_{N}$=$4_{3}$–$3_{2}$, HC3N $J$=5–4, CH3OH $J$=7–6 lines, as well as the NH3 ($J$, $K$)=(1, 1), (2, 2) and (3, 3) lines toward massive clumps associated with IRDCs. Although observations of several molecular lines were carried out toward massive clumps in the IRDCs (Pillai et al. 2006b, 2007; Leurini et al. 2007; Purcell et al. 2006; Ragan et al. 2006; Beuther & Sridharan 2007), a systematic survey of molecular lines focusing on chemical composition has been limited. In this paper, we present chemical compositions and physical properties of fifty-five nearby massive clumps associated with the IRDCs, and investigate their evolutionary stages. ## 2 Observations ### 2.1 The Sample Sridharan et al. (2005) identified the candidates of massive starless clumps from the objects found with MAMBO by Beuther et al. (2002). Their candidates involve clumps associated with IRDCs. Rathborne et al. (2006) also observed 38 IRDCs with known kinematic distance by using MAMBO, and listed the positions of clumps. We prepared a sample of massive clumps associated with IRDCs by use of these two lists on the basis of the following criteria: * 1\. Distance is less than 4.5 kpc. * 2\. Mass is larger than 100 $M_{\odot}$. * 3\. Galactic longitude is less than 34.5 degree. In addition to these, we also involved the massive clumps located near the clumps listed by Sridharan et al. (2005). Their masses are calculated from the data by Beuther et al. (2002), where the revised equation by Beuther et al. (2005) is employed. Finally, the sample consists of 55 objects, as listed in Table A MOLECULAR LINE OBSERVATION TOWARD MASSIVE CLUMPS ASSOCIATED WITH INFRARED DARK CLOUDS. In the course of this study, 15 objects are eventually found to have the different LSR velocity from those listed by Rathborne et al. (2006). As the result, such objects do not satisfy the above criteria. We investigated whether the observed objects are associated with point-like sources or extinction features in the MSX 8 $\mu$m data. Then, 12 objects are found to be associated with the MSX 8 $\mu$m sources, and 26 objects to be associated with extinction features. We call an object with an MSX 8 $\mu$m extinction feature as an MSX dark object. In the archival Spitzer 24 $\mu$m data, we also inspected whether the observed objects are associated with point-like sources or extinction features of 24 $\mu$m. As a result, the 31 objects are found to be associated with the point-like 24 $\mu$m sources. Note that there are no available Spitzer data for I18151-1208. There are 11 dark objects in the Spitzer 24 $\mu$m data, which are all MSX 8 $\mu$m dark objects. These results are summarized in Table A MOLECULAR LINE OBSERVATION TOWARD MASSIVE CLUMPS ASSOCIATED WITH INFRARED DARK CLOUDS. ### 2.2 NRO 45 m observations The N2H+ $J$=1–0, CCS $J_{N}$=$4_{3}$–$3_{2}$, HC3N $J$=5–4, NH3 ($J$, $K$) = (1, 1), (2, 2) and (3, 3) lines (Table A MOLECULAR LINE OBSERVATION TOWARD MASSIVE CLUMPS ASSOCIATED WITH INFRARED DARK CLOUDS) were observed by using the Nobeyama Radio Observatory (NRO) 45 m telescope from January to March in 2007. We used the two SIS receivers, S100 for N2H+ and S40 for CCS and HC3N. The N2H+, CCS, and HC3N lines were simultaneously observed. The half-power beam width is about $37^{\prime\prime}$ at 45 GHz and about $18^{\prime\prime}$ at 93 GHz, and the main beam efficiency is 0.75 at 45 GHz and 0.53 at 93 GHz. The NH3 (1,1), (2,2) and (3, 3) lines were observed with the HEMT receiver. The half-power beam width is about $73^{\prime\prime}$ at 22 GHz, and the main beam efficiency is 0.81 at 22 GHz. For all the observations, acousto-optical radiospectrometers were employed as a backend, each of which has bandwidth and frequency resolution of 40 MHz and 37 kHz, respectively. The velocity resolutions are 0.50, 0.25, and 0.12 km s-1 at 22 GHz, 45 GHz, and 93 GHz, respectively. The telescope pointing was checked by observing the nearby SiO maser sources every 2 hours, and was maintained to be better than $5^{\prime\prime}$. The line intensities were calibrated by the chopper wheel method. The system noise temperature was 200–300 K for the N2H+, CCS, and HC3N observations, whereas it was 100–150 K for the NH3 observation. All the observations were carried out with the position switching mode. The emission free regions in the Galactic Ring Survey 13CO $J$=1–0 data (Jackson et al. 2006) were employed as the OFF positions. The data reduction was carried out by use of the AIPS-based software package NewStar developed at NRO. ### 2.3 ASTE observations The CH3OH $J$=7–6 lines (Table A MOLECULAR LINE OBSERVATION TOWARD MASSIVE CLUMPS ASSOCIATED WITH INFRARED DARK CLOUDS) were observed with the Atacama Submillimeter Telescope Experiment (ASTE) 10 m telescope (Ezawa et al. 2004; Kohno 2005) from September to October in 2006. We also observed the CH3OH $J_{K}$=$1_{1}$–$0_{0}$ $A^{+}$ and $J_{K}$=$4_{0}$–$3_{-1}$ $E$ lines simultaneously. The beam size of the ASTE telescope is about 22′′ at 345 GHz. We used the SIS receiver, SC345, operated in the double-side-band mode. The backends were autocorrelators, whose band width and resolution each are 512 MHz and 0.5 MHz, respectively. By using three autocorrelators, we covered the frequency range from 337.9 to 338.8 GHz. The telescope pointing was calibrated by observing the thermal continuum emission from the Jupiter. The pointing observation was carried out every observing day, and the pointing accuracy was maintained within 3′′. All the observations were performed with the position switching mode. Antenna temperature was calibrated by the single load chopper-wheel method. The intensity scale was checked by observing the CH3OH lines toward the NGC 6334 hot core every 0.5–1 hr in order to correct a small variation of the telescope gain due to the temperature and elevation change. As a result, the relative error in intensity calibration can be maintained to be less than 9 %. The main beam efficiency of the ASTE telescope is estimated to be 0.60, although this includes an uncertainty of 20 %. The data reduction was carried out by use of the NewStar. ## 3 Results ### 3.1 Detection Rates & Spectra Table 3 shows the 3$\sigma$ detection rates of the observed lines. The N2H+ $J$=1–0 emission is detected toward 54 out of 55 objects, the HC3N $J$=5–4 emission toward 43 objects, and the NH3 ($J$, $K$) = (1, 1) emission toward all the 55 objects. On the other hand, the CCS $J_{N}$=$4_{3}$–$3_{2}$ emission is not detected at all. This survey indicates that the CCS emission is hardly detected in IRDCs, as in the case of high-mass star-forming regions. This is not a distance effect. If the AFGL 333 clump were located on the distance of 4.5 kpc, the peak temperature would be 0.29 K. This value is higher than the upper limits to the CCS peak intensity of all the observed objects, except for I18306-0835 MM1, MM2, and MM3. Thus, we can detect the CCS line, if a chemically young massive clump like the AFGL 333 clump would exist within the distance of 4.5 kpc. The CH3OH $J$=7–6 emissions are detected toward 18 objects. Although the upper state energy of the CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ line is as high as 65 K, the line is detected even in the objects without luminous heating sources seen in the MSX 8 $\mu$m image (MSX dark objects). In contrast, the Spitzer 24 $\mu$m sources are associated with all the CH3OH-detected objects except for I18151-1208 MM2, for which there are no available Spitzer data. Thus, it is likely that all the objects with the CH3OH $J$=$7_{0}$–$6_{0}$ $A^{+}$ emission harbor hot regions. Figure 1a shows the spectra of N2H+ $J$=1–0, HC3N $J$=5–4, CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$, and CCS $J_{N}$=$4_{3}$–$3_{2}$ for the selected objects, whereas Figure 2a shows those of NH3 ($J$, $K$) = (1, 1), (2, 2) and (3, 3). The spectra in Figures 1a and 2a are ”typical ones” taken arbitrarily from the spectra of our sample. The observed spectra toward all the objects are shown in Figures 1b–e and 2b–e, which are available in the electric edition of the Journal. Note that the observed lines toward G024.08+00.04 MM2 have two velocity components (52 km s-1 and 114 km s-1). The N2H+ $J$=1–0 line shows complicated hyperfine splittings due to the nuclear spin ($I$=1) of the nitrogen nuclei. The observed line shows triplet structure, where the center and higher velocity components consist of three hyperfine lines each unresolved due to the velocity width. The velocity width, optical depth, and excitation temperature of N2H+ are derived by fitting the observed spectrum pattern to a multi Gaussian function (see Appendix B in Sakai et al. 2006). The results are given in Table 4. We fit the spectra of the HC3N and CH3OH lines to a single Gaussian function, and these results are summarized in Tables 5 and 6, respectively. We represent the upper limits to peak temperature and integrated intensity for the non-detected lines in the tables. The upper limits to peak temperature for the non-detected lines are a 3$\sigma$ value at the velocity resolution of one channel width. The upper limits to integrated intensity for the non-detected lines are derived by using the following equation; $3T_{\rm rms}\sqrt{10({\rm kms^{-1}})\Delta V_{\rm res}}$, where $\Delta V_{\rm res}$ is a channel width in km s-1, and $T_{\rm rms}$ is a rms noise level at the velocity resolution of $\Delta V_{\rm res}$. As for the NH3 (1, 1) line, five hyperfine components are fitted to the single Gaussian each, and the parameters of the main hyperfine components are listed in Table 7. On the other hand, the NH3 (2, 2) and (3, 3) lines appear as a single line without hyperfine structure, and hence, they are fitted to a single Gaussian function each (Tables 8 and 9). Note that the NH3 (1, 1) and (2, 2) lines show a double-peak feature in G024.33+00.11 MM3 and MM9. Since the double peak structure is seen even in the weak hyperfine components, it seems to be a velocity structure. Then, we fit the spectra to a double Gaussian function. The parameters for each peak are listed in Tables 7 and 8. As mentioned earlier, we found that the LSR velocities of 15 objects including G024.33+00.11 are much different from the values in the lists of Rathborne et al. (2006). This is because they estimated the systemic velocity of the objects from the 13CO $J$=1–0 line. The critical density of N2H+ $J$=1–0 is much higher than that of 13CO $J$=1–0, and hence, the N2H+ $J$=1–0 emission well traces the dense clumps. We re-evaluate the kinematic distance of the dense clumps from the LSR velocity of the N2H+ or NH3 lines for all the observed objects by using the rotation curve obtained by Clemens (1985) with ($\Theta_{0}$, $R_{0}$) = (220 km s-1, 8.5 kpc). The derived distances are listed in Table 4. Figure 3 shows the spectra covering the CH3OH $J_{K}$=$7_{K}$–$6_{K}$ transitions toward the CH3OH-detected objects. Two strong lines are seen around 338.4 GHz; these are the $J_{K}$=$7_{-1}$–$6_{-1}$ $E$ and $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ lines. In addition, higher excitation lines ($7_{0}$–$6_{0}$ $E$ and $7_{\pm 2}$–$6_{\pm 2}$ $E$) are seen toward most of the sources. The CH3OH $J_{K}$=$1_{1}$–$0_{0}$ $A^{+}$ and $J_{K}$=$4_{0}$–$3_{-1}$ $E$ lines from the other side band are also observed. Although these two lines are lower-$J$ lines, the intensities of these lines are generally lower than those of the $J_{K}$=$7_{0}$–$6_{0}$ line because of their lower line strength. Therefore, the two lower-$J$ CH3OH lines are not detected toward the objects without CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ emission. The strongest CH3OH source is G034.43+00.24 MM1, where even the $J_{K}$=$7_{3}$–$6_{3}$ lines ($E_{u}$ = 112–127 K) are detected. The H2CS 101,10–91,9 line ($E_{u}$ = 102 K) is detected toward three objects, G024.33+00.11 MM1, G034.43+00.24 MM1, and I18182-1433 MM1, which seem to contain hot core activities. In fact, Rathborne et al. (2007) reported the detection of many molecular lines, such as HCOOCH3 and CH3CH2CN, toward G024.33+00.11 MM1. Since large organic molecules are thought to be formed from molecules evaporated from dust grains, such as C2H2, HCN, and CH3OH, in the hot gas phase (e.g. Nomura & Millar, 2004), the detections of HCOOCH3 and CH3CH2CN definitively indicate the hot core activity. ### 3.2 Integrated Intensities In Figure 4, we plot the integrated intensities of a few observed lines against the peak flux of the 1.2 mm continuum, $F_{1.2}$, reported by Beuther et al. (2002) and Rathborne et al. (2006). We plot only the values of the objects with $D$ $\leq$ 4.5 kpc. The mm-wave dust continuum emission is optically thin in general, and it is a good tracer of column density of a dense region. A correlation between the N2H+ $J$=1–0 emission and the dust continuum emission was suggested in several regions (Caselli et al. 1999, 2002a; Bergin et al. 2001; Sakai et al. 2007), although it was also reported that the N2H+ emission does not always trace the structures by dust continuum emission in very active star forming regions, such as Orion KL (Ungerechts et al. 1998). In IRDCs, the N2H+ emission is correlated with the 1.2 mm peak flux, as shown in Figure 4, confirming that the N2H+ emission well traces the column density of a dense regions. The HC3N integrated intensity is also roughly correlated with the 1.2 mm peak flux, suggesting that the HC3N emission also comes from a dense region, as well as the N2H+ emission. The CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ line is only detected toward the objects where $F_{1.2}$ is higher than 100 mJy. By using the relation reported by Beuther et al. (2005a), $F_{1.2}$ of 100 mJy corresponds to $A_{V}$ of 49, 68, and 114 at $T_{k}$ of 40, 30, and 20 K, respectively. Furthermore, all the objects with $F_{1.2}$ $>$ 130 mJy are the Spitzer sources. Hence, the appearance of the CH3OH $J$=7–6 emission is related to the star formation activity. A marginal correlation can be seen between the NH3 integrated intensities and the 1.2 mm peak flux. The NH3 integrated intensities are scattered even at the low $F_{1.2}$ values, probably because the NH3 emission traces less dense gas due to its lower critical density ($\sim$ 103 cm-3), and also because the beam size of the NH3 observation (73′′) is larger than those of the other observations (18′′–37′′). ## 4 Discussions ### 4.1 Temperatures We derive the NH3 rotation temperature, $T_{\rm rot}$(NH3), from the NH3 (1, 1) and (2, 2) lines by using the method presented by Ho & Townes (1983) and Li et al. (2003). We first evaluate the optical depth of the NH3 (1, 1) line from the peak intensity ratio of the main and satellite hyperfine components. Then we calculate the rotation temperature from the peak intensity ratio of the main hyperfine components of the (1, 1) and (2, 2) lines considering the optical depth of the (1, 1) line obtained above. The derived $T_{\rm rot}$(NH3) values are listed in Table 10. A range of $T_{\rm rot}$(NH3) is from 10.3 to 20.8 K with the median of 14–16 K. This is higher than the corresponding values of $T_{\rm rot}$(NH3) reported for dark clouds (10–15 K; Benson & Myers, 1989). Figure 5 represents a histogram of $T_{\rm rot}$(NH3) for the objects with $D$ $\leq$ 4.5 kpc. Figure 5 shows that $T_{\rm rot}$(NH3) for the MSX sources tends to be higher than that for the MSX dark objects. This confirms that the emitting regions are associated with the MSX sources. The $T_{\rm rot}$(NH3) values of the MSX and Spitzer dark objects tend to be lower than those of the other objects. The average $T_{\rm rot}$(NH3) value of the MSX and Spitzer dark objects is 13.9$\pm$1.5 K, which is comparable to that for the other samples of the MSX-dark IRDCs (13.9 K: Pillai et al. 2006b; 15.3 K: Sridharan et al. 2005), but is lower than that for the high-mass protostellar objects (22.5 K: Sridharan et al. 2002). The rotation temperatures of CH3OH, $T_{\rm rot}$(CH3OH), are derived by using the rotation diagram method (e.g. Turner 1991) under the assumption of local thermodynamic equilibrium (LTE). Since several E type lines are detected in many objects (Figure 3), we derive the rotation temperature from them. The results are listed in Table 11. The derived $T_{\rm rot}$(CH3OH) values range from 7 to 12 K. The range is lower than that for the high-mass star forming cores (24–203 K reported by van der Tak et al. (2000b), if we exclude one exception of GL 7009S (8 K)). The average value of $T_{\rm rot}$(CH3OH) in the MSX dark objects is 8.0$\pm$0.9 K, which seems to be slightly lower than that of the MSX sources (10$\pm$2 K). Although the upper state energy of the CH3OH $J$=$7_{-1}$–$6_{-1}$ $E$ line is 70 K, the derived temperatures are less than 12 K, which are even lower than $T_{\rm rot}$(NH3). This may suggest that the rotational levels of CH3OH are not thermalized to the gas kinetic temperature. Leurini et al. (2007) modeled the CH3OH emission toward several IRDCs, and they derived the gas kinetic temperatures for an extended component, an outflow, and a core to be 24$\pm$7, 15$\pm$5, and 47$\pm$9 K, respectively. The $T_{\rm rot}$(CH3OH) values derived in the present study are lower than their values, implying that the CH3OH lines are subthermally excited. ### 4.2 Column Densities We derive the column densities of the observed molecules. For simplicity, we assume the LTE condition for all the molecules. As for N2H+, we derive the column density by assuming that the excitation temperature is equal to the rotation temperature of NH3, and the lines are optically thin. This can be justified, because $\Delta V$(N2H+) is similar to $\Delta V$(NH3), as mentioned in the next section. The derived column densities are listed in Table 10. In the case of G034.43+00.24 MM1, a change in $T_{\rm ex}$ by $\pm$1 K results in a change in the derived column density by $\pm$4 %. Most of the N2H+ column densities are found to be larger than 1$\times$1013 cm-2. This is higher than a typical value found in low mass cores, which is several times 1012 cm-2 (Benson et al. 1998; Caselli et al. 2002b). The upper limit to the CCS column density is derived by assuming the optically thin condition. The rotation temperature of NH3 is employed as the excitation temperature. The derived upper limits are listed in Table 10. They are found to be lower than 7$\times$1012 cm-2 for all the objects. These are lower than the CCS column densities of some low-mass starless cores in the dark clouds, typically a few times 1013 cm-2 (Suzuki et al. 1992; Benson et al. 1998), and are also lower than the CCS column density of the massive clump in the AFGL 333 cloud, which is (1.2–3.0)$\times$1013 cm-2 (Sakai et al. 2007). Table 10 lists the upper limit to the [CCS]/[N2H+] ratio. Most of the ratios are lower than 1 even in the Spitzer dark objects, whereas the corresponding ratio is as high as 2.6–3.2 for some low-mass starless cores (Benson et al. 1998). Since we use the [CCS]/[N2H+] ratio to discuss the chemical evolution, the distance effect would be cancelled. It is therefore suggested that the massive clumps associated with the IRDCs are chemically evolved than low mass starless cores. This survey demonstrates that the AFGL 333 clump could be a rare case of a chemically young massive clump. The HC3N column density is evaluated by assuming the optically thin condition, where the excitation temperature is assumed to be equal to $T_{\rm rot}$(NH3). In the case of G034.43+00.24 MM1, a change in $T_{\rm ex}$ by $\pm$1 K results in a change in the derived column density by $\pm$3 %. In contrast to CCS, the derived HC3N column densities (Table 10) are comparable to those for the low- mass starless cores, $\sim$1013-14 cm-2(Suzuki et al. 1992), even in the objects with MSX sources, probably because HC3N is less sensitive to chemical evolution than CCS. In addition, HC3N could also be formed in a hot gas. In fact, strong HC3N emission is detected toward hot core sources like Orion KL (Ungerechts et al. 1998). The NH3 column density is derived from the optical depth and rotation temperature (Section 4.1). The derived column densities are listed in Table 10. The NH3 column densities (a few times 1015 cm-2) are higher than those of the low mass cores, $\sim$1014-15 cm-2 (Suzuki et al. 1992), and are comparable to those of the IRDCs observed by Pillai et al. (2006b). The $N$(CCS)/$N$(NH3) ratios are found to be lower than 0.007 for most cases, which are lower than those in some low-mass dark cloud cores ($>$0.01) (Suzuki et al. 1992). Along with the results of the $N$(CCS)/$N$(N2H+) ratio, this further supports that the observed IRDCs are chemically evolved. We derive the CH3OH column density from the integrated intensity of the CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ line. In this calculation, $T_{\rm rot}$(CH3OH) is not used as the excitation temperature, because the rotational levels of CH3OH are not thermalized as mentioned before. $T_{\rm rot}$(CH3OH) is an effective temperature determined by the population difference between the different $K$ ladders, and does not always represent the excitation temperature of the $7_{0}$–$6_{0}$ $A^{+}$ line. Therefore, we assume the excitation temperature ranging from 20 K to 50 K by referring the result by Leurini et al. (2007). We also estimate the upper limit to the CH3OH column density for the non-detected objects with adopting the rotation temperature of 20 K. The derived CH3OH column densities ranges from 2.9$\times$1014 to 4.9$\times$1015 cm-2 toward the CH3OH-detected objects, as shown in Table 10. They are comparable to those in the IRDCs obtained by Leurini et al. (2007), although they tend to be lower than those in several IRDCs obtained by Beuther & Sridharan (2007) (1.7$\times$1013–3.7$\times$1014 cm-2). In addition, they are also comparable to those in the massive star forming regions obtained by van der Tak et al. (2000b). On the other hand, Maret et al. (2005) reported that the CH3OH column densities ranges from 5.3$\times$1013 to 8.1$\times$1014 cm-2 in the low-mass star forming regions, NGC 1333, L1448, L1157, and IRAS16293-2422. Thus, the CH3OH column densities tend to be higher than those in the low-mass star forming regions. ### 4.3 Velocity Widths & Star Formation Velocity widths give information about turbulence of clumps including effects of star formation activity. In Figure 6, we plot the velocity widths of the HC3N $J$=5–4, CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$, NH3 ($J$, $K$) = (1, 1), (2, 2) and (3, 3) lines against that of the N2H+ $J$=1–0 line for the objects with $D$ $\leq$ 4.5 kpc. The velocity widths of the HC3N, NH3 (1, 1) and (2, 2) lines are correlated with that of N2H+, indicating that they come from almost the same region. On the other hand, the correlation is poor for high excitation lines such as CH3OH ($J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ and $7_{-1}$–$6_{-1}$ $E$) and NH3 (3, 3). Since the upper state energies of CH3OH $J$=$7_{K}$–$6_{K}$ (65 K) is higher than that of N2H+ $J$=1–0 (4 K) and that of NH3 (3, 3) is as high as 124.5 K, the CH3OH $J$=7–6 and NH3 (3, 3) lines would trace the denser and hotter region near the protostar. In fact, the velocity widths of the CH3OH $J$=7–6 and NH3 (3, 3) lines are correlated with each other (Figure 7). Pillai et al. (2006a) also found that the NH3 (3, 3) velocity width is larger than the NH3 (1, 1) and (2, 2) velocity widths toward the IRDC, G11.11-0.12. Beuther et al. (2002) reported a similar trend in the C34S and NH3 velocity widths, where the C34S $J$=2–1 velocity widths are about 1.5 times larger than the NH3 (1, 1) velocity widths. Since the C34S emission is optically thin and the critical density of C34S is as high as $3\times 10^{5}$ cm-3, it also traces dense regions. Furthermore, CS could be abundant in the hot gas phase in contrast to N2H+ (e.g. Nomura & Millar 2004). Thus it is likely that the C34S emission comes from regions close to the newly born stars, as well as the CH3OH and NH3 (3, 3) emissions. Figure 8 shows histograms of the velocity width of N2H+ for the MSX dark objects and the objects with MSX sources. The mean N2H+ velocity width of the MSX dark objects (2.3 km s-1) is almost comparable to that of the objects with MSX sources (2.5 km s-1). These values are also comparable to that of the high-mass star forming regions observed by Pirogov et al. (2003; 2.4 km s-1). The histogram shapes are also similar to each other. Since the N2H+ emission arises from the whole clump, it is not sensitive to the star formation activities. The mean N2H+ velocity widths of the MSX and Spitzer dark objects (2.1 km s-1) are significantly larger than those of the low-mass starless cores ($\sim$0.5 km s-1) (Benson et al. 1998), and are also larger than those of the cores in a giant molecular cloud, Orion A (Tatematsu et al. 1993). The massive clumps associated with the IRDCs seem to be initially turbulent. This gives an important constraint when one would construct a model of star- formation processes in IRDCs. For instance, McKee & Tan (2003) argued that massive stars can form in turbulence-supported cores with a high accretion rate. Thus, the large velocity widths observed in IRDCs are consistent with the theoretical prediction, if the IRDCs are birthplaces of high-mass stars. As shown in Figure 6, the CH3OH $J$=7–6 velocity widths for the MSX dark objects tend to be broader than those for the objects with MSX sources at a given N2H+ velocity width. This trend is clear in the case of G034.43+00.24. Figure 9 shows the Spitzer 24 $\mu$m image toward the G034.43+00.24 region and the spectra of the CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ and N2H+ $J$=1–0 $F_{1}$=0–1 $F$=1–2 lines toward G034.43+00.24 MM1, MM2 and MM3. In the Spitzer image, G034.43+00.24 MM3, which is a MSX dark object, is less luminous than MM1 and MM2. It is known that a ultracompact HII region is associated with MM2 (Miralles et al. 1994; Shepherd et al. 2007). Thus, MM2 is thought to be most evolved among three objects, whereas MM3 is thought to be youngest. In Figure 9, the CH3OH spectrum of G034.43+00.24 MM3 is clearly broader than the N2H+ spectrum for G034.43+00.24 MM3. On the other hand, the CH3OH velocity width is comparable to the N2H+ velocity width toward G034.43+00.24 MM1 and MM2. The CH3OH linewidths toward these two sources are relatively narrower than that toward MM3. As seen in Figure 9, the line shape of the CH3OH emission sometimes differs from the Gaussian shape. It may consist of the broad wing component and the narrow component. In the present analysis, we fit all the CH3OH spectra with a single Gaussian, so that the derived velocity widths have to be recognized as ”effective” values for some sources where the line shape is much deviated from the single Gaussian shape. We employ the single Gaussian fitting rather than the double Gaussian fitting, because the latter fitting is difficult for most of the objects due to the limited S/N ratio. Nevertheless, the ”effective” line width reflects the presence of the broad component, and hence, our discussion on the basis of the line width would be justified at least qualitatively. van der Tak et al. (2000a, 2000b) carried out multi-line observations of CH3OH, C34S, and C17O toward several massive star-forming regions, and found that the velocity widths of the CH3OH lines are broader than those of the C34S and C17O lines for the less luminous objects, whereas they are comparable in the luminous objects. van der Tak et al. (2000b) argued that shock by the interaction between an outflow and an ambient cloud is important as the origin of the gas-phase CH3OH enhancement in low-luminosity objects, while radiation heating is important in high-luminosity objects. In fact, the broad CH3OH emission are reported for such an interacting region in several low-mass star forming regions, including the famous case of L1157, where a clear interaction between an outflow and an ambient cloud is observed as broad lines of CH3OH, NH3, and HC3N (Umemoto et al. 1992; Mikami et al. 1992; Bachiller & Pérez Gutiérrez 1997; Umemoto et al. 1999). In contrast, the N2H+ line is not detected in the interaction region. Therefore, the CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ emission would well trace the shocked region caused by the interaction between an outflow and an ambient cloud in the case of the MSX dark objects. Furthermore, the broader line width can be seen in the observed lower-$J$ lines, $J_{K}$=$1_{1}$–$0_{0}$ $A^{+}$ and $4_{0}$–$4_{-1}$ $E$. This may imply that these CH3OH emissions do not come from the quiescent and extended regions. To understand the properties of CH3OH in the quiescent gas, we should observe the lower excitation transitions at the millimeter-wave region, since the critical density is proportional to $\nu^{3}$. Our result that the line width of the CH3OH line is broader in the MSX dark object means that the interaction between an outflow and an ambient cloud is important in the early stage of high-mass star formation. ## 5 Summary * • We have surveyed the N2H+ $J$=1–0, HC3N $J$=5–4, CCS $J_{N}$=$4_{3}$–$3_{2}$, CH3OH $J$=7–6, NH3 (J, K) = (1, 1), (2, 2), and (3, 3) lines toward the massive clumps associated with the IRDCs. The N2H+ $J$=1–0 emission is detected toward 54 out of 55 objects, the HC3N $J$=5–4 emission toward 43 objects, the NH3 ($J$, $K$) = (1, 1) emission toward all the 55 objects, the NH3 ($J$, $K$) = (2, 2) emission toward 52 objects, and the NH3 ($J$, $K$) = (3, 3) emission toward 34 objects. On the other hand, the CCS emission is not detected at all. The CH3OH $J$=7–6 emission is detected toward 18 out of 55 objects. All the CH3OH detected objects are associated with the Spitzer 24 $\mu$m sources. * • The mean NH3 rotation temperature of the MSX and Spitzer dark objects is 13.9 K, which tends to be lower than that of the objects with those of the MSX sources (17.9 K). * • The CCS column densities and the [CCS]/[N2H+] and [CCS]/[NH3] ratios in the massive clumps associated with the IRDCs are lower than those in the low-mass starless cores. Thus, most of the massive clumps would be more chemically evolved than the low-mass starless cores. This survey demonstrates that the massive clump in the AFGL 333 region, which was found by Sakai et al. (2006, 2007), could be very novel. * • The mean N2H+ velocity widths are significantly broader than those of the low- mass starless cores even in the MSX and Spitzer dark objects, indicating that massive cores associated with the IRDCs are initially turbulent. This along with the chemical ages inferred from the [CCS]/[N2H+] ratio would be useful to constrain a models of star formation processes in IRDCs. * • The velocity widths of the CH3OH $J$=7–6 and NH3 (3, 3) lines are broader than those of the other lines. This seems to originate from the interaction between an outflow and an ambient cloud. The CH3OH velocity widths of the MSX dark objects are larger than those of the objects with MSX sources. This suggests that outflow interaction plays an important role in production of the gas phase CH3OH in the early stage of high-mass star formation. We are grateful to the NRO staff for excellent support in the 45 m observations. We would like to thank the members of the ASTE team for supporting the ASTE observations. The 45 m radio telescope is operated by Nobeyama Radio Observatory, a branch of National Astronomical Observatory of Japan. The ASTE project is driven by Nobeyama Radio Observatory, a branch of National Astronomical Observatory of Japan, in collaboration with University of Chile, and Japanese institutes including University of Tokyo, Nagoya University, Osaka Prefecture University, Ibaraki University, and Kobe University. ## References * Aikawa et al. (2001) Aikawa, Y., Ohashi, N., Inutsuka, S., Herbst, E., Takakuwa, S. 2001, ApJ, 552, 639 * Aikawa et al. (2005) Aikawa, Y., Herbst, E., Roberts, H., Caselli, P. 2005, ApJ, 620, 330 * Anderson et al. (1990) Anderson, T., Herbst, E., De Lucia, F. C. 1990, ApJS, 72, 797 * Bachiller and Perez Gutierrez (1998) Bachiller, R., Pérez Gutiérrez, M. 1997, ApJ, 487, L93 * Benson and Myers (1989) Benson, P. J., Myers, P. C. 1989, ApJS, 71, 89 * Benson et al. (1998) Benson, P. J., Caselli, P., Myers, P. C. 1998, ApJ, 506, 743 * Bergin et al. (2001) Bergin, E. A., Ciardi, D. R., Lada, C. J., Alves, J., Lada, E. A. 2001, ApJ, 557, 209 * Bergin and Tafalla (2007) Bergin, E. A., Tafalla, M. 2007, ARA&A, 45, 339 * Beuther et al. (2002) Beuther, H., Schilke, P., Menten, K. M., Motte, F., Sridharan, T. K., Wyrowski, F. 2002, ApJ, 566, 945 * Beuther et al. (2005a) Beuther, H., Schilke, P., Menten, K. M., Motte, F., Sridharan, T. K., Wyrowski, F. 2005a, ApJ, 633, 535 * Beuther et al. (2005b) Beuther, H., Sridharan, T. K., Saito, M. 2005b, ApJ, 634, L185 * Beuther and Steinacker (2007) Beuther, H., Steinacker, J. 2007, ApJ, 656, L85 * Beuther and Sridharan (2007) Beuther, H., Sridharan, T. K. 2007, ApJ, 668, 348 * Carey et al. (1998) Carey, S. J., Clark, F. O., Egan, M. P., Price, S. D., Shipman, R. F., Kuchar, T. A. 1998, ApJ, 508, 721 * Caselli et al. (1995) Caselli, P., Myers, P. C., Thaddeus, P. 1995, ApJ, 455, L77 * Caselli et al. (1999) Caselli, P., Walmsley, C. M., Tafalla, M., Dore, L., Myers, P. C. 1999, ApJ, 523, L165 * Caselli et al. (2002a) Caselli, P., Walmsley, C. M., Zucconi, A., Tafalla, M., Dore, L., Myers, P. C. 2002a, ApJ, 565, 344 * Caselli et al. (2002b) Caselli, P., Benson, P. J., Myers, P. C., Tafalla, M. 2002b, ApJ, 572, 238 * Clemens (1985) Clemens, D. P. 1985, ApJ, 295, 422 * Egan et al. (1998) Egan, M. P., Shipman, R. F., Price, S. D., Carey, S. J., Clark, F. O., Cohen, M. 1998, ApJ, 494, L199 * Ellingsen (2006) Ellingsen, S. P. 2006, ApJ, 638, 241 * Ezawa et al. (2004) Ezawa, H., Kawabe, R., Kohno, K., Yamamoto, S. 2004, Proc. SPIE, 5489, 763 * Jackson et al. (2006) Jackson, J. M., Rathborne, J. M., Shah, R. Y., Simon, R., Bania, T. M., Clemens, D. P., Chambers, E. T., Johnson, A. M., Dormody, M., Lavoie, R., Heyer, M. H. 2006, ApJS, 163, 145 * Hirota et al. (2002) Hirota, T., Ito, T., Yamamoto, S. 2002, ApJ, 565, 359 * Hirota et al. (2004) Hirota, T., Maezawa, H., Yamamoto, S. 2004, ApJ, 617, 399 * Hirota and Yamamoto (2006) Hirota, T., Yamamoto, S. 2006, ApJ, 646, 258 * Ho and Townes (1983) Ho, P. T. P., Townes, C. H. 1983, ARA&A, 21, 239 * Kohno (2005) Kohno, K. 2005, ASP Conf. Ser., 344, 242 * Lai and Crutcher (2000) Lai, S.-P., Crutcher, R. M. 2000, ApJS, 128, 271 * Lee et al. (1996) Lee, H.-H., Herbst, E., Pineau des Forêts, G., Roueff, E., Le Bourlot, J. 1996, A&A, 311, 690 * Leurini et al. (2007) Leurini, S., Schilke, P., Wyrowski, F., Menten, K. M. 2007, A&A, 466, 215 * Li et al. (2003) Li, D., Goldsmith, P. F., Menten, K. 2003, ApJ, 587, 262 * Maret et al. (2005) Maret, S., Ceccarelli, C., Tielens, A. G. G. M., Caux, E., Lefloch, B., Faure, A., Castets, A., Flower, D. R. 2005, A&A, 442, 527 * McKee and Tan (2003) McKee, C. F., Tan, J. C. 2003, ApJ, 585, 850 * Mikami et al. (1992) Mikami, H., Umemoto, T., Yamamoto, S., Saito, S. 1992, ApJ, 392, L87 * Miralles et al. (1992) Miralles, M. P., Rodríguez, L. F., Scalise, E. 1992, ApJS, 92, 173 * Nomura and Millar (2004) Nomura, H., Millar, T. J. 2004, A&A, 414, 409 * Perault et al. (1996) Pérault, M., Omont, A., Simon, G., Séguin, P., Ojha, D., Blommaert, J., Felli, M., Gilmore, G., et al. 1996, A&A, 315, L165 * Pillai et al. (2006a) Pillai, T., Wyrowski, F., Menten, K. M., Krügel, E. 2006a, A&A, 447, 929 * Pillai et al. (2006b) Pillai, T., Wyrowski, F., Carey, S. J., Menten, K. M. 2006b, A&A, 450, 569 * Pillai et al. (2007) Pillai, T., Wyrowski, F., Hatchell, J., Gibb, A. G., Thompson, M. A. 2007, A&A, 467, 207 * Pirogov et al. (2003) Pirogov, L., Zinchenko, I., Caselli, P., Johansson, L. E. B., Myers, P. C. 2003, A&A, 405, 639 * Purcell et al. (2006) Purcell, C. R., Balasubramanyam, R., Burton, M. G., Walsh, A. J., Minier, V., Hunt-Cunningham, M. R., Kedziora-Chudczer, L. L., Longmore, S. N., Hill, T., et al. 2006, MNRAS, 367, 553 * Ragan et al. (2006) Ragan, S. E., Bergin, E. A., Plume, R., Gibson, D. L., Wilner, D. J., O’Brien, S., Hails, E. 2006 ApJS, 166, 567 * Rathborne et al. (2005) Rathborne, J. M., Jackson, J. M., Chambers, E. T., Simon, R., Shipman, R., Frieswijk, W. 2005, ApJ, 630, L181 * Rathborne et al. (2006) Rathborne, J. M., Jackson, J. M., Simon, R. 2006, ApJ, 641, 389 * Rathborne et al. (2007) Rathborne, J. M., Simon, R., Jackson, J. M. 2007, ApJ, 662, 1082 * Sakai et al. (2006) Sakai, T., Oka, T., Yamamoto, S. 2006, ApJ, 649, 268 * Sakai et al. (2007) Sakai, T., Oka, T., Yamamoto, S. 2007, ApJ, 662, 1043 * Shepherd et al. (2007) Shepherd, D. S., Povich, M. S., Whitney, B. A., Robitaille, T. P., Nünberger, D. E. A., Bronfman, L., Stark, D. P., Indebetouw, R., Meade, M. R., Babler, B. L. 2007, ApJ, 669, 464 * Simon et al. (2006) Simon, R., Jackson, J. M., Rathborne, J. M., Chambers, E. T. 2006, ApJ, 639, 227 * Sridharan et al. (2002) Sridharan, T. K., Beuther, H., Schilke, P., Menten, K. M., Wyrowski, F. 2002, ApJ, 566, 931 * Sridharan et al. (2005) Sridharan, T. K., Beuther, H., Saito, M., Wyrowski, F., Schilke, P. 2005, ApJ, 534, L57 * Suzuki et al. (1992) Suzuki, H., Yamamoto, S., Ohishi, M., Kaifu, N., Ishikawa, S., Hirahara, Y., Takano, S. 1992, ApJ, 392, 551 * Tatematsu et al. (1993) Tatematsu, K., Umemoto, T., Kameya, O., Hirano, N., Hasegawa, T., Hayashi, M., Iwata, T., Kaifu, N., Mikami, H., Murata, Y., Nakano, M., Nakano, T., Ohashi, N., Sunada, K., Takaba, H., Yamamoto, S. 1993, ApJ, 404, 643 * Turner (1991) Teyssier, D., Hennebelle, P., Pérault, M. 2002, A&A, 382, 624 * Turner (1991) Turner, B. E. 1991, ApJS, 76, 617 * Umemoto et al. (1992) Umemoto, T., Iwata, T., Fukui, Y., Mikami, H., Yamamoto, S., Kameya, O., Hirano, N. 1992, ApJ, 392, L83 * Umemoto et al. (1999) Umemoto, T., Mikami, H., Yamamoto, S., Hirano, N. 1999, ApJ, 525, L105 * Ungerechts et al. (1997) Ungerechts, H., Bergin, E. A., Goldsmith, P. F., Irvine, W. M., Schloerb, F. P., Snell, R. L. 1997, ApJ, 482, 245 * van der Tak et al. (2000a) van der Tak, F. F. S., van Dishoeck, E. F., Evans, N. J., II; Blake, G. A. 2000a, ApJ, 537, 283 * van der Tak et al. (2000b) van der Tak, F. F. S., van Dishoeck, E. F., Caselli, P. 2000b, A&A, 361, 327 * Yamamoto et al. (1990) Yamamoto, S., Saito, S., Kawaguchi, K., Chikada, Y., Suzuki, H., Kaifu, N., Ishikawa, S., Ohishi, M. 1990, ApJ, 361, 318 Figure 1: Spectra of the N2H+ $J$=1–0, HC3N $J$=5–4, CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$, and CCS $J_{N}$=$4_{3}$–$3_{2}$ lines toward the selected 9 objects (a) and all the other objects (b–e). For clarity, the spectra, except for N2H+, are offset from zero. In Figure 1a, the intensities for the spectra toward G034.43+00.24 MM2 and MM3 are divided by 2. Figures 1b–e are available in the electric edition of the journal. Figure 1b: Continued. Figure 1c: Continued. Figure 1d: Continued. Figure 1e: Continued. Figure 2: Spectra of the NH3 ($J$, $K$) = (1, 1), (2, 2), and (3, 3) lines toward the selected 9 objects (a) and all the other objects (b–e). For clarity, the spectra, except for the (1, 1) line, are offset from zero. Figures 2b–e are available in the electric edition of the journal. Figure 2b: Continued. Figure 2c: Continued. Figure 2d: Continued. Figure 2e: Continued. Figure 3: Spectra of the CH3OH $J$=7–6 lines. Figure 4: Correlation plots of the integrated intensities against the 1.2 mm continuum peak flux. The 1.2 mm continuum data are observed by Beuther et al. (2002) and Rathborne et al. (2006). The values of the objects with $D$ $\leq$ 4.5 kpc are plotted. Figure 5: Histogram of the NH3 rotation temperature. The objects with $D$ $\leq$ 4.5 kpc are used for the histogram. Figure 6: Correlation plots of the N2H+ $J$=1–0 velocity width against the velocity widths of HC3N $J$=5–4, CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$, CH3OH $J_{K}$=$7_{-1}$–$6_{-1}$ $E$, NH3 ($J$, $K$) = (1, 1), (2, 2) and (3, 3). The objects with $D$ $\leq$ 4.5 kpc are plotted. Figure 7: Correlation plots of the velocity width of CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ against the velocity width of NH3 ($J$, $K$) = (3, 3). The objects with $D$ $\leq$ 4.5 kpc are plotted. Figure 8: Histograms of the N2H+ velocity width of the MSX dark objects (top) and MSX sources (bottom). The objects with $D$ $\leq$ 4.5 kpc are plotted. Figure 9: Spizter 24 $\mu$m image toward G034.43+00.24 (left), and spectra of the CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ (black) and N2H+ $J$=1–0 $F_{1}$=0–1 $F$=1–2 lines (gray) toward G034.43+00.24 MM1, MM2 and MM3 (right). Table 1: Target list. Source | R. A. | Dec. | MSXaa”D” indicates a dark object, and ”S” indicates an object with point-like sources, whereas ”—” represents an object without point-like sources and dark spots. | Spitzeraa”D” indicates a dark object, and ”S” indicates an object with point-like sources, whereas ”—” represents an object without point-like sources and dark spots. | $F_{pk}$(1.2 mm) | Reference ---|---|---|---|---|---|--- | (J2000.0) | (J2000.0) | 8 $\mu$m | 24 $\mu$m | [mJy] | G015.05+00.07 MM1 | 18 17 50.4 | -15 53 38 | D | D | 115 | 1 G015.31-00.16 MM2 | 18 18 50.4 | -15 43 19 | D | S | 39 | 1 G015.31-00.16 MM3 | 18 18 45.3 | -15 41 58 | D | D | 37 | 1 G019.27+00.07 MM1 | 18 25 58.5 | -12 03 59 | D | S | 150 | 1 G019.27+00.07 MM2 | 18 25 52.6 | -12 04 48 | D | S | 89 | 1 G022.35+00.41 MM1 | 18 30 24.4 | -09 10 34 | D | S | 349 | 1 G022.35+00.41 MM2 | 18 30 24.2 | -09 12 44 | S | S | 67 | 1 G022.35+00.41 MM3 | 18 30 38.1 | -09 12 44 | — | — | 53 | 1 G023.60+00.00 MM1 | 18 34 11.6 | -08 19 06 | — | S | 375 | 1 G023.60+00.00 MM2 | 18 34 21.1 | -08 18 07 | D | S | 272 | 1 G023.60+00.00 MM3 | 18 34 10.0 | -08 18 28 | — | — | 81 | 1 G023.60+00.00 MM4 | 18 34 23.0 | -08 18 21 | — | — | 68 | 1 G023.60+00.00 MM7 | 18 34 21.1 | -08 17 11 | D | S | 58 | 1 G024.08+00.04 MM1 | 18 34 57.0 | -07 43 26 | S | S | 219 | 1 G024.08+00.04 MM2 | 18 34 51.1 | -07 45 32 | D | — | 68 | 1 G024.08+00.04 MM3 | 18 35 02.2 | -07 45 25 | D | D | 50 | 1 G024.08+00.04 MM4 | 18 35 02.6 | -07 45 56 | D | D | 48 | 1 G024.33+00.11 MM1 | 18 35 07.9 | -07 35 04 | S | S | 1199 | 1 G024.33+00.11 MM2 | 18 35 34.5 | -07 37 28 | — | — | 117 | 1 G024.33+00.11 MM3 | 18 35 27.9 | -07 36 18 | — | — | 96 | 1 G024.33+00.11 MM4 | 18 35 19.4 | -07 37 17 | D | D | 90 | 1 G024.33+00.11 MM5 | 18 35 33.8 | -07 36 42 | — | — | 79 | 1 G024.33+00.11 MM6 | 18 35 07.7 | -07 34 33 | D | — | 77 | 1 G024.33+00.11 MM8 | 18 35 23.4 | -07 37 21 | — | — | 72 | 1 G024.33+00.11 MM9 | 18 35 26.5 | -07 36 56 | S | S | 66 | 1 G024.33+00.11 MM11 | 18 35 05.1 | -07 35 58 | D | D | 48 | 1 G024.60+00.08 MM1 | 18 35 40.2 | -07 18 37 | — | S | 279 | 1 G024.60+00.08 MM2 | 18 35 35.7 | -07 18 09 | D | S | 230 | 1 G025.04-00.20 MM1 | 18 38 10.2 | -07 02 34 | — | S | 203 | 1 G025.04-00.20 MM2 | 18 38 17.7 | -07 02 51 | D | S | 92 | 1 G025.04-00.20 MM4 | 18 38 13.7 | -07 03 12 | D | D | 82 | 1 G034.43+00.24 MM1 | 18 53 18.0 | 01 25 24 | — | S | 2228 | 1 G034.43+00.24 MM2 | 18 53 18.6 | 01 24 40 | S | S | 964 | 1 G034.43+00.24 MM3 | 18 53 20.4 | 01 28 23 | D | S | 244 | 1 G034.43+00.24 MM4 | 18 53 19.0 | 01 24 08 | — | S | 221 | 1 G034.43+00.24 MM5 | 18 53 19.8 | 01 23 30 | — | — | 122 | 1 G034.43+00.24 MM6 | 18 53 18.6 | 01 27 48 | D | D | 57 | 1 G034.43+00.24 MM8 | 18 53 16.4 | 01 26 20 | D | D | 51 | 1 G034.43+00.24 MM9 | 18 53 18.4 | 01 28 14 | D | D | 50 | 1 I18102-1800 MM1 | 18 13 11.0 | -17 59 59 | — | S | 316 | 2,3,4 I18151-1208 MM1 | 18 17 58.0 | -12 07 27 | S | No data | 672 | 2,3,4 I18151-1208 MM2 | 18 17 50.4 | -12 07 55 | D | No data | 424 | 2,3,4 I18151-1208 MM3 | 18 17 52.2 | -12 06 56 | S | No data | 149 | 2,3,4 I18182-1433 MM1 | 18 21 09.2 | -14 31 57 | S | S | 1303 | 2,3,4 I18182-1433 MM2 | 18 21 14.9 | -14 33 06 | — | S | 100 | 2,3,4 I18223-1243 MM1 | 18 25 10.5 | -12 42 26 | S | S | 328 | 2,3,4 I18223-1243 MM2 | 18 25 09.5 | -12 44 15 | — | S | 124 | 2,3,4 I18223-1243 MM3 | 18 25 08.3 | -12 45 28 | D | S | 205 | 2,3,4 I18223-1243 MM4 | 18 25 07.2 | -12 47 54 | D | S | 52 | 2,3,4 I18306-0835 MM1 | 18 33 24.0 | -08 33 31 | S | S | 731 | 2,3,4 I18306-0835 MM2 | 18 33 17.2 | -08 33 26 | — | S | 212 | 2,3,4 I18306-0835 MM3 | 18 33 32.1 | -08 32 29 | D | D | 103 | 2,3,4 I18337-0743 MM1 | 18 36 41.0 | -07 39 20 | S | S | 485 | 2,3,4 I18337-0743 MM2 | 18 36 27.7 | -07 40 28 | S | S | 180 | 2,3,4 I18337-0743 MM3 | 18 36 18.2 | -07 41 01 | D | S | 110 | 2,3,4 References. — (1) Rathborne et al. (2006); (2) Sridharan et al. (2002); (3) Beuther et al. (2002); (4) Sridharan et al. (2005). Table 2: Observed lines. Species | Transition | $\nu$(rest) | $\mu^{2}S$ | $E_{u}$/$k$ | Reference | Telescope ---|---|---|---|---|---|--- | | [GHz] | [Debye2] | [K] | | CCS | $J_{N}$=$4_{3}$–$3_{2}$ | 45.379033 | 31.3 | 5.4 | 1 | NRO 45 m HC3N | $J$=5–4 | 45.490302 | 69.2 | 6.5 | 2 | NRO 45 m N2H+ | $J$=1–0 | 93.173777 | 11.6 | 4.47 | 3 | NRO 45 m NH3 | ($J$, $K$) = (1, 1) | 23.694495 | 3.24 | 23.4 | 4 | NRO 45 m NH3 | ($J$, $K$) = (2, 2) | 23.722633 | 7.20 | 64.9 | 4 | NRO 45 m NH3 | ($J$, $K$) = (3, 3) | 23.870129 | 11.3 | 124.5 | 4 | NRO 45 m CH3OH | $J_{K}$ = $7_{0}$–$6_{0}$ $E$ | 338.124502 | 5.480 | 77.1 | 5 | ASTE CH3OH | $J_{K}$ = $7_{-1}$–$6_{-1}$ $E$ | 338.344628 | 5.375 | 69.6 | 5 | ASTE CH3OH | $J_{K}$ = $7_{0}$–$6_{0}$ $A^{+}$ | 338.408681 | 5.484 | 65.0 | 5 | ASTE CH3OH | $J_{K}$ = $7_{2}$–$6_{2}$ $A^{-}$ | 338.512762 | 5.059 | 86.4 | 5 | ASTE CH3OH | $J_{K}$ = $7_{3}$–$6_{3}$ $A^{+}$ | 338.540795 | 4.457 | 114.8 | 5 | ASTE CH3OH | $J_{K}$ = $7_{3}$–$6_{3}$ $A^{-}$ | 338.543204 | 4.457 | 114.8 | 5 | ASTE CH3OH | $J_{K}$ = $7_{-3}$–$6_{-3}$ $E$ | 338.559928 | 4.503 | 126.8 | 5 | ASTE CH3OH | $J_{K}$ = $7_{3}$–$6_{3}$ $E$ | 338.583195 | 4.486 | 111.8 | 5 | ASTE CH3OH | $J_{K}$ = $7_{1}$–$6_{1}$ $E$ | 338.614999 | 5.508 | 85.1 | 5 | ASTE CH3OH | $J_{K}$ = $7_{2}$–$6_{2}$ $A^{+}$ | 338.639939 | 5.509 | 102.7 | 5 | ASTE CH3OH | $J_{K}$ = $7_{2}$–$6_{2}$ $E$ | 338.721630 | 4.980 | 86.3 | 5 | ASTE CH3OH | $J_{K}$ = $7_{-2}$–$6_{-2}$ $E$ | 338.722940 | 5.046 | 90.0 | 5 | ASTE CH3OH | $J_{K}$ = $4_{0}$–$3_{-1}$ $E$ | 350.687730 | 1.553 | 35.4 | 5 | ASTE CH3OH | $J_{K}$ = $1_{1}$–$0_{0}$ $A^{+}$ | 350.905119 | 1.977 | 16.8 | 5 | ASTE References. — (1) Yamamoto et al. (1990); (2) JPL Catalog (http://spec.jpl.nasa.gov/) ; (3) Caselli et al. (1995) ; (4) Ho & Townes (1983) ; (5) Anderson et al. (1990). Table 3: 3 $\sigma$ detection rates of the observed lines.aaTypical 3 $\sigma$ level in $T_{a}^{*}$ and channel width in velocity are 0.30 K and 0.070 km s-1 for N2H+, 0.22 K and 0.134 km s-1 for HC3N, 0.24 K and 0.134 km s-1 for CCS, 0.28 K and 0.443 km s-1 for CH3OH, and 0.08 K and 0.253 km s-1 for NH3. MSX | Spitzer | N2H+ | HC3N | CCS | CH3OH | NH3 | NH3 | NH3 ---|---|---|---|---|---|---|---|--- | | $J$=1–0 | $J$=5–4 | $J_{N}$=$4_{3}$–$3_{2}$ | $J$=7–6 | (1, 1) | (2, 2) | (3, 3) Dark | Dark | 11/11 | 7/11 | 0/11 | 0/11 | 11/11 | 10/11 | 3/11 Dark | Source | 12/12 | 12/12 | 0/12 | 6/12 | 12/12 | 11/12 | 8/12 Dark | Others | 3/3 | 2/3 | 0/3 | 1/3 | 3/3 | 3/3 | 2/3 Source | Source | 12/12 | 9/12 | 0/12 | 6/12 | 12/12 | 11/12 | 9/12 Others | Source | 9/9 | 7/9 | 0/9 | 5/9 | 9/9 | 9/9 | 7/9 Others | Others | 7/8 | 6/8 | 0/8 | 0/8 | 8/8 | 8/8 | 5/8 Total | | 54/55 | 43/55 | 0/55 | 18/55 | 55/55 | 52/55 | 34/55 Table 4: Line parameters of the N2H+ $J$=1–0 line and distance of each object.aaThe quoted errors denote one standard deviation. Source | $T_{\rm ex}$ | $\tau$ | $V_{LSR}$ | $\Delta V$ | $\int T_{a}^{*}dV$ bbVelocity range for integration is $\pm 10$ km s-1 centered at the $LSR$ velocity of N2H+ $J$=1–0 or NH3 (1, 1). | Distance ccThe distance is the assumed kinematic distance based on the rotation curve of Clemens (1985). ---|---|---|---|---|---|--- | [K] | | [km s-1] | [km s-1] | [K km s-1] | [kpc] G015.05+00.07 MM1 | 5.1 (0.2) | 2.5 (0.4) | 24.73 (0.03) | 2.79 (0.08) | 6.0 (0.1) | 2.5 G015.31-00.16 MM2 | 4.0 (0.2) | 6.9 (2.0) | 31.12 (0.02) | 0.81 (0.07) | 2.3 (0.1) | 3.1 G015.31-00.16 MM3 | 4.0 (0.1) | 11.4 (2.6) | 30.81 (0.02) | 0.62 (0.04) | 2.6 (0.1) | 3.0 G019.27+0.07 MM1 | 15 (1) | 1.1 (0.1) | 26.76 (0.01) | 3.15 (0.03) | 17.7 (0.1) | 2.3 G019.27+0.07 MM2 | 9.0 (0.6) | 1.5 (0.2) | 27.01 (0.02) | 2.62 (0.04) | 9.8 (0.1) | 2.3 G022.35+00.41 MM1 | 5.9 (0.1) | 3.1 (0.3) | 52.73 (0.02) | 2.74 (0.05) | 8.9 (0.1) | 3.7 G022.35+00.41 MM2 | 6 (2) | 1.50 (1.48) | 60.22 (0.02) | 0.83 (0.07) | 1.9 (0.1) | 4.1 G022.35+00.41 MM3 | 5.0 (0.2) | 3.4 (0.5) | 83.97 (0.03) | 2.23 (0.07) | 5.4 (0.1) | 5.3 G023.60+00.00 MM1 | 7.9 (0.5) | 1.5 (0.2) | 106.89 (0.06) | 3.82 (0.07) | 11.8 (0.1) | 6.7 G023.60+00.00 MM2 | 5.8 (0.3) | 2.2 (0.3) | 53.34 (0.03) | 3.03 (0.07) | 7.6 (0.1) | 3.7 G023.60+00.00 MM3 | 8 (3) | 0.7 (0.4) | 105.51 (0.04) | 3.4 (0.1) | 5.8 (0.1) | 6.6 G023.60+00.00 MM4 | 13 (24) | 0.2 (0.6) | 53.87 (0.05) | 2.6 (0.1) | 3.0 (0.1) | 3.7 G023.60+00.00 MM7 | 8 (1) | 1.3 (0.3) | 54.00 (0.05) | 1.90 (0.06) | 6.0 (0.1) | 3.7 G024.08+00.04 MM1 | 49 (26) | 0.21 (0.13) | 113.63 (0.01) | 2.67 (0.03) | 12.8 (0.1) | 7.8 G024.08+00.04 MM2(1) | 9.4 (8.5) | 0.6 (0.9) | 52.08 (0.03) | 1.5 (0.1) | 2.8 (0.1) | 3.6 G024.08+00.04 MM2(2) | 7 (3) | 0.8 (0.7) | 114.45 (0.04) | 2.1 (0.1) | 3.6 (0.1) | 7.8 G024.08+00.04 MM3 | 3.5 (0.1) | 3.5 (1) | 51.86 (0.06) | 2.2 (0.2) | 2.0 (0.1) | 3.6 G024.08+00.04 MM4 | 3.32 (0.04) | 19 (6) | 51.95 (0.03) | 0.85 (0.08) | 1.7 (0.1) | 3.6 G024.33+00.11 MM1 | 7.6 (0.1) | 3.4 (0.2) | 113.76 (0.02) | 3.53 (0.04) | 19.0 (0.1) | 7.7 G024.33+00.11 MM2 | 5.4 (0.3) | 2.3 (0.4) | 118.22 (0.04) | 3.23 (0.09) | 6.8 (0.1) | 7.7 G024.33+00.11 MM3 | 8 (2) | 0.8 (0.3) | 117.78 (0.05) | 4.3 (0.1) | 7.9 (0.1) | 7.7 G024.33+00.11 MM4 | 6 (2) | 0.9 (0.6) | 115.19 (0.07) | 3.6 (0.2) | 4.2 (0.1) | 7.7 G024.33+00.11 MM5 | 4.69 (0.09) | 4.6 (0.5) | 117.26 (0.04) | 3.00 (0.08) | 7.6 (0.1) | 7.7 G024.33+00.11 MM6 | 7 (2) | 0.9 (0.5) | 114.43 (0.04) | 2.97 (0.10) | 4.5 (0.1) | 7.7 G024.33+00.11 MM8 | — | — | — | — | 0.9 (0.1) | 7.7 G024.33+00.11 MM9 | 3.8 (0.2) | 4.2 (1.5) | 119.78 (0.05) | 1.7 (0.2) | 2.7 (0.1) | 7.7 G024.33+00.11 MM11 | 4.3 (0.3) | 2.4 (0.9) | 112.60 (0.05) | 2.18 (0.05) | 3.0 (0.1) | 7.7 G024.60+00.08 MM1 | 5.4 (0.1) | 3.9 (0.3) | 53.14 (0.03) | 3.10 (0.06) | 9.8 (0.1) | 3.6 G024.60+00.08 MM2 | 10 (2) | 1.2 (0.2) | 115.16 (0.01) | 2.37 (0.04) | 9.3 (0.1) | 7.7 G025.04-00.20 MM1 | 11 (1) | 1.1 (0.2) | 64.00 (0.02) | 2.36 (0.05) | 9.8 (0.1) | 4.2 G025.04-00.20 MM2 | 6.5 (0.4) | 2.0 (0.3) | 63.83 (0.03) | 2.77 (0.07) | 8.1 (0.1) | 4.2 G025.04-00.20 MM4 | 6.2 (0.5) | 1.9 (0.4) | 63.69 (0.02) | 2.18 (0.08) | 6.0 (0.1) | 4.1 G034.43+00.24 MM1 | 39 (2) | 0.75 (0.05) | 57.77 (0.05) | 3.15 (0.01) | 40.5 (0.1) | 3.5 G034.43+00.24 MM2 | 29 (2) | 0.67 (0.06) | 57.832 (0.009) | 4.04 (0.02) | 34.2 (0.1) | 3.5 G034.43+00.24 MM3 | 36 (7) | 0.42 (0.09) | 59.686 (0.008) | 2.90 (0.02) | 20.3 (0.1) | 3.6 G034.43+00.24 MM4 | 14.2 (0.5) | 1.5 (0.1) | 57.689 (0.008) | 2.81 (0.02) | 20.8 (0.1) | 3.5 G034.43+00.24 MM5 | 9 (2) | 0.9 (0.3) | 58.12 (0.03) | 2.58 (0.08) | 6.0 (0.1) | 3.5 G034.43+00.24 MM6 | 10 (2) | 0.8 (0.4) | 58.49 (0.02) | 2.07 (0.07) | 5.6 (0.1) | 3.6 G034.43+00.24 MM8 | 6 (1) | 1.1 (0.5) | 57.63 (0.05) | 3.2 (0.1) | 4.5 (0.1) | 3.5 G034.43+00.24 MM9 | 12 (3) | 0.7 (0.3) | 59.00 (0.02) | 2.45 (0.06) | 7.3 (0.1) | 3.6 I18102-1800 MM1 | 9.3 (0.5) | 1.7 (0.2) | 22.36 (0.02) | 3.49 (0.05) | 15.9 (0.1) | 2.7 I18151-1208 MM1 | 13 (2) | 0.8 (0.2) | 33.19 (0.01) | 2.10 (0.04) | 9.8 (0.1) | 2.8 I18151-1208 MM2 | 14 (2) | 0.9 (0.2) | 29.74 (0.02) | 3.09 (0.04) | 14.2 (0.1) | 2.6 I18151-1208 MM3 | 5.7 (0.2) | 4.9 (0.9) | 30.65 (0.01) | 0.87 (0.04) | 4.4 (0.1) | 2.7 I18182-1433 MM1 | 30 (8) | 0.4 (0.1) | 60.00 (0.01) | 3.06 (0.04) | 16.6 (0.1) | 4.6 I18182-1433 MM2 | 5.8 (0.4) | 2.3 (0.4) | 41.10 (0.03) | 2.28 (0.08) | 6.2 (0.1) | 3.6 I18223-1243 MM1 | 14 (1) | 1.2 (0.2) | 45.32 (0.01) | 2.16 (0.03) | 13.4 (0.1) | 3.6 I18223-1243 MM2 | 8.1 (0.7) | 1.3 (0.2) | 45.29 (0.02) | 2.99 (0.06) | 8.7 (0.1) | 3.6 I18223-1243 MM3 | 9.4 (0.3) | 2.2 (0.2) | 45.73 (0.01) | 2.64 (0.03) | 14.5 (0.1) | 3.7 I18223-1243 MM4 | 5.3 (0.3) | 2.4 (0.5) | 45.83 (0.03) | 1.97 (0.07) | 5.0 (0.1) | 3.7 I18306-0835 MM1 | 27 (31) | 0.2 (0.3) | 78.02 (0.03) | 2.81 (0.07) | 7.2 (0.1) | 4.9 I18306-0835 MM2 | 10 (11) | 0.4 (0.6) | 76.67 (0.05) | 2.6 (0.1) | 3.4 (0.1) | 4.9 I18306-0835 MM3 | 5 (2) | 0.95 (1.5) | 53.98 (0.09) | 2.1 (0.3) | 1.6 (0.1) | 3.7 I18337-0743 MM1 | 7.03 (0.08) | 5.9 (0.3) | 58.37 (0.02) | 3.35 (0.04) | 21.5 (0.1) | 3.9 I18337-0743 MM2 | 8 (1) | 1.3 (0.4) | 58.60 (0.03) | 2.71 (0.08) | 6.9 (0.1) | 3.9 I18337-0743 MM3 | 6.7 (0.6) | 1.7 (0.4) | 56.37 (0.03) | 2.44 (0.08) | 6.6 (0.1) | 3.8 Table 5: Line parameters of the HC3N $J$=5–4 and CCS $J_{N}$=$4_{3}$–$3_{2}$ lines.aaThe quoted errors denote one standard deviation. | HC3N | | CCS ---|---|---|--- Source | $T_{\rm pk}$ | $V_{LSR}$ | $\Delta V$ | $\int T_{a}^{*}dV$bbVelocity range for integration is $\pm 5$ km s-1 centered at the $LSR$ velocity of N2H+ $J$=1–0 or NH3 (1, 1). | | $T_{\rm pk}$ | $\int T_{a}^{*}dV$bbVelocity range for integration is $\pm 5$ km s-1 centered at the $LSR$ velocity of N2H+ $J$=1–0 or NH3 (1, 1). | [K] | [km s-1] | [km s-1] | [K km s-1] | | [K] | [K km s-1] G015.05+00.07 MM1 | 0.25 (0.02) | 24.4 (0.1) | 3.5 (0.3) | 1.05 (0.09) | | $<$0.22 | $<$0.25 G015.31-00.16 MM2 | 0.28 (0.03) | 30.92 (0.08) | 1.4 (0.2) | 0.61 (0.09) | | $<$0.24 | $<$0.28 G015.31-00.16 MM3 | $<$0.23 | — | — | 0.51 (0.09) | | $<$0.25 | $<$0.29 G019.27+0.07 MM1 | 0.60 (0.02) | 26.39 (0.05) | 3.7 (0.1) | 2.60 (0.07) | | $<$0.19 | $<$0.23 G019.27+0.07 MM2 | 0.64 (0.02) | 27.02 (0.04) | 2.9 (0.1) | 2.35 (0.07) | | $<$0.20 | $<$0.23 G022.35+00.41 MM1 | 0.46 (0.02) | 53.00 (0.07) | 3.6 (0.2) | 1.94 (0.08) | | $<$0.20 | $<$0.23 G022.35+00.41 MM2 | $<$0.20 | — | — | $<$0.24 | | $<$0.21 | $<$0.24 G022.35+00.41 MM3 | 0.34 (0.03) | 84.36 (0.06) | 1.6 (0.1) | 0.59 (0.07) | | $<$0.19 | $<$0.22 G023.60+00.00 MM1 | 0.57 (0.02) | 106.7 (0.06) | 3.6 (0.1) | 2.28 (0.08) | | $<$0.23 | $<$0.26 G023.60+00.00 MM2 | 0.81 (0.02) | 53.37 (0.04) | 3.4 (0.1) | 3.35 (0.08) | | $<$0.20 | $<$0.23 G023.60+00.00 MM3 | $<$0.21 | — | — | 0.41 (0.08) | | $<$0.23 | $<$0.27 G023.60+00.00 MM4 | 0.33 (0.02) | 53.91 (0.07) | 2.3 (0.2) | 0.76 (0.07) | | $<$0.21 | $<$0.25 G023.60+00.00 MM7 | 0.55 (0.02) | 54.08 (0.05) | 2.1 (0.1) | 1.56 (0.08) | | $<$0.21 | $<$0.24 G024.08+00.04 MM1 | 0.30 (0.02) | 113.68 (0.08) | 2.8 (0.2) | 0.89 (0.07) | | $<$0.21 | $<$0.24 G024.08+00.04 MM2(1) | 0.22 (0.03) | 52.18 (0.07) | 0.9 (0.2) | 0.44 (0.07) | | $<$0.21 | $<$0.24 G024.08+00.04 MM2(2) | $<$0.19 | — | — | 0.38 (0.07) | | $<$0.21 | $<$0.24 G024.08+00.04 MM3 | 0.39 (0.02) | 51.67 (0.04) | 1.6 (0.1) | 0.78 (0.07) | | $<$0.24 | $<$0.28 G024.08+00.04 MM4 | 0.56 (0.03) | 51.90 (0.02) | 1.08 (0.07) | 0.83 (0.07) | | $<$0.21 | $<$0.24 G024.33+00.11 MM1 | 0.71 (0.02) | 113.69 (0.06) | 4.0 (0.1) | 3.22 (0.09) | | $<$0.28 | $<$0.32 G024.33+00.11 MM2 | 0.34 (0.02) | 118.1 (0.1) | 3.4 (0.2) | 1.27 (0.09) | | $<$0.25 | $<$0.29 G024.33+00.11 MM3 | 0.24 (0.02) | 117.2 (0.2) | 5.1 (0.4) | 1.31 (0.09) | | $<$0.27 | $<$0.31 G024.33+00.11 MM4 | $<$0.23 | — | — | 1.11 (0.09) | | $<$0.26 | $<$0.30 G024.33+00.11 MM5 | 0.34 (0.02) | 117.4 (0.1) | 4.0 (0.3) | 1.48 (0.09) | | $<$0.23 | $<$0.27 G024.33+00.11 MM6 | $<$0.24 | — | — | 0.73 (0.09) | | $<$0.28 | $<$0.32 G024.33+00.11 MM8 | $<$0.23 | — | — | $<$0.27 | | $<$0.29 | $<$0.34 G024.33+00.11 MM9 | 0.21 (0.03) | 119.4 (0.1) | 1.3 (0.2) | 0.50 (0.09) | | $<$0.29 | $<$0.33 G024.33+00.11 MM11 | $<$0.23 | — | — | $<$0.27 | | $<$0.28 | $<$0.32 G024.60+00.08 MM1 | 0.63 (0.02) | 53.22 (0.05) | 3.5 (0.1) | 2.59 (0.08) | | $<$0.17 | $<$0.20 G024.60+00.08 MM2 | 0.29 (0.02) | 115.2 (0.1) | 3.0 (0.2) | 1.23 (0.09) | | $<$0.28 | $<$0.32 G025.04-00.20 MM1 | 0.46 (0.02) | 63.67 (0.07) | 3.0 (0.2) | 1.67 (0.09) | | $<$0.23 | $<$0.26 G025.04-00.20 MM2 | 0.46 (0.03) | 63.63 (0.07) | 2.4 (0.2) | 1.21 (0.09) | | $<$0.29 | $<$0.34 G025.04-00.20 MM4 | 0.35 (0.03) | 63.76 (0.08) | 2.0 (0.2) | 0.67 (0.09) | | $<$0.23 | $<$0.27 G034.43+00.24 MM1 | 1.22 (0.02) | 58.04 (0.03) | 3.85 (0.06) | 5.35 (0.06) | | $<$0.18 | 0.35 (0.07) G034.43+00.24 MM2 | 1.10 (0.01) | 57.86 (0.03) | 4.22 (0.06) | 5.09 (0.06) | | $<$0.18 | 0.43 (0.07) G034.43+00.24 MM3 | 0.57 (0.01) | 60.05 (0.07) | 5.8 (0.2) | 3.29 (0.07) | | $<$0.18 | 0.41 (0.07) G034.43+00.24 MM4 | 0.80 (0.02) | 57.50 (0.03) | 3.22 (0.08) | 3.03 (0.07) | | $<$0.17 | 0.30 (0.07) G034.43+00.24 MM5 | 0.31 (0.03) | 57.81 (0.08) | 1.8 (0.2) | 0.67 (0.08) | | $<$0.22 | 0.38 (0.08) G034.43+00.24 MM6 | 0.45 (0.02) | 58.34 (0.06) | 2.5 (0.1) | 1.25 (0.08) | | $<$0.23 | $<$0.26 G034.43+00.24 MM8 | 0.23 (0.02) | 57.7 (0.2) | 4.4 (0.4) | 1.1 (0.08) | | $<$0.21 | $<$0.24 G034.43+00.24 MM9 | 0.50 (0.02) | 58.40 (0.05) | 2.4 (0.1) | 1.31 (0.08) | | $<$0.21 | $<$0.24 I18102-1800 MM1 | 0.83 (0.02) | 21.53 (0.05) | 4.4 (0.1) | 4.14 (0.09) | | $<$0.29 | $<$0.34 I18151-1208 MM1 | 0.53 (0.03) | 33.26 (0.05) | 2.1 (0.1) | 1.39 (0.09) | | $<$0.26 | $<$0.3 I18151-1208 MM2 | 0.25 (0.02) | 30.3 (0.1) | 3.2 (0.3) | 0.93 (0.10) | | $<$0.27 | $<$0.32 I18151-1208 MM3 | $<$0.24 | — | — | 0.39 (0.09) | | $<$0.28 | $<$0.32 I18182-1433 MM1 | 1.04 (0.03) | 59.86 (0.03) | 2.58 (0.07) | 3.07 (0.09) | | $<$0.26 | $<$0.3 I18182-1433 MM2 | $<$0.25 | — | — | 0.39 (0.10) | | $<$0.26 | $<$0.3 I18223-1243 MM1 | 0.44 (0.03) | 45.38 (0.06) | 2.1 (0.1) | 1.09 (0.08) | | $<$0.22 | $<$0.25 I18223-1243 MM2 | 0.24 (0.03) | 45.5 (0.1) | 2.1 (0.3) | 0.68 (0.09) | | $<$0.21 | $<$0.25 I18223-1243 MM3 | 0.44 (0.02) | 45.65 (0.07) | 3.2 (0.2) | 1.59 (0.08) | | $<$0.24 | $<$0.28 I18223-1243 MM4 | $<$0.22 | — | — | 0.71 (0.08) | | $<$0.22 | $<$0.25 I18306-0835 MM1 | 0.39 (0.02) | 78.5 (0.1) | 4.0 (0.2) | 1.7 (0.09) | | $<$0.30 | $<$0.34 I18306-0835 MM2 | $<$0.25 | — | — | 0.41 (0.10) | | $<$0.31 | $<$0.36 I18306-0835 MM3 | $<$0.25 | — | — | 0.32 (0.10) | | $<$0.31 | $<$0.36 I18337-0743 MM1 | 1.06 (0.02) | 58.31 (0.04) | 3.51 (0.08) | 4.21 (0.09) | | $<$0.22 | $<$0.26 I18337-0743 MM2 | 0.50 (0.03) | 58.13 (0.06) | 2.3 (0.1) | 1.22 (0.09) | | $<$0.24 | $<$0.28 I18337-0743 MM3 | 0.25 (0.02) | 55.6 (0.2) | 3.7 (0.4) | 1.07 (0.09) | | $<$0.28 | $<$0.33 Table 6: Line parameters of the CH3OH $J_{K}$=$7_{0}$–$6_{0}$ $A^{+}$ line.aaThe quoted errors denote one standard deviation. Source | $T_{\rm pk}$ | $V_{LSR}$ | $\Delta V$ | $\int T_{a}^{*}dV$bbVelocity range for integration is $\pm 5$ km s-1 centered at the $LSR$ velocity of N2H+ $J$=1–0 or NH3 (1, 1). ---|---|---|---|--- | [K] | [km s-1] | [km s-1] | [K km s-1] G015.05+00.07 MM1 | $<$0.29 | — | — | $<$0.6 G015.31-00.16 MM2 | $<$0.30 | — | — | $<$0.6 G015.31-00.16 MM3 | $<$0.40 | — | — | $<$0.8 G019.27+0.07 MM1 | 0.57 (0.04) | 26.7 (0.2) | 5.2 (0.4) | 3.1 (0.2) G019.27+0.07 MM2 | $<$0.28 | — | — | $<$0.6 G022.35+00.41 MM1 | 0.64 (0.04) | 52.6 (0.2) | 5.4 (0.4) | 3.6 (0.2) G022.35+00.41 MM2 | $<$0.34 | — | — | $<$0.7 G022.35+00.41 MM3 | $<$0.28 | — | — | $<$0.6 G023.60+00.00 MM1 | 0.67 (0.04) | 107.1 (0.2) | 4.9 (0.4) | 3.5 (0.2) G023.60+00.00 MM2 | 0.34 (0.03) | 53.0 (0.3) | 6.7 (0.7) | 2.2 (0.2) G023.60+00.00 MM3 | $<$0.25 | — | — | $<$0.5 G023.60+00.00 MM4 | $<$0.25 | — | — | $<$0.5 G023.60+00.00 MM7 | $<$0.39 | — | — | $<$0.8 G024.08+00.04 MM1 | $<$0.34 | — | — | $<$0.7 G024.08+00.04 MM2(1) | $<$0.46 | — | — | $<$1.0 G024.08+00.04 MM2(2) | $<$0.46 | — | — | $<$1.0 G024.08+00.04 MM3 | $<$0.31 | — | — | $<$0.7 G024.08+00.04 MM4 | $<$0.39 | — | — | $<$0.8 G024.33+00.11 MM1 | 0.48 (0.04) | 114.2 (0.3) | 6.6 (0.6) | 3.0 (0.2) G024.33+00.11 MM2 | $<$0.29 | — | — | $<$0.6 G024.33+00.11 MM3 | $<$0.19 | — | — | $<$0.4 G024.33+00.11 MM4 | $<$0.28 | — | — | $<$0.6 G024.33+00.11 MM5 | $<$0.28 | — | — | $<$0.6 G024.33+00.11 MM6 | $<$0.36 | — | — | $<$0.8 G024.33+00.11 MM8 | $<$0.24 | — | — | $<$0.5 G024.33+00.11 MM9 | $<$0.30 | — | — | $<$0.6 G024.33+00.11 MM11 | $<$0.26 | — | — | $<$0.5 G024.60+00.08 MM1 | 0.32 (0.03) | 52.1 (0.3) | 5.9 (0.7) | 1.9 (0.2) G024.60+00.08 MM2 | $<$0.32 | — | — | $<$0.7 G025.04-00.20 MM1 | $<$0.28 | — | — | $<$0.6 G025.04-00.20 MM2 | $<$0.27 | — | — | $<$0.6 G025.04-00.20 MM4 | $<$0.33 | — | — | $<$0.7 G034.43+00.24 MM1 | 1.79 (0.09) | 57.6 (0.1) | 4.8 (0.3) | 9.5 (0.1) G034.43+00.24 MM2 | 1.26 (0.06) | 57.7 (0.1) | 5.7 (0.3) | 7.3 (0.2) G034.43+00.24 MM3 | 0.73 (0.03) | 58.8 (0.2) | 8.6 (0.5) | 5.4 (0.1) G034.43+00.24 MM4 | 0.57 (0.05) | 57.2 (0.2) | 4.1 (0.4) | 2.5 (0.2) G034.43+00.24 MM5 | $<$0.30 | — | — | $<$0.6 G034.43+00.24 MM6 | $<$0.31 | — | — | $<$0.7 G034.43+00.24 MM8 | $<$0.28 | — | — | $<$0.6 G034.43+00.24 MM9 | $<$0.32 | — | — | $<$0.7 I18102-1800 MM1 | 0.67 (0.03) | 21.7 (0.2) | 6.2 (0.4) | 4.2 (0.1) I18151-1208 MM1 | 0.30 (0.05) | 33.1 (0.3) | 3.6 (0.6) | 1.1 (0.2) I18151-1208 MM2 | 0.51 (0.04) | 30.4 (0.3) | 7.1 (0.6) | 3.3 (0.2) I18151-1208 MM3 | $<$0.27 | — | — | $<$0.6 I18182-1433 MM1 | 0.43 (0.04) | 60.0 (0.2) | 4.4 (0.5) | 2.0 (0.2) I18182-1433 MM2 | $<$0.39 | — | — | $<$0.8 I18223-1243 MM1 | $<$0.37 | — | — | $<$0.8 I18223-1243 MM2 | $<$0.42 | — | — | $<$0.9 I18223-1243 MM3 | 0.27 (0.03) | 44.9 (0.3) | 5.4 (0.6) | 1.5 (0.1) I18223-1243 MM4 | $<$0.29 | | | $<$0.6 I18306-0835 MM1 | 0.50 (0.04) | 78.0 (0.2) | 3.6 (0.4) | 1.6 (0.1) I18306-0835 MM2 | $<$0.45 | — | — | $<$0.9 I18306-0835 MM3 | $<$0.43 | — | — | $<$0.9 I18337-0743 MM1 | 0.34 (0.03) | 58.1 (0.3) | 5.7 (0.6) | 1.9 (0.1) I18337-0743 MM2 | $<$0.25 | — | — | $<$0.5 I18337-0743 MM3 | 0.23 (0.03) | 56.1 (0.3) | 5.0 (0.8) | 1.3 (0.1) Table 7: Line parameters of the NH3 ($J$, $K$) = (1, 1) line.aaThe quoted errors denote one standard deviation. Source | $T_{\rm pk}$ | $V_{LSR}$ | $\Delta V$ | $\int T_{a}^{*}dV$bbVelocity range for integration is $\pm 5$ km s-1 centered at the $LSR$ velocity of N2H+ $J$=1–0 or NH3 (1, 1). ---|---|---|---|--- | [K] | [km s-1] | [km s-1] | [K km s-1] G015.05+00.07 MM1 | 0.41 (0.02) | 24.72 (0.04) | 2.3 (0.1) | 1.16 (0.04) G015.31-00.16 MM2 | 0.45 (0.02) | 31.05 (0.03) | 1.34 (0.07) | 0.72 (0.04) G015.31-00.16 MM3 | 0.38 (0.02) | 30.93 (0.03) | 1.27 (0.07) | 0.50 (0.04) G019.27+0.07 MM1 | 0.93 (0.03) | 26.37 (0.03) | 2.57 (0.08) | 2.85 (0.04) G019.27+0.07 MM2 | 0.86 (0.02) | 26.79 (0.04) | 2.71 (0.09) | 2.68 (0.04) G022.35+00.41 MM1 | 0.36 (0.01) | 52.67 (0.05) | 3.0 (0.1) | 1.31 (0.03) G022.35+00.41 MM2 | 0.29 (0.02) | 60.22 (0.04) | 1.5 (0.1) | 0.53 (0.04) G022.35+00.41 MM3 | 0.78 (0.02) | 84.29 (0.03) | 2.45 (0.07) | 2.06 (0.04) G023.60+00.00 MM1 | 0.60 (0.02) | 106.10 (0.05) | 4.1 (0.1) | 2.72 (0.03) G023.60+00.00 MM2 | 1.02 (0.02) | 53.49 (0.03) | 2.47 (0.07) | 2.93 (0.04) G023.60+00.00 MM3 | 0.76 (0.02) | 105.45 (0.05) | 3.6 (0.1) | 3.06 (0.04) G023.60+00.00 MM4 | 0.78 (0.02) | 53.65 (0.03) | 2.2 (0.07) | 2.06 (0.04) G023.60+00.00 MM7 | 1.04 (0.03) | 53.74 (0.03) | 2.22 (0.07) | 2.74 (0.04) G024.08+00.04 MM1 | 0.55 (0.02) | 113.61 (0.04) | 2.9 (0.1) | 1.92 (0.04) G024.08+00.04 MM2(1) | 0.31 (0.02) | 51.80 (0.06) | 1.6 (0.1) | 0.69 (0.04) G024.08+00.04 MM2(2) | 0.39 (0.02) | 114.02 (0.05) | 2.4 (0.1) | 1.14 (0.04) G024.08+00.04 MM3 | 0.52 (0.02) | 51.59 (0.03) | 1.89 (0.07) | 1.21 (0.04) G024.08+00.04 MM4 | 0.55 (0.02) | 51.59 (0.03) | 1.67 (0.07) | 1.18 (0.04) G024.33+00.11 MM1 | 1.25 (0.04) | 113.61 (0.05) | 3.4 (0.1) | 4.88 (0.06) G024.33+00.11 MM2 | 0.71 (0.02) | 118.11 (0.05) | 3.5 (0.1) | 2.50 (0.05) G024.33+00.11 MM3 p1 | 0.58 (0.01) | 116.63 (0.04) | 2.16 (0.09) | 2.26 (0.05) G024.33+00.11 MM3 p2 | 0.47 (0.01) | 119.09 (0.05) | 1.9 (0.1) | G024.33+00.11 MM4 | 0.55 (0.02) | 114.89 (0.07) | 4.6 (0.2) | 2.84 (0.05) G024.33+00.11 MM5 | 0.97 (0.01) | 117.47 (0.03) | 3.06 (0.06) | 3.18 (0.05) G024.33+00.11 MM6 | 1.02 (0.03) | 113.85 (0.05) | 3.2 (0.1) | 3.77 (0.06) G024.33+00.11 MM8 | 0.29 (0.02) | 119.93 (0.09) | 2.8 (0.2) | 1.65 (0.03) G024.33+00.11 MM9 p1 | 0.38 (0.01) | 116.18 (0.04) | 1.98 (0.09) | 1.99 (0.03) G024.33+00.11 MM9 p2 | 0.45 (0.01) | 119.28 (0.03) | 1.90 (0.08) | G024.33+00.11 MM11 | 0.59 (0.02) | 112.99 (0.04) | 2.4 (0.1) | 1.60 (0.06) G024.60+00.08 MM1 | 0.93 (0.02) | 52.99 (0.03) | 2.37 (0.07) | 2.61 (0.04) G024.60+00.08 MM2 | 0.34 (0.01) | 115.11 (0.04) | 2.3 (0.1) | 0.92 (0.04) G025.04-00.20 MM1 | 1.18 (0.03) | 63.53 (0.02) | 2.08 (0.06) | 2.87 (0.04) G025.04-00.20 MM2 | 0.86 (0.02) | 63.44 (0.03) | 2.56 (0.08) | 2.48 (0.04) G025.04-00.20 MM4 | 1.00 (0.03) | 63.55 (0.03) | 2.03 (0.06) | 2.31 (0.04) G034.43+00.24 MM1 | 1.77 (0.04) | 57.79 (0.04) | 3.39 (0.09) | 6.83 (0.03) G034.43+00.24 MM2 | 1.58 (0.04) | 57.50 (0.04) | 3.8 (0.1) | 6.70 (0.03) G034.43+00.24 MM3 | 1.16 (0.03) | 58.98 (0.03) | 2.56 (0.07) | 3.62 (0.04) G034.43+00.24 MM4 | 1.43 (0.03) | 57.50 (0.04) | 3.28 (0.09) | 5.32 (0.04) G034.43+00.24 MM5 | 0.73 (0.03) | 57.57 (0.04) | 2.3 (0.1) | 1.97 (0.06) G034.43+00.24 MM6 | 0.85 (0.03) | 58.31 (0.04) | 2.59 (0.09) | 2.55 (0.06) G034.43+00.24 MM8 | 1.29 (0.03) | 57.84 (0.04) | 3.14 (0.09) | 4.62 (0.06) G034.43+00.24 MM9 | 1.10 (0.03) | 58.53 (0.04) | 2.62 (0.09) | 3.28 (0.06) I18102-1800 MM1 | 1.63 (0.04) | 21.53 (0.04) | 2.88 (0.08) | 5.61 (0.04) I18151-1208 MM1 | 0.58 (0.02) | 32.84 (0.03) | 2.37 (0.08) | 1.59 (0.04) I18151-1208 MM2 | 0.45 (0.01) | 29.93 (0.04) | 2.9 (0.1) | 1.48 (0.04) I18151-1208 MM3 | 0.23 (0.01) | 30.86 (0.06) | 2.1 (0.1) | 0.42 (0.04) I18182-1433 MM1 | 0.75 (0.02) | 59.72 (0.04) | 3.18 (0.09) | 2.59 (0.04) I18182-1433 MM2 | 0.53 (0.02) | 40.93 (0.04) | 2.08 (0.09) | 1.22 (0.05) I18223-1243 MM1 | 0.93 (0.03) | 45.28 (0.03) | 1.87 (0.07) | 3.03 (0.03) I18223-1243 MM2 | 0.82 (0.03) | 45.84 (0.05) | 2.9 (0.1) | 2.16 (0.06) I18223-1243 MM3 | 0.92 (0.02) | 45.66 (0.04) | 2.95 (0.09) | 2.62 (0.06) I18223-1243 MM4 | 0.62 (0.02) | 45.78 (0.03) | 1.73 (0.08) | 1.34 (0.06) I18306-0835 MM1 | 0.91 (0.02) | 77.88 (0.03) | 2.57 (0.08) | 2.75 (0.04) I18306-0835 MM2 | 0.29 (0.02) | 76.58 (0.07) | 2.3 (0.2) | 0.82 (0.06) I18306-0835 MM3 | 0.37 (0.02) | 54.45 (0.06) | 2.1 (0.1) | 0.73 (0.06) I18337-0743 MM1 | 1.38 (0.04) | 58.37 (0.05) | 3.7 (0.1) | 5.68 (0.04) I18337-0743 MM2 | 0.32 (0.01) | 58.53 (0.07) | 3.2 (0.2) | 1.15 (0.04) I18337-0743 MM3 | 0.45 (0.02) | 55.85 (0.05) | 3.2 (0.1) | 1.67 (0.05) Table 8: Line parameters of the NH3 ($J$, $K$) = (2, 2) line.aaThe quoted errors denote one standard deviation. Source | $T_{\rm pk}$ | $V_{LSR}$ | $\Delta V$ | $\int T_{a}^{*}dV$bbVelocity range for integration is $\pm 5$ km s-1 centered at the $LSR$ velocity of N2H+ $J$=1–0 or NH3 (1, 1). ---|---|---|---|--- | [K] | [km s-1] | [km s-1] | [K km s-1] G015.05+00.07 MM1 | 0.18 (0.01) | 24.60 (0.08) | 2.1 (0.2) | 0.48 (0.05) G015.31-00.16 MM2 | $<$0.09 | — | — | $<$0.14 G015.31-00.16 MM3 | $<$0.08 | — | — | $<$0.13 G019.27+0.07 MM1 | 0.43 (0.01) | 26.37 (0.03) | 2.24 (0.07) | 1.08 (0.04) G019.27+0.07 MM2 | 0.40 (0.01) | 26.92 (0.04) | 2.44 (0.09) | 1.19 (0.04) G022.35+00.41 MM1 | 0.17 (0.01) | 52.57 (0.07) | 2.0 (0.2) | 0.43 (0.04) G022.35+00.41 MM2 | $<$0.08 | — | — | 0.27 (0.05) G022.35+00.41 MM3 | 0.30 (0.01) | 84.32 (0.03) | 2.1 (0.1) | 0.81 (0.05) G023.60+00.00 MM1 | 0.31 (0.01) | 106.37 (0.04) | 3.3 (0.1) | 1.11 (0.03) G023.60+00.00 MM2 | 0.46 (0.01) | 53.47 (0.03) | 2.62 (0.07) | 1.47 (0.04) G023.60+00.00 MM3 | 0.32 (0.01) | 105.73 (0.06) | 3.8 (0.1) | 1.29 (0.04) G023.60+00.00 MM4 | 0.30 (0.01) | 53.44 (0.04) | 2.2 (0.1) | 0.71 (0.04) G023.60+00.00 MM7 | 0.33 (0.01) | 53.6 (0.04) | 2.0 (0.1) | 0.73 (0.05) G024.08+00.04 MM1 | 0.30 (0.01) | 113.66 (0.05) | 2.5 (0.1) | 0.87 (0.04) G024.08+00.04 MM2(1) | 0.12 (0.02) | 51.75 (0.08) | 1.1 (0.2) | 0.12 (0.04) G024.08+00.04 MM2(2) | 0.17 (0.01) | 114.17 (0.07) | 2.1 (0.2) | 0.38 (0.04) G024.08+00.04 MM3 | 0.12 (0.01) | 51.62 (0.09) | 1.7 (0.2) | 0.16 (0.04) G024.08+00.04 MM4 | 0.15 (0.01) | 51.63 (0.07) | 1.5 (0.2) | 0.23 (0.04) G024.33+00.11 MM1 | 0.84 (0.02) | 113.46 (0.03) | 3.04 (0.07) | 2.84 (0.06) G024.33+00.11 MM2 | 0.34 (0.01) | 118.00 (0.06) | 3.1 (0.2) | 1.05 (0.06) G024.33+00.11 MM3 p1 | 0.18 (0.01) | 116.9 (0.1) | 2.4 (0.3) | 0.62 (0.06) G024.33+00.11 MM3 p2 | 0.17 (0.02) | 119.24 (0.08) | 1.2 (0.2) | G024.33+00.11 MM4 | 0.16 (0.01) | 114.61 (0.2) | 4.8 (0.4) | 0.80 (0.05) G024.33+00.11 MM5 | 0.47 (0.01) | 117.4 (0.05) | 2.9 (0.1) | 1.28 (0.05) G024.33+00.11 MM6 | 0.62 (0.02) | 113.76 (0.04) | 3.02 (0.09) | 2.08 (0.06) G024.33+00.11 MM8 | 0.10 (0.01) | 119.4 (0.1) | 2.0 (0.3) | 0.41 (0.03) G024.33+00.11 MM9 p1 | 0.13 (0.01) | 116.26 (0.07) | 2.2 (0.2) | 0.62 (0.03) G024.33+00.11 MM9 p2 | 0.19 (0.01) | 119.22 (0.04) | 1.4 (0.1) | G024.33+00.11 MM11 | 0.24 (0.02) | 112.91 (0.09) | 2.5 (0.2) | 0.67 (0.06) G024.60+00.08 MM1 | 0.37 (0.01) | 52.98 (0.04) | 2.4 (0.1) | 1.10 (0.04) G024.60+00.08 MM2 | 0.14 (0.01) | 114.7 (0.1) | 3.4 (0.3) | 0.55 (0.04) G025.04-00.20 MM1 | 0.44 (0.01) | 63.6 (0.03) | 2.18 (0.08) | 1.12 (0.04) G025.04-00.20 MM2 | 0.25 (0.01) | 63.52 (0.07) | 2.9 (0.2) | 0.88 (0.05) G025.04-00.20 MM4 | 0.28 (0.01) | 63.61 (0.05) | 2.1 (0.1) | 0.64 (0.05) G034.43+00.24 MM1 | 1.02 (0.01) | 57.71 (0.02) | 3.54 (0.04) | 4.06 (0.04) G034.43+00.24 MM2 | 0.93 (0.01) | 57.53 (0.02) | 3.66 (0.05) | 3.73 (0.04) G034.43+00.24 MM3 | 0.46 (0.01) | 58.97 (0.03) | 2.91 (0.08) | 1.60 (0.04) G034.43+00.24 MM4 | 0.76 (0.01) | 57.49 (0.02) | 3.17 (0.05) | 2.64 (0.04) G034.43+00.24 MM5 | 0.28 (0.02) | 57.59 (0.07) | 1.9 (0.2) | 0.57 (0.06) G034.43+00.24 MM6 | 0.27 (0.02) | 58.62 (0.08) | 2.5 (0.2) | 0.85 (0.07) G034.43+00.24 MM8 | 0.63 (0.02) | 57.94 (0.04) | 3.3 (0.1) | 2.37 (0.07) G034.43+00.24 MM9 | 0.36 (0.02) | 58.99 (0.06) | 2.3 (0.1) | 0.82 (0.07) I18102-1800 MM1 | 0.93 (0.01) | 21.57 (0.02) | 3.04 (0.05) | 3.28 (0.05) I18151-1208 MM1 | 0.34 (0.01) | 32.92 (0.04) | 2.2 (0.1) | 0.90 (0.05) I18151-1208 MM2 | 0.20 (0.01) | 30.0 (0.1) | 4.2 (0.2) | 0.97 (0.05) I18151-1208 MM3 | 0.11 (0.01) | 31.2 (0.2) | 3.6 (0.4) | 0.45 (0.04) I18182-1433 MM1 | 0.39 (0.01) | 59.69 (0.04) | 3.0 (0.1) | 1.15 (0.04) I18182-1433 MM2 | 0.17 (0.01) | 40.9 (0.1) | 2.6 (0.2) | 0.42 (0.05) I18223-1243 MM1 | 0.48 (0.02) | 45.20 (0.04) | 2.2 (0.1) | 1.21 (0.04) I18223-1243 MM2 | 0.33 (0.02) | 45.72 (0.07) | 2.8 (0.2) | 1.28 (0.06) I18223-1243 MM3 | 0.42 (0.01) | 45.69 (0.03) | 2.57 (0.07) | 0.93 (0.06) I18223-1243 MM4 | 0.23 (0.02) | 45.75 (0.08) | 1.8 (0.2) | 0.56 (0.06) I18306-0835 MM1 | 0.49 (0.01) | 77.87 (0.04) | 2.82 (0.08) | 1.56 (0.04) I18306-0835 MM2 | 0.12 (0.02) | 76.7 (0.2) | 2.5 (0.4) | 0.23 (0.06) I18306-0835 MM3 | 0.13 (0.02) | 55.2 (0.2) | 2.3 (0.4) | 0.30 (0.06) I18337-0743 MM1 | 0.77 (0.01) | 58.37 (0.03) | 3.52 (0.06) | 2.92 (0.05) I18337-0743 MM2 | 0.18 (0.01) | 59.99 (0.09) | 2.4 (0.2) | 0.37 (0.05) I18337-0743 MM3 | 0.16 (0.01) | 56.3 (0.1) | 2.9 (0.3) | 0.59 (0.05) Table 9: Line parameters of the NH3 ($J$, $K$) = (3, 3) line.aaThe quoted errors denote one standard deviation. Source | $T_{\rm pk}$ | $V_{LSR}$ | $\Delta V$ | $\int T_{a}^{*}dV$bbVelocity range for integration is $\pm 5$ km s-1 centered at the $LSR$ velocity of N2H+ $J$=1–0 or NH3 (1, 1). ---|---|---|---|--- | [K] | [km s-1] | [km s-1] | [K km s-1] G015.05+00.07 MM1 | $<$0.08 | — | — | 0.34 (0.04) G015.31-00.16 MM2 | $<$0.08 | — | — | $<$0.13 G015.31-00.16 MM3 | $<$0.08 | — | — | $<$0.12 G019.27+0.07 MM1 | 0.16 (0.01) | 26.6 (0.1) | 4.0 (0.3) | 0.69 (0.04) G019.27+0.07 MM2 | 0.16 (0.01) | 27.0 (0.1) | 3.9 (0.3) | 0.77 (0.04) G022.35+00.41 MM1 | 0.15 (0.01) | 52.97 (0.07) | 4.5 (0.2) | 0.86 (0.04) G022.35+00.41 MM2 | $<$0.08 | — | — | $<$0.12 G022.35+00.41 MM3 | $<$0.08 | — | — | 0.24 (0.04) G023.60+00.00 MM1 | 0.19 (0.01) | 106.56 (0.09) | 5.0 (0.2) | 0.99 (0.03) G023.60+00.00 MM2 | 0.23 (0.01) | 53.51 (0.08) | 5.2 (0.2) | 1.29 (0.03) G023.60+00.00 MM3 | 0.17 (0.01) | 106.3 (0.1) | 4.3 (0.3) | 0.82 (0.04) G023.60+00.00 MM4 | 0.19 (0.01) | 53.46 (0.09) | 2.9 (0.2) | 0.72 (0.05) G023.60+00.00 MM7 | 0.14 (0.01) | 53.3 (0.1) | 4.2 (0.3) | 0.72 (0.04) G024.08+00.04 MM1 | 0.11 (0.01) | 113.9 (0.1) | 2.7 (0.3) | 0.28 (0.03) G024.08+00.04 MM2(1) | $<$0.07 | — | — | $<$0.11 G024.08+00.04 MM2(2) | $<$0.07 | — | — | $<$0.11 G024.08+00.04 MM3 | $<$0.07 | — | — | 0.13 (0.04) G024.08+00.04 MM4 | $<$0.07 | — | — | $<$0.11 G024.33+00.11 MM1 | 0.70 (0.01) | 113.81 (0.04) | 4.2 (0.1) | 3.30 (0.06) G024.33+00.11 MM2 | 0.18 (0.01) | 117.6 (0.2) | 5.6 (0.4) | 1.01 (0.05) G024.33+00.11 MM3 | 0.12 (0.01) | 118.2 (0.2) | 5.7 (0.5) | 0.76 (0.05) G024.33+00.11 MM4 | 0.14 (0.02) | 114.9 (0.1) | 2.1 (0.3) | 0.31 (0.05) G024.33+00.11 MM5 | 0.19 (0.01) | 119.1 (0.1) | 5.0 (0.3) | 1.02 (0.05) G024.33+00.11 MM6 | 0.48 (0.01) | 113.98 (0.05) | 3.7 (0.1) | 2.07 (0.06) G024.33+00.11 MM8 | $<$0.07 | — | — | 0.19 (0.03) G024.33+00.11 MM9 | $<$0.07 | — | — | 0.26 (0.03) G024.33+00.11 MM11 | $<$0.12 | — | — | 0.36 (0.06) G024.60+00.08 MM1 | 0.11 (0.01) | 52.2 (0.2) | 7.6 (0.5) | 0.78 (0.04) G024.60+00.08 MM2 | $<$0.08 | — | — | 0.41 (0.04) G025.04-00.20 MM1 | 0.13 (0.01) | 63.2 (0.1) | 3.0 (0.3) | 0.44 (0.04) G025.04-00.20 MM2 | 0.10 (0.01) | 63.7 (0.2) | 3.4 (0.4) | 0.41 (0.04) G025.04-00.20 MM4 | $<$0.08 | — | — | 0.18 (0.04) G034.43+00.24 MM1 | 0.66 (0.01) | 57.9 (0.03) | 5.17 (0.08) | 3.58 (0.04) G034.43+00.24 MM2 | 0.59 (0.01) | 57.65 (0.04) | 5.76 (0.09) | 3.45 (0.04) G034.43+00.24 MM3 | 0.215 (0.006) | 59.4 (0.1) | 7.2 (0.2) | 1.42 (0.04) G034.43+00.24 MM4 | 0.399 (0.007) | 57.55 (0.05) | 5.2 (0.1) | 2.15 (0.04) G034.43+00.24 MM5 | $<$0.12 | — | — | 0.39 (0.06) G034.43+00.24 MM6 | $<$0.12 | — | — | 0.33 (0.06) G034.43+00.24 MM8 | 0.33 (0.02) | 57.8 (0.1) | 4.3 (0.2) | 1.53 (0.07) G034.43+00.24 MM9 | 0.19 (0.02) | 59.2 (0.1) | 2.9 (0.3) | 0.64 (0.07) I18102-1800 MM1 | 0.61 (0.01) | 21.47 (0.03) | 4.71 (0.08) | 3.01 (0.04) I18151-1208 MM1 | 0.14 (0.01) | 32.8 (0.1) | 3.4 (0.3) | 0.58 (0.04) I18151-1208 MM2 | 0.11 (0.01) | 29.4 (0.2) | 6.8 (0.5) | 0.72 (0.05) I18151-1208 MM3 | $<$0.08 | — | — | 0.29 (0.04) I18182-1433 MM1 | 0.17 (0.01) | 59.4 (0.1) | 6 (0.3) | 1.02 (0.04) I18182-1433 MM2 | $<$0.09 | — | — | 0.19 (0.05) I18223-1243 MM1 | 0.16 (0.02) | 45.2 (0.1) | 1.9 (0.3) | 0.36 (0.06) I18223-1243 MM2 | $<$0.12 | — | — | 0.32 (0.06) I18223-1243 MM3 | 0.15 (0.01) | 45.1 (0.1) | 5.3 (0.3) | 0.8 (0.03) I18223-1243 MM4 | $<$0.12 | — | — | 0.24 (0.07) I18306-0835 MM1 | 0.31 (0.01) | 78.04 (0.06) | 3.5 (0.1) | 1.22 (0.04) I18306-0835 MM2 | 0.09 (0.02) | 77.6 (0.2) | 2.8 (0.5) | 0.3 (0.06) I18306-0835 MM3 | $<$0.11 | — | — | $<$0.18 I18337-0743 MM1 | 0.39 (0.01) | 58.29 (0.06) | 4.8 (0.1) | 1.99 (0.05) I18337-0743 MM2 | 0.13 (0.01) | 58.3 (0.1) | 2.5 (0.3) | 0.32 (0.04) I18337-0743 MM3 | $<$0.08 | — | — | 0.67 (0.04) Table 10: The NH3 rotation temperatures and column densities. Source | $T_{\rm rot}$(NH3) | $N$(N2H+)aaTex(N2H+) is assumed to be equal to $T_{\rm ex}$(NH3). The quoted error is dominated by the error in $T_{\rm ex}$(NH3). | $N$(CCS)aaTex(N2H+) is assumed to be equal to $T_{\rm ex}$(NH3). The quoted error is dominated by the error in $T_{\rm ex}$(NH3). | $N$(HC3N)aaTex(N2H+) is assumed to be equal to $T_{\rm ex}$(NH3). The quoted error is dominated by the error in $T_{\rm ex}$(NH3). | $N$(NH3) | $N$(CH3OH)bbTex(CH3OH) is assumed to be from 20 K to 50 K. | $N$(CCS)/$N$(N2H+) | $N$(CCS)/$N$(NH3) ---|---|---|---|---|---|---|---|--- | [K] | [1012 cm-2] | [1012 cm-2] | [1012 cm-2] | [1015 cm-2] | [1014 cm-5] | | [10-2] G015.05+00.07 MM1 | 15.2${}_{-1.5}^{+1.7}$ | 17.4${}_{-1.0}^{+1.2}$ | $<$3.8 | 9.7${}_{-0.4}^{+0.5}$ | 2.7${}_{-0.8}^{+0.6}$ | $<$2.2 | $<$0.24 | $<$0.20 G015.31-00.16 MM2 | 10.3${}_{-0.6}^{+0.7}$ | 5.4${}_{-0.1}^{+0.1}$ | $<$1.6 | 5.0${}_{-0.1}^{+0.1}$ | 1.1${}_{-0.6}^{+0.5}$ | $<$2.3 | $<$0.30 | $<$0.32 G015.31-00.16 MM3 | 12.6${}_{-2.0}^{+2.4}$ | 6.9${}_{-0.5}^{+0.7}$ | $<$1.7 | 4.4${}_{-0.2}^{+0.3}$ | 0.6${}_{-0.5}^{+0.4}$ | $<$3.1 | $<$0.27 | $<$1.74 G019.27+0.07 MM1 | 15.4${}_{-1.0}^{+1.1}$ | 52.0${}_{-2.1}^{+2.4}$ | $<$3.7 | 24.2${}_{-0.7}^{+0.8}$ | 3.1${}_{-0.6}^{+0.6}$ | 7.9—16 | $<$0.07 | $<$0.15 G019.27+0.07 MM2 | 15.8${}_{-0.9}^{+1.0}$ | 29.2${}_{-1.1}^{+1.2}$ | $<$3.5 | 22.1${}_{-0.6}^{+0.7}$ | 3.0${}_{-0.6}^{+0.5}$ | $<$2.2 | $<$0.12 | $<$0.15 G022.35+00.41 MM1 | 16.0${}_{-1.4}^{+1.6}$ | 26.7${}_{-1.4}^{+1.7}$ | $<$3.6 | 18.3${}_{-0.7}^{+0.8}$ | 3.3${}_{-0.8}^{+0.7}$ | 9.2—18.7 | $<$0.14 | $<$0.15 G022.35+00.41 MM2 | 12.8${}_{-1.7}^{+2.2}$ | 4.9${}_{-0.3}^{+0.4}$ | $<$1.7 | $<$2.2 | 1.3${}_{-0.9}^{+0.6}$ | $<$2.6 | $<$0.37 | $<$0.42 G022.35+00.41 MM3 | 15.2${}_{-1.0}^{+1.1}$ | 15.8${}_{-0.6}^{+0.7}$ | $<$3.0 | 5.5${}_{-0.1}^{+0.2}$ | 2.2${}_{-0.5}^{+0.5}$ | $<$2.1 | $<$0.20 | $<$0.18 G023.60+00.00 MM1 | 17.0${}_{-1.2}^{+1.3}$ | 36.7${}_{-1.7}^{+1.9}$ | $<$5.2 | 22.2${}_{-0.7}^{+0.8}$ | 4.3${}_{-0.7}^{+0.7}$ | 8.9—18 | $<$0.15 | $<$0.14 G023.60+00.00 MM2 | 16.0${}_{-0.9}^{+1.0}$ | 22.8${}_{-0.8}^{+0.9}$ | $<$3.8 | 31.6${}_{-0.8}^{+0.9}$ | 2.5${}_{-0.5}^{+0.5}$ | 5.5—11.2 | $<$0.17 | $<$0.19 G023.60+00.00 MM3 | 16.4${}_{-1.1}^{+1.2}$ | 17.5${}_{-0.7}^{+0.8}$ | $<$4.8 | 3.9${}_{-0.1}^{+0.1}$ | 2.6${}_{-0.7}^{+0.6}$ | $<$1.9 | $<$0.28 | $<$0.25 G023.60+00.00 MM4 | 14.9${}_{-0.9}^{+1.1}$ | 8.6${}_{-0.3}^{+0.4}$ | $<$3.5 | 7.0${}_{-0.2}^{+0.2}$ | 2.1${}_{-0.5}^{+0.5}$ | $<$2.0 | $<$0.42 | $<$0.22 G023.60+00.00 MM7 | 13.1${}_{-0.7}^{+0.8}$ | 16.0${}_{-0.5}^{+0.5}$ | $<$2.6 | 13.6${}_{-0.2}^{+0.3}$ | 2.7${}_{-0.6}^{+0.5}$ | $<$3.0 | $<$0.17 | $<$0.13 G024.08+00.04 MM1 | 16.4${}_{-1.4}^{+1.6}$ | 38.9${}_{-2.1}^{+2.4}$ | $<$3.8 | 8.5${}_{-0.3}^{+0.4}$ | 4.0${}_{-0.9}^{+0.7}$ | $<$3.6 | $<$0.10 | $<$0.12 G024.08+00.04 MM2(1) | 14.9${}_{-2.6}^{+3.4}$ | 8.1${}_{-0.8}^{+1.1}$ | $<$2.6 | 4.0${}_{-0.3}^{+0.4}$ | 1.3${}_{-0.9}^{+0.7}$ | $<$3.6 | $<$0.36 | $<$0.65 G024.08+00.04 MM2(2) | 15.1${}_{-1.6}^{+1.9}$ | 10.5${}_{-0.7}^{+0.8}$ | $<$3.1 | 3.5${}_{-0.1}^{+0.2}$ | 2.8${}_{-0.9}^{+0.7}$ | $<$3.6 | $<$0.32 | $<$0.16 G024.08+00.04 MM3 | 12.1${}_{-0.9}^{+1.0}$ | 5.1${}_{-0.2}^{+0.2}$ | $<$3.1 | 6.7${}_{-0.1}^{+0.2}$ | 2.0${}_{-0.6}^{+0.5}$ | $<$2.4 | $<$0.63 | $<$0.22 G024.08+00.04 MM4 | 12.9${}_{-0.9}^{+1.1}$ | 4.6${}_{-0.2}^{+0.2}$ | $<$1.7 | 7.2${}_{-0.1}^{+0.2}$ | 1.7${}_{-0.5}^{+0.4}$ | $<$3.0 | $<$0.39 | $<$0.14 G024.33+00.11 MM1 | 18.8${}_{-1.6}^{+1.8}$ | 63.3${}_{-3.6}^{+4.1}$ | $<$6.6 | 33.0${}_{-1.4}^{+1.7}$ | 5.0${}_{-0.8}^{+0.7}$ | 7.7—15.6 | $<$0.11 | $<$0.16 G024.33+00.11 MM2 | 16.6${}_{-1.2}^{+1.3}$ | 20.8${}_{-0.9}^{+1.1}$ | $<$5.1 | 12.2${}_{-0.4}^{+0.5}$ | 3.4${}_{-0.8}^{+0.7}$ | $<$2.6 | $<$0.26 | $<$0.20 G024.33+00.11 MM3 p1 | 13.9${}_{-0.7}^{+0.7}$ | 21.9${}_{-0.6}^{+0.6}$ | $<$5.5 | 11.7${}_{-0.2}^{+0.2}$ | 2.1${}_{-0.4}^{+0.3}$ | $<$1.5 | $<$0.26 | $<$0.32 G024.33+00.11 MM3 p2 | 14.5${}_{-1.3}^{+1.4}$ | | | | 1.8${}_{-0.4}^{+0.3}$ | | | G024.33+00.11 MM4 | 13.4${}_{-0.9}^{+1.0}$ | 11.5${}_{-0.4}^{+0.5}$ | $<$4.6 | 9.8${}_{-0.2}^{+0.3}$ | 4.2${}_{-1.1}^{+0.9}$ | $<$2.1 | $<$0.42 | $<$0.15 G024.33+00.11 MM5 | 17.1${}_{-0.5}^{+0.5}$ | 23.7${}_{-0.5}^{+0.5}$ | $<$4.7 | 14.4${}_{-0.2}^{+0.2}$ | 2.7${}_{-0.2}^{+0.2}$ | $<$2.2 | $<$0.20 | $<$0.19 G024.33+00.11 MM6 | 17.3${}_{-1.4}^{+1.6}$ | 14.3${}_{-0.8}^{+0.9}$ | $<$5.7 | 7.2${}_{-0.3}^{+0.3}$ | 4.7${}_{-0.8}^{+0.7}$ | $<$2.8 | $<$0.42 | $<$0.15 G024.33+00.11 MM8 | — | — | — | — | — | $<$1.9 | — | — G024.33+00.11 MM9 p1 | 15.1${}_{-1.1}^{+1.2}$ | | | | | | | G024.33+00.11 MM9 p2 | 15.2${}_{-0.9}^{+1.0}$ | 7.7${}_{-0.3}^{+0.3}$ | $<$3.9 | 4.6${}_{-0.1}^{+0.1}$ | 2.2${}_{-0.4}^{+0.3}$ | $<$2.3 | $<$0.52 | $<$0.21 G024.33+00.11 MM11 | 14.8${}_{-1.4}^{+1.7}$ | 8.6${}_{-0.5}^{+0.6}$ | $<$4.2 | $<$2.5 | 2.7${}_{-0.8}^{+0.6}$ | $<$2.0 | $<$0.51 | $<$0.22 G024.60+00.08 MM1 | 15.3${}_{-0.8}^{+0.9}$ | 28.5${}_{-0.9}^{+1.0}$ | $<$3.2 | 24.0${}_{-0.5}^{+0.6}$ | 2.1${}_{-0.5}^{+0.4}$ | 4.9—9.9 | $<$0.12 | $<$0.20 G024.60+00.08 MM2 | 14.2${}_{-1.1}^{+1.2}$ | 25.9${}_{-1.1}^{+1.3}$ | $<$4.2 | 11.1${}_{-0.3}^{+0.4}$ | 3.1${}_{-0.6}^{+0.5}$ | $<$3.0 | $<$0.17 | $<$0.17 G025.04-00.20 MM1 | 14.7${}_{-0.8}^{+0.9}$ | 27.9${}_{-0.9}^{+1.0}$ | $<$3.6 | 15.2${}_{-0.3}^{+0.4}$ | 2.0${}_{-0.4}^{+0.4}$ | $<$2.5 | $<$0.13 | $<$0.22 G025.04-00.20 MM2 | 13.1${}_{-0.7}^{+0.8}$ | 21.5${}_{-0.6}^{+0.7}$ | $<$4.5 | 10.5${}_{-0.2}^{+0.2}$ | 2.7${}_{-0.6}^{+0.5}$ | $<$2.1 | $<$0.21 | $<$0.21 G025.04-00.20 MM4 | 13.5${}_{-0.8}^{+0.9}$ | 16.2${}_{-0.5}^{+0.6}$ | $<$3.3 | 5.9${}_{-0.1}^{+0.1}$ | 1.7${}_{-0.5}^{+0.4}$ | $<$2.5 | $<$0.21 | $<$0.27 G034.43+00.24 MM1 | 18.5${}_{-1.0}^{+1.1}$ | 133.6${}_{-5.0}^{+5.4}$ | $<$3.9 | 54.3${}_{-1.6}^{+1.7}$ | 3.4${}_{-0.6}^{+0.6}$ | 24.2—49.2 | $<$0.03 | $<$0.14 G034.43+00.24 MM2 | 18.8${}_{-1.0}^{+1.1}$ | 113.9${}_{-4.2}^{+4.5}$ | $<$4.5 | 52.0${}_{-1.5}^{+1.6}$ | 3.8${}_{-0.6}^{+0.6}$ | 18.7—37.9 | $<$0.04 | $<$0.14 G034.43+00.24 MM3 | 15.5${}_{-0.9}^{+0.9}$ | 59.6${}_{-2.0}^{+2.2}$ | $<$3.2 | 30.6${}_{-0.7}^{+0.8}$ | 2.2${}_{-0.5}^{+0.4}$ | 13.9—28.1 | $<$0.06 | $<$0.19 G034.43+00.24 MM4 | 17.6${}_{-1.0}^{+1.1}$ | 66.0${}_{-2.4}^{+2.7}$ | $<$3.4 | 29.9${}_{-0.8}^{+0.9}$ | 3.3${}_{-0.6}^{+0.6}$ | 6.4—13 | $<$0.05 | $<$0.13 G034.43+00.24 MM5 | 14.3${}_{-1.3}^{+1.5}$ | 16.8${}_{-0.9}^{+1.0}$ | $<$3.5 | 6.1${}_{-0.2}^{+0.3}$ | 2.7${}_{-0.8}^{+0.6}$ | $<$2.3 | $<$0.22 | $<$0.18 G034.43+00.24 MM6 | 14.0${}_{-1.1}^{+1.3}$ | 15.4${}_{-0.7}^{+0.8}$ | $<$3.2 | 11.2${}_{-0.3}^{+0.4}$ | 2.3${}_{-0.7}^{+0.5}$ | $<$2.4 | $<$0.22 | $<$0.20 G034.43+00.24 MM8 | 17.2${}_{-1.1}^{+1.2}$ | 14.0${}_{-0.6}^{+0.6}$ | $<$4.4 | 10.7${}_{-0.3}^{+0.4}$ | 2.8${}_{-0.6}^{+0.5}$ | $<$2.1 | $<$0.33 | $<$0.20 G034.43+00.24 MM9 | 13.9${}_{-0.9}^{+1.1}$ | 20.1${}_{-0.8}^{+0.9}$ | $<$3.2 | 11.7${}_{-0.3}^{+0.3}$ | 2.5${}_{-0.6}^{+0.5}$ | $<$2.4 | $<$0.16 | $<$0.17 I18102-1800 MM1 | 17.9${}_{-1.1}^{+1.2}$ | 51.2${}_{-2.1}^{+2.3}$ | $<$6.6 | 41.2${}_{-1.2}^{+1.4}$ | 3.2${}_{-0.6}^{+0.5}$ | 10.7—21.7 | $<$0.13 | $<$0.25 I18151-1208 MM1 | 20.8${}_{-1.7}^{+1.8}$ | 34.9${}_{-2.1}^{+2.2}$ | $<$5.3 | 15.0${}_{-0.7}^{+0.8}$ | 1.3${}_{-0.5}^{+0.5}$ | 2.9—5.9 | $<$0.16 | $<$0.66 I18151-1208 MM2 | 17.8${}_{-1.4}^{+1.5}$ | 45.7${}_{-2.4}^{+2.6}$ | $<$5.8 | 9.3${}_{-0.4}^{+0.4}$ | 1.5${}_{-0.6}^{+0.6}$ | 8.4—17 | $<$0.13 | $<$0.65 I18151-1208 MM3 | 16.0${}_{-2.1}^{+2.7}$ | 13.2${}_{-1.1}^{+1.4}$ | $<$2.8 | 3.6${}_{-0.2}^{+0.3}$ | 2.4${}_{-1.0}^{+0.8}$ | $<$2.1 | $<$0.23 | $<$0.20 I18182-1433 MM1 | 19.0${}_{-1.4}^{+1.5}$ | 55.8${}_{-2.7}^{+3.0}$ | $<$5.8 | 31.6${}_{-1.2}^{+1.3}$ | 1.9${}_{-0.6}^{+0.6}$ | 5.1—10.3 | $<$0.11 | $<$0.45 I18182-1433 MM2 | 13.8${}_{-1.1}^{+1.2}$ | 17.0${}_{-0.7}^{+0.8}$ | $<$3.8 | 3.4${}_{-0.1}^{+0.1}$ | 2.0${}_{-0.6}^{+0.5}$ | $<$3.0 | $<$0.23 | $<$0.27 I18223-1243 MM1 | 17.5${}_{-1.5}^{+1.6}$ | 42.4${}_{-2.3}^{+2.6}$ | $<$3.8 | 10.7${}_{-0.4}^{+0.5}$ | 1.8${}_{-0.4}^{+0.4}$ | $<$3.3 | $<$0.09 | $<$0.27 I18223-1243 MM2 | 15.1${}_{-1.3}^{+1.4}$ | 25.2${}_{-1.2}^{+1.5}$ | $<$3.8 | 6.3${}_{-0.2}^{+0.3}$ | 2.9${}_{-0.7}^{+0.6}$ | $<$3.3 | $<$0.16 | $<$0.17 I18223-1243 MM3 | 16.2${}_{-0.9}^{+1.0}$ | 43.9${}_{-1.6}^{+1.7}$ | $<$4.3 | 15.1${}_{-0.4}^{+0.4}$ | 2.9${}_{-0.6}^{+0.5}$ | 3.8—7.7 | $<$0.10 | $<$0.19 I18223-1243 MM4 | 15.5${}_{-1.5}^{+1.7}$ | 14.6${}_{-0.9}^{+1.0}$ | $<$3.3 | 6.6${}_{-0.3}^{+0.3}$ | 1.2${}_{-0.5}^{+0.4}$ | $<$2.2 | $<$0.24 | $<$0.47 I18306-0835 MM1 | 17.3${}_{-1.1}^{+1.2}$ | 22.6${}_{-0.9}^{+1.0}$ | $<$5.9 | 16.6${}_{-0.5}^{+0.6}$ | 2.8${}_{-0.5}^{+0.5}$ | 4.1—8.4 | $<$0.27 | $<$0.26 I18306-0835 MM2 | 16.4${}_{-2.9}^{+3.8}$ | 10.4${}_{-1.2}^{+1.6}$ | $<$5.6 | 4.0${}_{-0.3}^{+0.4}$ | 1.6${}_{-1.2}^{+0.9}$ | $<$3.5 | $<$0.60 | $<$1.39 I18306-0835 MM3 | 13.4${}_{-1.7}^{+2.2}$ | 4.3${}_{-0.3}^{+0.4}$ | $<$4.2 | 2.8${}_{-0.1}^{+0.2}$ | 2.7${}_{-1.1}^{+0.7}$ | $<$3.3 | $<$1.05 | $<$0.26 I18337-0743 MM1 | 17.1${}_{-0.9}^{+1.0}$ | 67.2${}_{-2.4}^{+2.6}$ | $<$4.7 | 41.1${}_{-1.1}^{+1.2}$ | 4.6${}_{-0.7}^{+0.6}$ | 4.8—9.8 | $<$0.07 | $<$0.12 I18337-0743 MM2 | 17.1${}_{-1.8}^{+2.2}$ | 21.5${}_{-1.5}^{+1.8}$ | $<$4.6 | 11.9${}_{-0.6}^{+0.8}$ | 4.1${}_{-1.2}^{+1.0}$ | $<$1.9 | $<$0.23 | $<$0.16 I18337-0743 MM3 | 15.0${}_{-1.4}^{+1.6}$ | 19.1${}_{-1.1}^{+1.3}$ | $<$4.6 | 9.8${}_{-0.4}^{+0.5}$ | 2.5${}_{-0.9}^{+0.8}$ | 3.2—6.5 | $<$0.25 | $<$0.29 Table 11: Integrated Intensities and rotation temperature of the CH3OH $J$=7–6 line.aaThe quoted errors denote one standard deviation. Velocity range for integration is $\pm 5$ km s-1 centered at the $LSR$ velocity of N2H+ $J$=1–0 or NH3 (1, 1). Source | $7_{-1}$–$6_{-1}$ $E$ | $7_{1}$–$6_{1}$ $E$ | $7_{2}$–$6_{2}$ $E$ | $7_{0}$–$6_{0}$ $E$ | $T_{\rm rot}$ ---|---|---|---|---|--- | [K km s-1] | [K km s-1] | [K km s-1] | [K km s-1] | [K] G019.27+0.07 MM1 | 3.2 (0.2) | $<$0.4 | 0.7 (0.1) | 1.7 (0.2) | 7.3 (0.8) G022.35+00.41 MM1 | 2.9 (0.2) | 0.5 (0.1) | 0.7 (0.1) | 1.1 (0.2) | 8.4 (0.3) G023.60+00.00 MM1 | 3.6 (0.2) | $<$0.4 | 0.5 (0.1) | 0.9 (0.2) | 6.9 (0.6) G023.60+00.01 MM2 | 2.4 (0.2) | $<$0.3 | $<$0.3 | $<$0.5 | $<$7.4 G024.33+00.11 MM1 | 2.9 (0.2) | $<$0.4 | 1.0 (0.1) | 2.3 (0.2) | 8.3 (2.4) G024.60+00.08 MM1 | 1.9 (0.2) | $<$0.4 | 0.5 (0.1) | 0.6 (0.2) | 9.2 (1.1) G034.43+00.24 MM1 | 8.7 (0.1) | 2.5 (0.1) | 3.4 (0.1) | 5.4 (0.1) | 10.9 (1.2) G034.43+00.25 MM2 | 6.8 (0.2) | 1.2 (0.1) | 1.6 (0.1) | 3.1 (0.2) | 8.3 (0.5) G034.43+00.26 MM3 | 6.3 (0.1) | 0.7 (0.1) | 1.2 (0.1) | 2.7 (0.1) | 7.0 (0.5) G034.43+00.27 MM4 | 2.8 (0.2) | $<$0.4 | 0.7 (0.1) | $<$0.5 | 9.3 (3.1) I18102-1800 MM1 | 4.5 (0.1) | 0.4 (0.1) | 1.0 (0.1) | 1.5 (0.1) | 7.3 (1.0) I18151-1208 MM1 | 1.6 (0.2) | $<$0.4 | 0.8 (0.1) | 0.9 (0.2) | 12.1 (0.8) I18151-1209 MM2 | 4.3 (0.2) | $<$0.4 | 1.8 (0.1) | 2.7 (0.2) | 8.1 (2.8) I18182-1433 MM1 | 2.0 (0.2) | 0.8 (0.1) | 0.8 (0.1) | 1.5 (0.2) | 12.4 (3.5) I18223-1245 MM3 | 1.3 (0.1) | 0.3 (0.1) | $<$0.3 | 0.4 (0.1) | 9.5 (2.5) I18306-0835 MM1 | 1.8 (0.2) | 0.5 (0.1) | 0.6 (0.1) | 0.9 (0.2) | 10.2 (1.1) I18337-0743 MM1 | 1.8 (0.1) | $<$0.4 | $<$0.4 | 0.5 (0.1) | 8.6 (1.9) I18337-0745 MM3 | 1.7 (0.1) | 0.3 (0.1) | $<$0.3 | 0.6 (0.1) | 8.2 (1.2)

arxiv-papers

2024-09-04T02:48:53.906667

0802.3084

eurm10 msam10 119–126 # Non-Hydrostatic Effects in the Interaction between Flow and Orography IVANGLADICH1 F.STEL2 D.GIAIOTTI2 G.FURLAN3 1 Department of Mathematics, University of Trieste, Italy 2 ARPA-CRMA, via Cairoli 14, I-33057, Palmanova (UD), Italy 3the Abdus Salam International Centre for Theoretical Physics (ICTP) Strada Costiera, 11 I-34014 Trieste Italy (1996; ?? and in revised form ??) ###### Abstract The interaction between flows and orography is a fundamental aspect of theoretical fluid dynamics for its direct applications (e.g., in dynamical meteorology); a comprehensive description is nowadays still lacking in some aspects. In this work, in particular, the authors would like to face the problem of flow-blocking and of the streamlines pattern formation, examining the role of stratification (i.e., Brunt-Vaisala frequency) and Froude number on these problems. In particular this work wants to investigate the role of vertical advection on flow-blocking and on streamlines geometry. The importance of streamlines curvature and stratification for the formation of pressure perturbation, then their role in flow-blocking will be shown. Moreover it will be shown how flow-blocking cannot be easly predict using only a stratification parameter or the Froude number. ††volume: 538 ## 1 Introduction The interaction beetween flow and orography is an important topic of theoretical fluid dynamics because of its direct applications in everyday life. As an example, orographic rain is originated by a moist flow that, interacting with orography, gives rise to a vertical motion, then to condensation and precipitation formation. Even if these phenomena are very common, their explanation is nowadays not complete. Infact a full description requires the knowleadge of the solution of Navier-Stokes equation with complex boundary conditions (i.e. top of troposphere and the orography). The literature facing the interactions between flows and orography can be divided in: i) numerical works, ii) analytical works and iii) experimental works: Riley et al. (1976), Baines (1979), Hunt & Snyder (1980), Castro et al. (1983) and Snyder et al. (1985). Among the numerical studies there are several contributions produced using hydrostatic numerical models Smolarkiewcz & Rotunno (1989) and more recently some contributions realized using non- hydrostatic numerical models (e.g., the Weather, Research and Forecasting model, WRF) Miglietta & Rotunno (2005). The problem with numerical models, both hydrostatic and non-hydrostatic, is that their output is extremely complex then, generally, very difficult to interpret weighting the physical role of every possible parameter used in the tuning of the model Giaiotti et al. (2007). On the contrary analytic works permit to keep a more complete control of the role of each parameter inserted into the analytical model even if some approximations need to be taken to reduce the mathematical difficulty of the starting equations. Concerning the analytical works, several of them make use of the hydrostatic approximation and of different kind of obstacles Lilly & Klemp (1979), using both stratified and rotational fluids Inttyre (1972), using thermal forcing Reisner & Smolarkiewicz (1993) or imposing turbulent boundary layer at the surface Carrunthers & Hunt (1990). Only a few analitical works avoid the use of the hydrostatic approximation, this because the vertical advective term makes the analytic approch more difficult. A comprehensive review of all these works can be found in Baines (1995). Among the analytical works, three of them, Smith (1989a), Wurtele et al. (1987) and Keller (1994) deserve a special mention. In particular Smith (1989a), developing the previous work of Smith (1988) and Smith (1989b), studies the interaction between hydrostatic and stratified flow on an idealized 3-D topography. In his contribution the attention is focused on the case in which the flow stops its upward motion while moving on the topography (i.e., stagnation of the flow). Moreover, in this work, the use of Froude number as a discriminating factor between stagnation and non-stagnation, proposed by Sheppard (1956), is critically reviewed. In the work of Keller (1994) the study of interaction between non-hydrostatic and stratified flows on an idealized 2-D topography is presented but, in this case, the attention is focused on the effects of non-hydrostaticity on the formation of downstream lee waves. Keller (1994) also analizes the behaviour of the flow with different vertical velocity profiles. The interesting aspect of Wurtele et al. (1987), instead, stays in the approach to the gravity wave propagation in stratosphere. Following the line defined by the three analytical works above introduced, the aim of this work is to study analytically the influence of non-hydrostatic effects on the geometry of streamlines, essentially upstream to topography, and on flow blocking. In the § 2 a briefly review of Smith (1989a) is presented and its results are extended to the non-hydrostatic case. Then in § 3, an analytical model for a stratified flow on a 2-D profile is presented adopting a uniform incident velocity profile for the unperturbed flow and using the Fourier trasforms. In this part the mathematical difficulty of antitrasformation is evidenced and a new approach to overtake it is presented. With this approach the integrand is substituted with a new one for which the integration can be easly carried out. In this way, making use of the new integral form, the streamlines pattern for the non-hydrostatic case can be obtained and it is presented in § 4. With this approach the formation of lee waves, already reproduced by Keller (1994), as well as the intensification of wind speed at the top of topography profile is well described by the model. This phenomenon can be obtained even making use of hydrostatic models, but the non-hydrostatic approach showes a minor intensification of wind. Furthermore the changes in the pattern of streamline due to the super-critical (Froude number $\mbox{{F}}>1$) and sub-critical ($\mbox{{F}}<1$) regimes are presented. At the end of this work the relevance of non-hydrostatic effects are shown to be important in the dynamics of flow-blocking, § 5, specially for topographic profiles with horizontal scale comparable with the vertical one. Moreover, thanks to this study, stagnation can be put in relationship with the formation of vorticity as guessed before by Schar & Smith (1993a) and Schar & Smith (1993b). At the end it is demonstrated that simple parameters (e.g., Froude number or Brunt-Vaisala frequency) are not sufficient to describe completely the stagnation mechanism. This result is not merely theorethical but there are some concrete cases in which it might had played a role, as is the case of the Valcanale flood (29 August 2003, Valcanale-UD-, Italy) when two people died. ## 2 From hydrostatic to non-hydrostatic approach In this section the work of Smith (1989) is adapted to the non-hydrostatic case. A 2-D steady and parallel flow with a constant vertical velocity profile is assumed. This flow interacts with an obstacle whose analytic form is $h=h(x,y)$. Differently from Smith (1989a) a bounded flow is here considered. Infact, imposing that all the peturbation on pressure field and streamline displacement due to orography damp at infinite height can create problems to the energy conservation Baines (1995). For this reason, in a more realistic way, we impose that all the orographic effects fade at the top of fluid that, in the atmospheric case coincides with the end of troposphere at $10$ km. As done by Smith (1989a), the flow is here assumed always parallel (i.e., turbulent diffusion is neglected), incompressible and stable stratified with constant Brunt-Vaisala frequency $N$. In this way far from orography the density profile $\rho_{\infty}(z)$ is horizontally homogeneous and given by the following relationship $\rho_{\infty}(z)=\rho_{0}\Big{(}1-\frac{N^{2}(z-z_{0})}{g}\Big{)}$ (1) where $z$ is the vertical coordinate, $z_{0}$ is a reference level, $\rho_{0}$ the density at the reference level far from the obstacles and $g$ is the gravity acceleration. The flow is assumed as composed by dry air in isothermal condition and all the sources and sinks of heat are neglected. The viscous effects and the turbulent diffusion of momentum are neglected as well. Moreover Coriolis force is assumed as null, infact Rossby number is lower than unity for an incident velocity profile of $15$ m s-1and for a horizontal scale smaller than $200$ km. To proceed further it is now necessary to introduce an energy conservation principle. For isentropic flows the energy for unit volume $E$ given by $E=\frac{\rho U^{2}}{2}+{P}+\rho gz$ (2) is constant along stream and vortex lines Batchelor (1994). In (3) $P$ is the pressure and $U$ is the intensity of velocity vector. In this case streamlines coincides with isopicnal lines because the flow is incompressible and steady. For this reason the energy balance written for a parcel on a streamline characterized by the unperturbed level $z_{0}$ (i.e., far from the obstacle) is conserved and has the form $\frac{\rho_{0}{U^{2}}_{0}}{2}+\rho_{0}gz_{0}+P_{0}=\frac{\rho_{0}U^{2}}{2}+\rho_{0}gz+P$ The pressure field far from orography is horizontally homogeneous as previously assumed, then it is possible to assume the hydrostatic balance to describe $P_{\infty}$ $\frac{\partial{P_{\infty}(z)}}{\partial{z}}=-\rho_{\infty}(z)g$ (3) The reasonableness of this assumption is given by the fact that no vertical motion, upstream and far from the obstacle, is assumed. Defining the perturbation pressure $P^{*}$ as the difference between pressure field and pressure field at the same level and far from the obstacles $P^{*}=P-P_{\infty}$ and using the equation (3), the velocity of a parcel along a streamline is given by $U^{2}={U^{2}}_{0}-N^{2}{\eta}^{2}-2\frac{P^{*}}{\rho_{0}}$ (4) where ${U^{2}}_{0}$ is the velocity of an incident upstream vertical profile while $\eta$ is the streamline vertical displacement relative to the unperturbated streamline characterized by the height $z_{0}$ far from the obstacle and density $\rho_{0}$. The equation (4) gives us a simple relationship to identify the flow blocking, i.e. the situation in which $U^{2}=0$. Neglecting the perturbation pressure term, the results of Sheppard (1956) are reproduced. In those results stagnation occours when the displacement is $\mbox{{F}}=\frac{U}{N\eta}<1$ The problem of this approach to flow-blocking is that neglecting pressure perturbation, the parcel behaves as if it were not immersed in a fluid environment. It is then important, for a more realistic description of the flow, try to estimate the pressure term. Starting from the vertical momentum equation Emanuel (1994) for a stratified flow $\frac{Dw}{Dt}=-\frac{1}{\rho_{\infty}}\frac{\partial{P^{*}}}{\partial{z}}-\Big{(}\frac{\rho{{}^{\prime}}}{\rho_{\infty}}\Big{)}g$ (5) where $\rho^{\prime}$ is the density perturbation, i.e. the difference between density field and density field at the same level and far from the obstacles, that is $\rho^{\prime}=\rho(x,y,z)-\rho_{\infty}(z)=\frac{\rho_{0}N^{2}\eta}{g}$ The result of integration of (5) from the general level $z$ up to the top of the fluid $D$ becomes $P^{*}(x,y,z,t)=\Gamma_{h}+\Gamma_{nh}$ where $\Gamma_{h}=g\int_{z}^{D}\rho^{\prime}dz$ (6) and $\Gamma_{nh}=\int_{z}^{D}\rho_{\infty}\frac{dw}{dt}dz$ (7) that corrispond respectively to hydrostatic and non-hydrostatic contribution on the perturbation pressure Using the isopicnal change of coordinates $z=z_{0}+\eta$ in the first integral $\Gamma_{h}$, equation (3) it becomes $U^{2}={U^{2}}_{0}-2N^{2}I_{\eta}-2\frac{\Gamma_{nh}}{\rho_{0}}$ (8) where $I_{\eta}=\int_{z_{0}}^{D}\eta dz_{0}$ (9) is the integral, obtained integrating streamlines displacement $\eta$ from the unperturbed streamline level $z_{0}$ to the top of the fluid where all streamlines perturbation are null. In Smith (1989a) the non-hydrostatic contribution $\Gamma_{nh}$ is neglected. This means that in equation (5) the vertical advective term is neglected. Using scale analysis and incompressibility condition it can be shown that the hydrostatic assumption is good only for obstacles with a vertical scale lenght smaller than the horizontal scale lenght. Using this approximation Smith (1989a) showed that stagnation occurs when the condition $\mbox{{F}}=\frac{{U^{2}}_{0}}{2N^{2}I_{\eta}}<1$ (10) is satisfied. Stagnation dynamics is then connected to a non local variations of pressure. Infact $I_{\eta}$ is the vertical integral of all the vertical displacements above the fixed general streamline. The shape of the obstacle can become very important because the integral of streamlines vertical displacements depends from it. Moreover to determine stagnation all the streamlines displacement field $\eta(x,y,z)$ have to be known. This means that the equations of motion (Navier-Stokes equations) had to be solved and a non linear solution of Navier-Stokes equation is not available. Then, to proceed further, a linear solution of the equation of motion has to be introduced to evaluate $\eta$ and to insert it into the conservation of energy obtaining an evaluation of the stagnation point. There is then a formal contraddiction: when stagnation occurs, streamlines become singular (e.g., they can split or intersect) then any linear solution is, in principle, not adapt to describe this behavior. To bypass this difficulty, in this work it is supposed that the linear solution, jointly with energy conservation, can give some information useful only to identify the onset of stagnation and not its behavior. 111It is possible, using other technique and approximation, taking in account part of the non-linear term and try to give some hint on behaviour of streamlines in the stagnation point. This aspect is not taking under consideration in this paper and further details can be found in Baines (1995). Even if this limitation is present, the result seem to be in agreement respect the numerical ones; for example Smith (1989b) has found that, introducing the values of streamlines displacement $\eta(x,y,z)$ obtained by his 3-D linear hydrostatic models into (8), the critical hill height (i.e. the minimum hill height at which the flow starting to stop) is about 30% lower than the critical height prediction using the vertical displacement field obtained by a numerical integration of the full non-linear hydrostatic equation using in Smith (1989b). This fact suggest that the final resolution of the splitting problem will probably be accomplished using real observational data: this is not so easy because real data for stagnation position are difficult to obtain. Even though there are evident experimental difficulties, real data can give some hint on what in reality happens and what numerical models can be able to see respect analytical ones and viceversa. This paper, following the above idea, will present a comparison of result of analytical models on the case of Valcanale flood (29 Agust 2003, Valcanale-UD-, Italy). Finally, It could be argued that if a linear theory is used to provide the field of motion, it would be more consistent to use the linearized set Bernoulli equation instead of the complete one (8). This could be done: Smith (1989b) found that the critical hill height is two times greater of the values prediction using the complete Bernoulli equation. Following Smith (1988), Smith (1989b) and Smith (1989a) we choose to use the exact result (8). Thus if any error is present in the derived fields it is because of the linearization leading to motion equation and not any subsequent linearizations and it is provides a common method for stagnation diagnosing (analytic) linear and (numerical) non-linear solution. ## 3 A non-hydrostatic model for the interaction between a flow and a 2-D topography As shown in the previous section, the non-hydrostatic approch can become fundamental for obstacles characterized by horizontal scale comparable with the vertical one. In this case the vertical acceleration term plays an important role in the flow blocking phenomena and it has to be taken into account in the evaluation of the perturbation pressure term of (4). To introduce the non-hydrostatic effects a 2-D model is developed and its governing equations will be solved explicity in the super-critical (Froude number $\mbox{{F}}>1$) and sub-critical ($\mbox{{F}}<1$) regime. Then the streamlines pattern for both these cases and the flow-blocking dynamics will be described. ### 3.1 Integral representation of solutions The development of the model starts from the governing equations of a 2-D incompressible, unviscid and stratified flow in Boussinesq’s approximation Emanuel (1994). $\left\\{\begin{array}[]{ll}\frac{Du}{Dt}=-\frac{1}{\rho_{\infty}}\frac{\partial{P^{*}}}{\partial{x}}\\\ \\\ \frac{Dw}{Dt}=-\frac{1}{\rho_{\infty}}\frac{\partial{P^{*}}}{\partial{z}}-\frac{g\rho^{\prime}}{\rho_{\infty}}\\\ \\\ \frac{\partial{u}}{\partial{x}}+\frac{\partial{v}}{\partial{y}}=0\\\ \end{array}\right.$ (11) where $\rho_{\infty}$ is the unperturbed density field far from the obstacle. The set (11) is linearized imposing for the unperturbed state the horizontal and constant vertical velocity profile $\boldsymbol{U}=(U_{0},0)$. Moreover the perturbation velocity field $\boldsymbol{u^{\prime}}$, using the streamfunction representation, can be described by $\boldsymbol{u^{\prime}}=(-\frac{\partial{\Psi}}{\partial{z}},\frac{\partial{\Psi}}{\partial{x}})$ (12) It is then necessary to impose that all the perturbations damp at the fluid top while the linearized impermeability boundary condition at the ground according to Baines (1995) is satisfied, so $\left\\{\begin{array}[]{ll}\Psi=U_{0}h(x)\qquad\textrm{ if }\quad z=0\quad\textrm{ ground }\\\ \Psi=0\quad\quad\qquad\textrm{ if }\quad z=D\quad\textrm{top of fluid }\\\ \end{array}\right.$ (13) In this way the perturbed streamfunction using a linear set of equations satisfys $\left\\{\begin{array}[]{ll}{\nabla}^{2}\Psi-\frac{N^{2}}{g}\frac{\partial\Psi}{\partial{z}}+\frac{N^{2}}{{U_{0}}^{2}}\Psi=0\\\ \\\ \Psi=0\quad\quad\qquad\quad\textrm{ when }\quad z=D\\\ \\\ \Psi=U_{0}h(x)\qquad\textrm{ when }\quad z=0\\\ \\\ \end{array}\right.$ (14) This set of equations describes the behaviour of a 2-D stratified, steady, incompressible, parallel flow interacting with a general 2-D obstacle described by the shape $z=h(x)$. The linearity constraints the value of the parameters which determine the model. In particular the constraint is represented by the following two conditions on Brunt-Vaisala frequency $N$ $\left\\{\begin{array}[]{ll}\frac{NH}{U_{0}}\ll 1\\\ \\\ \frac{N^{2}D}{g}\ll 1\\\ \end{array}\right.$ (15) When the obstacle is represented by a 2-D hill of $H=2000$ m high and the flow is characterized by a upstream velocity $U_{0}=15$ m s-1the require startification frequency that preserves linearity is $N<7.5\cdot 10^{-3}$ s-1. This condition then limits the applicability of the above developed linear model. In the section § 4 and § 5 this constrains will be consider. To proceed further the Fourier trasform in reciprocal space $k$ is here used to find an integral representation of the solution. Assuming an orograhy at the ground with this shape $z=h(x)=\frac{H}{1+(x/a)^{2}}$ (16) where $a$ is the half width at half orography high. The solution has the form $\Psi=\Psi_{1}+\Psi_{2}$ where $\left\\{\begin{array}[]{ll}\Psi_{1}=U_{0}Ha\int_{0}^{c}\cos(kx)e^{-ak}G_{1}dk\\\ \\\ \Psi_{2}=U_{0}Ha\int_{c}^{+\infty}\cos(kx)e^{-ak}G_{2}dk\\\ \\\ \end{array}\right.$ (17) with $c=N/U_{0}$ $G_{1}=\frac{\sin((D-z)\lambda_{1})}{\sin(D\lambda_{1})}$ (18) $G_{2}=\frac{\sinh((D-z)\lambda_{2})}{\sinh(D\lambda_{2})}$ (19) and $\lambda_{1}=\sqrt{c^{2}-k^{2}}$ (20) $\lambda_{2}=\sqrt{k^{2}-c^{2}}$ (21) ### 3.2 Analytic integration Obtaining an explicit analytic solution of (17) is not possible because the non-hydrostatic term is preserved in the strating equation (11). Moreover it can be noticed that, adopting the definition of Froude numeber given in Baines (1995) $\mbox{{F}}=\frac{\pi U_{0}}{ND}$ (22) two different behaviours exist, according to the different values of the Froude number. When $\mbox{{F}}>1$ (super-critical regime) the functions $G_{1}$ and $G_{2}$ do not show any singularity on their integral path and the integral of (17) is defined. For the super-critical regime it is then possible to adopt a numerical integration of (17). In particular it is here adopted a “Monte- Carlo, importance sampling” integration with the use of an algorithm for the generation of pseudo-random numbers. This choice was taken because for low values of the wave number $k$ and for low values of $z$, the integral function is nearly an harmonic function then, an equispatial integration method is not able to reproduce correctly the orographic profile near to the ground because it continuously adds harmonic components in phase. The use of a pseudo-random integration method, in which harmonic components are added randomly, seems a better approach. A comprehensive tractation of this integration method can be found in Gould & Tobochnik (1996). In the case $\mbox{{F}}<1$ (sub-critical regime) the function $G_{1}$ admits poles of first order for $k_{n}=\sqrt{c^{2}-(n\frac{\pi}{D})^{2}}\qquad\textrm{with}\quad n\in\aleph$ (23) then the integral of (17) exists only in an improper way called “Cauchy principal value integral”. The presence of a pole in the integration path makes impossible the use of numerical integration methods and the analytic integration of the “Cauchy pricinpal value integral” is needed to deal with the singularity behaviour but an explicit solution of (17) is not avaible. To overcome this problem in this work a new approach is proposed based on the substitution of $G_{1}$ and $G_{2}$ with two new functions $\widetilde{G_{1}}$ and $\widetilde{G_{2}}$ such that * • The function $\widetilde{G_{2}}$ has to tend to infinity with the same behaviour of ${G_{2}}$; in this way there is a correct estimate of short wave lengths. * • The function $\widetilde{G_{1}}$ has to reproduce the same singular behaviour and the limit of $\widetilde{G_{1}}/G_{1}$ as $k$ tends to the singularity has to be unitary. * • The function $\widetilde{G_{1}}$ and $\widetilde{G_{2}}$ have to have the same value in $c=N/U_{0}$. * • The function $\widetilde{G_{1}}$ has to have the same value of ${G_{1}}$ in $k=0$; this assures that the short wavenumber (i.e., wavelengths greater or equal to the horizontal scale of the obstacle) are correctly reproduced. * • With the new functions $\widetilde{G_{1}}$ and $\widetilde{G_{2}}$ the integral (17) can be solved explicitly. At this point is necessary to fix some parameter of analytical model: with an upstream velocity $U_{0}=15$ m s-1, an hill high $H=2000$ m and a top of fluid due to the end of Troposphere $D=10$ km, linearization conditions (15) imposes that the stratification frequency must be $N<7.5\cdot 10^{-3}$ s-1. So, notice the (23), there is only one pole $k_{0}$ $k_{0}=\sqrt{c^{2}-(\frac{\pi}{D})^{2}}$ (24) In the following section § 4 and § 5 the above value of parameter are considered, so the function $G_{1}$ have only one pole in the sub-critical case $(\mbox{{F}}<1)$. Following the previously introduced criteria, the new functions that are going to be uses instead of $G_{1}$ and $G_{2}$ become $\widetilde{G_{2}}=\beta e^{-z(k-c)}$ (25) and $\widetilde{G_{1}}=Mk+\alpha+W\frac{k(c-k)}{k-k_{0}}$ (26) where $\left\\{\begin{array}[]{ll}M=\frac{\beta-\alpha}{c}\\\ \\\ \alpha=\frac{\sin(c(D-z))}{\sin(Dc)}\\\ \\\ \beta=1-z/D\\\ \\\ \gamma=\sin(\beta\pi)\frac{\pi}{k_{0}^{2}D^{2}}\\\ \\\ P=\frac{\gamma}{c-k_{0}}\\\ \end{array}\right.$ (27) As we will be shown later on, the wave’s dynamic is governed by the parameter $W$; if $W\neq 0$, there is a wave pattern formation, if $W=0$ the wave pattern disappears. Then, inserting the new functions in (17) it is possible to solve explicity the integral Gradshteyn & Ryzhik (2000), obtaining an explicit solution of $\Psi$ for the sub-critical regime. It is important to notice that the different behaviour for $\mbox{{F}}>1$ and $\mbox{{F}}<1$ is a typical example of bifurcation, i.e., differnt physical behaviour as a consequence of an arbitrarily small change in the parameters value. The case $\mbox{{F}}=1$ can not be described by a simply linear theory infact, for this value of $F$, equation (17) shows a second order pole in $k=0$ and the integral does not exist, neither in the sense of “Cauchy principal value integral”. ## 4 The comparison of the streamlines pattern in hydrostatic and non- hydrostatic approaches In this section the streamlines pattern of the non-hydrostatic model are compared with those obtained under the hydrostatic assumption. This is done using the 2-D hydrostatic model proposed by Baines (1995) making use of the same assumptions (steady flow, unviscid, parallel, stratified and incompressible) here adopted. $\Psi=U_{0}h(x)\frac{\sin(c(D-z))}{\sin(cD)}$ (28) From the linearization of boundary condition, fixing as a starting unperturbed level $z=z_{0}$, the streamline is then represented by $z=z_{0}+\Psi/U_{0}$. All the results presented in the following sections had been obtained adopting an hill-shaped obstacle whose functional form is (16) and with the geometrical paramenters $a=1700$ m and $H=2000$ m. This value of hill parameter corrispond to a characteristic narrow mountains of Alpine ridge. It is important to notice that, as it is told before, the topography must be narrow to have non- hydrostaticity importance. ### 4.1 Super-critical regime ($\mbox{{F}}>1$) Figure 1: Super-critical regime of the flow with low stratification. Solid lines represent the streamlines of the non-hydrostatic model, while dashed lines represent the streamlines of the hydrostatic model. Horizontal and vertical axes scale are in meters. Panel (a) shows the streamlines pattern for $N=1\cdot 10^{-3}$ s-1while panel (b) shows the results obtained for $N=3\cdot 10^{-3}$ s-1. In figure 1 are displayed the hydrostatic and non-hydrostatic streamlines patterns. It is clear that the non-hydrostatic model is characterized by a lower vertical displacement respect the hydrostatic one at the same stratification frequency $N$. The origin of this behaviour is in the vertical acceleration term that the hydrostatic model can not take into account. Infact the increase of vertical streamlines displacement and velocity, due to the continuity equation as a consequence of narrowing of troposphere, gives rise to the formation of a low pressure zone at hill’s top. This low pressure zone constrains the vertical streamlines displacement because the fluid in the upper part of troposphere is pushed downward by this pressure deficit. This lower pressure is more accentuated in the non-hydrostatic model because the extra term represented by the velocity vertical variation. Then streamlines displacement is reduced in the non-hydrostatic model by the action of this counteracting pressure gradient. Figure 2: Sub-critical regime of the flow. Solid lines represent the streamlines for the non-hydrostatic model while dashed lines represent the streamlines for the hydrostatic model. Horizontal and vertical axes scale are in meters. Panel (a) shows the streamlines pattern for $N=6\cdot 10^{-3}$ s-1while panel (b) shows the streamlines pattern for $N=7\cdot 10^{-3}$ s-1. Before to conclude it has to be noted that the vertical displacement increases with the increasing of $N$. This behaviour can be observed both in the hydrostatic and non-hydrostatic model and might seem counter-intuitive (one could think that when stratification increases then the restoring forces become larger). This behavior is explained considering that the inertia of the lower-levels parcels increases more than the inertia of the upper-levels parcels when stratification increases. So, even if the restoring force becomes larger as stratification increase, the difference of masses of different air level parcel is crucial for the developing of perturbation pressure that play an important rule in the streamlines formation and, as it will be possible to see later, in flow-blocking dynamics. ### 4.2 Sub-critical regime ($\mbox{{F}}<1$) Figure 3: Sub-critical regime of the flow with high stratification. Solid lines represent the streamlines of the non-hydrostatic model while dashed lines represents the streamlines for the hydrostatic model.Horizontal and vertical axes scale are in meters. Panel (a) shows the streamlines pattern for $N=6\cdot 10^{-3}$ s-1while panel (b) shows the streamlines pattern for $N=7\cdot 10^{-3}$ s-1. Considering all the terms present in (26) the results shown in figure 2 are obtained. In that picture it is clearly recognizable the formation of stationary waves. These results are very similar to those found in Keller (1994). The formation of these waves is characteristic of the singularity, infact for a wind velocity of $15$ m s-1and a stratification frequency of the order of $N=7\cdot 10^{-3}$ s-1there is a pole at $k_{0}=\sqrt{c^{2}-(\frac{\pi}{D})^{2}}=3.45\cdot 10^{-4}\nobreak\mbox{$\;$m${}^{-1}$}$ (29) which corresponds to a wavelength of $\lambda\approx 18$ kmthat is clearly recognizable in figure 2. Before to proceed further it has to be noted that the streamline waves pattern is symmetric to the top of the hill while from real cases it is clear that the waves pattern can be present only downstream to the hill. This is a spurious effect of the linearization, infact the (14) does not change for the reflection of the variable $x$. To avoid this spurious effect it is sufficient to fix $W=0$ in (26) and the streamlines pattern that one obtain is shown in figure 3. It is then interesting to notice that even if the waves are produced by the obstacle, their wavelength has not the same order of magnitude of the hill’s horizontal scale. Infact, being triggered by the singular point $k_{0}$, the waves depend only from the stratification and from the velocity profile. In figure 3 it is possible to clearly recognize the velocity intensification at the top of the obstacle represented by the streamlines concentration in that region. This intensification is a well known effect to mountain-hickers and it is due to the falling down of the upper-level streamlines which is not present in the super-critical stratification. This different behaviour comes out because in the sub-critical regime the upper-level parcels fall down in a denser environment that gives them the necessary up-ward lifting force to return to their initial level when they return far from obstacle. The obtained stationary pattern is represented in figure 3. This situation can not take place in the super-critical regime, when a downward displacement similar to that of the sub-critical case, cannot receive the same lifting force because of a less stratified environment. It is interesting to notice that the non-hydrostatic model gives a lower intensification of wind speed at the top of the hill respect the hydrostatic ones at the same stratification frequency. A possible description of this effects can be explained, as was done in the super-critical case, considering the Bernoulli equation even if, in sub-critical case, the streamlines pattern is more complicated respect the super-critical one. In this case the acceleration of the fluid and the increase of vertical displacement of lower level streamlines gives the formation of a low pressure area at hill’s top while the falling of upper streamlines could be enough to compensate the fluid acceleration and so this falling streamlines could give an increase of pressure: infact the mid-level streamlines, after an initial falling, seem to moves upward due to the acting of this increase of pressure as it is possible to see in figure 3. So, this more complicated pressure pattern gives rise to a lower intensification of wind speed at the top of the hill and, as it will be noted later, is the responsable of flow-blocking in sub-critical flow. Finally it is worth to be noticed that, also in this case, the vertical displacement increases with the increasing of stratification (the altitude of the wind speed intensification increase grows with $N$). The explanation given again makes use of the increasing of inertia with the increasing of stratification. Lower-level parcels moving upward find themselves immersed in a layer of fluid less dense, then with a low capability of counteracting with the buoyancy force their upward motion. Before to conclude this section it is again important to state that, both in the super-critical and sub-critical regime, the non-hydrostatic effect consists in a smoothing of the streamlines geometry and the smoothing effect is concetreted near to the top of the obstacle. ## 5 Non-hydrostatic effects on flow-blocking Knowing the streamfunction for the non-hydrostatic case, it is possible to can calculate the non-hydrostatic term $\Gamma_{nh}$ of (7) due to the vertical acceleration and use it in (8). Four cases for different stratification frequencies $N=1,3,6,7\cdot 10^{-3}$ s-1are taken into account. The first two frequencies (i.e., $N=1,3\cdot 10^{-3}$ s-1) correspond to the super-critical regime while the last two (i.e., $N=6,7\cdot 10^{-3}$ s-1) to the sub-critical regime. ### 5.1 Hydrostatic result Figure 4: Flow-blocking and hydrostatic approximation. The unperturbed streamine level is reported in the horizontal axis, in meters, while the abscissa of stagnation (i.e, the distance from the top of the hill where the flow-blocking occurrs) is reported, in meters, in the vertical axis. This picture is obtained with a stratification $N=3\cdot 10^{-3}$ s-1. In the equation (8) the vertical advection term (7) is null because a hydrostatic model is here used, then the streamlines displacement is given by (28). The relationship between the unperturbed streamline level and stagnation abscissa is shown in figure 4. The only stratification frequency that admits stagnation is $N=3\cdot 10^{-3}$ s-1, which corresponds to the streamlines pattern where the vertical displacement larger, as can be seen in figure 1. The explanation of this fact can be found in the behaviour of streamline pattern. In a super-critical hydrostatic flow, stagnation occours when the streamlines does not have a sufficient kinetic energy to trasform into potential to produce the vertical displacement request. On the contrary, the sub-critical flow blocking, as it will be remark in the next section, happens when the falling of upper streamlines creates a pressure configuration at the top of hill that could stop low-level streamlines A last comment can be done on stagnation abscissa: there are streamlines, in the hydrostaic case, that stopped in the proximity of hill’s top. ### 5.2 Non-Hydrostatic result Figure 5: Non-hydrostatic case. In the horizontal axis it is reported the unpertubated streamline level (in meters) while the vertical axis reports, in meters, the abscissa of stagnation, i.e. the distance from the top of the hill where the fluid stops. Panel (a) reports the super-critical regime. Dash upper lines correspond to the stratification $N=3\cdot 10^{-3}$ s-1while solid lower lines correspond to $N=1\cdot 10^{-3}$ s-1. Panel (b) reports the sub-critical regime. Dash upper lines correspond to the stagnation $N=7\cdot 10^{-3}$ s-1while solid lower lines to $N=6\cdot 10^{-3}$ s-1. In the non-hydrostatic case flow stagnation can take place for all the considered stratification frequencies. In the super-critical regime the behaviour of the abscissa of stagnation is shown in the panel $a$ of figure 5. This behaviour is similar to that evidenced in the hydrostatic regime even if in the non-hydrostatic case the abscissa of stagnation is upward limited and does not pass the abscissa of $2500$ m from the top of the hill. The explanation of this fact can be found in the role of the streamlines curvature. In figure 5a it is possible to observe that as all the stagnation abscissas are positionated before of the streamlines’ flexums (for the ground streamline the flexum is at $a\approx 1700$ km, value that corresponds to the flexum of (16)). This effect is similar to that experienced by a driver when he or she is running on the positive curvature of a road and his/her car is forced downward while on the negative curvature of the road the car is lifted. The same effect takes place in fluids but only a non-hydrostatic model can keep into account this mechanism because it is the only model that can consider vertical accelerations. In the sub-critical regime the situation is quite different. The inertia of the lower-levels fluid parcels pushes the fluid beyond the flexum and stagnation occurs later than in the super-critical regime. The other important things to be noticed is that large portions of the fluid are blocked in the sub-critical regime (nearly the $8$ kmof fluid nearer to the ground). This situation corresponds to the pattern in figure 3 where the upper streamlines fall toward the obstacle. This means that large amounts of kinetic energy become potentially available. This fact is in agreement with the findigs of Schar & Smith (1993a), Smolarkiewcz & Rotunno (1989) and, Castro et al. (1983) where stagnation is associated with the vorticity generation. Concerning the role of stratification frequency on stagnation in figure 5 it can be shown that when $N$ increases, stagnation occours later, i.e., for smaller abscissas. This fact is in agreement with the streamlines pattern, in fact when stratification increases the vertical displacement is larger. This result might seem countrintuitive but it can be explained taking into account the fact that when stratification increases, restoring forces increase but the inertia of the lower parts of the fluid increases as well. This inertia is strictly connected to the generation of pressure perturbation, that play a crucial roll in flow-blocking dynamics, and a non-hydrostatic model can better keep into account this fact in comparison with hydrostatic models. Even if this result is obtained with a linear model valid only for a restrict range of stratification frequencies it can give a useful hint toward the interpretaion of real cases, in particular of what happend in Valcanale (UD), Italy, during the $29^{th}$ August 2003. Figure 6: Radar image: a moist and stable stratified flow with $N=0.03$ s-1interact with 2-D ridge. The flow overtaking the first ridge but stopped on the second (red and dark region) where starting convection and an intense orographic rains. Photo made by doppler radar station of Fossalon (Ud), Italy by Meteorological regional service O.S.M.E.R. In that day a flow strongly stratified in the lower levels ($N=3\cdot 10^{-2}$ s-1compared to the mean value of $N=7\cdot 10^{-3}$ s-1) moving from south and interacting with the orographic ridge (Julian Preals and Alps) overcame the first ridge producing large amount of rain (nearly $400mm$ in four hours) and two casualties only on the further inner ridge (Julian Alps) where convection took place, as can be seen from the radar image shown in figure 6. ## 6 Conclusions This paper presents a study on the interactions beetween stratified flows and orography carried out developing an analytical model. In particular, starting from the previous works of Smith (1989a) and Keller (1994), here the non- hydrostatic terms are kept into account in a simplified 2-D model to evaluate their effects on streamlines patters and flow-blocking. Following this idea an integral solution of the 2-D non-hydrostatic model had been found, whose behaviour is described by way of the Monte-Carlo sampling integration method for the super-critical regime and by way of a newly developed explicit integration approch in the sub-critical regime. The main results of this work can be summarized in the following points: * • The non-hydrostatic effects are important for a topography characterized by a horizontal scale comparable with the vertical scale. In this case, in fact, the change in the topography curvature, then in the streamlines curvature that is connected to perturbative pressure, becomes important. * • The non-hydrostatic effects produces a general smoothing of the streamlines pattern. These effects can be explained taking into account the presence of the high/low pressure area, due to the complexity of streamlines picture, that a non-hydrostatic model are much able to see. * • The streamlines pattern shows, for the sub-critical regime, the formation of waves downstream to the topography, that hydrostatic models can not reproduce. The wavelength of this undulatory pattern is connected to the properties of a singular point in the integral function which depends only from the stratification frequency and from the upstream velocity of the flow. The intensification of flow velocity at the topography top is observed in the sub- critical regime. * • Stagnation occurs upstream to the topography and hydrostatic and non- hydrostatic models show a very different behavior in flow-blocking. In hydrostatic models stagnation occurs on the top of topography and only for large vertical displacements of streamlines (i.e, in the super-critical regime). * • In the non-hydrostatic model, for the super-critical regime, only lower-levels streamlines are blocked and stagnation does not occur beyond the flexum abscissa. This behaviour can be explained taking into account the role of streamline’s curvature which is connected to the vertical perturbation pressure (lifting in the negative curvature part of streamlines and downward restoring force in the positive curvature part of streamlines) that can not be observed in hydrostatic models. * • In the non-hydrostatic model, for the sub-critical regime, the stagnation is due by a formation of complex pressure pattern at the top of hill. Moreover stagnation in sub-critical case occurs beyond the flexum’s abscissa upstream to topography. This because the inertia of the lower-levels parcels is larger than the inertia of the upper-levels parcels. In the sub-critical regime the amount of fluid that is blocked is very large: this means that a large amount of kinetic energy is made avaible. This source of kinetic energy might become a source of vorticity production connected with the flow-blocking as previously observed by Schar & Smith (1993a), Smolarkiewcz & Rotunno (1989) and Castro et al. (1983). * • As stratification increases, the flow stops later than a less stratified flow. For this reason stratification parameters alone are not sufficient to detect the flow-blocking situation because stratification and curvature effects are linked to the perturbation pressure that can have an important role on flow- blocking, as shown in this work. So this work suggest the opportunity to connect perturbation pressure to stratification and curvature parameters. * • This work suggest the possibility to construct a faster evaluation method of stagnation phenomena: the result of a linear model, that can be run faster and for a more smaller spatial mesh, could be use to describe the flow-blocking onset. This could be an important application for the wheater forecasting where the evaluation time is an important parameter to minimize in many circumstances. ###### Acknowledgements. The authors aknowledge the FORALPS Project (Interreg IIIB - Alpine Space) in the frame of which this work was partially carried out. ## References * Baines (1979) Baines, P. G. 1979 observationas of stratified flow past three-dimensional barriers. J. Geophys. Res. 84, 7834–7838. * Baines (1995) Baines, P. G. 1995 Topographic effects in stratified flows. Cambridge University Press, London, New York, Melbourbne. * Batchelor (1994) Batchelor, G. K. 1994 An Introduction to Fluid Dynamics. Cambridge University Press, London, New York, Melbourbne. * Carrunthers & Hunt (1990) Carrunthers, D. J. & Hunt, J. C. R. 1990 Fluid Mechanics of Airflow Over Hills:Turbulent, Fluxes and Waves in the Boundary Layer. In Atmospheric Process over Complex terrain, Meteorology Monograph, vol. 45, pp. 83–104. American Meteorology Society. * Castro et al. (1983) Castro, I. P., Snyder, W. H. & Marsh, G. L. 1983 Stratified flow over three dimensional ridges. J. Fluid Mech. 135, 261–282. * Emanuel (1994) Emanuel, K. E. 1994 Atmospheric Convection. Oxford University Press. * Giaiotti et al. (2007) Giaiotti, D., Stel, F. & Steinacker, R. 2007 Atmospheric convection: research and operational forecasting aspects. Springer. * Gould & Tobochnik (1996) Gould, H. & Tobochnik, J. 1996 An Introduction to Computer Simulation. Addison Wesley. * Gradshteyn & Ryzhik (2000) Gradshteyn, I. S. & Ryzhik, I. M. 2000 Table of Integral Series and Product. Academic Press. * Hunt & Snyder (1980) Hunt, J. C. & Snyder, W. H. 1980 Experiments on stably and neutrally stratified flow over a model three-dimensional hill. J. Fluid Mech. 96, 671–704. * Inttyre (1972) Inttyre, M. E. M. 1972 On long’s model hypothesis of no upstream influence in uniformly stratified or rotating flow. J. Fluid Mech. 52, 209–243. * Keller (1994) Keller, T. L. 1994 Implications of the hydrostatic assumption on atmospheric gravity waves. J. Atmos. Sci. 51, 1915–29. * Lilly & Klemp (1979) Lilly, D. K. & Klemp, J. B. 1979 The effects of terrain on nonlinear hydrostatic mountains waves. J. Atmos. Sci. 95, 241–261. * Miglietta & Rotunno (2005) Miglietta, M. & Rotunno, R. 2005 Simulation of moist nearly flow over ridge. J. Atmos. Sci. 62, 1410–1427. * Reisner & Smolarkiewicz (1993) Reisner, J. M. & Smolarkiewicz, P. K. 1993 Thermally forced low froude number past three dimensional obstacles. J. Atmos. Sci. 51, 117–113. * Riley et al. (1976) Riley, J. J., Liu, H. T. & Geller, E. W. 1976 A numerical and experimental study of stratified flow around complex terrain. Tech. Rep. EPA-60014-76-021. Environmental Protection Agency Report. * Schar & Smith (1993a) Schar, C. & Smith, R. B. 1993a Shallow water flow past isolated topography. part i: Vorticity production and wake formation. J. Atmos. Sci. 50, 1373–1400. * Schar & Smith (1993b) Schar, C. & Smith, R. B. 1993b Shallow water flow past isolated topography. part ii: Transition to vortex shedding. J. Atmos. Sci. 50, 1401–1412. * Sheppard (1956) Sheppard, P. A. 1956 Airflow over mountain. Q. J. R. Met. Soc. 82, 528–529. * Smith (1988) Smith, R. B. 1988 Linear theory of stratified flow past an isolated mountain in isosteric coordinates. J. Atmos. Sci. 45, 3889–3896. * Smith (1989a) Smith, R. B. 1989a Hydrostatic airflow over mountains. Advanced in Geophysics 31, 1–41. * Smith (1989b) Smith, R. B. 1989b Mountain-induced stagnation points in hydrostatic flow. Tellus 41A, 270–274. * Smolarkiewcz & Rotunno (1989) Smolarkiewcz, P. K. & Rotunno, R. 1989 Low froude number flow past three-dimensional obstacles. part i: Baroclinically generated lee vortices. J. Atmos. Sci. 46, 1154–1164. * Snyder et al. (1985) Snyder, W. H., Thompson, R. S., Eskridge, R. E., Lawson, R. E., Castro, I. P., Lee, J. T. & Ogawa, Y. 1985 The structure of strongly stratified flow over hills: Dividing streamline concept. J. Fluid. Mech. 152, 249–288. * Wurtele et al. (1987) Wurtele, M. G., Sharman, R. D. & Keller, T. L. 1987 Analysis and simulation of a troposhere-stratosphere gravity waves. J. Atmos. Sci. 44, 3296–3281.

arxiv-papers

2024-09-04T02:48:53.916837

0802.3129

# An explicit finite difference scheme for the Camassa-Holm equation G. M. Coclite Dipartimento di Matematica Università degli Studi di Bari Via E. Orabona 4 70125 Bari, Italy coclitegm@dm.uniba.it , K. H. Karlsen Centre of Mathematics for Applications (CMA) University of Oslo P.O. Box 1053, Blindern N–0316 Oslo, Norway kennethk@math.uio.no http://folk.uio.no/kennethk/ and N. H. Risebro Centre of Mathematics for Applications (CMA) University of Oslo P.O. Box 1053, Blindern N–0316 Oslo, Norway nilshr@math.uio.no http://folk.uio.no/nilshr/ ###### Abstract. We put forward and analyze an explicit finite difference scheme for the Camassa-Holm shallow water equation that can handle general $H^{1}$ initial data and thus peakon-antipeakon interactions. Assuming a specified condition restricting the time step in terms of the spatial discretization parameter, we prove that the difference scheme converges strongly in $H^{1}$ towards a dissipative weak solution of Camassa-Holm equation. ###### Key words and phrases: Camassa-Holm equation, hyperbolic-elliptic system, weak solution, finite difference scheme, convergence ###### 2000 Mathematics Subject Classification: 35G25, 35L05, 65M06, 65M12 This work was supported by the Research Council of Norway through the project WaveMaker and an Outstanding Young Investigators Award of K. H. Karlsen. ###### Contents 1. 1 Introduction 2. 2 Preliminaries 3. 3 Explicit scheme and main result 4. 4 Total energy estimate and some consequences 5. 5 One-sided sup-norm estimate 6. 6 Higher integrability estimate 7. 7 Basic convergence results 8. 8 Strong convergence result 9. 9 Numerical examples ## 1\. Introduction In this paper we present and analyze an explicit finite difference scheme for the Camassa-Holm partial differential equation [7] (1.1) $\partial_{t}u-\partial_{txx}^{3}u+3u\partial_{x}u=2\partial_{x}u\partial_{xx}^{2}u+u\partial_{xxx}^{3}u,\qquad(t,x)\in(0,T)\times\mathbb{R},$ which we augment with an initial condition: (1.2) $u|_{t=0}=u_{0},\qquad u_{0}\in H^{1}(\mathbb{R}),u_{0}\neq 0.$ Rewriting equation (1.1) as $(1-\partial_{xx}^{2})\left[\partial_{t}u+u\partial_{x}u\right]+\partial_{x}\left(u^{2}+\frac{1}{2}(\partial_{x}u)^{2}\right)=0,$ we see that (for smooth solutions) (1.1) is equivalent to the elliptic- hyperbolic system (1.3) $\partial_{t}u+u\partial_{x}u+\partial_{x}P=0,\qquad-\partial_{xx}^{2}P+P=u^{2}+\frac{1}{2}(\partial_{x}u)^{2}.$ Recalling that $e^{-|x|}/2$ is the Green’s function of the operator $1-\partial_{xx}^{2}$, (1.3) can be written as (1.4) $\partial_{t}u+\partial_{x}F(u,\partial_{x}u)=0,\quad F(u,\partial_{x}u)=\frac{1}{2}\left[u^{2}+e^{-\left|x\right|}\star\left(u^{2}+\frac{1}{2}(\partial_{x}u)^{2}\right)\right],$ which can be viewed as a conservation law with nonlocal flux function. In this paper the relevant formulation of the Camassa-Holm equation (1.1) is the one provided by the hyperbolic-elliptic system (1.3) or (1.4). The Camassa-Holm equation can be viewed as a model for the propagation of unidirectional shallow water waves [7, 32]; it is a member of the class of weakly nonlinear and weakly dispersive shallow water models, a class which already contains the Korteweg-de Vries (KdV) and Benjamin-Bona-Mahony (BBM) equations. In another interpretation the Camassa-Holm equation models finite length, small-amplitude radial deformation waves in cylindrical compressible hyperelastic rods [21]. It arises also in the context of differential geometry as an equation for geodesics of the $H^{1}$-metric on the diffeomorphism group, see for example [17, 18, 30, 36]. The Camassa-Holm equation possesses several striking properties such as an inifinite number of conserved integrals, a bi-Hamiltonian structure, and complete integrability [2, 7, 19, 14, 26]. Moreover, it enjoys an infinite number of non-smooth solitary wave solutions, called peakons, which are weak solutions of (1.4). From a mathematical point of view the Camassa-Holm equation has by now become rather well-studied. While it is impossible to give a complete overview of the mathematical literature, we shall here mention a few typical results, starting with the local(-in-time) existence results in [15, 34, 37] and those using Besov spaces in [23, 22]. It is well-known that global solutions do not exist and wave-breaking occurs [7]. Wave-breaking means that the solution itself stays bounded while the spatial derivative becomes unbounded in finite time. In view of what we have said so far (peakon solutions/wave-breaking) it is clear that a theory based on weak solutions is essential. In the literature there are a number of results on (dissipative and conservative) weak solutions of the Camassa-Holm equation, see [3, 4, 5, 16, 11, 20, 29, 39, 40] and the references cited therein. In this paper we are interested specifically in the class of dissipative weak solutions studied by Xin and Zhang [39, 40]. Their results show, among other things, that there exists a global dissipative weak solution of (1.1)-(1.2) for any $H^{1}$ initial data $u_{0}$ (peakon- antipeakon interactions are covered). These solutions are global in the sense that they are defined past the blow-up time (wave-breaking). More precisely, suppose $u_{0}\in H^{1}(\mathbb{R})$. Then there exists a global weak (distributional) solution $u\in L^{\infty}(0,T;H^{1}(\mathbb{R}))$ of (1.1) satisfying the following properties: $t\mapsto\left\|u(t,\cdot)\right\|_{H^{1}(\mathbb{R})}$ is non-increasing; $\partial_{x}u\in L^{p}_{\mathrm{loc}}(\mathbb{R}_{+}\times\mathbb{R})$, $p<3$; (1.5) $\partial_{x}u(t,x)\leq\frac{2}{t}+C\left\|u_{0}\right\|_{H^{1}(\mathbb{R})},\quad\text{for}\quad t>0,$ for some positive constant $C$. This last item presumably singles out a unique weak solution. As an example of how this may work we consider the “peakon- antipeakon” solution given by (1.6) $u(t,x)=\tanh(t-1)\left(e^{-\left|x-y(t-1)\right|}-e^{-\left|x+y(t-1)\right|}\right),\quad y(t)=\log\left(\cosh(t)\right).$ This formula represents a peakon ($e^{-\left|x+y\right|}$) colliding with an antipeakon ($-e^{-\left|x-y\right|}$) at $x=0$ and $t=1$. Note that $u(1,x)=0$. How this solution is extended to $t>1$ depends on which solution concept we adopt. If we use the formula (1.6) also for $t>1$ we get the conservative solution for which $\left\|u(t,\cdot)\right\|_{H^{1}(\mathbb{R})}$ is constant for almost all $t$. We can also extend the solution by defining $u(t,x)=0$ for $t>1$. Obviously, the “entropy condition” (1.5) will only be satisfied for this dissipative solution. Let us now turn to the topic of the present paper, which is the design and analysis of numerical schemes. The first numerical results for the Camassa- Holm equation are presented in [8] using a pseudo-spectral scheme. Numerical simulations with pesudo-spectral schemes are also reported in [25, 31]. Numerical schemes based on multipeakons (thereby exploiting the Hamiltonian structure of the Camassa-Holm equation) are examined in [6, 9, 10]. In [28], the authors prove that the multipeakon algorithm from [9, 10] converges to the solution of the Camassa-Holm equation as the number of peakons tends to infinity. This convergence result applies to the specific situation where the initial function $u_{0}\in H^{1}$ is such that $(1-\partial_{xx}^{2})u_{0}$ is a positive measure. For the same class of initial data, in [27] the authors prove that a semi-discrete finite difference scheme based on the variable $m:=(1-\partial_{xx}^{2})u$ converges strongly in $H^{1}$ to the weak solution identified in [16, 20]. In [33], the authors establish error estimates for a spectral projection scheme for smooth solutions. In a different direction, an adaptive high-resolution finite volume scheme is developed and used in [1]. The local discontinuous Galerkin method is adapted to the Camassa-Holm equation in [41]. Although this work does not provide a rigorous convergence result for general (non-smooth) solutions, they show that the discrete total energy is nonincreasing in time, thereby suggesting that the approximate solutions are of dissipative nature. Besides, they establish an error estimate for smooth solutions. Finally, multi-symplectic schemes possessing good conservative properties are suggested and demonstrated in the recent work [13]. It seems rather difficult to construct numerical schemes for which one can prove the convergence to a (non-smooth) solution of the Camassa-Holm equation. This statement is particularly accurate in the case of general $H^{1}$ initial data and peakon-antipeakon interactions. Indeed, in this context we are only aware of the recent work [12] in which we prove convergence of a tailored semi-discrete difference scheme to a dissipative weak solution. Before we can outline this scheme, let us discretize the spatial domain $\mathbb{R}$ by specifying the mesh points $x_{j}=j{\Delta x}$, $x_{j+1/2}=(j+1/2){\Delta x}$, $j=0,\pm 1,\pm 2,\dots$, where ${\Delta x}>0$ is the length between two consecutive mesh points (the spatial discretization parameter). Let $D_{-}$, $D$, and $D_{+}$ denote the corresponding backward, central, and forward difference operators, respectively. The scheme proposed in [12], which is based on the formulation (1.3), reads (1.7) $\begin{split}&\frac{d}{dt}u_{j+1/2}+\left(u_{j+1/2}\vee 0\right)D_{-}u_{j+1/2}+\left(u_{j+1/2}\wedge 0\right)D_{+}u_{j+1/2}+D_{+}P_{j}=0,\\\ &-D_{-}D_{+}P_{j}+P_{j}=\left(u_{j+1/2}\vee 0\right)^{2}+\left(u_{j-1/2}\wedge 0\right)^{2}+\frac{1}{2}\left(D_{-}u_{j+1/2}\right)^{2},\end{split}$ where $u_{j+1/2}(t)\approx u(t,x_{j+1/2}),\quad P_{j}(t)\approx P(t,x_{j}),\quad\text{for $t\geq 0$ and $j\in\mathbb{Z}$}.$ If we interpret the Camassa-Holm equation (1.4) as a “perturbation” of the inviscid Burgers equation, then the $u$-part of (1.7) might not come across as a reasonable (upwind) difference scheme. On the other hand, as pointed out in [12], the key point is that with (1.7) the quantity $q_{j}:=D_{-}u_{j+1/2}$ satisfies a difference scheme which contains proper upwinding of the transport term in the equation for $q:=\partial_{x}u$, which reads $\partial_{t}q+u\partial_{x}q+\frac{q^{2}}{2}+P-u^{2}=0$. Consequently, as is proved in [12], the scheme (1.7) satisfies a total energy inequality, in which only the $q$-part of the total energy is dissipated (not the $u$-part, which is after all continuous). This is the essential starting point for the entire convergence analysis in [12]. The “semi-discrete” equation in (1.7) constitutes an infinite system of ordinary differential equations which must be solved by some numerical method. The main purpose of the present paper is to show that a fully discrete version of the scheme used in [12] produces a convergent sequence of approximate solutions, and that the limit is a dissipative weak solution to (1.1). The fully discrete version that is analyzed in this paper is based on replacing the time derivative in (1.7) by a forward difference, i.e., $u_{j+1/2}^{\prime}(t)\to D_{+}^{t}u_{j+1/2}^{n}:=\frac{u^{n+1}_{j+1/2}-u_{j+1/2}^{n}}{{\Delta t}},$ and evaluating the rest of (1.7) at $t^{n}:=n{\Delta t}$. Now $u_{j+1/2}^{n}$ should approximate the exact solution $u$ at the point $(t^{n},x_{j+1/2})$. This gives the fully discrete scheme (1.8) $\begin{split}D_{+}^{t}u_{j+1/2}^{n}+\left(u_{j+1/2}^{n}\vee 0\right)D_{-}u_{j+1/2}^{n}+\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}u_{j+1/2}^{n}+D_{+}P_{j}^{n}=0\\\ -D_{-}D_{+}P_{j}^{n}+P_{j}^{n}=\left(u_{j-1/2}^{n}\vee 0\right)^{2}+\left(u_{j+1/2}^{n}\wedge 0\right)^{2}+\frac{1}{2}\left(D_{-}u_{j+1/2}^{n}\right)^{2},\end{split}$ where $P_{j}^{n}$ approximates $P(t^{n},x_{j})$. As in [12] this is a difference scheme which is tailored so that it gives an upwind scheme for the equation satisfied by $q:=\partial_{x}u$. The main aim of this paper is prove that the fully discrete (explicit) scheme (1.8) converges to a dissipative weak solution of the Camassa-Holm equation. The starting point of the analysis is a total energy estimate, showing that the $H^{1}$ norm of the approximate solutions is (almost) nonincreasing in time. To this end, we must assume that (1.9) ${\Delta t}=\mathcal{O}({\Delta x}^{2}\log(1+{\Delta x}^{\theta}))$ for some $\theta>0$ as ${\Delta x}\to 0$. This is a very severe condition, and it may seem that when using this method in practice one should use very small time steps. However, this is not a Neumann type stability criterion, and we do not have blow up if it is violated. Indeed, practical experiments indicate stability and convergence if ${\Delta t}=\mathcal{O}({\Delta x})$. By appropriately extending the difference solution (1.8) to a function $u_{{\Delta x}}(t,x)$ defined at all points $(t,x)$ in the domain, we prove under condition (1.9) that $\left\\{u_{{\Delta x}}\right\\}_{{\Delta x}>0}$ converges strongly in $H^{1}$ to a dissipative weak solution of the Camassa- Holm equation (1.1)-(1.2). Regarding the proof, we adapt the “renormalization” approach used in [12] for the semi-discrete scheme, but there are several essential deviations and many parts of the convergence proof are substantially more involved and/or different. These differences are mainly due to the fact that the semi-discrete scheme, when viewed as a fully discrete scheme with “infinitely small time steps”, has a large and stabilizing numerical viscosity. Regarding the fully discrete (explict) scheme (1.8), to account for this lack of numerical viscosity the convergence analysis relies heavily on the CFL condition (1.9) and differennt With reference to the differences between the semi-discrete and fully discrete schemes, let us here point out just one aspect, namely that the $H^{1}$ norm of the fully discrete approximation is not entirely nonincreasing but can grow slightly with a growth factor that, however, tends to zero as ${\Delta x}\to 0$. Compared to semi-implcit case [12], the proof is notably more complicated and involves working with a version of the scheme (1.8) in which the quadratic terms have been suitably truncated. The paper is organized as follows: In Section 2 we introduce some notation to improve the readability and recall a few mathematical results relevant for the convergence analysis. The finite difference scheme and its convergence theorem are stated in Section 3. The convergence theorem is a consequence of the results proved in Sections 4-8. Finally, we present a numerical example in Section 9. Throughout this paper we use $C$ to denote a generic constant; the actual value of $C$ may change from one line to the next in a calcuation. We also use the notation that $a_{i}\lesssim b_{i}$ to mean that $a_{i}\leq Cb_{i}$ for some positive constant $C$ which is independent of $i$. ## 2\. Preliminaries In what follows, ${\Delta x}$ and ${\Delta t}$ denote two small positive numbers. Unless otherwise stated, the indices $j$ and $n$ will run over $\mathbb{Z}$ and $0,\ldots N$, respectively, where $N{\Delta t}=T$ for a fixed final time $T>0$. For such indices we set $x_{j}=j{\Delta x}$, $x_{j+1/2}=(j+1/2){\Delta x}$, $t^{n}=n{\Delta t}$, and introduce the grid cells $I_{j}=[x_{j-1/2},x_{j+1/2}),\quad I^{n}=[t^{n},t^{n+1}),\quad\text{and}\quad I^{n}_{j}=I_{j}\times I^{n}.$ The following notations will be used frequently: $a\vee 0=\max\left\\{a,0\right\\}=\frac{a+\left|a\right|}{2},\qquad a\wedge 0=\min\left\\{a,0\right\\}=\frac{a-\left|a\right|}{2}.$ For $n\in\left\\{0,\dots,N\right\\}$, let $v^{n}=\left\\{v_{j}^{n}\right\\}_{j\in\mathbb{Z}}$ denote an arbitrary sequence, where $n$ refers to “time” and $j$ to “space”. We will frequently employ the following finite difference operators: $\begin{gathered}D_{+}v^{n}_{j}:=\frac{v^{n}_{j+1}-v^{n}_{j}}{{\Delta x}},\qquad D_{-}v^{n}_{j}:=\frac{v^{n}_{j}-v^{n}_{j-1}}{{\Delta x}},\\\ Dv^{n}_{j}:=\frac{D_{+}v^{n}_{j}+D_{-}v^{n}_{j}}{2}=\frac{v^{n}_{j+1}-v^{n}_{j-1}}{2{\Delta x}}\quad\text{and}\quad D_{+}^{t}v^{n}_{j}=\frac{v^{n+1}_{j}-v^{n}_{j}}{{\Delta t}}.\end{gathered}$ We also use the notations $\displaystyle\left\|v^{n}\right\|_{\ell^{p}}$ $\displaystyle:=\left(\Delta x\sum\limits_{j\in\mathbb{Z}}\left|v^{n}_{j}\right|^{p}\right)^{\frac{1}{p}},\,1\leq p<\infty,\qquad\left\|v^{n}\right\|_{\ell^{\infty}}:=\sup_{j}\left|v_{j}^{n}\right|,$ $\displaystyle\left\|v^{n}\right\|_{h^{1}}$ $\displaystyle:=\left({\Delta x}\sum\limits_{j\in\mathbb{Z}}\left[\left(v^{n}_{j}\right)^{2}+(D_{-}v^{n}_{j})^{2}\right]\right)^{\frac{1}{2}}.$ Occasionally, we also use the “space-time” $\ell^{p}$ norms of $v=\left\\{v^{n}\right\\}_{n=0}^{N}=\left\\{v_{j}^{n}\right\\}_{j,n}$: $\left\|v\right\|_{\ell^{p}}:=\left({\Delta t}\sum_{n=0}^{N}\left\|v^{n}\right\|_{\ell^{p}}^{p}\right)^{1/p}.$ Note that if $v\in\ell^{p}$, $p<\infty$, then $\lim_{j\to\pm\infty}v_{j}=0$. Let $\left\\{v_{j}\right\\}_{j\in\mathbb{Z}}$ and $\left\\{w_{j}\right\\}_{j\in\mathbb{Z}}$ denote two arbitrary (spatial) sequences. Suppose $\left\|\left\\{v_{j}\right\\}_{j}\right\|_{h^{1}}<\infty$. Then the following discrete Sobolev inequality holds: (2.1) $\left\|\left\\{v_{j}\right\\}_{j}\right\|_{\ell^{\infty}}\leq\frac{1}{\sqrt{2}}\left\|\left\\{v_{j}\right\\}_{j}\right\|_{h^{1}}.$ The discrete product rule takes the form (2.2) $D_{\pm}\left(v_{j}w_{j}\right)=v_{j}D_{\pm}w_{j}+D_{\pm}v_{j}w_{j\pm 1}.$ Moreover, the discrete chain rule states (2.3) $D_{\pm}f(v_{j})=f^{\prime}(v_{j})D_{\pm}v_{j}\pm\frac{{\Delta x}}{2}f^{\prime\prime}(\xi_{j}^{\pm})(D_{\pm}v_{j})^{2},\qquad f\in C^{2},$ for some number $\xi_{j}^{\pm}$ between $v_{j\pm 1}$ and $v_{j}$. We continue to collect some handy results for later use, starting with a discrete Gronwall inequality. ###### Lemma 2.1. Assume that $c^{k}\geq 0$ and $f^{k}\geq 0$ for all $k=0,\dots,N$, and that the sequence $\left\\{u^{n}\right\\}_{n=0}^{N}$ satisfies the difference inequality (2.4) $D_{+}^{t}u^{n}+f^{n}\leq c^{n}u^{n},\qquad n=0,\dots,N-1.$ If $u^{n}\geq 0$ for all $n=0,\dots,N$, then $u^{N}+\exp\left({\Delta t}\sum_{n=0}^{N-1}c^{n}\right){\Delta t}\sum_{n=0}^{N-1}\exp\left(-{\Delta t}\sum_{k=0}^{n}c^{k}\right)f^{n}\leq\exp\left({\Delta t}\sum_{n=0}^{N-1}c^{n}\right)u^{0}.$ ###### Proof. Set $R^{n}=\exp\left(-{\Delta t}\sum_{k=0}^{n-1}c^{k}\right)$. Then we have $\displaystyle D_{+}^{t}R^{n}$ $\displaystyle=\frac{1}{{\Delta t}}\left(R^{n+1}-R^{n}\right)$ $\displaystyle=\exp\left(-{\Delta t}\sum_{k=0}^{n}c^{k}\right)\frac{1}{{\Delta t}}\left(1-\exp\left({\Delta t}c^{n}\right)\right)$ $\displaystyle\leq R^{n+1}\frac{1}{{\Delta t}}\left(1-(1+c^{n}{\Delta t})\right)=-c^{n}R^{n+1}.$ Hence, multiplying (2.4) with $R^{n+1}$ we arrive at $D_{+}^{t}\left(R^{n}u^{n}\right)=D_{+}^{t}u^{n}R^{n+1}+D_{+}^{t}R^{n}u^{n}\leq-f^{n}R^{n+1}.$ Multiplying this by ${\Delta t}$ and summing over $n$, we see that the lemma holds. ∎ The next lemma contains estimates for the solution of a discrete version of the differential equation $P-\partial_{xx}^{2}P=f$. ###### Lemma 2.2. Let $\left\\{f_{j}\right\\}_{j\in\mathbb{Z}}$ be a sequence in $\ell^{1}\cap\ell^{2}$, and denote by $\left\\{P_{j}\right\\}_{j\in\mathbb{Z}}$ the solution to the difference equation (2.5) $P_{j}-D_{-}D_{+}P_{j}=f_{j},\qquad j\in\mathbb{Z}.$ Introducing the notations $h=\left(1+2\frac{1-e^{-\kappa}}{({\Delta x})^{2}}\right)^{-1},\quad\kappa=\ln\left(1+\frac{{\Delta x}^{2}}{2}+\frac{{\Delta x}}{2}\sqrt{4+{\Delta x}^{2}}\right),$ the solution $\left\\{P_{j}\right\\}_{j\in\mathbb{Z}}$ takes the form (2.6) $P_{j}=h\sum\limits_{i\in\mathbb{Z}}e^{-\kappa\left|j-i\right|}f_{i},\qquad j\in\mathbb{Z}.$ Moreover, the following estimates hold: (2.7) $\displaystyle\left\|\left\\{P_{j}\right\\}_{j}\right\|_{\ell^{\infty}},\left\|\left\\{P_{j}\right\\}_{j}\right\|_{\ell^{1}}$ $\displaystyle\leq C\left\|\left\\{f_{j}\right\\}\right\|_{\ell^{1}},$ (2.8) $\displaystyle\left\|\left\\{D_{+}P_{j}\right\\}_{j}\right\|_{\ell^{\infty}},\left\|\left\\{D_{+}P_{j}\right\\}_{j}\right\|_{\ell^{1}}$ $\displaystyle\leq C\left\|\left\\{f_{j}\right\\}_{j}\right\|_{\ell^{1}},$ (2.9) $\displaystyle\left\|\left\\{P_{j}\right\\}_{j}\right\|_{h^{1}}$ $\displaystyle\leq C\left\|\left\\{f_{j}\right\\}_{j}\right\|_{\ell^{2}},$ where $C>0$ is a constant independent of ${\Delta x}$. ###### Proof. To verify the solution formula (2.6), we define $p_{i}$ by $p_{i}=ce^{-\sigma\left|i\right|},$ for some constants $c$ and $\sigma$ yet to be found. We shall choose these so that $(I-D_{+}D_{-})p_{i}=\begin{cases}1,&\text{if $i=0$,}\\\ 0,&\text{otherwise.}\end{cases}$ If we find that this holds with $\sigma=\kappa$ and $c=h$ then (2.6) holds. We observe that for $i\neq 0$ $D_{+}D_{-}p_{i}=ce^{-\sigma\left|i\right|}2\frac{\cosh(\sigma)-1}{{\Delta x}^{2}}.$ Hence $\sigma$ must satisfy $\sigma=\cosh^{-1}\left(1+\frac{{\Delta x}^{2}}{2}\right)=\kappa.$ For $i=0$ we find that $p_{0}-D_{+}D_{-}p_{0}=c\left(1-\frac{2}{{\Delta x}^{2}}\left(e^{-\kappa}-1\right)\right).$ If this is to be equal $1$ then $c=h$. For later use, one should observe that (2.10) $h=\frac{{\Delta x}}{2}+{\mathcal{O}}({\Delta x}^{2}),\quad\frac{\left|e^{\kappa}-1\right|}{{\Delta x}}=1+\mathcal{O}({\Delta x}),\quad\frac{\left|e^{-\kappa}-1\right|}{{\Delta x}}=1+\mathcal{O}({\Delta x}).$ For any $j\in\mathbb{Z}$, we have $\left|P_{j}\right|\lesssim\left\|\left\\{f_{j}\right\\}\right\|_{\ell^{1}}$ Furthermore, $\displaystyle\left\|\left\\{P_{j}\right\\}\right\|_{\ell^{1}}\leq 2h\sum\limits_{i}\bigg{[}\Delta x\sum\limits_{j}e^{-\kappa\left|j-i\right|}\bigg{]}\left|f_{i}\right|\lesssim\left\|\left\\{f_{i}\right\\}\right\|_{\ell^{1}},$ Hence, we have proved (2.7). From (2.6), $\displaystyle D_{+}P_{j}$ $\displaystyle=\frac{P_{j+1}-P_{j}}{{\Delta x}}$ $\displaystyle=h\sum\limits_{i}\frac{e^{-\kappa\left|i-j-1\right|}-e^{-\kappa\left|i-j\right|}}{{\Delta x}}f_{i}$ $\displaystyle=h\sum\limits_{i=j}^{\infty}\frac{e^{-\kappa(i-j-1)}-e^{-\kappa(i-j)}}{{\Delta x}}f_{i}+h\sum\limits_{i=-\infty}^{j-1}\frac{e^{\kappa(i-j-1)}-e^{\kappa(i-j)}}{{\Delta x}}f_{i}$ $\displaystyle=h\sum\limits_{i=j}^{\infty}e^{-\kappa(i-j)}\frac{e^{\kappa}-1}{{\Delta x}}f_{i}+h\sum\limits_{i=-\infty}^{j-1}e^{\kappa(i-j)}\frac{e^{-\kappa}-1}{{\Delta x}}f_{i}.$ Using (2.10) we acquire from this the following two estimates: $\left|D_{+}P_{j}\right|\lesssim h\sum\limits_{i}e^{-\kappa\left|i-j\right|}\left|f_{i}\right|\lesssim\left\|\left\\{f_{i}\right\\}\right\|_{\ell^{1}}$ and $\left\|\left\\{D_{+}P_{j}\right\\}_{j}\right\|_{\ell^{1}}\lesssim h{\Delta x}\sum\limits_{j,i}e^{-\kappa\left|i-j\right|}\left|f_{i}\right|\lesssim\left\|\left\\{f_{i}\right\\}\right\|_{\ell^{1}},$ Therefore (2.8) holds. It remains to prove (2.9). To this end, we multiply the equation (2.5) by ${\Delta x}P_{j}$ and perform a summation by parts to discover $\displaystyle\left\|\left\\{P_{j}\right\\}_{j}\right\|_{h^{1}}^{2}$ $\displaystyle={\Delta x}\sum_{j}P_{j}f_{j}\leq\frac{1}{2}\left\|\left\\{P_{j}\right\\}_{j}\right\|_{\ell^{2}}^{2}+\frac{1}{2}\left\|\left\\{f_{j}\right\\}_{j}\right\|_{\ell^{2}}^{2},$ from which (2.9) follows. ∎ We shall routinely use some well-known results related to weak convergence, which we collect in a lemma (for proofs, see, e.g., [24]). Throughout the paper we use overbars to denote weak limits. ###### Lemma 2.3. Let $O$ be a bounded open subset of $\mathbb{R}^{M}$, with $M\geq 1$. Let $\left\\{v_{n}\right\\}_{n\geq 1}$ be a sequence of measurable functions on $O$ for which $\sup_{n\geq 1}\int_{O}\Phi(\left|v_{n}(y)\right|)\,dy<\infty,$ for some given continuous function $\Phi:[0,\infty)\to[0,\infty)$. Then along a subsequence as $n\to\infty$ $g(v_{n})\rightharpoonup\overline{g(v)}$ in $L^{1}(O)$ for all continuous functions $g:\mathbb{R}\to\mathbb{R}$ satisfying $\lim_{\left|v\right|\to\infty}\frac{\left|g(v)\right|}{\Phi(\left|v\right|)}=0.$ Let $g\colon\mathbb{R}\to(-\infty,\infty]$ be a lower semicontinuous convex function and $\left\\{v_{n}\right\\}_{n\geq 1}$ a sequence of measurable functions on $O$, for which $\textit{$v_{n}\rightharpoonup v$ in $L^{1}(O)$, $g(v_{n})\in L^{1}(O)$ for each $n$, $g(v_{n})\rightharpoonup\overline{g(v)}$ in $L^{1}(O)$}.$ Then $g(v)\leq\overline{g(v)}$ a.e. on $O$. Moreover, $g(v)\in L^{1}(O)$ and $\int_{O}g(v)\,dy\leq\liminf_{n\to\infty}\int_{O}g(v_{n})\,dy.$ If, in addition, $g$ is strictly convex on an open interval $(a,b)\subset\mathbb{R}$ and $\text{$g(v)=\overline{g(v)}$ a.e.~{}on $O$},$ then, passing to a subsequence if necessary, $v_{n}(y)\to v(y)$ for a.e. $y\in\left\\{y\in O\mid v(y)\in(a,b)\right\\}$. Let $X$ be a Banach space and denote by $X^{\star}$ its dual. The space $X^{\star}$ equipped with the weak-$\star$ topology is denoted by $X^{\star}_{\mathrm{weak}}$, while $X$ equipped with the weak topology is denoted by $X_{\mathrm{weak}}$. By the Banach-Alaoglu theorem, a bounded ball in $X^{\star}$ is $\sigma(X^{\star},X)$-compact. If $X$ separable, then the weak-$\star$ topology is metrizable on bounded sets in $X^{\star}$, and thus one can consider the metric space $C\left([0,T];X^{\star}_{\mathrm{weak}}\right)$ of functions $v:[0,T]\to X^{\star}$ that are continuous with respect to the weak topology. We have $v_{n}\to v$ in $C\left([0,T];X^{\star}_{\mathrm{weak}}\right)$ if $\langle v_{n}(t),\phi\rangle_{X^{\star},X}\to\langle v(t),\phi\rangle_{X^{\star},X}$ uniformly with respect to $t$, for any $\phi\in X$. The following lemma is a consequence of the Arzelà-Ascoli theorem: ###### Lemma 2.4. Let $X$ be a separable Banach space, and suppose $v_{n}\colon[0,T]\to X^{\star}$, $n=1,2,\dots$, is a sequence of measurable functions such that $\left\|v_{n}\right\|_{L^{\infty}([0,T];X^{\star})}\leq C,$ for some constant $C$ independent of $n$. Suppose the sequence $[0,T]\ni t\mapsto\langle v_{n}(t),\Phi\rangle_{X^{\star},X},\quad n=1,2,\dots,$ is equi-continuous for every $\Phi$ that belongs to a dense subset of $X$. Then $v_{n}$ belongs to $C\left([0,T];X^{\star}_{\mathrm{weak}}\right)$ for every $n=1,2,\dots$, and there exists a function $v\in C\left([0,T];X^{\star}_{\mathrm{weak}}\right)$ such that along a subsequence as $n\to\infty$ $\text{$v_{n}\to v$ in $C\left([0,T];X^{\star}_{\mathrm{weak}}\right)$}.$ ## 3\. Explicit scheme and main result In this section we present the fully discrete (explicit) difference scheme for the Cammassa-Holm equation (1.3), which generates sequences $\left\\{u_{j+1/2}^{n}\right\\}$ and $\left\\{P_{j}^{n}\right\\}$ for $(n,j)\in\left\\{0,\dots,N\right\\}\times\mathbb{Z}$. We let $\left\\{u_{j+1/2}^{n}\right\\}$ solve the explicit difference equation (3.1) $D_{+}^{t}u_{j+1/2}^{n}+\left(u_{j+1/2}^{n}\vee 0\right)D_{-}u_{j+1/2}^{n}+\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}u_{j+1/2}^{n}+D_{+}P_{j}^{n}=0,$ where the initial values are specified as follows: (3.2) $u_{j+1/2}^{0}=u_{0}(x_{j+1/2}),$ Given $\left\\{u_{j+1/2}^{n}\right\\}$, we determine $\left\\{P_{j}^{n}\right\\}$ by solving (3.3) $-D_{-}D_{+}P_{j}^{n}+P_{j}^{n}=\left(u_{j+1/2}^{n}\vee 0\right)^{2}+\left(u_{j-1/2}^{n}\wedge 0\right)^{2}+\frac{1}{2}\left(D_{-}u_{j+1/2}^{n}\right)^{2},$ which is a linear system of equations that can be solved as outlined in Lemma 2.2. Next, let us derive the difference scheme satisfied by (3.4) $q_{j}^{n}=D_{-}u_{j+1/2}^{n}.$ This will be done by applying the difference operator $D_{-}$ to the $u$-equation (3.1). To this end, we apply the discrete product rule to find $D_{-}\left[\left(u_{j+1/2}^{n}\vee 0\right)D_{-}u_{j+1/2}^{n}\right]=\left(u_{j-1/2}^{n}\vee 0\right)D_{-}q_{j}+D_{-}\left(u_{j+1/2}^{n}\vee 0\right)q_{j}^{n}$ and $D_{-}\left[\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}u_{j+1/2}^{n}\right]=\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}q_{j}+D_{-}\left(u_{j+1/2}^{n}\wedge 0\right)q_{j},$ so that (3.5) $\begin{split}&D_{-}\left[\left(u_{j+1/2}^{n}\vee 0\right)D_{-}u_{j+1/2}^{n}+\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}u_{j+1/2}^{n}\right]\\\ &\qquad\qquad=\left(u_{j-1/2}^{n}\vee 0\right)D_{-}q_{j}^{n}+\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}q_{j}^{n}+\left({q_{j}^{n}}\right)^{2}.\end{split}$ The $P$-equation (3.3) rephrased in terms of $q$ reads (3.6) $-D_{-}D_{+}P_{j}^{n}+P_{j}^{n}=\left(u_{j+1/2}^{n}\vee 0\right)^{2}+\left(u_{j-1/2}^{n}\wedge 0\right)^{2}+\frac{1}{2}\left({q_{j}^{n}}\right)^{2}.$ Employing (3.5) and (3.6) when applying $D_{-}$ to the $u$-equation in (3.1) yields (3.7) $\begin{split}D_{+}^{t}q_{j}^{n}&+\left(u_{j-1/2}^{n}\vee 0\right)D_{-}q_{j}^{n}+\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}q_{j}^{n}\\\ &+\frac{\left({q_{j}^{n}}\right)^{2}}{2}+P_{j}^{n}-\left(u_{j+1/2}^{n}\vee 0\right)^{2}-\left(u_{j-1/2}^{n}\wedge 0\right)^{2}=0.\end{split}$ Regarding the initial values, in view of (3.4) and (3.2), we observe that (3.8) $q_{j}^{0}=\frac{1}{{\Delta x}}\int_{I_{j}}\partial_{x}u_{0}(x)\,dx,\qquad j\in\mathbb{Z}.$ Since the variable $q=\partial_{x}u$ can be discontinuous, (3.7) represents a natural upwind discretization of the equation for $q$, $\partial_{t}q+u\partial_{x}q+\frac{q^{2}}{2}-u^{2}+P=0$. The main result of this paper is the convergence of the scheme to a dissipative weak solution of (1.1)-(1.2), which is defined in the following sense [39, 40]: ###### Definition 3.1. Fix a final time $T>0$. We call a function $u\colon[0,T]\times\mathbb{R}\rightarrow\mathbb{R}$ a weak solution of the Cauchy problem for (1.1)-(1.2) on $[0,T]\times\mathbb{R}$ if 1. (D.1) $u\in C([0,T]\times\mathbb{R})\cap L^{\infty}(0,T;H^{1}(\mathbb{R}))$; 2. (D.2) For all $s$ and $t$ in $[0,T]$, with $s\leq t$, we have $\left\|u(t,\cdot)\right\|_{H^{1}(\mathbb{R})}\leq\left\|u(s,\cdot)\right\|_{H^{1}(\mathbb{R})}$. 3. (D.3) $u$ satisfies (1.3) in the sense of distributions on $(0,T)\times\mathbb{R}$; 4. (D.4) $u(0,x)=u_{0}(x)$ for every $x\in\mathbb{R}$; 5. (D.5) If, in addition, there exists a positive constant $K$ such that $u_{x}(t,x)\leq\frac{2}{t}+K\left\|u_{0}\right\|_{H^{1}(\mathbb{R})}^{2},\qquad(t,x)\in(0,T]\times\mathbb{R},$ then we call $u$ a dissipative weak solution of the Cauchy problem (1.1)-(1.2). In addition to $\partial_{x}u\in L^{\infty}(0,T;L^{2}(\mathbb{R}))$, cf. (D.2), the dissipative weak solutions $u$ that we construct in this paper will possess an improved integrability property, namely $\partial_{x}u\in L^{p}_{\mathrm{loc}}((0,T)\times\mathbb{R})$ for $p<3$, i.e., $\int_{0}^{T}\int_{a}^{b}\left|\partial_{x}u\right|^{p}\,dx\,dt\leq C(a,b,T,p),\quad\text{$\forall a,b\in\mathbb{R}$, $a<b$.}$ To state our main convergence result and also for later use, we need to introduce some functions (interpolations of the difference approximations) that are defined at all points $(t,x)$ in the domain. We begin by defining the functions $\left.\begin{aligned} q_{j}(t)&=q_{j}^{n}+\left(t-t^{n}\right)D_{+}^{t}q_{j}^{n},\\\ u_{j+1/2}(t)&=u_{j+1/2}^{n}+\left(t-t^{n}\right)D_{+}^{t}u_{j+1/2}^{n}\end{aligned}\right\\}\quad\text{for $t\in I^{n}$.}$ With the aid of these we define (3.9) $q_{{\Delta x}}(t,x)=q_{j}(t),\qquad(t,x)\in I_{j}^{n}$ and (3.10) $u_{{\Delta x}}(t,x)=u_{j-1/2}(t)+\left(x-x_{j-1/2}\right)q_{j}(t),\qquad\text{for $(t,x)\in I_{j}^{n}$,}$ for $j\in\mathbb{Z}$, $n=0,\dots,N-1$. Note that $t\mapsto u_{{\Delta x}}(t,x)$ is a continuous function, since $u_{j-1/2}(t)$ and $q_{j}(t)$ are continuous. Regarding the continuity in $x$ we have that $\displaystyle\lim_{x\uparrow x_{j+1/2}}u_{{\Delta x}}(t,x)$ $\displaystyle=u_{j-1/2}(t)+\left(x_{x+1/2}-x_{j-1/2}\right)q_{j}(t)$ $\displaystyle=u_{j-1/2}^{n}+\left(t-t^{n}\right)D_{+}^{t}u_{j-1/2}^{n}+{\Delta x}\left(q_{j}^{n}+\left(t-t^{n}\right)D_{+}^{t}q_{j}^{n}\right)$ $\displaystyle=u_{j-1/2}^{n}+\left(t-t^{n}\right)D_{+}^{t}u_{j-1/2}^{n}+\left(u_{j+1/2}^{n}-u_{j-1/2}^{n}\right)$ $\displaystyle\qquad\quad+\left(t-t^{n}\right)D_{+}^{t}\left(u_{j+1/2}^{n}-u_{j-1/2}^{n}\right)=u_{j+1/2}(t),$ and therefore $u_{{\Delta x}}$ is continuous, and furthermore $\partial_{x}u_{{\Delta x}}=q_{{\Delta x}}$ almost everywhere. Observe also that, due to (3.8), there holds $q_{{\Delta x}}(0,x)\to\partial_{x}u_{0}$ in $L^{2}(\mathbb{R})$ as ${\Delta x}\to 0$. Similarly to $u_{{\Delta x}}$, we define a function $P_{{\Delta x}}$ by bilinear interpolation. First, let $P_{j}(t)=P^{n}_{j}+(t-t^{n})D_{+}^{t}P_{j}^{n},\qquad t\in I^{n},$ and then define (3.11) $P_{{\Delta x}}(t,x)=P_{j}(t)+(x-x_{j})D_{+}P_{j}(t),\qquad(t,x)\in I_{j}^{n},$ for $j\in\mathbb{Z}$, $n=0,\dots,N-1$. We are now in a position to state our main result. ###### Theorem 3.1. Suppose (1.2) holds. Let $\left\\{u_{{\Delta x}}\right\\}_{{\Delta x}>0}$ be a sequence defined by (3.10) and (3.1)-(3.4). Then, along a subsequence as ${\Delta x}\downarrow 0$, $\text{$u_{{\Delta x}}\to u$ in $H_{\mathrm{loc}}^{1}((0,T)\times\mathbb{R})$},$ where $u$ is a dissipative weak solution of the Cauchy problem (1.1)-(1.2). This theorem is a consequence of the results stated and proved in Sections 4-8. ## 4\. Total energy estimate and some consequences The purpose of this section is establish a discrete total energy estimate for the difference scheme (3.1)-(3.4). ###### Lemma 4.1. Assume that ${\Delta x}$ and ${\Delta t}$ are related through the CFL type condition (4.1) ${\Delta t}<\frac{\log(1+{\Delta x}^{\theta})\,{\Delta x}^{2}}{C\left\|u_{0}\right\|_{H^{1}(\mathbb{R})}^{2}\left(1+{\Delta x}^{2}\right)},$ where $C$ is a constant (to be detailed in the proof of the lemma) that is independent of ${\Delta x}$ and $u_{0}$ and $\theta>0$. Then, for any $N_{0}\in\left\\{0,\dots,N\right\\}$, and for all sufficiently small ${\Delta x}$, (4.2) $\left\|u^{N_{0}}\right\|_{h^{1}}^{2}+{\Delta x}^{2}{\Delta t}\sum_{n=0}^{N_{0}-1}\sum_{j\in\mathbb{Z}}\left|u_{j+1/2}^{n}\right|\left(D_{-}D_{+}u_{j+1/2}^{n}\right)^{2}\leq e^{t^{N}{\Delta x}^{\theta}}\left\|u^{0}\right\|_{h^{1}}^{2}.$ ###### Proof. For the proof of (4.2), we shall need to introduce an auxiliary difference scheme. To this end, we start by defining the cut-off function $f^{M}(u)=\begin{cases}-M,\;&u<M,\\\ u,\;&u\in[-M,M],\\\ +M\;&u>M,\end{cases}$ where $M>0$ is a fixed constant (to be determined later on). Now, let $\left\\{\tilde{u}_{j+1/2}^{n}\right\\}$ and $\left\\{\tilde{P}_{j}^{n}\right\\}$ solve the following system of difference equations: (4.3) $\displaystyle D_{+}^{t}\tilde{u}_{j+1/2}^{n}$ $\displaystyle+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)D_{-}\tilde{u}_{j+1/2}^{n}$ $\displaystyle+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)D_{+}\tilde{u}_{j+1/2}^{n}+D_{+}\tilde{P}_{j}^{n}=0,$ for $n=0,\dots,N-1$ and $j\in\mathbb{Z}$, and $\displaystyle-D_{-}D_{+}\tilde{P}_{j}^{n}+\tilde{P}_{j}^{n}$ $\displaystyle=\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)$ $\displaystyle\qquad+\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)\left(\tilde{u}_{j-1/2}^{n}\wedge 0\right)$ $\displaystyle\qquad+\frac{1}{2}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)D_{-}\tilde{u}_{j+1/2}^{n},$ for $n=0,\dots,N$ and $j\in\mathbb{Z}$. Regarding the initial data, we set $\tilde{u}_{j}^{0}=u_{j}^{0}$ for $j\in\mathbb{Z}$. If we define $\tilde{q}_{j}^{n}:=D_{-}\tilde{u}_{j+1/2}^{n}$, then it is straightforward to see that $\left\\{\tilde{q}_{j}^{n}\right\\}$ satisfies the difference equation (4.4) $\begin{split}D_{+}^{t}\tilde{q}_{j}^{n}&+\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)D_{-}\tilde{q}_{j}^{n}+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)D_{+}\tilde{q}_{j}^{n}\\\ &+\frac{1}{2}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)\tilde{q}_{j}^{n}\\\ &-\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)\\\ &-\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)\left(\tilde{u}_{j-1/2}^{n}\wedge 0\right)+\tilde{P}_{j}^{n}=0.\end{split}$ Multiplying (4.3) by $\tilde{u}_{j+1/2}^{n}$ we find that (4.5) $\begin{split}&\tilde{u}_{j+1/2}^{n}D_{+}^{t}\tilde{u}_{j+1/2}^{n}+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)\tilde{q}_{j}^{n}\\\ &\quad+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)\left(\tilde{u}_{j+1/2}^{n}\wedge 0\right)\tilde{q}_{j+1}^{n}+D_{+}\tilde{P}_{j}^{n}\tilde{u}_{j+1/2}^{n}=0,\end{split}$ while multiplying (4.4) by $\tilde{q}_{j}^{n}$ gives us (4.6) $\begin{split}\tilde{q}_{j}^{n}D_{+}^{t}\tilde{q}_{j}^{n}&+\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)(D_{-}\tilde{q}_{j}^{n})\tilde{q}_{j}^{n}+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)(D_{+}\tilde{q}_{j}^{n})\tilde{q}_{j}^{n}\\\ &+\frac{1}{2}D_{-}\left(f^{M}(\tilde{u}_{j+1/2}^{n})\right)\left(\tilde{q}_{j}^{n}\right)^{2}\\\ &-\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)\tilde{q}_{j}^{n}\\\ &-\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)\left(\tilde{u}_{j-1/2}^{n}\wedge 0\right)\tilde{q}_{j}^{n}+\tilde{P}_{j}^{n}\tilde{q}_{j}^{n}=0.\end{split}$ Adding (4.5) and (4.6), multiplying the result with ${\Delta x}$, and summing over $j$ yields ${\Delta x}\sum_{j\in\mathbb{Z}}\left(\tilde{u}_{j+1/2}^{n}D_{+}^{t}\tilde{u}_{j+1/2}^{n}+\tilde{q}_{j}^{n}D_{+}^{t}\tilde{q}_{j}^{n}\right)+\text{I}+\text{II}+\text{III}=0,$ where I $\displaystyle={\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)(D_{-}\tilde{q}_{j}^{n})\tilde{q}_{j}^{n}$ $\displaystyle\qquad+{\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)(D_{+}\tilde{q}_{j}^{n})\tilde{q}_{j}^{n}+\frac{{\Delta x}}{2}\sum_{j\in\mathbb{Z}}D_{-}\left(f^{M}\left(u_{j+1/2}\right)\right)\left(\tilde{q}_{j}^{n}\right)^{2},$ II $\displaystyle={\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)\tilde{q}_{j}^{n}$ $\displaystyle\qquad+{\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)\left(\tilde{u}_{j+1/2}^{n}\wedge 0\right)\tilde{q}_{j+1}^{n}$ $\displaystyle\qquad-{\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)\tilde{q}_{j}^{n}$ $\displaystyle\qquad-{\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)\left(\tilde{u}_{j-1/2}^{n}\wedge 0\right)\tilde{q}_{j}^{n}\equiv 0\quad\text{(by shifting indices)},$ III $\displaystyle={\Delta x}\sum_{j\in\mathbb{Z}}D_{+}\tilde{P}_{j}^{n}\tilde{u}_{j+1/2}^{n}+{\Delta x}\sum_{j\in\mathbb{Z}}\tilde{P}_{j}^{n}\tilde{q}_{j}^{n}\equiv 0\quad\text{(by summation by parts, cf.~{}\eqref{eq:qjdef}).}$ Let us now deal with term I. The discrete chain rule (2.3) tells us that $(D_{\pm}\tilde{q}_{j}^{n})\tilde{q}_{j}^{n}=D_{\pm}\left(\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right)\mp\frac{{\Delta x}}{2}(D_{\pm}\tilde{q}_{j}^{n})^{2}.$ Hence I $\displaystyle={\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)\left[D_{-}\left(\frac{(\tilde{q}_{j}^{n})^{2}}{2}\right)+\frac{{\Delta x}}{2}(D_{-}\tilde{q}_{j}^{n})^{2}\right]$ $\displaystyle\qquad+{\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)\left[D_{+}\left(\frac{(\tilde{q}_{j}^{n})^{2}}{2}\right)-\frac{{\Delta x}}{2}(D_{+}\tilde{q}_{j}^{n})^{2}\right]$ $\displaystyle\qquad+\frac{{\Delta x}}{2}\sum_{j\in\mathbb{Z}}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)(\tilde{q}_{j}^{n})^{2}=\text{I}_{1}+\text{I}_{2},$ where $\displaystyle\text{I}_{1}$ $\displaystyle={\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)D_{-}\left(\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right)$ $\displaystyle\qquad+{\Delta x}\sum_{j\in\mathbb{Z}}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)D_{+}\left(\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right)$ $\displaystyle\qquad+\frac{{\Delta x}}{2}\sum_{j\in\mathbb{Z}}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)(\tilde{q}_{j}^{n})^{2},$ $\displaystyle\text{I}_{2}$ $\displaystyle=\frac{{\Delta x}^{2}}{2}\sum_{j\in\mathbb{Z}}\Bigl{[}\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)(D_{-}\tilde{q}_{j}^{n})^{2}-\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)(D_{+}\tilde{q}_{j}^{n})^{2}\Bigr{]}$ $\displaystyle=\frac{{\Delta x}^{2}}{2}\sum_{j\in\mathbb{Z}}\Bigl{[}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)(D_{+}\tilde{q}_{j}^{n})^{2}-\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)(D_{+}\tilde{q}_{j}^{n})^{2}\Bigr{]}$ $\displaystyle=\frac{{\Delta x}^{2}}{2}\sum_{j\in\mathbb{Z}}\left|f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right|\left(D_{+}\tilde{q}_{j}^{n}\right)^{2}\geq 0.$ To handle the $\text{I}_{1}$-term, we use the discrete product rule (2.2): $\displaystyle D_{-}\left[\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right]$ $\displaystyle\quad=\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)D_{-}\left(\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right)+D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2},$ $\displaystyle D_{+}\left[\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)\frac{(\tilde{q}_{j}^{n})^{2}}{2}\right]$ $\displaystyle\quad=\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)D_{+}\left(\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right)+D_{+}\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}$ $\displaystyle\quad=\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)D_{+}\left(\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right)+D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}.$ Using this we find that $\displaystyle\text{I}_{1}$ $\displaystyle={\Delta x}\sum_{j\in\mathbb{Z}}D_{-}\left[\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right]$ $\displaystyle\qquad-{\Delta x}\sum_{j\in\mathbb{Z}}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}$ $\displaystyle\qquad+{\Delta x}\sum_{j\in\mathbb{Z}}D_{+}\left[\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}\right]$ $\displaystyle\qquad-{\Delta x}\sum_{j\in\mathbb{Z}}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}$ $\displaystyle\qquad+\frac{{\Delta x}}{2}\sum_{j\in\mathbb{Z}}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)\left(\tilde{q}_{j}^{n}\right)^{2}$ $\displaystyle=-{\Delta x}\sum_{j\in\mathbb{Z}}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)\frac{\left(\tilde{q}_{j}^{n}\right)^{2}}{2}$ $\displaystyle\qquad\qquad+\frac{{\Delta x}}{2}\sum_{j\in\mathbb{Z}}D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)\left(\tilde{q}_{j}^{n}\right)^{2}=0.$ Summarizing our findings so far: (4.7) $\begin{split}&{\Delta x}\sum_{j\in\mathbb{Z}}\left(\tilde{u}_{j+1/2}^{n}D_{+}^{t}\tilde{u}_{j+1/2}^{n}+\tilde{q}_{j}^{n}D_{+}^{t}\tilde{q}_{j}^{n}\right)\\\ &\qquad+\frac{{\Delta x}^{2}}{2}\sum_{j\in\mathbb{Z}}\left|f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right|\left(D_{+}\tilde{q}_{j}^{n}\right)^{2}=0.\end{split}$ Next, by (2.3), (4.8) $\begin{split}&{\Delta x}\sum_{j\in\mathbb{Z}}\left(\tilde{u}_{j+1/2}^{n}D_{+}^{t}\tilde{u}_{j+1/2}^{n}+q_{j}D_{+}^{t}\tilde{q}_{j}^{n}\right)\\\ &\quad=D_{+}^{t}\left[\frac{{\Delta x}}{2}\sum_{j\in\mathbb{Z}}\left((\tilde{u}_{j+1/2}^{n})^{2}+(\tilde{q}_{j}^{n})^{2}\right)\right]\\\ &\quad\quad\quad-\frac{1}{2}{\Delta t}{\Delta x}\sum_{j\in\mathbb{Z}}\left(\left(D_{+}^{t}\tilde{u}_{j+1/2}^{n}\right)^{2}+\left(D_{+}^{t}\tilde{q}_{j}^{n}\right)^{2}\right).\end{split}$ Hence, we must now estimate ${\Delta t}{\Delta x}\sum_{j\in\mathbb{Z}}\left(\left(D_{+}^{t}\tilde{u}_{j+1/2}^{n}\right)^{2}+\left(D_{+}^{t}\tilde{q}_{j}^{n}\right)^{2}\right).$ Using (4.3), (4.4), and the basic inequality $\left(\sum_{\ell=1}^{l}a_{l}\right)^{2}\leq 2^{l-1}\sum_{\ell=1}^{l}(a_{\ell})^{2}$, which holds for any sequence $\left\\{a_{\ell}\right\\}_{\ell=1}^{l}$ of positive real numbers, there is a positive constant $c_{1}$ that does not depend on ${\Delta x}$ such that $\displaystyle{\Delta t}{\Delta x}\sum_{j\in\mathbb{Z}}\left(\left(D_{+}^{t}\tilde{u}_{j+1/2}^{n}\right)^{2}+\left(D_{+}^{t}\tilde{q}_{j}^{n}\right)^{2}\right)$ $\displaystyle\qquad\leq c_{1}{\Delta t}{\Delta x}\sum\limits_{j\in\mathbb{Z}}\Biggl{[}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)^{2}\left(\tilde{q}_{j}^{n}\right)^{2}$ $\displaystyle\quad\qquad\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)^{2}\left(\tilde{q}_{j+1}^{n}\right)^{2}+\left(D_{+}\tilde{P}_{j}^{n}\right)^{2}$ $\displaystyle\quad\qquad\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)^{2}\left(D_{-}\tilde{q}_{j}^{n}\right)^{2}$ $\displaystyle\quad\qquad\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)^{2}\left(D_{+}\tilde{q}_{j}^{n}\right)^{2}$ $\displaystyle\quad\qquad\qquad\qquad\qquad+\left(D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)\right)^{2}\left(\tilde{q}_{j}^{n}\right)^{2}$ $\displaystyle\quad\qquad\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)^{2}\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)^{2}$ $\displaystyle\quad\qquad\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)^{2}\left(\tilde{u}_{j-1/2}^{n}\wedge 0\right)^{2}+\left(\tilde{P}_{j}^{n}\right)^{2}\Biggr{]}$ $\displaystyle\qquad\leq c_{1}{\Delta t}\left(J_{1}+J_{2}+J_{3}\right),$ where $\displaystyle J_{1}$ $\displaystyle={\Delta x}\sum\limits_{j\in\mathbb{Z}}\Biggl{[}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)^{2}\left(\tilde{q}_{j}^{n}\right)^{2}+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)^{2}\left(\tilde{q}_{j+1}^{n}\right)^{2}$ $\displaystyle\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)^{2}\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)^{2}$ $\displaystyle\qquad\qquad\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)^{2}\left(\tilde{u}_{j-1/2}^{n}\wedge 0\right)^{2}\Biggr{]},$ $\displaystyle J_{2}$ $\displaystyle={\Delta x}\sum\limits_{j\in\mathbb{Z}}\Biggl{[}\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)^{2}\left(D_{-}\tilde{q}_{j}^{n}\right)^{2}$ $\displaystyle\qquad\qquad\quad+\left(f^{M}\left(u_{j+1/2}\right)\wedge 0\right)^{2}\left(D_{+}\tilde{q}_{j}^{n}\right)^{2}+\left(D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)\right)^{2}\left(\tilde{q}_{j}^{n}\right)^{2}\Biggr{]},$ $\displaystyle J_{3}$ $\displaystyle={\Delta x}\sum\limits_{j\in\mathbb{Z}}\Biggl{[}\left(D_{+}\tilde{P}_{j}^{n}\right)^{2}+\left(\tilde{P}_{j}^{n}\right)^{2}\Biggr{]}.$ Since $\left|f^{M}(u)\right|\leq M$, the following bounds hold: (4.9) $J_{1}\leq 2M^{2}{\Delta x}\sum\limits_{j\in\mathbb{Z}}\left[\left(\tilde{u}_{j+1/2}^{n}\right)^{2}+(\tilde{q}_{j}^{n})^{2}\right]=2M^{2}\left\|\tilde{u}^{n}\right\|_{h^{1}}$ and (4.10) $\begin{split}J_{2}&=\frac{{\Delta x}}{{\Delta x}^{2}}\sum\limits_{j\in\mathbb{Z}}\Biggl{[}\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\vee 0\right)^{2}\left(D_{-}\tilde{q}_{j}^{n}{\Delta x}\right)^{2}\\\ &\qquad\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\wedge 0\right)^{2}\left(D_{+}\tilde{q}_{j}^{n}{\Delta x}\right)^{2}\\\ &\qquad\qquad\qquad\qquad\qquad+\left(D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right){\Delta x}\right)^{2}\left(\tilde{q}_{j}^{n}\right)^{2}\Biggr{]}\\\ &\leq c_{2}M^{2}\frac{{\Delta x}}{{\Delta x}^{2}}\sum\limits_{j\in\mathbb{Z}}(q_{j})^{2}\leq c_{2}\frac{M^{2}}{{\Delta x}^{2}}\left\|\tilde{u}^{n}\right\|_{h^{1}},\end{split}$ for some constant $c_{2}>0$ independent of ${\Delta x}$. To estimate $J_{3}$ we use Lemma 2.2, specifically (2.9), which implies that (4.11) $\begin{split}J_{3}&\leq c_{3}{\Delta x}\sum\limits_{j\in\mathbb{Z}}\Biggl{[}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\vee 0\right)^{2}\left(\tilde{u}_{j+1/2}^{n}\vee 0\right)^{2}\\\ &\qquad\qquad\qquad\qquad+\left(f^{M}\left(\tilde{u}_{j-1/2}^{n}\right)\wedge 0\right)^{2}\left(\tilde{u}_{j-1/2}^{n}\wedge 0\right)^{2}\\\ &\qquad\qquad\qquad\qquad\qquad+\left(D_{-}\left(f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right)\right)^{2}\left(\tilde{q}_{j}^{n}\right)^{2}\Biggr{]}\\\ &\leq 2c_{3}M^{2}{\Delta x}\sum\limits_{i\in\mathbb{Z}}\left(\tilde{u}_{j+1/2}^{n}\right)^{2}+c_{3}M^{2}\frac{{\Delta x}}{{\Delta x}^{2}}\sum\limits_{i\in\mathbb{Z}}\left(\tilde{q}_{j}^{n}\right)^{2}\\\ &\leq c_{3}M^{2}\left(1+\frac{1}{{\Delta x}^{2}}\right)\left\|\tilde{u}^{n}\right\|_{h^{1}},\end{split}$ for some constant $c_{3}>0$ independent of ${\Delta x}$. Blending (4.9), (4.10), and (4.11) we derive the bound (4.12) ${\Delta t}{\Delta x}\sum_{j\in\mathbb{Z}}\left(\left(D_{+}^{t}\tilde{u}_{j+1/2}^{n}\right)^{2}+\left(D_{+}^{t}\tilde{q}_{j}^{n}\right)^{2}\right)\leq CM^{2}\left({\Delta t}+\frac{{\Delta t}}{{\Delta x}^{2}}\right)\left\|\tilde{u}^{n}\right\|_{h^{1}},$ where the constant $C$ is independent of ${\Delta x}$. Combining (4.7), (4.8), and (4.12), it follows that $\left\\{\tilde{u}_{j+1/2}^{n}\right\\}$ obeys the following discrete energy estimate: $D_{+}^{t}\left\|\tilde{u}^{n}\right\|_{h^{1}}^{2}+\underbrace{\frac{{\Delta x}^{2}}{2}\sum_{j\in\mathbb{Z}}\left|f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)\right|\left(D_{+}\tilde{q}_{j}^{n}\right)^{2}}_{=:Z^{n}}\leq\underbrace{CM^{2}\left({\Delta t}+\frac{{\Delta t}}{{\Delta x}^{2}}\right)}_{=:\omega}\left\|\tilde{u}^{n}\right\|_{h^{1}}^{2}.$ By the discrete Gronwall inequality, cf. Lemma 2.1, $\left\|\tilde{u}^{N_{0}}\right\|_{h^{1}}^{2}+\sum_{n=0}^{N_{0}-1}e^{CM^{2}\omega(t^{N_{0}-1}-t^{n})}Z^{n}\leq e^{CM^{2}\omega t^{N_{0}-1}}\left\|u^{0}\right\|_{h^{1}}^{2}.$ Choosing $M=\left\|u_{0}\right\|_{H^{1}(\mathbb{R})}$ and recalling the CFL type condition (4.1), we deduce $\displaystyle e^{CM^{2}\omega t^{N}}$ $\displaystyle=e^{C\left\|u_{0}\right\|_{H^{1}(\mathbb{R})}^{2}{\Delta t}^{2}((1+{\Delta x}^{2})/{\Delta x}^{2})N}$ $\displaystyle\leq(1+{\Delta x}^{\theta})^{N{\Delta t}}$ $\displaystyle\leq e^{t^{N}{\Delta x}^{\theta}}\leq 2\quad\text{if}\quad{\Delta x}^{\theta}\leq\frac{\log 2}{T}.$ Therefore, in particular $\left\|\tilde{u}^{n}\right\|_{h^{1}}\leq\sqrt{2}\left\|u^{0}\right\|_{h^{1}}\quad\text{for $n=0,\dots,N_{0}$.}$ By the discrete Sobolev inquality (2.1), we find that $\left\|\tilde{u}^{n}\right\|_{\ell^{\infty}}\leq\frac{1}{\sqrt{2}}\left\|\tilde{u}^{n}\right\|_{h^{1}}\leq\left\|u^{0}\right\|_{h^{1}}\leq M\quad\text{for $n=0,\dots,N_{0}$.}$ This means that $\tilde{u}^{n}$ will “never notice” $f^{M}$, since $f^{M}\left(\tilde{u}_{j+1/2}^{n}\right)=\tilde{u}_{j+1/2}^{n}\quad\text{for $j\in\mathbb{Z}$ and $n=0,\dots,N_{0}$.}$ Therefore, $\tilde{u}_{j+1/2}^{n}=u_{j+1/2}^{n}\quad\text{and}\quad\tilde{P}_{j}^{n}=P_{j}^{n}\quad\text{for $j\in\mathbb{Z}$ and $n=0,\dots,N_{0}$.}$ Finally, (4.2) follows by noting that $e^{CM^{2}\omega(t^{N_{0}-1}-t^{n})}\geq 1\quad\text{for $n=0,\dots,N_{0}-1$.}$ ∎ We conclude this section by stating some immediate consequences of (4.2). ###### Lemma 4.2. For $n=0,\dots,N$, $\displaystyle\left\|P^{n}\right\|_{\ell^{\infty}},\left\|P^{n}\right\|_{\ell^{1}}$ $\displaystyle\leq C\left\|u_{0}\right\|^{2}_{H^{1}(\mathbb{R})},$ $\displaystyle\left\|D_{+}P^{n}\right\|_{\ell^{\infty}},\left\|D_{+}P^{n}\right\|_{\ell^{1}}$ $\displaystyle\leq C\left\|u_{0}\right\|^{2}_{H^{1}(\mathbb{R})},$ where $C>0$ is a constant independent of ${\Delta x}$. ###### Proof. This follows immediately from Lemma 2.2, noting that in this case $f_{j}=\left(u_{j+1/2}^{n}\vee 0\right)^{2}+\left(u_{j-1/2}^{n}\wedge 0\right)^{2}+\frac{1}{2}\left(D_{-}u_{j+1/2}^{n}\right)^{2},$ and thus $\left\|f_{j}\right\|_{\ell^{1}}\leq\left\|u^{n}\right\|^{2}_{h^{1}}\leq(1+{\Delta x})^{2}\left\|u^{0}\right\|_{H^{1}(\mathbb{R})}^{2}.$ ∎ ## 5\. One-sided sup-norm estimate ###### Lemma 5.1. Assume that ${\Delta t}$ satisfies the CFL type condition (4.1) and that ${\Delta x}$ is sufficiently small. For $n=0,\dots,N$ and $j\in\mathbb{Z}$, we then have (5.1) $q_{j}^{n}\leq\frac{2}{t^{n}}+C\left\|u^{0}\right\|_{h^{1}},$ where $C>0$ is a finite constant. ###### Proof. We can write the difference equation for $\left\\{q_{j}^{n}\right\\}$, see (3.7), as $\begin{split}q^{n+1}_{j}&=q_{j}^{n}\left(1-\lambda a-\lambda b\right)+q_{j-1}^{n}\lambda a+q_{j+1}^{n}\lambda b-{\Delta t}\frac{{q_{j}^{n}}^{2}}{2}\\\ &\qquad+{\Delta t}\left(\left(u_{j-1/2}^{n}\vee 0\right)^{2}+\left(u_{j+1/2}^{n}\wedge 0\right)^{2}-P_{j}^{n}\right),\end{split}$ where $a=\lambda\left(u_{j-1/2}^{n}\vee 0\right),\quad b=-\lambda\left(u_{j+1/2}^{n}\wedge 0\right),\quad\lambda={\Delta t}/{\Delta x}.$ Now we have uniform bounds on $\left\|u^{n}\right\|_{\ell^{\infty}}$ and $\left\|P^{n}\right\|_{\ell^{\infty}}$ and thus (5.2) $q^{n+1}_{j}\leq q_{j}^{n}\left(1-\lambda a-\lambda b\right)+q_{j-1}^{n}\lambda a+q_{j+1}^{n}\lambda b-{\Delta t}\frac{{q_{j}^{n}}^{2}}{2}+{\Delta t}L,$ for some finite constant $L\lesssim\left\|u^{0}\right\|_{h^{1}}^{2}$. Set $\bar{q}^{n}_{j}=\max\left\\{q_{j}^{n},q_{j-1}^{n},q_{j+1}^{n}\right\\}$. We claim that (5.3) $q^{n+1}_{j}\leq\bar{q}^{n}_{j}-{\Delta t}\frac{\left(\bar{q}^{n}_{j}\right)^{2}}{2}+{\Delta t}L,$ if ${\Delta t}$ is chosen sufficiently small. First we choose ${\Delta t}$ so small that $\lambda(a+b)<1/2$. Then if $\bar{q}^{n}_{j}=q_{j}^{n}$ the claim follows immediately from (5.2). Next, assume $\bar{q}^{n}_{j}=q_{j-1}^{n}$. Then note that $\displaystyle\left(q_{j}^{n}\right)^{2}$ $\displaystyle=\left(q_{j-1}^{n}\right)^{2}+\left(q_{j}^{n}\right)^{2}-\left(q_{j-1}^{n}\right)^{2}$ $\displaystyle=\left(q_{j-1}^{n}\right)^{2}+\frac{q_{j}^{n}+q_{j-1}^{n}}{2}{\Delta x}D_{-}q_{j}^{n}$ $\displaystyle=\left(q_{j-1}^{n}\right)^{2}+\frac{u_{j+1/2}^{n}-u^{n}_{j-3/2}}{2}D_{-}q_{j}^{n}.$ Since $D_{-}q_{j}^{n}<0$, we find that $\left(q_{j}^{n}\right)^{2}\leq\left(q_{j-1}^{n}\right)^{2}-\left\|u^{n}\right\|_{\ell^{\infty}}D_{-}q_{j}^{n}.$ Using this we can rephrase (5.2) as $\displaystyle q^{n+1}_{j}$ $\displaystyle\leq q_{j}^{n}\left(1-\lambda\left(a+\left\|u^{n}\right\|_{\ell^{\infty}}+b\right)\right)+q_{j-1}^{n}\lambda\left(a+\left\|u^{n}\right\|_{\ell^{\infty}}\right)+q_{j+1}^{n}\lambda b$ $\displaystyle\qquad\quad-{\Delta t}\frac{\left(q_{j-1}^{n}\right)^{2}}{2}+{\Delta t}L$ $\displaystyle\leq\bar{q}^{n}_{j}-{\Delta t}\frac{\left(\bar{q}^{n}_{j}\right)^{2}}{2}+{\Delta t}L=:F\left(\bar{q}^{n}_{j}\right),$ if $\lambda\left\|u^{n}\right\|_{\ell^{\infty}}<\frac{1}{2}$. The proof of (5.3) if $\bar{q}^{n}_{j}=q_{j+1}^{n}$ is similar. Note that $F^{\prime}(q)=1-{\Delta t}q$, and thus $F$ is increasing for $q<1/{\Delta t}$. Furthermore, by the CFL type condition (4.1), ${\Delta t}=\mathcal{O}({\Delta x}^{3})$, and by the bounds on $\left\|u^{n}\right\|_{\ell^{\infty}}$, $q_{j}^{n}\leq\left|q_{j}^{n}\right|\leq\mathcal{O}\left(\frac{1}{{\Delta x}}\right)=\mathcal{O}\left(\frac{1}{{\Delta t}^{1/3}}\right)\leq\frac{1}{{\Delta t}},$ for sufficiently small ${\Delta t}$. Therefore, setting111This maximum exists since $q^{n}\in\ell^{2}$. $M^{n}=\max_{j}q_{j}^{n}$, from (5.3) we get $M^{n+1}\leq F\left(M^{n}\right).$ Now set $Z^{n}=M^{n}-\sqrt{2L}$. Then $\displaystyle Z^{n+1}$ $\displaystyle\leq F\left(Z^{n}+\sqrt{2L}\right)-\sqrt{2L}$ $\displaystyle=Z^{n}+\sqrt{2L}-\frac{{\Delta t}}{2}\left(Z^{n}+\sqrt{2L}\right)^{2}+L{\Delta t}-\sqrt{2L}$ $\displaystyle=Z^{n}\left(1-{\Delta t}\sqrt{2L}\right)-\frac{{\Delta t}}{2}\left(Z^{n}\right)^{2}$ $\displaystyle\leq Z^{n}-\frac{{\Delta t}}{2}\left(Z^{n}\right)^{2}.$ Now, clearly if $Z^{n}\leq 0$, then $Z^{n+1}\leq 0$. Hence, if $Z^{0}\leq 0$, then $Z^{n}\leq 0$ for all $n>0$. If $Z^{n}>0$, by [38, page 271], $Z^{n}\leq\frac{2}{t^{n}+1/Z^{0}}\leq\frac{2}{t^{n}}.$ This finishes the proof. ∎ ## 6\. Higher integrability estimate We begin this section by deriving a “renormalized form” of the finite difference scheme for $q_{j}$, so let $f$ be a nonlinear function (renormalization) of appropriate regularity and growth. Multiplying (3.7) by $f^{\prime}(q_{j})$ and using the discrete chain rule, which in the present context reads $\begin{split}f^{\prime}(q_{j}^{n})D_{\pm}q_{j}^{n}&=D_{\pm}f(q_{j}^{n})\mp\frac{{\Delta x}}{2}f^{\prime\prime}\left(q^{n}_{j\pm 1/2}\right)\left(D_{\pm}q_{j}^{n}\right)^{2},\\\ f^{\prime}(q_{j}^{n})D_{+}^{t}q_{j}^{n}&=D_{+}^{t}f\left(q_{j}^{n}\right)-\frac{{\Delta t}}{2}f^{\prime\prime}\left(q^{n+1/2}_{j}\right)\left(D_{+}^{t}q_{j}^{n}\right)^{2},\end{split}$ where $q^{n}_{j\pm 1/2}$ is a number between $q_{j}$ and $q_{j\pm 1}$, and $q^{n+1/2}_{j}$ is a number between $q_{j}^{n}$ and $q^{n+1}_{j}$. Multiplying the scheme (3.7) with $f^{\prime}(q_{j}^{n})$ we obtain (6.1) $\begin{split}D_{+}^{t}f\left(q_{j}^{n}\right)&+\left(u_{j-1/2}^{n}\vee 0\right)D_{-}f(q_{j}^{n})+\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}f(q_{j}^{n})\\\ &+\frac{\left(q_{j}^{n}\right)^{2}}{2}f^{\prime}(q_{j}^{n})+\left[P_{j}^{n}-\left(u_{j+1/2}^{n}\vee 0\right)^{2}-\left(u_{j-1/2}^{n}\wedge 0\right)^{2}\right]f^{\prime}(q_{j}^{n})\\\ &\qquad+I_{{\Delta x},f^{\prime\prime},j}=\frac{{\Delta t}}{2}f^{\prime\prime}\left(q^{n+1/2}_{j}\right)\left(D_{+}^{t}q_{j}^{n}\right)^{2},\end{split}$ where $\displaystyle I_{{\Delta x},f^{\prime\prime},j}$ $\displaystyle:=\frac{{\Delta x}}{2}\Biggl{\\{}\left(u_{j-1/2}^{n}\vee 0\right)f^{\prime\prime}(q^{n}_{j-1/2})(D_{-}q_{j}^{n})^{2}$ $\displaystyle\qquad\qquad\quad-\left(u_{j+1/2}^{n}\wedge 0\right)f^{\prime\prime}(q^{n}_{j+1/2})(D_{+}q_{j}^{n})^{2}\Biggr{\\}}.$ Let us now write (6.1) in divergence-form. To this end, observe that the discrete product rule (2.2) implies the following relations: $\displaystyle D_{-}\left[\left(u_{j+1/2}^{n}\vee 0\right)f(q_{j}^{n})\right]$ $\displaystyle=\left(u_{j-1/2}^{n}\vee 0\right)D_{-}f(q_{j}^{n})+D_{-}\left(u_{j+1/2}^{n}\vee 0\right)f(q_{j}^{n}),$ $\displaystyle D_{+}\left[\left(u_{j-1/2}^{n}\wedge 0\right)f(q_{j}^{n})\right]$ $\displaystyle=\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}f(q_{j}^{n})+D_{+}\left(u_{j-1/2}^{n}\wedge 0\right)f(q_{j}^{n})$ $\displaystyle=\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}f(q_{j}^{n})+D_{-}\left(u_{j+1/2}^{n}\wedge 0\right)f(q_{j}^{n}),$ and therefore, using that $q_{j}^{n}=D_{-}u_{j+1/2}^{n}$, $\displaystyle\left(u_{j-1/2}^{n}\vee 0\right)D_{-}f(q_{j}^{n})+\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}f(q_{j}^{n})$ $\displaystyle\qquad=D_{-}\left[\left(u_{j+1/2}^{n}\vee 0\right)f(q_{j}^{n})\right]+D_{+}\left[\left(u_{j-1/2}^{n}\wedge 0\right)f(q_{j}^{n})\right]-q_{j}^{n}f(q_{j}^{n}).$ Hence, we end up with the following divergence-form variant of the renormalized difference scheme (6.1): (6.2) $\begin{split}D_{+}^{t}f(q_{j}^{n})&+D_{-}\left[\left(u_{j+1/2}^{n}\vee 0\right)f(q_{j}^{n})\right]+D_{+}\left[\left(u_{j-1/2}^{n}\wedge 0\right)f(q_{j}^{n})\right]\\\ &+\frac{\left(q_{j}^{n}\right)^{2}}{2}f^{\prime}(q_{j}^{n})-q_{j}^{n}f(q_{j}^{n})\\\ &\qquad+\left[P_{j}^{n}-\left(u_{j+1/2}^{n}\vee 0\right)^{2}-\left(u_{j-1/2}^{n}\wedge 0\right)^{2}\right]f^{\prime}(q_{j}^{n})\\\ &\qquad\qquad+I_{{\Delta x},f^{\prime\prime},j}=\frac{{\Delta t}}{2}f^{\prime\prime}\left(q^{n+1/2}_{j}\right)\left(D_{+}^{t}q_{j}^{n}\right)^{2}.\end{split}$ To ensure that the limit of $(\partial_{x}u_{{\Delta x}})^{2}$, cf. (4.2), is nonsingular (i.e., not a measure), we shall need the following higher integrability estimate: ###### Lemma 6.1. Let $q_{{\Delta x}}$ be defined by (3.9), and assume that the CFL type condition (4.1) holds. Then, for all finite numbers $a,b,\alpha$ with $a<b$ and $\alpha\in(0,\theta)$, (6.3) $\int_{0}^{T}\int_{a}^{b}\left|q_{{\Delta x}}\right|^{2+\alpha}\,dx\,dt\leq C,$ for some constant $C=C(a,b,T,\alpha)$ that is independent of ${\Delta x}$. ###### Proof. Define $\eta_{\varepsilon}(q)=\sqrt{\varepsilon^{2}+q^{2}}-\varepsilon$. Note that $\eta_{\varepsilon}(q)\approx\left|q\right|$ for small $\varepsilon$ and $-1<\eta_{\varepsilon}^{\prime}(q)=\frac{q}{\sqrt{\varepsilon^{2}+q^{2}}}<1\quad\text{and}\quad 0<\eta_{\varepsilon}^{\prime\prime}(q)=\frac{\varepsilon^{2}}{(\varepsilon^{2}+q^{2})^{3/2}}\leq\frac{1}{\varepsilon}.$ In (6.2), we then specify $f(q)=\eta_{\varepsilon}(q)\left(1+q^{2}\right)^{\alpha/2}.$ One can easily check that $\displaystyle f^{\prime}(q)$ $\displaystyle=\eta_{\varepsilon}^{\prime}(q)\left(1+q^{2}\right)^{\alpha/2}+\alpha\eta_{\varepsilon}(q)q\left(1+q^{2}\right)^{\alpha/2-1}$ $\displaystyle f^{\prime\prime}(q)$ $\displaystyle=\eta_{\varepsilon}^{\prime\prime}(q)\left(1+q^{2}\right)^{\alpha/2}+2\alpha\eta_{\varepsilon}^{\prime}(q)q\left(1+q^{2}\right)^{\alpha/2-1}$ $\displaystyle\qquad\quad+\alpha\eta_{\varepsilon}(q)\left(\left(1+q^{2}\right)^{\alpha/2-1}+2\left(\frac{\alpha}{2}-1\right)\left(1+q^{2}\right)^{\alpha/2-2}\right),$ so that in particular $f^{\prime\prime}(q)\geq 0$ and (6.4) $f^{\prime\prime}(q)=\eta_{\varepsilon}^{\prime\prime}(q)\left(1+q^{2}\right)^{\alpha/2}+\text{bounded terms.}$ Next set $\displaystyle H(q)$ $\displaystyle:=\frac{q^{2}}{2}f^{\prime}(q)-qf(q)$ $\displaystyle=\frac{q^{2}}{2}\left(\eta_{\varepsilon}^{\prime}(q)\left(1+q^{2}\right)^{\alpha/2}+\alpha\eta_{\varepsilon}(q)\left(1+q^{2}\right)^{\alpha/2-1}q\right)-q\eta_{\varepsilon}(q)\left(1+q^{2}\right)^{\alpha/2}$ $\displaystyle=q\eta_{\varepsilon}(q)\left(1+q^{2}\right)^{\alpha/2}\left[\frac{q\eta_{\varepsilon}^{\prime}(q)}{2\eta_{\varepsilon}(q)}+\frac{\alpha q^{2}}{2(1+q^{2})}-1\right]$ $\displaystyle=q\eta_{\varepsilon}(q)\left(1+q^{2}\right)^{\alpha/2}\frac{1}{2}\left[\frac{\alpha q^{2}}{1+q^{2}}+\frac{\varepsilon}{\sqrt{\varepsilon^{2}+q^{2}}}-1\right]=:H_{\varepsilon}(q)h_{\varepsilon}(q),$ with $H_{\varepsilon}(q)=q\eta_{\varepsilon}(q)\left(1+q^{2}\right)^{\alpha/2},\qquad h_{\varepsilon}(q)=\frac{1}{2}\left[\frac{\alpha q^{2}}{1+q^{2}}+\frac{\varepsilon}{\sqrt{\varepsilon^{2}+q^{2}}}-1\right].$ Now note that $\lim_{q\to-\infty}h_{\varepsilon}(q)\leq\lim_{q\to-\infty}h_{1}(q)=\frac{\alpha-1}{2}<0.$ Hence, for $\alpha,\varepsilon<1$, we can find a constant $K>0$ such that (6.5) $h_{\varepsilon}(q)<\frac{\alpha-1}{4}\quad\text{for all $q<-K$}.$ Let us continue by defining the sets $\begin{gathered}\mathcal{N}^{n}=\left\\{j\in\mathbb{Z}\;\bigm{|}\;q_{j}^{n}<-K\right\\},\quad\mathcal{C}^{n}=\left\\{j\in\mathbb{Z}\;\bigm{|}\;-K\leq q_{j}^{n}\leq 0\right\\},\\\ \text{and}\quad\mathcal{P}^{n}=\left\\{j\in\mathbb{Z}\;\bigm{|}\;q_{j}^{n}>0\right\\},\end{gathered}$ where $K$ is defined in (6.5). Moreover, let $0\leq\chi(x)\leq 1$ be a smooth cutoff function satisfying $\chi(x)=\begin{cases}0,&x<a-1,\\\ 1,&x\in[a,b],\\\ 0,&x>b+1.\end{cases}$ We multiply (6.2) by $\chi(x_{j}){\Delta t}{\Delta x}$ and sum over $(n,j)\in\left\\{0,\dots,N-1\right\\}\times\mathbb{Z}$ to get $\displaystyle{\Delta t}{\Delta x}\sum_{n}\sum_{j\in\mathcal{N}^{n}}h_{\varepsilon}$ $\displaystyle\left(q_{j}^{n}\right)H_{\varepsilon}\left(q_{j}^{n}\right)\chi(x_{j})$ (6.6) $\displaystyle\leq-{\Delta t}{\Delta x}\sum_{n}\sum_{j\in\mathcal{C}^{n}}h_{\varepsilon}\left(q_{j}^{n}\right)H_{\varepsilon}\left(q_{j}^{n}\right)\chi\left(x_{j}\right)$ (6.7) $\displaystyle\quad-{\Delta t}{\Delta x}\sum_{n}\sum_{j\in\mathcal{P}^{n}}h_{\varepsilon}\left(q_{j}^{n}\right)H_{\varepsilon}\left(q_{j}^{n}\right)\chi\left(x_{j}\right)$ (6.8) $\displaystyle\quad\begin{aligned} -{\Delta t}{\Delta x}\sum_{j,n}\Bigl{[}&D_{-}\left(\left(u_{j+1/2}^{n}\vee 0\right)f\left(q_{j}^{n}\right)\right)\chi(x_{j})\\\ &\quad D_{+}\left(\left(u_{j-1/2}^{n}\wedge 0\right)f\left(q_{j}^{n}\right)\right)\chi(x_{j})\Bigr{]}\end{aligned}$ (6.9) $\displaystyle\quad-{\Delta t}{\Delta x}\sum_{n,j}A^{n}_{j}\chi\left(x_{j}\right)$ (6.10) $\displaystyle\quad+{\Delta x}\sum_{j}\left(f\left(q^{0}_{j}\right)-f\left(q^{N}_{j}\right)\right)\chi(x_{j})$ (6.11) $\displaystyle\quad+{\Delta x}{\Delta t}\sum_{n,j}\frac{{\Delta t}}{2}f^{\prime\prime}\left(q^{n+1/2}_{j}\right)\chi(x_{j})\left(D_{+}^{t}q_{j}^{n}\right)^{2},$ where (6.12) $A^{n}_{j}=P_{j}^{n}-\left(u_{j+1/2}^{n}\vee 0\right)^{2}-\left(u_{j-1/2}^{n}\wedge 0\right)^{2}.$ Now for $j\in\mathcal{N}^{n}$ we have that $\frac{1-\alpha}{4}\left|q_{j}^{n}\right|\eta_{\varepsilon}\left(q_{j}^{n}\right)\left(1+\left(q_{j}^{n}\right)^{2}\right)^{\alpha/2}\leq h_{\varepsilon}\left(q_{j}^{n}\right)H_{\varepsilon}\left(q_{j}^{n}\right).$ Therefore $\displaystyle\frac{1-\alpha}{4}$ $\displaystyle{\Delta t}{\Delta x}\sum_{n}\sum_{j\in\mathcal{N}^{n}}\left|q_{j}^{n}\right|\eta_{\varepsilon}\left(q_{j}^{n}\right)\left(1+\left(q_{j}^{n}\right)^{2}\right)^{\alpha/2}\chi(x_{j})$ $\displaystyle\leq\left|\text{\eqref{eq:Cterm}}\right|+\left|\text{\eqref{eq:Pterm}}\right|+\left|\text{\eqref{eq:diffterm}}\right|+\left|\text{\eqref{eq:Aterm}}\right|+\left|\text{\eqref{eq:fterm}}\right|+\left|\text{\eqref{eq:fppterm}}\right|.$ We shall now find bounds on all the terms on the right hand side; in what follows, we let $C$ denote a generic constant independent of ${\Delta x}$, $\varepsilon$, and $\alpha$. We start with (6.11). By (3.7) $\displaystyle\left(D_{+}^{t}q_{j}^{n}\right)^{2}$ $\displaystyle\leq C\Biggl{[}\left(\left(u_{j-1/2}^{n}\vee 0\right)D_{-}q_{j}^{n}\right)^{2}+\left(\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}q_{j}^{n}\right)^{2}$ $\displaystyle\qquad\qquad+\left(P_{j}^{n}\right)^{2}+\left(u_{j+1/2}^{n}\vee 0\right)^{4}+\left(u_{j-1/2}^{n}\wedge 0\right)^{4}+\left(q_{j}^{n}\right)^{4}\Biggr{]}.$ To bound the “integrals” of these terms we must use the CFL type condition (4.1), which implies that ${\Delta t}=\mathcal{O}({\Delta x}^{2+\theta})$. First, $\displaystyle{\Delta x}{\Delta t}\sum_{j}\left(\left(u_{j-1/2}^{n}\vee 0\right)D_{-}q_{j}^{n}\right)^{2}$ $\displaystyle\quad\leq C{\Delta x}^{1+\theta}\sum_{j}\left({\Delta x}D_{-}q_{j}^{n}\right)^{2}\leq C{\Delta x}^{1+\theta}\sum_{j}\left(q_{j}^{n}\right)^{2}\leq C{\Delta x}^{\theta}.$ Therefore $\displaystyle{\Delta t}{\Delta x}\sum_{n,j}{\Delta t}\Biggl{[}\left(\left(u_{j-1/2}^{n}\vee 0\right)D_{-}q_{j}^{n}\right)^{2}+\left(\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}q_{j}^{n}\right)^{2}\Biggr{]}\chi(x_{j})$ $\displaystyle\qquad\leq C{\Delta x}^{\theta}T\to 0\quad\text{as ${\Delta x}\to 0$.}$ We also find that $\displaystyle{\Delta x}\sum_{j}\Biggl{[}\left(P_{j}^{n}\right)^{2}+\left(u_{j+1/2}^{n}\vee 0\right)^{4}+\left(u_{j-1/2}^{n}\wedge 0\right)^{4}\Biggr{]}$ $\displaystyle\qquad\leq C{\Delta x}\sum_{j}\Biggl{[}\left|P_{j}^{n}\right|+\left(u_{j+1/2}^{n}\vee 0\right)^{2}+\left(u_{j-1/2}^{n}\wedge 0\right)^{2}\Biggr{]}$ $\displaystyle\qquad\leq C{\Delta x}\sum_{j}\Biggl{[}\left|P_{j}^{n}\right|+\left(u_{j+1/2}^{n}\right)^{2}\Biggr{]}\leq C,$ since $u_{{\Delta x}}$ and $P^{n}_{j}$ are uniformly bounded. Thus ${\Delta t}{\Delta x}\sum_{n,j}{\Delta t}\left[\left(P_{j}^{n}\right)^{2}+\left(u_{j+1/2}^{n}\vee 0\right)^{4}+\left(u_{j-1/2}^{n}\wedge 0\right)^{4}\right]\chi(x_{j})\to 0,$ as ${\Delta x}\to 0$. Additionally, $\displaystyle{\Delta x}\sum_{j}{\Delta t}\left(q_{j}^{n}\right)^{4}$ $\displaystyle\leq C{\Delta x}\sum_{j}{\Delta x}^{\theta}\left({\Delta x}q_{j}^{n}\right)^{2}\left(q_{j}^{n}\right)^{2}$ $\displaystyle\leq C{\Delta x}^{1+\theta}\sum_{j}\left[\left(u_{j+1/2}^{n}\right)^{2}+\left(u_{j-1/2}^{n}\right)^{2}\right]\left(q_{j}^{n}\right)^{2}$ $\displaystyle\leq C{\Delta x}^{1+\theta}\sum_{j}\left(q_{j}^{n}\right)^{2}\leq C{\Delta x}^{\theta}.$ Therefore ${\Delta x}{\Delta t}\sum_{j}{\Delta t}\left(q_{j}^{n}\right)^{4}\chi(x_{j})\leq CT{\Delta x}^{\theta}\to 0\quad\text{as ${\Delta x}\to 0$.}$ Now we have established that (6.13) ${\Delta x}{\Delta t}\sum_{n,j}{\Delta t}\left(D_{+}^{t}q_{j}^{n}\right)^{2}\chi(x_{j})=\mathcal{O}\left({\Delta x}^{\theta}\right)\quad\text{as ${\Delta x}\to 0$.}$ Recalling (6.4) this implies that $\left|\text{\eqref{eq:fppterm}}\right|\leq C{\Delta x}{\Delta t}\sum_{n,j}\frac{{\Delta t}}{\varepsilon}\left(1+\left(q^{n+1/2}_{j}\right)^{2}\right)^{\alpha/2}\left(D_{+}^{t}q_{j}^{n}\right)^{2}\chi(x_{j})+\mathcal{O}\left({\Delta x}^{\theta}\right).$ When we established (6.13) we always had a “${\Delta x}^{\theta}$ to spare”, which we can use now. With $\beta=\theta-\alpha>0$, we get $\displaystyle{\Delta t}{\Delta x}\sum_{n,j}$ $\displaystyle\frac{{\Delta t}}{\varepsilon}\left(1+\left(q^{n+1/2}_{j}\right)^{2}\right)^{\alpha/2}\left(D_{+}^{t}q_{j}^{n}\right)^{2}$ $\displaystyle\leq C{\Delta x}{\Delta t}\sum_{n,j}\frac{{\Delta x}^{\beta}}{\varepsilon}{\Delta x}^{\alpha}\left(1+\left(q^{n+1/2}_{j}\right)^{2}\right)^{\alpha/2}{\Delta x}^{2}\left(D_{+}^{t}q_{j}^{n}\right)^{2}$ $\displaystyle\leq C{\Delta x}{\Delta t}\sum_{n,j}\frac{{\Delta x}^{\beta}}{\varepsilon}\left({\Delta x}^{2}+\left({\Delta x}q^{n+1/2}_{j}\right)^{2}\right)^{\alpha/2}{\Delta x}^{2}\left(D_{+}^{t}q_{j}^{n}\right)^{2}$ $\displaystyle\leq C\frac{{\Delta x}^{\beta}}{\varepsilon}{\Delta x}{\Delta t}\sum_{n,j}{\Delta x}^{2}\left(D_{+}^{t}q_{j}^{n}\right)^{2}\leq C\frac{{\Delta x}^{\beta}}{\varepsilon},$ since $\left({\Delta x}^{2}+\left({\Delta x}q^{n+1/2}_{j}\right)^{2}\right)^{\alpha/2}\leq\left({\Delta x}^{2}+4\left(\left\|u^{n}\right\|_{\ell^{\infty}}\right)^{2}\right)^{\alpha/2}\leq C.$ Now choosing $\varepsilon={\Delta x}^{\beta}$, we finally conclude that $\left|\eqref{eq:fppterm}\right|$ is bounded. Next we turn to (6.6). For $-K\leq q\leq 0$, we have that $\left|h_{\varepsilon}(q)H_{\varepsilon}(q)\right|\leq C$, where $C$ is independent of $\varepsilon$. Therefore $\left|\text{\eqref{eq:Cterm}}\right|\leq CT(b-a+2).$ To estimate $\left|\text{\eqref{eq:Pterm}}\right|$, observe that $\left|\text{\eqref{eq:Pterm}}\right|\leq C{\Delta t}{\Delta x}\sum_{n}\sum_{j\in\mathcal{P}^{n}}\left(1+\left|q_{j}^{n}\right|^{2+\alpha}\right)\chi\left(x_{j}\right)$ Now, equipped with (4.2) and (5.1), it is possible to bound the right-hand side by a “${\Delta x}$ independent” constant exactly as was done in [12]. Regarding (6.8), observe that $\displaystyle\biggl{|}{\Delta t}$ $\displaystyle{\Delta x}\sum\limits_{j,n}D_{-}\left[\left(u_{j+1/2}^{n}\vee 0\right)f\left(q_{j}^{n}\right)\right]\chi(j{\Delta x})\biggr{|}$ $\displaystyle=\bigg{|}{\Delta t}{\Delta x}\sum\limits_{j,n}\left(u_{j+1/2}^{n}\vee 0\right)D_{+}\chi(j{\Delta x})f\left(q_{j}^{n}\right)\biggr{|}$ $\displaystyle\leq C{\Delta t}{\Delta x}\sum\limits_{j,n}\left|u_{j+1/2}^{n}\right|\left|D_{+}\chi(j{\Delta x})\right|\left(1+\left|q_{j}^{n}\right|^{1+\alpha}\right)$ $\displaystyle\leq CT\Biggl{(}\sup\limits_{n}\left\|u^{n}\right\|_{\ell^{\infty}}\left\|\left\\{D_{+}\chi(x_{j})\right\\}_{j}\right\|_{\ell^{1}}$ $\displaystyle\qquad\qquad\qquad\qquad+\sup\limits_{n}\left\|\left\\{\left|q_{j}^{n}\right|^{1+\alpha}\right\\}_{j}\right\|_{\ell^{\frac{2}{1+\alpha}}}\left\|\left\\{D_{+}\chi(j{\Delta x})\right\\}_{j}\right\|_{\ell^{\frac{2}{1-\alpha}}}\Biggr{)}$ $\displaystyle=cT\Biggl{(}\sup\limits_{n}\left\|u^{n}\right\|_{\ell^{\infty}}\left\|\left\\{D_{+}\chi(x_{j})\right\\}_{j}\right\|_{\ell^{1}}+\sup\limits_{n}\left\|q^{n}\right\|^{1+\alpha}_{\ell^{2}}\left\|\left\\{D_{+}\chi(j{\Delta x})\right\\}_{j}\right\|_{\ell^{\frac{2}{1-\alpha}}}\Biggr{)}.$ Therefore, also $\left|\eqref{eq:diffterm}\right|$ is bounded independently of ${\Delta x}$. Next we focus on (6.9). Remembering that $\left|f^{\prime}(q)\right|\leq C(1+\left|q\right|^{\alpha})$, we find $\displaystyle\left|\text{\eqref{eq:Aterm}}\right|$ $\displaystyle\leq{\Delta t}{\Delta x}\sum\limits_{n,j}\left(\left\|P^{n}\right\|_{\ell^{\infty}}+2\left\|u^{n}\right\|_{\ell^{\infty}}^{2}\right)\left|f^{\prime}(q_{j}^{n})\right|\chi(x_{j})$ $\displaystyle\leq C{\Delta t}{\Delta x}\sum\limits_{n,j}\left(\left\|P^{n}\right\|_{\ell^{\infty}}+2\left\|u^{n}\right\|_{\ell^{\infty}}^{2}\right)\left(1+\left|q_{j}^{n}\right|^{\alpha}\right)\chi(x_{j})$ $\displaystyle\leq CT\left(\sup\limits_{n}\left\|P^{n}\right\|_{\ell^{\infty}}+2\sup\limits_{n}\left\|u^{n}\right\|_{\ell^{\infty}}^{2}\right)$ $\displaystyle\qquad\quad\times\left(\left\|\left\\{\chi(x_{j})\right\\}_{j}\right\|_{\ell^{1}}+\sup\limits_{n}\left\|\left\\{\left|q_{j}^{n}\right|^{\alpha}\right\\}_{j}\right\|_{\ell^{\frac{2}{\alpha}}}\left\|\left\\{\chi(j{\Delta x})\right\\}_{j}\right\|_{\ell^{\frac{2}{2-\alpha}}}\right)$ $\displaystyle\leq CT\left(\sup\limits_{n}\left\|P^{n}\right\|_{\ell^{\infty}}+2\sup\limits_{n}\left\|u^{n}\right\|_{\ell^{\infty}}^{2}\right)$ $\displaystyle\qquad\quad\times\left(\left\|\left\\{\chi(j{\Delta x})\right\\}_{j}\right\|_{\ell^{1}}+\sup\limits_{n}\left\|q^{n}\right\|^{\alpha}_{\ell^{2}}\left\|\left\\{\chi(j{\Delta x})\right\\}_{j}\right\|_{\ell^{\frac{2}{2-\alpha}}}\right).$ Finally, keeping in mind that $f\geq 0$, we treat (6.10) as follows: $\displaystyle\left|\text{\eqref{eq:fterm}}\right|$ $\displaystyle\leq{\Delta x}\sum_{j}\left(f\left(q^{N}_{j}\right)+f\left(q^{0}_{j}\right)\right)\chi(x_{j})$ $\displaystyle\leq C{\Delta x}\sum_{j}\left(2+\left|q^{N}_{j}\right|^{\alpha}+\left|q^{0}_{j}\right|^{\alpha}\right)\chi(x_{j})$ $\displaystyle\leq\left\|\left\\{\chi(x_{j})\right\\}_{j}\right\|_{\ell^{1}}$ $\displaystyle\quad+\left(\left\|\left\\{\left|q^{N}_{j}\right|^{1+\alpha}\right\\}_{j}\right\|_{\ell^{\frac{2}{1+\alpha}}}+\left\|\left\\{\left|q^{0}_{j}\right|^{1+\alpha}\right\\}_{j}\right\|_{\ell^{\frac{2}{1+\alpha}}}\right)\left\|\left\\{\chi(x_{j})\right\\}_{j}\right\|_{\ell^{\frac{2}{1-\alpha}}}$ $\displaystyle=\left\|\left\\{\chi(x_{j})\right\\}_{j}\right\|_{\ell^{1}}+\left(\left\|q^{N}\right\|_{\ell^{2}}^{1+\alpha}+\left\|q^{0}\right\|_{\ell^{2}}^{1+\alpha}\right)\left\|\left\\{\chi(x_{j})\right\\}_{j}\right\|_{\ell^{\frac{2}{1-\alpha}}}.$ Summarizing, we have established ${\Delta t}{\Delta x}\sum_{n}\sum_{j}\left|q_{j}^{n}\right|\eta_{\varepsilon}\left(q_{j}^{n}\right)\left(1+\left(q_{j}^{n}\right)^{2}\right)^{\alpha/2}\leq C.$ The statement of the lemma follows by noting that $\left|q\right|^{2+\alpha}\leq\left|q\right|\eta_{\varepsilon}(q)\left(1+q^{2}\right)^{\alpha/2}+\left|q\right|^{1+\alpha},$ and using, in combination with (4.2), the bound ${\Delta t}{\Delta x}\sum_{j,n}\left|q_{j}^{n}\right|^{1+\alpha}\chi(x_{j})\leq CT\sup_{n}\left\|q^{n}\right\|_{\ell^{2}}^{1+\alpha}\left\|\left\\{\chi(x_{j})\right\\}_{j}\right\|_{\ell^{\frac{2}{1-\alpha}}}.$ ∎ ## 7\. Basic convergence results The purpose of this section is to present some straightforward consequences of the a priori estimates established in the foregoing sections. More precisely, we prove that the two sequences $\left\\{u_{{\Delta x}}\right\\}_{{\Delta x}>0}$, cf. (3.10), and $\left\\{P_{{\Delta x}}\right\\}_{{\Delta x}>0}$, cf. (3.11), have strongly converging subsequences, starting with the former. ###### Lemma 7.1. There exists a limit function $u\in L^{\infty}\left(0,T;H^{1}(\mathbb{R})\right)\cap C([0,T]\times\mathbb{R}),$ such that along a subsequence as ${\Delta x}\to 0$ (7.1) $\displaystyle u_{{\Delta x}}$ $\displaystyle\overset{\star}{\rightharpoonup}u\quad\text{in $L^{\infty}\left(0,T;H^{1}(\mathbb{R})\right)$,}$ (7.2) $\displaystyle u_{{\Delta x}}$ $\displaystyle\to u\quad\text{uniformly in $[a,b]\times[0,T]$, for any $a<b$.}$ Additionally, (7.3) $t\mapsto\left\|u(t,\cdot)\right\|_{H^{1}(\mathbb{R})}$ is non- increasing, and (7.4) $\displaystyle\lim_{t\to 0}u(t,x)=u_{0}(x),\qquad x\in\mathbb{R}.$ ###### Proof. First we note that for $t\in I^{n}$, $\displaystyle\int_{\mathbb{R}}(u_{{\Delta x}}(t,x))^{2}\,dx$ $\displaystyle=\sum_{j}\\!\int_{x_{j-1/2}}^{x_{j+1/2}}\\!\\!\\!\left(\frac{1}{{\Delta x}}(x_{j+1/2}-x)u_{j-1/2}(t)\\!+\\!(x-x_{j-1/2})u_{j+1/2}(t)\right)^{2}\,dx$ $\displaystyle\leq\frac{{\Delta x}}{2}\sum_{j}\left(\left(u_{j-1/2}(t)\right)^{2}+\left(u_{j+1/2}(t)\right)^{2}\right)$ $\displaystyle\leq\frac{1}{{\Delta t}}\left(\left(t-t^{n}\right){\Delta x}\sum_{j}\left(u^{n+1}_{j+1/2}\right)^{2}+\left(t^{n+1}-t\right){\Delta x}\sum_{j}\left(u_{j+1/2}^{n}\right)^{2}\right)$ and $\displaystyle\int_{\mathbb{R}}\left(\partial_{x}u_{{\Delta x}}(t,x)\right)^{2}\,dx$ $\displaystyle={\Delta x}\sum_{j}\left(q_{j}(t)\right)^{2}$ $\displaystyle\leq\frac{1}{{\Delta t}}\left(\left(t-t^{n}\right){\Delta x}\sum_{j}\left(q^{n+1}_{j}\right)^{2}+\left(t^{n+1}-t\right){\Delta x}\sum_{j}\left(q_{j}^{n}\right)^{2}\right).$ Hence, $\left\|u_{{\Delta x}}(t,\cdot)\right\|_{H^{1}(\mathbb{R})}^{2}\leq\frac{1}{{\Delta t}}\left(\left(t-t^{n}\right)\left\|u^{n+1}\right\|_{H^{1}(\mathbb{R})}^{2}+\left(t^{n+1}-t\right)\left\|u^{n}\right\|_{H^{1}(\mathbb{R})}^{2}\right).$ Let $s\in I^{m}$ with $m\leq n$. Then, using (4.2), $\displaystyle\left\|u_{{\Delta x}}(t,\cdot)\right\|_{H^{1}(\mathbb{R})}^{2}$ $\displaystyle\leq\frac{1}{{\Delta t}}\left(\left(t-t^{n}\right)\left\|u^{n+1}\right\|_{H^{1}(\mathbb{R})}^{2}+\left(t^{n+1}-t\right)\left\|u^{n}\right\|_{H^{1}(\mathbb{R})}^{2}\right)$ $\displaystyle\leq e^{{\Delta t}(n-m){\Delta x}^{\theta}}\left\|u_{{\Delta x}}(s,\cdot)\right\|_{H^{1}(\mathbb{R})}^{2}$ $\displaystyle\quad+\frac{e^{{\Delta t}(n-m){\Delta x}^{\theta}}}{{\Delta t}}\biggl{[}\left(t-t^{n}\right)\left(\left\|u^{m+1}\right\|_{H^{1}(\mathbb{R})}^{2}-\left\|u_{{\Delta x}}(s,\cdot)\right\|_{H^{1}(\mathbb{R})}^{2}\right)$ $\displaystyle\hphantom{\quad+\frac{1}{{\Delta t}}\biggl{[}}\quad+\left(t^{n+1}-t\right)\left(\left\|u^{m}\right\|_{H^{1}(\mathbb{R})}^{2}-\left\|u_{{\Delta x}}(s,\cdot)\right\|_{H^{1}(\mathbb{R})}^{2}\right)\biggr{]}$ $\displaystyle\leq e^{{\Delta t}(n-m){\Delta x}^{\theta}}\left\|u_{{\Delta x}}(s,\cdot)\right\|_{H^{1}(\mathbb{R})}^{2}$ $\displaystyle\qquad+e^{{\Delta t}(n-m){\Delta x}^{\theta}}\left|\left\|u^{m+1}\right\|_{H^{1}(\mathbb{R})}^{2}-\left\|u^{m}\right\|_{H^{1}(\mathbb{R})}^{2}\right|$ $\displaystyle\leq e^{{\Delta t}(n-m){\Delta x}^{\theta}}\left\|u_{{\Delta x}}(s,\cdot)\right\|_{H^{1}(\mathbb{R})}^{2}$ $\displaystyle\qquad+e^{{\Delta t}(n-m){\Delta x}^{\theta}}\left(e^{{\Delta t}{\Delta x}^{\theta}}-1\right)\left\|u^{m}\right\|_{H^{1}(\mathbb{R})}^{2}.$ This implies (7.1) and (7.3). Next we prove that $\left\\{u_{{\Delta x}}\right\\}_{{\Delta x}>0}$ is uniformly bounded in $W^{1,2+\alpha}((0,T)\times(a,b))$. We can assume that $a-1=x_{j_{a}}$ and $b+1=x_{j_{b}}$ for some integers $j_{a}$ and $j_{b}$. Since $q\mapsto\left|q\right|^{2+\alpha}$ is convex, (7.5) $\displaystyle\int_{0}^{T}\int_{a}^{b}\left|\partial_{x}u_{{\Delta x}}\right|^{2+\alpha}\,dx$ $\displaystyle\quad\leq\sum_{n}{\Delta x}\sum_{j}\left(\left(t^{n+1}-t\right)\left|q_{j}^{n}\right|^{2+\alpha}+\left(t-t^{n}\right)\left|q_{j}^{n}\right|^{2+\alpha}\right)\leq C,$ for some constant $C=C(\alpha,a,b,u_{0})$, where we have also used (6.3). Now set $\sigma=(x-x_{j-1/2})/{\Delta x}$. Then, for $x\in I_{j}$, we have $\displaystyle\left|\partial_{t}u_{{\Delta x}}(t,x)\right|$ $\displaystyle=\left|(1-\sigma)D_{+}^{t}u_{j-1/2}^{n}+\sigma D_{+}^{t}u_{j+1/2}^{n}\right|$ $\displaystyle\leq(1-\sigma)\left|D_{+}^{t}u_{j-1/2}^{n}\right|+\sigma\left|D_{+}^{t}u_{j+1/2}^{n}\right|.$ Furthermore, by the uniform bounds on $D_{+}P_{j}^{n}$ and $u_{{\Delta x}}$, $\left|D_{+}^{t}u_{j+1/2}^{n}\right|\leq C(1+\left|q_{j}^{n}\right|+\left|q^{n}_{j-1}\right|).$ Using this, $\int_{0}^{T}\int_{a}^{b}\left|\partial_{t}u_{{\Delta x}}\right|^{2+\alpha}\,dx\leq C\left(1+{\Delta t}\sum_{n=0}^{N}{\Delta x}\sum_{{j_{a}}}^{{j_{b}}}\left|q_{j}^{n}\right|^{2+\alpha}\right)\leq C.$ Now $\left\\{u_{{\Delta x}}\right\\}_{{\Delta x}>0}\subset W^{1,2+\alpha}\subset\subset C^{0,\ell}$ on $(0,T)\times(a,b)$ with $\ell=1-2/(2+\alpha)$. Therefore, along a subsequence, $u_{{\Delta x}}\to u$ uniformly in $(0,T)\times(a,b)$ as ${\Delta x}\to 0$. Let us show that the limit satisfies the initial condition (7.4). Fix $\bar{x}\in\mathbb{R}$ and let $t\in(0,1)$. We have $\bar{x}\in I_{j}$ for some $j$ and $u_{{\Delta x}}(x_{j-1/2},0)=u_{0}(x_{j-1/2})$, so that $\displaystyle\left|u_{{\Delta x}}(0,\bar{x})-u_{0}(\bar{x})\right|$ $\displaystyle\leq\left|u_{{\Delta x}}(0,\bar{x})-u_{{\Delta x}}(0,x_{j-1/2})\right|+\left|u_{0}(x_{j-1/2})-u_{0}(\bar{x})\right|$ $\displaystyle\leq C\left(\bar{x}-x_{j-1/2}\right)^{\ell}.$ Consequently, $\displaystyle\left|u(t,\bar{x})-u_{0}(\bar{x})\right|$ $\displaystyle\leq\left|u(t,\bar{x})-u_{{\Delta x}}(t,\bar{x})\right|$ $\displaystyle\qquad+\left|u_{{\Delta x}}(t,\bar{x})-u_{{\Delta x}}(0,\bar{x})\right|+\left|u_{{\Delta x}}(0,\bar{x})-u_{0}(\bar{x})\right|$ $\displaystyle\leq\left|u(t,\bar{x})-u_{{\Delta x}}(t,\bar{x})\right|+Ct^{\ell}+C{\Delta x}^{\ell}.$ Now we can let ${\Delta x}\to 0$ and then $t\to 0$ to conclude that $u(\bar{x},0)=u_{0}(\bar{x})$. This draws to a close the proof of the lemma. ∎ ###### Lemma 7.2. There exists a limit function $P\in L^{\infty}(0,T;W^{1,\infty}(\mathbb{R}))\cap L^{\infty}(0,T;W^{1,1}(\mathbb{R})),$ such that along a subsequence as ${\Delta x}\to 0$ (7.6) $\text{$P_{{\Delta x}}\to P$ in $L^{p}_{\mathrm{loc}}((0,T)\times\mathbb{R})$},\quad 1\leq p<\infty.$ ###### Proof. By the bounds on $P_{j}^{n}$ in Lemma 4.2, we see that $P_{{\Delta x}}$ is bounded in $L^{\infty}$ uniformly in ${\Delta x}$. Next we show that $\left\\{\partial_{t}P_{{\Delta x}}\right\\}_{{\Delta x}>0}$ is bounded in $L^{1}((0,T)\times\mathbb{R})$. For $t\in[t^{n},t^{n+1})$ and $x\in I_{j+1/2}$, $\partial_{t}P_{{\Delta x}}(t,x)=D_{+}^{t}P_{j}^{n}+(x-x_{j})D_{+}^{t}D_{+}P_{j}^{n}=\left(1-\sigma\right)D_{+}^{t}P_{j}^{n}+\sigma D_{+}^{t}P^{n}_{j+1},$ where $\sigma=(x-x_{j})/{\Delta x}$. Write $D_{+}^{t}P_{j}^{n}=X^{n}_{j}+Y^{n}_{j}$, where $X^{n}_{j}$ and $Y^{n}_{j}$ solve $\displaystyle X^{n}_{j}-D_{-}D_{+}X^{n}_{j}$ $\displaystyle=D_{+}^{t}\left(\left(u_{j-1/2}^{n}\wedge 0\right)^{2}+\left(u_{j+1/2}^{n}\vee 0\right)^{2}\right),$ $\displaystyle Y^{n}_{j}-D_{-}D_{+}Y^{n}_{j}$ $\displaystyle=\frac{1}{2}D_{+}^{t}\left(\left(q_{j}^{n}\right)^{2}\right).$ Then $\left\|X^{n}\right\|_{\ell^{1}}$ is bounded by the $\ell^{1}$ norm of the corresponding right hand side avbove. By the discrete chain rule $D_{+}^{t}\left(u_{j-1/2}^{n}\wedge 0\right)^{2}\leq 2\left(u_{j-1/2}^{n}\wedge 0\right)D_{+}^{t}u_{j-1/2}^{n}+{\Delta t}\left(D_{+}^{t}u_{j-1/2}^{n}\right)^{2}.$ Estimating the first term here $\displaystyle\left(u_{j-1/2}^{n}\wedge 0\right)D_{+}^{t}u_{j-1/2}^{n}$ $\displaystyle\qquad=-\left(u_{j-1/2}^{n}\wedge 0\right)\left[\left(u_{j-1/2}^{n}\vee 0\right)q^{n}_{j-1}+\left(u_{j-1/2}^{n}\wedge 0\right)q_{j}^{n}+D_{+}P_{j}^{n}\right]$ $\displaystyle\qquad=-\left(u_{j-1/2}^{n}\wedge 0\right)\left[\left(u_{j-1/2}^{n}\wedge 0\right)q_{j}^{n}+D_{+}P_{j}^{n}\right].$ This means that $\left\|\left(u^{n}\wedge 0\right)D_{+}^{t}u^{n}\right\|_{\ell^{1}}\leq C\Biggl{[}\left\|u^{n}\right\|_{\ell^{2}}\left\|q^{n}\right\|_{\ell^{2}}+\left\|D_{+}P^{n}\right\|_{\ell^{1}}\Biggr{]}\leq C.$ Similarly, we have $\displaystyle\left(D_{+}^{t}u_{j+1/2}^{n}\right)^{2}$ $\displaystyle\\!\leq\\!C\left[\left(u_{j+1/2}^{n}\vee 0\right)^{2}+\left(q_{j}^{n}\right)^{2}+\left(u_{j+1/2}^{n}\wedge 0\right)^{2}+\left(q^{n}_{j+1}\right)^{2}+\left(D_{+}P_{j}^{n}\right)^{2}\right]$ $\displaystyle\leq C\left[\left(u_{j+1/2}^{n}\right)^{2}+\left(q_{j}^{n}\right)^{2}+\left(q^{n}_{j+1}\right)^{2}+\left|D_{+}P_{j}^{n}\right|\right],$ and thus $\left\|\left(D_{+}^{t}u^{n}\right)^{2}\right\|_{\ell^{1}}\leq C$. We have shown that (7.7) $\left\|X^{n}\right\|_{\ell^{1}}\leq\left\|\left\\{D_{+}^{t}\left[(u^{n}_{j-1/2}\wedge 0)^{2}+(u^{n}_{j+1/2}\vee 0)^{2}\right]\right\\}_{j}\right\|_{\ell^{1}}\leq C.$ Next, using (6.2) with $f(q)=q^{2}/2$ we have that (7.8) $\displaystyle\frac{1}{2}D_{+}^{t}\left(q_{j}^{n}\right)^{2}$ $\displaystyle=\underbrace{-D_{-}\left[\left(u_{j+1/2}^{n}\vee 0\right)\left(q_{j}^{n}\right)^{2}\right]+D_{+}\left[\left(u_{j-1/2}^{n}\wedge 0\right)\left(q_{j}^{n}\right)^{2}\right]}_{a^{n}_{j}}$ $\displaystyle\qquad\qquad\quad+\underbrace{A^{n}_{j}q_{j}^{n}-I_{{\Delta x},f^{\prime\prime},j}+{\Delta t}\left(D_{+}^{t}q_{j}^{n}\right)^{2}}_{b^{n}_{j}},$ where $A^{n}_{j}$ is defined in (6.12) and $I_{{\Delta x},f^{\prime\prime},j}={\Delta x}\left\\{\left(u_{j-1/2}^{n}\vee 0\right)\left(D_{-}q_{j}^{n}\right)^{2}-\left(u_{j+1/2}^{n}\wedge 0\right)\left(D_{+}q_{j}^{n}\right)^{2}\right\\}\geq 0.$ We write $Y^{n}_{j}=Y^{a,n}_{j}+Y^{b,n}_{j}$ where $Y^{a,n}_{j}=\left(I-D_{-}D_{+}\right)^{-1}a^{n}_{j}\quad\text{and}\quad Y^{b,n}_{j}=\left(I-D_{-}D_{+}\right)^{-1}b^{n}_{j}.$ Now $\left\|Y^{b,n}\right\|_{\ell^{1}}\leq\left\|b^{n}\right\|_{\ell^{1}}$, and therefore we compute $\displaystyle\left\|A^{n}q^{n}\right\|_{\ell^{1}}$ $\displaystyle={\Delta x}\sum_{j}\left|A^{n}_{j}q_{j}^{n}\right|\leq C\left\|q^{n}\right\|_{\ell^{2}}\left({\Delta x}\sum_{j}\left[\left(P_{j}^{n}\right)^{2}+\left(u_{j+1/2}^{n}\right)^{2}\right]\right)^{1/2}$ $\displaystyle\leq C\left\|q^{n}\right\|_{\ell^{2}}\left({\Delta x}\sum_{j}\left[\left|P_{j}^{n}\right|+\left(u_{j+1/2}^{n}\right)^{2}\right]\right)^{1/2}\leq C.$ By (6.13), the $L^{1}$ norm of ${\Delta t}(D_{+}^{t}q_{j}^{n})^{2}$ is of the same order as ${\Delta x}^{\theta}$. Then, summing (7.8) over $n$ and $j$, we arrive at ${\Delta t}{\Delta x}\sum_{n,j}I_{{\Delta x},f^{\prime\prime},j}\leq CT+{\Delta x}\sum_{j}\left[\left(q^{N}_{j}\right)^{2}+\left(q^{0}_{j}\right)^{2}\right]+\mathcal{O}\left({\Delta x}^{\theta}\right)\leq C.$ This means that (7.9) ${\Delta t}{\Delta x}\sum_{n,j}\left|Y^{b,n}_{j}\right|\leq C.$ Now let $L_{j}=h\sum_{i}e^{-\kappa\left|i-j\right|}D_{\pm}K_{i}=-h\sum_{i}D_{\mp}\bigl{(}e^{-\kappa\left|i-j\right|}\bigr{)}K_{i},$ for some sequence $\left\\{K_{j}\right\\}_{j}\in\ell^{1}$. Since $\left|D_{\pm}e^{-\kappa\left|i-j\right|}\right|\leq Ce^{-\kappa\left|i-j\right|},$ we get $\left\|L\right\|_{\ell^{1}}={\Delta x}\sum_{j}\left|L_{j}\right|\leq{\Delta x}hC\sum_{i,j}e^{-\kappa\left|i-j\right|}\left|K_{i}\right|\leq C\left\|K\right\|_{\ell^{1}}.$ Using this, $\left\|Y^{a,n}\right\|_{\ell^{1}}\leq C\left\|q^{n}\right\|_{\ell^{2}}^{2}.$ Combining this with (7.9) and (7.7) we see that ${\Delta t}{\Delta x}\sum_{n,j}\left|D_{+}^{t}P_{j}^{n}\right|\leq C,$ and therefore $\int_{0}^{T}\int_{\mathbb{R}}\left|\partial_{t}P_{{\Delta x}}\right|\,dx\,dt\leq C.$ Hence $\left\\{P_{{\Delta x}}\right\\}_{{\Delta x}>0}$ is bounded in $W^{1,1}((0,T)\times\mathbb{R})$. Combining this with the $L^{\infty}$ estimates found in Lemma 4.2 yields the existence of a convergent subsequence as claimed in (7.6). ∎ ## 8\. Strong convergence result We now show that the sequence $\left\\{q_{{\Delta x}}\right\\}_{{\Delta x}>0}$, cf. (3.9), has a strongly converging subsequence. This result is a key point of the convergence analysis. ###### Lemma 8.1. Fix $1\leq p<3$ and $1\leq r<1+\frac{\theta}{2}$. Then there exist two functions $q\in L^{p}_{\mathrm{loc}}((0,T)\times\mathbb{R})$, $\overline{q^{2}}\in L^{r}_{\mathrm{loc}}((0,T)\times\mathbb{R})$ such that for a subsequence as ${\Delta x}\to 0$, (8.1) $\displaystyle\text{$q_{{\Delta x}}\overset{\star}{\rightharpoonup}q$ in $L^{\infty}(0,T;L^{2}(\mathbb{R}))$},$ $\displaystyle\text{$q_{{\Delta x}}\rightharpoonup q$ in $L^{p}_{\mathrm{loc}}((0,T)\times\mathbb{R})$},$ (8.2) $\displaystyle\text{$q_{{\Delta x}}^{2}\rightharpoonup\overline{q^{2}}$ in $L^{r}_{\mathrm{loc}}((0,T)\times\mathbb{R})$},$ for all $a,b\in\mathbb{R}$, $a<b$. Moreover, (8.3) $q^{2}(t,x)\leq\overline{q^{2}}(t,x)$ for a.e. $(t,x)\in(0,T)\times\mathbb{R}$ and (8.4) $\partial_{x}u=q\quad\text{in the sense of distributions on $(0,T)\times\mathbb{R}$.}$ Finally, there is a positive constant $C$ such that (8.5) $q(t,x)\leq\frac{2}{t}+C\left\|u_{0}\right\|_{H^{1}(\mathbb{R})},\qquad t\in(0,T),\,x\in\mathbb{R}.$ ###### Proof. Claims (8.1), (8.2) are direct consequences of Lemmas 4.1 and 6.1. Claim (8.3) is true thanks to (8.2) and the convexity of $g=q^{2}$, cf. Lemma 2.3, while (8.4) is a consequence of the definitions of $q_{{\Delta x}}$ and $u_{{\Delta x}}$, cf. (3.9) and (3.10). We conclude by proving (8.5). Fix $t>0$, and let ${\Delta t}$ be so small that $t\in I^{n}$ with $n>0$. Not that $n\to\infty$ as ${\Delta t}\to 0$, and $t^{n}$ and $t^{n+1}$ both tend to $t$. From the definition of $q_{{\Delta x}}$ and (5.1) we have that (8.6) $\begin{split}q_{{\Delta x}}(t,x)&=q_{j}^{n}+(t-t^{n})D_{+}^{t}q_{j}^{n}\\\ &=\frac{t^{n+1}-t}{{\Delta t}}q_{j}^{n}+\frac{t-t^{n}}{{\Delta t}}q^{n+1}_{j}\\\ &\leq\frac{t^{n+1}-t}{{\Delta t}}\left(\frac{2}{t^{n}}+\hat{C}\right)+\frac{t-t^{n}}{{\Delta t}}\left(\frac{2}{t^{n}}+\hat{C}\right)\\\ &=\frac{2}{t}+\hat{C}+2f_{{\Delta t}}(t),\end{split}$ where $\hat{C}:=C\left\|u_{0}\right\|_{H^{1}(\mathbb{R})}$ and for every $t\in[t^{n},t^{n+1})$, $x\in I_{j}$, with $f_{{\Delta t}}(t)=\frac{t-t^{n}}{{\Delta t}}\frac{1}{t^{n+1}}+\frac{t^{n+1}-t}{{\Delta t}}\frac{1}{t^{n}}-\frac{1}{t}.$ Observe that $f_{{\Delta t}}^{\prime}(t)=\frac{1}{{\Delta t}}\left(\frac{1}{t^{n+1}}-\frac{1}{t^{n}}\right)+\frac{1}{t^{2}}=-\frac{1}{t^{n}t^{n+1}}+\frac{1}{t^{2}},$ so $f_{{\Delta t}}^{\prime}(t)=0\Longleftrightarrow t=\sqrt{t^{n}t^{n+1}}\in(t_{n},t_{n+1})$, and in particular $\displaystyle\sup\limits_{t\in[t^{n},t^{n+1})}f_{{\Delta t}}(t)$ $\displaystyle=f_{{\Delta t}}(\sqrt{t^{n}t^{n+1}})=\frac{\left(\sqrt{t^{n+1}}-\sqrt{t^{n}}\right)^{2}}{t^{n}t^{n+1}}$ $\displaystyle\leq\left(\frac{{\Delta t}}{2\sqrt{t^{n}}}\right)^{2}\frac{1}{t^{n}t^{n+1}}\leq\frac{{\Delta t}^{2}}{8(t^{n})^{2}t^{n+1}}\to 0.$ Therefore (8.5) follows from (8.1) and (8.6). ∎ In view of the weak convergences stated in (8.1), we have that for any function $f\in C^{1}(\mathbb{R})$ with $f^{\prime}$ bounded (8.7) $\begin{split}&f(q_{{\Delta x}})\overset{\star}{\rightharpoonup}\overline{f(q)}\quad\text{in $L^{\infty}(0,T;L^{2}(\mathbb{R}))$},\\\ &\text{$f(q_{{\Delta x}})\rightharpoonup\overline{f(q)}$ in $L^{p}_{\mathrm{loc}}((0,T)\times\mathbb{R})$, $1\leq p<3$},\end{split}$ where the same subsequence of ${\Delta x}\to 0$ applies to any $f$ from the specified class. In what follows, we let $\overline{qf(q)}$ and $\overline{f^{\prime}(q)q^{2}}$ denote the weak limits of $q_{{\Delta x}}f(q_{{\Delta x}})$ and $f^{\prime}(q_{{\Delta x}})q_{{\Delta x}}^{2}$, respectively, in $L^{r}_{\mathrm{loc}}((0,T)\times\mathbb{R})$, $1\leq r<\frac{3}{2}$. ###### Lemma 8.2. For any convex function $f\in C^{1}(\mathbb{R})$ with $f^{\prime}$ bounded we have that $\begin{split}\iint_{(0,T)\times\mathbb{R}}&\left(\overline{f(q)}\partial_{t}\varphi+u\overline{f(q)}\partial_{x}\varphi\right)\,dx\,dt+\int_{\mathbb{R}}f(q_{0}(x))\varphi(0,x)\,dx\\\ &\qquad\geq\iint_{(0,T)\times\mathbb{R}}\left(\frac{1}{2}\overline{f^{\prime}(q)q^{2}}-\overline{qf(q)}+\left(P-u^{2}\right)\overline{f^{\prime}(q)}\right)\varphi\,dx\,dt,\end{split}$ for any nonnegative $\varphi\in C^{\infty}_{c}([0,T)\times\mathbb{R})$. ###### Proof. Set (8.8) $\varphi_{j}(t)=\frac{1}{{\Delta x}}\int_{I_{j}}\varphi(t,x)\,dx,\quad\varphi_{j}^{n}=\frac{1}{{\Delta x}{\Delta t}}\iint_{I_{j}^{n}}\varphi(t,x)\,dx\,dt.$ We multiply (6.2) by ${\Delta x}{\Delta t}\varphi_{j}^{n}$, sum over $n,j$, and take into account the convexity of $f$. After partial summations, the final result reads (8.9) $E_{0}+E_{1}+E_{2}+E_{3}+E_{4}+E_{5}\geq 0,$ where $\displaystyle E_{0}=$ $\displaystyle{\Delta x}\sum_{j}f(q_{j}^{0})\varphi_{j}^{0},$ $\displaystyle E_{1}=$ $\displaystyle{\Delta t}{\Delta x}\sum\limits_{n,j}f(q_{j}^{n})D_{t}^{-}\varphi_{j}^{n},$ $\displaystyle E_{2}=$ $\displaystyle{\Delta t}{\Delta x}\sum\limits_{n,j}\left[\left(u_{j+1/2}^{n}\vee 0\right)f(q_{j}^{n})D_{+}\varphi_{j}^{n}+\left(u_{j-1/2}^{n}\wedge 0\right)f(q_{j}^{n})D_{-}\varphi_{j}^{n}\right],$ $\displaystyle E_{3}=$ $\displaystyle-{\Delta t}{\Delta x}\sum\limits_{n,j}\left[\frac{(q_{j}^{n})^{2}}{2}f^{\prime}(q_{j}^{n})-q_{j}^{n}f(q_{j}^{n})\right]\varphi_{j}^{n},$ $\displaystyle E_{4}=$ $\displaystyle-{\Delta t}{\Delta x}\sum\limits_{n,j}A_{j}^{n}f^{\prime}(q_{j}^{n})\varphi_{j}^{n},$ $\displaystyle E_{5}=$ $\displaystyle{\Delta t}^{2}{\Delta x}\sum\limits_{n,j}f^{\prime\prime}\left(q^{n+1/2}_{j}\right)\left(D_{+}^{t}q_{j}^{n}\right)^{2}\varphi_{j}^{n}.$ By (3.8), $\left\|q_{0}-q_{{\Delta x},0}\right\|_{L^{2}(\mathbb{R})}\to 0\quad\text{as ${\Delta x}\to 0$,}$ where $q_{0}:=\partial_{x}u_{0}$ and $q_{{\Delta x},0}:=\partial_{x}u_{{\Delta x}}|_{t=0}$, so that ${\Delta x}\sum_{j}f(q_{j}^{0})\varphi_{j}^{0}\to\int_{\mathbb{R}}f(q_{0}(x))\varphi(0,x)\,dx\quad\text{as ${\Delta x}\to 0$}.$ We split $E_{1}$ in three parts: (8.10) $E_{1}=E_{1,1}+E_{1,2}+E_{1,3},$ where $\displaystyle E_{1,1}=$ $\displaystyle\iint_{(0,T)\times\mathbb{R}}f(q_{{\Delta x}})\partial_{t}\varphi\,dx\,dt,$ $\displaystyle E_{1,2}=$ $\displaystyle\sum\limits_{n,j}\iint_{I_{j}^{n}}\left(f(q_{j}^{n})-f(q_{{\Delta x}})\right)D_{t}^{-}\varphi_{j}^{n}\,dx\,dt\,$ $\displaystyle E_{1,3}=$ $\displaystyle\sum\limits_{n,j}\iint_{I_{j}^{n}}f(q_{{\Delta x}})\left(D_{t}^{-}\varphi_{j}^{n}-\partial_{t}\varphi\right)\,dt.$ Due to (8.7) (8.11) $E_{1,1}\to\iint_{(0,T)\times\mathbb{R}}\overline{f(q)}\partial_{t}\varphi\,dx\,dt.$ Due to the boundedness of $f^{\prime}$, (8.12) $\left|f(q_{j}^{n})-f(q_{{\Delta x}})\right|\leq c_{1}(t-t^{n})|D_{+}^{t}q_{j}^{n}|,$ for each $(t,x)\in[t^{n},t^{n+1})\times I_{j}$, where $c_{1}>0$ is a finite constant. In view of (6.13), (8.13) $\begin{split}\left|E_{1,2}\right|&\leq c_{1}{\Delta x}\sum\limits_{n,j}\left|D_{t}^{-}\varphi_{j}^{n}\right||D_{+}^{t}q_{j}^{n}|\int_{t^{n}}^{t^{n+1}}(t-t^{n})\,dt\\\ &\leq c_{1}{\Delta x}{\Delta t}^{2}\sum\limits_{n,j}\left|D_{t}^{-}\varphi_{j}^{n}\right||D_{+}^{t}q_{j}^{n}|\leq c_{1}{\Delta t}\left\|D_{t}^{-}\varphi\right\|_{\ell^{2}}\left\|D_{+}^{t}q\right\|_{\ell^{2}}\to 0,\end{split}$ as ${\Delta x}\to 0$. Finally, since $\left|D_{t}^{-}\varphi-\partial_{t}\varphi\right|\leq c_{2}{\Delta x},$ for some constant $c_{2}>0$, we have that (8.14) $\left|E_{1,3}\right|\leq c_{2}{\Delta x}\iint_{\mathrm{supp}(\varphi)}\left|f(q_{{\Delta x}})\right|\,dx\,dt\to 0\quad\text{as ${\Delta x}\to 0$.}$ Clearly (8.10), (8.11), (8.13), and (8.14) imply (8.15) $E_{1}\to\iint_{(0,T)\times\mathbb{R}}\overline{f(q)}\partial_{t}\varphi\,dx\,dt\quad\text{as ${\Delta x}\to 0$.}$ Next, we split $E_{2}$ into four parts: (8.16) $E_{2}=E_{2,1}+E_{2,2}+E_{2,3}+E_{2.4},$ where $\displaystyle E_{2,1}=$ $\displaystyle\iint_{(0,T)\times\mathbb{R}}u_{{\Delta x}}f(q_{{\Delta x}})\partial_{x}\varphi\,dx\,dt,$ $\displaystyle E_{2,2}=$ $\displaystyle\sum\limits_{n,j}\iint_{I_{j}^{n}}\Biggl{[}\left((u_{j+1/2}^{n}\vee 0)-(u_{{\Delta x}}\vee 0)\right)f(q_{j}^{n})D_{+}\varphi_{j}^{n}$ $\displaystyle\qquad\qquad\qquad+\left((u_{j-1/2}^{n}\wedge 0)-(u_{{\Delta x}}\wedge 0)\right)f(q_{j}^{n})D_{-}\varphi_{j}^{n}\Biggr{]}\,dx\,dt,$ $\displaystyle E_{2,3}=$ $\displaystyle\sum\limits_{n,j}\iint_{I_{j}^{n}}\Biggl{[}(u_{{\Delta x}}\vee 0)\left(f(q_{j}^{n})-f(q_{{\Delta x}})\right)D_{+}\varphi_{j}^{n}$ $\displaystyle\qquad\qquad\qquad+(u_{{\Delta x}}\wedge 0)\left(f(q_{j}^{n})-f(q_{{\Delta x}})\right)D_{-}\varphi_{j}^{n}\Biggr{]}\,dx\,dt,$ $\displaystyle E_{2,4}=$ $\displaystyle\sum\limits_{n,j}\iint_{I_{j}^{n}}\Biggl{[}\left(u_{{\Delta x}}\vee 0\right)f(q_{{\Delta x}})\left(D_{+}\varphi_{j}^{n}-\partial_{x}\varphi\right)$ $\displaystyle\qquad\qquad\qquad+(u_{{\Delta x}}\wedge 0)f(q_{{\Delta x}})\left(D_{-}\varphi_{j}^{n}-\partial_{x}\varphi\right)\Biggr{]}\,dx\,dt.$ Due to (8.7), (8.17) $E_{2,1}\to\iint_{(0,T)\times\mathbb{R}}\overline{qf(q)}\partial_{x}\varphi\,dx\,dt.$ Using the definition of $u_{{\Delta x}}$, $\begin{split}\left|u_{j-1/2}^{n}-u_{{\Delta x}}\right|&\leq{\Delta x}\left(\left|q_{j}^{n}\right|+\left|q_{j}^{n+1}\right|\right),\\\ \left|u_{j-1/2}^{n}-u_{{\Delta x}}\right|&\leq{\Delta x}q_{j}^{n}\left(\left|q_{j}^{n}\right|+\left|q_{j}^{n+1}\right|\right)+\left|u_{j+1/2}^{n}-u_{j-1/2}^{n}\right|,\end{split}$ so, by Lemmas 4.1 and 7.1, (8.18) $\begin{split}\left|E_{2,2}\right|&\leq{\Delta x}^{2}{\Delta t}\sum\limits_{n,j}\left(\left|q_{j}^{n}\right|+\left|q_{j}^{n+1}\right|\right)\left|f(q_{j}^{n})\right|\left(\left|D_{+}\varphi_{j}^{n}\right|+\left|D_{-}\varphi_{j}^{n}\right|\right)\\\ &\qquad\qquad\quad+{\Delta x}{\Delta t}\sum\limits_{n,j}\left|u_{j+1/2}^{n}-u_{j-1/2}^{n}\right|\left|f(q_{j}^{n})\right|\left|D_{+}\varphi_{j}^{n}\right|\\\ &\leq 2{\Delta x}\left\|D_{+}\varphi\right\|_{\ell^{\infty}}\left\|q\right\|_{\ell^{2}}\left\|f(q)\right\|_{\ell^{2}}\\\ &\qquad\qquad\quad+\left\|f(q)\right\|_{\ell^{2}}\left\|\left\\{\left(u_{j+1/2}^{n}-u_{j-1/2}^{n}\right)D_{+}\varphi_{j}^{n}\right\\}_{n,j}\right\|_{\ell^{2}}\to 0,\end{split}$ as ${\Delta x}\to 0$. Using (8.12) and (6.13), we deduce (8.19) $\begin{split}\left|E_{2,3}\right|&\leq c_{1}{\Delta x}\sum\limits_{n,j}\left(\left|D_{-}\varphi_{j}^{n}\right|+\left|D_{+}\varphi_{j}^{n}\right|\right)\\\ &\qquad\qquad\quad\times\left(\left|u_{j+1/2}^{n}\right|+\left|u_{j-1/2}^{n}\right|\right)\int_{t^{n}}^{t^{n+1}}(t-t^{n})\,dt\\\ &=c_{1}{\Delta x}{\Delta t}^{2}\sum\limits_{n,j}\left(\left|D_{-}\varphi_{j}^{n}\right|+\left|D_{+}\varphi_{j}^{n}\right|\right)\left(\left|u_{j+1/2}^{n}\right|+\left|u_{j-1/2}^{n}\right|\right)\\\ &\leq 4c_{1}{\Delta t}\left\|D_{+}\varphi\right\|_{\ell^{2}}\left\|u\right\|_{\ell^{2}}\to 0,\end{split}$ as ${\Delta x}\to 0$. Finally, since $\left|D_{\pm}\varphi-\partial_{t}\varphi\right|\leq c_{3}{\Delta x}$, for some constant $c_{3}>0$, we obtain (8.20) $\left|E_{2,4}\right|\leq c_{3}{\Delta x}\iint_{\mathrm{supp}(\varphi)}\left|q_{{\Delta x}}\right|\left|f(q_{{\Delta x}})\right|\,dt\,dx\to 0\quad\text{as ${\Delta x}\to 0$.}$ Note that (8.16), (8.17), (8.18), (8.19), and (8.20) imply (8.21) $E_{2}\to\iint_{(0,T)\times\mathbb{R}}\overline{qf(q)}\partial_{x}\varphi\,dx\,dt\quad\text{as ${\Delta x}\to 0$.}$ We split $E_{3}$ into three parts: (8.22) $E_{3}=E_{3,1}+E_{3,2}+E_{3,3},$ where $\displaystyle E_{3,1}$ $\displaystyle=-\iint_{(0,T)\times\mathbb{R}}\left[\frac{(q_{{\Delta x}})^{2}}{2}f^{\prime}(q_{{\Delta x}})-q_{{\Delta x}}f(q_{{\Delta x}})\right]\varphi\,dx\,dt,$ $\displaystyle E_{3,2}$ $\displaystyle=-\sum\limits_{n,j}\iint_{I_{j}^{n}}\Biggl{[}\frac{(q_{j}^{n})^{2}}{2}f^{\prime}(q_{j}^{n})-q_{j}^{n}f(q_{j}^{n})$ $\displaystyle\qquad\qquad\qquad\qquad\quad-\frac{(q_{{\Delta x}})^{2}}{2}f^{\prime}(q_{{\Delta x}})+q_{{\Delta x}}f(q_{{\Delta x}})\Biggr{]}\varphi_{j}^{n}\,dx\,dt,$ $\displaystyle E_{3,3}$ $\displaystyle=-\sum\limits_{n,j}\iint_{I_{j}^{n}}\left[\frac{(q_{{\Delta x}})^{2}}{2}f^{\prime}(q_{{\Delta x}})-q_{{\Delta x}}f(q_{{\Delta x}})\right]\left(\varphi_{j}^{n}-\varphi\right)\,dx\,dt,$ Due to (8.7), (8.23) $E_{3,1}\to-\iint_{(0,T)\times\mathbb{R}}\left(\frac{\overline{q^{2}f^{\prime}(q)}}{2}-\overline{qf(q)}\right)\varphi\,dx\,dt\quad\text{as ${\Delta x}\to 0$.}$ Using the boundedness of $f^{\prime}$, we can estimate as follows: $\displaystyle\left|\frac{(q_{j}^{n})^{2}}{2}f^{\prime}(q_{j}^{n})-q_{j}^{n}f(q_{j}^{n})-\frac{(q_{{\Delta x}})^{2}}{2}f^{\prime}(q_{{\Delta x}})+q_{{\Delta x}}f(q_{{\Delta x}})\right|$ $\displaystyle\leq\frac{\left|(q_{j}^{n})^{2}-q_{{\Delta x}}^{2}\right|}{2}|f^{\prime}(q_{j}^{n})|+\frac{(q_{{\Delta x}})^{2}}{2}\left|f^{\prime}(q_{j}^{n})-f^{\prime}(q_{{\Delta x}})\right|$ $\displaystyle\leq\frac{\left|q_{j}^{n}-q_{{\Delta x}}\right|(\left|q_{j}^{n}\right|+\left|q_{{\Delta x}}\right|)}{2}\left\|f^{\prime}\right\|_{L^{\infty}}+\frac{(q_{{\Delta x}})^{2}}{2}\left\|f^{\prime\prime}\right\|_{L^{\infty}}\left|q_{j}^{n}-q_{{\Delta x}}\right|$ $\displaystyle\leq\frac{(t-t^{n})\left|D_{+}^{t}q_{j}^{n}\right|(\left|q_{j}^{n}\right|+\left|q_{{\Delta x}}\right|)}{2}\left\|f^{\prime}\right\|_{L^{\infty}}+\frac{(q_{{\Delta x}})^{2}}{2}\left\|f^{\prime\prime}\right\|_{L^{\infty}}(t-t^{n})\left|D_{+}^{t}q_{j}^{n}\right|.$ Hence, taking into account (6.13) and (6.3), (8.24) $\begin{split}\left|E_{3,2}\right|&\leq{\Delta x}\sum\limits_{n,j}\int_{I^{n}}\Biggl{[}\frac{(t-t^{n})\left|D_{+}^{t}q_{j}^{n}\right|(\left|q_{j}^{n}\right|+\left|q_{{\Delta x}}\right|)}{2}\left\|f^{\prime}\right\|_{L^{\infty}}\\\ &\qquad\qquad\qquad\qquad\qquad+\frac{(q_{{\Delta x}})^{2}}{2}\left\|f^{\prime\prime}\right\|_{L^{\infty}}(t-t^{n})\left|D_{+}^{t}q_{j}^{n}\right|\Biggr{]}\varphi_{j}^{n}\,dt\\\ &\leq{\Delta x}{\Delta t}^{2}\sum\limits_{n,j}\left[\frac{\left|D_{+}^{t}q_{j}^{n}\right|(\left|q_{j}^{n}\right|+\left|q_{{\Delta x}}\right|)}{2}\left\|f^{\prime}\right\|_{L^{\infty}}+\frac{(q_{{\Delta x}})^{2}}{2}\left\|f^{\prime\prime}\right\|_{L^{\infty}}\left|D_{+}^{t}q_{j}^{n}\right|\right]\varphi_{j}^{n}\\\ &\leq{\Delta t}\Biggl{[}\frac{\left\|D_{+}^{t}q\right\|_{\ell^{2}}(\left\|q^{n}\right\|_{\ell^{2}}+\left\|q_{{\Delta x}}\right\|_{L^{2}(\mathbb{R})})}{2}\left\|f^{\prime}\right\|_{L^{\infty}}\\\ &\qquad\qquad\qquad\qquad\qquad+\mathcal{O}\left(\frac{1}{{\Delta x}}\right)\left\|u\right\|_{\ell^{\infty}}\left\|q\right\|_{\ell^{2}}\left\|D_{+}^{t}q^{n}\right\|_{\ell^{2}}\Biggl{]}\left\|\varphi\right\|_{L^{\infty}}\\\ &\leq{\Delta t}\left[\mathcal{O}\left(\sqrt{\frac{{\Delta x}^{\theta}}{{\Delta t}}}+\sqrt{\frac{{\Delta x}^{\theta}}{{\Delta t}}}\frac{1}{{\Delta x}}\right)\right]=\mathcal{O}\left({\Delta x}^{\theta}\right)\to 0\quad\text{as ${\Delta x}\to 0$.}\end{split}$ Since $\left|\varphi_{j}^{n}-\varphi\right|={\mathcal{O}}({\Delta x})$, (8.25) $\left|E_{3,3}\right|\leq{\mathcal{O}}({\Delta x})\iint_{\mathrm{supp}\,(\varphi)}\left|\frac{(q_{{\Delta x}})^{2}}{2}f^{\prime}(q_{{\Delta x}})-q_{{\Delta x}}f(q_{{\Delta x}})\right|\,dx\,dt\to 0\quad\text{as ${\Delta x}\to 0$}.$ We have that (8.22), (8.23), (8.24), and (8.25) imply (8.26) $E_{3}\to\iint_{(0,T)\times\mathbb{R}}\overline{qf(q)}\partial_{x}\varphi\,dx\,dt\quad\text{as ${\Delta x}\to 0$.}$ We split the term $E_{4}$ into three parts: $E_{4}=E_{4,1}+E_{4,2}+E_{4,3},$ where $\displaystyle E_{4,1}$ $\displaystyle=-\iint_{(0,T)\times\mathbb{R}}A_{{\Delta x}}f^{\prime}(q_{{\Delta x}})\varphi\,dx\,dt,$ $\displaystyle E_{4,2}$ $\displaystyle=-\sum\limits_{n,j}\iint_{I_{j}^{n}}\left(A_{j}^{n}f^{\prime}(q_{j}^{n})-A_{{\Delta x}}f^{\prime}(q_{{\Delta x}})\right)\varphi_{j}^{n}\,dx\,dt,$ $\displaystyle E_{4,3}$ $\displaystyle=-\sum\limits_{n,j}\iint_{I_{j}^{n}}A_{{\Delta x}}f^{\prime}(q_{{\Delta x}})\left(\varphi_{j}^{n}-\varphi\right)\,dx\,dt,$ where $A_{{\Delta x}}=P_{{\Delta x}}-\left(u_{{\Delta x}}\right)^{2}$. Lemmas 7.1 and 7.2, cf. also (8.7), imply that $E_{4,1}\to-\iint_{(0,T)\times\mathbb{R}}\left(P-u^{2}\right)\overline{f^{\prime}(q)}\varphi\,dx\,dt\quad\text{as ${\Delta x}\to 0$}.$ Continuing, it is not hard to see that $\left|E_{4,2}\right|=\mathcal{O}({\Delta x})\to 0$ as ${\Delta x}\to 0$. Moreover, since $\varphi_{j}^{n}-\varphi=\mathcal{O}({\Delta x})$, $E_{4,3}\to 0$ as ${\Delta x}\to 0$. Summarizing, (8.27) $E_{4}\to-\iint_{(0,T)\times\mathbb{R}}\left(P-u^{2}\right)\overline{f^{\prime}(q)}\varphi\,dx\,dt\quad\text{as ${\Delta x}\to 0$.}$ Finally, regarding $E_{5}$, due to (6.13), we conclude that as ${\Delta x}\to 0$ (8.28) $\left|E_{5}\right|\leq\left\|\varphi\right\|_{L^{\infty}((0,T)\times\mathbb{R})}{\Delta t}^{2}{\Delta x}\sum\limits_{n,j}f^{\prime\prime}\left(q^{n+1/2}_{j}\right)\left(D_{+}^{t}q_{j}^{n}\right)^{2}\varphi_{j}^{n}\to 0.$ The lemma now follows from (8.9), (8.15), (8.21), (8.26), (8.27), and (8.28). ∎ We know that $\left\\{(q_{{\Delta x}})^{2}\right\\}_{{\Delta x}>0}$ is bounded in $L^{\infty}(0,T;L^{1}(\mathbb{R}))\cap L_{\mathrm{loc}}^{r}((0,T)\times\mathbb{R})$, for any $1\leq r<1+\theta/2$. Additionally, using (6.2) with $f(q)=\tfrac{q^{2}}{2}$, we can show that the mapping $t\mapsto\int_{\mathbb{R}}(q_{{\Delta x}})^{2}\varphi\,dx$ is equi- continuous on $[0,T]$, for every $\varphi\in C^{\infty}_{c}(\mathbb{R})$. Hence, in view of Lemma 2.4, (8.29) $\int_{\mathbb{R}}(q_{{\Delta x}})^{2}\varphi\,dx\to\int_{\mathbb{R}}\overline{q^{2}}\varphi\,dx\quad\text{uniformly on $[0,T]$,}$ and (8.30) $t\mapsto\int_{\mathbb{R}}\overline{q^{2}}\varphi\,dx\quad\text{is continuous on $[0,T]$.}$ The statements (8.29) and (8.30) hold with $q_{{\Delta x}}^{2}$, $\overline{q^{2}}$ replaced respectively by $f(q_{{\Delta x}})$, $\overline{f(q_{{\Delta x}})}$, for any convex function $f\in C^{1}(\mathbb{R})$ with $f^{\prime}$ bounded. ###### Lemma 8.3. Let $q$ and $\overline{q^{2}}$ be the weak limits identified in Lemma 8.1. Then (8.31) $\begin{split}&\iint_{(0,T)\times\mathbb{R}}\left(q\partial_{t}\varphi+uq\partial_{x}\varphi\right)\,dx\,dt+\int_{\mathbb{R}}q_{0}(x)\varphi(0,x)\,dx\\\ &\quad=\iint_{(0,T)\times\mathbb{R}}\left(-\frac{1}{2}\overline{q^{2}}+\left(P-u^{2}\right)\right)\varphi\,dx\,dt,\quad\forall\varphi\in C^{\infty}_{c}((0,T)\times\mathbb{R}).\end{split}$ ###### Proof. Starting from (6.2) with $f(q)=q$, we argue as in the proof of Lemma 8.2 to conclude the validity of (8.31). ∎ The next lemma tells us that the weak limits in Lemma 8.1 satisfy the initial data in an appropriate sense. ###### Lemma 8.4. Let $q$ and $\overline{q^{2}}$ be the weak limits identified in Lemma 8.1. Then (8.32) $\begin{split}&\lim_{t\to 0}\int_{\mathbb{R}}q^{2}(t,x)\,dx=\int_{\mathbb{R}}(\partial_{x}u_{0})^{2}\,dx,\\\ &\lim_{t\to 0}\int_{\mathbb{R}}\overline{q^{2}}(t,x)\,dx=\int_{\mathbb{R}}(\partial_{x}u_{0})^{2}\,dx.\end{split}$ ###### Proof. The proof is similar to that in [12]. ∎ We can now wrap up the proof of the strong convergence of $\left\\{q_{{\Delta x}}\right\\}_{{\Delta x}>0}$. ###### Lemma 8.5. Let $q$ and $\overline{q^{2}}$ be the weak limits identified in Lemma 8.1. Then (8.33) $\overline{q^{2}}(t,x)=q^{2}(t,x)$ for a.e. $(t,x)\in(0,T)\times\mathbb{R}$. Consequently, as ${\Delta x}\to 0$ (along a subsequence if necessary) (8.34) $\text{$q_{{\Delta x}}\to q$ in $L^{2}_{\mathrm{loc}}((0,T)\times\mathbb{R})$ and a.e.~{}in $(0,T)\times\mathbb{R}$}.$ ###### Proof. By Lemma 8.2, (8.35) $\partial_{t}\overline{f(q)}+\partial_{x}\left(u\overline{f(q)}\right)\leq\overline{qf(q)}-\frac{1}{2}\overline{f^{\prime}(q)q^{2}}+\left(u^{2}-P\right)\overline{f^{\prime}(q)},$ in the sense of distributions on $(0,T)\times\mathbb{R}$, for any convex function $f\in C^{1}(\mathbb{R})$ with $f^{\prime}$ bounded Moreover, by Lemma 8.3, (8.36) $\partial_{t}q+\partial_{x}(uq)=\frac{1}{2}\overline{q^{2}}+u^{2}-P,$ in the sense of distributions on $(0,T)\times\mathbb{R}$. Equipped with (8.35), (8.36), (8.32), and (8.5), we can argue exactly as in Xin and Zhang [39] to arrive at (8.33). In view of Lemma 2.3, claim (8.34) follows immediately from (8.33) and (7.5). ∎ We now prove that the limit $u$ satisfies (D.3). ###### Lemma 8.6. For any $\varphi\in C^{\infty}_{c}((0,T)\times\mathbb{R})$, (8.37) $\begin{gathered}\int_{0}^{T}\int_{\mathbb{R}}u\varphi_{t}+\left(\frac{u^{2}}{2}+P\right)\varphi_{x}\,dx\,dt=0,\\\ \int_{0}^{T}\int_{\mathbb{R}}P\left(\varphi-\varphi_{xx}\right)\,dx\,dt=\int_{0}^{T}\int_{\mathbb{R}}\left(u^{2}+\frac{1}{2}\left(\partial_{x}u\right)^{2}\right)\varphi\,dx\,dt\end{gathered}$ ###### Proof. It is not difficult to establish the equation for $P$, since we have already established that $\partial_{x}u_{{\Delta x}}\to\partial_{x}u$ in $L^{2}_{\mathrm{loc}}((0,T)\times\mathbb{R})$, cf. (8.34). Indeed, we have $\displaystyle\int_{0}^{T}$ $\displaystyle\int_{\mathbb{R}}P_{{\Delta x}}\left(\varphi-\varphi_{xx}\right)\,dx\,dt$ $\displaystyle=\sum_{n,j}\iint_{I^{n}_{j-1/2}}P_{{\Delta x}}\left(\varphi-\varphi_{xx}\right)\,dx\,dt$ $\displaystyle=\sum_{n,j}P_{j}^{n}\iint_{I^{n}_{j-1/2}}\varphi-\varphi_{xx}\,dx\,dt$ $\displaystyle\qquad+\underbrace{\sum_{n,j}\iint_{I^{n}_{j-1/2}}\left(P_{j}^{n}-P_{{\Delta x}}\right)\left(\varphi-\varphi_{xx}\right)\,dx\,dt}_{w_{1}}$ $\displaystyle={\Delta t}{\Delta x}\sum_{n,j}P_{j}^{n}\left(\varphi^{n}_{j}-D_{-}D_{+}\varphi^{n}_{j}\right)$ $\displaystyle\qquad+w_{1}+\underbrace{\sum_{n,j}P_{j}^{n}\iint_{I^{n}_{j-1/2}}\left(\left(\varphi-\varphi^{n}_{j}\right)+\left(\varphi_{xx}-D_{-}D_{+}\varphi^{n}_{j}\right)\right)\,dx\,dt}_{w_{2}}.$ Since $\left|P_{{\Delta x}}(t,x)-P_{j}^{n}\right|\leq C{\Delta x}$ for $(t,x)\in I^{n}_{j-1/2}$, $w_{1}\to 0$ as ${\Delta x}\to 0$. Similarly, since $\varphi-\varphi_{xx}$ is close to $\varphi^{n}_{j}-D_{-}D_{+}\varphi^{n}_{j}$ in $I^{n}_{j-1/2}$ and $P_{j}^{n}$ is bounded, we conclude that $w_{2}\to 0$ as ${\Delta x}\to 0$. Set $\displaystyle f^{n}_{j}$ $\displaystyle=\left(u_{j-1/2}^{n}\wedge 0\right)^{2}+\left(u_{j+1/2}^{n}\vee 0\right)^{2}+\frac{1}{2}\left(q_{j}^{n}\right)^{2},$ $\displaystyle f_{{\Delta x}}$ $\displaystyle=\left(u_{{\Delta x}}\right)^{2}+\frac{1}{2}\left(\partial_{x}u_{{\Delta x}}\right)^{2}.$ Using the scheme for $P_{j}^{n}$, cf. (3.3), $\displaystyle{\Delta t}{\Delta x}\sum_{n,j}P_{j}^{n}\left(\varphi^{n}_{j}-D_{-}D_{+}\varphi^{n}_{j}\right)={\Delta t}{\Delta x}\sum_{n,j}f^{n}_{j}\varphi^{n}_{j}$ $\displaystyle\qquad=\int_{0}^{T}\int_{\mathbb{R}}f_{{\Delta x}}\varphi\,dx\,dt$ $\displaystyle\qquad\qquad+\underbrace{\sum_{n,j}\iint_{I^{n}_{j-1/2}}\left(f^{n}_{j}-f_{{\Delta x}}\right)\varphi\,dx\,dt}_{w_{3}}+\sum_{n,j}f^{n}_{j}\underbrace{\iint_{I^{n}_{j-1/2}}\left(\varphi-\varphi^{n}_{j}\right)\,dx\,dt}_{=0}.$ By the definition of $q_{{\Delta x}}$ we have that $q_{{\Delta x}}=q_{j}^{n}+(t-t^{n})D_{+}^{t}q_{j}^{n}$ for $t\in I^{n}$. Hence $(q_{{\Delta x}})^{2}-\left(q_{j}^{n}\right)^{2}=2(t-t^{n})q_{j}^{n}D_{+}^{t}q_{j}^{n}+\left(t-t^{n}\right)^{2}\left(D_{+}^{t}q_{j}^{n}\right)^{2}\quad\text{in $I^{n}_{j-1/2}$.}$ so that, assuming $\mathrm{supp}(\varphi)\subset[x_{j_{a}},x_{j_{b}}]$ for some intergers $j_{a}$ and $j_{b}$, $\displaystyle\sum_{n,j}\iint_{I^{n}_{j-1/2}}\left|\left(q_{j}^{n}\right)^{2}-\left(q_{{\Delta x}}\right)^{2}\right|\left|\varphi\right|\,dx\,dt$ $\displaystyle\quad\leq C{\Delta x}\sum_{n}\sum_{j_{a}}^{j_{b}}\int_{t^{n}}^{t^{n+1}}\left(2(t-t^{n})q_{j}^{n}D_{+}^{t}q_{j}^{n}+\left(t-t^{n}\right)^{2}\left(D_{+}^{t}q_{j}^{n}\right)^{2}\right)\,dt$ $\displaystyle\quad=C{\Delta t}{\Delta x}\sum_{n}\sum_{j=j_{a}}^{j_{b}}\left({\Delta t}q_{j}^{n}D_{+}^{t}q_{j}^{n}+\left({\Delta t}D_{+}^{t}q_{j}^{n}\right)^{2}\right)$ $\displaystyle\quad=\mathcal{O}\left(\sqrt{{\Delta t}{\Delta x}^{\theta}}+{\Delta t}{\Delta x}^{\theta}\right).$ By the Hölder continuity of $u_{{\Delta x}}$; recall that $u_{{\Delta x}}\in C^{0,\ell}$ with $\ell=1-2/(2+\alpha)$, we find $\left|\left(u_{j-1/2}^{n}\wedge 0\right)^{2}+\left(u_{j+1/2}^{n}\vee 0\right)^{2}-\left(u_{{\Delta x}}\right)^{2}\right|=\mathcal{O}({\Delta x}^{\ell}+{\Delta t}^{\ell}),$ and therefore $w_{3}\to 0$ as ${\Delta x}\to 0$. Hence, using (7.2), (8.4), and (8.34), $\displaystyle\int_{0}^{T}\int_{\mathbb{R}}P\left(\varphi-\varphi_{xx}\right)\,dx\,dt$ $\displaystyle=\lim_{{\Delta x}\to 0}\int_{0}^{T}\int_{\mathbb{R}}P_{{\Delta x}}\left(\varphi-\varphi_{xx}\right)\,dx\,dt$ $\displaystyle=\lim_{{\Delta x}\downarrow 0}\int_{0}^{T}\int_{\mathbb{R}}\left(\left(u_{{\Delta x}}\right)^{2}+\frac{1}{2}\left(\partial_{x}u_{{\Delta x}}\right)^{2}\right)\varphi\,dx\,dt$ $\displaystyle=\int_{0}^{T}\int_{\mathbb{R}}\left(u^{2}+\frac{1}{2}\left(\partial_{x}u\right)^{2}\right)\varphi\,dx\,dt.$ This means that the second equation in (8.37) holds. To establish the first equality in (8.37), we derive a divergence-form version of the scheme (3.1). To this end, introduce the functions $f_{\vee}(u)=\frac{1}{2}(u\vee 0)^{2}$ and $f_{\wedge}(u)=\frac{1}{2}(u\wedge 0)^{2}$. Observe that $f_{\vee}$ and $f_{\wedge}$ are piecewise $C^{2}$, and the absolute value of the second derivatives are bounded by $1$. By the discrete chain rule, $\left(u_{j+1/2}^{n}\vee 0\right)D_{-}u_{j+1/2}^{n}=D_{-}f_{\vee}(u_{j+1/2}^{n})+\mathcal{O}\left({\Delta x}\left(D_{-}u_{j+1/2}^{n}\right)^{2}\right)$ and $\left(u_{j+1/2}^{n}\wedge 0\right)D_{+}u_{j+1/2}^{n}=D_{+}f_{\wedge}(u_{j+1/2}^{n})+\mathcal{O}\left({\Delta x}\left(D_{+}u_{j+1/2}^{n}\right)^{2}\right).$ Consequently, we can replace (3.1) by (8.38) $\begin{split}D_{+}^{t}u_{j+1/2}^{n}&+D_{-}f_{\vee}(u_{j+1/2}^{n})+D_{+}f_{\wedge}(u_{j+1/2}^{n})+D_{+}P_{j}^{n}\\\ &=\mathcal{O}\left({\Delta x}\left\\{\left(D_{-}u_{j+1/2}^{n}\right)^{2}+\left(D_{+}u_{j+1/2}^{n}\right)^{2}\right\\}\right).\end{split}$ Observe that (8.39) $D=\frac{D_{-}+D_{+}}{2},\qquad{\Delta x}D_{-}D_{+}=D_{+}-D_{-},\qquad f_{\vee}+f_{\wedge}=\frac{u^{2}}{2}.$ Using these identities, we can restate (8.38) as (8.40) $\begin{split}D_{+}^{t}u_{j+1/2}^{n}&+D_{-}\left[\frac{\left(u_{j+1/2}^{n}\right)^{2}}{4}+\frac{1}{2}\left(f_{\vee}(u_{j+1/2}^{n})-f_{\wedge}(u_{j+1/2}^{n})\right)\right]\\\ &+D_{+}\left[\frac{\left(u_{j+1/2}^{n}\right)^{2}}{4}+\frac{1}{2}\left(f_{\wedge}(u_{j+1/2}^{n})-f_{\vee}(u_{j+1/2}^{n})\right)\right]+D_{+}P_{j}^{n}\\\ &\qquad\qquad={\mathcal{O}}\left({\Delta x}\left\\{\left(D_{-}u_{j+1/2}^{n}\right)^{2}+\left(D_{+}u_{j+1/2}^{n}\right)^{2}\right\\}\right).\end{split}$ Using, cf. (8.39), $\displaystyle D_{-}\left(f_{\vee}(u_{j+1/2}^{n})-f_{\wedge}(u_{j+1/2}^{n})\right)+D_{+}\left(f_{\wedge}(u_{j+1/2}^{n})-f_{\vee}(u_{j+1/2}^{n})\right)$ $\displaystyle\qquad={\Delta x}D_{-}D_{+}f_{\wedge}(u_{j+1/2}^{n})-{\Delta x}D_{-}D_{+}f_{\vee}(u_{j+1/2}^{n}),$ equation (8.40) becomes (8.41) $\begin{split}D_{+}^{t}u_{j+1/2}^{n}&+D\left(\frac{\left(u_{j+1/2}^{n}\right)^{2}}{2}\right)+D_{+}P_{j}^{n}\\\ &=\mathcal{O}\left({\Delta x}\left\\{\left(D_{-}u_{j+1/2}^{n}\right)^{2}+\left(D_{+}u_{j+1/2}^{n}\right)^{2}\right\\}\right)\\\ &\qquad+{\Delta x}\left\\{D_{-}D_{+}f_{\vee}(u_{j+1/2}^{n})-D_{-}D_{+}f_{\wedge}(u_{j+1/2}^{n})\right\\}.\end{split}$ Now fix $\varphi\in C^{2}_{c}((0,T)\times\mathbb{R})$ and define $\varphi^{n}_{j}$ as before, cf. (8.8). Multiplying (8.41) by $\varphi^{n}_{j}{\Delta t}{\Delta x}$ and performing partial summations gives $\displaystyle\underbrace{{\Delta t}{\Delta x}\sum_{n,j}u_{j+1/2}^{n}D_{+}^{t}\varphi^{n}_{j}}_{E_{1}}\\!+\\!\underbrace{{\Delta t}{\Delta x}\sum_{n,j}\frac{\left(u_{j+1/2}^{n}\right)^{2}}{2}D\varphi^{n}_{j}}_{E_{2}}\\!+\\!\underbrace{{\Delta t}{\Delta x}\sum_{n,j}P_{j}^{n}D_{-}\varphi^{n}_{j}}_{E_{3}}=\mathcal{O}({\Delta x}),$ by using (4.2). We have $\left|u_{{\Delta x}}-u_{j-1/2}^{n}\right|\leq C\left({\Delta x}\left|q_{j}^{n}\right|+{\Delta t}\left|D_{+}^{t}u_{j+1/2}^{n}\right|\right)$. Using this and (7.2), we compute as follows: $\displaystyle E_{1}$ $\displaystyle=\int_{0}^{T}\int_{\mathbb{R}}u_{{\Delta x}}\varphi_{t}\,dx\,dt+\sum_{n,j}\iint_{I^{n}_{j}}\left(u_{j+1/2}^{n}-u_{{\Delta x}}\right)\varphi_{t}\,dx\,dt$ $\displaystyle\qquad\qquad+\sum_{n,j}u_{j+1/2}^{n}\iint_{I^{n}_{j}}\left(D_{+}^{t}\varphi^{n}_{j}-\varphi_{t}\right)\,dx\,dt$ $\displaystyle=\int_{0}^{T}\int_{\mathbb{R}}u_{{\Delta x}}\varphi_{t}\,dx\,dt+\mathcal{O}({\Delta x})\to\int_{0}^{T}\int_{\mathbb{R}}u\varphi_{t}\,dx\,dt\quad\text{as ${\Delta x}\to 0$.}$ In the same way, equipped with (7.2) and (7.6), we can show that $E_{2}\to\int_{0}^{T}\int_{\mathbb{R}}\frac{u^{2}}{2}\varphi_{x}\,dx\,dt,\quad E_{3}\to\int_{0}^{T}\int_{\mathbb{R}}P\varphi_{x}\,dx\,dt\quad\text{as ${\Delta x}\to 0$.}$ thus proving the first equality in (8.37). This concludes the proof of the lemma. ∎ ## 9\. Numerical examples We have tried the difference method presented here on several examples, and in doing this found that the first order method analyzed in this paper exhibits very slow convergence, and thus requires a very small mesh size ${\Delta x}$ to compute reasonable solutions. This is not surprising and appears to be the case with other schemes in the literature as well. Therefore we have implemented a second order extension of the method. This second order extension is based on the conservative version of the scheme (9.1) $D_{+}^{t}u_{j+1/2}^{n}+D_{-}\left[\left(u_{j+1/2}^{n}\vee 0\right)u_{j+1/2}^{n}+\left(u_{j+1/2}^{n}\wedge 0\right)u^{n}_{j+3/2}+P^{n}_{j+1}\right]\\\ =D_{-}\left(u_{j+1/2}^{n}\vee 0\right)u_{j-1/2}^{n}+D_{-}\left(u_{j+1/2}^{n}\wedge 0\right)u_{j+1/2}^{n},$ which can be viewed as a balance equation with a flux across $x=x_{j+1/2}$ given by $F_{j+1/2}^{n}=\left(u_{j+1/2}^{n}\vee 0\right)u_{j+1/2}^{n}+\left(u_{j+1/2}^{n}\wedge 0\right)u^{n}_{j+3/2}+P^{n}_{j+1}.$ Taking this viewpoint, we _define_ the second order finite volume scheme by (9.2) $D_{+}^{t}u_{j}^{n}+D_{-}F_{j+1/2}^{n+1/2}=u_{j}^{n}D_{-}\left(u^{n+1/2}_{j+1/2}\right).$ Here $u^{n+1/2}_{j+1/2}$ is a first order approximation of the value at the point $x=x_{j+1/2}$, $t=t^{n}+{\Delta t}/2$. This approximation is found by setting $u_{j+1/2}^{n}=\frac{1}{2}\left(u^{n}_{j}+u^{n}_{j+1}\right),$ and then using the scheme (3.1) for half a time step (i.e., ${\Delta t}/2$). This scheme is a formally second order accurate finite volume approximation, and this simple adaptation produces significantly more accurate approximations. In Figure 1 we show the approximations calculated by the first order scheme (3.1) and the second order scheme (9.2) for the single peakon example. In this case the exact solution reads $u(x,t)=e^{-\left|x-t\right|}.$ Figure 1 shows the solutions calculated using $2^{9}$ equally spaced grid points in the interval $[-10,30]$ for $t=20$. We see that the second order method is much more accurate than the first order method. Figure 1. Approximations using ${\Delta x}=40/2^{9}$, at $t=20$ to the single peakon. In passing, we note that we have not used the strict CFL-condition (4.1), but the more natural condition ${\Delta t}\leq\max_{j}\left\\{u^{n}_{j}\right\\}{\Delta x}.$ This holds for the second order scheme as well. In order to investigate the convergence properties of the two methods, we computed errors in $L^{1}$ for the two schemes. Table 1 shows the computed $L^{1}$ errors in the case of a single peakon under mesh refinement. In this context, the $L^{1}$ error is defined as $\text{$L^{1}$ error}={\Delta x}\sum_{i}\left|u_{{\Delta x}}(x_{i},t)-u(x_{i},t)\right|,$ where $u$ is the exact solution. We used $t=20$ and ${\Delta x}=40/2^{k}$ for $k=5,\ldots,12$. As expected, and as reported in [12], the first order method converges very slowly. $k$ | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 ---|---|---|---|---|---|---|---|---|--- $1^{\mathrm{st}}$ | 2.92 | 3.23 | 3.41 | 3.53 | 3.57 | 3.51 | 3.32 | 3.01 | 2.64 $2^{\mathrm{nd}}$ | 5.36 | 5.17 | 3.29 | 1.27 | 0.60 | 0.36 | 0.21 | 0.13 | 0.09 Table 1. $L^{1}$ errors for the single peakon case, at $t=20$, for $x\in[-10,30]$, ${\Delta x}=40/2^{k}$, $k=5,\ldots,13$ One other notable feature of Table 1 is that the second order method seems to converge at a rate slightly less than $1$. The two-peakon solution is considerably more complicated than the single peakon, and this is also a much harder challenge computationally, see e.g., [1] and [33]. We use the two-peakon solution given by (9.3) $u(x,t)=m_{1}(t)e^{-\left|x-x_{1}(t)\right|}+m_{2}(t)e^{-\left|x-x_{2}(t)\right|},$ with $\begin{gathered}x_{1}(t)=\log\left(\frac{18e^{t-10}}{e^{(t-10)/2}+6}\right),\quad x_{2}(t)=\log\left(40e^{t-10}+60e^{(t-10)/2}\right)\\\ m_{1}(t)=\frac{e^{(t-10)/2}+6}{2e^{(t-10)/2}+3},\quad m_{2}(t)=\frac{e^{(t-10)/2}+\frac{2}{3}}{e^{(t-10)/2}+3}.\end{gathered}$ These formulas were taken from [35]. Figure 2 shows a contour plot of the approximate solutions found by using the first and second order methods, and ${\Delta x}=40/2^{10}$ for $x\in[-15,25]$ and $t\in[0,25]$. | ---|--- Figure 2. Approximations to (9.3) using ${\Delta x}=40/2^{10}$. Left: first order method(3.1). Right : second order method (9.2). We see that the interaction between the two peakons is poorly represented by the first order method. Both the location as well as the magnitude of the peaks are far from the correct value. This is also illustrated Figure 3 where we show the approximations using ${\Delta x}=40/2^{8}$ at $t=25$. Figure 3. The approximations to (9.3) at $t=25$ and ${\Delta x}=40/2^{8}$. We have also calculated errors for the two-peakon case. Indeed for ${\Delta x}\geq 40/2^{12}$, the first order method did not seem to converge, and in order to give meaningful answers, this method demands very fine discretizations. These results are reported in Table 2. For ${\Delta x}>40/2^{8}$ none of the methods gave satisfactory results. $k$ | 8 | 9 | 10 | 11 | 12 | 13 ---|---|---|---|---|---|--- $1^{\text{st}}$ | 4.56 | 3.64 | 3.97 | 4.18 | 4.05 | 3.70 $2^{\text{nd}}$ | 1.88 | 1.04 | 0.63 | 0.38 | 0.22 | 0.16 Table 2. $L^{1}$ errors for the approximation to (9.2), $t=25$, $x\in[-15,25]$, ${\Delta x}=40/2^{k}$, $k=8,\ldots,13$. In our final example we choose initial data corresponding to a peakon- antipeakon collision: (9.4) $u_{0}(x)=-\tanh(6)\left(e^{-\left|x+y(6)\right|}-e^{-\left|x-y(6)\right|}\right),$ where $y(t)=\log(\cosh(t))$. In this case we have a “peakon anti-peakon collision” at $t=6$. In Figure 4 we exhibit the approximations generated by the first order (left) and the second order method for $t\in[0,10]$ and ${\Delta x}=24/2^{12}$. It is clear that the first order scheme generates the dissipative solution, and for $t$ larger than the collision time, the first order approximation vanishes. Regarding the second order approximation, it seems to continue as a peakon moving to the right, and an anti-peakon moving to the left. The magnitudes and speeds of these features are however far from the conservative solution, and we have indicated the conservative solution in the right hand figure. | ---|--- Figure 4. The numerical solutions to the initial value problem (9.4). Left: first order method, right: second order version. ## References * [1] R. Artebrant and H. J. Schroll. Numerical simulation of Camassa-Holm peakons by adaptive upwinding. Appl. Numer. Math., 56(5):695–711, 2006. * [2] R. Beals, D. H. Sattinger, and J. Szmigielski. Multipeakons and the classical moment problem. Adv. Math., 154(2):229–257, 2000. * [3] A. Bressan and A. Constantin. Global conservative solutions of the Camassa-Holm equation. Arch. Ration. Mech. Anal., 183(2):215–239, 2007. * [4] A. Bressan and A. Constantin. Global dissipative solutions of the Camassa-Holm equation. Anal. Appl. (Singap.), 5(1):1–27, 2007. * [5] A. Bressan and M. Fonte. An optimal transportation metric for solutions of the Camassa-Holm equation. Methods Appl. Anal., 12(2):191–219, 2005. * [6] R. Camassa. Characteristics and the initial value problem of a completely integrable shallow water equation. Discrete Contin. Dyn. Syst. Ser. B, 3(1):115–139, 2003. * [7] R. Camassa and D. D. Holm. An integrable shallow water equation with peaked solitons. Phys. Rev. Lett., 71(11):1661–1664, 1993. * [8] R. Camassa, D. D. Holm, and J. Hyman. A new integrable shallow water equation. Adv. Appl. Mech, 31:1–33, 1994. * [9] R. Camassa, J. Huang, and L. Lee. On a completely integrable numerical scheme for a nonlinear shallow-water wave equation. J. Nonlinear Math. Phys., 12(suppl. 1):146–162, 2005. * [10] R. Camassa, J. Huang, and L. Lee. Integral and integrable algorithms for a nonlinear shallow-water wave equation. J. Comput. Phys., 216(2):547–572, 2006. * [11] G. M. Coclite, H. Holden, and K. H. Karlsen. Global weak solutions to a generalized hyperelastic-rod wave equation. SIAM J. Math. Anal., 37(4):1044–1069 (electronic), 2005. * [12] G. M. Coclite, K. H. Karlsen, and N. H. Risebro. A convergent finite difference scheme for the Camassa-Holm equation with general $H^{1}$ initial data. Submitted, 2006. * [13] D. Cohen, B. Owren, and X. Raynaud. Multi-symplectic integration of the Camassa-Holm equation. Submitted, 2007. * [14] A. Constantin. On the scattering problem for the Camassa-Holm equation. R. Soc. Lond. Proc. Ser. A Math. Phys. Eng. Sci., 457(2008):953–970, 2001. * [15] A. Constantin and J. Escher. Global existence and blow-up for a shallow water equation. Ann. Scuola Norm. Sup. Pisa Cl. Sci. (4), 26(2):303–328, 1998. * [16] A. Constantin and J. Escher. Global weak solutions for a shallow water equation. Indiana Univ. Math. J., 47(4):1527–1545, 1998. * [17] A. Constantin and B. Kolev. On the geometric approach to the motion of inertial mechanical systems. J. Phys. A, 35(32):R51–R79, 2002. * [18] A. Constantin and B. Kolev. Geodesic flow on the diffeomorphism group of the circle. Comment. Math. Helv., 78(4):787–804, 2003. * [19] A. Constantin and H. P. McKean. A shallow water equation on the circle. Comm. Pure Appl. Math., 52(8):949–982, 1999. * [20] A. Constantin and L. Molinet. Global weak solutions for a shallow water equation. Comm. Math. Phys., 211(1):45–61, 2000. * [21] H. H. Dai. Model equations for nonlinear dispersive waves in a compressible mooney–rivlin rod. Acta Mechanica, 127:193–207, 1998. * [22] R. Danchin. A few remarks on the Camassa-Holm equation. Differential Integral Equations, 14(8):953–988, 2001. * [23] R. Danchin. A note on well-posedness for Camassa-Holm equation. J. Differential Equations, 192(2):429–444, 2003. * [24] E. Feireisl. Dynamics of viscous compressible fluids, volume 26 of Oxford Lecture Series in Mathematics and its Applications. Oxford University Press, Oxford, 2004. * [25] O. B. Fringer and D. D. Holm. Integrable vs. nonintegrable geodesic soliton behavior. Phys. D, 150(3-4):237–263, 2001. * [26] B. Fuchssteiner and A. S. Fokas. Symplectic structures, their Bäcklund transformations and hereditary symmetries. Phys. D, 4(1):47–66, 1981/82. * [27] H. Holden and X. Raynaud. Convergence of a finite difference scheme for the Camassa-Holm equation. SIAM J. Num. Anal., 44(4):1655—1680, 2006. * [28] H. Holden and X. Raynaud. A convergent numerical scheme for the Camassa-Holm equation based on multipeakons. Discrete Contin. Dyn. Syst., 14(3):505–523, 2006. * [29] H. Holden and X. Raynaud. Global conservative solutions of the Camassa-Holm equation - a Lagrangian point of view. Comm. Partial Differential Equations, to appear. * [30] D. D. Holm, J. E. Marsden, and T. S. Ratiu. The Euler-Poincaré equations and semidirect products with applications to continuum theories. Adv. Math., 137(1):1–81, 1998. * [31] D. D. Holm and M. F. Staley. Wave structure and nonlinear balances in a family of evolutionary PDEs. SIAM J. Appl. Dyn. Syst., 2(3):323–380 (electronic), 2003. * [32] R. S. Johnson. Camassa-Holm, Korteweg-de Vries and related models for water waves. J. Fluid Mech., 455:63–82, 2002. * [33] H. Kalisch and X. Raynaud. Convergence of a spectral projection of the Camassa-Holm equation. Numer. Methods Partial Differential Equations, 22(5):1197–1215, 2006. * [34] Y. A. Li and P. J. Olver. Well-posedness and blow-up solutions for an integrable nonlinearly dispersive model wave equation. J. Differential Equations, 162(1):27–63, 2000. * [35] H. Lundmark. Formation and dynamics of shock waves in the Degasperis-Procesi equation. J. Nonlinear Sci., 17(3):169–198, 2007. * [36] G. Misiołek. A shallow water equation as a geodesic flow on the Bott-Virasoro group. J. Geom. Phys., 24(3):203–208, 1998. * [37] G. Rodríguez-Blanco. On the Cauchy problem for the Camassa-Holm equation. Nonlinear Anal., 46(3, Ser. A: Theory Methods):309–327, 2001. * [38] J. Smoller. Shock waves and reaction-diffusion equations, volume 258 of Grundlehren der Mathematischen Wissenschaften. Springer-Verlag, New York, second edition, 1994. * [39] Z. Xin and P. Zhang. On the weak solutions to a shallow water equation. Comm. Pure Appl. Math., 53(11):1411–1433, 2000. * [40] Z. Xin and P. Zhang. On the uniqueness and large time behavior of the weak solutions to a shallow water equation. Comm. Partial Differential Equations, 27(9-10):1815–1844, 2002\. * [41] Y. Xu and C.-W. Shu. A local discontinuous Galerkin method for the Camassa-Holm equation. Siam J. Numer. Anal., to appear.

arxiv-papers

2024-09-04T02:48:53.924992

0802.3183

# Violation of the Cauchy-Schwarz Inequality in the Macroscopic Regime A. M. Marino V. Boyer P. D. Lett Joint Quantum Institute, National Institute of Standards and Technology, and University of Maryland, Gaithersburg, Maryland 20899, USA ###### Abstract We have observed a violation of the Cauchy-Schwarz inequality in the macroscopic regime by more than 8 standard deviations. The violation has been obtained while filtering out only the low frequency noise of the quantum- correlated beams that results from the technical noise of the laser used to generate them. We use bright intensity-difference squeezed beams produced by four-wave mixing as the source of the correlated fields. We also demonstrate that squeezing does not necessarily imply a violation of the Cauchy-Schwarz inequality. ###### pacs: 42.50.Xa, 42.50.Dv, 42.65.Yj The comparison between the predictions of quantum and classical theories has been a subject of study since the development of quantum mechanics. To that end, a number of different classical inequalities have been developed that provide an experimental discrimination between these theories Loudon (1980); Reid and Walls (1986). Experiments showing a violation of these classical inequalities have verified quantum theory. However, to date, most of these experiments have been carried out in the regime in which single particles are detected one at a time. It is thus interesting to study whether or not the quantum signature given by these tests is still present in the limit in which the system under study becomes macroscopic. Among the inequalities that offer a test between quantum and classical theories is the Cauchy-Schwarz inequality (CSI) Loudon (1980); Reid and Walls (1986). The first observation of a violation of this inequality was obtained by Clauser using an atomic two-photon cascade system Clauser (1974). More recently, large violations using four-wave mixing have been obtained Kolchin et al. (2006); Thompson et al. (2006), still in the photon-counting regime. For bright fields the natural approach for analyzing their quantum nature is through noise measurements. In this case the boundary between quantum and classical is taken to be the noise of a coherent state, or standard quantum limit (SQL), such that having a field with less noise than the SQL (squeezed light) is considered non-classical. However, the presence of squeezing does not provide a direct discrimination between quantum and classical theories since the SQL is a result of quantum theory Reid and Walls (1986). The possibility of using a macroscopic quantum state to violate the CSI has been previously analyzed McNeil and Gardiner (1983); Gerry and Grobe (1995); An and Duc (2002); Olsen et al. (2003). To date, however, only a few experiments have probed this macroscopic regime. Recently anti-bunching of a small number of photons was observed in the continuous-variable regime Grosse et al. (2007). In addition, a frequency analysis has been used to infer a violation of the CSI over limited frequency ranges Li et al. (2000). In this Letter we present the first observation, to our knowledge, of a direct violation of the two-beam Cauchy-Schwarz inequality in the limit of a macroscopic quantum state. We show that the quantum-correlated fluctuations between two different modes of the electromagnetic field are responsible for the violation of the CSI. In addition to having a bright coherent carrier, we work in the high gain regime in which the mean number of spontaneous correlated photons within the inverse of the bandwidth (correlation time) of the process is much larger than one. Thus photon counting is not an option and continuous variable detection schemes need to be used. The CSI for the degree of second-order coherence, $g^{(2)}$, for two distinct fields, $a$ and $b$, is of the form Loudon (2000) $[g^{(2)}_{ab}(\tau)]^{2}\leq g^{(2)}_{aa}(0)g^{(2)}_{bb}(0),$ (1) where $g^{(2)}$ is the normalized intensity correlation function. This inequality indicates that for a classical system the cross-correlation between two fields, $g^{(2)}_{ab}$, cannot be larger than the geometric mean of the zero-time auto-correlations, $g^{(2)}_{aa}$ and $g^{(2)}_{bb}$. According to quantum theory, however, it is possible to violate this inequality. In this case the correlation function is defined in terms of normally ordered operators $g^{(2)}_{ab}(\tau)=\frac{\langle\hat{a}^{{\dagger}}(t)\hat{b}^{{\dagger}}(t+\tau)\hat{b}(t+\tau)\hat{a}(t)\rangle}{\langle\hat{a}^{{\dagger}}(t)\hat{a}(t)\rangle\langle\hat{b}^{{\dagger}}(t)\hat{b}(t)\rangle},$ (2) where $\hat{a}$ and $\hat{b}$ are the photon annihilation operators for the two fields. A violation of the CSI indicates the presence of non-classical correlations between the fields. Most of the experiments to date have been done in the photon-counting regime, in which the separation between photon pairs is much larger than the correlation time between photons. This makes it possible to obtain a large cross-correlation while the zero time auto-correlation functions are in principle equal to 2, giving as a result a large violation of the CSI Loudon (2000). In contrast, in the large gain regime all of the correlation functions tend to the same value (they are equal or larger than one), making it harder to observe a violation of the CSI. Figure 1: (color online). Experimental setup. A 4WM process is used to generate quantum-correlated bright beams. PBS = polarizing beam splitter, BS = 50/50 beam splitter. We use a seeded four-wave mixing (4WM) process in a double-$\Lambda$ system in rubidium vapor, as described in McCormick et al. (2007a, b), as our source of bright correlated beams. Four-wave mixing is a parametric process, such that the initial and final states of the atomic system are the same. This leads to the emission of probe and conjugate photons in pairs and thus to intensity correlations between the two fields which are stronger than any correlations possible between classical optical fields. The configuration and experimental parameters for the 4WM are the same as the ones described in Ref. McCormick et al. (2007b). A single Ti:Sapphire laser and an acousto-optic modulator are used to generate a bright pump and a weak probe which are resonant with a two-photon Raman transition between the $F=2$ and $F=3$ electronic ground states of 85Rb. The pump laser is tuned 800 MHz to the blue of the D1 line at 795 nm while the probe is downshifted in frequency by 3 GHz. The two beams are then mixed at a small angle in a pure 85Rb vapor cell, as shown in Fig. 1. In our double-$\Lambda$ configuration, the 4WM converts two photons from the pump into one probe photon and one conjugate photon (upshifted by 3 GHz with respect to the pump). We have measured up to 8 dB of intensity-difference squeezing at 1 MHz with this method. After the vapor cell we separate the probe and conjugate from the pump beam with a polarizer with $\approx 10^{5}:1$ extinction ratio for the pump.. We then use beamsplitters to split the probe and conjugate, each into two beams of equal power, and detect the resulting four beams with separate photodiodes, as shown in Fig. 1. This setup directly measures the normally ordered correlation function defined in Eq. (2), as described in Ref. Reid and Walls (1986). After each photodiode a bias-T is used to separate the DC part of the photocurrent, which is recorded and then used to normalize the correlation functions. The rest of the signal is amplified, digitized with a resolution of 9 bits, and recorded on a computer. The amplified time traces are sampled at a rate of 1 GS/s and 500 sets of traces, each with 10,000 points, are recorded. This setup allows us to simultaneously obtain all the information needed to calculate the correlation functions and the noise power spectra of the different beams. The bright correlated beams that are obtained from the seeded 4WM process consist of a large coherent part plus quantum-correlated fluctuations. The large coherent part makes the $g^{(2)}$ functions tend to 1, the value for a coherent state, as its intensity increases. It is thus useful to separate the correlation functions into contributions for the coherent part of the field and the fluctuations, that is $g^{(2)}_{ab}=1+\epsilon_{ab}$. Since the quantum correlations between the fields are in the fluctuations, we can rewrite the CSI in terms of the fluctuation terms such that it takes the form $\epsilon_{ab}\leq\frac{\epsilon_{aa}+\epsilon_{bb}}{2},$ (3) where we have kept only terms to first order in $\epsilon$. We define a violation factor $V\equiv\frac{\epsilon_{aa}+\epsilon_{bb}}{2\epsilon_{ab}}$ (4) such that $V<1$ indicates a violation of the CSI. In the ideal case, the 4WM process can be described by the two-photon squeeze operator $\hat{S}_{ab}=\exp(s\hat{a}\hat{b}-s\hat{a}^{{\dagger}}\hat{b}^{{\dagger}})$, where $s$ is the squeezing parameter ($s>0$). The bright quantum-correlated beams are obtained by applying this operator to an input coherent state, $|\alpha\rangle$, for the probe and the vacuum for the conjugate. In the limit in which the number of photons in the probe seed is much larger than one ($|\alpha|\gg 1$) $V$ takes the form $V=1-\frac{1}{2G},$ (5) where $G=\cosh^{2}s$ is the gain of the process and we have taken the single frequency approximation for each beam. For an ideal seeded 4WM process $V$ is always less than one, so that a violation of the CSI should always be obtained. In general, however, the presence of squeezing does not guaranty a violation of the CSI. As Eq. (5) shows, the amount of violation is inversely proportional to the gain. This is in contrast to the amount of intensity- difference squeezing that is expected from a seeded 4WM process, for which the noise scales as $1/(2G-1)$. Thus, a large amount of squeezing does not imply a large violation of the CSI, as has been pointed out in Ref. Carmichael et al. (2000). Figure 2: (color online). Correlation functions of the fluctuations for the probe (dotted), conjugate (dashed), and cross $g^{(2)}$ (solid). The inset shows an expanded view of the peaks of the correlation functions. The horizontal dashed line shows the mean value of the zero time auto-correlation functions for the probe and the conjugate. The indicated uncertainties are discussed in the text. A typical set of correlation functions that shows a violation of the CSI is shown in Fig. 2. Here the horizontal dashed line indicates the mean value of the zero-time auto-correlation functions of the fluctuations for the probe and the conjugate, such that a cross-correlation larger than this level indicates a violation of the CSI. A violation of the CSI can clearly be seen in the inset of Fig. 2. In obtaining these correlation functions we have only filtered out the low frequency technical noise below 500 kHz. The bandwidth of the detection system ($>$ 40 MHz) is larger than the bandwidth of the quantum correlations. The uncertainties indicated in Fig. 2 are obtained by directly calculating the correlation functions and obtaining the standard deviation over the 500 sets of traces. These uncertainties are not statistically independent since the probe and conjugate contain classical fluctuations that are strongly correlated as a result of slow intensity fluctuations of the pump and probe seed beams between data sets. This leads to a violation of the CSI that is more significant than what can be inferred from the inset. An accurate measure of the uncertainty of $V$ is obtained by calculating $V$ for each set of traces and using these results to derive the standard deviation of $V$ over the 500 sets. For the results shown in Fig. 2 the gain of the process is around 10 and $V=0.987\pm 1.4\times 10^{-3}$, giving a violation of the CSI by more than 8 standard deviations. The cross-correlation function shows a delay in the arrival time between probe and conjugate fluctuations; for the case shown in Fig. 2 the delay is around 8 ns. The delay results from the combination of 4WM in the double-$\Lambda$ system and propagation through the vapor cell van der Wal et al. (2003); Boyer et al. (2007). An important property of the double-$\Lambda$ system is that the relative delay between probe and conjugate for fluctuations of different frequencies is almost fixed. Such a fixed delay only causes the cross- correlation to be shifted in time and will not have an effect on $V$. In contrast, any large spread in the delay between different frequencies (dispersion) would make the cross-correlation peak wider and reduce its maximum value, degrading the amount of violation. The dips on the correlation functions are due to an offset of the carrier frequency with respect to the gain peak of the process. These effects will be examined in detail elsewhere. One of the difficulties in obtaining a violation of the CSI is that any source of excess uncorrelated noise will decrease the violation. In order to see why this is the case, we need to consider the noise power spectra of the different beams. We can rewrite the CSI in terms of the noise power spectra for the probe ($S_{p}$), conjugate ($S_{c}$), and intensity-difference ($S_{\rm{diff}}$) such that $\displaystyle\int{d\Omega\left(\frac{S_{\rm{diff}}(\Omega)}{\langle\hat{n}_{p}\rangle+\langle\hat{n}_{c}\rangle}-1\right)}$ $\displaystyle\geq\frac{\langle\hat{n}_{p}\rangle-\langle\hat{n}_{c}\rangle}{\langle\hat{n}_{p}\rangle+\langle\hat{n}_{c}\rangle}\int{d\Omega\left[\left(\frac{S_{p}(\Omega)}{\langle\hat{n}_{p}\rangle}-1\right)-\left(\frac{S_{c}(\Omega)}{\langle\hat{n}_{c}\rangle}-1\right)\right]}.$ (6) The terms in parenthesis represent the excess noise (or noise reduction) with respect to the corresponding SQL. For the ideal seeded 4WM process the normalized noise power spectra for the probe and the conjugate are equal, so that the term in square brackets is zero, and $\langle\hat{n}_{p}\rangle>\langle\hat{n}_{c}\rangle$. The presence of squeezing in the intensity difference can make the integral on the left hand side negative, leading to a violation of the CSI. Excess noise can have an impact on the violation in two different ways. The presence of excess uncorrelated noise on either beam can lead the intensity-difference noise to go above the SQL for some frequency ranges such that the integral on the left hand side can become positive. In addition, excess noise on the conjugate can make the right hand side of the inequality negative enough (given that $\langle\hat{n}_{p}\rangle>\langle\hat{n}_{c}\rangle$) so that even if squeezing is present a violation might not be obtained. Figure 3: (color online). Normalized noise power spectra for the probe ($S_{p}$), conjugate ($S_{c}$), and intensity-difference ($S_{\rm{diff}}$) for a gain of (a) 10 and (b) 2. All the spectra are normalized to their respective SQL, represented by the dashed line. For the results shown in Fig. 2, the corresponding normalized noise power spectra are shown in Fig. 3(a). All the noise power spectra are calculated by taking the FFT of the time traces and averaging over the 500 sets. The SQL for the probe and conjugates is calculated by taking the difference of the corresponding photocurrents while the one for the intensity-difference noise is given by the sum of the SQLs for the probe and conjugate. As is expected for a 4WM process both the probe and the conjugate have excess noise with respect to the SQL and their spectra are almost the same. The measured intensity-difference squeezing has a bandwidth of 15 MHz, consistent with the gain bandwidth of the 4WM process Boyer et al. (2007), with a maximum squeezing of 6 dB. For this case the system acts almost as an ideal 4WM medium which makes it possible to observe a violation of the CSI. The amount of squeezing that is measured is limited by a total detection efficiency, including optical path transmission and photodiode efficiencies, of $(80\pm 3)\%$. We have verified that $g^{(2)}$ is not affected by loss so that any source of loss will not have an impact on the violation of the CSI. When the gain of the process is reduced to 2, we find a situation in which the noise power spectra of the probe and the conjugate are noticeably different, as shown in Fig. 3(b). This difference in noise leads to a reduction of the intensity-difference squeezing bandwidth from 15 MHz to 7 MHz and a small amount of excess noise at higher frequencies. For this particular case we find that the small amount of excess noise is enough to prevent a violation of the CSI, such that $V=1.075\pm 3.3\times 10^{-3}$, even though there is more than 4 dB of squeezing at low frequencies. The relative delay between the probe and the conjugate (8 ns for $G=10$ and 13 ns for $G=2$) has been compensated when calculating the intensity-difference noise power spectra shown in Fig. 3. This makes it possible to see the real squeezing bandwidth that results from the 4WM process in Fig. 3(a). While the relative delay has no effect on the violation of the CSI, it introduces a frequency dependent phase shift such that the intensity-difference noise power spectrum oscillates between intensity-difference and intensity-sum noise levels Machida and Yamamoto (1989). Figure 4: (color online). Effect of frequency filtering on the violation of the CSI. Violation parameter ($V$) as a function of high-frequency cutoff for $G=2$, $G=5$, $G=8$, and $G=10$. $V<1$ indicates a violation of the CSI. The size of the squares represent the statistical uncertainties. The effect of the excess noise can be further analyzed by filtering out the high frequencies, where most of the uncorrelated excess noise is present. The filtering is done on the digitized traces by applying a 10th order Butterworth bandpass filter with a low-frequency cutoff of 500 kHz that filters out the technical noise of the laser and a variable high-frequency cutoff. We have done this analysis for a number of different gains, as shown in Fig. 4. The gain is changed by modifying the temperature of the cell and thus the atomic number density. If we look at the lowest high-frequency cutoff points in Fig. 4, we see that the violation follows the trend given by Eq. (5) for an ideal 4WM process, that is, the violation gets better with smaller gains. However, once we increase the high-frequency cutoff, $V$ starts to degrade, with lower gains degrading faster. Increasing the high-frequency cutoff takes into account higher frequencies of the noise power spectrum that correspond to different regions of the gain profile. This leads to competition with other processes, such as Raman gain on the conjugate, that add excess noise. Except for the case $G=2$ a violation of the CSI is obtained for the different gains shown in Fig. 4 when only the low frequency technical noise of the laser is filtered. Even for the case in which the system contains excess uncorrelated noise, a violation of the CSI can be recovered with enough filtering, as shown for the case of $G=2$. This approaches a spectral analysis of the noise, as is regularly done when measuring bright beams. If we compare the case of $G=2$ in Fig. 3 and Fig. 4 we find that the violation is lost at a high-frequency cutoff around 6 MHz while the squeezing is present over a larger frequency range than the filtering bandwidth used to calculate V, up to around 7 MHz, once the relative delay between probe and conjugate has been compensated. This gives a region in which squeezing is present but not a violation of the CSI. The amount of excess noise on the conjugate is enough to destroy the violation but not the squeezing. In conclusion, we have observed a violation of the CSI in the macroscopic regime. The necessary information to observe the violation is contained in the quantum-correlated fluctuations of the field. We have shown that the presence of excess uncorrelated noise can prevent the observation of a violation of the CSI. The ability to obtain a violation of the CSI shows that the 4WM process used here provides a low-noise source of quantum correlated bright beams over a large frequency range. Finally, we have shown that the presence of squeezing does not necessarily imply a violation of the CSI. ## References * Loudon (1980) R. Loudon, Rep. Prog. Phys. 43, 913 (1980). * Reid and Walls (1986) M. D. Reid and D. F. Walls, Phys. Rev. A 34, 1260 (1986). * Clauser (1974) J. F. Clauser, Phys. Rev. D 9, 853 (1974). * Kolchin et al. (2006) P. Kolchin et al., Phys. Rev. Lett. 97, 113602 (2006). * Thompson et al. (2006) J. K. Thompson et al., Science 313, 74 (2006). * McNeil and Gardiner (1983) K. J. McNeil and C. W. Gardiner, Phys. Rev. A 28, 1560 (1983). * Gerry and Grobe (1995) C. C. Gerry and R. Grobe, Phys. Rev. A 51, 1698 (1995). * An and Duc (2002) N. B. An and T. M. Duc, J. Opt. B 4, 289 (2002). * Olsen et al. (2003) M. K. Olsen, L. I. Plimak, and A. Z. Khoury, Opt. Commun. 215, 101 (2003). * Grosse et al. (2007) N. B. Grosse et al., Phys. Rev. Lett. 98, 153603 (2007). * Li et al. (2000) Y. Q. Li et al., J. Opt. B 2, 292 (2000). * Loudon (2000) R. Loudon, _The Quantum Theory of Light_ (Oxford Academic Press, Oxford, 2000). * McCormick et al. (2007a) C. F. McCormick et al., Opt. Lett. 32, 178 (2007a). * McCormick et al. (2007b) C. F. McCormick et al., arXiv:quant-ph/0703111 (2007b). * Carmichael et al. (2000) H. J. Carmichael et al., Phys. Rev. Lett. 85, 1855 (2000). * van der Wal et al. (2003) C. H. van der Wal et al., Science 301, 196 (2003). * Boyer et al. (2007) V. Boyer et al., Phys. Rev. Lett. 99, 143601 (2007). * Machida and Yamamoto (1989) S. Machida and Y. Yamamoto, Opt. Lett. 14, 1045 (1989).

arxiv-papers

2024-09-04T02:48:53.937126

0802.3234

# The dynamics of human body weight change Carson C. Chow111corresponding author: email:carsonc@mail.nih.gov and Kevin D. Hall Laboratory of Biological Modeling, NIDDK. NIH, Bethesda, MD 20892 ###### Abstract An imbalance between energy intake and energy expenditure will lead to a change in body weight (mass) and body composition (fat and lean masses). A quantitative understanding of the processes involved, which currently remains lacking, will be useful in determining the etiology and treatment of obesity and other conditions resulting from prolonged energy imbalance. Here, we show that the long-term dynamics of human weight change can be captured by a mathematical model of the macronutrient flux balances and all previous models are special cases of this model. We show that the generic dynamical behavior of body composition for a clamped diet can be divided into two classes. In the first class, the body composition and mass are determined uniquely. In the second class, the body composition can exist at an infinite number of possible states. Surprisingly, perturbations of dietary energy intake or energy expenditure can give identical responses in both model classes and existing data are insufficient to distinguish between these two possibilities. However, this distinction is important for the efficacy of clinical interventions that alter body composition and mass. Summary Understanding the dynamics of human body weight change has important consequences for conditions such as obesity, cancer, AIDS, anorexia and bulimia nervosa. While we know that changes of body weight result from imbalances between the energy derived from food and the energy expended to maintain life and perform physical work, quantifying this relationship has proved difficult. Part of the difficulty stems from the fact that the body is comprised of multiple components and we must quantify how weight change is reflected in terms of alterations of body composition (i.e. fat versus lean mass). Mathematical modeling is a means to address this issue. Here we show that a mathematical model of the flux balances of macronutrients, namely fat, protein and carbohydrates, can provide a general description of the way the body weight will change over time. For a fixed food intake rate and physical activity level, the body weight and body composition will approach a steady state. However, the steady state can correspond to a unique body weight or a continuum of body weights depending on how fat oxidation depends on the body weight and composition changes. Interestingly, the existing experimental data on human body weight dynamics cannot presently distinguish between these two possibilities. We propose experiments that could determine if steady state body weight is unique and use computer simulations to demonstrate how such experiments could be performed. ## I Introduction Obesity, anorexia nervosa, cachexia, and starvation are conditions that have a profound medical, social and economic impact on our lives. For example, the incidence of obesity and its co-morbidities has increased at a rapid rate over the past two decades hill2003 ; hill2006 . These conditions are characterized by changes in body weight (mass) that arise from an imbalance between the energy derived from food and the energy expended to maintain life and perform work. However, the underlying mechanisms of how changes in energy balance lead to changes in body mass and body composition are not well understood. In particular, it is of interest to understand how body composition is apportioned between fat and lean components when the body mass changes and if this energy partitioning can be altered. Such an understanding would be useful in optimizing weight loss treatments in obese subjects to maximize fat loss or weight gain treatments for anorexia nervosa and cachexia patients to maximize lean tissue gain. To address these issues and improve our understanding of human body weight regulation, mathematical and computational modeling has been attempted many times over the past several decades hall2006 ; hall2002 ; alpert1979 ; alpert1990 ; alpert2005 ; Forbes1970 ; Livingston2001 ; hbc ; jordan ; antonetti1973 ; payne1977b ; Girardier1994 ; westerterp1995 ; kozusko2001 ; flatt2004 ; christiansen2005 ; song2007 . Here we show how models of body composition and mass change can be understood and analyzed within the realm of dynamical systems theory and can be classified according to their geometric structure in the two dimensional phase plane. We begin by considering a general class of macronutrient flux balance equations and progressively introduce assumptions that constrain the model dynamics. We show that two compartment models of fat and lean masses can be categorized into two generic classes. In the first class, there is a unique body composition and mass (i.e. stable fixed point) that is specified by the diet and energy expenditure. In the second class, there is a continuous curve of fixed points (i.e. invariant manifold) with an infinite number of possible body compositions and masses at steady state for the same diet and energy expenditure. We show that almost all of the models in the literature are in the second class. Surprisingly, the existing data are insufficient to determine which of the two classes pertains to humans. For models with an invariant manifold, we show that an equivalent one dimensional equation for body composition change can be derived. We give numerical examples and discuss possible experimental approaches that may distinguish between the classes. ## II Results ### II.1 General model of macronutrient and energy flux balance The human body obeys the law of energy conservation atwater , which can be expressed as $\Delta U=\Delta Q-\Delta W$ (1) where $\Delta U$ is the change in stored energy in the body, $\Delta Q$ is a change in energy input or intake, and $\Delta W$ is a change in energy output or expenditure. The intake is provided by the energy content of the food consumed. Combustion of dietary macronutrients yields chemical energy and Hess’s law states that the energy released is the same regardless of whether the process takes place inside a bomb calorimeter or via the complex process of oxidative phosphorylation in the mitochondria. Thus, the energy that can be derived from combustion of food in the body can be precisely measured in the laboratory. However, there is an important caveat. Not all macronutrients in food are completely absorbed by the body. Furthermore, the dietary protein that is absorbed does not undergo complete combustion in the body, but rather produces urea and ammonia. In accounting for these effects, we refer to the metabolizable energy content of dietary carbohydrate, fat, and protein which are slightly less than the values obtained by bomb calorimetry. The energy expenditure includes the work to maintain basic metabolic function (resting metabolic rate), to digest, absorb and transport the nutrients in food (thermic effect of feeding), to synthesize or break down tissue, and to perform physical activity, together with the heat generated. The energy is stored in the form of fat as well as in lean body tissue such as glycogen and protein. The body need not be in equilibrium for Eq. (1) to hold. While we are primarily concerned with adult weight change, Eq. (1) is also valid for childhood growth. In order to express a change of stored energy $\Delta U$ in terms of body mass $M$ we must determine the energy content per unit body mass change, i.e. the energy density $\rho_{M}$. We can then set $\Delta U=\Delta(\rho_{M}M)$. To model the dynamics of body mass change, we divide Eq. (1) by some interval of time and take the limit of infinitesimal change to obtain a one dimensional energy flux balance equation: $\frac{d}{dt}(\rho_{M}M)=I-E$ (2) where $I=dQ/dt$ is the rate of metabolizable energy intake and $E=dW/dt$ is the rate of energy expenditure. It is important to note that $\rho_{M}$ is the energy density of body mass change, which need not be a constant but could be a function of body composition and time. Thus, in order to use Eq. (2), the dynamics of $\rho_{M}$ must also be established. When the body changes mass, that change will be composed of water, protein, carbohydrates (in the form of glycogen), fat, bone, and trace amounts of micronutrients, all having their own energy densities. Hence, a means of determining the dynamics of $\rho_{M}$ is to track the dynamics of the components. The extracellular water and bone mineral mass have no metabolizable energy content and change little when body mass changes in adults under normal conditions keys1950 . The change in intracellular water can be specified by changes in the tissue protein and glycogen. Thus the main components contributing to the dynamics of $\rho_{M}$ are the macronutrients - protein, carbohydrates, and fat, where we distinguish body fat (e.g. free fatty acids and triglycerides) from adipose tissue, which includes water and protein in addition to triglycerides. We then represent Eq. (2) in terms of macronutrient flux balance equations for body fat $F$, glycogen $G$, and protein $P$: $\displaystyle\rho_{F}\frac{dF}{dt}$ $\displaystyle=$ $\displaystyle I_{F}-f_{F}E$ (3) $\displaystyle\rho_{G}\frac{dG}{dt}$ $\displaystyle=$ $\displaystyle I_{C}-f_{C}E$ (4) $\displaystyle\rho_{P}\frac{dP}{dt}$ $\displaystyle=$ $\displaystyle I_{P}-(1-f_{F}-f_{C})E$ (5) where $\rho_{F}=39.5$ MJ/kg, $\rho_{G}=17.6$ MJ/kg, $\rho_{P}=19.7$ MJ/kg are the energy densities, $I_{F},I_{C},I_{P}$ are the intake rates, and $f_{F}$, $f_{C}$, $1-f_{F}-f_{C}$ are the fractions of the energy expenditure rate obtained from the combustion of fat, carbohydrates (glycogen) and protein respectively. The fractions and energy expenditure rate are functions of body composition and intake rates. They can be estimated from indirect calorimetry, which measures the oxygen consumed and carbon dioxide produced by a subject atwater1905 . The intake rates are determined by the macronutrient composition of the consumed food, the efficiency of the conversion of the food into utilizable form and the thermic effect of feeding. Transfer between compartments such as de novo lipogenesis where carbohydrates are converted to fat or gluconeogenesis where amino acids are converted into carbohydrates can be accounted for in the forms of $f_{F}$ and $f_{C}$. The sum of Eqs. (3), (4) and (5) recovers the energy flux balance equation (2), where the body mass $M$ is the sum of the macronutrients $F$, $G$, $P$, with the associated intracellular water, and the inert mass that does not change such as the extracellular water, bones, and minerals, and $\rho_{M}=(\rho_{F}F+\rho_{G}G+\rho_{P}P)/M$. The intake and energy expenditure rates are explicit functions of time with fast fluctuations on a time scale of hours to days periwal2006 . However, we are interested in the long term dynamics over weeks, months and years. Hence, to simplify the equations, we can use the method of averaging to remove the fast motion and derive a system of equations for the slow time dynamics. We do this explicitly in the Methods section and show that the form of the averaged equations to lowest order are identical to (3)-(5) except that the three components are to be interpreted as the slowly varying part and the intake and energy expenditure rates are moving time averages over a time scale of a day. The three compartment flux balance model was used by Hall hall2006 to numerically simulate data from the classic Minnesota human starvation experiment keys1950 . In Hall’s model, the forms of the energy expenditure and fractions were chosen for physiological considerations. For clamped food intake, the body composition approached a unique steady state. The model also showed that apart from transient changes lasting only a few days, carbohydrate balance is precisely maintained as a result of the limited storage capacity for glycogen. We will exploit this property to reduce the three dimensional system to an approximately equivalent two dimensional system where dynamical systems techniques can be employed to analyze the system dynamics. ### II.2 Reduced models #### II.2.1 Two compartment macronutrient partition model The three compartment macronutrient flux balance model Eqs. (3)-(5) can be reduced to a two dimensional system for fat mass $F$ and lean mass $L=M-F$, where $M$ is the total body mass. The lean mass includes the protein and glycogen with the associated intracellular water along with the mass that does not change appreciably such as the extracellular water and bone. Hence the rate of change in lean mass is given by $\frac{dL}{dt}=(1+h_{P})\frac{dP}{dt}+(1+h_{G})\frac{dG}{dt}$ (6) where $h_{P}=1.6$ and $h_{G}=2.7$ are reasonable estimates of the hydration coefficients for the intracellular water associated with the protein and glycogen respectively hall2006 ; hall2007 . (We note that fat is not associated with any water.) The glycogen storage capacity is extremely small compared to the fat and protein compartments. Thus the slow component of glycogen can be considered to be a constant (see Methods). In other words, on time scales much longer than a day, which are of interest for body weight change, we can consider glycogen to be in quasi-equilibrium so that $dG/dt=0$, as observed in numerical simulations hall2006 . This implies that $f_{C}=I_{C}/E$, which can be substituted into Eq. (5) to give $\rho_{P}\frac{dP}{dt}=I_{P}+I_{C}-(1-f_{F})E$ (7) Substituting Eq. (7) and $dG/dt=0$ into Eq. (6) leads to the two compartment macronutrient partition model $\displaystyle\rho_{F}\frac{dF}{dt}$ $\displaystyle=$ $\displaystyle I_{F}-fE$ (8) $\displaystyle\rho_{L}\frac{dL}{dt}$ $\displaystyle=$ $\displaystyle I_{L}-(1-f)E$ (9) where $\rho_{L}=\rho_{P}/(1+h_{P})=7.6$ MJ/kg, $I_{F}$ and $I_{L}=I_{P}+I_{C}$ are the intake rates into the fat and lean compartments respectively, $E=E(I_{F},I_{L},F,L)$ is the total energy expenditure rate, and $f=f(I_{F},I_{L},F,L)\equiv f_{F}$ is the fraction of energy expenditure rate attributed to fat utilization. We note that $dG/dt=0$ may be violated if the glycogen content is proportional to the protein content, which is plausible because the glycogen mass is stored in muscle tissue and may scale with protein mass. We show that this assumption leads to the same two dimensional system. Substituting $\frac{dG}{dt}=k\frac{dP}{dt}$ (10) for a proportionality constant $k$, into Eq. (4) gives $f_{C}E=I_{C}-\rho_{C}kdP/dt$ which inserted into Eq. (5) leads to $(\rho_{P}+k\rho_{C})\frac{dP}{dt}=I_{L}-(1-f)E$ (11) Substituting Eqs. (10) and (11) into Eq. (6) will again result in Eq. (9) but with $\rho_{L}=(\rho_{P}+k\rho_{C})/((1+h_{p}+k+kh_{G})$. For $k=0.044<<1$ as suggested by Snyder et al. snyder1984 , $\rho_{L}$ has approximately the same value as before. Previous studies have considered two dimensional models of body mass change although they were not derived from the three dimensional macronutrient partition model. Alpert alpert1979 ; alpert1990 ; alpert2005 considered a model with $E$ linearized in $F$ and $L$ and different $f$ depending on context. Forbes Forbes1970 and Livingston et al. Livingston2001 modeled weight loss as a double exponential decay. Although, they did not consider macronutrient flux balance, the dynamics of their models are equivalent to the two dimensional model with $I_{F}$ and $I_{L}$ zero, and $E$ linear in $F$ and $L$. #### II.2.2 Energy partition model The two compartment macronutrient partition model can be further simplified by assuming that trajectories in the $L$ \- $F$ phase plane follow prescribed paths satisfying $\frac{\rho_{F}}{\rho_{L}}\frac{dF}{dL}=\alpha(F,L)$ (12) where $\alpha(F,L)$ is a continuous function Forbes1987 ; hbc ; jordan that depends on the mechanisms of body weight change. This stringent constraint was first hypothesized by Forbes after analyzing body composition data collected across a large number of subjects Forbes1987 ; Forbes2000 . Forbes postulated that for adults $\alpha=\frac{\rho_{F}}{\rho_{L}}\frac{F}{10.4}$ (13) so that $F=D\exp(L/10.4)$ (14) where $D$ is a free parameter, and the lean and fat masses are in units of kg. Forbes found that his general relationship (14) is similar whether weight loss is induced by diet or exercise Forbes2000 . It is possible that resistance exercise or a significant change in the protein content of the diet may result in a different relationship for $\alpha$ hansen2007 ; layman2005 ; stiegler2006 . Infant growth is an example where $\alpha$ is not well described by the Forbes relationship. Jordan and Hall jordan used longitudinal body composition data in growing infants to determine an appropriate form for $\alpha$ during the first two years of life. Equation (12) describes a family of $F$ vs $L$ curves, parameterized by an integration constant (e.g. D in Eq. (14)). Depending on the initial condition, the body composition moves along one of these curves when out of energy balance. Dividing Eq. (8) by Eq. (9) and imposing Eq. (12) results in $f(F,L)=\frac{I_{F}-\alpha I_{L}+\alpha E}{(1+\alpha)E}=\frac{I_{F}}{E}-\frac{\alpha}{1+\alpha}\frac{I-E}{E}$ (15) Hall, Bain and Chow hbc showed that the two compartment macronutrient partition model with Eq. (15) using Forbes’s law (13) matched a wide range of data without any adjustable parameters. Substituting Eq. (15) into the macronutrient partition model (8) and (9) leads to the Energy Partition model: $\displaystyle\rho_{F}\frac{dF}{dt}$ $\displaystyle=$ $\displaystyle(1-p)(I-E)$ (16) $\displaystyle\rho_{L}\frac{dL}{dt}$ $\displaystyle=$ $\displaystyle p(I-E)$ (17) where $p=p(F,L)=1/(1+\alpha)$ is known as the p-ratio dugdale1977 . In the energy partition model, an energy imbalance $I-E$ is divided between the compartments according to a function $p(F,L)$ that defines the fraction assigned to lean body tissue (mostly protein). Most of the previous models in the literature are different versions of the energy partition model antonetti1973 ; payne1977b ; alpert1990 ; Girardier1994 ; westerterp1995 ; kozusko2001 ; flatt2004 ; christiansen2005 , although none of the authors have noted the connection to macronutrient flux balance or analyzed their models using dynamical systems theory. Some of these previous models are expressed as computational algorithms that can be translated to the form of the energy partition model. Despite the ubiquity of the energy partition model, the physiological interpretation of the p-ratio remains obscure and is difficult to measure directly. It can be inferred indirectly from $f$ (which can be measured by indirect calorimetry) by using Eq. (15) hbc . Previous uses of the energy partition model often considered $p$ to be a constant antonetti1973 ; payne1977b ; alpert1990 ; Girardier1994 ; kozusko2001 ; flatt2004 ; christiansen2005 , which implies that the partitioning of energy is independent of current body composition and macronutrient composition. This is in contradiction to weight loss data that finds that the fraction of body fat lost does depend on body composition with more fat lost if the body fat is initially higher Forbes1987 ; elia1999 ; hall2007b . However, if $\alpha$ is a weak function of body composition then a constant p-ratio may be a valid approximation for small changes. Flatt flatt2004 considered a model where the p-ratio was constant but included the dynamics for glycogen. His model would be useful when dynamics on short time scales are of interest. It may sometimes be convenient to express the macronutrient partition model with a unique fixed point as $\displaystyle\rho_{F}\frac{dF}{dt}$ $\displaystyle=$ $\displaystyle(1-p)(I-E)+\psi$ (18) $\displaystyle\rho_{L}\frac{dL}{dt}$ $\displaystyle=$ $\displaystyle p(I-E)-\psi$ (19) for a function $\psi=\psi(I_{F},I_{L},F,L)$, which is zero at the fixed point $(F_{0},L_{0})$. We use this form in numerical examples in Sec. II.4.1. The fasting model of Song and Thomas song2007 ) used this form with $I=0$ and $\psi$ was a function of $F$ representing ketone production. Comparing to Eq. (8) and Eq. (18) gives $f=\frac{I_{F}}{E}-(1-p)\frac{I-E}{E}-\frac{\psi}{E}$ (20) #### II.2.3 One Dimensional models The dynamics of the energy partition model Eqs. (16) and (17) move along fixed trajectories in the $L$ \- $F$ plane. Thus a further simplification to a one dimensional model is possible by finding a functional relationship between $F$ and $L$ so that one variable can be eliminated in favor of the other. Such a function exists if Eq. (12) has a unique solution, which is guaranteed in some interval of $L$ if $\alpha(F,L)$ and $\partial\alpha/\partial F$ are continuous functions of $F$ and $L$ on a rectangle containing this interval. These are sufficient but not necessary conditions. Suppose a relationship $F=\phi(L)$ can be found between $F$ and $L$. Substituting this relationship into Eq. (16) and Eq. (17) and adding the two resulting equations yields the one dimensional equation $\frac{dL}{dt}=\frac{I-E(\phi(L),L)}{\rho_{F}\phi^{\prime}(L)+\rho_{L}}$ (21) We can obtain a dynamical equation for body mass by expressing the body mass as $M=L+\phi(L)$. If we can invert this relationship uniquely and obtain $L$ as a function of $M$, then this can be substituted into Eq. (21) to obtain a dynamical equation for $M$. As an example, assume $p$ to be a constant, which was used in antonetti1973 ; payne1977b ; alpert1990 ; Girardier1994 ; kozusko2001 ; flatt2004 ; christiansen2005 . This implies that the phase orbits are a family of straight lines of the form $F=\beta L+C\equiv\phi(L)$ where $\beta=\rho_{L}(1-p)/(\rho_{F}p)$ and $C$ is a constant that is specified by the initial body composition. This results in $\frac{dM}{dt}=\left(\frac{1-p}{\rho_{F}}+\frac{p}{\rho_{L}}\right)\left[I-E\left(\frac{\beta}{1+\beta}(M-C)+C,\frac{M-C}{1+\beta}\right)\right]$ (22) Linearizing Eq. (22) around a mass $M_{0}$ gives $\rho_{M}\frac{dM}{dt}=\mu-\epsilon(M-M_{0})$ (23) where $\rho_{M}=\rho_{F}\rho_{L}/(\rho_{L}+(\rho_{F}-\rho_{L})p)$, $\mu=I-E(F(M_{0}),L(M_{0}))$ and $\epsilon=dE/dM|_{M=M_{0}}$. This is the form used in christiansen2005 . If Eqs. (16) and (17) are constrained to obey the phase plane paths of Forbes’s law, then a reduction to a one dimensional equation can also be made. Using Eq. (14) (i.e. $\phi(L)=D\exp(L/10.4)$) in Eq. (21) yields $\frac{dL}{dt}=\frac{10.4}{\rho_{F}D\exp(L/10.4)+10.4\rho_{L}}[I-E(D\exp(L/10.4),L)]$ (24) Similarly, a one dimensional equation for the fat mass has the form $\frac{dF}{dt}=\frac{F}{\rho_{F}F+10.4\rho_{L}}[I-E(F,10.4\log(F/D))]$ (25) Since the mass functions $M=L+D\exp(L/10.4)$ or $M=F+10.4\log(F/D)$ cannot be inverted in closed form, an explicit one dimensional differential equation in terms of the mass cannot be derived. However, the dynamics of the mass is easily obtained using either Eq. (24) or Eq. (25) together with the relevant mass function. For large changes in body composition, the dynamics could differ significantly from the constant $p$ models (22) or (23). The one dimensional model gives the dynamics of the energy partition model along a fixed trajectory in the $F$ \- $L$ plane. The initial body composition specifies the constant $C$ or $D$ in the above equations. A one dimensional model will represent the energy partition model even if the intake rate is time dependent. Only for a perturbation that directly alters body composition will the one dimensional model no longer apply. However, after the perturbation ceases, the one dimensional model with a new constant will apply again. ### II.3 Existence and stability of body weight fixed points The various flux balance models can be analyzed using the methods of dynamical systems theory, which aims to understand dynamics in terms of the geometric structure of possible trajectories (time courses of the body components). If the models are smooth and continuous then the global dynamics can be inferred from the local dynamics of the model near fixed points (i.e. where the time derivatives of the variables are zero). To simplify the analysis, we consider the intake rates to be clamped to constant values or set to predetermined functions of time. We do not consider the control and variation of food intake rate that may arise due to feedback from the body composition or from exogenous influences. We focus only on what happens to the food once it is ingested, which is a problem independent of the control of intake. We also assume that the averaged energy expenditure rate does not depend on time explicitly. Hence, we do not account for the effects of development, aging or gradual changes in lifestyle, which could lead to an explicit slow time dependence of energy expenditure rate. Thus, our ensuing analysis is mainly applicable to understanding the slow dynamics of body mass and composition for clamped food intake and physical activity over a time course of months to a few years. Dynamics in two dimensions are particularly simple to analyze and can be easily visualized geometrically strogatz ; gh . The one dimensional models are a subclass of two dimensional dynamics. Three dimensional dynamical systems are generally more difficult to analyze but Hall hall2006 found in simulations that the glycogen levels varied over a small interval and averaged to an approximate constant for time periods longer than a few days, implying that the slow dynamics could be effectively captured by a two dimensional model. Reduction to fewer dimensions is an oft used strategy in dynamical systems theory. Hence, we focus our analysis on two dimensional dynamics. In two dimensions, changes of body composition and mass are represented by trajectories in the $L$ \- $F$ phase plane. For $I_{F}$ and $I_{L}$ constant, the flux balance model is a two dimensional autonomous system of ordinary differential equations and trajectories will flow to attractors. The only possible attractors are infinity, stable fixed points or stable limit cycles strogatz ; gh . We note that fixed points within the context of the model correspond to states of flux balance. The two compartment macronutrient partition model is completely general in that all possible autonomous dynamics in the two dimensional phase plane are realizable. Any two or one dimensional autonomous model of body composition change can be expressed in terms of the two dimensional macronutrient partition model. Physical viability constrains $L$ and $F$ to be positive and finite. For differentiable $f$ and $E$, the possible trajectories for fixed intake rates are completely specified by the dynamics near fixed points of the system. Geometrically, the fixed points are given by the intersections of the nullclines in the $L$ \- $F$ plane, which are given by the solutions of $I_{F}-fE=0$ and $I_{L}-(1-f)E=0$. Example nullclines and phase plane portraits of the macronutrient model are shown in Fig. 1. If the nullclines intersect once then there will be a single fixed point and if it is stable then the steady state body composition and mass are uniquely determined. Multiple intersections can yield multiple stable fixed points implying that body composition is not unique hall2002 . If the nullclines are collinear then there can be an attracting one-dimensional invariant manifold (continuous curve of fixed points) in the $L$ \- $F$ plane. In this case, there are an infinite number of possible body compositions for a fixed diet. As we will show, the energy partition model implicitly assumes an invariant manifold. If a single fixed point exists but is unstable then a stable limit cycle may exist around it. Figure 1: Possible trajectories (solid lines) for different initial conditions and nullclines (dotted lines) in the $L$ \- $F$ phase plane for models with a stable fixed point a), multi-stability with two stable fixed points separated by one unstable saddle point b), an attracting invariant manifold c), and a limit cycle attractor d). The fixed point conditions of Eqs. (8) and (9) can be expressed in terms of the solutions of $\displaystyle E(F,L)=I$ (26) $\displaystyle f(F,L)=\frac{I_{F}}{I}$ (27) where $I=I_{F}+I_{L}$, and we have suppressed the functional dependence on intake rates. These fixed point conditions correspond to a state of flux balance of the lean and fat components. Equation (26) indicates a state of energy balance while Eq. (27) indicates that the fraction of fat utilized must equal the fraction of fat in the diet. Stability of a fixed point is determined by the dynamics of small perturbations of body composition away from the fixed point. If the perturbed body composition returns to the original fixed point then the fixed point is deemed stable. We give the stability conditions in Methods. The functional dependence of $E$ and $f$ on $F$ and $L$ determine the existence and stability of fixed points. As shown in Methods, an isolated stable fixed point is guaranteed if $f$ is a monotonic increasing function of $F$ and a monotonic decreasing function of $L$. If one of the fixed point conditions automatically satisfies the other, then instead of a fixed point there will be a continuous curve of fixed points or an invariant manifold. For example, if the energy balance condition (26) automatically satisfies the fat fraction condition (27), then there is an invariant manifold defined by $I=E$. The energy partition model has this property and thus has an invariant manifold rather than an isolated fixed point. This can be seen by observing that for $f$ given by Eq. (15), $I=E$ automatically satisfies condition (27). An attracting invariant manifold implies that the body can exist at any of the infinite number of body compositions specified by the curve $I=E(F,L)$ for clamped intake and energy expenditure rates (see Figure 1 c)). Each of these infinite possible body compositions will result in a different body mass $M=F+L$ (except for the unlikely case that $E$ is a function of the sum $F+L$). The body composition is marginally stable along the direction of the invariant manifold. This means that in flux balance, the body composition will remain at rest at any point on the invariant manifold. A transient perturbation along the invariant manifold will simply cause the body composition to move to a new position on the invariant manifold. The one dimensional models have a stable fixed point if the invariant manifold is attracting. We also show in Methods that for multiple stable fixed points or a limit cycle to exist, $f$ must be nonmonotonic in $L$ and be finely tuned. The required fine tuning makes these latter two possibilities much less plausible than a single fixed point or an invariant manifold. Data suggest that $E$ is a monotonically increasing function of $F$ and $L$ cunningham1991 . The dependence of $f$ on $F$ and $L$ is not well established and the form of $f$ depends on multiple interrelated factors. In general, the sensitivity of various tissues to the changing hormonal milieu will have an overall effect on both the supply of macronutrients as well as the substrate preferences of various metabolically active tissues. On the supply side, we know that free fatty acids derived from adipose tissue lipolysis increase with increasing body fat mass which thereby increase the daily fat oxidation fraction, $f$, as $F$ increases astrup1994 . Furthermore, reduction of $F$ with weight loss has been demonstrated to decrease $f$ astrup1992 . Similarly, whole-body proteolysis and protein oxidation increases with lean body mass welle1990 ; short2004 implying that $f$ should be a decreasing function of $L$. In further support of this relationship, body builders with significantly increased $L$ have a decreased daily fat oxidation fraction versus control subjects with similar $F$ bosselaers1994 . Thus a stable isolated fixed point is consistent with this set of data. ### II.4 Implications for body mass and composition change We have shown that all two dimensional autonomous models of body composition change generically fall into two classes - those with fixed points and those with invariant manifolds. In the case of a stable fixed point, any temporary perturbation of body weight or composition will be corrected over time (i.e. for all things equal, the body will return to its original state). An invariant manifold allows the possibility that a transient perturbation could lead to a permanent change of body composition and mass. At first glance, these differing properties would appear to point to a simple way of distinguishing between the two classes. However, the traditional means of inducing weight change, namely diet or altering energy expenditure through aerobic exercise, turn out to be incapable of revealing the distinction. For an invariant manifold, any change of intake or expenditure rate will only elicit movement along one of the prescribed $F$ vs $L$ trajectories obeying Eq. (12), an example being Forbes’s law (14). As shown in Fig. 2, a change of intake or energy expenditure rate will change the position of the invariant manifold. The body composition that is initially at one point on the invariant manifold will then flow to a new point on the perturbed invariant manifold along the trajectory prescribed by (12). If the intake rate or energy expenditure is then restored to the original value then the body composition will return along the same trajectory to the original steady state just as it would in a fixed point model (see Fig 2 solid curves). Only a perturbation that moves the body composition off of the fixed trajectory could distinguish between the two classes. In the fixed point case (Fig. 2 a) dashed-dot curve), the body composition would go to the same steady state following the perturbation to body composition but for the invariant manifold case (Fig 2 b) dashed-dot curve), it would go to another steady state. Figure 2: An example of a situation where the intake or energy expenditure rate is changed from one clamped value to another and then returned for the fixed point case a) and the invariant manifold case b). Dotted lines represent nullclines. In both cases, the body composition follows a fixed trajectory and returns to the original steady state (solid curves). However, if the body composition is perturbed directly (dashed-dot curves) then the body composition will flow to same point in a) but to a different point in b). Perturbations that move the body composition off the fixed trajectory can be done by altering body composition directly or by altering the fat utilization fraction $f$. For example, body composition could be altered directly through liposuction or administering compounds such as growth hormone. Resistance exercise may cause an increase in lean muscle tissue at the expense of fat. Exogenous hormones, compounds, or infectious agents that change the propensity for fat versus carbohydrate oxidation (for example, by increasing adipocyte proliferation and acting as a sink for fat that is not available for oxidation atkinson ; pasarica ; vangipuram ), would also perturb the body composition off of a fixed $F$ vs $L$ curve by altering $f$. If the body composition returned to its original state after such a perturbation then there is a unique fixed point. If it does not then there could be an invariant manifold although multiple fixed points are also possible. We found an example of one clinical study that bears on the question of whether humans have a fixed point or an invariant manifold. Biller et al. investigated changes of body composition pre- and post- growth hormone therapy in forty male subjects with growth hormone deficiency biller2000 . Despite significant changes of body composition induced by 18 months of growth hormone administration, the subjects returned very closely to their original body composition 18 months following the removal of therapy. However, there was a slight (2%) but significant increase in their lean body mass compared with the original value. Perhaps not enough time had elapsed for the lean mass to return to the original level. Alternatively, the increased lean mass may possibly have been the result of increased bone mineral mass and extracellular fluid expansion, both of which are known effects of growth hormone, but were assumed to be constant in the body composition models. Therefore, this clinical study provides some evidence in support of a fixed point, but it has not been repeated and the result was not conclusive. Using data from the Minnesota experiment keys1950 and the underlying physiology, Hall hall2006 proposed a form for $f$ that predicts a fixed point. On the other hand, Hall, Bain and Chow hbc showed that an invariant manifold model is consistent with existing data of longitudinal weight change but these experiments only altered weight through changes in caloric intake so this cannot rule out the possibility of a fixed point. Thus it appears that existing data is insufficient to decide the issue. #### II.4.1 Numerical simulations We now consider some numerical examples using the macronutrient partition model in the form given by Eqs. (18) and (19), with a p-ratio consistent with Forbes’s law (13): $p=2/(2+F)$, where $F$ is in units of kg. Consider two cases of the model. If $\psi=0$ then the model has an invariant manifold and body composition moves along a fixed trajectory in the $L$ \- $F$ plane. If $\psi$ is nonzero, then there can be an isolated fixed point. We will show an example where if the intake energy is perturbed, the approach of the body composition to the steady state will be identical for both cases but if body composition is perturbed, the body will arrive at different steady states. For every model with an invariant manifold, a model with a fixed point can be found such that trajectories in the $L$ \- $F$ plane resulting from energy intake perturbations will be identical. All that is required is that $\psi$ in the fixed point model is chosen such that the solution of $\psi(F,L)=0$ defines the fixed trajectory of the invariant manifold model. Using Forbes’s law (14), we choose $\psi=0.05(F-0.4\exp(L/10.4))/F$. We then take a plausible energy expenditure rate of $E=0.14L+0.05F+1.55$, where energy rate has units of MJ/day and mass has units of kg. This expression is based on combining cross-sectional data cunningham1991 for resting energy with a contribution of physical activity of a fairly sedentary person hall2006 . Previous models propose similar forms for the energy expenditure payne1977b ; alpert1979 ; alpert2005 ; christiansen2005 . Figure 3 shows the time dependence of body mass and the $F$ vs $L$ trajectories of the two model examples given a reduction in energy intake rate from 12 MJ/day to 10 MJ/day starting at the same initial condition. The time courses are identical for body composition and mass. The mass first decreases linearly in time but then saturates to a new stable fixed point. The dashed line represents the same intake rate reduction but with 10 kg of fat removed at day 100. For the invariant manifold model, the fat perturbation permanently alters the final body composition and body mass, whereas in the fixed point model it only has a transient effect. In the fixed point model, the body composition can ultimately exist only at one point given by the intersection of the nullclines (i.e. solution of $I=E$ and $\psi=0$). For the invariant manifold, the body composition can exist at any point on the $I=E$ curve (dotted line in Fig. 2 d)). Since a $\psi$ can always be found so that a fixed point model and an invariant manifold model have identical time courses for body composition and mass, a perturbation in energy intake can never discriminate between the two possibilities. The time constant to reach the new fixed point in the numerical simulations is very long. This slow approach to steady state (on the order of several years for humans) has been pointed out many times previously hall2006 ; payne1977b ; alpert1979 ; alpert2005 ; christiansen2005 . A long time constant will make experiments to distinguish between a fixed point and an invariant manifold difficult to conduct. Experimentally reproducing this example would be demanding but if the time variation of the intake rates and physical activity levels were small compared to the induced change then the same result should arise qualitatively. Additionally, the time constant depends on the form of the energy expenditure. There is evidence that the dependence of energy expenditure on $F$ and $L$ for an individual is steeper than for the population due to an effect called adaptive thermogenesis leibel , thus making the time constant shorter. Figure 3: Time dependence of body mass for fixed point model a). In all the figures, the solid line is for an intake reduction from 12 MJ/day to 10 J/day and the dashed line is for the same reduction but with a removal of 10 kg of fat at day 100. Trajectories in the $F$ vs $L$ phase plane for the fixed point model b). Dotted lines are the nullclines. Time dependence c) and phase plane d) of the invariant manifold model for the same conditions. ## III Discussion In this paper we have shown that all possible two dimensional autonomous models for lean and fat mass are variants of the macronutrient partition model. The models can be divided into two general classes - models with isolated fixed points (most likely a single stable fixed point) and models with an invariant manifold. There is the possibility of more exotic behavior such as multi-stability and limit cycles but these require fine tuning and thus are less plausible. Surprisingly, experimentally determining if the body exhibits a fixed point or an invariant manifold is nontrivial. Only perturbations of the body composition itself apart from dietary or energy expenditure interventions or alterations of the fraction of energy utilized as fat can discriminate between the two possibilities. The distinction between the classes is not merely an academic concern since this has direct clinical implications for potential permanence of transient changes of body composition via such procedures as liposuction or temporary administration of therapeutic compounds. Our analysis considers the slow dynamics of the body mass and composition where the fast time dependent hourly or daily fluctuations are averaged out for a clamped average food intake rate. We also do not consider a slow explicit time dependence of the energy expenditure. Such a time dependence could arise during development, aging or gradual changes in lifestyle where activity levels differ. Thus our analysis is best suited to modeling changes over time scales of months to a few years in adults. We do not consider any feedback of body composition on food intake, which is an extremely important topic but beyond the scope of this paper. Previous efforts to model body weight change have predominantly used energy partition models that implicitly contain an invariant manifold and thus body composition and mass are not fully specified by the diet. If the body does have an invariant manifold then this fact puts a very strong constraint on the fat utilization fraction $f$. Hall hall2006 considered the effects of carbohydrate intake on lipolysis and other physiological factors to conjecture a form of $f$ that does not lead to an invariant manifold. However, our analysis and numerical examples show that the body composition could have an invariant manifold but behave indistinguishably from having a fixed point. Also, the decay to the fixed point could take a very long time, possibly as long as a decade giving the appearance of an invariant manifold. Only experiments that perturb the fat or lean compartments independently can tell. ## IV Methods ### IV.1 Method of Averaging The three compartment macronutrient flux balance equations (3)-(5) are a system of nonautonomous differential equations since the energy intake and expenditure are explicitly time dependent. Food is ingested over discrete time intervals and physical activity will vary greatly within a day. However, this fast time dependence can be viewed as oscillations or fluctuations on top of a slowly varying background. It is this slower time dependence that governs long-term body mass and composition changes that we are interested in. For example, if an individual had the exact same schedule with the same energy intake and expenditure each day, then averaged over a day, the body composition would be constant. If the daily averaged intake and expenditure were to gradually change on longer time scales of say weeks or months then there would be a corresponding change in the body composition and mass. Given that we are only interested in these slower changes, we remove the short time scale fluctuations by using the method of averaging to produce an autonomous system of averaged equations valid on longer time scales. We do so by introducing a second “fast” time variable $\tau=t/\epsilon$, where $\epsilon$ is a small parameter that is associated with the slow changes and let all time dependent quantities be a function of both $t$ and $\tau$. For example, if $t$ is measured in units of days and $\tau$ is measured in units of hours then $\epsilon\sim 1/24$. Inserting into (3)-(5) and using the chain rule yields $\displaystyle\rho_{F}\left(\frac{\partial F}{\partial t}+\frac{1}{\epsilon}\frac{\partial F}{\partial\tau}\right)$ $\displaystyle=$ $\displaystyle I_{F}(t,\tau)-f_{F}E(t,\tau)$ (28) $\displaystyle\rho_{G}\left(\frac{\partial G}{\partial t}+\frac{1}{\epsilon}\frac{\partial G}{\partial\tau}\right)$ $\displaystyle=$ $\displaystyle I_{C}(t,\tau)-f_{C}E(t,\tau)$ (29) $\displaystyle\rho_{P}\left(\frac{\partial P}{\partial t}+\frac{1}{\epsilon}\frac{\partial P}{\partial\tau}\right)$ $\displaystyle=$ $\displaystyle I_{P}(t,\tau)-(1-f_{F}-f_{C})E(t,\tau)$ (30) We then consider the three body compartments to have expansions of the form $\displaystyle F(t,\tau)$ $\displaystyle=$ $\displaystyle F^{0}(t)+\epsilon F^{1}(t,\tau)+O(\epsilon^{2})$ (31) $\displaystyle G(t,\tau)$ $\displaystyle=$ $\displaystyle G^{0}(t)+\epsilon G^{1}(t,\tau)+O(\epsilon^{2})$ (32) $\displaystyle P(t,\tau)$ $\displaystyle=$ $\displaystyle P^{0}(t)+\epsilon P^{1}(t,\tau)+O(\epsilon^{2})$ (33) where $\langle F^{1}\rangle=\langle P^{1}\rangle=\langle G^{1}\rangle=0$ for a time average defined by $\langle X\rangle=(1/T)\int_{0}^{T}X\ d\tau$ and $T$ represents an averaging time scale of a day. The fast time dependence can be either periodic or stochastic. The important thing is that the time average over the fast quantities is of order $\epsilon$ or higher. We then expand the energy expenditure rate and expenditure fractions to first order in $\epsilon$: $E(F,G,P,t,\tau)=E^{0}(t,\tau)+\epsilon\left(\frac{\partial E}{\partial F}F^{1}+\frac{\partial E}{\partial G}G^{1}+\frac{\partial E}{\partial P}P^{1}\right)+O(\epsilon^{2})$ (34) $\displaystyle f^{0}_{i}(F,G,P)=f_{i}(F^{0},G^{0},P^{0})+\epsilon\left(\frac{\partial f_{i}}{\partial F}F^{1}+\frac{\partial f_{i}}{\partial G}G^{1}+\frac{\partial f_{i}}{\partial P}P^{1}\right)+O(\epsilon^{2})$ (35) where $E^{0}(t,\tau)\equiv E(F^{0},G^{0},P^{0},t,\tau)+O(\epsilon^{2})$ and $i\in\\{F,G,P\\}$. We assume that the expenditure fractions depend on time only through the body compartments. Substituting these expansions into Eqs. (28)-(30) and taking lowest order in $\epsilon$ gives $\displaystyle\rho_{F}\left(\frac{\partial F^{0}}{\partial t}+\frac{\partial F^{1}}{\partial\tau}\right)$ $\displaystyle=$ $\displaystyle I_{F}(t,\tau)-f^{0}_{F}E^{0}(t,\tau)$ (36) $\displaystyle\rho_{G}\left(\frac{\partial G^{0}}{\partial t}+\frac{\partial G^{1}}{\partial\tau}\right)$ $\displaystyle=$ $\displaystyle I_{C}(t,\tau)-f^{0}_{C}E^{0}(t,\tau)$ (37) $\displaystyle\rho_{P}\left(\frac{\partial P^{0}}{\partial t}+\frac{\partial P^{1}}{\partial\tau}\right)$ $\displaystyle=$ $\displaystyle I_{P}(t,\tau)-(1-f_{F}-f^{0}_{C})E^{0}(t,\tau)$ (38) Taking the moving time average of Eqs. (36)-(38) and requiring that $\langle\partial F^{1}/\partial\tau\rangle$, $\langle\partial G^{1}/\partial\tau\rangle$, and $\langle\partial P^{1}/\partial\tau\rangle$ are of order $\epsilon$ or higher leads to the averaged equations: $\displaystyle\rho_{F}\frac{dF^{0}}{dt}$ $\displaystyle=$ $\displaystyle\langle I_{F}\rangle-f^{0}_{F}\langle E^{0}\rangle$ (39) $\displaystyle\rho_{G}\frac{dG^{0}}{dt}$ $\displaystyle=$ $\displaystyle\langle I_{C}\rangle-f^{0}_{C}\langle E^{0}\rangle$ (40) $\displaystyle\rho_{P}\frac{dP^{0}}{dt}$ $\displaystyle=$ $\displaystyle\langle I_{P}\rangle-(1-f^{0}_{F}-f^{0}_{C})\langle E^{0}\rangle$ (41) In the main text we only consider the slow time scale dynamics so we drop the superscript and bracket notation for simplicity. Hence, the system (3)-(5) can be thought of as representing the lowest order time averaged macronutrient flux balance equations. We note that in addition to the daily fluctuations of meals and physical activity, there can also be fluctuations in food intake from day to day periwal2006 . Our averaging scheme can be used to average over these fluctuations as well by extending the averaging time $T$. A difference in the choice of $T$ will only result in a different interpretation of the averaged quantities. ### IV.2 Stability conditions for fixed points The dynamics near a fixed point $(F_{0},L_{0})$ are determined by expanding $fE$ and $(1-f)E$ to linear order in $\delta F=F-F_{0}$ and $\delta L=L-L_{0}$ strogatz ; gh . Assuming solutions of the form $\exp(\lambda t)$ yields an eigenvalue problem with two eigenvalues given by $\lambda=\frac{1}{2}\left({\rm Tr}J\pm\sqrt{{\rm Tr}J^{2}-4\det J}\right)$ where ${\rm Tr}J=-\left[\frac{1}{\rho_{F}}\frac{\partial}{\partial F}(fE)+\frac{1}{\rho_{L}}\frac{\partial}{\partial L}((1-f)E)\right]_{(F_{0},L_{0})}$ (42) and $\det J=\frac{E}{\rho_{L}\rho_{F}}\left[\frac{\partial E}{\partial L}\frac{\partial f}{\partial F}-\frac{\partial E}{\partial F}\frac{\partial f}{\partial L}\right]_{(F_{0},L_{0})}$ (43) A fixed point is stable if and only if ${\rm Tr}J<0$ and $\det J>0$. In the case of an invariant manifold, $\det J=0$, so the eigenvalues are ${\rm Tr}J$ and $0$. The zero eigenvalue reflects the marginal stability along the invariant manifold, which is an attractor if ${\rm Tr}J<0$. An attracting invariant manifold implies a stable fixed point in the corresponding one dimensional model. Unstable fixed points are either unstable nodes, saddle points or unstable spirals. In the case of unstable spirals, a possibility is a limit cycle surrounding the spiral arising from a Hopf bifurcation, where ${\rm Tr}J=0$ and $\det J>0$. In this case, body composition and mass would oscillate even if the intake rates were held constant. The frequency and amplitude of the oscillations may be estimated near a supercritical Hopf bifurcation by transforming the equations to normal form. Stability of a fixed point puts constraints on the form of $f$. Physiological considerations and data imply that $\partial E/\partial L>\partial E/\partial F>0$ cunningham1991 ; hall2006 . Thus we can set $\partial E/\partial F=\delta\partial E/\partial L$ where $\delta<1$ (where the derivatives are evaluated at the fixed point). Then $\det J>0$ implies that $\partial f/\partial F>\delta\partial f/\partial L$ (44) and ${\rm Tr}J<0$ implies $\partial f/\partial F>\gamma\partial f/\partial L-K,$ (45) where $K=[\delta f+\gamma(1-f)](\partial E/\partial L)/E>0$ and $\gamma=\rho_{F}/\rho_{L}\simeq 5.2$. Hence $\partial f/\partial F>0$ and $\partial f/\partial L<0$ guarantees stability of a fixed point. From Eq. (44) and Eq. (45) and the fixed point conditions (26) and (27), if $f$ increases monotonically with $F$ and decreases monotonically with $L$ then there will be a unique stable fixed point. For an invariant manifold, $f$ is given by Eq. (15), which immediately satisfies $\det J=0$; ${\rm Tr}J<0$ is guaranteed if $E$ is monotonically increasing in $F$ and $L$. For a Hopf bifurcation, we require $\partial f/\partial F=\gamma\partial f/\partial L-K$ and Eq. (44), implying $(\gamma-\delta)\partial f/\partial L-K>0$. Since $\gamma>\delta$, $f$ must increase with $L$ for the possibility of a limit cycle. However, to ensure that trajectories remain bounded $f$ must decrease with $L$ for very small and large values of $L$. Hence, $f$ must be nonmonotonic in $L$ for a limit cycle to exist. This can also be seen from an application of Bendixson’s criterion gh , which states that a limit cycle cannot exist in a given region of the $L$ \- $F$ plane if $\frac{1}{\rho_{F}}\frac{\partial}{\partial F}(fE)+\frac{1}{\rho_{L}}\frac{\partial}{\partial L}((1-f)E)$ (46) does not change sign in that region. In addition, the other parameters must be fine tuned for a limit cycle (see Fig. 1 d)). Similarly, as seen in Fig. 1 c), for multi-stability to exist, nonmonotonicity and fine tuning are also required. ###### Acknowledgements. This research was supported by the Intramural Research Program of the NIH, NIDDK. ## References * [1] James O. Hill, Holly R. Wyatt, George W. Reed, and John C. Peters. Obesity and the environment: where do we go from here? Science, 299:853–55, 2003. * [2] J. O. Hill. Understanding and addressing the epidemic of obesity: an energy balance perspective. Endocr Rev, 27(7):750–61, 2006. * [3] K. D. Hall. Computational model of in vivo human energy metabolism during semistarvation and refeeding. Am J Physiol Endocrinol Metab, 291(1):E23–37, 2006. * [4] K Hall and D Polidori. Can a temporary high fat diet cause persistent obesity via acquired leptin resistance? insights from a mathematical model. Diabetes, 51(S2):A607, 2002. * [5] S. S. Alpert. A two-reservoir energy model of the human body. Am J Clin Nutr, 32(8):1710–8, 1979. * [6] S. S. Alpert. Growth, thermogenesis, and hyperphagia. Am J Clin Nutr, 52(5):784–92, 1990. * [7] S. S. Alpert. A limit on the energy transfer rate from the human fat store in hypophagia. J Theor Biol, 233(1):1–13, 2005. * [8] G. B. Forbes. Weight loss during fasting: implications for the obese. Am J Clin Nutr, 23(9):1212–9, 1970. * [9] E. H. Livingston, J. L. Sebastian, S. Huerta, I. Yip, and D. Heber. Biexponential model for predicting weight loss after gastric surgery for obesity. J Surg Res, 101(2):216–24, 2001. * [10] K. D. Hall, H. L. Bain, and C. C. Chow. How adaptations of substrate utilization regulate body composition. Int J Obes (Lond), 31(9):1378–83, 2007. * [11] P Jordan and KD Hall. Dynamic coordination of macronutrient balance during infant growth: Insights from a mathematical model. Am J Clin Nutr, in press. * [12] V. W. Antonetti. The equations governing weight change in human beings. Am J Clin Nutr, 26(1):64–71, 1973. * [13] P. R. Payne and A. E. Dugdale. A model for the prediction of energy balance and body weight. Ann Hum Biol, 4(6):525–35, 1977. * [14] L. Girardier. [autoregulation of body weight and body composition in man. a systematic approach through models and simulation]. Arch Int Physiol Biochim Biophys, 102(4):A23–35, 1994. * [15] K. R. Westerterp, J. H. Donkers, E. W. Fredrix, and P. Boekhoudt. Energy intake, physical activity and body weight: a simulation model. Br J Nutr, 73(3):337–47, 1995. * [16] F. P. Kozusko. Body weight setpoint, metabolic adaption and human starvation. Bull Math Biol, 63(2):393–403, 2001. * [17] J. P. Flatt. Carbohydrate-fat interactions and obesity examined by a two-compartment computer model. Obes Res, 12(12):2013–22, 2004. * [18] E. Christiansen, L. Garby, and T. I. Sorensen. Quantitative analysis of the energy requirements for development of obesity. J Theor Biol, 234(1):99–106, 2005. * [19] B. Song and D. M. Thomas. Dynamics of starvation in humans. J Math Biol, 54(1):27–43, 2007. * [20] W. O. Atwater and F. G. Benedict. Experiments on the metabolism of matter and energy in the human body. 1900-1902. Bulletin No. 136, U.S. Department of Agriculture, Office of Experiment Stations, Washington, DC., 1903. * [21] A Keys. The Biology of Human Starvation. University of Minnesota Press, Minneapolis, MN, 1950. * [22] W. O. Atwater and F. G Benedict. A respiration calorimeter with appliances for the direct determination of oxygen. Publication No. 42, Carnegie Institution of Washington, Washington, DC, 1905. * [23] V. Periwal and C. C. Chow. Patterns in food intake correlate with body mass index. American Journal of Physiology: Endocrinology and Metabolism, 291:E929–E936, 2006. * [24] K. D. Hall. What is the required energy deficit per unit weight loss? Int J Obes (Lond), 2007. * [25] W. S. Snyder, M. J. Cook, E. S. Nasset, L. R. Karhausen, G. P. Howells, and I. H. Tipton. Report on the Task Group on Reference Man. Pergamon, Oxford, UK, 1984. * [26] G. B. Forbes. Lean body mass-body fat interrelationships in humans. Nutr Rev, 45(8):225–31, 1987. * [27] G. B. Forbes. Body fat content influences the body composition response to nutrition and exercise. Ann N Y Acad Sci, 904:359–65, 2000. * [28] D. Hansen, P. Dendale, J. Berger, J. L. van Loon, and R. Meeusen. The effects of exercise training on fat-mass loss in obese patients during energy intake restriction. Sports Med (Auckland, NZ), 37:31–46, 2007. * [29] D.K. Layman, E. Evans, J. I. Baum, J. Seyler, D. J. Erickson, and R. A. Boileau. Dietary protein and exercise have additive effects on body composition during weight loss in adult women. J Nutr, 135:1903–1910, 2005. * [30] P. Stiegler and A. Cunliffe. The role of diet and exercise for the maintenance of fat-free mass and resting metabolic rate during weight loss. Sports Med (Auckland, NZ), 36:239–262, 2006. * [31] AE Dugdale and PR Payne. Pattern of lean and fat deposition in adults. Nature, 266:349–51, 1977. * [32] M Elia, RJ Stubbs, and CJ Henry. Differences in fat, carbohydrate, and protein metabolism between lean and obese subjects undergoing total starvation. Obes Res, 7:597–604, 1999. * [33] K. D. Hall. Body fat and fat-free mass inter-relationships: Forbes’s theory revisited. Brit J Nutr, 97:1059–63, 2007. * [34] J Guckenheimer and Holmes P. Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields. Springer, New York, 1997. * [35] S. H. Strogatz. Nonlinear Dynamics and Chaos: With Applications to Physics, Biology, Chemistry and Engineering. Perseus Books, Cambridge, MA, 2000. * [36] J. J. Cunningham. Body composition as a determinant of energy expenditure: a synthetic review and a proposed general prediction equation. Am J Clin Nutr, 54(6):963–9, 1991. * [37] A. Astrup, B. Buemann, P. Western, S. Toubro, A. Raben, and N. J. Christensen. Obesity as an adaptation to a high-fat diet: evidence from a cross-sectional study. Am J Clin Nutr, 59:350–5, 1994. * [38] A. Astrup, B. Buemann, N. J. Christensen, and J. Madsen. 24-hour energy expenditure and sympathetic activity in postobese women consuming a high-carbohydrate diet. Am J Physiol, 262:E282–8, 1992. * [39] S. Welle and K. S. Nair. Relationship of resting metabolic rate to body composition and protein turnover. Am J Physiol Endocrinol Metab, 258:E990–E998, 1990. * [40] K. R. Short, J. L. Vittone, M. L. Bigelow, D. N. Proctor, and K. S. Nair. Age and aerobic exercise training effects on whole body and muscle protein metabolism. Am J Physiol Endocrinol Metab, 286:E92–E101, 2004. * [41] I. Bosselaers, .B Buemann, O. J. Victor, and A. Astrup. Twenty-four-hour energy expenditure and substrate utilization in body builders. Am J Clin Nutr, 59:10–2, 1994. * [42] RL Atkinson. Viruses as an etiology of obesity. Mayo Clin Proc, 82:1192–8, 2007. * [43] M Pasarica, M Yu, Ho Yang, M Mahida, S MohanKumar, PS MohanKumar, K-L C Jen, and NV Dhurandhar. Human adenovirus ad-36 induces adiposity, up-regulates genes of adipocyte differentiation, and increases insulin sensitivity in rats. Obesity, 14:1905–13, 2006. * [44] SD Vangipuram, M Yu, J Tian, KL Stanhope, Pasarica M, PJ Havel, AR Heydari, and NV Dhurandhar. Adipogenic human adenovirus-36 reduces leptin expression and secretion and increases glucose uptake by fat cells. Int J Obes (Lond), 31:87–96, 2007. * [45] B. M. K. Biller, G. Sesmilo, H. B. A. Baum, D. Hayden, D. Schoenfeld, and A. Klibanski. Withdrawal of long-term physiological growth hormone (gh) administration: Differential effects on bone density and body composition in men with adult- onset gh deficiency. J Clin Endotrinol Metab, 85:970–976, 2000. * [46] Rudolph L. Leibel, Michael. Rosenbaum, and Jules Hirsch. Changes in energy expenditure resulting from altered body weight. The New England Journal of Medicine, 332(10):621–628, 1995.

arxiv-papers

2024-09-04T02:48:53.942387

0802.3238

# Three-dimensional stability of magnetically confined mountains on accreting neutron stars M. Vigelius 1 and A. Melatos 1 1 School of Physics, University of Melbourne, Parkville, VIC 3010, Australia E-mail: mvigeliu@physics.unimelb.edu.au (Submitted to MNRAS) ###### Abstract We examine the hydromagnetic stability of magnetically confined mountains, which arise when material accumulates at the magnetic poles of an accreting neutron star. We extend a previous axisymmetric stability analysis by performing three-dimensional simulations using the ideal-magnetohydrodynamic (ideal-MHD) code zeus-mp, investigating the role played by boundary conditions, accreted mass, stellar curvature, and (briefly) toroidal magnetic field strength. We find that axisymmetric equilibria are susceptible to the undular sub-mode of the Parker instability but are not disrupted. The line- tying boundary condition at the stellar surface is crucial in stabilizing the mountain. The nonlinear three-dimensional saturation state of the instability is characterized by a small degree of nonaxisymmetry ($\la 0.1$ per cent) and a mass ellipticity of $\epsilon\sim 10^{-5}$ for an accreted mass of $M_{a}=10^{-5}M_{\odot}$. Hence there is a good prospect of detecting gravitational waves from accreting millisecond pulsars with long-baseline interferometers such as Advanced LIGO. We also investigate the ideal-MHD spectrum of the system, finding that long-wavelength poloidal modes are suppressed in favour of toroidal modes in the nonaxisymmetric saturation state. ###### keywords: accretion, accretion disks – stars: magnetic fields – stars: neutron – pulsars: general ## 1 Introduction There exists strong observational evidence that the magnetic dipole moment of accreting neutron stars in X-ray binaries, $\mu$, decreases with accreted mass, $M_{a}$ (Taam & van de Heuvel, 1986; van den Heuvel & Bitzaraki, 1995), although Wijers (1997) noted that extra variables may enter this relation. Numerous mechanisms for field reduction have been proposed, such as accelerated Ohmic decay (Konar & Bhattacharya, 1997; Urpin & Konenkov, 1997), vortex-fluxoid interactions in the superconducting core (Muslimov & Tsygan, 1985; Srinivasan et al., 1990), and magnetic screening or burial (Bisnovatyi- Kogan & Komberg, 1974; Romani, 1990; Zhang, 1998; Payne & Melatos, 2004; Lovelace et al., 2005). The reader is referred to Melatos & Phinney (2001) for a comparative review. Magnetic burial occurs when accreted plasma flowing inside the Alfvén radius is chanelled onto the magnetic poles of the neutron star. The hydrostatic pressure at the base of the accreted column overcomes the magnetic tension of the magnetic field lines and spreads equatorwards, thereby distorting the frozen-in magnetic flux (Melatos & Phinney, 2001). Payne & Melatos (2004), hereafter PM04, computed self-consistently the _unique_ quasistatic sequence of ideal-MHD equilibria that describes how burial proceeds as a function of $M_{a}$, _while respecting the flux freezing constraint_ of ideal MHD. They found that the magnetic field is compressed into an equatorial belt, which confines the accreted mountain at the poles. A key result is that $\mu$ is reduced significantly once $M_{a}$ exceeds $\sim 10^{-5}M_{\odot}$, five orders of magnitude above previous estimates (Brown & Bildsten, 1998; Litwin et al., 2001). Generally speaking, one expects highly distorted hydromagnetic equilibria like those in PM04 to be disrupted on the Alfvén time-scale by a plethora of MHD instabilities. Surprisingly, however, Payne & Melatos (2007) (hereafter PM07) found the equilibria to be stable to axisymmetric ideal-MHD modes. When kicked, the mountain performs radial and lateral oscillations corresponding to global Alfvén and compressional modes, but it remains intact. An axisymmetric analysis, however, excludes instabilities involving toroidal modes and is therefore incomplete. In general, three-dimensional effects alter MHD stability, quantitatively and qualitatively. For example, Matsumoto & Shibata (1992) found the growth rate of the three-dimensional Parker instability to be higher than that of its two-dimensional counterpart. Masada et al. (2006) proved that newly born neutron stars containing a toroidal field are stable to the axisymmetric magneto-rotational instability yet unstable to its nonaxisymmetric counterpart. Differences between the two- and three- dimensional stability of MHD equilibria are also observed in a variety of solar contexts (Priest, 1984) and in tokamaks (Lifschitz, 1989; Goedbloed & Poedts, 2004). The central aim of this paper is to perform fully three-dimensional, ideal-MHD simulations to assess the stability of magnetic mountains, generalizing PM07. Importantly, we compute not just the linear growth rate but also the _nonlinear saturation state_ of any unstable modes. The latter property is what matters over the long accretion time-scale when evaluating magnetic burial as the cause of the observed reduction in $\mu$ (Payne, 2005), the persistence of millisecond oscillations in type-I X-ray bursts (Payne & Melatos, 2006b), and gravitational radiation from magnetic mountains (Melatos & Payne, 2005; Payne & Melatos, 2006a). The structure of the paper is as follows. We introduce our numerical setup in section 2 and validate it against previous axisymmetric results in section 3, characterizing the controlling influence of the boundary conditions for the first time. In section 4, we present three-dimensional simulations, which display growth of unstable toroidal modes. The instability is classified according to its dispersive properties and energetics, and the nonlinear saturation state is computed as a function of $M_{a}$. We compute the spectrum of global MHD oscillations in section 5. Resistive effects are postponed to a future paper. ## 2 Numerical model The accretion problem contains two fundamentally different time-scales: the long accretion time ($\sim 10^{8}$ yr) and the short Alfven time ($\sim 10^{-3}$ s). The wide discrepancy prevents us from treating the accretion problem dynamically, i.e. in a full MHD simulation, where mass is added through the outer boundary onto an initially dipolar field. Instead, for a given value of $M_{a}$, we compute the magnetohydrostatic equilibria, using the Grad-Shafranov solver developed by PM04, then load it into the ideal-MHD solver zeus-mp (Hayes et al., 2006) to test its hydromagnetic stability on the Alfvén time-scale $\tau_{\mathrm{A}}$. We find below that all quantities reach their saturation values after $\sim 10\tau_{A}\sim 10^{-2}$ s at a particular value of $M_{a}$ (e.g. $10^{-4}M_{\odot}$). As $M_{a}$ changes slowly, over $\sim 10^{8}$ yr, the saturation values adjust in a quasistatic way on the Alfvén time-scale. In practice, to study a different value of $M_{a}$ numerically, we recalculate the Grad-Shafranov eqilibrium and load the new equilibrium into zeus-mp. ### 2.1 Magnetic mountain equilibria Analytic and numerical recipes for calculating self-consistent ideal-MHD equilibria for magnetic mountains are set out in PM04. Here, we briefly restate the main points for the convenience of the reader. An axisymmetric equilibrium is generated by a scalar flux function $\psi(r,\theta)$, such that the magnetic field $\mathbf{B}=\frac{\nabla\psi}{r\sin\theta}\times\hat{\mathbf{e}}_{\phi}$ (1) automatically satisfies $\nabla\mathbf{\mn@boldsymbol{\cdot}}\mathbf{B}=0$. We employ the usual spherical coordinates $(r,\theta,\phi)$, where $\theta=0$ corresponds to the symmetry axis of the magnetic field before accretion. In the static limit, the mass conservation and MHD induction equations are identically satisfied, while the component of the momentum equation transverse to $\mathbf{B}$ reduces to a second order, nonlinear, elliptic partial differential equation for $\psi$, the Grad-Shafranov (GS) equation (PM04): $\Delta^{2}\psi=-F^{\prime}(\psi)\exp[-(\varphi-\varphi_{0})/c_{s}^{2}],$ (2) with $\Delta^{2}=\frac{1}{\mu_{0}r^{2}\sin^{2}\theta}\left[\frac{\partial^{2}}{\partial r^{2}}+\frac{\sin\theta}{r^{2}}\frac{\partial}{\partial\theta}\left(\frac{1}{\sin\theta}\frac{\partial}{\partial\theta}\right)\right].$ (3) Formally, $F(\psi)$ is an arbitrary function. However, the ideal-MHD flux- freezing constraint, that matter cannot cross flux surfaces, imposes an additional conservation law on the mass-flux ratio $\mathrm{d}M/\mathrm{d}\psi$, $F(\psi)=\frac{c_{s}^{2}}{2\pi}\frac{\mathrm{d}M}{\mathrm{d}\psi}\left\\{\int\mathrm{d}s\,r\sin\theta|\nabla\psi|^{-1}\mathrm{e}^{-(\varphi-\varphi_{0})/c_{s}^{2}}\right\\}^{-1},$ (4) which determines $F(\psi)$ uniquely when solved simultaneously with (2). The integration in (4) is performed along the field line $\psi=\mathrm{const}$, $\phi=\mathrm{const}$. The solution is insensitive to the exact form of $\mathrm{d}M/\mathrm{d}\psi$; we thus distribute the accreted mass $M_{a}$ uniformly over $0\leq\psi\leq\psi_{a}$, where $\psi_{a}$ is the flux enclosed within the polar cap, and take $\psi$ to be dipolar, initially, with hemispheric flux $\psi_{\ast}$. In writing (2) and (5), we approximate the gravitational field as uniform over the height of the mountain, and hence write the gravitational potential $\varphi$ as $\varphi=GM_{\ast}r/R_{\ast}^{2},$ (5) with $\varphi_{0}=GM_{\ast}/R_{\ast}$, where $M_{\ast}$ and $R_{\ast}$ denote the stellar mass and radius respectively. We also assume an isothermal equation of state, $\rho=c_{s}^{2}p$, where $c_{s}$ denotes the sound speed. By working in the ideal-MHD limit, we neglect elastic stresses (Melatos & Phinney, 2001; Haskell et al., 2006; Owen, 2006), the Hall drift (Geppert & Rheinhardt, 2002; Cumming et al., 2004; Pons & Geppert, 2007), and Ohmic diffusion (Romani, 1990; Geppert & Urpin, 1994). In particular, Ohmic diffusion causes the mass quadrupole moment of the magnetic mountain (and $\mu$) to saturate above a certain value of $M_{a}$ and may also affect the stability to resistive MHD (e.g. ballooning) modes.111J. Arons, private communication. We defer investigating these resistive effects to a forthcoming paper [Vigelius & Melatos (in preparation)]. We solve (2) and (4) simultaneously using the relaxation algorithm described in PM04, subject to the boundary conditions $\psi(R_{\ast},\theta)=\psi_{\ast}\sin^{2}\theta$ (line tying at the surface), $\partial\psi/\partial r(R_{m},\theta)=0$ at the outer boundary $r=R_{m}$, $\psi(r,0)=0$, and $\partial\psi/\partial\theta(r,\pi/2)=0$ (north-south symmetry). ### 2.2 Evolution in ZEUS-MP In this paper, we explore numerically how the axisymmetric GS equilibria evolve when subjected to a variety of initial and boundary conditions in three dimensions. To achieve this, we employ the parallelized, general purpose, time-dependent, ideal-MHD solver zeus-mp (Hayes et al., 2006). zeus-mp integrates the equations of ideal MHD, discretized on a fixed staggered grid. The hydrodynamic part is based on a finite-difference advection scheme accurate to second order in time and space. The magnetic tension force and the induction equation are solved via the method of characteristics and constrained transport (MOCCT) (Hawley & Stone, 1995), whose numerical implementation in zeus-mp is described in detail by Hayes et al. (2006). We initialize zeus-mp with an equilibrium computed by the GS code and described by $\mathbf{B}(r,\theta)$ and $\rho(r,\theta)$, rotated about the $z$ axis to generate cylindrical symmetry. We introduce initial perturbations by taking advantage of the numerical noise produced by the transition between grids in the GS code and zeus-mp. We adopt dimensionless variables in zeus-mp satisfying $\mu_{0}=G=c_{s}=h_{0}=1$, where $h_{0}=c_{s}^{2}R_{\ast}^{2}/GM_{\ast}$ denotes the hydrostatic scale height. The basic units of mass, magnetic field, and time are then $M_{0}=h_{0}c_{s}^{2}/G$, $B_{0}=[c_{s}^{4}/(Gh_{0}^{2})]^{1/2}$, and $\tau_{0}=h_{0}/c_{s}$. The grid and boundary conditions are specified in appendix A. ### 2.3 Curvature rescaling In general, the characteristic length-scale for radial gradients ($h_{0}$) is much smaller than the length-scale for latitudinal gradients $R_{\ast}$, creating numerical difficulties. However, in the small-$M_{a}$ limit, it can be shown analytically (PM04, PM07) that the structure of the magnetic mountain depends on $R_{\ast}$ and $M_{\ast}$ through the combination $h_{0}\propto R_{\ast}^{2}/M_{\ast}$, not separately. We therefore artificially reduce $R_{\ast}$ and $M_{\ast}$, while keeping $h_{0}$ fixed, to render the problem tractable computationally. It is important to bear in mind that invariance of the equilibrium structure under this curvature rescaling does not imply invariance of the dynamical behaviour, nor is it necessarily applicable at large $M_{a}$. A standard neutron star has $M_{\ast}=1.4M_{\odot}$, $R_{\ast}=10^{6}$ cm, $B_{\ast}=10^{12}$ G, and $c_{s}=10^{8}$ cm s-1, giving $h_{0}=53.82$ cm, $a=R_{\ast}/h_{0}=1.9\times 10^{4}$, and $\tau_{0}=5.4\times 10^{-7}$ s. We rescale the star to $M_{\ast}^{\prime}=1.0\times 10^{-5}M_{\odot}$ and $R_{\ast}^{\prime}=2.7\times 10^{3}$ cm, reducing $a$ to 50 while keeping it large. The base units for this rescaled star (see section 2.2) are then $M_{0}=8.1\times 10^{24}$ g, $\rho_{0}=5.2\times 10^{19}$ g cm-3, $B_{0}=7.2\times 10^{17}$ G, and $\tau_{0}=5.4\times 10^{-7}$ s. The critical accreted mass above which the star’s magnetic moment starts to change, $M_{c}$, is defined by equation (30) of PM04: $\frac{M_{c}}{M_{\odot}}=6.2\times 10^{-15}\left(\frac{a}{50}\right)^{4}\left(\frac{B_{\ast}}{10^{12}\mathrm{G}}\right)^{2}\left(\frac{c_{s}}{10^{8}\mathrm{cm\;s}^{-1}}\right)^{-4}.$ (6) A characteristic time-scale for the MHD response of the mountain is the Alfvén pole-equator crossing time, $\tau_{\mathrm{A}}=\pi R_{\ast}/(2v_{\mathrm{A}})$, where $v_{\mathrm{A}}=(B^{2}/\mu_{0}\rho)^{1/2}$ is the Alfvén speed. Clearly, $v_{\mathrm{A}}$ is a function of position and time, so the definition of $\tau_{\mathrm{A}}$ is somewhat arbitrary. Typically, at the equator, we find $B\sim 10^{-6}B_{0}$ and $\rho\sim 10^{-11}\rho_{0}$, empirically implying $\tau_{\mathrm{A}}\approx 250\tau_{0}$. ## 3 Axisymmetric stability and global oscillations Table 1: Simulation parameters. $M_{a}$ is the accreted mass, in units of the characteristic mass $M_{c}$ (section 2.3), and $a=R_{\ast}/h_{0}$ measures the curvature of the rescaled star (section 2.3). The conditions at the outer boundary ($r=R_{m}$) are either outflow (zero gradient in all field variables) or inflow (pinned magnetic field); cf. also appendix A. Model | $M_{a}/M_{c}$ | $a$ | Axisymmetry | Boundary ---|---|---|---|--- A | 1.0 | 50 | yes | outflow B | 1.0 | 50 | yes | inflow D | 1.0 | 50 | no | outflow E | 1.0 | 50 | no | inflow F | 0.6 | 50 | no | outflow G | 1.4 | 50 | no | outflow J | 1.0 | 75 | no | outflow K | 1.0 | 100 | no | outflow PM07 demonstrated the axisymmetric stability of magnetic mountains using the serial ideal-MHD solver zeus-3d. Here, we start by repeating these axisymmetric simulations in the parallel solver zeus-mp, in order to verify the mountain implementation in zeus-3d and zeus-mp, generate an axisymmetric reference model, and understand the effect of the boundary conditions, which were not investigated fully in previous work. The simulation parameters are detailed in Table 1 (models A and B). Figure 1: Meridional section of model A at $t/\tau_{\mathrm{A}}=0,0.4,1,1.8,2.4,3.6$ (top left to bottom right). Shown are density contours (dashed curves) with values $\log_{10}(\rho/\rho_{0}^{\prime})=-13,-12,-11,-10.7,-10.5,-10.3$, and flux surfaces with footpoints at $r=R_{\ast}$, $\theta=0.10,0.12,0.15,0.20,0.39,0.79$ (solid curves). Lateral oscillations of the equatorial field lines are clearly visible. The outflow boundary condition at $r=R_{m}$ makes the field lines flare towards the magnetic pole. This is visible most clearly for the line whose footpoint lies at $\theta=0.39$. ### 3.1 Reference model Model A, in which we set $M_{a}/M_{c}=1.0$ and the outer boundary condition to outflow, serves as a reference case. Fig. 1 displays a time series of six $r$-$\theta$ sections for $0\leq t/\tau_{A}\leq 3.6$, showing density contours (dashed curves) and the magnetic field lines projected into the plane $\phi=0$ (solid curves). The axisymmetric equilibrium (top-left panel) reveals how the bulk matter is contained at the magnetic pole by the tension of the distorted magnetic field. The mountain in Fig. 1 performs damped lateral oscillations without being disrupted. The (unexpected) stability of this configuration is due to two factors. First, the configuration is already the final, saturated state of the nonlinear Parker instability, which is reached quasistatically during slow accretion (PM04; Mouschovias, 1974). Second, line-tying of the magnetic field at $r=R_{\ast}$ significantly changes the structure of the MHD wave spectrum in a way that enhances stability. Goedbloed & Halberstadt (1994) found that, in a homogenous plasma, a superposition of Alfvén- and magnetosonic waves is needed to satisfy the line-tying boundary conditions. As a consequence, the basic, unmixed MHD modes are not eigenfunctions of the linear force operator, and thus the spectrum is modified. The mass quadrupole moments, $Q_{ij}=\int d^{3}x^{\prime}\,(3x_{i}^{\prime}x_{j}^{\prime}-r^{\prime 2}\delta_{ij})\rho(\mathbf{x^{\prime}}),$ (7) of the mountain in Fig. 1 are plotted versus time in Fig. 2. We note first that $Q_{12}=0$ and $Q_{22}=-Q_{33}/2$, as expected for an axisymmetric system. We can compare Fig. 2 directly with the ellipticity $\epsilon\propto Q_{33}$ computed by Payne & Melatos (2006a).222Payne & Melatos (2006a) simulated a polar cap with $b=10$ as against $b=3$ in model A.. These authors found two dominant global modes, an Alfvén and an acoustic mode, claimed to be analogous to the fundamental modes of a gravitating, magnetized plasma slab. We cannot resolve the compressional modes in Fig. 2, but the latitudinal (Alfvén) mode is clearly visible through the oscillations in $Q_{22}$ and $Q_{33}$. Figure 2: Mass quadrupole moments for model A, normalised to the maximum of $Q_{33}$ ($1.3\times 10^{25}$ g cm2), as a function of time in units Alfvén time. We find $Q_{22}=-Q_{33}/2$ and $Q_{12}=0$, as expected for an axisymmetric configuration. The global Alfvén oscillation is damped by numerical viscosity. An oscillation cycle proceeds as follows. The first minimum of $Q_{33}$, and hence $\epsilon$, at $t=0.4\tau_{\mathrm{A}}$ in Fig. 2, corresponds to the top right panel of Fig. 1. The mountain withdraws radially and poleward. Polar field lines move closer to the magnetic pole, while equatorial field lines are drawn towards the equator. At $\tau=\tau_{\mathrm{A}}$, the mountain spreads and $Q_{33}$ reaches a maximum in Fig. 2. The damping observed in Fig. 2 arises solely from numerical dissipation; neither viscosity nor resistivity are included in our version of zeus-mp. ### 3.2 Outer boundary The “flaring up” of magnetic field lines at the pole, observed by PM07, is an artifact of the outflow boundary condition at $r=R_{m}$. In order to check this, we repeat the simulation of model A but switch to an inflow boundary condition (model B). The density distribution is similar in the two models, as is clear from Fig. 3. However, the inflow BC artifically pins the magnetic field to the outer boundary, introducing magnetic field discrepancies (mostly in the outer layers, where $\rho$ is negligible). Figure 3: Meridional section of model B, showing density (left) and projected magnetic field (right) at $t/\tau_{\mathrm{A}}=3.6$ (solid curves). The field lines are pinned to the outer $r$ boundary by the inflow boundary condition. For comparison, the density and magnetic field of model A are overplotted (dotted curves). Alfvén waves are similar to transverse waves on a string, where the magnetic tension provides the restoring force. Models A and B are therefore equivalent to a vibrating string with one end free and fixed respectively. We expect the oscillation frequency of the fundamental mode in model B to be twice that of model A. This is indeed observed in the oscillations of the quadrupole moments, displayed in Fig. 4: $Q_{22}$ and $Q_{33}$ in Fig. 4 oscillate at $0.6$ times the period in Fig. 2. A comprehensive analytic computation of the MHD spectrum, including discrete and continuous components, will be attempted in a forthcoming paper. Figure 4: Mass quadrupole moments for model B, normalised to the maximum of $Q_{33}$ ($1.3\times 10^{25}$ g cm2) as a function of time in units of the Alfvén time. We find $Q_{22}=-Q_{33}/2$ and $Q_{12}=0$, as expected for an axisymmetric configuration. The mountain performs damped lateral oscillations with twice the frequency of model A. ### 3.3 Uniform toroidal field There are strong theoretical indications that the magnetorotational instability (Balbus & Hawley, 1998) acts during core collapse supernova explosions to generate a substantial toroidal field component $B_{\phi}\sim B_{p}$ beneath the stellar surface (Cutler, 2002; Akiyama et al., 2003). The hydromagnetic stability of equilibria with $B_{\phi}\neq 0$ will be discussed thoroughly in a forthcoming paper. In this subsection, for completeness, we present the results of a preliminary investigation. Let us rerun model A with the same initial conditions while applying a uniform $B_{\phi}=10^{-7}B_{0}=0.35B_{p}$ throughout the integration volume and at $r=R_{\ast}$, where the poloidal field component is defined as $B_{p}=(B_{r}^{2}+B_{\theta}^{2})^{1/2}$, taken at the point $[\tilde{x}=(r-R_{\ast})/h_{0},\theta,\phi]=(10^{-3},0.012,0)$. $B_{\phi}$ is uniform only initially and is allowed to evolve nonuniformly as zeus-mp proceeds. This procedure leads to a non-equilibrium configuration, because we do not generalise and solve again the GS equation (2) to accomodate $B_{\phi}\neq 0$. Nevertheless, it provides us with some insight into the stability of a field with nonzero pitch angle. Figure 5: Model A repeated with the same parameters as in Table 1 including a uniform toroidal field $B_{\phi}=10^{-7}B_{0}=0.35B_{p}$. The mountain is defined by the orange isosurface $\rho(r,\theta,\phi)=1.04\times 10^{9}$ g cm3, while red denotes the neutron star surface $r=R_{\ast}$. In order to improve visibility, all length scales of the mountain and the field lines (blue) are magnified five-fold. The field exhibits a helical topology, which is most distinct in the polar flux tubes where the poloidal contribution is weakest. Figure 6: Mass ellipticity of model A with (solid) and without (dotted) a uniform $B_{\phi}$. The toroidal field component leads to a shorter oscillation period and to a higher saturation ellipticity. Fig. 5 displays the result of this numerical experiment after $t=7.4\tau_{A}$. The toroidal field component creates a helical field topology in the polar region far from the surface, where the poloidal field is comparably weak. In the equatorial region, however, the poloidal field is still dominant and the structure remains unchanged from Fig. 1. Remarkably, the toroidal field does not alter the stability of the system qualitatively, at least for the parameters of model A [We expect a stronger effect in other parameter regimes; see Lifschitz (1989); Goedbloed & Poedts (2004)]. The ellipticity, displayed in Fig. 6 (solid curve), exhibits characteristic oscillations with a period $\sim 30$ per cent smaller than that of the purely poloidal configuration (dotted curve), which can be explained simply by the increase in the Alfven speed. In addition, the saturation ellipticity is $\sim 3$ per cent higher than in model A; the magnetic tension increases with $B$, sustaining the mountain at a lower colatitude. ## 4 Nonaxisymmetric stability We turn now to the three-dimensional evolution of a magnetised mountain in zeus-mp. The chief finding, presented below, is that the initial (axisymmetric) configuration becomes unstable to toroidal perturbations, but that, after a brief transition phase, the system settles into a new (nearly axisymmetric) state, which is stable in the long term. Section 4.1 compares the results to the axisymmetric reference model A. The magnetic and mass multipole moments are computed in section 4.2, the influence of the boundary conditions is considered in section 4.3, and the component-wise evolution of the energy is examined in section 4.4. A scaling of the mass quadrupole moment versus $M_{a}$ is derived empirically in section 4.5. The curvature rescaling is verified in section 4.6. ### 4.1 General features Figure 7: Density and magnetic field of model D at $\tau/\tau_{\mathrm{A}}=0,1,2,3,4,5$ (from top left to bottom right). The mountain is defined by the orange isosurface $\rho(r,\theta,\phi)=1.04\times 10^{9}$ g cm3, while red indicates the neutron star surface $r=R_{\ast}$. In order to improve visibility, all length scales of the mountain and the field lines are magnified five-fold. The mountain becomes unstable to toroidal modes at $\tau\approx 0.8\tau_{\mathrm{A}}$. It subsequently relaxes to a new nonaxisymmetric equilibrium. The footpoint of the blue fieldlines is at the stellar surface while green fieldlines are traced starting from the equator. Green field lines eventually become topologically disconnected (see text). Figure 8: Evolution of the azimuthal magnetic field component $B_{\phi}$ at $(\tilde{x},\theta)=(10^{-3},0.01)$ in model D as a function of longitude $\phi$ (in radians) and time $t$ (in units of the Alfvén time.) Model D starts from the same configuration as model A ($M_{a}/M_{c}=1.0$, outflow at $r=R_{m}$) but is evolved in three dimensions. Six snapshots of a density isosurface (orange) and magnetic field lines (blue) are depicted in Fig. 7. At $\tau\approx 0.8\tau_{\mathrm{A}}=200\tau_{0}$, the system undergoes a violent transition. The field lines bend in the $\phi$ direction, indicating that the initial axisymmetric configuration is unstable to toroidal modes, a channel that is evidently not present in axisymmetric simulations. This hypothesis is supported by Fig. 8 which plots $B_{\phi}$ at $\tilde{x}=10^{-3}$ and $\theta=0.01$ as a function of $\phi$ and $t$. The magnetic field takes the form of an azimuthal travelling wave $B_{\phi}\sim\exp[\mathrm{i}(m\phi-\omega t)]$. From Fig. 8, we measure the phase speed to be approximately $v_{\mathrm{p}}=\omega R_{\ast}/m=27v_{\mathrm{A}}$, where the Alfvén speed $v_{\mathrm{A}}$ is measured at $(\tilde{x},\theta,\phi)=(10^{-3},0.01,0.1)$ and we assume $m=1$. An inhomogenous plasma generally supports mixed magnetosonic/Alfvén modes, so $v_{\phi}$ does not necessarily equal $v_{\mathrm{A}}$ or the fast/slow magnetosonic speed. The magnitude of $B_{\phi}$ is comparable to the magnitude of the polar magnetic field $B_{\mathrm{p}}=2.9\times 10^{-7}B_{0}$. The deviations of the mountain isosurface, defined by $\rho(r,\theta,\phi)=1.04\times 10^{9}$ g cm3, from axisymmetry are small ($\la 5$ per cent laterally and $\la 0.06$ per cent radially during the transition phase). The isosurface spreads outward by $\sim 32$ per cent relative to ites initial position. Figure 9: $|B_{\phi}|$ at $(\tilde{x},\theta,\phi)=(10^{-3},1.5,1.3)$ for model D, simulated at higher resolution ($N_{\phi}=32$, solid) and lower resolution ($N_{\phi}=8$, dashed). Small wavelength perturbations grow faster. We track the absolute value of $|B_{\phi}|$ in order to isolate better the dominant mode, as the system exists in a superposition of stable and unstable modes, and $B_{\phi}$ switches sign. Fig. 9 demonstrates how the instability grows. We plot $|B_{\phi}|$ at the (arbitrary) position $(r,\theta,\phi)=(R_{\ast},1.5,1.3)$ versus time. The solid curve corresponds to a higher toroidal resolution ($N_{\phi}=32$ grid cells in $\phi$ direction) than the dashed curve ($N_{\phi}=8$). We note first that $|B_{\phi}|$ grows exponentially with time, as expected in the linear regime. The growth rate is measured to be $\Gamma=\mathrm{Im}(\omega)=0.05\tau_{0}^{-1}=12.5\tau_{\mathrm{A}}^{-1}$. Second, the instability is manifestly associated with toroidal modes. The magnetic perturbation, $\delta\mathbf{B}$, induced by a linear Lagrangian displacement $\mn@boldsymbol{\xi}$ is $\delta\mathbf{B}=\nabla\times(\mn@boldsymbol{\xi}\times\mathbf{B})$. By writing out the vector components, one sees that $\delta B_{\phi}\neq 0$ implies $\xi_{\phi}\neq 0$, provided the unperturbed field has the form $\mathbf{B}=B_{r}(r,\theta)\hat{\mathbf{e}}_{r}+B_{\theta}(r,\theta)\hat{\mathbf{e}}_{\theta}$. The dashed curve in Fig. 9 tracks $|B_{\phi}|$ for a simulation carried out at a lower resolution ($N_{\phi}=8$). The instability grows significantly slower with $\Gamma=\mathrm{Im}(\omega)=0.02\tau_{0}^{-1}=5\tau_{\mathrm{A}}^{-1}$. We conclude that $\Gamma$ scales with the wavelength $\lambda$ of the perturbation roughly as $\lambda^{-1/2}$. The piecewise-straight appearance of the dashed curve in Fig. 9 shows that the global oscillations are governed by a superposition of unstable (growing) and stable wave modes. What type of instability is at work here? In order to answer that question, we first write down the change in potential energy associated with a Lagrangian displacement $\mn@boldsymbol{\xi}$ in a form that reveals the physical meaning of the different contributions (Biskamp, 1993; Lifschitz, 1989; Greene & Johnson, 1968): $\displaystyle\delta W_{p}$ $\displaystyle=$ $\displaystyle\frac{1}{2}\int\mathrm{d}V\left[|\mathbf{Q}_{\perp}|^{2}+|\nabla\mn@boldsymbol{\cdot}\mn@boldsymbol{\xi}_{\perp}+2\mn@boldsymbol{\kappa}\mn@boldsymbol{\cdot}\mn@boldsymbol{\xi}_{\perp}|^{2}B_{0}^{2}\right.$ (8) $\displaystyle+c_{s}^{2}\rho|\nabla\mn@boldsymbol{\cdot}\mn@boldsymbol{\xi}|^{2}-j_{\parallel}(\mn@boldsymbol{\xi}_{\perp}^{\ast}\times\mathbf{b})\mn@boldsymbol{\cdot}\mathbf{Q}$ $\displaystyle\left.-2(\mn@boldsymbol{\xi}_{\perp}\mn@boldsymbol{\cdot}\nabla p)(\mn@boldsymbol{\kappa}\mn@boldsymbol{\cdot}\mn@boldsymbol{\xi}_{\perp}^{\ast})-(\mn@boldsymbol{\xi}^{\ast}\mn@boldsymbol{\cdot}\nabla\varphi)\nabla\mn@boldsymbol{\cdot}(\rho\mn@boldsymbol{\xi})\right],$ We include the term due to gravity (Goedbloed & Poedts, 2004) and define $\mathbf{j}=\mu_{0}^{-1}\nabla\times\mathbf{B}$ (current density), $\mathbf{Q}=\nabla\times(\mn@boldsymbol{\xi}\times\mathbf{B})$ (change in $\mathbf{B}$ as a response to $\mn@boldsymbol{\xi}$), $\mn@boldsymbol{\kappa}=(\mathbf{b}\mn@boldsymbol{\cdot}\nabla)\mathbf{b}$ (field line curvature), and $\mathbf{b}=\mathbf{B}/B$. The subscripts $\perp$, $\parallel$ refer to the magnetic field, such that $\mn@boldsymbol{\alpha}_{\perp}=\mn@boldsymbol{\alpha}-(\mn@boldsymbol{\alpha}\mn@boldsymbol{\cdot}\mathbf{b})\mathbf{b}$ and $\alpha_{\parallel}=\mn@boldsymbol{\alpha}\mn@boldsymbol{\cdot}\mathbf{b}$. The first three (stabilising) terms are the potential energy of the shear Alfvén mode, the fast magnetosonic mode, and the (unmagnetized) sound mode. They are all positive definite. The last three terms may have either sign. The term proportional to $j_{\parallel}$ causes the current-driven instabilities, the curvature term causes pressure-driven instabilities (when $\mn@boldsymbol{\kappa}\mn@boldsymbol{\cdot}\nabla\rho>0$), and the final term causes gravitational instabilities. We first note that $j_{\parallel}=0$, ruling out current-driven instabilities. Furthermore, we have $\mn@boldsymbol{\kappa}=\kappa_{r}(r,\theta)\hat{\mathbf{e}}_{r}+\kappa_{\theta}(r,\theta)\hat{\mathbf{e}}_{\theta}$ in our particular field geometry. In principle, this term admits pressure- driven instabilities. However, we would expect such instabilities, if they exist, to also grow in an axisymmetric system, yet they do not. This suggests that the instability we see in Figs. 7–9 is associated with a toroidal dependence in $\mn@boldsymbol{\xi}$, leaving the gravitational term, which indeed contains $\partial_{\phi}\mn@boldsymbol{\xi}$ contributions. One prominent gravitational mode is the Parker or magnetic buoyancy instability (PM04; Mouschovias, 1974). Its physics was elucidated by Hughes & Cattaneo (1987) for a plane-parallel, stratified atmosphere with a horizontal field increasing with depth $z$. The instability involves an interchange sub- mode and an undular sub-mode. The interchange sub-mode satisfies $k_{y}=\xi_{y}=0$. We do not observe this mode in our system because (i) it should also be present in two dimensions, as it does not rely on a toroidal dependence, yet it is absent; and (ii) it is inconsistent with the line-tying boundary condition at $r=R_{\ast}$. On the other hand, undular modes compress the plasma along field lines, even in systems which are interchange stable. In two dimensions, they are restricted to $k_{x}=Q_{x}=0$, whereas a non- vanishing $Q_{x}$ is allowed in three dimensions. Hughes & Cattaneo (1987) showed that $\delta W_{p}$ is minimized for $k_{x}\rightarrow\infty$, consistent with the results in Fig. 9; the instability grows faster, if we allow smaller wavelength perturbations by increasing $N_{\phi}$. When $v_{A}$ is uniform the growth rate of the Parker instability reaches an asymptotic maximum $\Gamma_{\mathrm{P}}\simeq(g/\Lambda)^{1/2}$ for $k_{x}\Lambda,k_{y}\Lambda\gg 1$. Here, $\Lambda=v_{\mathrm{A}}^{2}/g$ is the scale height for a stratified atmosphere with uniform gravitational acceleration $g$. We recognize $(\Lambda/g)^{1/2}$ as the characteristic free fall time over one scale height. In the units specified in section 2.3, we find $\Gamma_{\mathrm{P}}\tau_{0}\simeq\Lambda\simeq g\simeq 1$, two orders of magnitude higher than the observed growth rate $\Gamma\simeq 10^{-2}\tau_{0}^{-1}$. The discrepancy arises because the Parker instability cannot grow freely in the belt region, since the adjacent plasma at higher latitudes effectively acts as a line-tying boundary for the magnetic field. The snapshot at $t=\tau_{A}$ (top-middle panel in Fig. 7) demonstrates how the instability starts in the equatorial region, whose magnetic belt represents the endpoint of the two-dimensional Parker instability. The undular Parker sub-mode releases gravitational energy by radial plasma flow towards the neutron star’s surface. At the same time, the magnetic field is rearranged such as to minimize the (radial) gradient in $\mathbf{B}$. Importantly, the undular mode is not available in the axisymmetric case. The extra degree of freedom in the $\phi$ direction allows perturbations to develop which do no work against the magnetic pressure, destabilising the belt region. Particularly interesting here is the formation of topologically disconnected field lines (green curves in Fig. 7). These occur when field lines are pushed out of the radial boundary surface. They are then disrupted and can subsequently reconnect at the equatorial boundary, forming O-type neutral points (“bubbles”) and associated Y-type points. It is important to note, however, that the formation of these bubbles is not an unphysical boundary effect. Instead, in a realistic setting, even a small resistivity leads to reconnection and thus to a topological rearrangement of the field. The effect is similar to the formation of plasmoids (Schindler et al., 1988). $B_{r}$ switches sign at the magnetic equator, implying the existence of a current sheet. Reconnection then leads to the creation of magnetic X-type neutral points and the associated bubbles. We discuss resistive effects and the importance of these bubbles to resistive instabilities in an accompanying paper. Figure 10: Magnetic pitch angle $B_{\phi}/B_{p}$ (right panel) as a function of the coordinate $\eta$ along four magnetic field lines 1–4 for model D, for a snapshot taken at $t=5\tau_{\mathrm{A}}$. The positions of the field lines are depicted in the left panel. The mountain is defined by the orange isosurface $\rho(r,\theta,\phi)=1.04\times 10^{9}$ g cm3, while red denotes the neutron star surface $r=R_{\ast}$. In order to improve visibility, all length scales of the mountain and the field lines (blue) are magnified five- fold. The concept of rational magnetic surfaces, where the field lines close upon themselves, plays an important role in a local plasma stability analysis. The bending of field lines as a result of a Lagrangian displacement is associated with an increase in potential energy. Hence, for almost all instabilities to occur, this contribution, which can be expressed as $\mathbf{B}\mn@boldsymbol{\cdot}\nabla\mn@boldsymbol{\xi}$, needs to be small. In a tokamak geometry, it can be shown that this term vanishes on a rational surface. The spatial location of rational surfaces is directly related to the pitch angle $B_{\phi}/B_{p}$. We defer a detailed analysis of the rational magnetic surfaces in our problem to a forthcoming paper and restrict ourselves to a brief discussion in the following paragraph. In Fig. 10, we plot the pitch angle as a function of the coordinate $\eta$ along the field line (right panel) for four different field lines (left panel) in model D. We first note that the toroidal component stays below 20 per cent of the poloidal component along all four field lines. Furthermore, the absolute magnitude of the pitch angle tends to increase with colatitude. This is consistent with the previous discussion. The Parker instability (and hence $B_{\phi}$) dominates close to the magnetic equator. The wave-like character of the instability is vividly demonstrated by the zero crossings of the pitch angle. ### 4.2 Mass and magnetic quadrupole moments Figure 11: Mass quadrupole moments for model D, normalised to the maximum of $Q_{33}$ ($1.30\times 10^{25}$ g cm2) as a function of time in units of the Alfvén time. The sytem develops a substantial asymmetry, characterised by the off-diagonal element $Q_{12}$, during the relaxation phase, before settling down to a nearly axisymmetric state. At $t\approx 2\tau_{\mathrm{A}}$, the system in Fig. 7 settles down to a stable state which differs from the initial configuration, primarily by being nonaxisymmetric with respect to the pre-accretion magnetic axis. The field lines whose footpoints are at a low colatitude move towards the magnetic poles. This behaviour is reflected in the mass quadrupole moments, plotted against time in Fig. 11. The transition to a nonaxisymmetric magnetic field configuration at $t\approx\tau_{A}$ is accompanied by a sudden rise in the off-diagonal moment $Q_{12}$. However, by the time the mountain settles down at $t\approx 2\tau_{A}$, axisymmetry is largely restored and $Q_{12}$ decreases. Fig. 11 shows that $Q_{12}$ oscillates before damping down, with a remarkably low deviation from axisymmetry of $Q_{12}/Q_{33}<0.1$ per cent in the final state. We reiterate that the final state is not the same as the initial state, even though it is nearly axisymmetric. Furthermore, the final state is stable. This is the main result of the paper, as far as astrophysical applications are concerned. Figure 12: Meridional section for model D at $t/\tau_{A}=0,1,2,3,4,5$ (top left to bottom right). Shown are the density contours (dashed curves) for $\log_{10}(\rho/\rho_{0}^{\prime})=-13,-12,-11,-10.7,-10.5,-10.3$, and the normalised Lorentz force per unit volume $(\mathbf{J}\times\mathbf{B})_{\phi}/|\mathbf{J}\times\mathbf{B}|$ (solid curves) for the values 0.1, 0.5, 0.9. The Lorentz force develops a toroidal component as $B_{\phi}$ increases, but its poloidal component diminishes, allowing the poloidal pressure gradient to push the mountain equatorwards. Why does $Q_{ij}$ decrease? Naively, one would not expect a significant change, given that the Parker instability predominantly acts in the equatorial belt region, while most of the plasma is located at the magnetic pole. The answer can be found in the Lorentz force, which balances the lateral pressure gradient. Fig. 12 shows how the relative strength of the toroidal component of the Lorentz force, $(\mathbf{J}\times\mathbf{B})_{\phi}/|\mathbf{J}\times\mathbf{B}|$ (dashed curve), grows as a function of time in model D. As $B_{\phi}$ grows, following the onset of the instability, the force per unit volume develops a toroidal component while its lateral component decreases. Hence the (approximately unchanged) lateral hydrostatic pressure gradient forces the mountain to slip towards the equator. After the system settles down, $B_{\phi}$ decreases and the lateral components of $\mathbf{J}\times\mathbf{B}$ and $\nabla p$ readjust to balance each other, leading to the stable equilibrium state. The ellipticity $\epsilon\propto Q_{22}$ reaches a local maximum during the transition phase at $t\approx\tau_{A}$ and subsequently drops. The mass quadrupole moment of the final configuration is $\approx 33$ per cent lower than in model A. The asymptotic values of $Q_{ij}$ for the eight models in Table 1 are tabulated in Table 2, normalized to $Q_{33}(t=8\tau_{\mathrm{A}})$ for model A. Table 2: Asymptotic values of $Q_{ij}$ for the eight models in Table 1, normalized to $\hat{Q}_{33}=Q_{33}(t=8\tau_{\mathrm{A}})$ for model A. We select $t=8\tau_{\mathrm{A}}$ (models A–E), $t=5\tau_{\mathrm{A}}$ (models F & G), and $t=4\tau_{\mathrm{A}}$ (models J & K) to compute the asymptotic value. Model | $Q_{12}/\hat{Q}_{33}$ | $Q_{22}/\hat{Q}_{33}$ | $Q_{33}/\hat{Q}_{33}$ ---|---|---|--- A | $-3.1\times 10^{-8}$ | $-0.50$ | $1.00$ B | $-3.1\times 10^{-8}$ | $-0.51$ | $1.02$ D | $1.6\times 10^{-3}$ | $-0.21$ | $0.64$ E | $-3.1\times 10^{-8}$ | $-0.51$ | $1.02$ F | $-1.9\times 10^{-3}$ | $-0.23$ | $0.60$ G | $8.7\times 10^{-4}$ | $-0.17$ | $0.68$ J | $5.7\times 10^{-2}$ | $-1.27$ | $3.9$ K | $2.5\times 10^{-4}$ | $-4.72$ | $12$ Figure 13: Magnetic dipole moment $d_{10}/R_{m}^{3}$ (bottom) and magnetic quadrupole moment $d_{21}/R_{m}^{4}$ (top, middle) for model D, normalised to the initial value of $d_{10}/R_{m}^{3}=5.29\times 10^{-7}B_{0}$, as a function of time (in units of the Alfvén time scale). All other components vanish due to symmetry. The nonvanishing components of the magnetic dipole and quadrupole moments $d_{lm}(r=R_{m})$, defined as $d_{lm}(r=R_{m})=R_{m}^{l+2}\int\mathrm{d}\Omega\;Y_{lm}^{\ast}B_{r}$ (9) (see appendix C), are displayed as functions of time in Fig. 11. The dipole moment $d_{10}=4(\pi/3)^{1/2}\mu$ increases rapidly during the transition phase, reaching an asymptotic maximum of 5.5 times the initial value. Likewise, the quadrupole $d_{21}$ peaks during the transition phase before settling down to a constant value. The final field is highly axisymmetric, deviating from perfect symmetry by $|d_{21}|R_{m}/d_{10}=0.8$ per cent. The asymptotic values of $d_{lm}$ for models A–K are listed in Table 3. Table 3: Non-vanishing components of the asymptotic magnetic dipole moments $d_{10}/R_{m}^{3}$ and magnetic quadrupole moments $q_{10}/R_{m}^{4}$, both normalised to $\hat{d}_{10}/R_{m}^{3}=d_{10}(t=8\tau_{\mathrm{A}})/R_{m}^{3}$. We select $t=8\tau_{\mathrm{A}}$ (models A–E), $t=5\tau_{\mathrm{A}}$ (models F & G), and $t=4\tau_{\mathrm{A}}$ (models J & K) to compute the asymptotic value. Model | $d_{10}/\hat{d}_{10}$ | $\mathrm{Re}(d_{21})/(\hat{d}_{10}R_{m})$ | $\mathrm{Im}(d_{21})/(\hat{d}_{10}R_{m})$ ---|---|---|--- A | $1.0$ | $-4.6\times 10^{-8}$ | $-4.8\times 10^{-9}$ B | $1.3$ | $-5.9\times 10^{-8}$ | $-6.1\times 10^{-9}$ D | $7.4$ | $4.8\times 10^{-2}$ | $-4.3\times 10^{-2}$ E | $1.3$ | $-5.9\times 10^{-8}$ | $-6.1\times 10^{-9}$ F | $7.3$ | $1.1\times 10^{-1}$ | $8.2\times 10^{-3}$ G | $7.1$ | $2.1\times 10^{-2}$ | $-5.5\times 10^{-2}$ J | $21$ | $-1.2$ | $-0.12$ K | $57$ | $-6.3\times 10^{-4}$ | $9.1\times 10^{-4}$ ### 4.3 Boundary conditions Figure 14: Meridional section of density contours (left) and magnetic field lines (right) for model D (solid curve) and model E (dotted curve). The inflow boundary condition corresponds to line-tying at the outer boundary. Deviations between the two models occur in the outermost, low density regions. We perform a simulation (model E) with the same initial configuration as model D ($M_{a}/M_{c}=1.0$ and $b=3$) but with inflow boundary conditions at $R=R_{m}$. Fig. 14 compares the density (left panel) and magnetic field (right panel) of models D and E. Again, inflow pins the magnetic field at the outer boundary, as opposed to outflow, which leaves the field free. The density distribution is almost unaffected. The magnetic field is mainly affected in the outermost region, where the plasma density is low. The overall time evolution (a nonaxisymmetric transition phase which leads to a nearly axisymmetric equilibrium) remains as before, too. We therefore conclude that the outer boundary condition can be chosen opportunistically. By contrast, the inner boundary condition contributes fundamentally to stability. The tension of the magnetic field, which is tied to the stellar surface, suppresses those modes which are driven by a pressure gradient perpendicular to the magnetic flux surfaces, such as the interchange and ballooning mode. If line-tying is taken away, the latter modes disrupt the mountain in short order. If we rerun model A (for example) by applying a reflecting boundary condition at $r=R_{\ast}$, the mountain rapidly dissolves on a timescale $\sim\tau_{0}$. The same experiment for model D results in high velocities and steep field gradients, causing the numerical algorithm of zeus- mp to break down. ### 4.4 Energetics Mouschovias (1974) showed that an isothermal, gravitating, MHD system possesses a total energy $W$, which can be written as the sum of gravitational, kinetic, magnetic, and acoustic contributions, defined by the following volume integrals, evaluated over the simulation volume: $W_{\mathrm{g}}=\int\mathrm{d}V\rho\varphi,$ (10) $W_{\mathrm{k}}=\frac{1}{2}\int\mathrm{d}V\rho v^{2},$ (11) $W_{\mathrm{m}}=\frac{1}{2\mu_{0}}\int\mathrm{d}VB^{2},$ (12) $W_{\mathrm{a}}=\int\mathrm{d}Vp\log p.$ (13) Here, $\mathbf{v}$ is the plasma velocity and $p=c_{s}^{2}\rho$ is the pressure. Figure 15: The evolution of the total energy $W$ and its components $W_{\mathrm{m}}$, $W_{\mathrm{g}}$, $W_{\mathrm{k}}$, and $W_{\mathrm{a}}$ (top to bottom) for model D, all normalised to $W_{0}=2.2\times 10^{36}$ erg, as a function of time (in units of the Alfvén time scale). The total energy increases artifically, due to mass loss through the outer boundary (see text). The evolution of (10)–(13) for model D is shown in Fig. 15. The magnetic energy (second panel from top) steadily decreases to $20$ per cent of its original value, as the axisymmetric equilibrium evolves to a lower energy, nonaxisymmetric state. The kinetic energy peaks at $t=1.2\tau_{A}$, during the transition phase when the magnetic reconfiguration occurs. However, the gravitational and acoustic contributions, which dominate $W$, increase with time. The reason for this becomes apparent if we track the total mass in the simulation volume. Approximately 3.7 per cent of the mass is lost through the outflow boundary at $r=R_{m}$ by $t=6\tau_{\mathrm{A}}$. The mass loss is responsible for the increase of $W_{\mathrm{a}}\propto\rho^{2}$ and $W_{\mathrm{g}}\propto\rho$, both of which are negative ($W_{\mathrm{g}}$ because the plasma is gravitationally bound and $W_{\mathrm{a}}$ since $\rho<1$ in our units). Figure 16: The evolution of the total energy $W$ and its components $W_{\mathrm{m}}$, $W_{\mathrm{g}}$, $W_{\mathrm{k}}$, and $W_{\mathrm{a}}$ (top to bottom) for model D, all normalised to $W_{0}=2.2\times 10^{36}$ erg, as a function of time (in units of the Alfvén time scale). $W$, $W_{\mathrm{g}}$, and $W_{\mathrm{a}}$ are now corrected for the mass loss through the outer boundary (cf. Fig. 15). Let us try to correct for the mass loss by multiplying $W_{g}$, $W_{k}$, and $W_{a}$ by $M(t=0)/M(t)$, where $M(t)$ is the mass in the simulation volume at time $t$. The result is presented in Fig. 16. $W_{g}$ and $W_{a}$ now decrease, and the total energy, $W=W_{\mathrm{g}}+W_{\mathrm{k}}+W_{\mathrm{m}}+W_{\mathrm{a}}$, decreases by just 2.5 per cent. Figure 17: The evolution of the total energy $W$ and its components $W_{\mathrm{m}}$, $W_{\mathrm{g}}$, $W_{\mathrm{k}}$, and $W_{\mathrm{a}}$ (top to bottom) for model E, all normalised to $W_{0}=2.2\times 10^{36}$ erg, as a function of time (in units of the Alfvén time scale). The approximate correction above assumes $\rho$ decreases uniformly, which is not strictly true. We therefore check our claim that mass loss is responsible by tracking the energy evolution of model E, which has the same initial configuration as model D, but an inflow outer boundary which blocks mass loss. From Fig. 17, it is clear that the total energy rises then falls, consistent with the observed dynamical evolution. The mountain oscillates until toroidal modes grow sufficiently to disrupt the initial configuration and force it into a nonaxisymmetric state. There is no spurious increase in $W$. We conclude that mass loss through the outer boundary is indeed responsible for the observed behaviour of $W$ in model D in Fig. 15. ### 4.5 Dependence on $M_{a}$ Figure 18: The mass quadrupole moments for models F (solid), D (dotted), and G (dashed), normalised to $1.33\times 10^{25}$ g cm2, as a function of time. While all models show similar dynamical behaviour, the quadrupole moment of the final state increases with $M_{a}$. Does the final, nonaxisymmetric configuration of the mountain become unstable once the accreted mass exceeds a critical threshold? There are two ways that this can happen. First, the sequence of nonaxisymmetric GS equilibria passed through as $M_{a}$ increases can terminate above a critical value of $M_{a}$; i.e. there is a loss of equilibrium. PM07 observed this phenomenon in axisymmetric magnetic mountains with $M_{a}\ga 10^{-4}M_{\odot}$, when the source term in the GS equation forces the flux function outside the range $0\leq\psi\leq\psi_{\ast}$ permitted by the boundary condition at $r=R_{\ast}$. Second, the nonaxisymmetric state reached in Fig. 7 (for example) may be metastable. That is, it may be a local energy minimum which can be reached from an axisymmetric starting point via the Parker instability but which the system can exit (in favor of some other, global energy minimum) if the system is kicked hard enough. One way to kick the system hard is to increase $M_{a}$ substantially. We are not really in a position to answer this question definitively, because the GS fails to converge to valid equilibria for $M_{a}\gg 10^{-4}M_{\odot}$, due to numerical difficulties (steep gradients, which would be smoothed in a more realistic, non-ideal-MHD simulation). Nevertheless, we begin to address the issue by performing two simulations, models F and G, with the same parameters as model D but with lower and higher masses viz. $M_{a}/M_{c}=0.6$ and $M_{a}/M_{c}=1.4$ respectively. The mass quadrupole moments are plotted versus time in Fig. 18. The solid and dashed curves are for models F and G respectively, with model D (dotted curve) overplotted for comparison. The dynamical behaviour of all three models is similar: a violent transition phase which settles down to a nonaxisymmetric state. However, the start of the transition phase, defined as the instant where $Q_{33}$ is maximized, scales roughly $\propto 0.5M_{a}/M_{c}$ in units of $\tau_{A}$. Physically, this means that the onset of the toroidal instability depends on $M_{a}$. We can understand the trend in terms of the Parker instability (section 4.1), whose growth rate scales as $\Gamma_{\mathrm{P}}\propto v_{\mathrm{A}}^{-1}$. By measuring $v_{A}$ at $\theta=\pi/2$ in models D,F, and G, we find $v_{A}\propto M_{a}$ empirically and $\Gamma_{\mathrm{P}}\propto M_{a}^{-1}$, consistent with the Parker scalings. Figure 19: Ellipticity $\epsilon$ for models F (solid), D (dotted), and G (dashed) as a function of time. As expected, $\epsilon$ increases with $M_{a}$. The evolution of the ellipticity $\epsilon\propto Q_{33}$ for models D, F, and G is displayed in Fig. 19. Of chief interest here is the ellipticity of the final state. It increases along with $M_{a}$, consistent with Melatos & Payne (2005). A linear fit yields the following rule of thumb for our _downscaled_ star (section 2.3): $\frac{\epsilon}{10^{-10}}=1.12\frac{M_{a}}{M_{c}}.$ (14) Note, however, that the fit is valid in the range $0.6\leq M_{a}/M_{c}\leq 1.4$. Numerical difficulties prevent us from extending it to larger values of $M_{a}$. Payne & Melatos (2006a) found $\epsilon/10^{-10}=7.8M_{a}/M_{c}(1+1.1M_{a}/M_{c})^{-1}$ in the $M_{a}\sim M_{c}$ regime for the axisymmetric equilibrium. Equation (14) yields values roughly 70 per cent lower than the latter formula. Figure 20: Azimuthal magnetic field component $|B_{\phi}|$ for models F (solid), D (dotted), and G (dashed), in units of $B_{0}$, plotted as a function of time, in units of the Alfvén time. For completeness, we plot the magnitude of the toroidal field component $|B_{\phi}|$ versus time in Fig. 20. Interestingly, the peak value is achieved for the intermediate mass model, D, not for model G. However, $B_{\phi}$ in the final state depends weakly on $M_{a}$. We find $B_{\phi,\mathrm{F}}=3.4\times 10^{-7}B_{0}$, $B_{\phi,\mathrm{D}}=4.5\times 10^{-7}B_{0}$, and $B_{\phi,\mathrm{G}}=3.2\times 10^{-7}B_{0}$, where $B_{0}$ is defined in section 2.3. These values are comparable to the magnitude of the polar magnetic field $B_{\mathrm{p}}=2.9\times 10^{-7}B_{0}$. ### 4.6 Dependence on curvature As discussed in section 2.3, PM07 argued that reducing $R_{\ast}$ and $M_{\ast}$ does not affect the equilibrium structure as long as $h_{0}$ remains constant, at least in the small-$M_{a}$ limit. To test whether this also holds for the dynamical behaviour of the system, we perform two runs, models J and K, with $a=75$ and $a=100$ respectively. Figure 21: Mass quadrupole moments $Q_{ij}$ for model J ($a=75$, solid curve) and model K ($a=100$, dashed curve), plotted as a function of time (in units of the Alfvén time). The scale for model J (K) appears on the left (right) vertical axis. Although the transition phase is less distinct than in Fig. 11, these models basically share the same dynamics as lower curvature runs. Fig. 21 plots $Q_{ij}$ versus time for these models. The transition phase is more gradual than model D (Fig. 11). Again, however, $Q_{12}$ rises significantly, marking a deviation from axisymmetry. Melatos & Payne (2005) found $\epsilon\propto a^{2}$ analytically in the small-$M_{a}$ regime, so we fit a parabola to the simulation data (for $M_{a}=M_{c}$): $\frac{\epsilon}{10^{-13}}=1.82a^{2}.$ (15) A realistic star has $a=1.9\times 10^{4}$ (cf. section 2.3). Extrapolating (15), we find $\epsilon=6.6\times 10^{-5}$. (An ellipticity this large is close to the upper limit inferred from existing gravitational-wave nondetections; see section 6 for more details.) However, it should be remembered that equation (15) is an overestimate, because the computations in this paper neglect nonideal MHD effects. ## 5 Global MHD oscillations In this section, we explore the natural oscillation modes of a nonaxisymmetric magnetic mountain. We do this by loading the final state from models D, F, and G into zeus-mp and setting $\mathbf{v}=0$ on the whole grid. This procedure introduces numerical perturbations that are sufficient to excite small linear oscillation modes, albeit an uncontrolled distribution thereof. We then compute the power spectrum $P[S](\omega)=\left|\frac{1}{N}\sum_{i=0}^{N-1}S(t_{i})\mathrm{e}^{-i\omega t_{i}/N}\right|^{2}$ (16) by evaluating the discrete Fourier transform of the scalar function $S(t)$ [e.g. $B_{r}(t)$] at $N$ sample times $t_{i}$. In order to explore the magnetic modes, we examine $B_{r}$, $B_{\theta}$, and $B_{\phi}$. We choose one point on the grid where the amplitude of the oscillations is high, namely $(r,\theta,\phi)=(50.03h_{0},0.26,2.06)$ and compute $P[B_{r}]$, $P[B_{\theta}]$, and $P[B_{\phi}]$. The results are displayed in Fig. 22. We can distinguish five different spectral peaks at $10^{3}\omega\tau_{0}=0.78,1.1,1.2,2.1,3.2$, which are more or less distinct for the different components. Figure 22: Power spectrum of $B_{r}$, $B_{\theta}$, and $B_{\phi}$ (top to bottom) for model D (arbitrary units), plotted as a function of angular Fourier frequency (in units of $\tau_{0}^{-1}$). For a magnetized gravitating slab in a plane-parallel geometry, one can distinguish three different MHD modes (Goedbloed & Poedts, 2004): slow magnetosonic, Alfvén, and fast magnetosonic. Each mode consists of a discrete set of eigenmodes and a continuous spectrum, which are clearly separated. Unfortunately, such clean separation cannot be expected for a highly inhomogenous plasma in spherical geometry. Generally, different parts of the spectrum overlap or degenerate into a single point in a nontrivial way. We therefore restrict the discussion below to some qualitative remarks. The MHD spectrum contains genuine singularities, when the eigenfrequency coincides with the Alfvén or slow magneto-sonic frequency at some location within the magnetic mountain. In this case, the boundary value problem becomes singular; the boundary conditions can be fulfilled for a continuous range of frequencies. The singular frequencies depend on the components of the wave vector perpendicular to the direction of inhomogenity. It is unclear whether the band $\omega<0.002\tau_{0}^{-1}$, which looks “filled” in Fig. 22, belongs to the continuous part of the spectrum or else is an artifact of the nonzero line width from numerical damping (which can be estimated from the sample times $t_{i}$ to be $\sim 1.2\times 10^{-4}\omega\tau_{0}$). We do not observe any singular behaviour in the field variables, but we note that singularities would be suppressed by the shock- capturing algorithm (i.e. the artificial viscosity) in zeus-mp. We conclude that the features in Fig. 22 are probably discrete lines. Figure 23: Power spectra of $B_{r}$ and $B_{\theta}$ (top to bottom) for model A (arbitrary units), plotted as a function of angular Fourier frequency (in units of $\tau_{0}^{-1}$). Let us compare these results to the spectrum of the axisymmetric model A (Fig. 23). We first note that the Alfvén frequency $\omega_{A}=0.018\tau_{0}^{-1}(M_{a}/M_{c})^{-1/2}$ and acoustic frequency $\omega_{s}=0.48\tau_{0}^{-1}$ found by PM07 are outside the range of this plot, which is set by the Nyquist frequency $\omega_{\mathrm{N}}=(4\pi N\Delta t)^{-1}$ ($\Delta t=50\tau_{0}$ for model A and $\Delta t=10\tau_{0}$ for models D–G). Here, we are restricted to low frequency oscillations which are generally associated with global magnetic modes. Most distinct is the peak at $10^{3}\omega\tau_{0}=0.5$, which is not visible in Fig. 22. This long wavelength poloidal mode is suppressed in favor of toroidal modes in the three-dimensional configuration. However, the small peak at $10^{3}\omega\tau_{0}=1.1$ is present in both systems. This example illustrates vividly how relaxing the axisymmetric constraint leads to a different MHD spectrum. Figure 24: Power spectra of $B_{r}$, $B_{\theta}$, and $B_{\phi}$ (top to bottom) for models F (solid curve) and G (dashed curve) in arbitrary units, plotted as a function of angular Fourier frequency in units of $\tau_{0}^{-1}$. We overplot the spectrum of model G, stretched by a factor of 1.4 in $\omega$, as dotted curve, by way of comparison. Fig. 24 shows the power spectrum for models F (solid) and G (dashed). The most distinct peaks are again concentrated in the low frequency region. We can roughly match the peaks of models G and F by stretching the former spectrum by a factor of 1.4 in frequency. The higher $M_{a}$ equilibrium has a similar structure, but the Alfvén timescale is lower because the plasma density is 85 per cent higher. A complete analytic determination of the discrete and continuous components of the MHD spectrum via a full linear mode analysis will be attemped in a forthcoming paper. ## 6 Discussion Magnetically confined mountains on accreting neutron stars screen the magnetic dipole moment of the star. Potentially, therefore, the process of polar magnetic burial can explain the observed reduction of $\mu$ with $M_{a}$ in neutron stars with an accretion history. However, before magnetic burial can be invoked as a viable explanation, the question of stability must be resolved. In this article, we concentrate on the important aspect of three- dimensional stability, deferring resistive processes to future work (especially the issue of resistive g-modes333J. Arons, private communication). We find that the axisymmetric configurations in PM04 are susceptible to the three-dimensional magnetic buoyancy instability. The instability proceeds via the undular submode, with growth rate $\propto\lambda^{1/2}$, limited by the toroidal grid resolution. However, instead of breaking up and reverting to an isothermal atmosphere threaded by a dipolar magnetic field, the magnetic field reconfigures (over a few Alfvén times) and settles down into a new nonaxisymmetric equilibrium which is still highly distorted. Just as the axisymmetric solutions in PM04 are the final saturated states of the nonlinear evolution of the Parker instability in two dimensions, we find here the three- dimensional equivalent. This surprising result is the main conclusion of the paper. It holds irrespective of the outer boundary condition and curvature rescaling factor, but it depends critically on the line-tying boundary condition at the stellar surface. The final state is predominantly axisymmetric, with $1.5\leq|Q_{12}/Q_{33}|/10^{-3}\leq 3.2$ for models D, F, and G ($0.6\leq M_{a}/M_{c}\leq 1.4$). The ellipticity for model G reaches $1.6\times 10^{-10}$ in the downscaled star. Figure 25: Amplitude of the gravitational wave signal $h_{0}$ for $M_{a}/M_{\odot}=10^{-9},10^{-8},10^{-7},10^{-6},10^{-5},10^{-4},10^{-3}$, for the axisymmetric (solid lines) and nonaxisymmetric equilibrium (dashed lines). The sensitivities of Initial and Advanced LIGO, assuming 14 days coherent integration, are also plotted (upper and lower curves respectively). The growth of the mountain is arrested for $M_{a}\ga 1.2\times 10^{-5}M_{\odot}$ (light shaded region), due to Ohmic dissipation (Melatos & Payne, 2005; Vigelius & Melatos, 2008), while the right-hand edge is excluded at present because no accreting millisecond pulsars have been discovered with $f_{\ast}>0.7$ kHz (dark shaded region). The stability of magnetic mountains is important for the emission of gravitational waves from accreting millisecond pulsars, as pointed out previously by Melatos & Payne (2005). Persistent X-ray pulsations from accreting binary pulsars imply that the angle between the spin vector $\mathbf{\Omega}$ and the magnetic symmetry axis $\mn@boldsymbol{\mu}$ is not zero (Romanova et al., 2004; Kulkarni & Romanova, 2005). Hence a magnetic mountain constitutes a time-varying mass quadrupole which emits gravitational waves. Furthermore, the star precesses in general, emitting gravitational waves at the spin frequency and its first harmonic. The amplitude of the resulting signal (with curvature upscaled to a realistic neutron star at a distance $d=10$ kpc using $\epsilon\propto a^{2}$) is plotted in Fig. 25 for $10^{-9}\leq M_{a}/M_{\odot}\leq 10^{-3}$. The amplitude of the average signal that can be detected by the Laser Interferometer Gravitational Wave Observatory (LIGO) from a periodic source with a false alarm rate of 1 per cent and a false dismissal rate of 10 per cent over an integration time of $T_{0}=14$ days (Jaranowski et al., 1998; Abbott, B. et al., 2004), is overplotted in Fig. 25. This $T_{0}$ can realistically be achieved computationally. At this point, it is important to acknowledge that $M_{a}=1.4M_{c}\approx 1.7\times 10^{-4}M_{\odot}$ is still well below $M_{a}\sim 0.1M_{\odot}$, the mass required to spin up a neutron star to millisecond periods (Burderi et al., 1999). At present, this high-mass regime is not accessible numerically; neither the GS solver nor zeus-mp can handle the steep magnetic gradients involved. By the same token, Ohmic diffusion becomes important in this high-$M_{a}$ regime (Melatos & Payne, 2005; Vigelius & Melatos, 2008), smoothing the gradients and mitigating the numerical challenge. We postpone studying realistic values of $M_{a}$ to an accompanying paper, which will concentrate on non-ideal MHD simulations. However, to make a rough estimate regarding detectability here, we assume that non-ideal effects stall the growth of the mountain at $M_{a}\approx M_{c}$, following Melatos & Payne (2005). This includes the region shaded light grey in Fig. 25. Furthermore, no accreting millisecond pulsars have been discovered spinning faster than $f_{\ast}\ga 720$ Hz, possibly due to braking by gravitational waves (Bildsten, 1998; Chakrabarty et al., 2003). The region with $2f_{\ast}\ga 1400$ Hz is shaded dark grey in Fig. 25. Even with those exclusions, Fig. 25 demonstrates that there is a fair prospect of detecting gravitational waves from accreting X-ray millisecond pulsars in the near future, for accreted masses as low as $M_{a}\approx 10^{-4}M_{\odot}$. Recent directed searches for gravitational waves from the nearby X-ray source Sco-X1 found no signal at the level $h_{0}\ga 10^{-22}$ (Abbott, B. et al., 2007), thereby setting an upper bound on the ellipticity of $\epsilon=3.6\times 10^{-3}$. ## References * Abbott, B. et al. (2004) Abbott, B. et al. 2004, Phys. Rev. D, 69, 082004 * Abbott, B. et al. (2007) Abbott, B. et al. 2007, Phys. Rev. D, 76, 082001 * Akiyama et al. (2003) Akiyama S., Wheeler J. C., Meier D. L., Lichtenstadt I., 2003, ApJ, 584, 954 * Balbus & Hawley (1998) Balbus S. A., Hawley J. F., 1998, Rev. Mod. Phys., 70, 1 * Bildsten (1998) Bildsten L., 1998, ApJ, 501, L89+ * Biskamp (1993) Biskamp D., 1993, Nonlinear magnetohydrodynamics. Cambridge University Press, Cambridge. * Bisnovatyi-Kogan & Komberg (1974) Bisnovatyi-Kogan G. S., Komberg B. V., 1974, Soviet Astronomy, 18, 217 * Bonazzola & Gourgoulhon (1996) Bonazzola S., Gourgoulhon E., 1996, A&A, 312, 675 * Bouwkamp & Casimir (1954) Bouwkamp C. J., Casimir H. B. G., 1954, Physica, 20, 539 * Brown & Bildsten (1998) Brown E. F., Bildsten L., 1998, ApJ, 496, 915 * Burderi et al. (1999) Burderi L., Possenti A., Colpi M., di Salvo T., D’Amico N., 1999, ApJ, 519, 285 * Chakrabarty et al. (2003) Chakrabarty D., Morgan E. H., Muno M. P., Galloway D. K., Wijnands R., van der Klis M., Markwardt C. B., 2003, Nature, 424, 42 * Cumming et al. (2004) Cumming A., Arras P., Zweibel E., 2004, ApJ, 609, 999 * Cutler (2002) Cutler C., 2002, Phys. Rev. D, 66, 084025 * Geppert & Rheinhardt (2002) Geppert U., Rheinhardt M., 2002, A&A, 392, 1015 * Geppert & Urpin (1994) Geppert U., Urpin V., 1994, MNRAS, 271, 490 * Goedbloed & Halberstadt (1994) Goedbloed J. P., Halberstadt G., 1994, A&A, 286, 275 * Goedbloed & Poedts (2004) Goedbloed J. P. H., Poedts S., 2004, Principles of Magnetohydrodynamics. Cambridge University Press, Cambridge. * Greene & Johnson (1968) Greene J. M., Johnson J. L., 1968, Plasma Physics, 10, 729 * Haskell et al. (2006) Haskell B., Jones D. I., Andersson N., 2006, MNRAS, 373, 1423 * Hawley & Stone (1995) Hawley J. F., Stone J. M., 1995, Comp. Phys. Comm., 89, 127 * Hayes et al. (2006) Hayes J. C., Norman M. L., Fiedler R. A., Bordner J. O., Li P. S., Clark S. E., ud-Doula A., Mac Low M.-M., 2006, ApJS, 165, 188 * Hughes & Cattaneo (1987) Hughes D. W., Cattaneo F., 1987, Geophysical and Astrophysical Fluid Dynamics, 39, 65 * Jackson (1998) Jackson J. D., 1998, Classical Electrodynamics. Wiley-VCH, New York. * Jaranowski et al. (1998) Jaranowski P., Królak A., Schutz B. F., 1998, Phys. Rev. D, 58, 063001 * Konar & Bhattacharya (1997) Konar S., Bhattacharya D., 1997, MNRAS, 284, 311 * Kulkarni & Romanova (2005) Kulkarni A. K., Romanova M. M., 2005, ApJ, 633, 349 * Lifschitz (1989) Lifschitz A. E., 1989, Magnetohydrodynamics and Spectral Theory. Kluwer Academic Publishers, London. * Litwin et al. (2001) Litwin C., Brown E. F., Rosner R., 2001, ApJ, 553, 788 * Lovelace et al. (2005) Lovelace R. V. E., Romanova M. M., Bisnovatyi-Kogan G. S., 2005, ApJ, 625, 957 * Masada et al. (2006) Masada Y., Sano T., Takabe H., 2006, ApJ, 641, 447 * Matsumoto & Shibata (1992) Matsumoto R., Shibata K., 1992, PASJ, 44, 167 * Melatos & Payne (2005) Melatos A., Payne D. J. B., 2005, ApJ, 623, 1044 * Melatos & Phinney (2001) Melatos A., Phinney E. S., 2001, Publ. Astronom. Soc. Aust., 18, 421 * Mouschovias (1974) Mouschovias T. C., 1974, ApJ, 192, 37 * Muslimov & Tsygan (1985) Muslimov A. G., Tsygan A. I., 1985, Sov. Astron. Lett., 11, 80 * Owen (2006) Owen B. J., 2006, Classical and Quantum Gravity, 23, 1 * Payne (2005) Payne D. J. B., 2005, PhD thesis, School of Physics. University of Melbourne. * Payne & Melatos (2004) Payne D. J. B., Melatos A., 2004, MNRAS, 351, 569 * Payne & Melatos (2006a) Payne D. J. B., Melatos A., 2006a, ApJ, 641, 471 * Payne & Melatos (2006b) Payne D. J. B., Melatos A., 2006b, ApJ, 652, 597 * Payne & Melatos (2007) Payne D. J. B., Melatos A., 2007, MNRAS, 376, 609 * Pons & Geppert (2007) Pons J. A., Geppert U., 2007, A&A, 470, 303 * Priest (1984) Priest E. R., 1984, Solar magneto-hydrodynamics. Geophysics and Astrophysics Monographs, Dordrecht: Reidel. * Romani (1990) Romani R. W., 1990, Nature, 347, 741 * Romanova et al. (2004) Romanova M. M., Ustyugova G. V., Koldoba A. V., Lovelace R. V. E., 2004, ApJ, 610, 920 * Schindler et al. (1988) Schindler K., Hesse M., Birn J., 1988, J. Geophys. Res., 93, 5547 * Shapiro & Teukolsky (1983) Shapiro S. L., Teukolsky S. A., 1983, Black holes, white dwarfs, and neutron stars: The physics of compact objects. Wiley-Interscience, New York. * Srinivasan et al. (1990) Srinivasan G., Bhattacharya D., Muslimov A. G., Tsygan A. J., 1990, Curr. Sci., 59, 31 * Taam & van de Heuvel (1986) Taam R. E., van de Heuvel E. P. J., 1986, ApJ, 305, 235 * Urpin & Konenkov (1997) Urpin V., Konenkov D., 1997, MNRAS, 284, 741 * van den Heuvel & Bitzaraki (1995) van den Heuvel E. P. J., Bitzaraki O., 1995, A&A, 297, L41+ * Vigelius & Melatos (2008) Vigelius M., Melatos A., 2008, in preparation * Wijers (1997) Wijers R. A. M. J., 1997, MNRAS, 287, 607 * Zhang (1998) Zhang C. M., 1998, Ap&SS, 262, 97 ## Appendix A Defining the grid and boundary conditions in zeus-mp In this appendix, we briefly outline the key variables and settings in zeus- mp, to aid the reader in reproducing our numerical results. Our grid consists of ggen1:nbxl, ggen2:nbxl, and ggen3:nbxl blocks in the $r$, $\theta$, and $\phi$ direction, respectively. The integration volume is defined by $R_{\ast}/h_{0}\leq r\leq R_{m}$, $0\leq\theta\leq\pi/2$, $0\leq\phi<2\pi$. The radial coordinate in the GS code, $\tilde{x}=(r-R_{\ast})/h_{0}$ ($0\leq\tilde{x}\leq X$), is stretched logarithmically according to $\tilde{x}_{1}=\log(\tilde{x}+\mathrm{e}^{-L_{x}})+L_{x}$, where $L_{x}$ controls the zooming (PM04). This grid is implemented by setting the zeus-mp parameters $\mathtt{ggen1:x1min}=R_{\ast}/h_{0}$ and $\mathtt{ggen1:x1max}=\mathtt{ggen1:x1min}+X$. Stretching is achieved via the parameter ggen1:x1rat, which sets the radial length ratio of two neighbouring zones. In order to get consistent radial grid positions in the GS code and zeus-mp, we set $\mathtt{ggen1:x1rat}=(X\mathrm{e}^{L_{x}}+1)^{(G_{x}-1)^{-1}}$. Boundary conditions are enforced in zeus-mp via ghost cells, which frame the active grid cells. Several predefined prescriptions are supplied to implement a variety of standard boundary conditions. In the $\phi$ direction, we choose periodic boundary conditions [ikb.niks(1)=4 and okb.noks(1)=4]. The $\theta=\pi/2$ surface is reflecting, with normal magnetic field [ojb.nojs(1)= 5], which translates to $\mathbf{v}_{\perp}=\mathbf{B}_{\parallel}=0$. The line $\theta=0$ is also reflecting [ijb.nijs(1)= -1] with tangential magnetic field ($\mathbf{v}_{\perp}=\mathbf{B}_{\perp}=0$). Additionally, the toroidal component $B_{\phi}$ is reversed at the boundary, i.e. $B^{<}_{\phi}=-B^{>}_{\phi}$, where $B_{\phi}^{<}$ and $B_{\phi}^{>}$ are the field components for $\theta<0$ and $\theta>0$, respectively. The outer surface $r=R_{m}$ is usually an outflow [oib nois(1)= 2] boundary, i.e. zero gradient. The stellar surface is impenetrable, so the inner $r=R_{\ast}$ boundary is inflow [iib.niis(1)= 3]. This enables us to impose line-tying at $r=R_{\ast}$ by fixing the density and magnetic field there. We also use an isothermal equation of state (XISO=.true.). ## Appendix B Mass multipole moments We work out the mass quadrupole moment in Cartesian coordinates from the code output in spherical coordinates. Following Jackson (1998), we define the spherical mass multipole moments according to $q_{lm}=\int\mathrm{d}^{3}\mathbf{x^{\prime}}Y^{*}_{lm}(\theta^{\prime},\phi^{\prime})r^{\prime l}\rho(\mathbf{x}^{\prime}),$ (17) where $Y_{lm}$ denotes the usual orthonormal set of spherical harmonics. The spherical quadrupole moments are related to the traceless, Cartesian quadrupole moment tensor, $Q_{ij}=\int d^{3}x^{\prime}\,(3x_{i}^{\prime}x_{j}^{\prime}-r^{\prime 2}\delta_{ij})\rho(\mathbf{x^{\prime}}),$ (18) by $Q_{11}=6(2\pi/15)^{1/2}\mathrm{Re}(q_{22})-2(4\pi/5)^{1/2}q_{20}$, $Q_{12}=-6(2\pi/15)^{1/2}\mathrm{Im}(q_{22})$, $Q_{13}=-3(8\pi/15)^{1/2}\mathrm{Re}(q_{21})$, $Q_{22}=-6(2\pi/15)^{1/2}\mathrm{Re}(q_{22})-(4\pi/5)^{1/2}q_{20}$, and $Q_{23}=3(8\pi/15)^{1/2}\mathrm{Im}(q_{21})$. In the axisymmetric case (when the star and the mountain form a prolate spheroid), we have $\rho=\rho(r,\theta)$ and the $\phi$ integrals in (17) vanish. $Q$ is then diagonal with components $Q_{\hat{x}\hat{x}}=Q_{\hat{y}\hat{y}}=-Q_{\hat{z}\hat{z}}/2$, with respect to the body coordinate system, and we can introduce the ellipticity444Note that the definition of the ellipticity varies in the literature. The ellipticity defined here is consistent with Bonazzola & Gourgoulhon (1996) and Shapiro & Teukolsky (1983) and is related to the ellipticity in Melatos & Payne (2005) and Jaranowski et al. (1998) by $|\epsilon|=3\epsilon_{\mathrm{MP}}$. Abbott, B. et al. (2007) used a different ellipticity defined for a _triaxial_ rotator, $\epsilon=(I_{xx}-I_{yy})/I_{zz}$. $\epsilon$, where we assume that the $\hat{z}$ axis is the symmetry axis: $\epsilon=\frac{Q_{\hat{z}\hat{z}}}{2I_{\hat{z}\hat{z}}}=\frac{3(I_{\hat{z}\hat{z}}-I_{\hat{x}\hat{x}})}{I_{\hat{z}\hat{z}}},$ (19) with $I_{\hat{z}\hat{z}}=2M_{\ast}R_{\ast}^{2}/5$ is the moment of inertia along the rotation axis for a biaxial ellipsoid with mass $M_{\ast}$ and minor axis $R_{\ast}$. We can compute the ellipticity directly from the code output through $\epsilon=\frac{1}{2I_{\hat{z}\hat{z}}}\int d\theta\,d\phi\,dr\,\rho r^{4}\sin\theta(3\cos^{2}\theta-1).$ (20) ## Appendix C Magnetic multipole moments In a source-free region $\mathbf{J}=0$, a magnetic field $\mathbf{B}$ is determined solely by its radial component $B_{r}$ (Bouwkamp & Casimir, 1954), which, from Maxwell’s equations, satisfies the Laplace equation $\nabla^{2}B_{r}=0.$ (21) One can therefore define the magnetic multipoles as the expansion coefficients in the general solution of the boundary value problem (21), viz. $B_{r}=\sum_{l=0}^{\infty}\sum_{m=-l}^{l}d_{lm}r^{-(l+1)}Y_{lm}(\theta,\phi),$ (22) with $d_{lm}=r^{l+1}\int\mathrm{d}\Omega\;Y_{lm}^{\ast}\mathbf{r}\mn@boldsymbol{\cdot}\mathbf{B}.$ (23) Note that $d_{10}$ is related to the magnetic moment $\mu$ of a dipole field $\mathbf{B}(\mathbf{r})=\mu r^{-3}(2\cos\theta\mathbf{e}_{r}+\sin\theta\mathbf{e}_{\theta})$ by $d_{10}=4(\pi/3)^{1/2}\mu$. In the case of north-south symmetry, we find $d_{10}=2\hat{d}_{10}$ and $d_{21}=2\hat{d}_{21}$, where a hat denotes the moment evaluated on the hemisphere. All other coefficients vanish.

arxiv-papers

2024-09-04T02:48:53.950134

0802.3245

# Optimal angular window for observing Dark Matter annihilation from the Galactic Center region: the case of gamma-ray lines. Pasquale D. Serpico Center for Particle Astrophysics, Fermi National Accelerator Laboratory, Batavia, IL 60510-0500 USA Gabrijela Zaharijas HEP Division, Argonne National Laboratory, 9700 Cass Ave., Argonne, IL 60439 ###### Abstract Although the emission of radiation from dark matter annihilation is expected to be maximized at the Galactic Center, geometric factors and the presence of point-like and diffuse backgrounds make the choice of the angular window size to optimize the chance of a signal detection a non-trivial problem. We find that the best strategy is to focus on an annulus around the Galactic Center of $\sim 1^{\circ}$ to $\raise 1.29167pt\hbox{$\;>$\kern-7.5pt\raise-4.73611pt\hbox{$\sim\;$}}30^{\circ}$, where the optimal size depends on the angular distribution of the signal and the backgrounds. Although our conclusions are general, we illustrate this point in the particular case of annihilation into two monochromatic photons in the phenomenologically most interesting range of energy $45\,{\rm GeV}\raise 1.29167pt\hbox{$\;<$\kern-7.5pt\raise-4.73611pt\hbox{$\sim\;$}}E\raise 1.29167pt\hbox{$\;<$\kern-7.5pt\raise-4.73611pt\hbox{$\sim\;$}}80\,$GeV, which is of great interest for the GLAST satellite. We find for example that Dark Matter models with sufficiently strong line annihilation signals, like the Inert Doublet Model, may be detectable without or with reasonable boost factors. ###### pacs: 95.35.+d, 14.80.Cp, 98.70.Rz ANL-HEP-PR-08-16, FERMILAB-PUB-08-036-A ## I Introduction Although the existence of Dark Matter seems to be required by a wealth of astrophysical and cosmological data (for reviews see Bertone:2004pz ; Bergstrom:2000pn ), the identification of its physical nature remains still elusive. The most popular candidates are the so-called weakly interacting massive particles (or WIMPs), believed to be neutral and colorless thermal relics from the early universe with electroweak-scale masses ($m_{X}\sim 10-1000\,$GeV) and couplings. These generic assumptions seem indeed to roughly reproduce the amount of DM needed in the Universe and in our Galaxy in particular, while arising at the same time naturally in many extensions of the Standard Model of Particle Physics. Among the different methods proposed to identify WIMP dark matter (production in accelerators, direct detection via recoil of nuclei underground, indirect detection via their annihilation or decay products Hooper:2008sn ), a particularly promising technique is the detection of gamma rays from annihilation in the galactic halo. However, it is nowadays clear that the DM signal should be likely singled out of a larger astrophysical background, either point-like (the case of the Galactic Center) or diffuse. In assessing the capability of detecting a given WIMP candidate it is important to optimize the search strategy based on our current expectation for the energy range, but also the angular distribution of the signal. This information proves actually to be at least as useful in guiding DM searches as the energy spectrum and the choice of astrophysical targets. Here we illustrate this point with reference to the inner Galaxy region. Although the emission of radiation per unit solid angle from dark matter annihilation is expected to be maximized at the Galactic Center, geometric factors and the presence of point-like and diffuse backgrounds make the choice of the window size to optimize the chance of a signal detection a non-trivial problem. To the best of our knowledge, this issue has been only raised within over-simplified assumptions for the signal and background Stoehr:2003hf or, by converse, treated in some technical studies by experimental groups devoted to dark matter searches (see for example poster ), with the background profile determined e.g. via a numerical model. In no case a comparative study for different profiles and background assumptions, which illustrates the physical origin of the optimal cuts to adopt, has been presented. This is the main task we address in this article. We shall prove that—in all the commonly considered models for the DM distribution in our Galaxy and given the present knowledge of point-like and diffuse backgrounds—the optimal strategy for DM searches is never to focus on the inner 0.1∘-0.5∘ around the Galactic Center. Unless astrophysical processes dramatically enhance the DM distribution in the innermost region of our Galaxy, producing effectively a bright point-source, an annulus of radius from $\sim 1^{\circ}$ to $\raise 1.29167pt\hbox{$\;>$\kern-7.5pt\raise-4.73611pt\hbox{$\sim\;$}}30^{\circ}$ is always to be preferred. The optimal size depends on the angular distribution of the signal and the backgrounds, and the improvement in the sensitivity reaches a factor of a few, thus affecting the prospects for the detectability of specific models. To give an explicit example, we shall refer mostly to the detection of gamma-ray lines from annihilation. Gamma-ray lines of energy $E=m_{X}$ would be an unambiguous signature of WIMPs as DM particles populating our Galactic Halo, but the two-photon channel is loop(or radiatively)-suppressed with respect to the main WIMP annihilation channels: fermion pairs, weak gauge bosons or higgses. Even models where the line emission is enhanced, as the Inert Doublet Model (IDM) Barbieri:2006dq ; Gustafsson:2007pc are very challenging to detect via line emission. Yet, with an optimized choice of the angular size, the IDM model may be detectable without or with reasonable boost factors (i.e., enhancement of the flux with respect to the smooth halo signal due to the “clumpiness” of DM). Similar considerations apply for the more conventional continuum gamma-ray flux from DM annihilation, usually considered a more promising channel. This paper is structured as follows. In Sec. II we discuss the ingredients contributing to the DM signal. In Sec. III we describe our model for the point-like and diffuse backgrounds, while Sec. IV is devoted to present our results. In Sec. V we conclude. ## II The Signal It is customary to write the number flux of gamma rays (per energy and solid angle) produced in dark matter (DM) annihilations in our Galactic halo as $\Phi^{\rm DM}(E,\psi)=\left[\frac{{\rm d}N_{\gamma}}{{\rm d}E}\frac{\langle\sigma v\rangle}{8\pi m^{2}_{X}}\right]\,d_{\rm f}\,\rho_{\rm f}^{2}\,J(\psi)\,,$ (1) where the factor in square brackets depends only on the particle physics model: $\langle\sigma v\rangle$ is the WIMP annihilation cross section multiplied by the relative velocity of the two WIMPs (averaged over the WIMP velocity distribution), $m_{X}$ is the mass of the WIMP, ${\rm d}N_{\gamma}/{\rm d}E$ is the photon spectrum per annihilation. If the dark matter is not its own antiparticle (as assumed here), Eq. (1) should be multiplied further by a factor 1/2 (if $X$ and $\bar{X}$ are equally abundant). The factor outside the brackets only depends on the DM astrophysics, in particular on the distribution of DM in the halo and on the position of the Solar System within the halo. The quantities $d_{\rm f}$ and $\rho_{\rm f}$ are constants with dimensions of distance and energy density respectively, whose only purpose is of making $J$—which encodes the angular dependence of the DM signal—a dimensionless function. Following Bergstrom:1997fj , we fix these parameters at $d_{\rm f}=8.5\,$kpc and $\rho_{\rm f}=0.3\,$GeV/cm3 (indicative the galactocentric distance of the sun and of the local density of dark matter). The $J$-function is thus defined in terms of the DM density $\rho$ as $J(\psi)\equiv\frac{1}{d_{\rm f}}\frac{1}{\rho_{\rm f}^{2}}\int_{\rm{los}}\rho^{2}[r(\ell,\psi)]{\rm d}\ell,$ (2) where $r(\ell,\psi)=\sqrt{r_{\odot}^{2}+\ell^{2}-2\,r_{\odot}\,\ell\cos\psi}\,.$ (3) In the previous equations, $\psi$ is the angle between the direction in the sky and the Galactic Center so that, in terms of galactic latitude $b$ and longitude $l$, $\cos\psi=\cos b\cos l$; $r_{\odot}\approx 8.0\,$kpc Yao:2006px is the solar distance from the Galactic Center; $\ell$ is the distance from the Sun along the line-of-sight (l.o.s.). Note that we implicitly assumed that the halo is spherically symmetric and that the angular dependence only comes from the off-center position of the Sun in the halo, which implies for example the azimuthal symmetry of the signal (see e.g. Hooper:2007be and refs. therein for a discussion of sub-leading effects determining the angular distribution of the signal). A general class of spherically symmetric, smooth halo distributions can be fitted as $\rho(r)=\left(\frac{r_{s}}{r}\right)^{\gamma}\frac{\rho_{0}}{[1+(r/r_{s})^{\alpha}]^{(\beta-\gamma)/\alpha}},$ (4) where $\rho_{0}$ is a normalization constant and $r_{s}$ is a characteristic radius below which the profile scales as $r^{-\gamma}$. Two of the most well known profiles have been proposed by Navarro, Frenk and White (NFW) NFW and Moore et al. Moore . Another fit, proposed by Kravtsov et al. Kravtsov:1997dp mimics a flat core potential, which seems to be in better agreement with some observations. More recent simulations (see e.g. Power:2002sw ; Navarro:2003ew ; Reed:2003hp ; Merritt:2005xc ) suggest that halo density profiles are better represented by a function with a continuously-varying slope. At least above the resolved radius ($\sim 1\,$kpc) and for Galactic-size halos, the newly proposed fitting formulae provide only marginal improvement with respect to the more traditional ones, although it seems that the inner slope of a DM halo is very likely shallower that the Moore one, and probably shallower than the NFW one, too. For illustrative purposes, we shall consider the Moore, NFW and Kravtsov profiles with the parameters reported in Table I, since the former and the latter bracket the different cases, with the NFW (as well as the paramterization used in Merritt:2005xc ) falling in between. Similarly to Ref. Yuksel:2007ac , the normalizations are chosen so that the mass contained within the solar circle provides the appropriate DM contribution to the local rotational curves, such that $\rho_{\odot}\equiv\rho(r_{\odot})$—the DM density at the solar distance from the GC—yields to the canonical value $\rho_{\odot}=0.3\,$GeV/cm3 for the NFW profile. Table 1: Halo parameters, see text and Ref. Yuksel:2007ac . Model | $\alpha$ | $\beta$ | $\gamma$ | $\rho_{\odot}$ | $r_{s}$ ---|---|---|---|---|--- | | | | [GeV cm-3] | [kpc] Moore | 1.5 | 3 | 1.5 | 0.27 | 28 NFW | 1.0 | 3 | 1.0 | 0.30 | 20 Kravtsov | 2.0 | 3 | 0.4 | 0.47 | 10 Note that in none of the above-mentioned models astrophysical processes modifying the pure DM profile are included. For example, “adiabatic compression” may lead to the formation of a DM spike around the central supermassive black hole; mergers as well as scattering on the dense stellar cusp around the central black hole may destroy density enhancements, etc. While potentially relevant in determining the dark matter annihilation brightness of the Galactic Center, they only affect regions too close to the Galactic Center to be resolved angularly by present detectors. In general, they might lead to a measurable gamma ray flux from the innermost angular bin even in presence of relatively large astrophysical backgrounds, but in this case only the energy information is relevant for the separation, for which the present discussion is irrelevant. We not address these effects further; details can be found in the refs. cited in Bertone:2004pz or ref. Fornasa:2007nr . The spectrum per annihilation depends on the nature of the WIMP, although the gamma-ray yield is often largely dominated by the result of the fragmentation and hadronization of the WIMP annihilation products. Neutralinos in the MSSM, for example, typically annihilate to final states consisting of heavy fermions and gauge or Higgs bosons jungman . With the exception of the $\tau^{+}\tau^{-}$ channel, each of these annihilation modes result in a very similar spectrum of gamma rays. In most of the following considerations, however, it is the angular distribution of the DM radiation that determine the optimal angular window. To avoid unnecessary complications, we shall illustrate our point considering mostly the case of annihilation in two gammas, for which $\frac{{\rm d}N_{\gamma}}{{\rm d}E}=2\,b_{\gamma\gamma}\delta(E-m_{X})\,,$ (5) where $b_{\gamma\gamma}$ is the branching ratio for the 2-photon line. Although it would constitute a striking signature of a DM annihilation process, in general the photon line is considered hard to detect, since it is loop-suppressed with respect to the main annihilation modes Bergstrom:1997fj . This is not necessarily the case in some DM models, as for example the Inert Doublet Model (IDM) Barbieri:2006dq ; Gustafsson:2007pc . In general, the line feature may be enhanced whenever: (i) the DM almost decouples from fermions, while being mostly coupled to gauge bosons; (ii) the DM candidate is in the mass range $45\,{\rm GeV}\simeq m_{Z}/2<M_{X}<m_{W}\simeq{\rm 80}\,$GeV, the lower limit coming from the bound on the $Z^{0}$-width measurement and the upper limit to avoid the annihilation mode $X\,X\to W^{+}W^{-}$ by kinematical constraints (other cases where a prominent line emission is possible at lower or higher energies have been treated respectively in Pullen:2006sy and Bringmann:2007nk ). Interestingly, this is probably the best region where the GLAST satellite GLASTurl has a reasonable chance to detect a line emission and at the same time, potentially interesting for the next generation of low- threshold atmospheric cherenkov telescopes (ACTs), like CTA CTA or AGIS AGIS . In the following, for GLAST we shall use an integrated area exposure over time of $10^{4}\,{\rm cm}^{2}\,{\rm yr}$, corresponding to about 5 years of observations, and an energy resolution of $\pm 7\%$ at $1\sigma$. For ACTs, we shall assume an effective array area of $A_{\rm eff}=1\,$km2 , energy resolution of $15\%$ and 200 hour observations, with a $2^{\circ}\times 2^{\circ}$ field of view. Whenever needed, we shall compare our results with two benchmark models for IDM candidates among the ones considered in Gustafsson:2007pc , see Table 2. Also, to get a feeling of how things change if the continuum gamma-ray spectrum is considered instead, we shall also report the results obtained when a fiducial case of a 100 GeV MSSM neutralino mainly annihilating into $W^{+}W^{-}$ is considered. Table 2: Benchmark models in the IDM scenario, see Gustafsson:2007pc . Model | $m_{X}$ | $\langle\sigma v\rangle$ | $b_{\gamma\gamma}$ ---|---|---|--- | [GeV] | [cm3/s] | $I$ | $70$ | $1.6\times 10^{-28}$ | $0.36$ $II$ | $50$ | $8.2\times 10^{-29}$ | $0.29$ ## III The Backgrounds The Galactic Center is a complex region of the sky at all wavelengths, the gamma-ray window being no exception. In this section, we discuss how, in our analysis, we treat the backgrounds for dark matter searches due to known and unknown astrophysical sources of gamma rays. Since we focus on an extended signal, diffuse (or unresolved) backgrounds are our main concern. However, to illustrate how effective is a large field-of-view search, we shall compare it with the sensitivity achieved when focusing on the inner region around the Galactic Center, when at least one astrophysical point-like background is known to exist. This is the relatively bright, very high-energy gamma ray source observed by HESS, MAGIC, WHIPPLE, and CANGAROO-II GCtev . This source is consistent with point-like emission and is located at $l=359^{\circ}56^{\prime}41.1^{\prime\prime}\pm 6.4^{\prime\prime}$ (stat.), $b=-0^{\circ}2^{\prime}39.2^{\prime\prime}\pm 5.9^{\prime\prime}$ (stat.) with a systematic pointing error of 28′′ van Eldik:2007yi . It appears to be coincident with the position of Sgr A⋆, the black hole constituting the dynamical center of the Milky Way. Following the measurements of HESS, we describe the spectrum of this source as a power-law given by: $\Phi^{\rm GC}=1.0\times 10^{-8}\left({{E}\over{{\rm GeV}}}\right)^{-2.25}{\rm GeV}^{-1}\,{\rm cm}^{-2}\,{\rm s}^{-1}.$ (6) At energies below $\sim 200$ GeV, the spectrum of this source has not yet been measured. GLAST, however, will be capable of measuring the spectrum of this source at energies below the thresholds of HESS and other ACTs. In addition to the HESS source, a yet unidentified source has been detected by EGRET in the GeV range, approximately 0.2∘ away from the dynamical center of our galaxy dingus ; pohl . It has not been detected so-far at the energies explored by HESS and other ACTs, which means that its power-law spectrum cuts- off between $\sim$10 and $\sim$100 GeV. Although it should be included in more realistic analyses of the DM signal from the inner galactic regions Dodelson:2007gd , in the following we neglect it, thus overestimating the diagnostic power of a narrow field of view search and providing a conservative estimate of the improvement achieved by a large angle search. The overall diffuse gamma-ray radiation can be qualitatively divided into a galactic and an extragalactic contribution. Since the latter is not simply the isotropic part of the flux, the separation of these two components can be done at present only assuming a specific model for the production of secondaries by cosmic rays in the galactic disk and halo. (However, a measurement of the cosmological Compton-Getting effect that should be achievable for GLAST would provide a model-independent way to separate the two contributions Kachelriess:2006aq ). A significant fraction of the quasi-isotropic component, especially in the GeV range, may be due to high-latitude galactic emission coming from processes in the magnetized halo of the Milky Way. We employ here a fit of the galactic diffuse flux proposed in Bergstrom:1997fj and calibrated on EGRET data around the GeV Hunter:1997 , $I_{\rm gal}(E)=N_{0}(l,b)\times 10^{-6}\left(\frac{E}{{\rm GeV}}\right)^{-\alpha}\,{\rm cm}^{-2}{\rm s}^{-1}{\rm sr}^{-1}{\rm GeV}^{-1}\,,$ (7) where the arguments $l,b$ are in degrees, $-180^{\circ}\leq l\leq 180^{\circ}$ and $-90^{\circ}\leq b\leq 90^{\circ}$, $\alpha\simeq 2.7$, and $N_{0}(l,b)=\left\\{\begin{array}[]{cc}\frac{85.5}{\sqrt{1+(l/35)^{2}}\sqrt{1+[b/(1.1+0.022\,|l|)]^{2}}}+0.5\,,&\>\>|l|\geq 30^{\circ}\\\ \frac{85.5}{\sqrt{1+(l/35)^{2}}\sqrt{1+(b/1.8)^{2}}}+0.5\,,&\>\>|l|\leq 30^{\circ}\end{array}\right..$ (8) The EGRET collaboration derived the intensity of the extragalactic gamma-ray flux as Sreekumar:1997un $I_{\rm ex}(E)=k_{0}\times 10^{-6}\left(\frac{E}{0.451{\rm GeV}}\right)^{-\beta}{\rm cm}^{-2}{\rm s}^{-1}{\rm sr}^{-1}{\rm GeV}^{-1}\,,$ (9) with $\beta=2.10\pm 0.03$, $k_{0}=(7.32\pm 0.34)$ and the fits is valid from $E\sim\,$10 MeV to $E\sim\,$100 GeV. In the following, we shall consider two models of diffuse backgrounds: a “conservative” model assuming $\alpha=2.7$ in Eq. (7) and $k_{0}=7.32$ in Eq. (9), and an “optimistic” model, where a harder spectrum $\alpha=3.0$ is assumed for the Galactic background and $k_{0}=3.66$. Based on presently available models, one expects indeed that GLAST should resolve roughly half of the extragalactic background Stecker:2001dk ; analogously, one expects that unresolved sources contribute significantly to the diffuse Galactic spectrum as measured by EGRET, and once detected and removed by GLAST the Galactic background should be steeper (see e.g. discussion in Pullen:2006sy ). Finally, for ACTs the dominant background is the isotropic cosmic ray one. At energies well below $\sim 100\,$GeV one expects the cosmic-ray electron background to play a dominant role. In the following, we account for these backgrounds following the formulae reported in Bergstrom:1997fj . ## IV Results First, let us repeat a simple argument already sketched in Stoehr:2003hf . The DM signal coming from integrating over a solid angle $\Delta\Omega$ is proportional to $\langle J(\Omega)\rangle\Delta\Omega$, where $\langle J(\Omega)\rangle\equiv\frac{1}{\Delta\Omega}\int_{\Delta\Omega}J\,{\rm d}\Omega\,.$ (10) For an annulus around the GC with inner opening semi-angle $\psi_{\rm m}$ and outer one $\psi_{\rm M}$, the previous formula reduces to $\langle J\rangle=\frac{1}{\cos\psi_{\rm m}-\cos\psi_{\rm M}}\int_{\psi_{\rm m}}^{\psi_{\rm M}}J(\psi)\sin\psi{\rm d}\psi\,.$ (11) The significance $S/N$ of a Signal $S$ in presence of a background $B$ is defined as $\frac{S}{N}\equiv\frac{S}{\sqrt{S+B}}$ (12) The dark matter signal goes like $S\propto\langle J\rangle\,\Delta\Omega$, while for an isotropic background $B\propto\Delta\Omega$. In the limit $S\ll B$ (which is likely to hold in all realistic DM searches), we have $\frac{S}{N}\propto\langle J\rangle\sqrt{\Delta\Omega}\sim\langle J\rangle\psi_{\rm M}\,.$ (13) In Fig. 1, we plot this function vs. $\psi_{\rm M}$ for the three models considered in Table II, assuming $\psi_{\rm m}=0.1^{\circ}$ (solid lines) or $\psi_{\rm m}=0.3^{\circ}$ (dashed lines). Note the exact choice of the inner bound is irrelevant for cored profiles: the peak of this function is above 20∘. Even for profiles as cusped as the NFW (or more) the peak of this function is at a few degrees. Of course, the Moore profile gives a formally divergent signal when $\psi_{\rm m}\to 0$, so the inner angle cutoff does matter. Figure 1: Plot of the function $\langle J\rangle\psi_{\rm M}$ defined in Eq. (11) (proportional to the significance of a DM signal for an isotropic background in the limit $S\ll B$) for the three models considered in Table I: the solid lines assume $\psi_{\rm m}=0.1^{\circ}$, the dashed lines $\psi_{\rm M}=0.3^{\circ}$; the leftmost (red) lines are for the Moore profile, the rightmost (green) lines for the Kravtsov, in the middle (blue) lines for the NFW profile. Figure 1 already shows that the $S/N$ is typically maximized for values of $\psi_{\rm M}\gg\psi_{\rm m}$. However, the diffuse backgrounds are not $\\{l,b\\}$-independent, but really decline away from the Galactic Plane. When using the models for the galactic and extragalactic background introduced in the previous section, we see how this effect is appreciable. In Fig. 2 , in order to make the comparison among the different models easier, we plot the rescaled $S/N$ vs. $b_{\rm max}$ (in such a way that their maximum values are equal) for a region $0.4^{\circ}<|b|<b_{\rm max}$, $0^{\circ}<|l|<l_{\rm max}=b_{\rm max}$ for the fiducial IDM model I of Tab. II, and for the continuum spectrum of a 100 GeV MSSM neutralino annihilating into $W^{+}W^{-}$. In the former case, we consider the 1-$\sigma$ smoothing of the gamma-line due to the GLAST energy resolution, while for the continuum we actually show the S/N in the most significant energy bin (which happens to fall in the $\sim 1\,$GeV range). The inner cut has been chosen to $0.4^{\circ}$ to avoid contamination from known point sources close to the GC. Note how the peak of $S/N$ moves to larger angles in all cases compared with Fig. 1, developing a maximum at finite angle also for the Moore profile. For the continuum spectrum the peak of the significance moves to larger angles, due to the fact that the isotropic component of the background is harder, and thus at the lower energies relevant for this signal the overall background is dominated by the Galactic background which falls rapidly with $b$. It is also worth nothing that even for heavy dark matter particles (TeV mass scale) the continuum photon spectrum peaks at or below about 30 GeV. Since the $S/N$ for continuum spectra is dominated by its value at the peak, the angular window results reported in Figs. 1 and 2 are really more general than the case they were discussed for, reflecting the situation for the continuum spectrum from DM in all the interesting mass range. Figure 2: relative Signal/Noise vs. annulus size $b_{\rm max}$ (we equate their maximum value to ease the comparison) for a region $0.4^{\circ}<|b|<b_{\rm max}$, $0^{\circ}<|l|<l_{\rm max}=b_{\rm max}$ for the continuum spectrum of a 100 GeV MSSM neutralino annihilating into $W^{+}W^{-}$ (dot-dashed line, NFW profile) and for the fiducial IDM model I of Tab. II for the Moore, NFW and Kravtsov profile (solid lines respectively from the left to the right). In Fig. 3 we show the curves with $S/N=3$ for the branching ratio into two gammas, normalized in terms of the fiducial annihilation cross section value $\langle\sigma v\rangle=3\times 10^{-26}\,$cm3/s in the phenomenologically most interesting mass range below $m_{W}$. The triangles represent the fiducial points of the IDM model reported in Table II, and for the window size that maximizes $S/N$ (respectively $b_{\rm max}\sim 1.5^{\circ}$, 15∘ and 30∘, for Moore, NFW and Kravtsov profiles of Tab. I). The solid lines are for the conservative background estimate, the dashed lines for the optimistic one. It is interesting to note that for the Moore profile, the fiducial models should be detectable at the $3\,\sigma$ level without need for boost factors. For the NFW case, boost factors of order 10 should suffice. Even for the unfortunate case of a Kravtsov profile, a boost factor $\sim 20$ should be sufficient. These values appear to be viable in cold DM cosmologies, see e.g. Strigari:2006rd . Thus, our conclusions are more promising that what envisioned in Gustafsson:2007pc , where boost factors $\raise 1.29167pt\hbox{$\;>$\kern-7.5pt\raise-4.73611pt\hbox{$\sim\;$}}100$ were considered. Also, the top dot-dashed line shows the sensitivity achieved if one only includes the inner $0.1^{\circ}$ region around the GC, where the background is dominated by the central source. These limits, comparable with the results reported in gabi , show an evident improvement. It is also worth commenting on the fact that, according to our estimate, although ACTs are not optimized to search for line signals in the range below 100 GeV, the performance of the next generation, km2 instruments presently being considered may be comparable or better than GLAST. For example, in accordance with existing estimates ACTwp , for the fiducial exposure parameters we have used and the NFW profile we find a sensitivity a factor $\sim$1.5 better in the parameter space of Fig. 3, despite their worse energy resolution. It is also fortunate that the ground-based instruments, that have necessarily a much smaller field of view with respect to satellite observatories like GLAST, do not suffer too much from limited field of view, at least for relatively cusped profiles. The dominant backgrounds of ACTs are indeed isotropically distributed, and for this case the optimal annulus size is $\lesssim 2^{\circ}\div 3^{\circ}$ (for cusped profiles), within the field of view range considered by present designs for future ACT. Figure 3: Plot of the curves with $S/N=3$ in the $10-100$ GeV range, with vertical lines in correspondence of $m_{Z}/2$ and $m_{W}$. The dot-dashed line shows the sensitivity achieved if one only includes the inner $0.1^{\circ}$ region around the GC for the NFW profile. The couples of solid/dashed lines (solid for conservative background, dashed for optimistic one) represent instead the sensitivity for the optimal annulus size for each model: from top to bottom, for Kravtsov, NFW and Moore profiles, respectively. ## V Discussion and Conclusions A large field of view instrument like GLAST promises to revolutionize the field of high energy gamma ray astrophysics. But it could also turn into a discovery instrument for DM, if it is made by WIMPs which annihilate into standard model particles in the halo of our galaxy. Since the DM signal grows like the square of its density, it is usually assumed that the Galactic Center, where the DM density is expected to be the highest, is the most promising region to focus on. However, two arguments conspire to modify this naive expectation: i) the presence of astrophysical point-like backgrounds close to the GC; ii) geometric considerations, in particular related to the different angular shape of the DM signal and astrophysical backgrounds. It is thus important to establish the level of the diffuse galactic and extragalactic diffuse backgrounds, a task which will be accomplished with exquisite precision by GLAST itself. Yet, by employing simple models based on EGRET data, we showed that it is always to be preferred a window size $>1^{\circ}$ around the GC, even for profiles as steep as the Moore one. The flatter the profile, the larger the optimal window size: for example we find an optimal galactic latitude cut $b_{\rm max}\raise 1.29167pt\hbox{$\;>$\kern-7.5pt\raise-4.73611pt\hbox{$\sim\;$}}10^{\circ}$ for NFW, $b_{\rm max}\raise 1.29167pt\hbox{$\;>$\kern-7.5pt\raise-4.73611pt\hbox{$\sim\;$}}30^{\circ}$ for Kravtsov profile. The discovery of additional point-like sources is not expected to change things dramatically, since especially at high energies the good angular resolution of GLAST will allow one to remove the quasi-totality of the resolved point-source contaminations with only a minor loss in the solid angle of the diffuse signal. When these geometric effects are taken into account and the search strategy is adjusted consistently, the prospects for DM detection improve considerably. For example, we found that models with sufficiently strong line annihilation signals, like the Inert Doublet Model, may be detectable without or with still reasonable boost factors, of the order of 10 even for Kravtsov-like profiles. Also, the perspectives to detect a DM signal become less sensitive to the unknown profile of the DM halo. In particular the $S/N$ of quasi-cored profiles is only a factor $\sim 2$ worse than for the NFW case. We also showed with one example that our considerations remain true also for the widely considered case of continuum gamma-ray spectrum from DM annihilation. Given the generality of these considerations, we believe they should be taken into account when estimating the reach of DM searches via gamma-ray telescopes. Acknowledgments. Authors are grateful to James Buckley and Dan Hooper for careful reading of the manuscript and useful comments. This work is supported in part by the US Department of Energy, Division of High Energy Physics, under Contract DE-AC02-06CH11357. P.S. is supported by the US Department of Energy and by NASA grant NAG5-10842. Fermilab is operated by Fermi Research Alliance, LLC under Contract No. DE-AC02-07CH11359 with the United States Department of Energy. ## References * (1) G. Bertone, D. Hooper and J. Silk, Phys. Rept. 405, 279 (2005) [hep-ph/0404175]. * (2) L. Bergstrom, Rept. Prog. Phys. 63, 793 (2000) [hep-ph/0002126]. * (3) D. Hooper and E. A. Baltz, arXiv:0802.0702 [hep-ph]. * (4) F. Stoehr, S. D. M. White, V. Springel, G. Tormen and N. Yoshida, Mon. Not. Roy. Astron. Soc. 345, 1313 (2003) [astro-ph/0307026]. * (5) http://confluence.slac.stanford.edu/download/attachments/19303/HEAD_Poster.pdf?version=1 * (6) R. Barbieri, L. J. Hall and V. S. Rychkov, Phys. Rev. D 74, 015007 (2006) [hep-ph/0603188]. * (7) M. Gustafsson, E. Lundstrom, L. Bergstrom and J. Edsjo, Phys. Rev. Lett. 99, 041301 (2007) [astro-ph/0703512]. * (8) L. Bergstrom, P. Ullio and J. H. Buckley, Astropart. Phys. 9, 137 (1998) [astro-ph/9712318]. * (9) W. M. Yao et al. [Particle Data Group], J. Phys. G 33, 1 (2006). * (10) D. Hooper and P. D. Serpico, JCAP 0706, 013 (2007) [astro-ph/0702328]. * (11) J. F. Navarro et al., Astrophys. J. 462 563 (1996). * (12) B. Moore et al., Phys. Rev. D 64 063508 (2001). * (13) A. V. Kravtsov, A. A. Klypin, J. S. Bullock and J. R. Primack, Astrophys. J. 502, 48 (1998) [astro-ph/9708176]. * (14) C. Power et al., Mon. Not. Roy. Astron. Soc. 338, 14 (2003) [astro-ph/0201544]. * (15) D. Reed et al., Mon. Not. Roy. Astron. Soc. 357, 82 (2005) [astro-ph/0312544]. * (16) J. F. Navarro et al., Mon. Not. Roy. Astron. Soc. 349, 1039 (2004) [astro-ph/0311231]. * (17) D. Merritt, J. F. Navarro, A. Ludlow and A. Jenkins, Astrophys. J. 624, L85 (2005) [astro-ph/0502515]. * (18) H. Yuksel, S. Horiuchi, J. F. Beacom and S. Ando, Phys. Rev. D 76, 123506 (2007) [arXiv:0707.0196 [astro-ph]]. * (19) M. Fornasa and G. Bertone, arXiv:0711.3148 [astro-ph]. * (20) G. Jungman, M. Kamionkowski and K. Griest, Phys. Rept. 267, 195 (1996). * (21) A. R. Pullen, R. R. Chary and M. Kamionkowski, Phys. Rev. D 76, 063006 (2007) [astro-ph/0610295]. * (22) T. Bringmann, L. Bergstrom and J. Edsjo, arXiv:0710.3169 [hep-ph]. * (23) http://www-glast.slac.stanford.edu/ * (24) http://www.mpi-hd.mpg.de/hfm/CTA/ * (25) http://cherenkov.physics.iastate.edu/wp/ * (26) F. Aharonian et al. [The HESS Collaboration], astro-ph/0408145; J. Albert et al. [MAGIC Collaboration], Astrophys. J. 638, L101 (2006) [astro-ph/0512469]; K. Kosack et al. [The VERITAS Collaboration], Astrophys. J. 608, L97 (2004) [astro-ph/0403422]; K. Tsuchiya et al. [CANGAROO-II Collaboration], Astrophys. J. 606, L115 (2004) [astro-ph/0403592]. * (27) C. van Eldik, O. Bolz, I. Braun, G. Hermann, J. Hinton and W. Hofmann [for the HESS Collaboration], arXiv:0709.3729 [astro-ph]. * (28) D. Hooper and B. L. Dingus, Phys. Rev. D 70, 113007 (2004) [astro-ph/0210617]; D. Hooper and B. Dingus, Proc. of the 34th COSPAR Scientific Assembly, Houston, Texas (2002), astro-ph/0212509. * (29) M. Pohl, Astrophys. J. 626, 174 (2005). * (30) S. Dodelson, D. Hooper and P. D. Serpico, Phys. Rev. D 77, 063512 (2008) [arXiv:0711.4621 [astro-ph]]. * (31) M. Kachelrieß and P. D. Serpico, Phys. Lett. B 640, 225 (2006) [astro-ph/0605462]. * (32) S. D. Hunter et al. [EGRET Collaboration], Astrophys. J. 481, 205 (1997). * (33) P. Sreekumar et al. [EGRET Collaboration], Astrophys. J. 494, 523 (1998) [astro-ph/9709257]. * (34) F. W. Stecker and M. H. Salamon, astro-ph/0104368. * (35) L. E. Strigari, S. M. Koushiappas, J. S. Bullock and M. Kaplinghat, Phys. Rev. D 75, 083526 (2007) [astro-ph/0611925]. * (36) G. Zaharijas and D. Hooper, Phys. Rev. D 73, 103501 (2006). [astro-ph/0603540]. * (37) J. H. Buckley et al., whitepaper for “Dark Matter Searches with a Future VHE -Ray Observatory”, in preparation.

arxiv-papers

2024-09-04T02:48:53.957435

0802.3268

11institutetext: Geneva Observatory, University of Geneva, 1290 Sauverny, Switzerland # Chaos in Galaxies Daniel Pfenniger ###### Abstract After general considerations about limits of theories and models, where small changes may imply large effects, we discuss three cases in galactic astrophysics illustrating how galactic dynamics models may become insufficient when previously neglected effects are taken into account: 1. 1. Like in 3D hydrodynamics, the non-linearity of the Poisson-Boltzmann system may imply dissipation through the growth of discontinuous solutions. 2. 2. The relationship between the microscopic exponential sensitivity of N-body systems and the stability of mean field galaxy models. 3. 3. The role of quantum physics in the dynamics of structure formation, considering that cosmological neutrinos are massive and semi-degenerate fermions. ## 1 Introduction ### 1.1 Chaos in science Chaos plays a key role in many sciences, but particularly in galactic astrophysics where it appears under different aspects in various galaxy models. Before discussing chaos in galaxies, some space will be spent discussing chaos in science. The notion of chaos has deeply modified the scientist view of Nature, but also of the scientific process itself. But what makes chaos so special? In sciences chaos expresses commonly two important properties of dynamical models of natural phenomena: 1. 1. The sensitivity to initial conditions. Initially close solutions of a dynamical model separate in average at least exponentially with time, $\langle\lVert\Delta x(t)\rVert\rangle>\exp(t/\tau)$, such that after a few characteristic Liapunov time $\tau$ the small scale effects neglected either in the model, or in the numerical model of the mathematical model, become macroscopic. This restricts a deterministic use of chaotic models over a model dependent finite time, and means that in chaotic systems Laplace’s determinism is only applicable approximately over a limited time. 2. 2. The rapid mixing of the model solutions. An exponential divergence of nearby solutions is not sufficient to produce chaos in the technical sense. Indeed, if a uniformly diverging flow preserves the neighbourhood of solutions, a smooth transformation of coordinates allows to remove the formal divergence back: the regularized model is then no longer diverging. Therefore the divergence effect may be model dependent. To be truly chaotic, a system should remain chaotic even after smooth transformations of coordinates. It should mix the solutions, such that after a few Liapunov times not only initially close solutions become widely separated by the exponential divergence, but some initially distant solutions come close together at some time. With time, most of solutions become at least once arbitrarily close to each other, which expresses the mixing property of chaotic systems. Mixing occurs frequently in systems with bounded phase space, while in systems with unbounded phase space escaping solutions, i.e., solutions going to infinity, do not return often in almost all phase space regions. Chaos is therefore important because it touches to an essential aspect of the scientific activity: to represent faithfully natural phenomena with formal models. At the heart of the scientific process is the determination of the domain of applicability and the limits of models and theories. Actually we can enlarge the notion of sensitivity of solutions to initial conditions to the sensitivity of solutions to functionally close dynamical systems. The former case is a particular case of the latter one when the initial displacement has been produced by an initial impulse perturbation. Therefore in chaotic systems not only possible perturbations of the initial conditions lead to unpredictability, but also close but distinct functional approximations adopted in the model may lead to very different solutions after a finite time. Therefore when considering chaos in natural systems, it is not only important to discuss how the ignorance of the real initial conditions influences the model predictions, but also how the functional “distance” from the real system to the modelled one participates to unpredictability. This point is actually central to physical modelling, because a good model should be robust to perturbations of the model functional form, since a model is always an approximation of the real system. ### 1.2 Epistemological digression, theories vs. models Until recent times theories and models could be viewed as fundamentally different: physical theories, like quantum physics or general relativity, have been thought to be more fundamental than models, because applicable to many more different cases. For a long time, since about the rise of classical mechanics in the 17–18th centuries, theories were even taken as absolutely exact. Isaac Newton’s “laws” were widely considered as absolute at the same level as a God given, revealed truth. In the meanwhile, classical mechanics was found to be only an approximation of Nature, and had to be corrected by relativistic and quantum effects. Further, the new physical theories also could not claim to be exact, because quantum physics and general relativity have remained incompatible since then, the theory of quantum gravity is still a work in progress. The so-called fundamental physical theories, like the elementary particle Standard Model, are today known to be incomplete, and must all be considered as approximations of Nature. Models, like a galaxy or star model, concern often more specific phenomena than theories. Models are simplified formal approximations of particular phenomena, while theories aimed in the past to be exact isomorphisms with reality, but aren’t. In models, scientists deliberately simplify reality in order to keep essential features and discard inessential ones. Doing so allows to describe in a formal way, with usual mathematics but more and more with computer programs, the gist of the phenomena. For example, a planet or a star may be approximated first by a perfect sphere, which allows to concentrate the discussion on its most essential features that we want to understand, and leave out the inessential aspherical features, like rotation induced flattening, or mountains. This simplification or reality has the important virtue to reduce the complexity of phenomena to a level compatible with the finite capacity of human brains to grasp complex systems. In the end, the subjective feeling of understanding a phenomenon comes from a simplified, i.e., inexact, description that captures essential features and discards inessential ones. This essential-inessential separation is often not unique, contains subjective assumptions, and of course is constrained by our finite brain capacity, which is not uniquely specified among human beings. A minority of models and theories are successful, and often only for a limited time. The scientific process, like biological evolution, is selective. In this view, the understanding of Nature by human beings looks no longer as much miraculous as Albert Einstein thought111“The eternal mystery of the world is its comprehensibility…The fact that it is comprehensible is a miracle.” Albert Einstein. All theories are derived by human beings from the search of a formal simplified description of Nature that is both as faithful as possible, but also that is adapted to the brain of at least some other human beings. Furthermore, accepted theories are repeatedly checked, scrutinized for correctness by the scientific community, so in the end only effective theories survive. At this stage it is worth to put some attention to mathematics, which is used as the most solid formal language in physical models and theories. Mathematics is not a frozen field and its view by mathematicians has deeply changed during the 20th century. Since about the Greek school of mathematics around $-300$ BC, and particularly since Euclid, mathematics and logic have been based on a limited set of axioms from which all the theorems were expected to follow with rigorous demonstrations. But during the 20th century, famous mathematicians like Kurt Gödel, Alan Turing, and Gregory Chaitin showed however that such an axiomatic mathematics would always be incomplete (see chaitin for an introduction to these problems and for references). Not by a little part, since actually most theorems (“mathematical facts”) escape demonstration when starting from a limited set of axioms. Since physical theories try to build an isomorphism between Nature and a subset of mathematics, it follows that no “theory of everything” (TOE) can summarize with provable derived theorems the implicit complexity of its content. Similar ideas have been expressed by Stephen Hawking222http://www.damtp.cam.ac.uk/strings02/dirac/hawking/. Therefore even if a theory would be exactly true, most of the consequences of it would escape an axiomatic description. Only left is the possibility to compute by brute force methods the consequences of theories. This is why the exponentially growing capacity of computers plays an increasingly important role in science, because computers allow to explore better and better the vast domain of theories that were out of reach by older methods using mainly theorems (like analytical methods). The best illustration of the power of the computer approach vs. the traditional analytic approach is the very notion of deterministic chaos that arose from computer models in the 1960’s henon ; lorenz . ## 2 Chaos in galaxies Let us now apply these above general ideas to galaxies. These particular structures in the Universe offer an excellent case illustrating the scientific method at work applied to complex phenomena where chaos takes an important part. Galaxies indeed contain a large variety of physical problems at widely different scales, most of them chaotic at some level, which have been more or less successfully been described by distinct models. No galaxy model may claim to describe fully and exactly the galaxies, but the rich variety of models provide complementary descriptions of galaxies that overall improve our understanding of these objects. With these models, all deliberately simplified for making a part of galaxies understandable, we can much better think about galaxies than without. Let us mention some types of chaos in galaxies existing at very different levels. ### 2.1 Chaos linked to Newtonian dynamics Before and even after the discovery of interstellar matter and dark matter ($>1930$), galaxies have been described by purely stellar dynamical models of point masses following classical mechanics, the N-body model. Such a simplified description already demands substantial efforts to extract useful information helping us to understand how galaxies work. For a long time the mean field approximations, models of the N-body model, were studied, with further time-independence and spherical or axisymmetric symmetry assumptions: 1. 1. Orbit description. Assuming a fixed, smooth mean field potential allows to decouple the motion of point masses. At this stage one considers all the possible individual orbits in the galactic potential. As soon as the potential departs from a spherical shape, or differs from some particular functional form like the Stäckel potential family (for example due to the presence of a rotating bar), orbital chaos appears and occupies large regions of phase space associated to resonances. The neglected granularity of the mass become then relevant on chaotic orbits, the relaxation process estimated in integrable potentials is much faster pfenniger86 . Over several decades George Contopoulos and collaborators (e.g., contopoulos80 ; contopoulos_patsis ; patsis ; voglis99 ; voglis07 ) have explored by numerical means the orbital complexity in the phase space of galactic potentials, illustrating well how numerical “brute force” orbit integration did allow to make understandable a part of the complexity of the mean field model of galaxies. In contrast, traditional analytical approaches have been much less efficient in extracting the information implicitly contained in the classical mechanics models of galaxies. 2. 2. Phase space fluid description. Another approach is to approximate by infinity the large but finite number of stars or weakly interacting dark matter particles in a galaxy. In addition, these collisionless particles are supposed to be smoothly distributed in phase space, and make a continuous and differentiable flow of matter in the 6-dimensional phase space of space and velocity coordinates. The differentiability of such a collisionless flow is a strong hypothesis that is by far not justified, neither by the observations of the stars in the solar neighbourhood, where stellar streams abound, nor by the usual arguments used in collisional flows in which smoothness is expected to arise through the microscopic chaos resulting from the frequent particle collisions (the relaxation time is short), erasing quickly irregularities and decorrelating particles. Further, one should also keep in mind that the number of stars in galaxies ($<$$10^{12}$) is actually not very large for a smooth fluid description in a 6-dimensional phase space, and lies also in part in massive star clusters. If one would represent phase space with cells each containing, say, at least 100 particles in order to have reasonably smooth average quantities between contiguous cells, one would obtain a number of bins per coordinate of only $({10^{12}/10^{2}})^{1/6}\sim 50$. Dark matter particles are more numerous but are not expected to contribute much density in the optical part of galaxies like the Milky Way, therefore the graininess of the mass distribution in galaxies is certainly already a difference with the smooth model that must be taken into account. 3. 3. N-body description. The full N-body model of a galaxy is a much more faithful model than the two previous ones, at least when one wants to describe a galaxy with about $10^{12}$ point masses. In a not too distant future, 5–10 years, it is likely that it will be possible to integrate such as number of particles with the forces calculated by tree approximation techniques. Therefore the N-body model has certainly a bright future and will make the collisionless smooth flow model less relevant in galactic dynamics, and even less in star cluster dynamics. However the N-body model is also limited. a) The Miller’s exponential instability miller64 of individual particle trajectories and of the whole N-body system constrains to interpret N-body simulations in a statistical way. An ensemble of close but otherwise uncorrelated initial conditions produces an ensemble of round-off errors dependent evolutions, which may contain statistical useful information about typical representative evolutions. b) Another important limitation of the N-body approach when the bodies are supposed to represent real stars is that one often neglects the internal evolution of stars. At formation, most of the mass transformed into stars consists of small mass stars that eject a large fraction of their mass in the red giant phase several Gyr later. This ejected gas from planetary nebulae mixes mass, orbital momentum and kinetic energy in the interstellar gas, a highly energy dissipative process. Galactic models that ignore this dissipative aspect obviously miss an important part of reality over long time scales, especially in 5–12 Gyr old systems like elliptical galaxies that have been considered for a long time as prototypical systems for using pure dissipationless N-body dynamics. In fact ellipticals, like spiral galaxies but for different physical reasons, should be seen as substantially dissipative systems when described over several Gyr. ### 2.2 The complex physics of baryons Besides stars, galaxies contain gas in sometimes large amount, even exceeding the stellar content. Some very gas rich galaxies like Blue Compact Dwarfs galaxies are misnamed: they are called dwarf only because the visible stellar mass is tiny, but when the gas content revealed by HI emission is considered, they appear just as massive as normal galaxies, except for the fact that gas has not yet been turned into stars. So for galaxies even with lesser amount of gas, stellar dynamical models are more or less rapidly invalidated by the rest of the physics that baryons can be subject to. The interstellar gas physics is very complex and far from being under control. Typical interstellar gas is multiphased, has supersonic turbulence, and density and temperature contrasts covering several orders of magnitudes, a very chaotic state that defies description with the present physics tools. For example, thermodynamics supposes for its use that a local thermal equilibrium can be established, provided a local mechanical equilibrium has been reached. But supersonic turbulence means precisely that strong pressure gradients are ubiquitous, out of mechanical equilibrium regions frequent. Despite such incoherences, thermodynamical quantities like temperature are used in models and observations due to a lack of better theoretical tools about supersonic compressible turbulence. What is apparent from simple order of magnitude estimates is that dissipative effects and the exchange of energies between the stars and the gas is not negligible for the whole galaxy equilibrium over Gyr timescales. For example the power radiated by the stars at their different stages of evolution, known to be partly recycled by the dust in the infrared, or known to feed a part of the turbulence in the interstellar medium, is comparable to the power necessary to change the whole galaxy shape against its own gravity pfenniger91 . Therefore, the galaxy global parameters and shapes can be expected to depend also on its internal dissipative micro-physics, and not only on the initial conditions at earlier epochs, or external effects like accretion. ### 2.3 The dynamics of non-baryonic matter Solid cosmological and particle physics arguments exist for the existence of large amount of non-baryonic matter. 1 s after the Big Bang a number of neutrinos comparable to the photon number must have been produced mostly from electron-positron annihilations. The involved physics is the well known, far from exotic MeV nuclear physics. The discovery of neutrino oscillations between the e-, $\mu$\- and $\tau$-neutrinos was a proof of their positive mass, and solved the 40 year old solar neutrino deficit. With the present constraints about the neutrino mass ($\sim 0.01-0.1$ eV), this average leptonic density turns out to be comparable to the average identified baryon density pfenniger06 . The neutrino case shows that a particle predicted by Pauli in the 30’s for resolving an apparent violation of energy conservation during the $\beta$ decay demanded huge effort to arrive to the present solid conclusion that indeed much matter is in non-baryonic form. The gained knowledge about neutrinos suddenly doubles the amount of identified matter, which is this time leptonic. With similar arguments, many other particle candidates (axions, neutralinos, super-symmetric particles, …) have been proposed, often with strong theoretical motivations based on symmetries and conservation laws. For example, axions are invoked for explaining the zero neutron electric dipole moment, an empirical fact that escapes predictions of the Standard Model. Therefore, in view of all the oddities remaining to be explained in elementary particle physics, it is natural to expect a rich variety of different dark matter components that remain to be identified. The consequence for galaxies is that each kind of matter can imprints different effects during structure and galaxy formations. Some phases of structure formation, like during the formation of Zel’dovich’s pancakes, are highly sensitive to the neglected physics. Yet a high fraction of matter is expected to participate at least once, even over a brief time interval, to a sheet-like singularity where the outcome of such highly non-linear singularities is known to be very sensitive to the exact physics of the participating matter, so also from non-baryonic particles. The often adopted collisionless property of cold dark matter is just an assumption that may be acceptable in present day galaxy models, but may turn out to be invalid during perturbation sensitive events like pancake or filamentary collapses. ## 3 Cases of Sensitive Dependence in Galaxy Models In the following we will concentrate on illustrative cases of sensitivity to perturbations of galactic models, where slight changes in the model may turn out to lead to radically different conclusions. We will discuss the perfectly smooth fluid phase space description of the star ensemble and of collisionless matter used in the collisionless Boltzmann equation, where the collisionless limit may turn out to lead to severe approximations. The discrete point mass models used in N-body simulations is also limited by its strongly chaotic character. Finally the role of quantum mechanics at extra-galactic scales related with cosmological neutrinos and possibly other relic dark matter particles will be argued to be not so negligible as usually assumed. ### 3.1 Collisionless Chaos The main equation of collisionless galactic dynamics is the collisionless Boltzmann equation. Suppose that we describe the mass density at the instant $t$ in space $x\in\mathbb{R}^{3}$ and velocity space $v\in\mathbb{R}^{3}$, i.e., $\\{x,v\\}\in\mathbb{R}^{6}$ by a density distribution $f(t,x,v)\in\mathbb{R}$. The projection of $f$ onto the $x$-space $\rho(x,t)=\int\\!d^{3}v\,f(x,v,t),$ (1) provides the usual mass density $\rho(x,t)$. This projection is well defined even when $f$ is not differentiable. Poisson’s equation gives us a constraint on the gravitational acceleration $g$ induced by the mass density $\rho$, $\nabla g(x,t)=-4\pi G\rho(x,t)=-4\pi G\int d^{3}\,f(x,v,t)\ .$ (2) Finally, the collisionless Boltzmann equation $\partial_{t}f+v\cdot\partial_{x}f+g\cdot\partial_{v}f=0,$ (3) tells us that the mass flow in $\mathbb{R}^{6}$ is conserved, the characteristics curves of this equation are the trajectories of particles in the acceleration field $g$. The above three equations forms a system of non­linear integro-differential equations, similar to Euler’s equation in $\mathbb{R}^{3}$ for incompressible fluids. Many efforts have been dedicated by mathematicians to understand the Euler or Navier-Stokes equations with more rigorous tools that commonly used in physics. This has been useful to understand much more general facts about non- linear partial differential equations, such as the limits of their applicability. The simplest case taken from the Navier-Stokes equation but still preserving its non-linear character is the 1-dimensional Burger’s equation, $\partial_{t}u+u\partial_{x}u=\mu\partial^{2}_{x}u\ ,$ (4) which describes a constant density advection flow at velocity $u(x,t)$ along the direction $x$, with an optional viscosity term proportional to a parameter $\mu$ on the right-hand side. When $\mu=0$ the system corresponds to the energy conserving Euler equation. For simple initial conditions, say $u(x,0)=-\sin(x)$, the solution becomes multi-valued after a finite time when $\mu=0$, and develops a shock, a discontinuity when $\mu>0$. This shows a prototypical behaviour of non-linear partial differential equations: they tend to break the initial assumptions about the solution after a finite time by violating the assumption that the solution remains continuous and differentiable everywhere. We have here an example of sensitive dependence on the functional form of the flow model. If viscosity is zero the flow develops multi-valued velocities, while if viscosity is small but positive the flow remains single-valued but becomes discontinuous, and most of the energy is dissipated in the shocks. An old but relevant result by Onsager onsager 333An extensive review about Onsager’s work on turbulence is given in eyink . about 3-dimensional turbulence is that energy dissipation in Navier-Stokes fluids does not vanish to zero when viscosity tends toward zero. As viscosity tends toward zero, Navier-Stokes’ equation tends well toward Euler’s equation, but the solutions don’t. The most astonishing fact verified in experiments of developed turbulence is that the energy dissipation tends toward a positive constant independently of the value of viscosity. As in Burger’s equation entropy increasing discontinuous solutions (so-called “weak” solutions) are physically the relevant ones. Therefore it may be physically misleading to consider Euler’s equation and its energy conservation law, for application on systems where viscosity is small, because even a small viscosity term becomes essential when the flow becomes discontinuous, highly turbulent, which is precisely the rule in low viscosity fluids. There is no ground to believe that the growth of discontinuities is restricted to non-linear hydrodynamics. On the contrary, the growth of shocks and discontinuous solutions occur frequently in other non-linear partial differential equations. One should expect shocks in more complex, higher dimensional non-linear systems like the Poisson-Boltzmann system, in which the collisional or diffusive term is small, but is never exactly zero. In cases of strong phase space “turbulence” during collapses and violent events we can expect that the small residual collisionality has its diffusive effects strongly amplified. High phase space density gradients and multi-streams are susceptible to develop fast from the actual particle noise. Also we should remind that differentiability in usual fluids is often a valid assumption because the microscopic molecular chaos does erase the growth of correlations faster than the flow develops gradients of density or velocity. Except for very particular cases, laminar flows do require some positive viscosity, low Reynold’s number, to stay laminar. Otherwise low viscosity fluids become spontaneously turbulent, i.e., develop discontinuous, singular flows. Since precisely collisionless flows in galactic dynamics lack of a strong microscopic collisional chaos that would justify the usual smoothness assumption of distribution functions, we should rather expect irregular, non- smooth distributions (“weak” solutions) as a rule in galaxies. Actually, the local distribution function of stars in the solar neighbourhood is highly structured with several star streams, and is far from resembling a Maxwellian distribution. What we see is at best a partly relaxed distribution. To explain the smoothness in the galaxy distribution functions, the violent relaxation concept lynden-bell has been proposed. It was initially attributed to the time-dependence of the gravitational potential, but today we view it rather as resulting from the highly sensitive to perturbations, chaotic stages of galaxy evolution where microscopic perturbations become fast macroscopically relevant. Contrary to a still popular opinion, this is not directly the time-dependence of the potential that leads to an enhanced relaxation, but the highly chaotic, sensitive stage of the system. Counter- examples demonstrating that time-dependence does not necessarily relax a collisionless flow are analytical time-dependent periodic solutions of the Poisson-Boltzmann system (e.g., sridhar ). ### 3.2 N-body Chaos Miller (1964) miller64 discovered numerically the exponential divergence of particular gravitational N-body systems, and noticed that the divergence of close systems occurs not because of transient close 2-body encounters, but constantly by the $N(N-1)$ particle interactions. Gurzadyan & Savvidy (1986) gurzadian showed also with Riemannian geometry that the N-body problem is indeed generally chaotic in simple particle distributions. However these studies should not be seen as definitive, it is indeed not difficult to invent particular counter-examples of as weakly unstable as wished configurations of N-body systems, such as widely separated pairs of pairs etc. of binaries. By natural selection we do observe in the sky the least unstable multiple star systems, often arranged hierarchically. The solar system is also an example where the Liapunov time is much longer than its dynamical time.. An interesting problem is to specify the relationship between Miller’s type chaos seen at the microscopic level, and the global stability of a stellar system. In usual gases, the molecular very rapid chaos is the key property that guarantees that the system seen at macroscopic scales can be modelled with the quasi-deterministic rules of thermodynamics. To be effective thermodynamics requires a fast relaxation of molecules, in other words, a strong molecular chaos. Is it similar in gravitational systems? From numerical experiments, systems like hot spherical models of many equal mass stars are examples where indeed a kind of statistical robust state appears to be reached over time-scales longer than the crossing time. For these systems Miller’s microscopic chaos could be actually favouring a global statistical quasi- equilibrium. Other gravitational systems, such as disks, or systems with strongly anisotropic velocity dispersions, can present macroscopic instabilities leading to evolution. It is presently unclear whether Miller’s microscopic chaos is related in any way with large scale instabilities. ### 3.3 Sensitivity to Quantum Physics The relict cosmological neutrinos are fermions and have been produced $\sim 1-2$ s after the Big Bang from electron-positron annihilations at a redshift of $z\approx 10^{10}$. At this epoch all the particles were strongly relativistic and close to thermal equilibrium. This means that the distribution of neutrinos had to be very close to a Fermi-Dirac distribution with a negligible chemical potential. In other words, the natural creation state of such relativistic particles is always semi-degenerate weinberg , which means that quantum mechanics plays a significant role from the start. The phase space density of neutrinos is therefore sufficiently high for quantum effects to be important at the macroscopic level. As the Universe expands the neutrino phase space density is little modified, since neutrinos have only two weak possibilities of interaction with matter, the weak force and the gravitational force, and their number cannot decrease through decay since they are supposed to be stable. For a rest mass range such as $0.01-0.1$ eV at the present time the relict neutrinos should be non-relativistic and much colder than the often quoted temperature of 1.9 K, which would apply if they were massless or still relativistic pfenniger06 . Their speed is estimated low enough ($\sim 1000\,\rm km\,s^{-1}$) to be trapped at least in galaxy clusters. So if neutrinos are massive and participate to structure formation, clearly they are able to perturb the outcome of pancake collapses since their total mass is comparable to the one of baryons, and they are still today semi-degenerate fermions, so their physics differs from classical ballistic particles. Structure formation is a highly chaotic process where most of the matter goes through Zel’dovich’s pancakes, so one can expect that any perturbation to a purely classical model of structure formation, like when including the fermionic properties of neutrinos, may significantly change the results. But here we have the non-trivial challenge to merge two widely different descriptions of Nature. Sometimes quantum physics is said to be “holistic” because it is a non-local theory. A “particle” does not necessarily represent a localized mass, but can be a plane-wave. Pauli’s principle does not require to localize particles, and the wave-function in Schrödinger equation may extend over all space. In contrast, the notion of localized point mass is central in classical mechanics. In the recent years these questions have considerably advanced. The decoherence theory zeh70 ; zeh05 ; zurek02 ; zurek03 allows to specify when a finite system, supposed to be isolated, is well represented by quantum physics, and when classical physics is the best description: when the neglected but always existing perturbation rate of the outer world to a given model stays indeed negligible, quantum coherence build-up wins and quantum mechanics stays a faithful representation of the system. But when the outer world perturbation rate is faster and destroys quantum coherence, classical physics emerges as a good description of the system. A satisfactory aspect of the decoherence theory is that a conscious observer is no longer required. A measurement corresponds to coherence destruction by external perturbations to the system. Such considerations are central for building quantum computers, where coherence preservation is crucial, and clearly unrelated with the presence of a conscious observer. So in order to describe a particle as a localized mass distribution with classical mechanics, as implicit in all cosmological simulations up to now, one needs to check that decoherence is effective. Otherwise particles can not be localized and must be described with quantum physics as non-local ensembles. But since neutrinos have extremely low probability to interact with other particles by the weak interaction, the only remaining possibility is gravitational interaction. As long as the Universe is homogeneous, gravitational interaction cancels and is negligible. Only when structures form ($z\ll 100$) matter inhomogeneities perturb neutrinos by gravitation. The entanglement time-scale $\tau_{E}$ of identical neutrinos is estimated by a classical collision time where the cross section diameter is given by the de Broglie wavelength $\lambda_{\rm dB}=h/m_{\nu}v$ (see pfenniger06 ), $\tau_{E}\approx(n_{\nu}\lambda_{\rm dB}^{2}v_{\nu})^{-1}\approx 1.3\cdot 10^{-8}(1+z)^{-2}\left(\frac{n_{\nu}}{56}\right)^{-1}\left(\frac{m_{\nu}}{1\rm eV}\right)^{-1}\rm s,$ (5) where $n_{\nu}$ is the neutrino density, $m_{\nu}$ the neutrino mass, and $z$ the redshift. After a multiple of $\tau_{E}$ large numbers of neutrinos are entangled just because they are identical fermions, not because of their weak interaction. The decoherence theory zeh70 ; zeh05 ; zurek02 ; zurek03 gives an estimate of the decoherence time $\tau_{D}$ if we know the characteristic relaxation time $\tau_{R}$ over which the external world perturbs the system of size $\Delta x$: $\frac{\tau_{D}}{\tau_{R}}=\left(\frac{\lambda_{\rm dB}}{\Delta x}\right)^{2}.$ (6) For ensembles of neutrinos above a given size $\Delta x$ the decoherence time $\tau_{D}$ will be shorter than the entanglement time $\tau_{E}$, and above such a scale the neutrino ensemble may be considered as localized and included in a classical description, like hydrodynamics. In pfenniger06 we estimated that at the present epoch the fastest relaxing mass condensations for neutrinos are the galaxies, not the stars or the galaxy clusters. With this estimate of the “relaxation” time $\tau_{R}$ produced by the external world, we found that when considering scales much larger than $\sim 10^{13}$ cm, the ensemble of at least $10^{40}$ neutrinos is perturbed at a rate fast enough by the galaxy gravitational interactions to be considered as a classical fluid with a Fermi-Dirac equation of state. Thus a simple model of cosmological neutrinos above solar system scales is to describe them as a Fermi-Dirac fluid, like what is done in stellar models of white dwarfs and neutron stars. This is a very different physics than the collisionless classical mechanics representation that up to now has been used in cosmological simulations including neutrinos, but closer to the adhesion model sometimes used in pancake models. Of course the consequences for structure formation models may be drastic because fluids tend to develop shocks, contrary to collisionless flows, and shocks imply entropy production and dissipation. What has been said about neutrinos may be applicable to other dark matter particles like axions or neutralinos, with possible complications for bosons. Much depends on their effective rest mass, number density, and if they are sufficiently non-relativistic to participate to structure formation. ## 4 Conclusions The sensitivity to perturbations is a characteristics of chaos. Various aspects of galaxies can be represented by different models that may contain sensitive parts subject to limitations on the scope of applicability. Here are a few points to consider for future galaxy models: * • The assumed smoothness of distribution functions for collisionless systems is not grounded on theoretical arguments or observational evidences. The effects of irregular, non-smooth distributions on the dynamics of a model have been little investigated. The results of incompressible fluid turbulence could be useful to extend to collisionless phase space flows. * • Therefore the collisionless Boltzmann differential equation is a decreasingly attractive model in galactic dynamics in regard of the fast growing capabilities of the much less constrained N-body techniques. * • The dissipative baryonic physics in galaxies coming from the gas but also from the stellar evolution is important over Gyr timescale, not only for spirals, but also for ellipticals due to the stellar important mass loss in the red giant phase. * • Quantum physics in the semi-degenerate sea of cosmological neutrinos produces a kind of collisional relaxation due to decoherence and the fermionic nature of such particles. These neutrinos and possibly other dark matter particles should not be modelled as collisionless, but as a collisional fluid with the proper quantum statistics equation of state. This point is crucial during pancake shocks. ### Acknowledgements I am grateful to the organisers of this stimulating interdisciplinary conference for the invitation. This work has been supported by the Swiss National Science Foundation. ## References * (1) Chaitin, G.J. 1982, Int. J. Theor. Phys. 21, 941 (http://cs.umaine.edu/~chaitin/georgia.html) * (2) Contopoulos, G. 1980, Astron. & Astrophys. 81, 198 * (3) Contopoulos, G., Patsis, P. A. 2006 Monthly Not. Royal Astron. Soc. 369, 1039 * (4) Eyink G.L., Sreenivasan, K.R. 2006, Rev. Mod. Phys, 78, 87 * (5) Gurzadyan, V. G., Savvidy, G. K. 1986, Astron. & Astrophys. 160, 203 * (6) Hénon, M., Heiles, C. 1964, Astron. J. 69, 73 * (7) Lorenz, E. N. 1963, J. Atmos. Sci. 20, 130 * (8) Lynden-Bell, D. 1967, Monthly Not. Royal Astron. Soc. 136, 101 * (9) Miller, R. H. 1964, Astrophys. J. 140, 250 * (10) Onsager, L. 1949, Nuovo Cimento, Suppl. 6, 279 * (11) Patsis, P. A., Contopoulos, G., Grosbøl, P. 1991, Astron. & Astrophys. 243, 373-380. * (12) Pfenniger, D. 1986, Astron. & Astrophys. 165, 74 * (13) Pfenniger, D. 1991, in “Dynamics of Disc Galaxies”, ed. B. Sundelius, Göteborg, 389 * (14) Pfenniger, D., Muccione, V. 2006, Astron. & Astrophys. 456, 45 * (15) Sridhar, S. 1989, Monthly Not. Royal Astron. Soc.238, 1159 * (16) Voglis, N., Contopoulos, G., Efthymiopoulos, C. 1999, Cel. Mech. & Dyn. Astron. 73, 211 * (17) Voglis, N., Harsoula, M., Contopoulos, G. 2007, Monthly Not. Royal Astron. Soc. 381, 757 * (18) Weinberg S. 1962, Phys. Rev. 128, 1457 * (19) Zeh, H.D. 1970, Found. Phys. 1, 69 * (20) Zeh, H.D. 2005, Séminaire Poincaré, 2, 1 (quant-ph/0512078) * (21) Zurek, W.H. 2002, Los Alamos Science 27, 2 * (22) Zurek, W.H. 2003, Rev. Mod. Phys., 75, 715

arxiv-papers

2024-09-04T02:48:53.962838

0802.3423

# Correlations in the properties of static and rapidly rotating compact stars B. K. Agrawal bijay.agrawal@saha.ac.in Saha Institute of Nuclear Physics, Kolkata - 700064, India. Raj Kumar and Shashi K. Dhiman shashi.dhiman@gmail.com Department of Physics, Himachal Pradesh University, Shimla - 171005, India. ###### Abstract Correlations in the properties of the static compact stars (CSs) and the ones rotating with the highest observed frequency of 1122Hz are studied using a large set of equations of state (EOSs). These EOSs span various approaches and their chemical composition vary from the nucleons to hyperons and quarks in $\beta$-equilibrium. It is found that the properties of static CS, like, the maximum gravitational mass $M_{\rm max}^{\rm stat}$ and radius $R_{1.4}^{\rm stat}$ corresponding to the canonical mass and supramassive or non- supramassive nature of the CS rotating at 1122 Hz are strongly correlated. In particular, only those EOSs yield the CS rotating at 1122Hz to be non- supramassive for which $\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{1/2}\left(\frac{10{\rm km}}{R_{1.4}^{\rm stat}}\right)^{3/2}$ is greater than unity. Suitable parametric form which can be used to split the $M_{\rm max}^{\rm stat}$ $-$ $R_{1.4}^{\rm stat}$ plane into the regions of different supramassive nature of the CS rotating at 1122Hz is presented. Currently measured maximum gravitational mass 1.76$M_{\odot}$ of PSR J0437-4715 suggests that the CS rotating at 1122Hz can be non-supramassive provided $R_{1.4}^{\rm stat}\leqslant 12.4$ km. ###### pacs: 26.60.+c,91.60.Fe,97.10.Kc,97.10.Nf,97.10.Pg ## I Introduction The accurate knowledge of the properties of static and rotating compact stars (CSs) are of utmost importance to probe the behaviour of the equation of state (EOS) of superdense matter. Even the accurate information on the maximum gravitational mass $M_{\rm max}^{\rm stat}$ for the static CS and its radius $R_{1.4}^{\rm stat}$ with the canonical mass ($1.4M_{\odot}$), not yet well known, would narrow down the choices for the plausible EOSs to just a few. The newly measured CS mass 1.76$\pm 0.20M_{\odot}$ of PSR J0437-4715 Verbiest et al. is obtained by the precise determination of the orbital inclination angle, the highest measured mass for any known pulsar to date. Recent observations of the thermal emission from the quiescent LMXB X7 in the globular cluster 47 Tuc yield the value of R1.4 to be 14.5${}^{+1.8}_{-1.6}$ km Heinke et al. (2006). The binary pulsars PSR J0737-3039A,B with masses of the individual star being $1.338M_{\odot}$ and $1.249M_{\odot}$ are plausible candidates for the measurement the moment of inertia due to the spin-orbit coupling effects Lyne et al. (2004). It is expected that a reasonably accurate value of $R_{1.4}^{\rm stat}$ can be deduced from the moment of inertia measurement of PSR J0737-3039A Morrison et al. (2004). Rotating CSs discovered until recent past have periods of rotation close to one millisecond. The first millisecond pulsar PSR B1937+214 rotating at the frequency $\nu=641Hz$ was discovered in 1982 Backer et al. (1982). In year 2006, a more rapid pulsar PSR J1748-2446ad rotating at $\nu=716$ Hz was detected Hessels et al. (2006). Such frequencies are too low to affect significantly the structure of CSs with $M>1M_{\odot}$ Shapiro and Teukolsky (1983), since, these CSs have the Keplerian (mass-shedding) frequencies larger than 1000Hz. Very recent discovery of X-ray transient XTE J1739-285 by Kaaret et al Kaaret et al. (2006) suggests that it contains a CS rotating at 1122Hz. Following this discovery, the structure of the CS rotating with 1122Hz are studied using several EOSs Bejger et al. (2007). It is found that, for some of the EOSs, this CS is supramassive, i.e., $\delta M_{\rm B}=M_{\rm B,max}^{\rm stat}-M_{\rm B,min},$ (1) is less than $0$. In the above equation, $M_{\rm B,max}^{\rm stat}$ is the maximum baryonic mass of the static CS and $M_{\rm B,min}$ is the minimum mass for the CS rotating with 1122Hz for a given EOS. In the present work we search for the possible correlations in some key properties of static CSs and the ones rotating with the highest observed frequency of 1122Hz. The properties of these CSs are computed for a large set of EOSs, which are constructed using variety of approaches with the chemical compositions ranging from nucleons to hyperons and quarks in $\beta$-equilibrium. Our results suggest that the values of $M_{\rm max}^{\rm stat}$, $R_{1.4}^{\rm stat}$, and $\delta M_{\rm B}$ are strongly correlated. Though, $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$ alone does not show any systematic correlations Bejger et al. (2005). We also unmask the underlying feature of the EOS responsible for the supramassive or non- supramassive nature of the CS rotating with 1122Hz. ## II Equations of State In this work we consider 24 different EOSs with $M_{\rm max}^{\rm stat}\geqslant 1.6M_{\odot}$ which exceeds the recent mass measurements suggesting only 5$\%$ probability that the mass of pulsar PSR J1516+02B is below 1.59 M⊙ Freire (2007). These EOSs are constructed using various approaches which can broadly be grouped into (i) models based on variational approach, (ii) relativistic or non-relativistic mean-field models and (iii) Dirac-Brueckner-Hartree-Fock model. The first group contains EOSs involving neutrons, protons, electrons and muons. The EOSs considered in this group are: BJ-C Bethe and Johnson (1974), FPS Lorenz et al. (1993), BBB2 Baldo et al. (1997), AU, WS, and UU Wiringa et al. (1988), and APR Akmal et al. (1998). For the second group we consider the EOSs: O Bowers et al. (1975), GN3Glendenning (1985), GM1 Glendenning and Moszkowski (1991), TM1 Sumiyoshi et al. (1995), G2Furnstahl et al. (1997), BalbN1H1 Balberg and Gal (1997), GMU110Pons et al. (2000), DH Douchin and Haensel (2001), SSK and GSK1 Agrawal et al. (2006), UY, U0 and L0 Dhiman et al. (2007), GM1-H, UQM52, and CFL52. The EOS GM1-H is composed of nucleons and hyperons in $\beta$ equilibrium. The nucleon-meson interaction parameters are taken from the GM1 parameter set whereas hyperon meson couplings are obtained from SU(6) model. The EOS UQM52 involves noninteracting unpaired quark matter, composed of massive u, d, and s quarks, is based upon the MIT Bag Model of quarks. This EOS has been calculated by using model parameters, bag constant, B = 52 MeV/fm3, masses of three quarks, mu = md = 5.00 MeV/c2, ms = 150 MeV/c2, and QCD coupling constant $\alpha_{c}=0.1$. The Color-Flavor-Locked quark matter equation of state (CFL52) is based upon the free energy as described by Alford (2001) and Rajagopal and Wilczek (2001), by using pairing gap parameter $\Delta$ = 100 MeV. The other model parameters such as bag constant, quark masses, and QCD constant are same as used for UQM52 EOS. In the third group we consider only one EOS: DBHF by Krastev et al. Krastev and Sammarruca (2006). ## III Results and Discussions The properties of spherically symmetric static and axially symmetric rotating CSs are obtained by solving the Einstein’s equations in 1D and 2D, respectively. The numerical computations are performed by using RNS code written by Stergioulas and Friedman Stergioulas and Friedman (1995). In Fig. 1 we present the values of $M_{\rm max}^{\rm stat}$, $R_{1.4}^{\rm stat}$ and $\delta M_{\rm B}$ obtained for several EOSs. We notice that the value of $\delta M_{\rm B}$ for a given EOS is only weakly correlated with those of $M_{\rm max}^{\rm stat}$ as compared to $R_{1.4}^{\rm stat}$. It seems, larger is the value of $R_{1.4}^{\rm stat}$ smaller will be the $\delta M_{\rm B}$. On the other hand, $\delta M_{\rm B}$ large means large $M_{\rm max}^{\rm stat}$ but $R_{1.4}^{\rm stat}$ small. For instance, $\delta M_{\rm B}<0$ for all those EOSs for which $R_{1.4}^{\rm stat}\gtrsim 14$km. Whereas, $\delta M_{\rm B}\approx 0.75M_{\odot}$ for the cases with $M_{\rm max}^{\rm stat}>2M_{\odot}$ and $R_{1.4}^{\rm stat}\sim 10-11$km. Though $\delta M_{\rm B}$ is not correlated well with $M_{\rm max}^{\rm stat}$ or $R_{1.4}^{\rm stat}$ alone, but, it might be well correlated with some appropriate combination of $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$. In Table 1 we summarize the properties of the CS, rotating with 1122Hz, calculated at the maximum circumferential equatorial radius $R_{\rm eq}^{\rm max}$ and the minimum circumferential equatorial radius $R_{\rm eq}^{\rm min}$ for a few selected EOSs. The values of the radius $R_{\rm eq}^{\rm max}$ are determined by the mass shedding instability and that of $R_{\rm eq}^{\rm min}$ are determined by the secular axi-symmetric instability according to turning point theorem Friedman et al. (1988). It can be verified from the table that the variations in the gravitational mass, $\delta M=\left|M(R_{\rm eq}^{\rm max})-M(R_{\rm eq}^{\rm min})\right|,$ (2) of the CS rotating at 1122Hz are correlated with $\delta M_{\rm B}$ up to some extent. The difference $\delta M\lesssim 0.1M_{\odot}$ when $\delta M_{\rm B}$ is negative. For $\delta M_{\rm B}>0$, $\delta M$ increases with $\delta M_{\rm B}$. Therefore, $\delta M_{\rm B}$ not only determines whether the CS rotating at 1122Hz is supramassive or not, but, it also gives an estimate about the value of $\delta M$ for a given EOS. In Fig. 2 we consider the variations of $\delta M_{\rm B}$ with $\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{1/2}\left(\frac{10{\rm km}}{R_{1.4}^{stat}}\right)^{3/2}$. This combination of $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$ is analogous to the one derived within the Newtonian approximation to determine the value of the Keplerian frequency. The values of $\delta M_{\rm B}$ and $\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{1/2}\left(\frac{10{\rm km}}{R_{1.4}^{\rm stat}}\right)^{3/2}$ are well correlated. It is interesting to note that $\delta M_{\rm B}>0$ only if $\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{1/2}\left(\frac{10{\rm km}}{R_{1.4}^{\rm stat}}\right)^{3/2}$ is greater than unity. These correlations simply suggest that the high density behaviour of a EOS with respect to its behaviour at low density plays a predominant role in determining whether the CS rotating at 1122Hz is supramassive or not. Since, the $M_{\rm max}^{\rm{stat}}$ probes densest segment of the EOS whereas, $R_{1.4}^{\rm stat}$ probes relatively lower density region of EOS. We parameterize $\delta M_{\rm B}$ in terms of $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$ as, $\frac{\delta M_{\rm B}}{M_{\odot}}=a_{0}+a_{1}\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{\alpha}\left(\frac{10{\rm km}}{R_{1.4}^{\rm stat}}\right)^{\beta}.$ (3) The best fit values of the parameters appearing in Eq. 3 are calculated using the results displayed in Fig. 1. The values of parameters are $a_{0}=-2.75$, $a_{1}=2.5$, $\alpha=0.75$ and $\beta=1.56$. In Fig. 3, we plot the results for $R_{1.4}^{\rm stat}$ versus $M_{\rm max}^{\rm stat}$ obtained by solving Eq. 3 for fixed values of $\delta M_{\rm B}$. These plots can provide us immediately some idea of $\delta M_{\rm B}$ once the properties of the static CS like $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$ are known. We also superpose the results shown in Fig. 1 by dividing them in to three classes depending on the values of the $\delta M_{\rm B}$. The symbols, triangles, circles and squares represent the values of $R_{1.4}^{\rm stat}$ and $M_{\rm max}^{\rm stat}$ with $\delta M_{\rm B}$ lie in the range of $-0.5-0.0$, $0.0-0.5$ and 0.5 - 1.5 $M_{\odot}$ respectively. It is evident from Fig. 3 that Eqs. 3 can be used to divide the $M_{\rm max}^{\rm stat}$ $-$ $R_{1.4}^{\rm stat}$ plane in to the regions with different $\delta M_{\rm B}$. It is to be noted from Fig. 3 that the current measurement of maximum gravitational mass 1.76M⊙ Verbiest et al. would set the upper limit on R${}_{1.4}^{\rm stat}$ to be 12.4 km which corresponds to $\delta M_{B}=0$. Interestingly, this upper limit on R${}_{1.4}^{\rm stat}$ is closer to the lower limit of 12.9 km obtained by analyzing the high quality X-ray spectra from CS in qLMXB X7 Heinke et al. (2006). We plot in Fig. 4 the curves for moment of inertia ${\cal I}_{1.4}^{\rm stat}$ versus $M_{\rm max}^{\rm stat}$ with fixed values of $\delta M_{\rm B}$. These curves are generated by fitting the values of $\delta M_{\rm B}$ to the following expression, $\frac{\delta M_{\rm B}}{M_{\odot}}=a_{0}^{\prime}+a_{1}^{\prime}\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{\alpha^{\prime}}\left(\frac{{\cal I}_{1.4}^{\rm stat}}{{\cal I}_{0}}\right)^{\beta^{\prime}},$ (4) where, ${\cal I}_{0}=10^{45}$g cm-2 and the values of the best fit parameters are $a_{0}^{\prime}=-3.25$, $a_{1}^{\prime}=3.25$, $\alpha^{\prime}=0.63$ and $\beta^{\prime}=0.85$. Similar to the case of R${}_{1.4}^{\rm stat}$, we obtained the upper limit of ${\cal I}_{1.4}^{\rm stat}=1.53\times 10^{45}$ g cm-2 from maximum mass of CS measured to date Verbiest et al. . ## IV Summary The key properties such as $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$ of static CS and $\delta M_{\rm B}$ (Eq. 1) for the CS rotating with the highest observed frequency of 1122Hz are computed using 24 diverse EOSs. These EOSs are chosen in a manner that they correspond to a wide variety of approaches and their chemical composition vary from the nucleons to hyperons and quarks in $\beta$-equilibrium. The values of $\delta M_{\rm B}$ are found to be almost linearly correlated with $\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{1/2}\left(\frac{10{\rm km}}{R_{1.4}^{\rm stat}}\right)^{3/2}$; a combination of $M_{\rm max}^{\rm stat}$ and $R_{\rm 1.4}^{\rm stat}$ analogous to the one popularly used to determine Keplerian frequency. For a given EOS, the CS rotating at 1122Hz is non-supramassive (i.e., $\delta M_{\rm B}>0$) only if $\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{1/2}\left(\frac{10{\rm km}}{R_{1.4}^{\rm stat}}\right)^{3/2}$ is greater than unity. It is also noticed that the variations in the gravitational mass for the CS rotating with 1122Hz are up to some extent correlated with the values of $\delta M_{\rm B}$ (see Table 1). In view of these results, it appears that the observation of the rapidly rotating CSs constrain relative behaviour of EOS at high density with respect to it’s behaviour at low or moderate densities. Since, the $M_{\rm{max}}^{\rm stat}$ probes densest segment of the EOS, whereas, $R_{1.4}^{\rm stat}$ probes relatively lower density region of EOS. The suitable parametric forms for the $\delta M_{\rm B}$ in terms of $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$ or ${\cal I}_{1.4}^{\rm stat}$ (Eqs. 3 and 4) are also presented. Using these parametric forms, one can divide the $M_{\rm max}^{\rm stat}$ $-$ $R_{1.4}^{\rm stat}$ and $M_{\rm max}^{\rm stat}$ $-$ ${\cal I}_{1.4}^{\rm stat}$ planes into regions of different $\delta M_{\rm B}$. Thus, for a given EOS, only the knowledge of the key properties of static CSs can well estimate a priori the properties of the resulting CS rotating with 1122Hz. Currently measured maximum gravitational mass 1.76$M_{\odot}$ of PSR J0437-4715 suggests that the CS rotating at 1122Hz can be non-supramassive provided $R_{1.4}^{\rm stat}\leqslant 12.4$ km or equivalently ${\cal I}_{\rm 1.4}^{\rm stat}\leqslant 1.53\times 10^{45}$ g cm-2. It will be worth while to repeat the present investigations for the CS rotating at higher frequencies. ###### Acknowledgements. The authors greatly acknowledge Professors F. Sammarruca and J. A. Pons for providing the data for EOSs, DBHF and GMU110 respectivelly. This work was supported in part by the University Grant Commission under grant # F.17-40/98 (SA-I). ## References * (1) J. P. W. Verbiest, M. Bailes, W. van Straten, G. B. Hobbs, R. T. Edwards, R. N. Manchester, N. D. R. Bhat, J. M. Sarkissian, B. A. Jacoby, and S. R. Kulkarni, astro-ph/0801.2589 . * Heinke et al. (2006) C. O. Heinke, G. B. Rybicki, R. Narayan, and J. E. Grindlay, Astrophys. J. 644, 1090 (2006). * Lyne et al. (2004) A. Lyne, M. Burgay, M. Kramer, A. Possenti, R. N. Manchester, F. Camilo, M. A. McLaughlin, D. R. Lorimer, D’Amico, B. C. Joshi, et al., Science 303, 1153 (2004). * Morrison et al. (2004) I. A. Morrison, T. W. Baumgarte, S. L. Shapiro, and V. R. Pandharipande, Astrophys. J. 617, L135 (2004). * Backer et al. (1982) D. Backer, S. Kulkarni, C. Heiles, M. M. Davis, and W. M. Goss, Nature 300, 615 (1982). * Hessels et al. (2006) J. W. T. Hessels, S. M. Ransom, I. H. stairs, P. C. C. Freire, V. M. Kaspi, and F. Camilo, Science 311, 190 (2006). * Shapiro and Teukolsky (1983) S. L. Shapiro and S. A. Teukolsky, _Black Holes, White Dwarfs, and Neutron Stars_ (Wiley, New York, 1983). * Kaaret et al. (2006) P. Kaaret, Z. Prieskorn, J. I. T. Zand, S. Brandt, N. Lund, S. Mereghetti, D. Gotz, E. Kuulkers, and J. Tomsick, Astrophys. J. 657, L97 (2006). * Bejger et al. (2007) M. Bejger, P. Haensel, and J. Zdunik, Astron. Astrophys. 464, L49 (2007). * Bejger et al. (2005) M. Bejger, T. Bulik, and P. Haensel, Mon. Not. R. Astron. Soc. 364, 635 (2005). * Freire (2007) P. C. C. Freire, Astro-ph/ p. 0712.0024 (2007). * Bethe and Johnson (1974) H. A. Bethe and M. Johnson, Nucl. Phys. A230, 1 (1974). * Lorenz et al. (1993) C. P. Lorenz, D. G. Ravenhall, and C. J. Pethick, Phys. Rev. Lett. 70, 379 (1993). * Baldo et al. (1997) M. Baldo, I. Bombaci, and G. F. Burgio, Astron. Astophys. 328, 274 (1997). * Wiringa et al. (1988) R. B. Wiringa, V. Fiks, and A. Fabrocini, Phys. Rev. C 38, 1010 (1988). * Akmal et al. (1998) A. Akmal, V. R. Pandharipande, and D. G. Ravenhall, Phys. Rev. C 58, 1804 (1998). * Bowers et al. (1975) R. L. Bowers, A. M. Gleeson, and R. D. Pedigo, Phys. Rev. D 12, 3043 (1975). * Glendenning (1985) N. K. Glendenning, Astrophys. J 293, 470 (1985). * Glendenning and Moszkowski (1991) N. K. Glendenning and S. A. Moszkowski, Phys. Rev. Lett. 67, 2414 (1991). * Sumiyoshi et al. (1995) K. Sumiyoshi, H. Kuwabara, and H. Toki, Nucl. Phys. A581, 725 (1995). * Furnstahl et al. (1997) R. Furnstahl, B. D. Serot, and H.-B. Tang, Nucl. Phys. A615, 441 (1997). * Balberg and Gal (1997) S. Balberg and A. Gal, Nucl. Phys. A625, 435 (1997). * Pons et al. (2000) J. A. Pons, S. Reddy, P. J. Ellis, M. Prakash, and J. M. Lattimer, Phys. Rev. C 62, 035803 (2000). * Douchin and Haensel (2001) F. Douchin and P. Haensel, Aston. Astrophys. 380, 151 (2001). * Agrawal et al. (2006) B. K. Agrawal, S. K. Dhiman, and R. Kumar, Phys. Rev. C 73, 034319 (2006). * Dhiman et al. (2007) S. K. Dhiman, R. Kumar, and B. K. Agrawal, Phys. Rev. C 76, 045801 (2007). * Alford (2001) M. G. Alford, Annu. Rev. Nucl. Part. Sci. 51, 131 (2001). * Rajagopal and Wilczek (2001) K. Rajagopal and F. Wilczek, Phys. Rev. Lett. 86, 3492 (2001). * Krastev and Sammarruca (2006) P. G. Krastev and F. Sammarruca, Phys. Rev. C 74, 025808 (2006). * Stergioulas and Friedman (1995) N. Stergioulas and J. L. Friedman, Astrophys. J. 444, 306 (1995). * Friedman et al. (1988) H. L. Friedman, J. R. Ipser, and R. D. Sorkin, Astrophys. J. 325, 722 (1988). Table 1: The properties of the compact stars, rotating with frequency $\nu$ = 1122 Hz, calculated at the maximum and the minimum circumferential equatorial radius $R_{\rm eq}^{\rm max}$ and $R_{\rm eq}^{\rm min}$. The quantities $\delta M$ and $\delta M_{B}$ for a given EOS are determined by using Eqs.(1 and 2) respectively. The central mass densities $\epsilon_{c}$ at the $R_{eq}^{max}$ and $R_{\rm eq}^{\rm min}$ are also presented. EOS | $\epsilon_{c}$ | $M(R^{\rm max}_{\rm eq})$ | $R^{\rm max}_{\rm eq}$ | $\epsilon_{c}$ | $M(R^{\rm min}_{\rm eq})$ | $R^{\rm min}_{\rm eq}$ | $\delta M$ | $\delta M_{B}$ ---|---|---|---|---|---|---|---|--- | $(10^{15}gcm^{-3})$ | $(M_{\odot})$ | (km) | $(10^{15}gcm^{-3})$ | $(M_{\odot})$ | (km) | $(M_{\odot})$ | $(M_{\odot})$ GMU110 | 1.30322 | 2.071 | 17.53 | 1.48038 | 2.053 | 16.33 | 0.018 | -0.393 GM1-H | 1.25875 | 2.235 | 18.03 | 1.56496 | 2.158 | 15.23 | 0.074 | -0.328 BalbN1H1 | 1.42497 | 1.825 | 16.88 | 3.56904 | 1.704 | 9.97 | 0.123 | -0.079 G2 | 1.05609 | 2.051 | 17.48 | 1.81037 | 2.116 | 13.02 | 0.065 | -0.064 GM1 | 0.86585 | 2.456 | 18.55 | 1.53015 | 2.576 | 13.77 | 0.120 | 0.007 UQM52 | 1.14801 | 1.585 | 16.14 | 2.33042 | 1.864 | 11.48 | 0.279 | 0.246 BJ-C | 1.05732 | 1.695 | 16.46 | 2.55674 | 1.954 | 10.88 | 0.259 | 0.278 GSK1 | 0.98210 | 1.788 | 16.70 | 2.37515 | 2.089 | 11.15 | 0.301 | 0.335 O | 0.78443 | 2.138 | 17.81 | 1.69183 | 2.554 | 12.72 | 0.416 | 0.461 SSK | 0.94170 | 1.684 | 16.39 | 2.55229 | 2.127 | 10.72 | 0.443 | 0.537 FPS | 0.98885 | 1.395 | 15.51 | 3.04073 | 1.881 | 9.89 | 0.486 | 0.583 BBB2 | 0.9615 | 1.476 | 15.79 | 2.87618 | 2.002 | 10.08 | 0.526 | 0.654 DBHF | 0.7813 | 1.732 | 16.54 | 2.19589 | 2.412 | 11.24 | 0.680 | 0.861 CFL52 | 0.948 | 0.846 | 12.99 | 2.39711 | 1.990 | 10.39 | 1.144 | 1.357 AU | 0.9237 | 1.143 | 14.46 | 2.87473 | 2.215 | 9.76 | 1.072 | 1.402 Figure 1: (Color online) Values of the maximum gravitational mass $M_{\rm max}^{\rm stat}$ for static CSs, radius $R_{1.4}^{\rm stat}$ for static CSs with mass $1.4M_{\odot}$ and the difference $\delta M_{\rm B}$ (Eq. 1) obtained for several EOSs. Figure 2: (Color online) Correlations between $\delta M_{\rm B}$ and $\left(\frac{M_{\rm max}^{\rm stat}}{M_{\odot}}\right)^{1/2}\left(\frac{10{\rm km}}{R_{1.4}^{\rm stat}}\right)^{3/2}$. This combination of $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$ is analogous to the one commonly used to determine the Keplerian frequency. Figure 3: (Color online) Plots for $R_{1.4}^{\rm stat}$ versus $M_{\rm max}^{\rm stat}$ generated using Eq. 3 for $\delta M_{\rm B}=-0.5,0.0$ and 0.5 $M_{\odot}$ as indicated. Different symbols represent the values of $M_{\rm max}^{\rm stat}$ and $R_{1.4}^{\rm stat}$ with $\delta M_{B}$ lying in the range of $-0.5-0.0$ (triangles), $0.0-0.5$ (circles) and $0.5-1.5$ (squares) $M_{\odot}$ as also depicted in Fig. 1 Figure 4: (Color online) Same as fig. 3, but, values of moment of inertia ${\cal I}_{1.4}^{\rm stat}$ are used instead of $R_{1.4}^{\rm stat}$.

arxiv-papers

2024-09-04T02:48:53.972909

0802.3439

# Results from a search for the $\boldsymbol{0\nu\beta\beta}$-decay of $\mathbf{{}^{130}Te}$ (Physical Review C $\mathbf{78}$, 035502 (2008)) C. Arnaboldi1,2, D.R. Artusa3, F.T. Avignone III3aaaCorresponding author, M. Balata4, I.C. Bandac3, M. Barucci5,6, J.W. Beeman7, F. Bellini17,18, C. Brofferio1,2, C. Bucci4, S. Capelli1,2, L. Carbone2, S. Cebrian8, M. Clemenza1,2, O. Cremonesi2, R.J. Creswick3, A. de Waard9, S. Di Domizio10,11, M. J. Dolinski14,15, H.A. Farach3, E. Fiorini1,2, G. Frossati9, A. Giachero4, A. Giuliani12,2, P. Gorla4, E. Guardincerri7, T. D. Gutierrez19, E.E. Haller7,13, R.H. Maruyama20, R.J. McDonald7, S. Nisi4, C. Nones12,2, E.B. Norman14,21, A. Nucciotti1,2, E. Olivieri5,6, M. Pallavicini10,11, E. Palmieri16, E. Pasca5,6, M. Pavan1,2, M. Pedretti2, G. Pessina2, S. Pirro2, E. Previtali2, L. Risegari5,6, C. Rosenfeld3, S. Sangiorgio20, M. Sisti1,2, A.R. Smith7, L. Torres8, G. Ventura5,6, M. Vignati17,18 ###### Abstract A detailed description of the CUORICINO ${}^{130}Te$ neutrinoless double-beta ($0\nu\beta\beta$) decay experiment is given and recent results are reported. CUORICINO is an array of $62$ tellurium oxide ($TeO_{2}$) bolometers with an active mass of $40.7$ kg. It is cooled to $\sim 8-10$ mK by a dilution refrigerator shielded from environmental radioactivity and energetic neutrons. It is running in the Laboratori Nazionali del Gran Sasso (LNGS) in Assergi, Italy. These data represent an exposure of $11.83\textrm{ kg}\cdot\textrm{y}$ or $91$ mole-years of ${}^{130}Te$. No evidence for $0\nu\beta\beta$-decay was observed and a limit of $T^{0\nu}_{1/2}\left({}^{130}Te\right)\geq 3.0\times 10^{24}$ y ($90\%$ C.L.) is set. This corresponds to an upper limit on the effective mass, $\left\langle m_{\nu}\right\rangle$, between $0.19$ and $0.68$ eV when analyzed with the many published nuclear structure calculations. In the context of these nuclear models, the values fall within the range corresponding to the claim of evidence of $0\nu\beta\beta$-decay by H.V. Klapdor-Kleingrothaus, et al. The experiment continues to acquire data. 1 Dipartimento di Fisica dell’Università di Milano-Bicocca, I-20126 Milano, Italy 2 Sesione INFN di Milano-Bicocca, I-20126 Milano, Italy 3 Department of Physics and Astronomy, University of South Carolina, Columbia, SC 29208, USA 4 INFN Laboratori Nazionali del Gran Sasso, I-67010, Assergi (L’Aquila), Italy 5 Dipartimento di Fisica dell’Università di Firenze, I-50019 Firenze, Italy 6 Sezione INFN di Firenze, I-50019 Firenze, Italy 7 Lawrence Berkeley National Laboratory, Berkeley, CA 94720, USA 8 Laboratorio de Fisica Nuclear y Altas Energias, Universidad de Zaragoza, E-50001 Zaragoza, Spain 9 Kamerling Onnes Laboratory, Leiden University, 2300 RAQ Leiden, Netherlands 10 Dipartimento di Fisica dell’Università di Genova, I-16146 Genova, Italy 11 Sezione INFN di Genova, I-16146 Genova, Italy 12 Dipartimento di Fisica e Matematica dell’Università dell’Insubria, I-22100 Como, Italy 13 Department of Materials Science and Engineering, University of California, Berkeley, CA 94720, USA 14 Lawrence Livermore National Laboratory, Livermore, CA 94551, USA 15 Department of Physics, University of California, Berkeley, CA 94720, USA 16 INFN Laboratori Nazionali di Legnaro, I-35020 Legnaro (Padova), Italy 17 Dipartimento di Fisica dell’Università di Roma La Sapienza, I-00185 Roma, Italy 18 Sezione INFN di Roma, I-00185 Roma, Italy 19 Physics Department, California Polytechnic State University, San Luis Obisbo, CA 93407 USA 20 Department of Physics, University of Wisconsin, Madison, WI 53706, USA 21 Department of Nuclear Engineering, University of California, Berkeley, CA 94720, USA ## I INTRODUCTION There are three very important open questions in neutrino physics that can best be addressed by next generation neutrinoless double-beta $\left(0\nu\beta\beta\right)$ decay experiments. First, are neutrinos Majorana particles that differ from antineutrinos only by helicity? Second, what is their mass-scale? Third, is lepton number conservation violated? While searches for $\beta\beta$-decay have been carried out steadily throughout many decades [1, 2, 3], it is now a far more interesting time for the field. Atmospheric neutrino-oscillation data imply that there exist scenarios in which the effective Majorana mass of the electron neutrino could be larger than $0.05$ eV. Recent developments in detector technology make the observation of $0\nu\beta\beta$-decay at this scale now feasible. For recent comprehensive experimental and theoretical reviews see [4, 5, 6]. Optimism that a direct observation of $0\nu\beta\beta$-decay is possible was greatly enhanced by the observation and measurement of the oscillations of atmospheric neutrinos [7], the confirmation by SuperKamiokande [8] of the deficit of ${}^{8}B$ neutrinos observed by the chlorine experiment [9], the observed deficit of $p-p$ neutrinos by SAGE [10] and GALEX [11], and the results of the SNO experiment [12] that clearly showed that the total flux of ${}^{8}B$ neutrinos from the sun predicted by Bahcall and his co-workers [13] is correct. Finally, the data from the KamLAND reactor-neutrino experiment strongly favor the MSW large mixing-angle solution of solar neutrino oscillations [14]. This important list of results published since $1998$ weighs very heavily in favor of supporting two or more next generation $0\nu\beta\beta$-decay experiments (see the reports in references [15, 16]). The most sensitive limits have come from germanium detectors enriched in ${}^{76}Ge$. They were the Heidelberg-Moscow experiment $\left(T_{1/2}^{0\nu}\left({}^{76}Ge\right)\geq 1.9\times 10^{25}y\right)$ [17] and the IGEX experiment $\left(T_{1/2}^{0\nu}\left({}^{76}Ge\right)\geq 1.6\times 10^{25}y\right)$ [18]. These imply that the upper bound on the effective Majorana mass of the electron neutrino, $\langle m_{\nu}\rangle$, defined below, ranges from $\sim 0.3$ to $\sim 1.0$ eV, depending on the choice of nuclear matrix elements used in the analysis. However, a subset of the Heidelberg-Moscow Collaboration has reanalyzed the data and claimed evidence of a peak at the total decay energy, $2039$ keV, implying $0\nu\beta\beta$-decay [19, 20]. While there have been opposing views [21, 22, 23], there is no clear proof that the observed peak is not an indication of $0\nu\beta\beta$-decay. The GERDA experiment, also using ${}^{76}Ge$, is under construction in the Laboratori Nazionali del Gran Sasso (LNGS), and will test this claim [24]. The CUORICINO experiment, also located at LNGS, is the most sensitive $0\nu\beta\beta$-decay experiment with good energy resolution currently operating [25, 26]. It is searching for the $0\nu\beta\beta$-decay of ${}^{130}Te$ and has the capability of confirming the claim; however, a null result cannot be used to refute the claim because of the uncertainty in the nuclear matrix element calculations. The proposed Majorana ${}^{76}Ge$ experiment [27], CUORE ${}^{130}Te$ experiment [28], and EXO ${}^{136}Xe$ experiment [29] are all designed to reach the $\langle m_{\nu}\rangle\approx 0.05$ eV mass sensitivity and below. Descriptions of other proposed experiments with similar goals are given in the recent reviews [4, 5, 6]. There are other constraints on the neutrino-mass scale, irrespective of their Majorana or Dirac character. The Troitsk [30] and Mainz [31] ${}^{3}H$ single $\beta$-decay experiments have placed an upper limit of $2.2$ eV on the mass of the electron neutrino. The KATRIN experiment, a greatly enlarged ${}^{3}H$ $\beta$-decay experiment in preparation, is projected to have a sensitivity of $0.2$ eV [32]. Astrophysical data are also very relevant in a discussion of neutrino mass. In a recent paper by Barger _et al._ , [33] an upper limit on the sum of neutrino mass eigenvalues, $\Sigma\equiv m_{1}+m_{2}+m_{3}\leq 0.75$ eV ($95\%$ C.L.), was derived. The data used were from the Sloan Digital Sky Survey (SDSS) [34], the two degree Field Galaxy Red Shift Survey (2dFGRS) [35], and the Wilkinson Microwave Anisotropy Probe (WMAP) [36], as well as other CMB experiments and data from the Hubble Space Telescope. Hannestad [37] used the WMAP and 2dFGRS data to derive the bound $\Sigma<1.0$ eV ($95\%$ C.L.) and concluded that these data alone could not rule out the evidence claimed in [19, 20]. On the other hand, Allen, Schmidt and Briddle [38] found a preference for a non-zero neutrino mass, i.e., $\Sigma=0.56^{+0.30}_{-0.25}$ eV. This is interestingly close to the favored range of values given in [19, 20]. For recent papers on the subject see [39] and references therein. The constraint $\Sigma\leq 0.75$ eV would imply that the lightest neutrino eigenstate mass $m_{1}<0.25$ eV. On the other hand, if the claim of the positive value of $\Sigma$ would be correct, $\left\langle m_{\nu}\right\rangle\approx 0.17$ eV, and next generation $0\nu\beta\beta$-decay experiments would constitute a stringent test of lepton-number conservation, irrespective of the neutrino mass hierarchy (see the discussion of hierarchy below). In this paper we present a detailed description and present the results from the CUORICINO $0\nu\beta\beta$-decay experiment derived from data taken between April $2003$ and May $2006$. Finally, we note that ${}^{130}Te$ has a series of calculated matrix elements implying values of $\left\langle m_{\nu}\right\rangle$ derived from the CUORICINO half-life limit between $\sim 0.20$ keV, and $\sim 0.68$ keV. A detailed discussion of the implications from the recent developments in the theoretical nuclear structure calculations is given later. ## II NEUTRINO PHYSICS AND NEUTRINOLESS DOUBLE-BETA DECAY Neutrino-oscillation data very strongly imply that there are three neutrino flavor eigenstates, $\left|\nu_{e,\mu,\tau}\right\rangle$, that are super positions of three mass eigenstates, $\left|\nu_{1,2,3}\right\rangle$, of the weak Hamiltonian as expressed in equation (1): $\left|\nu_{l}\right\rangle=\sum^{3}_{j=1}\left|u_{lj}^{L}\right|e^{i\delta_{j}}\left|\nu_{j}\right\rangle,$ (1) where $l=e,\mu,\tau$, and the factor $e^{i\delta_{j}}$ is a CP phase, $\pm 1$ for CP conservation. The decay rate for the $0\nu\beta\beta$-decay mode driven by the exchange of a massive Majorana neutrino is expressed in the following approximation: $\left(T^{0\nu}_{1/2}\right)^{-1}=G^{0\nu}\left(E_{0},Z\right)\left|\frac{\left\langle m_{\nu}\right\rangle}{m_{e}}\right|^{2}\left|M_{f}^{0\nu}-\left(g_{A}/g_{V}\right)^{2}M_{GT}^{0\nu}\right|^{2},$ (2) where $G^{0\nu}$ is a phase space factor including the couplings, $\left|\left\langle m_{\nu}\right\rangle\right|$ is the effective Majorana mass of the electron neutrino discussed below, $M_{f}^{0\nu}$ and $M_{GT}^{0\nu}$ are the Fermi and Gamow-Teller nuclear matrix elements respectively, and $g_{A}$ and $g_{V}$ are the relative axial-vector and vector weak coupling constants respectively. After multiplication by a diagonal matrix of Majorana phases, $\left\langle m_{\nu}\right\rangle$ is expressed in terms of the first row of the $3\times 3$ matrix of equation (1) as follows: $\left|\left\langle m_{\nu}\right\rangle\right|\equiv\left|\left(u_{e1}^{L}\right)^{2}m_{1}+\left(u_{e2}^{L}\right)^{2}m_{2}e^{i\phi_{2}}+\left(u_{e3}^{L}\right)^{2}m_{3}e^{i\left(\phi_{3}+\delta\right)}\right|,$ (3) where $e^{i\phi_{2,3}}$ are the Majorana CP phases ($\pm 1$ for CP conservation in the lepton sector). Only the phase angle $\delta$ appears in oscillation expressions. The two Majorana phases, $e^{i\phi_{2,3}}$, do not, and hence do not affect neutrino oscillation measurements. The oscillation experiments have, however, constrained the mixing angles and thereby the coefficients $u_{lj}^{L}$ in equation (3). Using the best-fit values from the SNO and Super Kamiokande solar neutrino experiments and the CHOOZ [40], Palo Verde [41] and KamLAND [14] reactor neutrino experiments, we arrive at the following expression in the case of the normal hierarchy: $\left|\left\langle m_{\nu}\right\rangle\right|=\left|\left(0.70^{+0.02}_{-0.04}\right)m_{1}+\left(0.30^{+0.04}_{-0.02}\right)m_{2}e^{i\phi_{2}}+\left(\leq 0.05\right)m_{3}e^{i\left(\phi_{3}+\delta\right)}\right|,$ (4) where the errors are approximated from the published confidence levels (C.L.). The bound on $\left|u_{e3}\right|^{2}$ is at the $2\sigma$ C.L. and the errors on the first two coefficients are $1\sigma$. In the convention used here, the expression for the inverted hierarchy, discussed below, is obtained by exchanging $m_{1}\Leftrightarrow m_{3}$ in equation (4). The results of the solar neutrino and atmospheric neutrino experiments yield the mass square differences $\delta_{ij}^{2}=\left|m_{i}^{2}-m_{j}^{2}\right|$ but cannot distinguish between two mass patterns (hierarchies): the ”normal” hierarchy, in which $\delta m_{\textrm{solar}}^{2}=m_{2}^{2}-m_{1}^{2}$ and $m_{1}\cong m_{2}\ll m_{3}$, and the ”inverted” hierarchy where $\delta m_{\textrm{solar}}^{2}=m_{3}^{2}-m_{2}^{2}$ and $m_{3}\cong m_{2}\gg m_{1}$. In both cases we can approximate $\delta m_{AT}^{2}\cong m_{3}^{2}-m_{1}^{2}$ . Considering the values in equation (4), we make the simplifying approximation $\left(u_{e3}\right)^{2}\approx 0$. Using the central values of equation (4), we can write the following approximate expressions: $\left|\left\langle m_{\nu}\right\rangle\right|\cong m_{1}\left|0.7+0.3e^{i\phi_{2}}\sqrt{1+\frac{\delta_{solar}^{2}}{m_{1}^{2}}}\right|,$ (5) for the case of ”normal” hierarchy, and, $\left|\left\langle m_{\nu}\right\rangle\right|\cong\sqrt{m_{1}^{2}+\delta m_{AT}^{2}}\left|0.7+0.3e^{i\phi_{2}}\right|,$ (6) in the ”inverted” hierarchy case. At this time there is no experimental evidence favoring either hierarchy. In Table I, we use Eqs. (5) and (6) to show the predicted central values of $\left\langle m_{\nu}\right\rangle$ as a function of the lightest neutrino mass eigenvalue, $m_{1}$. These values roughly define the desired target sensitivities of next generation $0\nu\beta\beta$-decay experiments. Table I: Central values of the numerical predictions of $\left|\left\langle m_{\nu}\right\rangle\right|$ (meV) for both hierarchies and CP phase relations. ( $m_{1}$ is also given in meV.) Normal Hierarchy | | Inverted Hierarchy ---|---|--- $e^{i\phi_{2}}=-1$ | | $e^{i\phi_{2}}=+1$ | | $e^{i\phi_{2}}=-1$ | | $e^{i\phi_{2}}=+1$ $m_{1}$ | $\left|\left\langle m_{\nu}\right\rangle\right|$ | | $m_{1}$ | $\left|\left\langle m_{\nu}\right\rangle\right|$ | | $m_{1}$ | $\left|\left\langle m_{\nu}\right\rangle\right|$ | | $m_{1}$ | $\left|\left\langle m_{\nu}\right\rangle\right|$ $20.0$ | $7.90$ | | $20.0$ | $20.2$ | | $0.00$ | $20.0$ | | $0.00$ | $50.0$ $40.0$ | $16.0$ | | $40.0$ | $40.0$ | | $20.0$ | $21.6$ | | $20.0$ | $53.9$ $60.0$ | $24.0$ | | $60.0$ | $60.0$ | | $50.0$ | $28.3$ | | $50.0$ | $70.7$ $80.0$ | $32.0$ | | $80.0$ | $80.0$ | | $75.0$ | $36.0$ | | $75.0$ | $90.1$ $100.0$ | $40.0$ | | $100.0$ | $100.0$ | | $100.0$ | $44.7$ | | $100.0$ | $111.0$ $200.0$ | $80.0$ | | $200.0$ | $200.0$ | | $200.0$ | $82.5$ | | $200.0$ | $206.0$ $400.0$ | $160.0$ | | $400.0$ | $400.0$ | | $400.0$ | $161.1$ | | $400.0$ | $403.0$ It is clear that a next generation experiment should have at least the sensitivity for discovery in the case of an inverted hierarchy when $e^{i\phi_{2}}=e^{i\phi_{3}}$ and for $m_{1}=0$. In this case, $\left\langle m_{\nu}\right\rangle\approx\sqrt{\delta_{\textrm{AT}}^{2}}\approx 0.050\textrm{ eV}$. It should also be capable of being expanded in case this level is reached and no effect is found [15, 16]. It is convenient to define the nuclear structure factor, $F_{N}$, (sometimes denoted as $C_{mm}$ in the literature) as follows: $F_{N}\equiv G^{0\nu}\left|M_{f}^{0\nu}-\left(g_{A}/g_{V}\right)^{2}M_{GT}^{0\nu}\right|^{2}.$ (7) Accordingly, the effective Majorana mass of the electron neutrino is connected to the half-life as shown in equation (8): $\left\langle m_{\nu}\right\rangle=\frac{m_{e}}{\sqrt{F_{N}T_{1/2}^{0\nu}}}.$ (8) To extract values of $F_{N}$ from theoretical papers, we recommend using their calculated values of half lives for a given value of $\left\langle m_{\nu}\right\rangle$, thereby avoiding difficulties associated with conventions used in calculating phase-space factors. Possible interpretations of the null result of CUORICINO, in terms of the effective Majorana neutrino mass, may be understood with detailed analyses of the nuclear matrix elements discussed in a Secs. VIII and IX. In Sec. X, this null result will be compared with the positive claim report reported in [19, 20]. ## III THE EXPERIMENT The CUORICINO experiment is an array of cryogenic bolometers containing ${}^{130}Te$, the parent $0\nu\beta\beta$-decay isotope. This technique was suggested for $\beta\beta$-decay searches by Fiorini and Niinikoski [42] and applied earlier by the Milano group in the MIBETA experiment [43]. The bolometers are sensitive calorimeters that measure the energy deposited by particle or photon interactions by measuring the corresponding rise in temperature. The CUORICINO bolometers are single crystals of $TeO_{2}$; they are dielectric and diamagnetic, and are operated at temperatures between $8$ and $10$ mK [44, 45]. According to the Debye Law, the specific heat of $TeO_{2}$ crystals is given by $C(T)=\beta\left(T/\Theta_{D}\right)^{3}$, where $\beta=1994\textrm{ JK}^{-1}\textrm{mol}^{-1}$ and $\Theta_{D}$ is the Debye temperature. In these materials, $C(T)$ is due almost exclusively to lattice degrees of freedom. A special measurement determined the value of $\Theta_{D}$, as $232$ K [43]. This differs from the previously published value of $272$ K [46]. The specific heat followed the Debye Law down to $60$ mK. The heat capacity of these crystals, extrapolated to $10$ mK, is $2.3\times 10^{-9}\textrm{ JK}^{-1}$. With these values of the parameters, an energy deposition of a few keV will result in a measurable temperature increase, $\Delta T$. In CUORICINO, $\Delta T$ is measured by high-resistance germanium thermistors glued to each crystal. More details can be found in reference [44] and in earlier publications [47, 48]. Accordingly, the temperature increase caused by the deposition of energy equal to the total $\beta\beta$-decay energy, $Q_{\beta\beta}=2530.3\pm 2.0$ keV, would be $1.77\times 10^{-4}$ K. To obtain usable signals for such small temperature changes, very sensitive thermistors are required. The thermistors are heavily doped high-resistance germanium semiconductors with an impurity concentration slightly below the metal-insulator transition. High quality thermistors require a very homogeneous doping concentration. CUORICINO uses Neutron Transmutation Doped (NTD) germanium thermistors. This is achieved by means of uniform thermal neutron irradiation throughout the entire semiconductor volume, in a nuclear reactor. The electrical conductivity of these devices, which is due to variable range hopping (VHR) of the electrons, depends very sensitively on the temperature. The resistivity varies with temperature according to $\rho=\rho_{0}\exp{{\left(\frac{T_{0}}{T}\right)^{\gamma}}}$, where $\rho_{0}$ and $T_{0}$ depend on the doping concentration and $\gamma=1/2$. Thermistors can be parameterized by their sensitivity, $A(T)$, defined as follows: $A(T)\equiv\left|d\left(\ln{R}\right)/d\left(\ln{T}\right)\right|$ $=\gamma\left(T_{0}/T\right)^{\gamma}$, and where the resistance is $R(T)=R_{0}\exp{\left(T_{0}/T\right)^{\gamma}}$. The parameter $R_{0}\equiv\rho_{0}(d/a)$, where $d$ and $a$ are the distance between the contacts and the cross section of the thermistor, respectively. The values of $R_{0},T_{0}$ and $\gamma$ were experimentally measured for about one third of the thermistors, and the average values used for the rest. The measurements were done by coupling the thermistor to a low-temperature heat sink with a high-heat-conductivity varnish glue, which can be easily removed with alcohol. The base temperature of the heat sink is between $15$ and $50$ mK [50]. A current flows through the device and an I-V load curve is plotted. The curve becomes very non-linear due to the power dissipation, which causes the dynamic resistance, the slope of the I-V curve, to invert from positive to negative. The characterization, as discussed in Ref. [51] is done on the thermistors directly mounted on a heat sink, while the optimum bias is studied for the complete detector, thermistor and crystal, since the noise figure depends on all thermal conductances, glue, wires, Teflon, etc. This allows the maximization of the signal to noise ratio. The parameters of each thermistor are determined from a combined fit to a set of load curves measured at different base temperatures. A detailed description of the characterization process for $Si$ thermistors was described in Ref. [51] and same process was used for the CUORICINO $Ge$ thermistors. The thermistors used in the MIBETA and CUORICINO experiments were specially developed and produced for this application [52]. It is necessary to optimize the neutron doping of the $Ge$. This is facilitated by foils of metal with long-lived $(n,\gamma)$ radioactive daughter nuclides, allowing the neutron exposure to be evaluated without having to wait for the intense radiation of the ${}^{71}Ge$ in the $Ge$ sample to decay. Following the decay period, the $Ge$ is heat treated to repair the crystal structure and then cut into $3\times 3\times 1$ mm strips. Electrical connections are made with two $50\,\mu\textrm{m}$ gold wires, ball bonded to metalized surfaces on the thermistor. The thermistors are glued to each bolometer by nine spots of epoxy, deposited by an array of pins for better control of the thermal conductances and to minimize stresses at the interface between the two materials. ## IV THE CUORICINO DETECTOR CUORICINO is a pilot experiment for a larger experiment, CUORE (Cryogenic Underground Observatory for Rare Events) discussed later. It is a tower of $13$ planes [25, 26]. As shown in Fig. 1, the CUORICINO structure is as follows: each of the upper $10$ planes and the lowest one consists of four $5\times 5\times 5\textrm{ cm}^{3}$ $TeO_{2}$ crystals (of natural isotopic abundance of ${}^{130}Te$) as shown in the upper right hand figure, while the $11^{th}$ and $12^{th}$ planes have nine, $3\times 3\times 6\textrm{ cm}^{3}$ crystals, as shown in the lower right hand figure. In the $3\times 3\times 6\textrm{ cm}^{3}$ planes the central crystal is fully surrounded by the nearest neighbors for greater veto capability. Figure 1: (Color online) The Tower of CUORICINO and individual $4$ and $9$ detector modules. The smaller crystals are of natural isotopic abundance except for four. Two of them are enriched to $82.3\%$ in ${}^{128}Te$ and two are enriched to $75\%$ in ${}^{130}Te$. All crystals were grown with pre-tested low radioactivity material by the Shanghai Institute of Ceramics and shipped to Italy by sea to minimize the activation by cosmic ray interactions. They were lapped with specially selected low contamination polishing compound. All these operations, as well as the mounting of the tower, were carried out in a nitrogen atmosphere glove box in a clean room. The mechanical structure is made of oxygen-free high-conductivity copper and Teflon, and both were previously tested to be sure that radioactive contaminations were minimal and consistent with the required detector sensitivity. Thermal pulses are measured with NTD $Ge$ thermistors thermally coupled to each crystal. The thermistors are biased through two high-impedance load resistors at room temperature, with resistances typically in excess of one hundred times that of the thermistors. The large ratio of the resistances of the load resistors over those of the thermistors allows the parallel noise to be kept at an adequate level. Low frequency load-resistor noise was minimized by a specially designed circuit [53]. The voltage signals from the thermistors are amplified and filtered before being fed to an analog-to-digital converter (ADC). This part of the electronic system is DC coupled, and only low-pass anti-aliasing filters are used to reduce the high-frequency noise. The typical bandwidth is approximately $10$ Hz, with signal rise and decay times of order $30$ and $500$ ms, respectively. This entire chain of electronics makes a negligible contribution to the detector energy resolution. More details of the design and features of the electronic system are found in [54]. The gain of each bolometer is stabilized by means of a $Si$ resistor of $50$-$100$ $\textrm{k}\Omega$, attached to each bolometer that acts as a heater. Heat pulses are periodically supplied by a calibrated ultra-stable pulser [55]. This sends a calibrated voltage pulse to the $Si$ resistor. This pulse has a time duration very much shorter than the typical thermal response of the detector [44]. The Joule dissipation from the $Si$ resistor produces heat pulses in the crystal almost indistinguishable in characteristic shape from those from calibration $\gamma$-rays. The heater pulses are produced with a frequency of about one in every $300$ s in each of the CUORICINO bolometers. Any variation in the voltage amplitude recorded from the heater pulses indicates that the gain of that bolometer has changed. The heater pulses are used to measure (and later correct offline) for the gain drifts. Two other pulses, one at lower and one at higher energies, are sent to the same resistors with much lower frequency. The former is used to monitor threshold stability, and the latter to check the effectiveness of the gain stability correction. The tower is mechanically decoupled from the cryostat to avoid heating due to vibrations. The tower is connected through a $25$ mm copper bar to a steel spring fixed to the $50$ mK plate of the refrigerator. The temperature stabilization of the tower is made by means of a thermistor and a heater glued to it. An electronic channel is used for a feed back system [56]. The entire setup is shielded with two layers of lead of $10$ cm minimum thickness each. The outer layer is made of common low radioactivity lead, while the inner layer is made of special lead with a measured activity of $16\pm 4$ Bq/kg from ${}^{210}Pb$. The electrolytic copper of the refrigerator thermal shields provides an additional shield with a minimum thickness of $2$ cm. An external $10$ cm layer of borated polyethylene was installed to reduce the background due to environmental neutrons. The detector is shielded against the intrinsic radioactive contamination of the dilution unit materials by an internal layer of $10$ cm of Roman lead (${}^{210}Pb$ activity $<4$ mBq/kg [50]), located inside of the cryostat immediately above the tower of the array. The background from the activity in the lateral thermal shields of the dilution refrigerator is reduced by a lateral internal shield of Roman lead that is $1.2$ cm thick. The refrigerator is surrounded by a Plexiglas anti-radon box flushed with clean $N_{2}$ from a liquid nitrogen evaporator and is also enclosed in a Faraday cage to eliminate electromagnetic interference. A sketch of the assembly is shown in Fig. 2. Figure 2: (Color online) A sketch of the CUORICINO assembly showing the tower hanging from the mixing chamber and the various heat shields and the external shielding. When cooled to $8$ mK there is a temperature spread of $\sim 1$ mK among the different detectors. Routine calibrations are performed using two wires of thoriated tungsten inserted inside the external lead shield in immediate contact with the outer vacuum chamber (OVC) of the dilution refrigerator. Calibrations normally last one to two days, and are performed at the beginning and end of each run, which lasts for approximately two-three weeks. The CUORICINO array was first cooled down at the beginning of $2003$. However, during this operation electrical connections were lost to $12$ of the $44$ detectors of $5\times 5\times 5\textrm{ cm}^{3}$, and to one of the 3$\times 3\times 6\textrm{ cm}^{3}$ crystals. Thermal stresses broke the electrical connections on their thermalizer stages that allow the transition in temperature of the electric signals in several steps from the detectors at $\sim 8$ mK to room temperature. When the cause of the disconnection was found, new thermalizer stages were fabricated and tested at low temperature. However, since the performance of the remaining detectors was normal, and their total mass was $\sim 30$ kg, warming of the array and rewiring were postponed for several months while $0\nu\beta\beta$-decay data were collected. At the end of $2003$, CUORICINO data acquisition was stopped and the system was warmed to room temperature and the broken thermalizer stages were replaced with new ones. During this operation, the tower was kept enclosed in its copper box to prevent possible recontamination of the detectors. As a consequence, two detectors whose disconnections were inside the box were not recovered. The same was true for one of the small central detectors whose $Si$ resistor was electrically disconnected inside the box. In the middle of $2004$, CUORICINO was cooled down and data collection began again. Typical calibration spectra are shown in Fig. 3. Figure 3: Typical calibration spectra of the CUORICINO array with a ${}^{232}Th$ source: $5\times 5\times 5\textrm{ cm}^{3}$ crystals upper frame, $3\times 3\times 6\textrm{ cm}^{3}$ crystals lower frame. ## V DATA ACQUISITION AND ANALYSIS The signals coming from each bolometer are amplified and filtered with a six- pole Bessel low-pass filter and fed to a $16$-bit ADC. The signal is digitized with a sampling time of $8$ ms, and a circular buffer is filled. With each trigger pulse, a set of $512$ samples is recorded to disk; accordingly, the entire pulse shape is stored for offline analysis. Each channel (bolometer) has a ompletely independent trigger and trigger threshold, optimized according to the bolometer s typical noise and pulse shape. Starting with run No. $2$, the CUORICINO data acquisition (DAQ) now has a software trigger that implements a ”debounce” algorithm to reduce spurious fast signal triggering. The trigger is ready again within a few tens of ms, a delay due to the debounce time. Therefore, most of the pile-up events are re-triggered. The trigger efficiency above $100$ keV was evaluated as $99\pm 1\%$ by checking the fraction of recorded pulser signals. The offline analysis uses an Optimal Filter technique [44] to evaluate the pulse amplitudes and to compare pulse- shapes with detector response function. Events not caused by interactions in the crystals are recognized and rejected on the basis of this comparison. Pile-up pulses are identified and dealt with. This is important for calibration and high rate measurements because the pulses have long time durations and pile-up pulses can significantly increase the dead time. However, the pile-up fraction during the search for $0\nu\beta\beta$-decay is negligible given the low trigger rate from signals above threshold. The pile- up probability on the rise time is $\sim 0.01\%$, while that on the entire sampling window is quite a bit higher, $\sim 0.4\%$. However, these events are easily identified and the pile-up pulses are rejected. The total trigger rate, before any pulse-shape rejection, is time and channel dependent. On a single channel it ranges from a few mHz to hundreds of mHz, with a mean value of about $20$ mHz. Accepted-pulse amplitudes are then corrected using the variation in the gain measured with the heat pulses from the $Si$ resistors. Finally, spectra are produced for each detector. Any type of coincidence cut can be applied to the data written to disk, before the creation of the final spectra, depending on the specific analysis desired. In the case of $\beta\beta$-decay analyses, anticoincidence spectra are used. This allows the rejection of background counts from gamma rays that Compton scatter in more than one bolometer, for example. The probability of accidental coincidences over the entire detector is negligible ($<0.6\%$). Crosstalk pulses have been observed between a few channels; however, the resulting pulses are rejected on the basis of pulse-shape. ## VI SOURCE CALIBRATION AND DETECTOR PERFORMANCE The performance of each detector is periodically checked during the routine calibration with the ${}^{232}Th$ gamma rays from thoriated calibration wires. The most intense gamma ray peaks visible in the calibration spectra are used. They are the: $511,583,911,968,1588$ and $2615$ keV $\gamma$-rays, and the single escape peak of the $2615$ keV gamma ray at $2104$ keV. The resulting amplitude-energy relationship is obtained from the calibration data, and the pulse amplitudes are converted into energies. The dependence of the amplitude on energy is fit with a second order log-polynomial for which the parameters were obtained from the calibration data. The selection of the functional form was established by means of simulation studies based on a thermal model of the detectors. The details of how the thermal model was applied have been published elsewhere [44]. These calibration data are also used to determine the energy resolution of each bolometer. Data sets are collected for two to six weeks, separated by radioactive-source calibrations. The data collected by a single detector in this short time does not have the statistical significance to show the background gamma-ray lines because of the very low counting rates. The energy resolution, and the stability of the energy calibration, relies on the heater pulses, and on the initial and final source calibration measurements. Double-beta decay data collected with each detector during a single data collection period are rejected if any of the following criteria are not fulfilled: 1. (i) The position of the $2615$ keV background $\gamma$-ray line from the decay of ${}^{208}Tl$, in the initial and the final source-calibration measurements must be stable to within $1/3$ of the measured full width at half maximum (FWHM) of the $2615$ keV line for that detector. 2. (ii) The energy resolution of the $2615$ keV $\gamma$-ray lines in the initial and final energy calibration measurements must be stable within $30\%$. 3. (iii) The energy position of the heater pulses during the entire data collection period for that data set must be stable to within $1/3$ of the characteristic (FWHM) for that detector. 4. (iv) The energy resolution measured with the heater pulses for that entire data collection period must be stable within $30\%$ over the entire data collection period. Whenever one or more of these criteria is not fulfilled, the data from that detector are not included in the final data set. Approximately $17\%$ of the data were discarded because they failed one or more of these criteria. Frequent causes of failure to satisfy all of the criteria were noise pulses that degrade the energy resolution and temperature drifts that change the operating parameters of the bolometers. The particular bolometers involved cary; however, some are more sensitive to noise and temperature changes than others. The application of coincidence cuts does not change the efficiency; however, the difference in rise time between pulses from various bolometers can cause coincidences not to be recognized as such, cut this effect is small. in any case, the only result of the failure to recognize coincidences is the loss of background reduction, which would tend to make the quoted bound conservative. In both runs, the measured detector performances appear to be excellent; the average FWHM resolutions in the energy region around $2530$ keV during the calibration measurements are $7$ and $9$ keV, for the $5\times 5\times 5\textrm{ cm}^{3}$ and $3\times 3\times 6\textrm{ cm}^{3}$ detectors, respectively. The spread in the FWHM is about $2$ keV in both cases. The smaller detectors have somewhat worse resolution on average, while they also exhibit a very important nonlinearity. When the calibration spectra from all of the larger and smaller detectors are summed together, the summed spectrum resembled that of a single large detector as shown in Figure 3. Figure 4: The sum spectrum of the background from the $5\times 5\times 5\textrm{ cm}^{3}$ detectors, from both runs, to search for $0\nu\beta\beta$-decay. ## VII DOUBLE-BETA DECAY RESULTS Following the shutdown discussed earlier, and restart in May $2004$, a second interruption was required to remove the malfunctioning helium liquefier used to automatically refill the main bath of the dilution refrigerator. There were also short interruptions for routine maintenance of the $17$-year old refrigerator. Excluding these interruptions, the duty cycle was very satisfactory, $\sim 60\%$ , not withstanding the fact that $15$ to $20\%$ of the live time is necessary for calibration. The three spectra corresponding to large ($5\times 5\times 5\textrm{ cm}^{3}$) detectors and the smaller natural and enriched ($3\times 3\times 6\textrm{ cm}^{3}$) detectors are kept separate because of the different detection efficiencies for $\beta\beta$-decay events, and also because of their different background counting rates. For similar reasons, the spectra of the two runs are treated separately. Because the background rates in the spectra of Runs I and II do not show any statistically significant difference, it was concluded that no recontamination of the detector took place when the cryostat was opened to air during the interruption between Runs I and II. The full data set used in this analysis has a total effective exposure of $11.83\textrm{ kg}\cdot\textrm{yr}$ of ${}^{130}Te$ for the entire array. The full summed spectrum, shown in Fig. 4, clearly exhibits the $\gamma$-ray line from the decay of ${}^{40}K$, and those from the ${}^{238}U$ and ${}^{232}Th$ chains. Also visible are the lines of ${}^{57}Co$, ${}^{60}Co$, and ${}^{54}Mn$, due to the cosmogenic activation of the tellurium and the copper frame. The correct positions and widths of the peaks in the sum spectrum demonstrate the effectiveness of the calibration and linearity of the spectra. The accuracy of calibration in the $0\nu\beta\beta$-decay region was evaluated to be about $\pm 0.4$ keV. The details of the gamma-ray background resulting from a preliminary analysis of Run 2 are given in Tables II, III and IV. There is also clear evidence of alpha backgrounds at energies above the $2614.5$ keV gamma ray in the decay of ${}^{232}Tl$. A detailed analysis attributes the dominant background in the region of interest to degraded alpha particles on the surface of the copper frames. A major effort is underway to reduce this to a minimum. Table II: Gamma rays from the decay of ${}^{232}Th$ observed in Run II. Energy (keV) | | Isotope | | Counts/1000 h ---|---|---|---|--- $238.6$bbbContains a contribution from the $U$ chain. | | ${}^{212}Pb$ | | $6.84\pm 0.43$ $338.2$ | | ${}^{228}Ac$ | | $0.89\pm 0.40$ $463.0$cccContains a contribution from ${}^{125}Sb$. | | ${}^{228}Ac$ | | $1.33\pm 0.25$ $510.7$dddContains a contribution from annihilation radiation. | | ${}^{208}Tl$ | | $7.78\pm 0.38$ $583.2$ | | ${}^{208}Tl$ | | $3.88\pm 0.30$ $727.3$ | | ${}^{212}Bi$ | | $1.04\pm 0.21$ $785.4$eeeContains a contribution from ${}^{214}Bi$ in the $U$ chain. | | ${}^{212}Bi$ | | $1.02\pm 0.20$ $794.9$ | | ${}^{228}Ac$ | | $0.70\pm 0.25$ $833.0$fffContains a contribution from ${}^{54}Mn$. | | ${}^{228}Ac$ | | $2.85\pm 0.25$ $911.2$ | | ${}^{228}Ac$ | | $4.69\pm 0.26$ $964.8$ | | ${}^{228}Ac$ | | $1.37\pm 0.19$ $968.9$ | | ${}^{228}Ac$ | | $2.79\pm 0.21$ $1588.1$ | | ${}^{228}Ac$ | | $0.65\pm 0.12$ $1593.0$gggContains a contribution from ${}^{214}Bi$ in the $U$ chain. | | ${}^{208}Tl$ | | $0.25\pm 0.10$ $1620.6$ | | ${}^{212}Bi$ | | $0.58\pm 0.15$ $1631.0$ | | ${}^{228}Ac$ | | $0.39\pm 0.13$ $2614.5$ | | ${}^{208}Tl$ | | $6.90\pm 0.26$ Table III: Gamma rays from the ${}^{238}U$ chain in data of Run II. Most of the activity is attributed o a radon contamination due to temporary leak in the anti-radon box surrounding the refrigerator. Energy (keV) | | Isotope | | Rate Cts/1000 h | | Energy (keV) | | Isotope | | Rate Cts/1000 h ---|---|---|---|---|---|---|---|---|---|--- $241.9$hhhContains a contribution from ${}^{214}Pb$ in the $Th$ chain. | | ${}^{214}Pb$ | | $6.84\pm 0.43$ | | $1401.7$ | | ${}^{214}Bi$ | | $1.23\pm 0.13$ $295.2$ | | ${}^{214}Pb$ | | $2.69\pm 0.48$ | | $1408.0$ | | ${}^{214}Bi$ | | $1.85\pm 0.15$ $352.0$ | | ${}^{214}Pb$ | | $3.88\pm 0.42$ | | $1509.5$ | | ${}^{214}Bi$ | | $1.85\pm 0.13$ $609.4$ | | ${}^{214}Bi$ | | $13.09\pm 0.47$ | | $1583.2$ | | ${}^{214}Bi$ | | $0.99\pm 0.15$ $665.6$ | | ${}^{214}Bi$ | | $2.54\pm 0.33$ | | $1594.7$iiiContains a contribution from ${}^{208}Tl$ in the $Th$ chain. | | ${}^{214}Bi$ | | $0.25\pm 0.10$ $768.4$ | | ${}^{214}Bi$ | | $2.55\pm 0.33$ | | $1599.3$ | | ${}^{214}Bi$ | | $0.43\pm 0.90$ $786.0$jjjContains a contributions from ${}^{214}Bi$ in the $Th$ chain. | | ${}^{214}Bi$ | | $1.02\pm 0.20$ | | $1661.5$ | | ${}^{214}Bi$ | | $1.06\pm 0.13$ $803.0$ | | ${}^{210}Po$ | | $1.52\pm 0.19$ | | $1729.9$ | | ${}^{214}Bi$ | | $2.51\pm 0.14$ $934.1$ | | ${}^{214}Bi$ | | $1.75\pm 0.17$ | | $1764.7$ | | ${}^{214}Bi$ | | $14.28\pm 0.38$ $1120.4$ | | ${}^{214}Bi$ | | $10.84\pm 0.40$ | | $1838.4$ | | ${}^{214}Bi$ | | $0.40\pm 0.07$ $1155.3$ | | ${}^{214}Bi$ | | $1.38\pm 0.14$ | | $1847.7$ | | ${}^{214}Bi$ | | $1.98\pm 0.17$ $1238.2$ | | ${}^{214}Bi$ | | $4.83\pm 0.21$ | | $2118.9$ | | ${}^{214}Bi$ | | $1.21\pm 0.12$ $1281.1$ | | ${}^{214}Bi$ | | $1.32\pm 0.13$ | | $2204.5$ | | ${}^{214}Bi$ | | $4.55\pm 0.24$ $1377.8$ | | ${}^{214}Bi$ | | $3.37\pm 0.17$ | | $2448.0$ | | ${}^{214}Bi$ | | $1.51\pm 0.14$ $1385.3$ | | ${}^{214}Bi$ | | $0.88\pm 0.88$ | | | | | | Table IV: Background gamma rays from a variety of sources including isotopes produced by cosmogenic neutrons: ${}^{60}Co$, ${}^{54}Mn$, and fall out isotopes ${}^{137}Cs$, ${}^{207}Bi$. Energy (keV) | | Isotope | | Rate Cts/1000 h | | Energy (keV) | | Isotope | | Rate Cts/1000 h ---|---|---|---|---|---|---|---|---|---|--- $122.1$ | | ${}^{57}Co$ | | $5.39\pm 0.44$ | | $661.7$ | | ${}^{137}Cs$ | | $1.26\pm 0.19$ $427.9$ | | ${}^{125}Sb$ | | $1.95\pm 0.27$ | | $834.8$kkkContains a contribution from ${}^{228}Ac$ in the $Th$ chain. | | ${}^{54}Mn$ | | $2.86\pm 0.25$ $463.2$lllContains a contribution from ${}^{228}Ac$ in the $Th$ chain. | | ${}^{125}Sb$ | | $1.33\pm 0.25$ | | $1063.7$ | | ${}^{207}Bi$ | | $2.36\pm 0.29$ $511.0$mmmContains a contribution from ${}^{208}Tl$ in the $Th$ chain. | | annihilation | | $7.78\pm 0.38$ | | $1173.2$ | | ${}^{60}Co$ | | $11.6\pm 0.33$ $569.7$ | | ${}^{207}B$ | | $3.11\pm 0.27$ | | $1332.5$ | | ${}^{60}Co$ | | $11.9\pm 0.36$ $600.6$ | | ${}^{125}Sb$ | | $1.42\pm 0.20$ | | $1461.0$ | | ${}^{40}K$ | | $31.4\pm 0.58$ $635.9$ | | ${}^{125}Sb$ | | $0.64\pm 0.18$ | | $2505.7$ | | ${}^{60}Co$ | | $0.31\pm 0.05$ The average background counting rates in the region of $0\nu\beta\beta$ decay are: $0.18\pm 0.01$, and $0.20\pm 0.04$ counts per keV, per kg, per year ($\textrm{keV}^{-1}\textrm{kg}^{-1}\textrm{y}^{-1}$) for the $5\times 5\times 5\textrm{ cm}^{3}$ and $3\times 3\times 6\textrm{ cm}^{3}$ crystals, respectively. The sum background spectrum from about $2300$ to $2700$ keV, of the $5\times 5\times 5\textrm{ cm}^{3}$ and $3\times 3\times 6\textrm{ cm}^{3}$ crystals, is shown in Fig. 5. The shape of the background in the region of interest does not change when anticoincidence requirement is applied. An extensive analysis of the background contributions implies that the continuum background in the region of interest around $2530$ keV breaks down as follows: $10\pm 5\%$ is due to surface contamination of the $TeO_{2}$ crystals with ${}^{238}U$ and ${}^{232}Th$; $50\pm 20\%$ is due to surface contamination of the copper surfaces facing the bolometers also with ${}^{232}Th$ and ${}^{238}U$; and $30\pm 10\%$ is due to the tail of the $2614.5$ keV gamma ray in the decay of ${}^{232}Th$ from the contamination of the cryostat copper shields. Finally, there were no observable gamma-ray lines associated with neutron interactions. Monte-Carlo simulations of the neutron shield imply that the background from neutron interactions would be negligible. The energy resolution for the complete data set was computed from the FWHM of the $2615$ keV background $\gamma$-ray line in the decay of ${}^{203}Tl$ at the end of the thorium chain. The results are $8$ keV for the forty operating $5\times 5\times 5\textrm{ cm}^{3}$ crystals, and $12$ keV for the eighteen $3\times 3\times 6\textrm{ cm}^{3}$ crystals. Clearly visible is the peak at about $2505$ keV due the summing of the $1332.50$-$1173.24$ keV $\gamma$-ray cascade in the decay of ${}^{60}Co$. This is $25.46$ keV, i.e., about $7$ sigma of the Gaussian energy resolution peak from the $0\nu\beta\beta$-decay end-point energy of ${}^{130}Te$, and could make a negligible contribution to the region under the expected $0\nu\beta\beta$-decay peak . The sum spectrum from $2290$ to $2700$ keV is shown in Fig. 5. The sum spectrum from $2470$ to $2590$ keV is shown in Fig. 6. Figure 5: The summed background spectrum in the $\sim 400$ keV region of interest, which includes the $0\nu\beta\beta$-decay energy $2530.3\pm 2.0$ keV. The details of the operating conditions and parameters of the two CUORICINO data collection periods are given in Table V. The total usable exposure for Run I + Run II is $11.83\textrm{ kg}\cdot\textrm{yr}$ of ${}^{130}Te$. The event detection efficiencies were computed with Monte-Carlo simulations; they are $0.863$ and $0.845$ for the large and small crystals, respectively. The loss of efficiency of the bolometers is due to beta particles created near the surface that escape part of their energy. From the above exposure data we compute: $\ln{2}\times N_{L}\times\epsilon_{L}\times t=2.809\times 10^{25}$ yr, for the large and $\ln{2}\times N_{S}\times\epsilon_{S}\times t=4.584\times 10^{24}$ yr for the small crystals. Here, $\epsilon$ is the detection efficiency, while $N_{L}$ and $N_{S}$ are the numbers of ${}^{130}Te$ nuclei in the large and small detectors, respectively. Table V: Summary of operating parameters for the two CUORICINO data collection periods. From columns $1$ through $8$ are listed: the run number, number of large and small detectors, the active mass of ${}^{130}Te$, total run time, the calibration time, the time collecting $\beta\beta$ \- decay data, the total exposure in $\textrm{kg}\cdot\textrm{yr}$, and the usable exposure in $\textrm{kg}\cdot\textrm{yr}$ after rejection of data not fulfilling the quality requirements. The total usable exposure is then $11.83\textrm{ kg}\cdot\textrm{yr}$. Run $\\#$ | Detectors | Active mass | Run time | Calibration | t-$\beta\beta$ | Collected | Used ---|---|---|---|---|---|---|--- | large/small | [kg 130Te] | [d] | [d] | [d] | [kg$\cdot$yr 130Te] | [kg$\cdot$yr 130Te] 1 | 29/15 | 7.95 | 240 | 24.5 | 55.08 | 1.2 | 1.06 2 | 40/15 | 10.37 | 983 | 108.5 | 415.1 | 11.79 | 10.77 The $\beta\beta$-decay half-life limit was evaluated using a Bayesian approach. The peaks and continuum in the region of the spectrum centered on the $\beta\beta$-decay energy were fit using a maximum likelihood analysis [57, 58]. The likelihood functions of six spectra (the sum spectra of the three types of crystals in the two runs) were combined allowing for a different background level for each spectrum, and a different intensity of the $2505$ keV ${}^{60}Co$ sum peak. Other free parameters are the position of the ${}^{60}Co$ peak and the number of counts under a peak at the $\beta\beta$-decay energy. The same procedure is used to evaluate the $90\%$ CL limit to the number of counts present in the $0\nu\beta\beta$-decay peak. Assuming Poisson statistics for the binned data, the fit procedure was formulated in terms of the likelihood chi-square analysis as described in the following equation: $\chi_{L}^{2}=2\sum_{j=1}^{6}\sum\left(y_{i,j}-n_{i,j}+n_{i,j}\ln{\left(n_{i,j}/y_{i,j}\right)}\right),$ where $j$ indicates the $j^{th}$ spectrum, $n_{i,j}$ is the number of events in the $i^{th}$ bin of the $j^{th}$ spectrum and $y_{i,j}$ is the number of events predicted by the fit model. Fit parameters were estimated minimizing the $\chi_{L}^{2}$, while limits were obtained, after proper renormalization, considering the $\chi_{L}^{2}$ distribution in the physical region. The response function for each spectrum is assumed to be a sum of symmetric gaussian functions, each having the typical energy resolution of one of the detectors summed in that spectrum. The experimental uncertainty in the transition energy is considered by means of a quadratic (gaussian) term in the above equation. In the region between $2575$ and $2665$ keV, assuming a flat background, the best fit yields a negative number of counts under the peak ($-13.9\pm 8.7$). However, the resulting upper bound on the number of candidate events in the $0\nu\beta\beta$-decay peak is $n=10.7$ at $90\%$ C.L. These values are normalized to a hypothetical sum spectrum of the entire statistical data set in which each of the six spectra are weighted according to the corresponding exposure, geometric efficiency, and isotopic abundance. The resulting lower limit on the half-life is computed as: $\displaystyle T_{1/2}^{0\nu}\left({}^{130}Te\right)$ $\displaystyle\geq\ln 2{\left\\{N_{L}\epsilon_{L}+N_{S}\epsilon_{S}\right\\}t/n\left(90\%\textrm{ CL}\right)}$ $\displaystyle=\left(3.27\times 10^{25}/10.7\right)\textrm{ yr}=3.0\times 10^{24}\textrm{ yr}.$ The dependence of the value of the limit on systematic uncertainties that arise from the method of analyzing the data was investigated in detail. These uncertainties reside in the dead time, energy calibration, $Q$-value, and background spectral shape. The main factor influencing the limit is the uncertainty in the background spectral shape. For example, changing the degree of the polynomial used to fit the background in the $0\nu\beta\beta$-decay region from $0$ to $2$ as well as the selection of the energy window used in the analysis can vary the bound from $2.5$ to $3.3\times 10^{24}$ yr. The quoted $90\%$ CL lower bound was computed using the central value, $2530.3$ keV of the measured double beta decay energy [49]. There is a small dip in the data centered at $\sim 2530$ keV as shown in Figure 5. This has been treated as a statistical fluctuation. Figure 6: The total background spectrum from $2470$ to $2590$ keV. Clearly visible is the sum peak at $2505.74$ keV due to the sum of the $1173.24$ and $1332.50$ keV $\gamma$-ray cascade in the decay of ${}^{60}Co$. This activity is attributed to the ${}^{60}Co$ in the copper frames generated by cosmic ray neutrons while the frames were above ground. The solid lines are the best fit to the region using polynomials of the order $0$ to $2$. The three lines in the region of interest are for bounds ($68\%$ and $90\%$) CL on the number of candidate $\beta\beta$-decay events. ## VIII NUCLEAR STRUCTURE ISSUES There is one theoretical viewpoint that holds that the required model space for ${}^{130}Te$ is still very large for reliable shell model calculations and must be severely truncated. Accordingly, the quasiparticle random-phase approximation (QRPA) is commonly used [59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78]. The results from these calculations, from author to author had, until recently, differed significantly for the same nucleus. In Table VI, only the results from references [62, 73] differ significantly from the other $13$; they correspond to the largest matrix elements. In the QRPA approach, the particle-particle interaction is fixed by a parameter, $g_{pp}$, which is derived in various ways by different authors. Two recent papers by Rodin, _et al._ , give detailed assessments of the uncertainties in QRPA calculations of $0\nu\beta\beta$-decay matrix elements, and explain many of the reasons for the disagreements between the various authors over the years [60, 61]. The numerical values given in these articles were corrected in a later erratum [78]. In Table VI we list the values of $\left\langle m_{\nu}\right\rangle$ corresponding to $T_{1/2}^{0\nu}\left({}^{130}Te\right)\geq 3.0\times 10^{24}$ yr derived using the calculations of various authors. More details are discussed later, including the results from recent shell model calculations. Table VI: Various values of $\left\langle m_{\nu}\right\rangle$ corresponding to $T_{1/2}^{0\nu}\left({}^{130}Te\right)=3.0\times 10^{24}$ yr. Authors/Reference | Method | $\left\langle m_{\nu}\right\rangle$ (eV) ---|---|--- [78] Rodin et al., 2007 | using $2\nu\beta\beta$-decay to fix $g_{pp}$ | 0.46 [62] Staudt et al., 1992 | pairing (Bohm) | 0.19 [63] Pantis et al., 1996 | no p-n pairing | 0.52 [64] Vogel, 1986 | | 0.47 [65] Civitarese and Suhoen 2006 | | 0.42 [66] Tomoda, 1991 | | 0.42 [67] Barbero, et al., 1999 | | 0.33 [68] Simkovic, 1999 | pn - RQRPA | 0.68 [69] Suhoen et al., 1992 | | 0.64 [67] Muto et al., 1989 | | 0.39 [71] Stoica et al., 2001 | | 0.60 [72] Faessler et al., 1998 | | 0.55 [73] Engel et al., 1989 | seniority | 0.29 [74] Aunola et al., 1998 | | 0.41 [79] Caurier et al., 2008 | Nuclear Shell Model | 0.58 Extracting the effective Majorana mass of the electron neutrino from the half- life requires the calculation of the nuclear structure factor, $F_{N}\equiv G^{0\nu}\left(M_{F}^{0\nu}-\left(g_{A}/g_{F}\right)^{2}M_{GT}^{0\nu}\right)$, in Eq. (7). This is not straightforward for the nuclei that are the best candidates for $0\nu\beta\beta$-decay experiments, e.g., ${}^{130}Te$, because they have many valence nucleons. To create a tractable shell-model calculation for these heavy nuclei, it is necessary to truncate the model space to the point that could affect the reliability of the results. Accordingly, schematic models are employed. As stated above, QRPA has become the standard approach for both $2\nu\beta\beta$ and $0\nu\beta\beta$ decay. The results calculated with QRPA, however, depend on the selection of a number of parameters, and the fact that different authors select the parameters in various ways has resulted in large differences in the resulting matrix elements as discussed in Ref. [61]. In Table VI, we list $14$ different values of $\left\langle m_{\nu}\right\rangle$ derived with QRPA and with renormalized QRPA, (RQRPA), corresponding to $T_{1/2}^{0\nu}\left({}^{130}Te\right)=3.0\times 10^{24}$ yr, and also the recent shell-model calculations of Caurier _et al._ [79]. From the table it is clear that the different ways of applying the same basic model has lead to a spread in the resulting matrix elements, and hence in the corresponding value of $\left\langle m_{\nu}\right\rangle$, of a factor of three [61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74]. This corresponds to differences of a factor of nine in the predicted half-life for a given value of $\left\langle m_{\nu}\right\rangle$, if all calculations are given the same weight. This assumption, however, cannot be justified. It should be recognized that calculation techniques, as well as computational power have made significant progress over the years, improving the reliability of both QRPA and shell-model calculations. In their recent article, Rodin, Simkovic, Faessler, and Vogel (Tübingen) [61], give detailed discussions of how the choices of various parameters in similar models can lead to such discrepancies. These are the gap of the pairing interactions, the use of (renormalized) RQRPA that partially accounts for the violation of the Pauli principle in the evaluation of the two-fermion commutators, the nucleon-nucleon interaction potential, the strength of the particle-hole interactions of the core polarization, the size of the model space, and the strength of the particle-particle interaction, parameterized by the quantity $g_{pp}$. The matrix elements of the virtual transitions through states with $J^{\pi}=1^{+}$ in the intermediate nucleus are extremely sensitive to the value of $g_{pp}$, which makes $2\nu\beta\beta$-decay matrix elements also very sensitive to it because this decay mode only proceeds through $1^{+}$ intermediate states. On the other hand, $0\nu\beta\beta$-decay also proceeds via higher multipoles through states of higher spin. These transitions are found to be far less sensitive to the value of $g_{pp}$. For this reason, Rodin _et al._ select the value of $g_{pp}$ that makes the calculation of the $2\nu\beta\beta$-decay half-life agree with the experimental value. In addition, some calculations are greatly simplified by using an average energy in the denominator of the second-order matrix-element expression, and the sum over the intermediate states is done by closure. When the value, $g_{A}=1.245$, of the axial-vector coupling constant obtained from muon decay is used, it commonly lead to a value of the Gamow-Teller strength typically larger than the measured value. To ameliorate this situation, a quenched value $g_{A}=1.00$ is used. In calculated rates of $2\nu\beta\beta$-decay, which proceed only through $J^{\pi}=1^{+}$ states, this results in a factor of $2.44$ reduction in the rate. Using the technique of Rodin _et al._ [61], the choice of $g_{aA}=1.00$ reduces the rate by between $10$ to $30\%$, depending on the particular nucleus. Another serious difference between some of the $0\nu\beta\beta$-decay calculations is due to the treatment of the short-range correlations in the nucleon-nucleon interactions. It was also pointed out by Simkovic _et al._ [68], that including the momentum dependent higher order terms of the nucleon current typically result in a reduction in the calculated value of the $0\nu\beta\beta$-decay matrix element by about $30\%$. These were included in the calculations of Refs. [60, 61]. In recent paper by Alvarez _et al._ [75], a QRPA formalism for $2\nu\beta\beta$-decay in deformed nuclei was presented. A considerable reduction in the matrix elements was observed in cases in which there was a significant difference in the deformations of the parent and daughter nuclides. Exactly how this would affect $0\nu\beta\beta$-decay calculations is not yet clear. It must be understood that this uncertainty, when resolved could result in a further reduction in neutrinoless double-beta decay matrix elements calculated within the framework of QRPA and RQRPA. In general, however, the paper by Rodin _et al._ [61], represents a detailed study of the various factors that cause the large variations in the nuclear matrix elements of $0\nu\beta\beta$-decay calculated by different authors over the years, and must be taken seriously. The procedure of Rodin _et al._ [59, 60, 61] has the attractive feature that it gives a straightforward prescription for selecting the very important particle-particle parameter, $g_{pp}$. However, Civitarese and Suhonen (referred to as the Jyväskylä group) have given strong arguments in favor of using single $\beta^{\pm}$-decay and electron capture data for this purpose, while giving arguments against using experimental $2\nu\beta\beta$-decay half lives [65]. They argue that only states with spin and parity $1^{+}$ can be the intermediate states involved in $2\nu\beta\beta$-decay, and that in the neutrinoless process these states play a minor role, and that the higher spin states play a dominant role. The Jyväskylä group recently presented a preprint in which they show that the effects of short-range correlations have been significantly overestimated in the past [76, 77]. Accordingly, their matrix elements originally gave a very different picture of the of the physics impact of the CUORICINO data presented in this paper. However, recently there have been some very important developments discussed below. ## IX RECENT DEVELOPMENTS IN QRPA CALCULATIONS We adopt the position that the large dispersion in values in the nuclear matrix elements implied by the values in Table VI does not reflect the true state of the art. Instead, we assume that there has been significant progress in understanding the key theoretical issues, as well as large increases in available computational power. Until very recently, however, two of the recent extensive theoretical treatments of the $0\nu\beta\beta$-decay matrix elements disagreed significantly, and in particular in the case of ${}^{130}Te$. The relevant nuclear structure factors, $F_{N}$, from the Jyväskylä and Tübingen groups for $g_{A}=1.25$ were $F_{N}\left({}^{130}Te\right)=1.20\pm 0.27\times 10^{-13}\textrm{ yr}^{-1}$ of Rodin _et al._ [61], and $F_{N}\left({}^{130}Te\right)=5.13\times 10^{-13}\textrm{ yr}^{-1}$ of Civitarese and Suhonen [65]. Recently an erratum was submitted by Rodin _et al._ [78] with major corrections to Table I of Ref. [61]. A coding error was discovered in the computation of the short-range correlations that, for example, increased the predicted $0\nu\beta\beta$-decay rate of ${}^{130}Te$ by a factor of $4.03$. Their corrected value of the nuclear structure factor of ${}^{130}Te$, is now $F_{N}\left({}^{130}Te\right)=4.84^{+1.30}_{-0.64}\times 10^{-13}\textrm{ yr}^{-1}$, in good agreement with the above value given by Civitarese and Suhonen. However, there is still a small disagreement between these two groups concerning the technique for calculating short-range correlations. Rodin _et al._ , used a Jastrow-correlation function, which has subsequently been shown by Kortelainen _et al._ [76] to overestimate the effects of short-range correlations, and hence to result in an excessive reduction in the nuclear matrix elements. Kortelainen _et al._ [77] have also updated the calculations of Civitarese and Suhonen. They extended their model space, for the cases of ${}^{116}Cd$, ${}^{128,130}Te$ and ${}^{136}Xe$, to include the $1p$-$0f$-$2s$-$1d$-$0g$-$2p$-$1f$-$0h$ single particle orbitals, calculated with a spherical Coulomb-corrected Woods-Saxon potential. In Ref. [77], a complete discussion is given of their method of fixing the parameters of the Hamiltonian. In this treatment they fix particle-particle parameter $g_{pp}$ of the pnQRPA using the method of Rodin _et al._ [59, 60, 61], namely with the experimentally measured $2\nu\beta\beta$-decay half-lives. They did not use the Jastrow-correlation function to correct for the short-range correlations, but rather they employ a ”unitary correlation operator method” (UCOM), which in the case of ${}^{130}Te$ increases the matrix element by a factor of $1.38$ over that calculated with the Jastrow correlation function. Their new values for the nuclear structure factors are: $\displaystyle F_{N}\left({}^{130}Te\right)_{g_{A}=1.25}$ $\displaystyle=7.47\times 10^{-13}\textrm{ yr}^{-1},$ $\displaystyle F_{N}\left({}^{130}Te\right)_{g_{A}=1.00}$ $\displaystyle=4.93\times 10^{-13}\textrm{ yr}^{-1}.$ This is to be compared to the results of the earlier work of Civitarese and Suhonen [65]. In any case, the major disagreements between the Jyväskylä and Tübingen groups have finally been understood, and the present difference in the predicted $0\nu\beta\beta$-decay rates of ${}^{130}Te$ now differ by a factor of $1.06$, whereas the earlier disagreement was by a factor of $4.28$. Some remaining differences might well lie in the differing methods of applying the short- range correlations (see also the discussion in Ref. [80]). In any case these recent developments have had a major impact on the interpretation of the CUORICINO data. Furthermore, the group of Caurier _et al._ [79], have recently given new values for these matrix elements from improved nuclear shell model calculations. The shell-model matrix elements are somewhat smaller than those of the recent Jyväskylä and corrected Tübingen results, and according to their matrix elements, the CUORICINO data imply: $\left\langle m_{\nu}\right\rangle\leq 0.58$ eV. ## X CUORICINO AS A TEST OF THE CLAIM OF DISCOVERY The CUORICINO array is the only operating $0\nu\beta\beta$-decay experiment, with energy resolution adequate to potentially probe the range of effective Majorana mass, $\left\langle m_{\nu}\right\rangle$, implied by the observation of $0\nu\beta\beta$-decay claimed by Klapdor-Kleingrothaus _et al._ [19, 20]. In the $2006$ article by Klapdor-Kleingrothaus and Krivosheina (KK& K) [20], the peak in the spectrum centered at $Q_{\beta\beta}\cong 2039$ keV is interpreted as the $0\nu\beta\beta$-decay of ${}^{76}Ge$, consistent with the range: $T_{1/2}^{0\nu}\left({}^{76}Ge\right)=\left\\{1.30-3.55\right\\}\times 10^{25}\textrm{ yr }(3\sigma)$. The best-fit value is $\left(2.23_{-0.31}^{+0.44}\right)\times 10^{25}$ yr. In this discussion we offer no critique of the claim, however, since this claim has been criticized from several points of view [21, 22, 23], it is interesting to ask if it is feasible to observe a $0\nu\beta\beta$-decay with this half-life with a significant confidence level with the published parameters of the experiment. Below, we show that the answer is ”yes”, the experiment could have made the observation in the range of half-lives quoted [20]. It is straightforward to derive an approximate analytical expression for the half-life sensitivity for discovery at a given confidence level that an experiment can achieve (see Appendix). The achievable discovery half-life, when the background rate is nonzero, is expressed as: $T_{1/2}^{0\nu}\left(n_{\sigma}\right)=\frac{4.17\times 10^{26}\textrm{ yr}}{n_{\sigma}}\left(\frac{\epsilon a}{W}\right)\sqrt{\frac{Mt}{\left(1+\zeta\right)b\,\delta(E)}}.$ (9) It is more conventional to simply have $b\,\delta(E)$ in the denominator of the root of Eq. (9) as prescribed by the Particle Data Book [81]. However, when the background continuum is obtained by a best fit to all peaks and continuum in the region, we choose this alternative approach. In Eq. (9), $n_{\sigma}$ is the desired number of standard deviations of the (CL) ($3$ for $\textrm{CL}=99.73\%$, for example), $\epsilon$ is the event detection and identification efficiency, $a$ is the isotopic abundance, $W$ is the molecular weight of the source material, $M$ is the total mass of the source, $\zeta$ is the signal-to-background ratio, $b$, is the specific background rate in counts/keV/kg/yr, and $\delta(E)$ is the instrumental width of the region of interest related to the energy resolution at the energy of the expected $0\nu\beta\beta$-decay peak. The values for these parameters for the Heidelberg-Moscow experiment [17, 19, 20] are: $Mt=71.7\textrm{ kg}\cdot\textrm{yr}$, $b=0.11\textrm{ kg}^{-1}\textrm{keV}^{-1}\textrm{yr}^{-1}$, $\epsilon=0.95$, $a=0.86$, $W=76$, and $\delta(E)=3.27\textrm{ keV}$. The number of counts under the identified peak at $2039$ keV is $28.75\pm 6.86$. The average value of the background near the region of interest was $11.6$ counts, therefore $\zeta\cong 2$. Direct substitution into Eq. (9) yields: $T_{1/2}^{0\nu}\left(4\sigma,\,^{76}Ge\right)=0.9\times 10^{25}\textrm{ yr};\quad T_{1/2}^{0\nu}\left(3\sigma\right)=1.2\times 10^{25}\textrm{ yr}.$ (10) Using the less conservative approach with $b\,\delta(E)$ in the denominator, the predicted half-life sensitivity for a discovery is $T_{1/2}^{0\nu}\left(4\sigma,\,^{76}Ge\right)=1.6\times 10^{25}\textrm{ yr};\quad T_{1/2}^{0\nu}\left(3\sigma\right)=2.13\times 10^{25}\textrm{ yr}.$ (11) These are close to the claimed most probable value given in Ref. [20]. This simple analysis is independent of the claimed result, with the exception of the determination of the signal to background ratio, $\zeta$. The conclusion is that with the given experimental parameters, this experiment could have had a discovery potential. Since this analysis does not account for statistical fluctuations, the discovery confidence level could possibly fall between $3\sigma$ and $5\sigma$. Any criticism of the claim would involve a reanalysis of the data, and the interpretation of the background peaks in the region. This falls outside of the scope of this discussion. Accordingly, we do not question the claim, but rather ask how well the present CUORICINO data confront it, now and in the future after five years of running. While the many theoretical calculations of the nuclear matrix elements over the years have differed significantly, the recently corrected-QRPA calculations of _Tübingen_ [78], those of _Jyväskylä_ [65], and shell model calculations of Caurier _et al._ [79], differ by less than about $30\%$. We have chosen to use for further analysis of the physics impact of the present CUORICINO data. Equation (8) can be inverted to obtain the values of the nuclear structure factor, $F_{N}$, using the calculated half-lives for $0\nu\beta\beta$-decay calculated with a given $\left\langle m_{\nu}\right\rangle$ by the authors of the theoretical papers. The resulting values are as follows: ${}^{76}Ge_{g_{A}=1.245}$: Rodin, et al.: $\displaystyle F_{N}$ $\displaystyle=1.22^{+0.10}_{-0.11}\times 10^{-13}\textrm{ yr}^{-1},$ Caurier, et al.: $\displaystyle F_{N}$ $\displaystyle=4.29\times 10^{-14}\textrm{ yr}^{-1},$ (12) Civitarese and Suhonen: $\displaystyle F_{N}$ $\displaystyle=7.01\times 10^{-14}\textrm{ yr}^{-1}$ ${}^{130}Te_{g_{A}=1.245}$: Rodin, et al.: $\displaystyle F_{N}$ $\displaystyle=4.84^{+1.30}_{-0.64}\times 10^{-13}\textrm{ yr}^{-1}\textrm{ (corrected value)},$ Caurier, et al.: $\displaystyle F_{N}$ $\displaystyle=2.57\times 10^{-13}\textrm{ yr}^{-1},$ (13) Civitarese and Suhonen: $\displaystyle F_{N}$ $\displaystyle=5.13\times 10^{-13}\textrm{ yr}^{-1}.$ The resulting values and ranges of values of $\left\langle m_{\nu}\right\rangle$ implied by the KK&K data, and by the CUORICINO data are as follows: $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{KK\&K}}^{\textrm{Rod}}=\\{0.23-0.43\\}\textrm{ eV},$ $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{CUOR}}^{\textrm{Rod}}\leq\\{0.38-0.46\\}\textrm{ eV},$ $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{KK\&K}}^{\textrm{Civ}}=\\{0.32-0.54\\}\textrm{ eV},$ (14) $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{CUOR}}^{\textrm{Civ}}\leq 0.41\textrm{ eV},$ $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{KK\&K}}^{\textrm{Cau}}=\\{0.41-0.68\\}\textrm{ eV},$ $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{CUOR}}^{\textrm{Cau}}\leq 0.58\textrm{ eV}.$ The results of the analyses with the new corrected matrix elements of Ref. [78] imply that the CUORICINO sensitivity has entered well into the range of values of $\left\langle m_{\nu}\right\rangle$ implied by the claim of KK&K. In the other two analyses, the CUORICINO data also constrain part of the range of values of $\left\langle m_{\nu}\right\rangle$ implied by KK&K. It is also interesting to try to predict the sensitivity of CUORICINO if it were to continue to operate for a total of $5$ years. The three recent calculations of the nuclear matrix elements result in the following predicted decay rates if the Heidelberg claim is correct. In this case, the decay rates would be: $\displaystyle\tau_{\textrm{KK\&K}}^{-1}\left({}^{76}Ge\right)$ $\displaystyle=\\{1.95-5.32\\}\times 10^{-26}\textrm{ yr}^{-1},$ $\displaystyle\tau_{\textrm{Rod}}^{-1}\left({}^{130}Te\right)$ $\displaystyle=\\{0.62-2.94\\}\times 10^{-25}\textrm{ yr}^{-1},$ (15) $\displaystyle\tau_{\textrm{Civ}}^{-1}\left({}^{130}Te\right)$ $\displaystyle=\\{1.43-3.89\\}\times 10^{-25}\textrm{ yr}^{-1},$ $\displaystyle\tau_{\textrm{Cau}}^{-1}\left({}^{130}Te\right)$ $\displaystyle=\\{1.17-3.19\\}\times 10^{-25}\textrm{ yr}^{-1}.$ Accordingly, we can calculate the number of $0\nu\beta\beta$-decay counts with $5$ years of live-time operation expected in the CUORICINO data consistent with the claim of KK&K. The exposure would be: $Nt\epsilon=2.85\times 10^{26}$ y, resulting in the following predicted number of real $0\nu\beta\beta$-decay events: $\displaystyle\tau_{\textrm{Rod}}^{-1}Nt\epsilon$ $\displaystyle=\\{18-84\\}_{0\nu\beta\beta},$ $\displaystyle\tau_{\textrm{Civ}}^{-1}Nt\epsilon$ $\displaystyle=\\{41-110\\}_{0\nu\beta\beta},$ (16) $\displaystyle\tau_{\textrm{Cau}}^{-1}Nt\epsilon$ $\displaystyle=\\{33-91\\}_{0\nu\beta\beta}.$ These counts would be superimposed on an expected background of $35$ to $39$ counts per keV in the $8$ keV region of interest centered at $2530$ keV. The constraints placed by the current CUORICINO data might favor the lower numbers in the ranges above. This would make it more challenging for CUORICINO to confirm the discovery claim of KK&K, and renders it almost impossible to rule out the KK&K claim with a significant level of confidence. The solution to this problem is the construction and operation of the proposed first tower of CUORE, called CUORE-$0$, combine its data with that of CUORICINO, and later the complete CUORE Experiment. ## XI THE PROPOSED CUORE EXPERIMENT The proposed CUORE detector will be made of $19$ towers of $TeO_{2}$ bolometers, very similar to the CUORICINO tower [28]. Each will house $13$ modules of four $5\times 5\times 5\textrm{ cm}^{3}$ crystals with masses of $\sim 750$ g. CUORE will contain $\sim 200$ kg of ${}^{130}Te$. The $988$ bolometers will have a total detector mass of $\sim 750$ kg and will operate at $8$-$10$ mK. An intense research and development program is underway to reduce the background to $0.01$ counts/keV/kg/yr. Thus far a reduction has been achieved that has reached within a factor of $2.4$ of this goal in the region of $2530$ keV, the $Q$-value for the $0\nu\beta\beta$-decay of ${}^{130}Te$. With this background, CUORE would reach a sensitivity of $\sim T_{1/2}^{0\nu}\left({}^{130}Te\right)\approx 2.1\times 10^{26}$ yr in $5$ years. The secondary goal is to achieve a background level of $0.001$ counts/keV/kg/yr. This would allow a half-life sensitivity of $T_{1/2}^{0\nu}\approx 6.5\times 10^{26}$ yr. In case that the background would be reduced to $0.001$ counts/keV/kg/yr, the associated sensitivities in the effective Majorana mass of the electron neutrino, $\left\langle m_{\nu}\right\rangle$, would be $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{Rod}}$ $\displaystyle=\\{0.026-0.031\\}\textrm{ eV},$ $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{Civ}}$ $\displaystyle=0.028\textrm{ eV},$ (17) $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{Cau}}$ $\displaystyle=0.040\textrm{ eV}.$ The half-life sensitivity is directly proportional to the abundance, $a$, of the parent $\beta\beta$-decay isotope [see equation (9)]. Accordingly, enriching the detectors of CUORE from $33.8\%$ in ${}^{130}Te$ to $90\%$, CUORE would achieve the same sensitivity with a background of $0.01$ counts/keV/kg/yr as it would with natural $Te$ and a background of $0.0014$ counts/keV/kg/yr. An R&D program, to determine the feasibility and cost of isotopically enriching CUORE is underway . In addition, the CUORE collaboration has a rigorous R&D program to improve the energy resolution from an average of $8$ keV, as it is in CUORICINO, to $5$ keV. This resolution should be achievable because some of the CUORICINO bolometers have already achieved $5$ keV. An intense program is underway to determine the cause of the spread in energy resolution. If in the end, CUORE does achieve the background of $0.001$ counts/keV/kg/yr, in addition is enriched, and has an average energy resolution of $5$ keV, it could reach a half life sensitivity of $2.5\times 10^{27}$ yr in $10$ years. In this case the sensitivities become: $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{Rod}}$ $\displaystyle=\\{13-16\\}\textrm{ meV},$ $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{Civ}}$ $\displaystyle=14\textrm{ meV},$ (18) $\displaystyle\left\langle m_{\nu}\right\rangle_{\textrm{Cau}}$ $\displaystyle=20\textrm{ meV}.$ This brings the sensitivity into the normal hierarchy region, which exceeds the goals of some of the other next generation experiments. It is possible to proceed as planned with a natural abundance version of CUORE, and then the bolometers could be replaced with those isotopically enriched in ${}^{130}Te$. This would increase the half-life reach by a factor of $2.5$ for an enrichment of $85\%$. ## XII SUMMARY AND CONCLUSIONS The CUORICINO detector is an array of $62$ $TeO_{2}$ bolometers operating at a temperature of about $8$ mK. It has a total mass of $40.7$ kg of $TeO_{2}$, containing $11$ kg of ${}^{130}Te$. It has operated for a total exposure of $N\left({}^{130}Te\right)t\epsilon=5.47\times 10^{25}$ yr, with no observation of $0\nu\beta\beta$-decay events, results in a lower bound, $T_{1/2}^{0\nu}\left({}^{130}Te\right)\geq 3.0\times 10^{24}$ yr. The corresponding upper bound on the effective Majorana mass of the electron neutrino, $\left\langle m_{\nu}\right\rangle$, using the corrected nuclear structure calculations of Rodin _et al._ , is $\left\langle m_{\nu}\right\rangle\leq(0.38-0.46)$ eV, while using those of Civitarese and Suhonen yields $\left\langle m_{\nu}\right\rangle\leq 0.47$ eV. With the recent shell model calculations the CUORICINO data imply $\left\langle m_{\nu}\right\rangle\leq 0.58$ eV. In all cases, the present CUORICINO data probe a significant portion of the range of the half life measured by KK&K. If the Heidelberg claim is correct, the nuclear structure calculations of Ref. [78] imply that after $5$ years of live time CUORICINO would detect $\\{18-84\\}$, $0\nu\beta\beta$-decay events, while those of Ref. [65] imply it would detect $\\{41-110\\}$ events, and those of Ref. [79] imply it would detect $\\{33-91\\}$ $0\nu\beta\beta$-events. In all cases, these counts would appear in Gaussian peaks with $\textrm{FWHM}=8k$ keV, superimposed on an average background of $35-39$ counts keV-1. In any case, the current results imply that the continued operation of CUORICINO is very important since it represents the only possibility of testing the claim of evidence of $0\nu\beta\beta$-decay for the next $5$ years or more. ## ACKNOWLEDGEMENTS The CUORICINO Collaboration owes many thanks to the Directors and Staff of the Laboratori Nazionali del Gran Sasso over the years of the development, construction and operation of CUORICINO, and to the technical staffs of our Laboratories. The experiment was supported by the Istituto Nazionale di Fisica Nucleare (INFN), the Commission of the European Community under Contract No. HPRN-CT-$2002$-$00322$, by the U.S. Department of Energy under Contract No. DE-AC$03$-$76$-SF$00098$, and DOE W-$7405$-Eng-$48$, and by the National Science Foundation Grants Nos. PHY-$0139294$ and PHY-$0500337$. We also wish to thank the following colleagues for their help and advice: Juoni Suhonen, Osvaldo Civiterese, Petr Vogel, Amand Faessler. Vadim Rodin, and Fedor Simkovic, and Fernando Ferroni. ## Appendix A APPENDIX An approximate expression for estimating the $0\nu\beta\beta$-decay half-life at which a given experiment can achieve discovery at the confidence level corresponding to $n_{\sigma}\sigma$, can be derived by reference to Figure 7. Let $"C"$ be the total number of counts found in the region of the expected $0\nu\beta\beta$-decay peak; let $"B"$ be the total number of background counts in the same energy interval, $\delta(E)$. For the number of real $0\nu\beta\beta$-decay events to have a statistical significance of $n_{\sigma}$, the following must be true: $C-B=n_{\sigma}\sqrt{C}$. In the usual case where $B\neq 0$, a desired signal to background ratio, $\zeta\equiv(C-B)/B$, can be chosen; hence $C=\left(1+\zeta\right)B$. The usual expression for the corresponding half-life can be written in terms of these parameters as: $T_{1/2}^{0\nu}\left(n_{\sigma}\right)=\frac{(\ln{2})Nt\epsilon}{n_{\sigma}\sqrt{\left(1+\zeta\right)B}}$ (A.1) where $N$ is the total number of parent nuclei, $\epsilon$ is the total detection efficiency, and $t$ is the live time of the data collection. The number of parent nuclei can be written in terms of, $M$, the total mass of the source (in an oxide for example), as follows: $N=\left(10^{3}\textrm{ g/kg/Wq/mole}\right)\cdot\left(A_{0}\textrm{ at/mole}\right)\cdot a\left(\textrm{abundance}\right)\cdot M\textrm{kg}$. Substituting these values, and expressing the background in terms of the background rate, $B=bM\delta(E)t$, where $b=\textrm{(counts/keV/kg/yr)}$, the expression is written: $T_{1/2}^{0\nu}\left(n_{\sigma}\right)=\frac{4.17\times 10^{26}}{n_{\sigma}}\left(\frac{a\epsilon}{W}\right)\sqrt{\frac{Mt}{\left(1+\zeta\right)b\delta(E)}}$ (A.2) Figure 7: Diagram showing the scheme on which Eq. (A.2) is derived. Of course in the case of zero background, Eq. (A.1) is used, and the quantity, $\left(1+\zeta\right)B$, is replaced the number of real events in the peak. In case there are no real or background events, i.e., $C=B=0$, the denominator of Eq. (A.1) is replaced by the usual quantity, $\ln{\\{1/(1-CL)\\}}$, which is $2.3$, ($90\%$ C.L.) for example, and $T_{1/2}^{0\nu}$ becomes an experimental lower limit. In Eq. (A.2), we use the fluctuation in the real events instead of that of the background because in these experiments the background level used is that of a best fit curve to the background in the region, and the fluctuation is a fitting error and is much smaller than the statistical fluctuations in the region of interest. ## References * [1] M. Goeppert-Mayer, Phys. Rev. 48, 512 (1935). * [2] Ya. B. Zel dovich, Yu. Luk yanov, and Ya. A. Smorodinski, Usp. Fiz. Nauk. 54, 361 (1954) * [3] H. Primakoff and S.P. Rosen, Rep. Prog. Phys. 22, 121 (1959); Annu. Rev. Nucl. Part. Sci. 31,145 (1981); W. C. Haxton and G. J. Stephenson Jr., Prog. Nucl. Part. Phys. 12, 409 (1984). * [4] Yu. G. Zdesenko, Rev. Mod. Phys. 74, 663 (2002); F. T. Avignone III, G. S. King III, and Yu. G. Zdesenko, New Journal of Physics 7, 6 (2005). * [5] S. R. Elliott and P. Vogel, Annu. Rev. Part. Sci. 52, 115 (2000). * [6] S. R. Elliott and J. Engel, J. Phys.G: Nucl. Part. Phys. 30, R183 (2004); F. T. Avignone III, S. R. Elliott, and J. Engel, Rev. Mod. Phys. 80, 481 (2008). * [7] T. Kajita and Y. Totsuka, Rev. Mod. Phys. 73, 85 (2001), and references therein. * [8] S. Fakuda _et al._ (SuperKamiokande Collaboration), Phys. Rev. Lett. 86, 5651 (2001). * [9] B. T. Cleveland _et al._ , Astrophys. J. 496, 505 (1998) and the references therein. * [10] J. N. Abdurashitov _et al._ (Sage Collaboration), J. Exp. Theor. Phys. 95, 181 (2002). * [11] W. Hampel _et al._ (GALLEX Collaboration), Phys. Lett. B 447,127 (2001); 86, 5656 (2001). * [12] Q. R. Ahmad _et al._ (SNO Collaboration), Phys. Rev. Lett. 87, 071301 (2001). * [13] J. N. Bahcall, M.C. Gonzales-Garcia, and C. Peña - Garay, Phys. Rev. C 66, 035802 (2002); J. High Energy Phys. 07, 54 (2002); New J. of Phys., 6, 63 (2004). * [14] K. Eguchi _et al._ (KamLAND Collaboration), Phys. Rev. Lett. 90, 021802 (2003); 92, 071301 (2004); T. Araki _et al._ _ibid._ , Phys. Rev. Lett. 94, 081801 (2005). * [15] APS Multidivisional Neutrino Study, Joint Study on the future of Neutrino Physics: The Neutrino Matrix; also see C. Aalseth _et al._ , arXiv:hep-ph/0412300. * [16] Recommendations to the Department of Energy and the National Science Foundation on a United States Program on Neutrino-less Double Beta Decay; Submitted to the Nuclear Science Advisory Committee and the High Energy Physics Advisory Panel by the Neutrino Scientific Assessment Group, September 1, 2005. * [17] L. Baudis _et al._ (Heidelberg-Moscow Collaboration), Phys. Rev. Lett. 83, 41 (1999); H. V. Klapdor-Kleingrothaus _et al._ , Eur. Phys. J. A 12, 147 (2001). * [18] C.E. Aalseth _et al._ , (IGEX Collaboration), Phys. Rev. C 59, 2108 (1999); Phys. Rev. D 65, 092007 (2002); 70, 078302 (2004). * [19] H. V. Klapdor-Kleingrothaus, A. Deitz, H. L.Harney and I. V. Krivosheina, Mod. Phys. Lett. A 16, 2409 (2001). * [20] H. V. Klapdor-Kleingrothaus _et al._ , Phys. Lett. B 586, 198 (2004); Nucl. Instrum. Methods A 522, 371 (2004); H. V. Klapdor-Kleingrothaus and I. V. Krivosheina, Mod. Phys. Lett. 21, 1547 (2006). * [21] C.E. Aalseth _et al._ , Mod. Phys. Lett. A 17, 1475 (2002). * [22] Yu. G. Zdesenko, F. A. Danevich, and V. I. Tretyak, Phys. Lett. B 546, 206 (2002). * [23] Ferruccio Feruglio, Alessandro Strumia and Francesco Vissani, Nucl. Phys. B 637, 345 (2002). * [24] I. Abt _et al._ (GERDA Collaboration), arXiv:hep-ex/0404039. * [25] C. Arnaboldi _et al._ (CUORICINO Collaboration), Phys. Lett. B 584, 260 (2004). * [26] C. Arnaboldi _et al._ (CUORICINO Collaboration), Phys. Rev. Lett. 95, 142501 (2005). * [27] C. E. Aalseth _et al._ (Majorana Collaboration), Nucl. Phys. B, Proc. Suppl. 138, 217 (2005); also see arXiv:nucl-ex/0311013. * [28] R. Ardito _et al._ (CUORE Collaboration), arXiv:hep-ex/0501010. * [29] M. Danilov _et al._ , Phys. Lett. B 480, 12 (2000); D. Akimov _et al._ , Nucl. Phys. B, Proc. Suppl. 138, 224 (2005). * [30] A. I. Belesev _et al._ , Phys. Lett. B 350, 263 (1995); V. M. Lobashev _et al._ , _ibid._ Phys. Lett. B 460, 227 (1999). * [31] Ch. Kraus _et al._ , Eur. Phys. J. C 40, 447 (2005). * [32] A. Osipowicz _et al._ , arXiv:hep-ex/0109033; V. M. Lobashev, Nucl. Phys. A 719, 153 (2003), and references therein. * [33] V. Barger _et al._ , Phys. Lett. B 595, 55 (2004). * [34] M. Tegmark _et al._ , Phys. Rev. D 69, 103501 (2004). * [35] W. J. Percival _et al._ , Mon. Not. R. Astron. Soc,. 327, 1297 (2001); M. Colles _et al._ , astro-ph/1016498; Mon. Not. R. Astron. Soc. 328,1039 (2001). * [36] C.L. Bennett _et al._ , Astrophys. J. Suppl. 148, 1 (2003); D. N. Spergel _et al._ , Astrophys. J. Suppl. Ser. 148,175 (2003). * [37] S. Hannestad, J. Cosmol. Astropart. Phys. 05, 004 (2003); astro-ph/0303076. * [38] S. W. Allen, R. W. Schmidt, and S. L. Briddle, Mon. Not. R. Astron. Soc. 346, 593 (2003). * [39] V. Barger, D. Marfatia, and K. Whisnant, Int. J. Mod. Phys. E 12, 569 (2003); see also Patric Crotty, Julien Lesgourgues, and Sergio Pastor, Phys. Rev. D 69, 123007 (2004), and references therein. * [40] M. Apollonio _et al._ , Phys. Lett. B 466, 415 (1999). * [41] F. Boehm _et al._ , Phs. Rev D 64, 112001 (2001). * [42] E. Fiorini and T. Niinikoski, Nucl. Instrum. Methods 224, 83 (1984). * [43] C. Arnabold, _et al._ , Phys. Lett. B 557, 167 (2003), and references therein. * [44] C. Arnaboldi _et al._ , Nucl. Instr. Meth. A 518, 775 (2004). * [45] M. Barucci _et al._ , J. Low Low Temp. Phys. 123, 303 (2001). * [46] G. K. White, S. J. Collocott, and J. G. Collins, J. Phys. Condens. Matter 2, 7715 (1990). * [47] C. Arnaboldi _et al._ , Phys. Lett. B 557, 167 (2003). * [48] A. Allessandrello _et al._ , Nucl. Instrum. Meth. Phys. Res. A 142, 163 (1998); 412, 454 (1998). * [49] A. H. Wapstra, G. Audi, and C. Thibault, Nucl. Phys. A 729, 337 (2003). * [50] A. Alessendrello _et. al._ , Cryogenics 37, 27 (1997). * [51] A. Allessendrello _et al._ , J. Phys. D 32, 3099 (1999). * [52] E. E. Haller _et al._ , in _”Neutron transmutation doping of semiconductor materials”_ , edited by R. D. Larrabee (Plenum Press, New York, 1984), p. 21. * [53] C. Arnaboldi _et al._ , IEEE Trans. Nucl. Sci. 49, 1808 (2002). * [54] C. Arnaboldi _et al._ , IEEE Trans. Nucl. Sci. 49, 2440 (2002). * [55] C. Ardaboldi _et al._ , IEEE Trans. Nucl. Sci. 50, 979 (2003). * [56] C. Arnaboldi _et al._ , IEEE Trans. Nucl. Sci., 52, 1630 (2005). * [57] R. M. Barnett _et al._ (Particle Data Group), Phys. Rev. D 54,1 (1996). * [58] S. Baker and P. D. Cousins, Nucl. Instrum. Methods. Phys. Res. A 221, 437 (1984). * [59] V. A. Rodin, A. Faessler, F. Simkovic, and P. Vogel, Czeh. J. Phys. 56, 495 (2006). * [60] V. A. Rodin, A. Faessler, F. Simkovic, and P. Vogel, Phys. Rev. C 68, 044302 (2003). * [61] V. A. Rodin, A. Faessler, F. Simkovic, and P. Vogel, Nucl. Phys. A 766, 107 (2006). * [62] A. Staudt, T. T. S. Kuo, and H. V. Klapdor-Kleingrothaus, Phys. Rev. C 46, 871 (1992). * [63] G. Pantis, F. Simkovic, J. D. Vergados, and A. Faessler, Phys. Rev. C 53, 695 (1996). * [64] P. Vogel and M.R. Zirnbauer, Phys. Rev. Lett. 57, 3148 (1986); P. Vogel, M. R. Zirnbauer. J. Engel, Phys. Rev. C 37, 731 (1988); M. Moe and P. Vogel, Annu. Rev. Nucl. Part. Sci. 44, 247 (1994). * [65] O. Civitarese and J. Suhonen, Nucl. Phys. A 761,313 (2005). * [66] T. Tomoda, Rep. Prog. Phys. 54, 53 (1991). * [67] C. Barbero _et al._ , Nucl. Phys. A 650, 485 (1999). * [68] F. Simkovic, G. Pantis, J. D. Vergados and A. Faessler , Phys. Rev. C 60, 055502 (1999). * [69] J. Suhonen, O. Civitarese, and A. Faessler, Nucl. Phys. A 543, 645 (1992). * [70] K. Muto, E. Bender, and H.V. Klapdor, Z. Phys. A 334, 187 (1989). * [71] S. Stoica and H.V. Klapdor-Kleingrothaus, Phys. Rev. C 63, 064304 (2001). * [72] A. Faessler and F. Simkovic, J. Phys. G 24, 2139 (1998). * [73] J. Engel _et al._ , Phys. Lett. B 225, 5 (1989). * [74] M. Aunola and J. Suhonen, Nucl. Phys. A 643, 207 (1998). * [75] R. Alvarez-Rodreguez _et al._ , Phys. Rev. C 70, 064309 (2004). * [76] M. Kortelainen, O. Civitarese, J. Suhonen, and J. Toivanen, Phys. Lett. B 647, 128 (2007). * [77] M. Kortelainen and J. Suhonen, Phys. Rev. C 75, 051303 (2007); Phys. Rev. C76, 025315 (2007). * [78] V. A. Rodin, A. Faessler, F. Sikovic, and P. Vogel, Nucl. Phys. A 766, 107 (2006); [Erratum Nucl. Phys. A 793, 213 (2007)]. * [79] A. Poves , talk at the ILIAS meeting, Chambery, France, Feb. 25-28, 2007; E. Caurier, F. Nowacki, and A. Poves, arXiv:0709.0277 [nucl-th]; Eur. Phys. J. A 36, 195 (2008). * [80] F. Simkovic _et al._ , arXiv: 0710.2055 [nucl-th]. * [81] The Review of Prticle Physics, C. Amsler _et al._ , Phys. Lett. B 667, 1 (2008).

arxiv-papers

2024-09-04T02:48:53.978759

0802.3474

# Classification of electromagnetic fields in general relativity and its physical applications Nikolai V. Mitskievich Physics Department, CUCEI, University of Guadalajara, Guadalajara, Jal., Mexico.Postal address: Apartado Postal 1-2011, C.P. 44100, Guadalajara, Jalisco, México. E-mail: mitskievich03@yahoo.com.mx ( ) ###### Abstract The simplest electromagnetic fields’ (general- as well as special- relativistic) classification is formulated which is based on physically motivated ideas. According to this classification these fields can belong to three types (electric, magnetic and null), each of them being split in pure and impure subtypes. Only pure null type field propagates with the fundamental velocity $c$, all other fields have the propagation velocity less than that of light. The reference-frame-based methods of elimination of alternative three- fields (e.g., magnetic in the electric type case) are given for pure subtypes; for pure null type the generalized Doppler effect takes place instead. All three types of impure fields are shown to be E-B-parallelizable. Thus such an elimination in pure non-null and parallelization in all impure cases mean transformation to the reference frame co-moving with the electromagnetic field in which the Poynting vector vanishes. The methods we propose modernizing the Rainich–Misner–Wheeler approach, also permit to construct new exact Einstein–Maxwell solutions from already known seed solutions. As examples, the Kerr–Newman and Liénard–Wiechert solutions are considered, three “new” types of rotating charged black holes (with the same Kerr-Newman geometry) are presented, and new physical effects are evaluated. PACS 2008 Numbers: 04.20-q, 04.20.Ex, 04.40.Nr, 04.70.Bw ## 1 Introduction The main idea of this paper is to present an elementary practical classification of electromagnetic fields (see also [12]) equally applicable in both relativities and having deep physical roots. In fact, there already exist classifications of stress-energy tensors (essentially, those of Segrè and Plebański, [14, 15, 17]) and, specifically, of electromagnetic fields [18]. However, one does not encounter there a direct relation to such field properties as the Poynting vector and the velocities of propagation of concrete configurations of these fields (in particular, with respect to the reference frames co-moving with the corresponding fields when these velocities are $<c=1$ in the natural units used here; for a detailed discussion of the Liénard–Wiechert example see our recent article [12] and a preprint with complete simple deduction of that solution [10]). Now, our aim is also to give here natural and simple prescriptions for calculation of such physical characteristics of every concrete electromagnetic field (cf. [8, 16, 23]). We consider electromagnetic fields in a vacuum (but electric currents may be present, not the media with dielectric and magnetic properties), this theory being treated here on the classical (not quantum theoretical) basis. The state of motion (in general, inhomogeneous) of the electromagnetic field, which we call its propagation, is that of a concrete (preferably, exact) solution of the system of dynamical equations (of Maxwell in special theory and Einstein–Maxwell in general theory of relativity), but not the propagation of perturbations on the background of these solutions, and not the propagation of discontinuities, which belong to other problems of the field theory not considered in this paper (their treatment is already well developed and does not need immediate revision). Any motion is, naturally, relative (that occurring with the velocity of light is also relative, at least in the sense of its direction: the light aberration effect), while motions with under- luminal velocities always permit to introduce co-moving reference frames with respect to which the fields are “at rest” everywhere in the four-dimensional region of the frame determination, which includes the requirement that redistributions of the field (“deformations”) should be “caused” by deformation, acceleration and rotation of the respective co-moving reference frame, and vice versa. In this connection, we remind that the physical reference frame is an idealized image of a changing with time (let us not to put this concept here more precisely) distribution and motion of observers together with their observational and measuring devices, idealized primarily in the sense to be test objects, i.e. they should not practically perturb the characteristics of the objects (including fields and spacetime) in the classical (non-quantum) theory. We do not touch here upon the problem of quantum theoretical description of reference frames: this question still seems not to be adequately considered in physics, although the first step in this direction was already made by Bohr and Rosenfeld [1]. ### 1.1 A preview of the paper’s structure The reader will notice that in this paper the original conclusions are enlarged with a modernized review of already known facts to make the exposition more self-sufficient. In the next section the notations and definitions used in this paper are given. In section 3 we present a condensed information on description of reference frames and its application to electromagnetic fields. The short section 4 is dedicated to the classification of these fields in terms of their two invariants, $I_{1}$ and $I_{2}$. In section 5 we introduce the concept of propagation velocity of electromagnetic fields in a vacuum, and it is shown that absolute value the three-velocity of all pure null fields is equal to unity. The pure electromagnetic fields (when $I_{2}=0$) are considered in section 6 yielding for the pure electric and magnetic types a simple elimination of magnetic or electric field, respectively, by the corresponding choices of reference frame (subsection 6.1), and a specific rôle of the Doppler effect (with its inevitable generalization) for the pure null type in subsection 6.2. An approach to constructing exact Einstein–Maxwell solutions in the same 4-geometry as that of any exact seed Einstein–Maxwell solution one arbitrarily would choose (with the exception of pure null fields), is developed in sections 7 and 8. Section 9 is dedicated to the treatment of impure subtype ($I_{2}\neq 0$) of all three types of fields leading to parallelization of electric and magnetic vectors in the adequate (canonical) reference frames. In sections 10 and 11 we consider application of the methods developed in the preceding sections to some exact solutions of the Einstein–Maxwell equations in general relativity and Maxwell’s equations in special relativity: in subsection 10.1 of the well- known Kerr–Newman (KN) solution (involving, as we show, in different regions different types of electromagnetic field whose electric and magnetic vectors are always collinear and radially-directed), and in subsection 11.1, the Liénard–Wiechert solution (we show that it belongs to the pure electric type and there always exists a global co-moving with this electromagnetic field non-degenerate reference frame, so that the velocity of propagation of this field in a vacuum is everywhere less than that of light, with the exception of the future null infinity). In subsection 10.2 “new” exact solutions with pure electric and pure magnetic fields in the standard Kerr-Newman black hole geometry are presented, together with a similar black hole with impure-null- type electromagnetic field. In subsection 11.2, it is shown that a superposition of plane harmonic electromagnetic wave and homogeneous magnetic field has strictly sub-luminal velocity of its propagation in a vacuum. In the final section 12 the obtained results are summed up and concluding remarks are given. ## 2 Mathematical preliminaries Everything will be considered in four spacetime dimensions. We use the spacetime signature $+,-,-,-$, Greek indices being four-dimensional (running from 0 to 3), and Latin ones, three-dimensional, with the Einstein convention of summation over dummy indices. However, in the reference frame formalism, all indices usually are Greek, and the splitting into physical spacelike and timelike objects only means that the former ones are in all free indices orthogonal to the timelike monad vector (projected onto the physical three- space of the reference frame), while the physical timelike parts represent contractions with the monad in the indices which hence become absent (a change of the root-letter notation is then advisable). The indices put into individual parentheses belong to tetrad components. For the sake of convenience and writing and reading economy, the Cartan exterior forms formalism is frequently used. In it, the coordinated basis is the set of four covectors (1-forms) $dx^{0},\dots,dx^{3}$, and the orthonormal tetrad basis similarly is $\theta^{(0)},\dots,\theta^{(3)}$. Every such basis 1-form, (e.g., $dx^{2}$, $\theta^{(3)}$), itself represents an individual four-dimensional covector. The exterior (wedge) product simply is a skew- symmetrized tensorial product (antisymmetrization is also denoted by Bach’s square brackets which embrace the indices, while factor $\frac{1}{(\textnormal{their number})!}$ is supposed to be included in this definition). It is clear that the rank of a form can be from 0 (a scalar) to 4, inclusively; all forms of higher ranks vanish identically (in $D=4$). The scalar product of four-vectors or covectors is denoted by a central dot, if these vectors are written without indices ($A\cdot B$), but with such indices this has a wider meaning, for example, $dx^{\mu}\cdot dx^{\nu}=g^{\mu\nu}$: literally, this means that the scalar product of two coordinated-basis covectors equals a contravariant component of the metric tensor with the same indices as those of these factors. The dual conjugation in the sense of components (their indices) is denoted by an asterisk over the corresponding subindices, or under upper indices; the Hodge star stands for dual conjugation of a form written more abstractly, and is denoted by an asterisk before the form; it is convenient to have in mind that, after all, it applies to the form’s basis, though this is equivalent to a similar dual conjugation of the form’s components (not both at once!). An application of a pair of Hodge stars does not change an odd-rank form and results in the change of the sign of an even-rank form (for example, the electromagnetic field 2-form $F$). By this definition, $\ast(dx^{\alpha_{1}}\wedge\dots\wedge dx^{\alpha_{k}}):=\frac{1}{(4-p)!}{E^{\alpha_{1}\dots\alpha_{k}}}_{\beta_{1}\dots\beta_{l}}(dx^{\beta_{1}}\wedge\dots\wedge dx^{\beta_{l}})$ (2.1) where $E_{\kappa\lambda\mu\nu}:=\sqrt{-g}\epsilon_{\kappa\lambda\mu\nu},~{}~{}E^{\kappa\lambda\mu\nu}:=-\frac{1}{\sqrt{-g}}\epsilon_{\kappa\lambda\mu\nu}$ (2.2) are covariant and contravariant components of the axial Levi-Cività tensor, and the usual Levi-Cività symbol is defined as $\epsilon_{\kappa\lambda\mu\nu}=\epsilon_{[\kappa\lambda\mu\nu]},~{}~{}\epsilon_{0123}=+1$ (2.3) (always with the sub indices: this is a symbol, though simultaneously representing components of a contravariant axial tensor density of the weight $-1$ and a covariant axial tensor density of the weight $+1$). See some details in the beginning of the introductive chapter in [11]. Finally, coming back to a formula in the end of the second paragraph of this section, we have $\ast(dx^{\mu}\wedge\ast dx^{\nu})=-dx^{\mu}\cdot dx^{\nu}=-g^{\mu\nu}$. Of course, the rôle of metric properties of spacetime is somewhat hidden in the Hodge notations, as one can see from the formulae (2.1) and (2.2). ## 3 Algebra of reference frames; applications to the electromagnetic field The central point of our paper is the use of algebraic considerations, other applications being here only of auxiliary significance, for example, the exterior differentiation operator $d=\theta^{(\alpha)}\nabla_{X_{(\alpha)}}\wedge\equiv dx^{\alpha}\nabla_{\partial_{\alpha}}\wedge$. In the physical sense, a concrete reference frame (see [11]) has only to do with a state of motion (a timelike world lines’ congruence, or, equivalently, its unitary tangent vector field, the monad $\tau$) of a swarm of test observers together with their test measuring devices. Moreover, one additional ingredient, the metric tensor $g$, is needed to construct the projector $b:=g-\tau\otimes\tau$ which at the same time serves as the (formally, four- dimensional) metric tensor on the three-dimensional local subspace orthogonal to the monad field $\tau$; $b_{\mu\nu}\tau^{\nu}\equiv 0$, $\det b\equiv 0$. $b$ has the signature $0,-,-,-$, so that the “three-dimensional” scalar product of two vectors is $A\bullet B:=-b_{\alpha\beta}A^{\alpha}B^{\beta}\equiv\ast[(\tau\wedge A)\wedge\ast(\tau\wedge B)]$ (3.1) where these vectors are also automatically projected onto the local subspace mentioned above. If such vectors did already belong to the subspace, they usually are boldfaced: $\mathbf{A}^{\mu}=b^{\mu}_{\nu}A^{\nu}$. The “three- dimensional” axial vector product of two vectors now reads $A\times B:=\ast(A\wedge\tau\wedge B).$ (3.2) These algebraic operations are locally equivalent to the usual three- dimensional scalar and vector products, so we denote them by essentially the same symbols. In fact, in the complete reference frame theory we similarly use the operations of gradient, divergence and curl, but, being differential operators, they are more profoundly generalized, explicitly taking into account the characteristics of inhomogeneities of general reference frames, such as acceleration, rotation and deformation (expansion and shear) which naturally cannot be present in the algebraic treatment of geometry. The uniformity of general and old traditional notations radically simplifies the physical interpretation of general- and (in non-inertial frames) special- relativistic expressions as well as of theoretically predicted effects. Electromagnetic fields are described with the use of the covector potential $A=A_{\alpha}dx^{\alpha}$ and the 2-form (the field tensor) $F=dA=\frac{1}{2}F_{\alpha\beta}dx^{\alpha}\wedge dx^{\beta}.$ (3.3) With respect to a given reference frame $\tau$ (see [11]), the field tensor splits into two four-dimensional (co)vectors, electric $\textnormal{{\bf E}}_{\mu}=F_{\mu\nu}\tau^{\nu}~{}~{}\Longleftrightarrow~{}~{}\textnormal{{\bf E}}=\ast(\tau\wedge\ast F)$ (3.4) and magnetic $\textnormal{{\bf B}}_{\mu}=-F\\!\\!\stackrel{{\scriptstyle\textnormal{\small$\ast$}}}{{\textnormal{\scriptsize$\mu\nu$}}}\\!\tau^{\nu}~{}~{}\Longleftrightarrow~{}~{}\textnormal{{\bf B}}=\ast(\tau\wedge F),$ (3.5) both $\perp\tau$, thus $F=\mathbf{E}\wedge\tau+\ast(\mathbf{B}\wedge\tau).$ (3.6) It is obvious that E is a polar four-vector and B, an axial four-vector, both restricted to the local physical three-subspace of the $\tau$-reference frame. (In Cartesian coordinates and with the corresponding inertial monad, consequently, in the Minkowskian spacetime, we have the same relations as for usual contravariant three-vectors: $\textnormal{{\bf E}}^{i}=F_{i0}=-F^{i0},~{}~{}\textnormal{{\bf B}}^{i}=-\frac{1}{2}\epsilon_{ijk}F_{jk}=-\frac{1}{2}\epsilon_{ijk}F^{jk}$.) The splitting (3.4), (3.5), (3.6) follows from the observation that the Lorentz force can be expressed as $(\textnormal{{\bf E}}+\textnormal{{\bf v}}\times\textnormal{{\bf B}})_{\alpha}=F_{\mu\nu}\left(\tau^{\nu}+\textnormal{{\bf v}}^{\nu}\right)b^{\mu}_{\alpha}.$ (3.7) Here the three-velocity of the charged particle on which acts the Lorentz force, follows from the general definition $u=\stackrel{{\scriptstyle(\tau)}}{{u}}(\tau+\mathbf{v})~{}\Rightarrow~{}\textnormal{\bf v}=b(\frac{dx}{dt},\cdot)$ (3.8) where $\stackrel{{\scriptstyle(\tau)}}{{u}}=u\cdot\tau=\frac{dt}{ds}=(1-v^{2})^{-1/2}$, while $dt=\tau_{\mu}dx^{\mu}$ ($\tau\cdot dx$, that is non-total differential of the physical time along an infinitesimal displacement of the particle in spacetime), and $u$ is its four-velocity. We have to add here important comments related to the basic concepts of both relativities, and these comments could be more transparent just with the three-velocity as an intuitively clear example. The “physical” objects (such as $\mathbf{v}$, $\mathbf{E}$, etc.) belong to the section orthogonal to $\tau$ using which these objects are introduced. Thus, already in special relativity, the velocities considered in their composition law, may exist even in three distinct sections of spacetime, while three frames are participating in the composition, and it is absolutely obvious that one cannot simply add vectors from two subspaces obtaining the third one automatically lying in the third subspace, all of them having necessary properties with respect to these respective frames. And in the composition law the three-vectors being added together, frequently are “collinear” (an absurd if they belong to non-parallel sections of spacetime). This is, of course, understandable, since Einstein himself did not realize the fact of unification of space and time into the four-dimensional manifold before the famous discovery of Minkowski in 1908 (and even during several years after this discovery). The only two electromagnetic invariants being important in the Einstein–Maxwell theory can be easily introduced: $I_{1}=-2\ast(F\wedge\ast F)=F_{\mu\nu}F^{\mu\nu}=2\left(\mathbf{B}^{2}-\mathbf{E}^{2}\right),$ (3.9) $I_{2}=2\ast(F\wedge F)=F\\!\stackrel{{\scriptstyle\textnormal{\small$\ast$}}}{{\textnormal{\scriptsize$\mu\nu$}}}F^{\mu\nu}=4\textnormal{{\bf E}}\bullet\textnormal{{\bf B}}.\phantom{aaaaaa.}$ (3.10) These invariants enter the following important identities: $F_{\mu\nu}F^{\lambda\nu}-F\\!\stackrel{{\scriptstyle\textnormal{\small$\ast$}}}{{\textnormal{\scriptsize$\mu\nu$}}}F\\!\stackrel{{\scriptstyle\textnormal{\scriptsize$\lambda\nu$}}}{{\textnormal{\small$\ast$}}}=\frac{1}{2}I_{1}\,\delta^{\lambda}_{\mu},~{}~{}~{}F\\!\stackrel{{\scriptstyle\textnormal{\small$\ast$}}}{{\textnormal{\scriptsize$\mu\nu$}}}F^{\lambda\nu}=\frac{1}{4}I_{2}\,\delta^{\lambda}_{\mu}.$ (3.11) In fact, $I_{2}$ is an axial (pseudo-) invariant whose square behaves as a usual scalar. The electromagnetic stress-energy tensor is [9, 11] $T^{\nu}_{\mu}=\frac{1}{4\pi}\left(\frac{1}{4}F_{\kappa\lambda}F^{\kappa\lambda}\delta^{\nu}_{\mu}-F_{\mu\lambda}F^{\nu\lambda}\right)=-\frac{1}{8\pi}\left(F_{\mu\lambda}F^{\nu\lambda}+F\\!\stackrel{{\scriptstyle\textnormal{\small$\ast$}}}{{\textnormal{\scriptsize$\mu\lambda$}}}F\\!\stackrel{{\scriptstyle\textnormal{\scriptsize$\nu\lambda$}}}{{\textnormal{\small$\ast$}}}\right)$ (3.12) (in Gaussian units). Its (single) contraction with arbitrary monad includes the electromagnetic energy density and Poynting vector in that frame, $T^{\nu}_{\mu}\tau_{\nu}=\frac{1}{8\pi}\left[\left(\textnormal{{\bf E}}^{2}+\textnormal{{\bf B}}^{2}\right)\tau_{\mu}+2(\textnormal{{\bf E}}\times\textnormal{{\bf B}})_{\mu}\right],$ (3.13) and the squared expression is (see (3.11) and cf. [23, 16]) $T^{\nu}_{\mu}T^{\mu}_{\xi}\tau_{\nu}\tau^{\xi}=\frac{1}{(8\pi)^{2}}\left[\left(\textnormal{{\bf E}}^{2}+\textnormal{{\bf B}}^{2}\right)^{2}-4(\textnormal{{\bf E}}\times\textnormal{{\bf B}})^{2}\right]\\\ \equiv\frac{1}{(8\pi)^{2}}\left[\left(\textnormal{{\bf B}}^{2}-\textnormal{{\bf E}}^{2}\right)^{2}+4(\textnormal{{\bf E}}\bullet\textnormal{{\bf B}})^{2}\right]=\frac{1}{(16\pi)^{2}}\left({I_{1}}^{2}+{I_{2}}^{2}\right).$ (3.14) It is interesting that these constructions are not only scalars under transformations of coordinates, but they are also independent of the reference frame choice: the right-hand side does not involve any mention of the monad at all. ## 4 A classification of electromagnetic fields The simple and exhaustive classification of electromagnetic fields is based on existence of only two invariants, (3.9) and (3.10), built with the field tensor $F_{\mu\nu}$, while all other invariants are merely algebraic functions of these two invariants (if not vanish identically). Since $I_{2}$ itself is a pseudo-invariant (axial scalar) which acquires the factor sign$(J):=J/|J|$ under a general transformation of coordinates, $J$ being its Jacobian, the concrete sign of $I_{2}$ does not matter in our classification. In terms of $I_{1}$ the invariant classification suggests three types of fields: $I_{1}<0$ is the electric type (the electric field dominates), $I_{1}>0$ gives the magnetic type, and to $I_{1}=0$, the null type corresponds. The pseudo-invariant $I_{2}$ permits to work out the classification in more detail: we get additional subtypes, impure ($I_{2}\neq 0$) and pure ($I_{2}=0$). Below we shall see how this classification enables us to find reference frames most adequately suitable for description of concrete electromagnetic fields and even to construct new exact solutions of Einstein–Maxwell’s equations. It also gives a natural base for straightforward physical interpretation of these fields. ## 5 Propagation of electromagnetic fields Considering the propagation of electromagnetic field, we do not include the high-frequency limits related to field discontinuities (bicharacteristics). The Poynting vector plays an important rôle in electrodynamics having two distinct meanings: of the energy density flow and of the linear momentum density due to symmetry of the electromagnetic energy-momentum tensor (in natural units velocity is dimensionless and that of light in a vacuum is $c=1$). It is worth giving more comments on physical interpretation of the Poynting vector. It does not always describe propagation of extractable energy of the field and even a real motion (see also [19]); the exclusion is here related to the special case of static and stationary fields (whose frequency is equal to zero). Thus the Poynting vector, together with the electromagnetic energy density, determines (sometimes formally) the propagation three-velocity of electromagnetic field with respect to the reference frame in which the expression (3.13) is given. We take this velocity according to Landau and Lifshitz [7] (see the problem in p. 69) as $\frac{\mathbf{v}}{1+\mathbf{v}^{2}}=\frac{\mathbf{E}\times\mathbf{B}}{\mathbf{E}^{2}+\mathbf{B}^{2}}$ (5.1) (an alternative definition see in [13], p. 115, formula (312), $\mathbf{v}=2\frac{\mathbf{E}\times\mathbf{B}}{\mathbf{E}^{2}+\mathbf{B}^{2}},$ (5.2) but this definition is false as it can be seen from subsection 11.2 below). In the preceding pages in [7], an interesting discussion of electromagnetic invariants is worth being noted. From (3.14) and (5.1) we see that $0\leq\frac{|\mathbf{v}|}{1+\mathbf{v}^{2}}=\frac{1}{2}\sqrt{1-\frac{{I_{1}}^{2}+{I_{2}}^{2}}{4(\mathbf{E}^{2}+\mathbf{B}^{2})^{2}}}=\frac{|\mathbf{E}||\mathbf{B}|}{\mathbf{E}^{2}+\mathbf{B}^{2}}|\sin\psi|\leq\frac{1}{2},$ (5.3) $\psi$ being the angle between E and B in the strict local Euclidean sense; moreover, the function $|\mathbf{v}|/(1+\mathbf{v}^{2})$ is everywhere monotonic. In particular, this means that the propagation of all pure null fields ($|\mathbf{E}|=|\mathbf{B}|$, $\psi=\pi/2$) occurs with the unit absolute value of the three-velocity, the velocity of light, and all other electromagnetic fields propagate with sub-luminal velocities which can always be made equal to zero in corresponding co-moving reference frames. This is the general-relativistic conclusion, only expressed in three-dimensional notations characteristic to the general reference frame theory. ## 6 Dealing with pure electromagnetic fields Pure electromagnetic fields represent the simplest cases, especially in the non-null types when there always exist reference frames in which either magnetic or electric field can be easily eliminated. The pure null type requires more thorough examination involving a consideration of the Doppler effect (here, its generalized counterpart) which we have to discuss below in more detail. ### 6.1 Pure electric and magnetic type fields Vanishing of the second invariant, $I_{2}$, means that the electromagnetic field tensor, or its dual conjugate, is a simple bivector in all reference frames (the second of two necessary and sufficient conditions is four- dimensionality of the manifold under consideration), thus $F=U\wedge V\textnormal{ or }\ast F=P\wedge Q,$ (6.1) $U$, $V$, $P$, and $Q$ being four-(co)vectors. In the first case, $I_{1}=2\left((U\cdot U)(V\cdot V)-(U\cdot V)^{2}\right)$ (6.2) obviously is negative if one of these vectors is timelike (say, $U$) and another, spacelike ($V$), thus $F$ will pertain to the pure electric type (or, similarly, for $\ast F$, to the pure magnetic type; see also an alternative case considered in subsection 11.1 when vector $U=R$ is null). Normalizing timelike $U$ to unity (the extra coefficient may be included in $V$), we can take the normalized $U$ as a new monad in the choice of reference frame and immediately see that in this frame the magnetic vector automatically vanishes. It remains only to show that our supposition (timelike $U$ and spacelike $V$) is sufficiently general; this can be easily proven using the substitution $V\Longrightarrow V+aU$ which does not change $F$. Similarly we treat the problem of eliminating the electric field in the pure magnetic case using $\ast F$ in (6.1). ### 6.2 Pure null type fields and the Doppler effect Pure null type fields have both invariants equal to zero, but the very fields remain non-trivial in any non-degenerate system of coordinates as well as in any realistic reference frame (here, in the sense of E and B, simultaneously), although these three-vectors do transform under changes of reference frames, and their components transform under transformations of coordinates. The monad $\tau$ under these transformation should remain always timelike, and the Jacobian of the transformation of coordinates has to be non-zero and non-$\infty$. As an example we consider in this subsection a special-relativistic plane electromagnetic wave in a vacuum ($k=\omega$) written in Cartesian coordinates, $\mathbf{E}=\\{0,E\cos[\omega(x-t)],0\\},~{}~{}\mathbf{B}=\\{0,0,E\cos[\omega(x-t)]\\},$ (6.3) and apply to it the Lorentz transformation $t,\mathbf{r}\Rightarrow t^{\prime},\mathbf{r}^{\prime}$ with the three-velocity $\pm v$ in the positive/negative direction of the $x$ axis using the well-known change of E and B under this transformation. The resulting electromagnetic field then is $\mathbf{E}^{\prime}=\\{0,E^{\prime}\cos[\omega^{\prime}(x^{\prime}-t^{\prime})],0\\},~{}~{}\mathbf{B}^{\prime}=\\{0,0,E^{\prime}\cos[\omega^{\prime}(x^{\prime}-t^{\prime})]\\}$ (6.4) where $E^{\prime}=\sqrt{\frac{1\mp v}{1\pm v}}E,~{}~{}\omega^{\prime}=\sqrt{\frac{1\mp v}{1\pm v}}\omega.$ (6.5) It is clear that the expression of $\omega^{\prime}$ in (6.5) describes the longitudinal Doppler effect while $E^{\prime}$ gives the accompanying change of the wave intensity. Since the latter is an integral part of the longitudinal Doppler effect, we consider the complete expression (6.5) as its natural generalization; the description of transversal Doppler effect has to be generalized in the similar way. It seems that this generalization of the Doppler effect is not encountered in physics textbooks. Nevertheless, it is generally used as an important hint in the interpretation of the well-known astrophysical phenomenon of ultrarelativistic particles’ jet pairs emitted by cores of some galaxies (the jet moving away from the observer not only has lower frequency, but also correspondingly lower intensity, thus this jet sometimes escapes to be observed). There is also a static ($\omega=0$) particular case of pure null electromagnetic fields involving mutually orthogonal constant vectors E and B (let us call it “Cartesian case” whose cylindrically symmetric analogue is used in some experiments involving electromagnetic fields with non-zero angular momentum without a genuine rotation). Such Cartesian pure null fields manifest only intensity part of the Doppler effect since in this case $\omega=0=\omega^{\prime}$ in (6.5). Thus the pure null electromagnetic fields can be adjusted to any non-zero and non-$\infty$ values of their intensity and frequency (only the relation of frequency to intensity remaining constant, and if the frequency had not been equal to zero from the very beginning). A complete transformation away of initially non-trivial pure null fields is however impossible in any non-degenerate frame, representing only asymptotic and not real possibility. We considered above the case of a plane-polarized wave, but similar approach works in the circular polarization case as well. ## 7 Duality rotation and electromagnetic fields with the same spacetime geometry Let us introduce a new electromagnetic field tensor (2-form) $\mathcal{F}=(k+l\ast)F,~{}~{}\ast\mathcal{F}=(k\ast-l)F$ (7.1) ($\ast$ is the Hodge star), where $F$ is the electromagnetic field tensor belonging to some given exact self-consistent Einstein–Maxwell solution, $k$ and $l$ being some scalar functions to be further determined. We now set the condition that the new field $\mathcal{F}$ has to produce the same energy- momentum tensor which follows from the old field $F$. Since geometry is well determined by the energy-momentum tensor, from the Bianchi identities it then follows that the standard general relativistic Maxwell equations for both fields, old and new, will be equally satisfied if the old field has no electromagnetic sources, or the sources are localized at the singularity of the old and new fields where the standard classical theory is not applicable. The calligraphic letters will be used for all concomitants of the new electromagnetic field. Thus $\mathcal{F}_{\mu\nu}=kF_{\mu\nu}+lF\\!\\!\stackrel{{\scriptstyle\textnormal{\small$\ast$}}}{{\textnormal{\scriptsize$\mu\nu$}}},~{}~{}\ast\mathcal{F}=kF\\!\\!\stackrel{{\scriptstyle\textnormal{\small$\ast$}}}{{\textnormal{\scriptsize$\mu\nu$}}}-lF_{\mu\nu},$ (7.2) $\mathcal{E}=k\mathbf{E}-l\mathbf{B},~{}~{}~{}\mathcal{B}=k\mathbf{B}+l\mathbf{E},$ (7.3) $\mathcal{I}_{1}=\mathcal{F}_{\mu\nu}\mathcal{F}^{\mu\nu}=(k^{2}-l^{2})I_{1}+2klI_{2},~{}~{}\mathcal{I}_{2}=\mathcal{F}\\!\\!\stackrel{{\scriptstyle\textnormal{\small$\ast$}}}{{\textnormal{\scriptsize$\mu\nu$}}}\\!\mathcal{F}^{\mu\nu}=(k^{2}-l^{2})I_{2}-2klI_{1}.$ A simple calculation yields $\mathcal{T}^{\mu}_{\nu}=\frac{k^{2}+l^{2}}{4\pi}\left(\frac{1}{4}I_{1}\delta^{\mu}_{\nu}-F^{\alpha\mu}F_{\alpha\nu}\right)=(k^{2}+l^{2})T^{\mu}_{\nu}$ (the terms with $I_{2}$ cancel automatically for arbitrary $I_{2}$). Hence the coincidence of geometries created by the two fields is guaranteed iff $k=\cos\alpha,~{}~{}l=\sin\alpha,$ (7.4) so that $\mathcal{I}_{1}=\cos 2\alpha\,I_{1}+\sin 2\alpha\,I_{2},~{}~{}\mathcal{I}_{2}=\cos 2\alpha\,I_{2}-\sin 2\alpha\,I_{1}.$ (7.5) Duality rotation, how the “transformation” (7.1) with (7.4) is now interpreted (see [8]), does not change the 4-geometry compatible with the new electromagnetic field. This geometry remains the same as that created by the old field (see also [16, 8, 23]). The “angle” (complexion) $\alpha$ of the duality rotation is, of course, an axial scalar function of coordinates which we shall now concretely determine. ## 8 Construction of new Einstein–Maxwell solutions via duality rotation First, let us see how restrictive is the duality rotation. The relations (7.5) lead to a general conclusion ${\mathcal{I}_{1}}^{2}+{\mathcal{I}_{2}}^{2}={I_{1}}^{2}+{I_{2}}^{2}$ (8.1) from where it follows that the pure null property is invariant under the duality rotation and it is impossible to obtain from any other type a pure null solution using this method. Together with the considerations of subsection 6.2, this means that pure null fields sharply differ from all other electromagnetic fields. Now, let us see if pure subtypes ($\mathcal{I}_{2}=0~{}\Rightarrow~{}{\mathcal{I}_{1}}^{2}={I_{1}}^{2}+{I_{2}}^{2}$) can be obtained from the impure fields ($I_{2}\neq 0$). The second relation in (7.5) then yields $\cot 2\alpha=I_{1}/I_{2}$ (8.2) which with the first relation gives $\mathcal{I}_{1}=\frac{I_{2}}{\sin 2\alpha}=\frac{I_{1}}{\cos 2\alpha}=\cos 2\alpha\frac{{I_{1}}^{2}+{I_{2}}^{2}}{I_{1}}=\sin 2\alpha\frac{{I_{1}}^{2}+{I_{2}}^{2}}{I_{2}}.$ (8.3) (In fact, we have to perform straightforward calculations for every concrete solution $F$ and see if $I_{2}$ would be sign-definitive or not in the desired region. Though the last possibility seems to be excluded by our initial supposition, it could be, naturally, softened: the duality rotation should reduce to the identity transformation at loci where $I_{2}$ becomes equal to zero.) From (8.2) and (8.3) we come to the following conclusions: if the new field has to be pure electric ($\mathcal{I}_{1}<0$), while $I_{1}<0$ and $I_{2}>0$, the “angle” $\alpha$ has to be such that $\sin 2\alpha<0$ and $\cos 2\alpha>0$; for $I_{1}<0$ and $I_{2}<0$, $\alpha$ has to give $\sin 2\alpha>0$ and $\cos 2\alpha>0$; for $I_{1}>0$ and $I_{2}>0$, there has to be $\sin 2\alpha<0$ and $\cos 2\alpha<0$; for $I_{1}>0$ and $I_{2}<0$, $\sin 2\alpha>0$ and $\cos 2\alpha<0$. Similarly, we determine the position of $\alpha$ for the pure magnetic new field. For the null (now impure) type of the new field ($\mathcal{I}_{1}=0~{}\Rightarrow~{}{\mathcal{I}_{2}}^{2}={I_{1}}^{2}+{I_{2}}^{2}$) we have to use the first relation in (7.5) yielding $\tan 2\alpha=-I_{1}/I_{2}.$ (8.4) The second relation gives $\mathcal{I}_{2}=\frac{I_{2}}{\cos 2\alpha}=\frac{I_{1}}{\sin 2\alpha},~{}~{}etc;$ (8.5) the procedure of determination of the position of $\alpha$ is the same as above, here only $\mathcal{I}_{2}\neq 0$, and both signs of $\mathcal{I}_{2}$ are equally admissible. It is clear that we can perform the inverse duality rotation in all these cases (in particular, coming from the impure null to pure electric or magnetic type fields). Thus the pure and impure electric and magnetic types form together with the impure null type a mutually “transformable” (via duality rotation) group of electromagnetic fields disconnected from the pure null type. ## 9 Impure electromagnetic fields: parallelizing of E and B In this section we again use the classification of electromagnetic fields in two senses: in the proper one, i.e. with respect to $F$, and, simultaneously, in the sense of the new field $\mathcal{F}$ introduced in (7.1), but we now look for information received from $\mathcal{F}$ about the old field $F$. It is already clear that when $F$ is impure (electric, magnetic, or null), the field $\mathcal{F}$ can be chosen as pure (electric or magnetic in everyone of these cases). An interesting feature here is that the reference frame in which only one field (electric or magnetic in the sense of $\mathcal{F}$) survives, is precisely that in which E and B following from $F$ are mutually parallel, and the parallelization procedure becomes completely reduced just to determination of this canonical frame (say, $\tau^{\prime}$). Thus, while the relations $\mathcal{E}\bullet\mathcal{B}\equiv\mathcal{E}^{\prime}\bullet\mathcal{B}^{\prime}=0$ (9.1) are frame-invariant (the field $\mathcal{F}$ is chosen as belonging to the pure subtype), the property $\mathbf{E}^{\prime}\parallel\mathbf{B}^{\prime}$ is realized only in the canonical frame where either $\mathcal{E}^{\prime}$ or $\mathcal{B}^{\prime}$ vanishes. The field $\mathcal{F}$ may be considered as a merely auxiliary one (essentially, in special relativity where its rôle in generation of gravitational field is neglected), so that the parallelization procedure then may be managed even without the use of the strict duality rotation with the “angle” $\alpha$ and the relation (8.1), but when we simply take, e.g., $\mathcal{F}=(1+k\ast)F.$ (9.2) The calculations following from this ansatz are simple, but somewhat cumbersome, and we omit them, especially since they will not be used in this paper. ## 10 Examples in general relativity In this section we consider two particular electromagnetic fields self- consistently sharing one and the same four-dimensional geometry: in the first subsection, the standard Kerr–Newman (KN) rotating charged black hole, and in the subsection 10.2, its generalizations to the black holes created by specific mixtures of electric charge and magnetic monopole distributions rotating as the KN singular ring. The first example corresponds to an impure electromagnetic field, while the next three ones belong to the pure electric, pure magnetic and impure null types, thus representing new black-hole exact solutions of Einstein–Maxwell equations. The first two new solutions admit reference frames in which there is no magnetic or no electric fields in the whole spacetime. ### 10.1 The Kerr–Newman solution The KN metric tensor is taken in the Boyer–Lindquist (BL) coordinates as $\begin{array}[]{l}ds^{2}=\displaystyle\frac{\Delta}{\rho^{2}}\left(dt-a\sin^{2}\vartheta\,d\varphi\right)^{2}-\frac{\rho^{2}}{\Delta}dr^{2}-\rho^{2}d\vartheta^{2}\\\ \phantom{ds^{2}}-\displaystyle\frac{\sin^{2}\vartheta}{\rho^{2}}\left[(r^{2}+a^{2})d\varphi- adt\right]^{2}\end{array}$ (10.1) where $\rho^{2}=r^{2}+a^{2}\cos^{2}\vartheta$, $\Delta=r^{2}-2Mr+Q^{2}+a^{2}$. Thus the orthonormal 1-form basis outside the singularity $\rho=0$ reads $\left.\begin{array}[]{ll}\theta^{(0)}=\displaystyle\frac{\sqrt{\Delta}}{\rho}\left(dt-a\sin^{2}\vartheta\,d\varphi\right),&\theta^{(1)}=\displaystyle\frac{\rho}{\sqrt{\Delta}}dr,\\\ \theta^{(2)}=\rho d\vartheta,&\theta^{(3)}=\displaystyle\frac{\sin\vartheta}{\rho}\left[(r^{2}+a^{2})d\varphi- adt\right],\end{array}\right\\}$ (10.2) so that $d\varphi=\frac{a}{\rho\sqrt{\Delta}}\theta^{(0)}+\frac{1}{\rho\sin\vartheta}\theta^{(3)}$. Further, the electromagnetic 1-form four-potential (cf. the usual Coulomb potential) and the field tensor (2-form) are $\left.\begin{array}[]{l}A\equiv A_{(\alpha)}\theta^{(\alpha)}=\displaystyle\frac{Qr}{\rho\sqrt{\Delta}}\theta^{(0)}~{}~{}\textnormal{ and}\\\ {}\\\ F\equiv\frac{1}{2}F_{(\alpha)(\beta)}\theta^{(\alpha)}\wedge\theta^{(\beta)}=dA\\\ \phantom{F}=\displaystyle\frac{Q}{\rho^{4}}\left[(r^{2}-a^{2}\cos^{2}\vartheta)\theta^{(0)}\wedge\theta^{(1)}-2ar\cos\vartheta\,\theta^{(2)}\wedge\theta^{(3)}\right],\end{array}\right\\}$ (10.3) respectively (see for some details [2, 3, 4, 5, 6, 17, 20, 22]). Since $4a^{2}r^{2}\cos^{2}\vartheta+\left(r^{2}-a^{2}\cos^{2}\vartheta\right)^{2}=\rho^{4}$ (10.4) (this confirms the presence in $F$ of the factor $r^{2}-a^{2}\cos^{2}\vartheta$ which first could seem to be somewhat unnatural), we can now introduce an “angle” $\beta$ as $\sin\beta=\frac{2ar\cos\vartheta}{\rho^{2}},~{}~{}\cos\beta=\frac{r^{2}-a^{2}\cos^{2}\vartheta}{\rho^{2}}.$ (10.5) Then the electromagnetic field $F$ reads $F=\frac{Q}{\rho^{2}}(\cos\beta+\sin\beta\ast)\left(\theta^{(0)}\wedge\theta^{(1)}\right),$ (10.6) obviously involving a duality rotation, and the electromagnetic invariants read $I_{1}=-\frac{2Q^{2}}{\rho^{4}}\cos 2\beta,~{}~{}I_{2}=\frac{2Q^{2}}{\rho^{4}}\sin 2\beta,$ (10.7) so that the construction invariant under duality rotation (8.1) in the KN solution case is (cf. the Coulomb and Reissner–Nordström fields where, of course, $a=0$) ${I_{1}}^{2}+{I_{2}}^{2}=\frac{4Q^{4}}{\rho^{8}}.$ (10.8) From (10.7) and (10.5) we immediately find that on the “plane” $\cos\vartheta=0$ and the “sphere” $r=0$ (it is well known that the “negative region of space” with $r<0$ makes a certain sense in this spacetime) the invariants take values $I_{1}=-\frac{2Q^{2}}{r^{4}},~{}~{}I_{2}=0~{}\textnormal{ and }~{}I_{1}=-\frac{2Q^{2}}{a^{4}\cos^{4}\vartheta},~{}~{}I_{2}=0,$ respectively, thus the field $F$ belongs there to the pure electric type (in the BL frame, the magnetic field is already eliminated in this region). The intersection of these 2-surfaces is the well-known singular rotating Kerr ring where $I_{2}$ vanishes when we approach to the ring from these mutually orthogonal directions. This would be in conformity with the usual interpretation of the ring as rotating with the velocity of light if that $I_{1}$ also tended there to zero, though this is not quite the case. The vanishing of $I_{1}$ occurs only in the limits in four directions along which $r=\epsilon(3\pm 2\sqrt{2})a^{2}\cos^{2}\vartheta$ where $\epsilon=\pm 1$ (without admission to simultaneously take only both similar signs in the whole formula). Further, when $r^{2}=a^{2}\cos^{2}\vartheta$, we have the pure magnetic type field (with already eliminated electric field in the BL frame) with $I_{1}=\frac{3Q^{2}}{4a^{4}\cos^{4}\vartheta}=\frac{3Q^{2}}{4r^{4}}$, thus, if we come to the ring from the corresponding directions, the field will be purely magnetic. The electromagnetic field around the Kerr ring in KN solution is in fact very diverse, like a patchwork quilt. To find how behaves the propagation velocity v of this field, we have to calculate its energy density and Poynting vector, but let us begin with E and B in the BL frame $\tau=\theta^{(0)}$, see (10.2). Already before any calculations, only looking at the definitions (3.4) and (3.5), one understands that these two vectors and $\theta^{(1)}$ are everywhere collinear, so that the Poynting vector identically vanishes, as well as v does (this is natural, since both metric coefficients and the field tensor components do not depend on $t$ and $\varphi$, and we already are in the KN-field’s co-moving frame). The pure subtype can be realized (if we suppress duality rotations) only due to local vanishing of E or B in the BL frame. In this co-moving frame the electromagnetic energy density is everywhere equal to $T_{\mu}^{\nu}\tau^{\mu}\tau_{\nu}=\frac{1}{8\pi}(\mathbf{E}^{2}+\mathbf{B}^{2})=\frac{1}{16\pi}\sqrt{{I_{1}}^{2}+{I_{2}}^{2}}=\frac{Q^{2}}{8\pi\rho^{4}},$ (10.9) see (5.3). We have found here that the KN solution is “anisotropic” and “inhomogeneous” in the sense of distribution of electric and magnetic types of the $F$ field, although three-vector fields E and B are always collinear in the spacetime of KN black hole in the BL frame, but there are surfaces on which either magnetic or electric field vanishes. In order to better understand if this structure of electromagnetic field belonging to the KN solution is inevitable (however, see also [3]), we shall further apply to the KN solution the method developed in sections 7 and 8 (as a broadening and modernization of the Rainich–Misner–Wheeler duality rotation approach). We shall find that it is easy to radically modify the KN solution obtaining rotating black holes with electromagnetic fields of everywhere pure electric or pure magnetic, or impure null type. ### 10.2 “New” rotating black hole solutions with pure electric, pure magnetic, and impure null fields $F$ in KN geometry Here we apply duality rotation $\mathcal{F}=(\cos\alpha+\sin\alpha\ast)F,~{}\textnormal{or}~{}\ast\mathcal{F}=(\cos\alpha\ast-\sin\alpha)F$ (10.10) [the combination of (7.1) and (7.4)] to the KN electromagnetic field (10.3). Since our aim is to construct a pure subtype field ($\mathcal{I}_{2}=0$), we already have from (8.1) ${\mathcal{I}_{1}}^{2}={I_{1}}^{2}+{I_{2}}^{2}$ while the sign of $\mathcal{I}_{1}$ is determined by the choice of position of $\alpha$ in the angle diagram and (8.2). Taking the expression (10.6) and using the obvious algebra of duality rotation, we rewrite the expression (10.10) as $\mathcal{F}=\frac{Q}{\rho^{2}}\left[\cos(\alpha+\beta)+\sin(\alpha+\beta)\ast\right]\left(\theta^{(0)}\wedge\theta^{(1)}\right).$ (10.11) Now, putting $\alpha+\beta=0$, we immediately come to the pure electric field $\mathcal{F}_{\mathrm{el}}$ $\mathcal{F}_{\mathrm{el}}=\frac{Q}{\rho^{2}}\theta^{(0)}\wedge\theta^{(1)}$ (10.12) generating the same KN geometry which everybody associates with the KN “patchwork” electromagnetic field $F$ (10.3). The pure magnetic field $\mathcal{F}_{\mathrm{mag}}=\frac{Q}{\rho^{2}}\theta^{(2)}\wedge\theta^{(3)}$ (10.13) is similarly obtained with the use of $\alpha+\beta=-\pi/2$. Finally, we have to add the third new case of KN-like black holes, those with null type electromagnetic field when in (10.11) $\alpha+\beta=\pi/4+n\pi/2$ is taken (naturally, this field now belongs to the impure subtype). With $n=0$, it reads $\mathcal{F}_{\mathrm{null}}=\frac{Q}{\sqrt{2}\rho^{2}}\left(\theta^{(0)}\wedge\theta^{(1)}-\theta^{(2)}\wedge\theta^{(3)}\right).$ (10.14) The (contracted) Bianchi identities guarantee satisfaction of Maxwell’s equations outside the ring singularity for the new field $\mathcal{F}$ in this geometry, while the presence of magnetic monopole distribution existing here only on the singular Kerr ring should not create any problem since at the singularity the classical laws of physics obviously fail to work. One may say that the magnetic monopole distribution (as well as that of the electric charge) can, as this is shown above, exactly compensate the magnetic (electric) field created by a rotating charge (rotating monopole) distribution, but this is not precisely the case. In fact, we encounter in this situation a more complicated superposition of dynamical and kinematic effects, since the ($\tau=\theta^{(0)}$)-reference frame is not an inertial one: it involves both acceleration and rotation (the latter is present due to $\theta^{(0)}\wedge d\theta^{(0)}\neq 0$, see the definitions in [11]). In the same ref. 11 (chapter 4, pp. 86, 87, 90, and 91) it is shown that in the classical Maxwell equations and in laws of motion of electric charges, both written in non-inertial reference frames, there appear kinematic terms of the monopole nature. In equations of motion they bear the name of kinematic forces (forces of inertia), thus in the field equations let us speak of kinematic sources. While dynamical force and source are originated by the same interaction term in the action integral (only the variational procedure is performed with respect to particle’s world line and to field’s potential, correspondingly), their kinematic counterparts automatically appear in the respective dynamical equations written (and experimentally investigable) in non-inertial frames. It is interesting that kinematic and dynamical counterparts of forces, as well as of sources, have rather similar structure, despite their different origin, thus making them recognizable. ## 11 Examples in special relativity ### 11.1 Liénard–Wiechert’s field: the pure electric type The Liénard–Wiechert (LW) field is special relativistic electromagnetic field generated by an arbitrarily moving electric charge $Q$ (we restrict our consideration to an arbitrary timelike world line of the charge). See the details of deduction of this field in [10] where we used the future light cone (the case of retarded field, precisely like in the present paper); an arbitrary mixture of retarded and advanced fields can be found in [18]. Thus the retarded point on the charge world line $x^{\prime\mu}$ is connected with the four-dimensional point $x^{\mu}$ (where the field is determined) by the null vector $R^{\mu}=x^{\mu}-x^{\prime\mu}$ (we choose for simplicity the Cartesian coordinates in this special relativistic treatment). The four- potential then reads $A^{\mu}=\frac{Qu^{\prime\mu}}{D}$ (11.1) where $u^{\prime\mu}=dx^{\prime\mu}/ds^{\prime}$ ($u^{\prime}\cdot u^{\prime}=1$) is the retarded four-velocity of the charge and $D=u^{\prime}\cdot R\equiv\sqrt{-\mathbf{D}^{\mu}\mathbf{D}_{\mu}}$, while $\mathbf{D}^{\mu}=R^{\nu}b^{\mu}_{\nu}$, $b=g-u^{\prime}\otimes u^{\prime}$ simultaneously being the projector on local retarded three-dimensional subspace orthogonal to $u^{\prime}$ and the spatial three-metric on this subspace (with the signature $0,-,-,-$). The retarded four-acceleration of the charge is $a^{\prime\mu}=du^{\prime\mu}/ds^{\prime}$ (naturally, $a^{\prime}\cdot u^{\prime}\equiv 0$). A simple calculation yields the LW field tensor $F=R\wedge V,~{}~{}V=\frac{Q}{D^{2}}\left(a^{\prime}+u^{\prime}\frac{1-a^{\prime}\cdot R}{D}\right)$ (11.2) (the second field invariant $I_{2}$ automatically vanishes). The first invariant is $I_{1}=-\frac{2Q^{2}}{D^{4}}<0$ (11.3) (remarkably, its structure is exactly Coulombian). Thus the LW field pertains to the pure electric type everywhere outside the point charge’s world line. Combining $R$ with $V$ in (11.2), one can change the null vector $R$ to a timelike one, $U$, and thus reduce the problem to that discussed in subsection 6.1. However it is much simpler to (algebraically) regauge the vector $V~{}\rightarrow~{}W=V+\frac{Q}{D^{2}}lR$ (the fractional coefficient is put only for convenience, and $l$ is a scalar function still to be determined); this does neither change the field tensor, $F=R\wedge W$, nor produce any $l^{2}$-term in further calculations. Applying now the 1-form definition of the magnetic vector in a $\tau$-frame (3.5) and taking the monad as $\tau=NW$ where the scalar normalization factor is $N=(W\cdot W)^{-1/2}$, we come to $\mathbf{B}=0$ in this frame. The problem is thus reduced to a proper choice of $l$ such that $W$ will be a suitable real timelike vector. We see that $W=\frac{Q}{D^{2}}\left(a^{\prime}+\frac{1-a^{\prime}\\!\cdot\\!R}{D}\,u^{\prime}+lR\right).$ (11.4) Then its square takes an unexpectedly simple form $W\\!\cdot\\!W=\left(\frac{Q}{D^{2}}\right)^{2}\left[a^{\prime}\\!\cdot\\!a^{\prime}+\frac{(1-a^{\prime}\\!\cdot\\!R)^{2}}{D^{2}}+2l\right].$ (11.5) In fact, $l$ still remains arbitrary (this means that there is a continuum of such different co-moving frames). Let it be $l=\frac{1}{2}\left[\frac{1}{D^{2}}-a^{\prime}\\!\cdot\\!a^{\prime}-\frac{(1-a^{\prime}\\!\cdot\\!R)^{2}}{D^{2}}\right]$ (11.6) (the first term in the square brackets, $1/D^{2}$, got its denominator to fit the dimensional considerations). Finally, $W\cdot W=\left(\frac{Q}{D^{3}}\right)^{2}>0$ and $\tau=Da^{\prime}+\left(1-a^{\prime}\\!\cdot\\!R\right)u^{\prime}+\frac{1}{2D}\left[1-D^{2}a^{\prime}\\!\cdot\\!a^{\prime}-\left(1-a^{\prime}\\!\cdot\\!R\right)^{2}\right]R$ (11.7) (it is clear that $\tau\cdot\tau=+1$). By its definition, the monad $\tau$ describes the reference frame co-moving with the LW electromagnetic field: in this frame the Poynting vector of the field vanishes, the electromagnetic energy flux ceases to exist due to the absence of magnetic part $\overset{\textnormal{\tiny$\tau$}}{\mathbf{B}}$ of the field in this frame (applicable at any finite distance $D$, not asymptotically), and $F$ can be rewritten as $F=\frac{Q}{D^{3}}R\wedge\tau$. The expression (3.4) now yields $\overset{\textnormal{\tiny$\tau$}}{\mathbf{E}}=\ast(\tau\wedge\ast F)=\frac{Q}{D^{3}}\ast[\tau\wedge\ast(R\wedge\tau)]=\frac{Q}{D^{2}}\overset{\textnormal{\tiny$\tau$}}{\mathbf{n}}$ (11.8) which is, up to an understandable reinterpretation of notations, exactly the form known as the Coulomb field vector. Here the unitary vector $\overset{\textnormal{\tiny$\tau$}}{\mathbf{n}}=\overset{\textnormal{\tiny$\tau$}}{\mathbf{D}}/D$ is normal to the $\tau$-congruence, while $R\cdot u^{\prime}=\overset{\textnormal{\;\tiny$u^{\prime}$}}{D}=D=\overset{\textnormal{\;\tiny$\tau$}}{D}=R\cdot\tau$ and $\overset{\textnormal{\tiny$\tau$}}{\mathbf{D}}^{\mu}=b^{\mu}_{\nu}R^{\nu}$ with $b^{\mu}_{\nu}=\delta^{\mu}_{\nu}-\tau^{\mu}\tau_{\nu}$, hence in the frame co-moving with the LW field (the reader may choose other co-moving reference frames taking different $l$s, but our choice seems to be one of the simplest ones) $\overset{\textnormal{\tiny$\tau$}}{\mathbf{D}}=-D^{2}a^{\prime}-D\left(1-a^{\prime}\\!\cdot\\!R\right)u^{\prime}+\frac{1}{2}\left[1+D^{2}a^{\prime}\\!\cdot\\!a^{\prime}+\left(1-a^{\prime}\\!\cdot\\!R\right)^{2}\right]R,$ (11.9) so that this $\overset{\textnormal{\tiny$\tau$}}{\mathbf{D}}\neq\overset{\textnormal{\tiny$u^{\prime}$}}{\mathbf{D}}$ in fact given after the formula (11.1) as $\mathbf{D}$ and pertaining to another frame (co-moving with the retarded charge, not with its field, see for details [10]). The situation discovered in this subsection can be formulated in a short and exact form as existence in all spacetime outside the world line of the charge generating LW’s field, of a reference frame co-moving with this field, i.e., a frame in which the Poynting vector vanishes in all this region (with the exception of the future null infinity which can be described only asymptotically, using more topological111It was a gibe of the fate with respect to the authors who deliberately disregarded topology and nevertheless claimed that the LW field contains electromagnetic radiation, though in this field the Poynting vector can be easily transformed away by means of a proper choice of the reference frame in every finite region around the charge’s world line. About the attitude of L.D. Landau and E.M. Lifshitz toward topology see [21], pp. 470-471, — and see also [7], p. 173 ff. However, Misner and Wheeler did take topology into account in their general-relativistic considerations [8, 23]. than geometrical methods), thus in this frame there is no flow of electromagnetic energy anywhere. Of course, this frame is in general a rotating one (see in [10] the expression (4.27) and appendix A), thus the three-dimensional space is non-holonom (it does not form a global — at least, finite — three-dimensional subspace of the four-dimensional world; at most, in the presence of rotation there exist only strictly local (infinitesimal) elements of such a subspace which do not merge into a finite hypersurface, like scales of a sick fish in aquarium. Note that this occurs here even in the special relativity, not only in general theory. Moreover, the presence of the frame’s rotation does not permit synchronization of clocks being at rest in such a frame. (In the same spacetime there always exist also an infinite number of non-rotating frames in which you are welcomed to perform a synchronization, but in any rotating frame this very procedure is strictly forbidden. It is curious that while we live all our lives in our terrestrial rotating frame, its rotation remains sufficiently slow not to condition us to this non-holonom psychology.) Since everywhere outside the LW source (the pointlike charge world line on which the field singularity occurs) there is no magnetic field in this frame, and any redistribution of electromagnetic field cannot take place there, the LW field does not propagate in this frame, it only can be compressed or rarefied remaining at rest with respect to the frame in its contraction or expansion, similar to effects known in relativistic cosmology. This could seem to be in contradiction with the traditional decomposition of LW field into the near (induction) and distant (radiation) zones. The reason for this “contradiction” should be seen in the fact that, although the very Maxwell equations are linear, the physical characteristics of electromagnetic fields such as their energy density, Poynting vector (describing, up to a constant factor, either the energy flow density or linear momentum density of the field), and stress, are quadratic (or bilinear) in the field tensor sense. Thus we have not to overlook the “interaction” terms between the induction and radiation counterparts in these characteristics. Note that the elimination of the magnetic part of LW’s field is directly related to its pure electric type,— consequently, to the quadratic (bilinear) invariants of this field. Therefore the “contradiction” exists only in a wrong customary application of the linearity concept to the strictly nonlinear characteristics even of the electrovacuum electromagnetic fields. In a certain sense, there should be a way to reconcile this contradiction considering the asymptotic behaviour of the field; in any case, this has to correspond to merely technical details of the problem. Similarly, in his ironic paper [19], Synge with his great wit criticized the existing style of introduction of these same characteristics in the most widely used textbooks on field theory. Though his criticism was there somewhat superficial, we find Synge’s paper quite provocative in more profound determination of the quotidian concepts in our theory using their physical sense. ### 11.2 Propagation of a plane electromagnetic wave on the background of homogeneous magnetic field in a vacuum Finally, let us consider a simple, but not yet discussed in literature problem of electromagnetic waves’ propagation in a time-independent sourceless Maxwell field in a vacuum. For simplicity, we take the same wave as in subsection 6.2, and the additional Maxwell field is chosen merely as a magnetic one in the direction of propagation of the wave, with the constant three-vector B. Obviously, this superposition is an exact solution of Maxwell’s equations, and there cannot be any real interaction between these two fields since the equations are linear. We have however already noted that the velocity of propagation of electromagnetic field is non-linear in terms of this field’s tensor $F$, so that there should, naturally, exist an observable physical effect in the case of a superposition of such free Maxwell’s fields. There is, of course, an effect which was already considered and observed in the early history of optics, that of the standing electromagnetic waves, but nobody still worried about the seemingly absurd problem formulated above. Thus we take in Cartesian coordinates $t,x,y,z$ the superposition of the fields (6.3) and $\mathbf{B}_{\mathrm{H}}=Hdx$, i.e. $\mathbf{E}=E\cos[\omega(x-t)]dy,~{}~{}\mathbf{B}=Hdx+E\cos[\omega(x-t)]dz.$ (11.10) Obviously, $I_{2}=0$ due to the orthogonality of E and B, and the first invariant is $I_{1}=2H^{2}>0$: this is the pure magnetic type field. The result of superposition (11.10) is in fact a specific not precisely monochromatic wave whose behaviour can be best understood in the reference frame co-moving with it, and one can find such a frame using the pure-magnetic property of this wave’s field, see subsection 6.1. First, we write the field $\ast F$ [through (11.10) in the initial frame $\tau_{\mathrm{in}}=dt$] as a simple bivector: $\displaystyle\ast F$ $\displaystyle=\ast(\mathbf{E}\wedge dt)-\mathbf{B}\wedge dt$ $\displaystyle=-\left(E\cos[\omega(x-t)]dx\wedge dz+H\phantom{{}^{Z}}\\!\\!\\!dx\wedge dt+E\cos[\omega(x-t)]dz\wedge dt\right)$ $\displaystyle=-\left(Hdx+E\phantom{{}^{Z}}\\!\\!\\!\cos[\omega(x-t)]dz\right)\wedge(dt- dx)=-P\wedge Q.$ (11.11) If to $P$ we add $lQ$ ($l$ being an arbitrary function) and use this sum $P^{\prime}$ instead of the former $P$, $\ast F$ does not change. It is obvious that $P\cdot P<0$, but $P^{\prime}\cdot P^{\prime}=2lH-H^{2}-E^{2}\cos^{2}[\omega(t-x)]$. Thus if we choose $l=H+\frac{E^{2}}{2H}\cos^{2}[\omega(t-x)]$, the vector $P^{\prime}$ will be timelike, $P^{\prime}\cdot P^{\prime}=H^{2}>0$, and we can take $P^{\prime}/H$ as a properly normalized monad, $\tau=\left(1+\frac{E^{2}}{H^{2}}\cos^{2}[\omega(t-x)]\right)(dt- dx)+dx+\frac{E}{H}\cos[\omega(t-x)]dz.$ (11.12) Now the dually conjugated field tensor reads $\ast F=-H\tau\wedge(dt-dx)$, thus in the frame $\tau$ the electric field (3.4) vanishes, and this is the field’s co-moving frame. In all these calculations one has to remember that when only one (here, magnetic) field survives after the reference frame is transformed, there are other possible transformations which do not change this situation (in fact, all those which involve an additional motion in the direction of this field, even when this motion occurs to be with a non- constant magnitude of the three-velocity described by strictly local Lorentz transformations working in non-inertial frames). Thus there appears a continuum of such one-field frames (cf. [7], but working in general as well as in special relativity), and the search for more elegant ones depends on the individual taste of the researcher. Let us now calculate the three-velocity of the frame $\tau$ from the viewpoint of $\tau_{\mathrm{in}}$ using our general definition (3.8) and substitute the result into the left-hand side of (5.1), then putting into the right-hand side the expressions of E and B from (11.10) in the frame $\tau_{\mathrm{in}}$ to check if the Landau–Lifshitz definition (5.1) really works. Obviously, this way will not represent a vicious circle since these parts of (5.1) were initially deduced in [7] from a very different standpoint than ours (moreover, in this way the left-hand side of (5.2) will be automatically checked: both definitions of v cannot simultaneously work well). First, we rewrite (3.8) in these notations for frames and find $\mathbf{v}\,(\perp\tau_{\mathrm{in}})$: $\tau=(\tau\cdot\tau_{\mathrm{in}})(\tau_{\mathrm{in}}+\mathbf{v})~{}\Rightarrow~{}\mathbf{v}=\frac{\frac{E}{H}\cos[\omega(t-x)]}{1+\frac{E^{2}}{H^{2}}\cos^{2}[\omega(t-x)]}\left(dz-\frac{E}{H}\cos[\omega(t-x)]dx\right).$ This means that $\frac{|\mathbf{v}|}{1+\mathbf{v}^{2}}=\frac{\frac{E}{H}\cos[\omega(t-x)]\sqrt{1+\frac{E^{2}}{H^{2}}\cos^{2}[\omega(t-x)]}}{1+2\frac{E^{2}}{H^{2}}\cos^{2}[\omega(t-x)]}.$ Precisely the same is the result of calculating $\frac{|\mathbf{E}\times\mathbf{B}|}{\mathbf{E}^{2}+\mathbf{B}^{2}}$ — Landau and Lifshitz’s definition wins. (Pauli’s definition (5.2) cannot contain on its left-hand side the construction $1+2\frac{E^{2}}{H^{2}}\cos^{2}[\omega(t-x)]$ which inevitably appears on the right-hand side $2\frac{|\mathbf{E}\times\mathbf{B}|}{\mathbf{E}^{2}+\mathbf{B}^{2}}$, like in the Landau–Lifshitz case.) The mean value of $|\mathbf{v}|$ is simply $\frac{2}{\pi}\arcsin\frac{E}{\sqrt{E^{2}+H^{2}}}$. When $H\rightarrow 0$, the mean propagation velocity approaches that of light, while if $E\ll H$, the mean velocity can become as low as one wishes: to this end, it is necessary to use as strong magnetic field $H$ as possible and/or choose a low-intensity wave in the superposition. ## 12 Concluding remarks The results obtained in this paper are based on three simple observations: that the physical classification of electromagnetic fields should be formulated using the properties of only two well known invariants of these fields, the complete description of reference frame is related only to the state of motion of a continuous multitude of test observers, and that the duality rotation (in the vein of Rainich–Misner–Wheeler, but in a more modern and general form) applied to a seed solution of Maxwell’s equations, yields a new solution in the same four-geometry which was generated by the seed solution via Einstein’s equations. We have proven that these suppositions really work together, and the duality rotation permits to construct qualitatively new solutions, belonging also to other desired types of electromagnetic fields in accordance with our classification. There is only one restriction separating the pure null type fields from those of other five types. The pure null type does not change under the duality rotation, becoming in fact the same solution of this pure type, though corresponding to another reference frame and displaying the Doppler effect in its generalized form also considered in this paper. As illustrations of application of our approach we discuss concrete examples of the Kerr–Newman (KN) solution and the Liénard–Wiechert (LW) field (to show the efficiency of our method also in special relativity). Moreover, we deduce three qualitatively new types of electromagnetic field creating the same four-geometry as the seed KN solution, thus describing other kinds of KN-like black holes. Studying the LW field, we come upon a new conclusion that the linearity of Maxwell’s equations does not automatically mean that different constituent parts of this field can be properly interpreted separately. Other characteristics of the field (such as the energy density and Poynting vector) have non-linear nature, thus a study of these characteristics constructed only of one or another parts of the LW solution, with omission of the combination (“interaction”) of these parts, means a disregard of important physical properties of the field, in particular, of its true propagation velocity. We have explicitly shown that this velocity of the complete LW solution is less than that of light, and we have given the physically full-fledged frame co-moving with LW field in which its Poynting vector exactly vanishes everywhere outside the world line of the source of this field (strangely, this fact was never noticed before). The last, but not least example is related to a simple superposition of two exact solutions of special-relativistic Maxwell’s equations, plane electromagnetic wave and homogeneous magnetic field in a vacuum. We show that this superposition, being itself an exact solution, always propagates with the velocity lesser than that of light, and we show that the elementary expression for this velocity is properly defined in [7], but not in [13]. (I must admit that at first I liked the definition given in Pauli’s book much more than Landau–Lifshitz’s one: see, e.g., [12].) Finally, may I express my hope (to a certain extent, against hope), that the given here examples should lead our community of physicists to a more profound consideration of the non-trivial concept of reference frame and to its better understanding as a more physical than purely mathematical subject and an important ingredient in the description of physical reality. To console those who cannot accept the representation of reference frames through monads and Cartan’s forms, I would add that they can take instead any system of coordinates whose $t$-coordinate lines coincide with those of the $\tau$-congruence (the choice of spatial coordinates does not matter). In such a system, there will be realized precisely the same picture, though mathematics will feel awkward, while reference frames will seem to be silenced. ## References * [1] N. Bohr and L. Rosenfeld (1933) Kgl. Danske Videnskab. Selskab., Mat.-fys. Medd. 12, 8. * [2] G.C. Debney, R.P. Kerr, and A. Schild, (1969) J. Math. Phys. 10, 1842. * [3] M.P. Hobson, G.P. Estathiou, and A.N. Lasenby (2006) General Relativity. An Introduction to Physicists (Cambridge, U.K.: CUP). * [4] G. ’t Hooft (1998) Introduction to General Relativity (Institute for Theoretical Physics, Utrecht University, Princetonplein 5, 3584 CC Utrecht, the Netherlands). * [5] R.P. Kerr (2008) Discovering the Kerr and Kerr–Schild metrics arXiv: 0706.1109v2 [gr-qc]. * [6] D. Kramer, H. Stephani, M. MacCallum, and E. Herlt (1980) Exact Solutions of Einstein’s Field Equations (Berlin: VEB Deutscher Verlag der Wissenschaften; Cambridge, U.K.: CUP). * [7] L.D. Landau and E.M. Lifshitz (2006) The Classical Theory of Fields, Fourth Revised English Edition (Amsterdam: Elsevier). * [8] C.W. Misner and J.A. Wheeler (1957) Ann. Phys. (USA) 2, 525. * [9] N.V. Mitskievich (Mizkjewitsch) (1958) Ann. Phys. (Leipzig) 1, 319. In German. (The Noether theorem in general relativity, see also [11].) * [10] N.V. Mitskievich (2005) Liénard–Wiechert solution revisited arXiv: physics/0506144. * [11] N.V. Mitskievich (2006) Relativistic Physics in Arbitrary Reference Frames (New York: Nova Science Publishers). See also a less detailed early book preprint: arXiv gr-qc/9606051 (without the Noether theorem [9]). * [12] N.V. Mitskievich (2006) Electromagnetic Phenomena 6, no. 2, 124. * [13] W. Pauli (1981) Theory of Relativity (New York: Dover). * [14] J.F. Plebański (1964) Acta Phys. Polon. 26, 963. * [15] F.M. Paiva, M.J. Rebouças, G.S. Hall, and M.A.H. MacCallum (1998) Class. Quantum Grav. 15, 1031. * [16] G.Y. Rainich (1925) Trans. Am. Math. Soc. 27, 106. * [17] H. Stephani, D. Kramer, M. MacCallum, C. Hoenselaers, and E. Herlt (2003) Exact Solutions of Einstein’s Field Equations, Second Edition (Cambridge, U.K.: CUP). * [18] J.L. Synge (1965) Relativity: The Special Theory (Amsterdam: North-Holland). * [19] J.L. Synge (1974) Hermathena (a Dublin University review) No. cxvii. * [20] S.A. Teukolsky (1973) Astrophys. J. 185, 635. * [21] K.S. Thorne (1994) Black Holes & Time Warps: Einstein’s Outrageous Legacy (New York: W.W. Norton and Co.). * [22] P.K. Townsend (without date) Black Holes. Lecture notes (DAMTP, Univ. of Cambridge, Silver St., Cambridge, U.K.). * [23] J.A. Wheeler (1962) Geometrodynamics (New York: Academic Press).

arxiv-papers

2024-09-04T02:48:53.987529

0802.3488

# Classification of Quiver Hopf Algebras and Pointed Hopf Algebras of Type One Shouchuan Zhang a,c, Hui-xiang Chen b, Yao-Zhong Zhang c, $a$. Department of Mathematics, Hunan University Changsha 410082, P.R. China $b$. Department of Mathematics, Yangzhou University Yangzhou 225002, P.R. China $c$. Department of Mathematics, University of Queensland Brisbane 4072, Australia. ###### Abstract The quiver Hopf algebras are classified by means of ramification system with irreducible representations. This leads to the classification of Nichols algebras over group algebras and pointed Hopf algebras of type one. 2000 Mathematics Subject Classification: 16W30, 16G10 keywords: Quiver, Hopf algebra, Hopf bimodule, Nichols algebras ## 0 Introduction Hopf algebras have important applications in mathematics and mathematical physics. Indeed, quasitriangular Hopf algebras gives rise to braided tensor categories through their categories of representations (see [ENO, EO]). Semisimple Hopf algebras and non-semisimple Hopf algebras are related to conformal field theories ( see [Ga]). The classification of Hopf algebras is the main object in research of Hopf algebras. So far, much important results have been obtained in the classification of finite dimensional pointed Hopf algebras (see [AS98, AS02, AS00, AS05, H1, H2, AZ07]). The classification of PM quiver Hopf algebras was completed by means of ramification system with characters (RSC in short ) in [ZZC]. The classification of RSC’s over symmetric group $\mathbb{S}_{n}$ with $n\not=6$ was obtained in [ZWW]. Irreducible Hopf bimodules over a finite group were described in [DPR]. They correspond to pairs $(\mathcal{O}_{s},\rho)$, where $\mathcal{O}_{s}$ is a conjugacy class containing $s$ in G and $\rho$ is an irreducible representation of the centralizer $G^{s}$ in G. This result has been reobtained and reproved several times; these data are used in a previous paper of Lusztig more or less in the same direction as in [DPR]. The relation between bi-one arrow Nichols algebras and $\mathfrak{B}({\mathcal{O}}_{s},\rho)$, which is the Nichols algebra corresponding to pairs $(\mathcal{O}_{s},\rho)$, was given in [ZZWC]. In this paper, quiver Hopf algebras, Nichols algebras over group algebras and pointed Hopf algebras of type one are classified by means of ramification system with irreducible representations (RSR in short ). As examples we classify ${\rm RSR}$’s over symmetric group $\mathbb{S}_{n}$ with $n\not=6.$ The quivers (see [ARS, CR02, CR97, ZZ, OZ] Ri84 ) and the tensor algebras of Hopf bimodule (see [N, W]) have widely been applied in representation theory, Hopf algebras and quantum groups. We use the quivers to describe Yetter- Drinfeld $kG$-modules, $kG$-Hopf bimodules, Nichols algebras in braided tensor category ${}^{kG}_{kG}{\mathcal{Y}D}$, pointed Hopf algebras of type one and quiver Hopf algebras in this paper. The main results in this paper are summarized in the following statement. ###### Theorem 1. (Classification theorem) (i) Every Yetter-Drinfeld $kG$-module, (ii) every $kG$-Hopf bimodule, (iii) every Nichols algebra in braided tensor category ${}^{kG}_{kG}{\mathcal{Y}D}$, (iv) every pointed Hopf algebra of type one with coradical $kG$ and (v) every quiver Hopf algebra over $G$ are, respectively, isomorphic to one of the following: $(kQ_{1}^{1},{\rm ad}(G,r,\overrightarrow{\rho},u))$, $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u),$ ${\mathcal{B}}(kQ_{1}^{1},{\rm ad}(G,r,\overrightarrow{\rho},u))$, $kG[kQ_{1}^{c},G,r,\overrightarrow{\rho},u]$, $kQ^{c}(G,r,\overrightarrow{\rho},u)$, $kQ^{s}(G,r,\overrightarrow{\rho},$ $u)$, $kG[kQ_{1}^{c},$ $G,r,$ $\overrightarrow{\rho},u],$ $kQ^{a}(G,r,\overrightarrow{\rho},u)$, $kQ^{sc}(G,r,\overrightarrow{\rho},u)$, $(kG)^{*}[kQ_{1}^{a},G,r,\overrightarrow{\rho},u]$. Furthermore, they are uniquely determined by ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ up to isomorphisms of ${\rm RSR}$’s. Indeed, (i) follows from Proposition 2.4(i); (ii) follows from Proposition 1.3; (iii) follows from Proposition 2.4(ii); (iv) follows from Theorem 3; (v) follows from definition of quiver Hopf algebras in [ZZC, the argument after Theorem 2]. For the uniqueness, (i) and (iii) follow from Theorem 4; (ii), (iv) and (v) follow from Theorem 2. Note that the isomorphism in (i) is about Yetter-Drinfeld $kG$-modules; the isomorphism in (ii) is about $kG$-Hopf bimodules; the isomorphism in (iii) is about graded braided Hopf algebras in ${}^{kG}_{kG}{\mathcal{Y}D}$; the isomorphism in (iv) and (v) are about graded Hopf algebras. Furthermore, for any fixed map $u_{0}$ from ${\mathcal{K}}(G)$ to $G$ with $u_{0}(C)\in C$ for any $C\in{\mathcal{K}}(G)$, if $u_{0}$ replaces $u$ in the theorem above, then the theorem holds still but the isomorphisms in (i), (ii) and (iii) need change into graded pull-push isomorphisms. ## Preliminaries Throughout this paper we assume that $G$ is a finite group and $k$ is a field with ${\rm char}k\nmid\ \mid\\!G\\!\mid$. Let $\hat{{G}}$ denote the set of all isomorphic classes of irreducible representations of group $G$ and $Z_{s}$ the centralizer of $s$ in $G$. For $h\in G$ and an isomorphism $\phi$ from $G$ to $G^{\prime}$, define a map $\phi_{h}$ from $G$ to $G^{\prime}$ by sending $x$ to $\phi(h^{-1}xh)$ for any $x\in G$. ${\rm deg}(\rho)$ denotes the dimension of representation space of representation $\rho$. Let ${\mathbb{N}}$ and ${\mathbb{Z}}$ denote the sets of all non-negative integers and all integers, respectively. For a set $X$, we denote by $|X|$ the number of elements in $X$. If $X=\oplus_{i\in I}X_{(i)}$ as vector spaces, then we denote by $\iota_{i}$ the natural injection from $X_{(i)}$ to $X$ and by $\pi_{i}$ the corresponding projection from $X$ to $X_{(i)}$. We will use $\mu$ to denote the multiplication of an algebra and use $\Delta$ to denote the comultiplication of a coalgebra. For a (left or right) module and a (left or right) comodule, denote by $\alpha^{-}$, $\alpha^{+}$, $\delta^{-}$ and $\delta^{+}$ the left module, right module, left comodule and right comodule structure maps, respectively. The Sweedler’s sigma notations for coalgebras and comodules are $\Delta(x)=\sum x_{(1)}\otimes x_{(2)}$, $\delta^{-}(x)=\sum x_{(-1)}\otimes x_{(0)}$, $\delta^{+}(x)=\sum x_{(0)}\otimes x_{(1)}$. A quiver $Q=(Q_{0},Q_{1},s,t)$ is an oriented graph, where $Q_{0}$ and $Q_{1}$ are the sets of vertices and arrows, respectively; $s$ and $t$ are two maps from $Q_{1}$ to $Q_{0}$. For any arrow $a\in Q_{1}$, $s(a)$ and $t(a)$ are called its start vertex and end vertex, respectively, and $a$ is called an arrow from $s(a)$ to $t(a)$. For any $n\geq 0$, an $n$-path or a path of length $n$ in the quiver $Q$ is an ordered sequence of arrows $p=a_{n}a_{n-1}\cdots a_{1}$ with $t(a_{i})=s(a_{i+1})$ for all $1\leq i\leq n-1$. Note that a 0-path is exactly a vertex and a 1-path is exactly an arrow. In this case, we define $s(p)=s(a_{1})$, the start vertex of $p$, and $t(p)=t(a_{n})$, the end vertex of $p$. For a 0-path $x$, we have $s(x)=t(x)=x$. Let $Q_{n}$ be the set of $n$-paths. Let ${}^{y}Q_{n}^{x}$ denote the set of all $n$-paths from $x$ to $y$, $x,y\in Q_{0}$. That is, ${}^{y}Q_{n}^{x}=\\{p\in Q_{n}\mid s(p)=x,t(p)=y\\}$. A quiver $Q$ is finite if $Q_{0}$ and $Q_{1}$ are finite sets. A quiver $Q$ is locally finite if ${}^{y}Q_{1}^{x}$ is a finite set for any $x,y\in Q_{0}$. Let $G$ be a group. Let ${\mathcal{K}}(G)$ denote the set of conjugate classes in $G$. A formal sum $r=\sum_{C\in{\mathcal{K}}(G)}r_{C}C$ of conjugate classes of $G$ with cardinal number coefficients is called a ramification (or ramification data ) of $G$, i.e. for any $C\in{\mathcal{K}}(G)$, $r_{C}$ is a cardinal number. In particular, a formal sum $r=\sum_{C\in{\mathcal{K}}(G)}r_{C}C$ of conjugate classes of $G$ with non- negative integer coefficients is a ramification of $G$. For any ramification $r$ and a $C\in{\mathcal{K}}(G)$, since $r_{C}$ is a cardinal number, we can choice a set $I_{C}(r)$ such that its cardinal number is $r_{C}$ without loss generality. Let ${\mathcal{K}}_{r}(G):=\\{C\in{\mathcal{K}}(G)\mid r_{C}\not=0\\}=\\{C\in{\mathcal{K}}(G)\mid I_{C}(r)\not=\emptyset\\}$. If there exists a ramification $r$ of $G$ such that the cardinal number of ${}^{y}Q_{1}^{x}$ is equal to $r_{C}$ for any $x,y\in G$ with $x^{-1}y\in C\in{\mathcal{K}}(G)$, then $Q$ is called a Hopf quiver with respect to the ramification data $r$. In this case, there is a bijection from $I_{C}(r)$ to ${}^{y}Q_{1}^{x}$, and hence we write ${\ }^{y}Q_{1}^{x}=\\{a_{y,x}^{(i)}\mid i\in I_{C}(r)\\}$ for any $x,y\in G$ with $x^{-1}y\in C\in{\mathcal{K}}(G)$. Denote by $(Q,G,r)$ the Hopf quiver of $G$ with respect to $r$. If $\phi:A\rightarrow A^{\prime}$ is an algebra homomorphism and $(M,\alpha^{-})$ is a left $A^{\prime}$-module, then $M$ becomes a left $A$-module with the $A$-action given by $a\cdot x=\phi(a)\cdot x$ for any $a\in A$, $x\in M$, called a pullback $A$-module through $\phi$, written as ${}_{\phi}M$. Dually, if $\phi:C\rightarrow C^{\prime}$ is a coalgebra homomorphism and $(M,\delta^{-})$ is a left $C$-comodule, then $M$ is a left $C^{\prime}$-comodule with the $C^{\prime}$-comodule structure given by ${\delta^{\prime}}^{-}:=(\phi\otimes{\rm id})\delta^{-}$, called a push-out $C^{\prime}$-comodule through $\phi$, written as ${}^{\phi}M$. If $B$ is a Hopf algebra and $M$ is a $B$-Hopf bimodule, then we say that $(B,M)$ is a Hopf bimodules. For any two Hopf bimodules $(B,M)$ and $(B^{\prime},M^{\prime})$, if $\phi$ is a Hopf algebra homomorphism from $B$ to $B^{\prime}$ and $\psi$ is simultaneously a $B$-bimodule homomorphism from $M$ to ${}_{\phi}M^{\prime}{}_{\phi}$ and a $B^{\prime}$-bicomodule homomorphism from ${}^{\phi}M^{\phi}$ to $M^{\prime}$, then $(\phi,\psi)$ is called a pull-push Hopf bimodule homomorphism. If $\psi$ is a bijection, then we say that $(\phi,\psi)$ is a a pull-push Hopf bimodule isomorphism, written as $(B,M)\cong(B^{\prime},M^{\prime})$ as pull-push Hopf bimodules. In particular, if $B=B^{\prime}$ we also write $M\cong M^{\prime}$ as pull-push $B$-Hopf bimodules, in short. Similarly, we say that $(B,M)$ and $(B,X)$ are a Yetter-Drinfeld module and a Yetter-Drinfeld Hopf algebra if $M$ is a Yetter- Drinfeld $B$-module and $X$ is a braided Hopf algebra in Yetter-Drinfeld category ${}^{B}_{B}{\mathcal{Y}D}$, respectively. For any two Yetter-Drinfeld modules $(B,M)$ and $(B^{\prime},M^{\prime})$, if $\phi$ is a Hopf algebra homomorphism from $B$ to $B^{\prime}$, and $\psi$ is simultaneously a left $B$-module homomorphism from $M$ to ${}_{\phi}M^{\prime}$ and a left $B^{\prime}$-comodule homomorphism from ${}^{\phi}M$ to $M^{\prime}$, then $(\phi,\psi)$ is called a pull-push Yetter-Drinfeld module homomorphism. For any two Yetter-Drinfeld Hopf algebra $(B,X)$ and $(B^{\prime},X^{\prime})$, if $\phi$ is a Hopf algebra homomorphism from $B$ to $B^{\prime}$, $\psi$ is simultaneously a left $B$-module homomorphism from $X$ to ${}_{\phi}X^{\prime}$ and a left $B^{\prime}$-comodule homomorphism from ${}^{\phi}X$ to $X^{\prime}$, meantime, $\psi$ also is algebra and coalgebra homomorphism from $X$ to $X^{\prime}$, then $(\phi,\psi)$ is called a pull- push Yetter-Drinfeld Hopf algebra homomorphism (see [ZZC, the remark after Th.4]). Let $A$ be an algebra and $M$ be an $A$-bimodule. Then the tensor algebra $T_{A}(M)$ of $M$ over $A$ is a graded algebra with $T_{A}(M)_{(0)}=A$, $T_{A}(M)_{(1)}=M$ and $T_{A}(M)_{(n)}=\otimes^{n}_{A}M$ for $n>1$. That is, $T_{A}(M)=A\oplus(\bigoplus_{n>0}\otimes^{n}_{A}M)$ (see [N]). Let $D$ be another algebra. If $h$ is an algebra map from $A$ to $D$ and $f$ is an $A$-bimodule map from $M$ to $D=\ _{h}D_{h}$, then by the universal property of $T_{A}(M)$ (see [N, Proposition 1.4.1]) there is a unique algebra map $T_{A}(h,f):T_{A}(M)\rightarrow D$ such that $T_{A}(h,f)\iota_{0}=h$ and $T_{A}(h,f)\iota_{1}=f$. One can easily see that $T_{A}(h,f)=h+\sum_{n>0}\mu^{n-1}T_{n}(f)$. For the details, the reader is directed to [N, Section 1.4] or [ZZC]. Dually, let $C$ be a coalgebra and let $M$ be a $C$-bicomodule. Then the cotensor coalgebra $T_{C}^{c}(M)$ of $M$ over $C$ is a graded coalgebra with $T_{C}^{c}(M)_{(0)}=C$, $T_{C}^{c}(M)_{(1)}=M$ and $T_{C}^{c}(M)_{(n)}=\Box^{n}_{C}M$ for $n>1$. That is, $T_{C}^{c}(M)=C\oplus(\bigoplus_{n>0}\Box^{n}_{C}M)$ (see [N] or [ZZC] ). If $B$ is a Hopf algebra and $M$ is a $B$-Hopf bimodule, then both $T_{B}(M)$ and $T_{B}^{c}(M)$ are graded Hopf algebras. Furthermore, subagebra generated by $H$ and $M$ in $T_{B}^{c}(M)$, written as $H[M]$, is a Hopf subalgebra of $T_{B}^{c}(M)$ and $H[M]$ is called a Hopf algebra of type one. ## 1 Classification of Quiver Hopf Algebras ###### Definition 1.1. $(G,r,\overrightarrow{\rho},u)$ is called a ramification system with irreducible representations (or RSR in short ), if $r$ is a ramification of $G$; $u$ is a map from ${\mathcal{K}}(G)$ to $G$ with $u(C)\in C$ for any $C\in{\mathcal{K}}(G)$; $I_{C}(r,u)$ and $J_{C}(i)$ are sets with $\mid\\!J_{C}(i)\\!\mid$ = ${\rm deg}(\rho_{C}^{(i)})$ and $I_{C}(r)=\\{(i,j)\mid i\in I_{C}(r,u),j\in J_{C}(i)\\}$ for any $C\in{\mathcal{K}}_{r}(G)$, $i\in I_{C}(r,u)$; $\overrightarrow{\rho}=\\{\rho_{C}^{(i)}\\}_{i\in I_{C}(r,u),C\in{\mathcal{K}}_{r}(G)}\ \in\prod_{C\in{\mathcal{K}}_{r}(G)}(\widehat{{Z_{u(C)}}})^{\mid I_{C}(r,u)\mid}$ with $\rho_{C}^{(i)}\in\widehat{{Z_{u(C)}}}$ for any $i\in I_{C}(r,u),C\in{\mathcal{K}}_{r}(G)$. ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$ are said to be isomorphic if the following conditions are satisfied: $\bullet$ There exists a group isomorphism $\phi:G\rightarrow G^{\prime}$. $\bullet$ For any $C\in{\mathcal{K}}(G)$, there exists an element $h_{C}\in G$ such that $\phi(h_{C}^{-1}u(C)h_{C})=u^{\prime}(\phi(C))$. $\bullet$ For any $C\in{\mathcal{K}}_{r}(G)$, there exists a bijective map $\phi_{C}:I_{C}(r,u)\rightarrow I_{\phi(C)}(r^{\prime},u^{\prime})$ such that $\rho_{C}^{(i)}\cong\rho^{\prime}{}_{\phi(C)}^{(\phi_{C}(i))}\phi_{h_{c}}$ as representations of $kZ_{u(C)}$ for all $i\in I_{C}(r,u)$, where $\phi_{h_{C}}(h)=\phi(h^{-1}_{C}hh_{C})$ for any $h\in G$. Remark. Assume that $G=G^{\prime}$, $r=r^{\prime}$, $u(C)=u^{\prime}(C)$ and $I_{C}(r,u)=I_{C}(r^{\prime},u^{\prime})$ for any $C\in{\mathcal{K}}_{r}(G)$. If there is a permutation $\phi_{C}$ on $I_{C}(r,u)$ for any $C\in{\mathcal{K}}_{r}(G)$ such that $\rho^{\prime}{{}_{C}^{(\phi_{C}(i))}}\cong\rho_{C}^{(i)}$ for all $i\in I_{C}(r,u)$, then obviously ${\rm RSR}(G,r,\overrightarrow{\rho},u)\cong{\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u)$. ###### Example 1.2. Assume that $k$ is a complex field and $G=\mathbb{S}_{3}$, then there are 3 elements in ${\mathcal{K}}(G)$, which are $\\{(1)\\}$, $\\{(12),(13),(23)\\}$, $\\{(123),(132)\\}$, and there are 3 non-isomorphic irreducible representations, which are two 1 dimensional irreducible representations $\epsilon$ and $\rm{sgn}$; one 2 dimensional irreducible representation $\rho$. Obviously $Z_{u(\\{1\\})}=G$. The set $\\{{\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\mid\overrightarrow{\rho}=\rho,(sgn,sgn),(\epsilon,sgn),(sgn,\epsilon),(\epsilon,\epsilon)\\}$ is all ${\rm RSR}$’s with $r=r_{C}r$ and $C=\\{(1)\\}$. The set $\\{{\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\mid\overrightarrow{\rho}=\rho,(sgn,sgn),(\epsilon,sgn),(\epsilon,\epsilon)\\}$ is all representatives of isomorphic classes of ${\rm RSR}$’s with $r=r_{C}r$ and $C=\\{(1)\\}$. Furthermore, when $\overrightarrow{\rho}=\rho$, we can set $I_{C}(r,u)=\\{1\\}$ and $J_{C}(1)=\\{1\\}$. In this case $\rho_{C}^{(1)}=\rho$. When $\overrightarrow{\rho}=(\epsilon,sgn)$, we can set $I_{C}(r,u)=\\{1,2\\}$ and $J_{C}(1)=J_{C}(2)=\\{1\\}$. In this case $\rho_{C}^{(1)}=\epsilon$, $\rho_{C}^{(2)}=sgn$. Let $\displaystyle G$ $\displaystyle=$ $\displaystyle\bigcup_{\theta\in\Theta_{C}}Z_{u(C)}g_{\theta},$ (1.1) where $\Theta_{C}$ is an index set, be a coset decomposition of $Z_{u(C)}$ in $G$. It is easy to check that $|\Theta_{C}|=|C|$. We always assume that the representative element of the coset $Z_{u(C)}$ is the identity $1$ of $G$. For any $h\in G$ and $\theta\in\Theta_{C}$, there exist unique $h^{\prime}\in Z_{u(C)}$ and $\theta^{\prime}\in\Theta_{C}$ such that $g_{\theta}h=h^{\prime}g_{\theta^{\prime}}$. Let $\zeta_{\theta}(h)=h^{\prime}$. Then we have $\displaystyle g_{\theta}h$ $\displaystyle=$ $\displaystyle\zeta_{\theta}(h)g_{\theta^{\prime}}.$ (1.2) ###### Proposition 1.3. If $N$ is a $kG$-Hopf bimodule, then there exist a Hopf quiver $(Q,G,r)$, an ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and a $kG$-Hopf bimodule $(kQ_{1}^{c},\alpha^{-},\alpha^{+})$ with $\displaystyle\alpha^{-}(h\otimes a^{(i,j)}_{y,x}):=h\cdot a^{(i,j)}_{y,x}=a^{(i,j)}_{hy,hx},\ \ \alpha^{+}(a^{(i,j)}_{y,x}\otimes h):=a^{(i,j)}_{y,x}\cdot h=\sum_{s\in J_{C}(i)}k_{C,h}^{(i,j,s)}a^{(i,s)}_{yh,xh}$ (1.3) for some $k_{C,h}^{(i,j,s)}\in k$ such that $N\cong(kQ_{1}^{c},\alpha^{-},\alpha^{+})$ as $kG$-Hopf bimodules, where $x,y,h\in G$ with $x^{-1}y=g^{-1}_{\theta}u(C)g_{\theta}$, $\zeta_{\theta}$ is given by [ZZC, (0.3)], $C\in{\mathcal{K}}_{r}(G)$, $i\in I_{C}(r,u)$, $j\in J_{C}(i)$, $x_{C}^{(i,j)}\cdot\zeta_{\theta}(h)=\sum_{s\in J_{C}(i)}k_{C,h}^{(i,j,s)}x_{C}^{(i,s)}$. Proof. Since $N$ is a $kG$-Hopf bimodule, there exists an object $\prod_{C\in{\mathcal{K}}(G)}M(C)$ in $\prod_{C\in{\mathcal{K}}(G)}{\mathcal{M}}_{kZ_{u(C)}}$ such that $M(C)$ is a $kZ_{u(C)}$-module for any $C\in{\mathcal{K}}(G)$ and $N\cong\bigoplus_{y=xg_{\theta}^{-1}u(C)g_{\theta},\ x,y\in G}\ x\otimes M(C)\otimes_{kZ_{u(C)}}g_{\theta}$ as $kG$-Hopf bimodules by [CR97] or [ZZC, Th. 1]. Let $r=\sum_{C\in{\mathcal{K}}(G)}r_{C}C$ with $r_{C}={\rm dim}M(C)$ for any $C\in{\mathcal{K}}(G)$. Notice that ${\rm dim}M(C)$ denotes the cardinal number of a basis of $M(C)$ when $M(C)$ is infinite dimensional. Since $M(C)$ is a $kZ_{u(C)}$-module and $kZ_{u(C)}$ is semisimple, there exists a family of irreducible representations $\\{(X_{C}^{(i)},\rho_{C}^{(i)})\mid i\in I_{C}(r,u)\\}$ such that $M(C)=\bigoplus_{i\in I_{C}(r,u)}(X_{C}^{(i)},\rho_{C}^{(i)})$. Let $\\{{x_{C}^{(i,j)}\mid j\in J_{C}(i)}\\}$ be a k-basis of $X_{C}^{(i)}$ for any $i\in I_{C}(r,u)$. Then for any $h\in G$ there are some $k_{C,h}^{(i,j,s)}\in k$ such that $x_{C}^{(i,j)}\cdot\zeta_{\theta}(h)=\sum_{s\in J_{C}(i)}k_{C,h}^{(i,j,s)}x_{C}^{(i,s)}$ for all $i\in I_{C}(r,u)$ and $j\in J_{C}(i)$ since $x_{C}^{(i,j)}\cdot\zeta_{\theta}(h)\in X_{C}^{(i)}$. It remains to show that $(kQ_{1}^{c},\alpha^{-},\alpha^{+})$ is isomorphic to $\bigoplus_{y=xg_{\theta}^{-1}u(C)g_{\theta},\ x,y\in G}\ x\otimes M(C)\otimes_{kZ_{u(C)}}g_{\theta}$ as $kG$-Hopf bimodules. Observe that there is a canonical $kG$-bicomodule isomorphism $\varphi:kQ_{1}\rightarrow\bigoplus_{y=xg_{\theta}^{-1}u(C)g_{\theta},\ x,y\in G}\ x\otimes M(C)\otimes_{kZ_{u(C)}}g_{\theta}$ given by $\displaystyle\varphi(a^{(i,j)}_{y,x})=x\otimes x_{C}^{(i,j)}\otimes_{kZ_{u(C)}}g_{\theta}$ (1.4) where $x,y\in G$ with $x^{-1}y=g^{-1}_{\theta}u(C)g_{\theta}$, $C\in{\mathcal{K}}_{r}(G)$ , $i\in I_{C}(r,u)$ and $j\in J_{C}(r)$. See $\displaystyle\varphi(\alpha^{-}(h\otimes a^{(i,j)}_{y,x}))$ $\displaystyle=$ $\displaystyle\varphi(a^{(i,j)}_{hy,hx})=hx\otimes x^{(i,j)}_{C}\otimes_{kZ_{u(C)}}g_{\theta}\ \ $ $\displaystyle=$ $\displaystyle h\cdot(x\otimes x^{(i,j)}_{C}\otimes_{kZ_{u(C)}}g_{\theta})\ \ \ (\hbox{see }\cite[cite]{[\@@bibref{}{ZZC04}{}{}, (1.2)]})$ $\displaystyle=$ $\displaystyle h\cdot\varphi(a^{(i,j)}_{y,x})\ \ .$ Thus $\varphi$ is a left $kG$-module isomorphism. See $\displaystyle\alpha^{+}(\varphi(a^{(i,j)}_{y,x})\otimes h)$ $\displaystyle=$ $\displaystyle xh\otimes x_{C}^{(i,j)}\cdot\zeta_{\theta}(h)\otimes g_{\theta^{\prime}}$ $\displaystyle=$ $\displaystyle xh\otimes(\sum_{s\in J_{C}(i)}{k_{C,h}^{(i,j,s)}}x_{C}^{(i,s)})\otimes g_{\theta^{\prime}})$ $\displaystyle=$ $\displaystyle\varphi(\sum_{s\in J_{C}(i)}{k_{C,h}^{(i,j,s)}}a_{yh,xh}^{(i,s)})$ $\displaystyle=$ $\displaystyle\varphi(\alpha^{+}(a_{y,x}^{(i,j)}\otimes h))\ \ (\hbox{by }(\ref{e1.10.11})).$ Consequently, $\varphi$ is a $kG$-Hopf bimodule isomorphism. $\Box$ Let $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)$ denote the $kG$-Hopf bimodule $(kQ_{1}^{c},\alpha^{-},\alpha^{+})$ given in Proposition 1.3. Furthermore, if $(kQ_{1}^{c},kQ_{1}^{a})$ is an arrow dual pairing, i.e. $kQ_{1}^{c}$ is isomorphic to the dual of $kQ_{1}^{a}$ as $kG$-Hopf bimodules or $kQ_{1}^{a}$ is isomorphic to the dual of $kQ_{1}^{c}$ as $(kG)^{*}$-Hopf bimodules under the isomorphisms in [ZZC, Lemma 1.7] (cf [ZZC, the argument before Def. 1.8]), then we denote the $(kG)^{*}$-Hopf bimodule $kQ_{1}^{a}$ by $(kQ_{1}^{a},G,r,\overrightarrow{\rho},u)$. We obtain six quiver Hopf algebras $kQ^{c}(G,r,\overrightarrow{\rho},u)$, $kQ^{s}(G,r,\overrightarrow{\rho},u)$, $kG[kQ_{1}^{c},G,r,\overrightarrow{\rho},u]$, $kQ^{a}(G,r,\overrightarrow{\rho},u)$, $kQ^{sc}(G,r,\overrightarrow{\rho},u),$ $(kG)^{*}[kQ_{1}^{a},G,r,\overrightarrow{\rho},$ $u]$, called the quiver Hopf algebras determined by ${\rm RSR}(G,$ $r,$ $\overrightarrow{\rho},u)$. From Proposition 1.3, it seems that the right $kG$-action on $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)$ depends on the choice of the set $\\{g_{\theta}\mid\theta\in\Theta_{C}\\}$ of coset representatives of $Z_{u(C)}$ in $G$ (see, (1.1) or [ZZC, (0.1)]). The following lemma shows that $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)$ is, in fact, independent of the choice of the coset representative set $\\{g_{\theta}\mid\theta\in\Theta_{C}\\}$, up to $kG$\- Hopf bimodule isomorphisms. For a while, we write $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)=(kQ_{1}^{c},G,r,\overrightarrow{\rho},u,\\{g_{\theta}\\})$ given before. Now let $\\{h_{\theta}\in G\mid\theta\in\Theta_{C}\\}$ be another coset representative set of $Z_{u(C)}$ in $G$ for any $C\in{\mathcal{K}}(G)$. That is, $\displaystyle G=\bigcup_{\theta\in\Theta_{C}}Z_{u(C)}h_{\theta}.$ (1.5) ###### Lemma 1.4. With the above notations, $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u,\\{g_{\theta}\\})$ and $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u,$ $\\{h_{\theta}\\})$ are isomorphic $kG$-Hopf bimodules. Proof. We may assume $Z_{u(C)}h_{\theta}=Z_{u(C)}g_{\theta}$ for any $C\in{\mathcal{K}}(G)$ and $\theta\in\Theta_{C}$. Then $g_{\theta}h^{-1}_{\theta}\in Z_{u(C)}$. Now let $x,y,h\in G$ with $x^{-1}y=g^{-1}_{\theta}u(C)g_{\theta}$. Then $x^{-1}y=$ $h^{-1}_{\theta}(g_{\theta}h^{-1}_{\theta})^{-1}u(C)(g_{\theta}h^{-1}_{\theta})h_{\theta}$ $=h^{-1}_{\theta}u(C)h_{\theta}$ and $\displaystyle h_{\theta}h=(h_{\theta}g^{-1}_{\theta})g_{\theta}h=(h_{\theta}g^{-1}_{\theta})\zeta_{\theta}(h)g_{\theta^{\prime}}=(h_{\theta}g^{-1}_{\theta})\zeta_{\theta}(h)(g_{\theta^{\prime}}h^{-1}_{\theta^{\prime}})h_{\theta^{\prime}}\ ,$ (1.6) where $g_{\theta}h=\zeta_{\theta}(h)g_{\theta^{\prime}}$. For any $C\in{\mathcal{K}}(G)$, let $M(C)$ be a right $kZ_{u(C)}$-module. Let $N:=\bigoplus_{y=xg_{\theta}^{-1}u(C)g_{\theta},\ x,y\in G}\ x\otimes M(C)\otimes_{kZ_{u(C)}}g_{\theta}$ and $M:=\bigoplus_{y=xh_{\theta}^{-1}u(C)h_{\theta},\ x,y\in G}\ x\otimes M(C)\otimes_{kZ_{u(C)}}h_{\theta}$ be two $kG$-Hopf bimodules. It is sufficient to show $N\cong M$ as $kG$-Hopf bimodules by the proof of Proposition 1.3. Considering $x\otimes w\otimes_{kZ_{u(C)}}g_{\theta}=x\otimes w\cdot g_{\theta}h_{\theta}^{-1}\otimes_{kZ_{u(C)}}h_{\theta}$, we have that $f:N\rightarrow M$ given by $f(x\otimes w\otimes_{kZ_{u(C)}}g_{\theta})=x\otimes w\cdot g_{\theta}h_{\theta}^{-1}\otimes_{kZ_{u(C)}}h_{\theta}$ for any $w\in M(C)$, any $x,y\in G$ with $x^{-1}y=g^{-1}_{\theta}u(C)g_{\theta}$, $C\in{\mathcal{K}}_{r}(G)$ and $i\in I_{C}(r,u)$, is a $k$-linear isomorphism. It is clear that $f$ is a $kG$-bicomodule isomorphism and a left $kG$-module isomorphism from $N$ to $M$. See $\displaystyle(f(x\otimes x_{C}^{(i,j)}\otimes_{kZ_{u(C)}}g_{\theta}))\cdot h$ $\displaystyle=$ $\displaystyle(x\otimes(x_{C}^{(i,j)})\rho_{C}^{(i)}(g_{\theta}h_{\theta}^{-1})\otimes_{kZ_{u(C)}}h_{\theta})\cdot h$ $\displaystyle=$ $\displaystyle xh\otimes(x_{C}^{(i,j)})\rho_{C}^{(i)}(\zeta_{\theta}(h))\rho_{C}^{(i)}(g_{\theta^{\prime}}h_{\theta^{\prime}}^{-1})\otimes_{kZ_{u(C)}}h_{\theta^{\prime}}\ \ (\hbox{by (\ref{e1.131})})$ $\displaystyle=$ $\displaystyle f((x\otimes x_{C}^{(i,j)}\otimes_{kZ_{u(C)}}g_{\theta})\cdot h),$ for any $x,y,h\in G$, $i\in I_{C}(r,u)$, $j\in J_{C}(i)$, $C\in{\mathcal{K}}_{r}(G)$ with $x^{-1}y=g^{-1}_{\theta}u(C)g_{\theta}$. Thus $f$ is a right $kG$-module homomorphism. $\Box$ Now we state one of our main result, which classifies quiver Hopf algebras. ###### Theorem 2. Let $(G,r,\overrightarrow{\rho},u)$ and $(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$ be two RSR’s. Then the following statements are equivalent: (i) ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$. (ii) There exists a Hopf algebra isomorphism $\phi:kG\rightarrow kG^{\prime}$ such that $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)\cong\ _{\phi}^{\phi^{-1}}((kQ_{1}^{\prime}{}^{c},G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})){}_{\phi}^{\phi^{-1}}$ as $kG$-Hopf bimodules. (iii) $kQ^{c}(G,r,\overrightarrow{\rho},u)\cong k{Q^{\prime}}^{c}(G,r,\overrightarrow{\rho},u)$. (iv) $kQ^{s}(G,r,\overrightarrow{\rho},u)\cong k{Q^{\prime}}^{s}(G^{\prime},r^{\prime},\overrightarrow{\rho}^{\prime},u^{\prime})$. (v) $kG[kQ_{1}^{c},G,r,\overrightarrow{\rho},u]$ $\cong kG^{\prime}[kQ_{1}^{\prime}{}^{c},G^{\prime},r^{\prime},\overrightarrow{\rho}^{\prime},u^{\prime}]$. Furthermore, if $Q$ is finite, then the above are equivalent to the following: (vi) $kQ^{a}(G,r,\overrightarrow{\rho},u)$ $\cong k{Q^{\prime}}^{a}(G^{\prime},r^{\prime},\overrightarrow{\rho}^{\prime},u^{\prime})$. (vii) $kQ^{sc}(G,r,\overrightarrow{\rho},u)\cong k{Q^{\prime}}^{sc}(G^{\prime},r^{\prime},\overrightarrow{\rho}^{\prime},u^{\prime})$. (viii) $(kG)^{*}[kQ_{1}^{a},G,r,\overrightarrow{\rho},u]\cong(kG^{\prime})^{*}[kQ_{1}^{\prime}{}^{a},G^{\prime},r^{\prime},\overrightarrow{\rho}^{\prime},u^{\prime}]$. Notice that the isomorphisms above are ones of graded Hopf algebras but (i) (ii). Proof. By [ZZC, Lemma 1.5, Lemma 1.6], we only have to prove (i) $\Leftrightarrow$ (ii). (i) $\Rightarrow$ (ii). Assume that ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$. Let $(X_{C}^{(i)},\rho_{C}^{(i)})$ and $(X^{\prime}{}_{C^{\prime}}^{(i^{\prime})},\rho^{\prime}{}_{C^{\prime}}^{(i^{\prime})})$ be irreducible representations over $Z_{u(C)}$ and $Z_{u^{\prime}(C^{\prime})}$, respectively. Then there exist a group isomorphism $\phi:G\rightarrow G^{\prime}$, an element $h_{C}\in G$ such that $\phi(h^{-1}_{C}u(C)h_{C})=u^{\prime}(\phi(C))$ for any $C\in{\mathcal{K}}(G)$ and a bijective map $\phi_{C}:I_{C}(r,u)\rightarrow I_{\phi(C)}(r^{\prime},u^{\prime})$ such that $(X^{(i)}_{C},\rho_{C}^{(i)})\stackrel{{\scriptstyle\xi_{C}^{(i)}}}{{\cong}}(X^{\prime}{}_{\phi(C)}^{(\phi_{C}(i))},\rho^{\prime}{}_{\phi(C)}^{(\phi_{C}(i))}\phi_{h_{c}})$ as right $kZ_{u(C)}$-modules for all $i\in I_{C}(r,u)$. Now let $G=\bigcup_{\theta\in\Theta_{C}}Z_{u(C)}g_{\theta}$ as in (1.1) or [ZZC, (0.1)] for any $C\in{\mathcal{K}}(G)$. Let $x,y,h\in G$ with $x^{-1}y=g^{-1}_{\theta}u(C)g_{\theta}$, $C\in{\mathcal{K}}_{r}(G)$ and $\theta\in\Theta_{C}$. Assume that $g_{\theta}h^{-1}_{C}=\zeta_{\theta}(h^{-1}_{C})g_{\eta}$, $g_{\theta}h=\zeta_{\theta}(h)g_{\theta^{\prime}}$, $g_{\eta}(h_{C}hh^{-1}_{C})=\zeta_{\eta}(h_{C}hh^{-1}_{C})g_{\eta^{\prime}}$ and $g_{\theta^{\prime}}h^{-1}_{C}=\zeta_{\theta^{\prime}}(h^{-1}_{C})g_{\theta^{\prime\prime}}$ with $\zeta_{\theta}(h^{-1}_{C}),$ $\zeta_{\theta}(h),$ $\zeta_{\eta}(h_{C}hh^{-1}_{C}),$ $\zeta_{\theta^{\prime}}(h^{-1}_{C})\in Z_{u(C)}$ and $\eta,\theta^{\prime},\eta^{\prime},\theta^{\prime\prime}\in\Theta_{C}$. Then we have $\displaystyle g_{\theta}hh^{-1}_{C}$ $\displaystyle=$ $\displaystyle\zeta_{\theta}(h)g_{\theta^{\prime}}h^{-1}_{C}\ =\zeta_{\theta}(h)\zeta_{\theta^{\prime}}(h^{-1}_{C})g_{\theta^{\prime\prime}}$ (1.7) and $\displaystyle g_{\theta}hh^{-1}_{C}$ $\displaystyle=$ $\displaystyle(g_{\theta}h^{-1}_{C})(h_{C}hh^{-1}_{C})=\zeta_{\theta}(h^{-1}_{C})g_{\eta}(h_{C}hh^{-1}_{C})\ =\zeta_{\theta}(h^{-1}_{C})\zeta_{\eta}(h_{C}hh^{-1}_{C})g_{\eta^{\prime}}.$ (1.8) It follows that $\displaystyle\theta^{\prime\prime}=\eta^{\prime}\ \ \ \hbox{and }\ \ \ \zeta_{\theta}(h)\zeta_{\theta^{\prime}}(h^{-1}_{C})=\zeta_{\theta}(h^{-1}_{C})\zeta_{\eta}(h_{C}hh^{-1}_{C}).$ (1.9) Moreover, we have $(xh)^{-1}(yh)=h^{-1}g^{-1}_{\theta}u(C)g_{\theta}h=g^{-1}_{\theta^{\prime}}u(C)g_{\theta^{\prime}}$ and $g_{\theta}=g_{\theta}h^{-1}_{C}h_{C}=\zeta_{\theta}(h^{-1}_{C})g_{\eta}h_{C}$. Thus $\begin{array}[]{rcl}\phi(x)^{-1}\phi(y)&=&\phi(x^{-1}y)=\phi(g^{-1}_{\theta}u(C)g_{\theta})=\phi(h^{-1}_{C}g^{-1}_{\eta}u(C)g_{\eta}h_{C})\\\ &=&\phi(h^{-1}_{C}g^{-1}_{\eta}h_{C})\phi(h^{-1}_{C}u(C)h_{C})\phi(h^{-1}_{C}g_{\eta}h_{C})\\\ &=&\phi(h^{-1}_{C}g_{\eta}h_{C})^{-1}u^{\prime}(\phi(C))\phi(h^{-1}_{C}g_{\eta}h_{C}).\end{array}$ We also have $\displaystyle\phi(h^{-1}_{C}g_{\eta}h_{C})\phi(h)$ $\displaystyle=$ $\displaystyle\phi(h^{-1}_{C}g_{\eta}h_{C}hh^{-1}_{C}h_{C})$ (1.10) $\displaystyle=$ $\displaystyle\phi(h^{-1}_{C}\zeta_{\eta}(h_{C}hh^{-1}_{C})g_{\eta^{\prime}}h_{C})$ $\displaystyle=$ $\displaystyle\phi(h^{-1}_{C}\zeta_{\eta}(h_{C}hh^{-1}_{C})h_{C})\phi(h^{-1}_{C}g_{\eta^{\prime}}h_{C}).$ Since $\begin{array}[]{rcl}\phi_{h_{C}}(g_{\eta^{\prime}})\phi(xh)^{-1}\phi(yh)&=&\phi(h_{C}^{-1}(\zeta_{\eta}(h_{C}hh_{C}^{-1}))^{-1}(\zeta_{\theta}(h_{C}^{-1}))^{-1}u(C)g_{\theta}h)\ \ \ (\hbox{by }\ref{e1.101})\\\ &=&u^{\prime}(\phi(C))\phi_{h_{C}}(g_{\eta^{\prime}})\ \ \ (\hbox{by }\ref{e1.101}),\\\ \end{array}$ $\displaystyle\phi(xh)^{-1}\phi(yh)$ $\displaystyle=$ $\displaystyle\phi_{h_{C}}(g_{\eta^{\prime}})^{-1}u^{\prime}(\phi(C))\phi_{h_{C}}(g_{\eta^{\prime}}).$ (1.11) It is clear $\displaystyle G^{\prime}=\bigcup_{\theta\in\Theta_{C}}Z_{u^{\prime}(\phi(C))}(\phi(h^{-1}_{C}g_{\theta}h_{C}))$ (1.12) is a coset decomposition of $Z_{u^{\prime}(\phi(C))}$ in $G^{\prime}$ for any $\phi(C)\in{\mathcal{K}}(G^{\prime})$. Let $\displaystyle N:$ $\displaystyle=\bigoplus_{y=xg_{\theta}^{-1}u(C)g_{\theta},\ x,y\in G}\ x\otimes N(C)\otimes_{kZ_{u(C)}}g_{\theta}\ \ \ \ \ \hbox{ and }$ $\displaystyle M:$ $\displaystyle=\bigoplus_{\phi(y)=\phi(x)\phi_{h_{C}}(g_{\eta}^{-1})u^{\prime}(\phi(C)\phi_{h_{C}}(g_{\eta}),\ x,y\in G}\ \phi(x)\otimes M(\phi(C))\otimes_{kZ_{u^{\prime}(\phi(C))}}\phi_{h_{C}}(g_{\eta})$ with $N(C):=\bigoplus_{i\in I_{C}(r,u)}X_{C}^{(i)}$ and $M(\phi(C)):=\bigoplus_{i\in I_{C}(r,u)}X^{\prime}{}_{\phi(C)}^{(\phi_{C}(i))}$. It is suffices to show $N\cong\ ^{\phi^{-1}}_{\phi}M^{\phi^{-1}}_{\phi}$ as $kG$-Hopf bimodules by the proof of Proposition 1.3. Considering $x\otimes(w)\rho^{(i)}_{C}((\zeta_{\theta}(h_{C}^{-1}))^{-1})\otimes_{kZ_{u(C)}}g_{\theta}=x\otimes w\otimes_{kZ_{u(C)}}g_{\eta}h_{c}$, we have that $\psi:N\rightarrow M$ given by $\psi(x\otimes(w)\rho^{(i)}_{C}((\zeta_{\theta}(h_{C}^{-1}))^{-1})\otimes_{kZ_{u(C)}}g_{\theta})=\phi(x)\otimes\xi_{C}^{(i)}(w)\otimes_{kZ_{u^{\prime}(\phi(C))}}\phi_{h_{C}}(g_{\eta})$ for any $x,y\in G$ with $x^{-1}y=g^{-1}_{\theta}u(C)g_{\theta}$, and $i\in I_{C}(r,u)$, $w\in X_{C}^{(i)}$, where $C\in{\mathcal{K}}_{r}(G)$ and $g_{\theta}h^{-1}_{C}=\zeta_{\theta}(h^{-1}_{C})g_{\eta}$ with $\zeta_{\theta}(h^{-1}_{C})\in Z_{u(C)}$ and $\theta,\eta\in\Theta_{C}$, is a $k$-linear isomorphism. It is clear that $\psi$ is a homomorphism not only of $kG$-bicomodules from $N$ to $\ {}^{\phi^{-1}}M^{\phi^{-1}}$ but also of left $kG$-modules from $N$ to $\ {}_{\phi}M$. For any $h\in G$ and $w\in X_{C}^{(i)}$, see $\displaystyle\psi((x\otimes(w)\rho^{(i)}_{C}((\zeta_{\theta}(h_{C}^{-1}))^{-1})\otimes_{kZ_{u(C)}}g_{\theta})\cdot h)$ $\displaystyle=\psi((xh\otimes(w)\rho^{(i)}_{C}((\zeta_{\theta}(h_{C}^{-1}))^{-1}\zeta_{\theta}(h))\otimes_{kZ_{u(C)}}g_{\theta^{\prime}})$ $\displaystyle=\phi(xh)\otimes\xi_{C}^{(i)}((w)\rho^{(i)}_{C}((\zeta_{\theta}(h_{C}^{-1}))^{-1}\zeta_{\theta}(h)\zeta_{\theta^{\prime}}(h_{C}^{-1})))\otimes_{kZ_{u^{\prime}(\phi(C))}}\phi_{h_{C}}(g_{\eta^{\prime}})\ \ \ (\hbox{ by (\ref{e1.102})})$ $\displaystyle=\phi(x)\phi(h)\otimes(\xi_{C}^{(i)}(w))\rho^{(\phi_{C}(i))}_{\phi(C)}\phi_{h_{C}}((\zeta_{\theta}(h_{C}^{-1}))^{-1}\zeta_{\theta}(h)\zeta_{\theta^{\prime}}(h_{C}^{-1})))\otimes_{kZ_{u^{\prime}(\phi(C))}}\phi_{h_{C}}(g_{\eta^{\prime}})$ $\displaystyle\ \ (\hbox{ by Def. \ref{1.1}})$ and $\displaystyle\psi(x\otimes(w)\rho^{(i)}_{C}((\zeta_{\theta}(h_{C}^{-1}))^{-1})\otimes_{kZ_{u(C)}}g_{\theta})\cdot\phi(h)$ $\displaystyle=(\phi(x)\otimes\xi_{C}^{(i)}(w)\otimes_{kZ_{u^{\prime}(\phi(C))}}\phi_{h_{C}}(g_{\eta}))\cdot\phi(h)$ $\displaystyle=\phi(x)\phi(h)\otimes(\xi_{C}^{(i)}(w))\rho^{(\phi_{C}(i))}_{\phi(C)}\phi_{h_{C}}(\zeta_{\eta}(h_{C}hh_{C}^{-1}))\otimes_{kZ_{u^{\prime}(\phi(C))}}\phi_{h_{C}}(g_{\eta^{\prime}})\ \ \ (\hbox{ by (\ref{e1.1311})})$ $\displaystyle=\phi(x)\phi(h)\otimes(\xi_{C}^{(i)}(w))\rho^{(\phi_{C}(i))}_{\phi(C)}\phi_{h_{C}}((\zeta_{\theta}(h_{C}^{-1}))^{-1}\zeta_{\theta}(h)\zeta_{\theta^{\prime}}(h_{C}^{-1})))\otimes_{kZ_{u^{\prime}(\phi(C))}}\phi_{h_{C}}(g_{\eta^{\prime}})$ $\displaystyle\ \ \ (\hbox{ by (\ref{coeff})}),$ which show that $\psi$ is a right $kG$-module homomorphism. (ii) $\Rightarrow$ (i). Assume that there exist a Hopf algebra isomorphism $\phi:kG\rightarrow kG^{\prime}$ and a $kG$-Hopf bimodule isomorphism $\psi:(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)\rightarrow\ ^{\phi^{-1}}_{\phi}(kQ_{1}^{\prime}{}^{c},G^{\prime},r^{\prime},\overrightarrow{\rho}^{\prime},u^{\prime})^{\phi^{-1}}_{\phi}$. Then $\phi:G\rightarrow G^{\prime}$ is a group isomorphism. Let $C\in{\mathcal{K}}(G)$. Then $\phi(u(C))$, $u^{\prime}(\phi(C))\in\phi(C)\in{\mathcal{K}}(G^{\prime})$, and hence $u^{\prime}(\phi(C))=\phi(h_{C})^{-1}\phi(u(C))\phi(h_{C})=\phi(h^{-1}_{C}u(C)h_{C})$ for some $h_{C}\in G$. Since $\psi$ is a $kG^{\prime}$-bicomodule isomorphism from ${}^{\phi}(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)^{\phi}$ to $(kQ_{1}^{\prime}{}^{c},G^{\prime},r^{\prime},\overrightarrow{\rho}^{\prime},u^{\prime})$ and $\phi(h^{-1}_{C}u(C)h_{C})=u^{\prime}(\phi(C))$, by restriction one gets a $k$-linear isomorphism $\psi_{C}:\ ^{h^{-1}_{C}u(C)h_{C}}(kQ_{1})^{1}\rightarrow\ ^{u^{\prime}(\phi(C))}\\!(kQ^{\prime}_{1})^{1},\ x\mapsto\psi(x).$ We also have a $k$-linear isomorphism $f_{C}:\ ^{u(C)}\\!(kQ_{1})^{1}\rightarrow\ ^{h^{-1}_{C}u(C)h_{C}}(kQ_{1})^{1},\ x\mapsto h^{-1}_{C}\cdot x\cdot h_{C}.$ Since $\phi(h^{-1}_{C}u(C)h_{C})=u^{\prime}(\phi(C))$ and $h^{-1}_{C}Z_{u(C)}h_{C}=Z_{h^{-1}_{C}u(C)h_{C}}$, one gets $\phi(h^{-1}_{C}Z_{u(C)}h_{C})=Z_{u^{\prime}(\phi(C))}$. Hence $\phi_{h_{C}}$ is an algebra isomorphism from $kZ_{u(C)}$ to $kZ_{u^{\prime}(\phi(C))}$ by sending $h$ to $\phi(h^{-1}_{C}hh_{C})$. Using the hypothesis that $\psi$ is a $kG$-bimodules homomorphism from $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)$ to ${}_{\phi}(kQ_{1}^{\prime}{}^{c},$ $G^{\prime},r^{\prime},\overrightarrow{\rho}^{\prime},u^{\prime})_{\phi}$, one can easily check that the composition $\psi_{C}f_{C}$ is a right $kZ_{u(C)}$-module isomorphism from $(^{u(C)}\\!(kQ_{1})^{1},\lhd)$ to $((^{u^{\prime}(\phi(C))}\\!(kQ^{\prime}_{1})^{1})_{\phi_{h_{C}}},\lhd)$. Indeed, For any $z\in^{u(C)}\\!(kQ_{1})^{1}$, see $\displaystyle\psi_{C}f_{C}(z)\lhd\phi_{h_{C}}(h)$ $\displaystyle=\psi_{C}(z\lhd h_{C})\lhd\phi_{h_{C}}(h)$ $\displaystyle=\psi_{C}(z\lhd h_{C}\phi_{h_{C}}(h))\ \ \ (\hbox{since }\psi\hbox{ is a bimodule homomorphism})$ $\displaystyle=\psi_{C}f_{C}(z\lhd h).$ Obviously, both ${}^{u(C)}\\!(kQ_{1})^{1}$ and $(^{u^{\prime}(\phi(C))}\\!(kQ^{\prime}_{1})^{1})_{\phi_{h_{C}}}$ are semisimple right $kZ_{u(C)}$-modules. Assume ${}^{u(C)}\\!(kQ_{1})^{1}\cong\oplus_{i\in I_{C}(r,u)}(X_{C}^{(i)},\rho_{C}^{(i)})$ as right $kZ_{u(C)}$-modules and ${}^{u^{\prime}(\phi(C))}\\!(kQ^{\prime}_{1})^{1}\cong\oplus_{j\in I_{\phi(C)}(r^{\prime},u)}(X^{\prime}{}_{\phi(C)}^{(j)},\rho^{\prime}{}_{\phi(C)}^{(j)})$ as right $kZ_{u^{\prime}(\phi(C))}$-modules, where $(X_{C}^{(i)},\rho_{C}^{(i)})$ is an irreducible right $kZ_{u(C)}$-module for any $i\in I_{C}(r,u)$ and $(X^{\prime}{}_{\phi(C)}^{(j)},\rho^{\prime}{}_{\phi(C)}^{(j)})$ is an irreducible right $kZ_{u^{\prime}(\phi(C))}$-module for any $j\in I_{\phi(C)}(r^{\prime},u^{\prime})$. Therefore, there exists a bijective map $\phi_{C}:I_{C}(r,u)\rightarrow I_{\phi(C)}(r^{\prime},u^{\prime})$ such that $(X_{C}^{(i)},\rho_{C}^{(i)})\cong(X^{\prime}{}_{\phi(C)}^{(\phi_{C}(i))},\rho_{\phi(C)}^{(\phi_{C}(i))}\phi_{h_{C}})$ as right $kZ_{u(C)}-modules$ for all $i\in I_{C}(r,u)$. It follows that ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},$ $u^{\prime}).$ $\Box$ Up to now we have classified the quiver Hopf algebras by means of RSR’s. In other words, ramification systems with irreducible representations uniquely determine their corresponding quiver Hopf algebras up to graded Hopf algebra isomorphisms. ###### Proposition 1.5. Let ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u^{\prime})$ be two RSR’s. If $u^{\prime}(C)=h_{C}^{-1}u(C)h_{C}$ and $\rho_{C}{}^{(i)}=\rho^{\prime}{}_{C}^{(i)}{\rm ad}^{+}_{h_{C}}$ with $I_{C}(r,u)=I_{C}(r,u^{\prime})$ for any $C\in{\mathcal{K}}_{r}(G)$, $i\in I_{C}(r,u)$, where $h_{C}\in G$ and ${\rm ad}^{+}_{h_{C}}(g)=h_{C}^{-1}gh_{C}$, then ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u^{\prime})$. Proof. Let $\phi=id_{G}$ and $\phi_{C}=id_{I_{C}(r,u)}$ for any $C\in{\mathcal{K}}_{r}(G).$ It is clear that ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u^{\prime})$. $\Box$ Remark: This proposition means that the choice of map $u$ doesn’t affect the classification of ${\rm RSR}^{\prime}$s. That is, if we fix a map $u_{0}$ from ${\mathcal{K}}(G)$ to $G$ with $u_{0}(C)\in C$ for any $C\in{\mathcal{K}}(G)$, then for any ${\rm RSR}(G,r,\overrightarrow{\rho},u)$, there exists ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u_{0})$ such that ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u_{0})$. ## 2 The classification of pointed Hopf algebras of type one A graded Hopf algebra $A=\oplus_{n=0}^{\infty}A_{(n)}$ is called to be of Nichols type, if diagram of $A$ is a Nichols algebra over $A_{(0)}$ (the definition of diagram was in [AS98] and [ZZC, Subsection 3.1]). Furthermore, if the coradical of $A$ is a group algebra, then $A$ is called a pointed Hopf algebra of Nichols type. For an ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and a $kG$-Hopf bimodule $(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)$ with the module operations $\alpha^{-}$ and $\alpha^{+}$, define a new left $kG$-action on $kQ_{1}$ by $g\rhd x:=g\cdot x\cdot g^{-1},\ g\in G,x\in kQ_{1},$ where $g\cdot x=\alpha^{-}(g\otimes x)$ and $x\cdot g=\alpha^{+}(x\otimes g)$ for any $g\in G$ and $x\in kQ_{1}$. With this left $kG$-action and the original left (arrow) $kG$-coaction $\delta^{-}$, $kQ_{1}$ is a Yetter- Drinfeld $kG$-module. Let $Q_{1}^{1}:=\\{a\in Q_{1}\mid s(a)=1\\}$, the set of all arrows with starting vertex $1$. It is clear that $kQ_{1}^{1}$ is a Yetter-Drinfeld $kG$-submodule of $kQ_{1}$, denoted by $(kQ_{1}^{1},ad(G,r,\overrightarrow{\rho},u))$. ###### Lemma 2.1. (i) If $H$ is a Hopf algebra with bijective antipode and $(B,\alpha_{B}^{-},\delta_{B}^{-})$ is a graded braided Hopf algebra in ${}^{H}_{H}{\mathcal{Y}D}$ with $B_{(0)}=k1_{B}$, then ${\rm diag}(B\\#H)=B\\#1_{H}\cong B$ as graded braided Hopf algebras in ${}^{H}_{H}{\mathcal{Y}D}$. (ii) $A$ is a pointed Hopf algebra of Nichols type if and only if $A$ is isomorphic to biproduct ${\mathcal{B}}(V)\\#kG$ with Nichols algebra ${\mathcal{B}}(V)$ over group agebra $kG$, $A_{(0)}=kG$ and $V=A_{(1)}$. Proof. (i) Obviously, ${\rm diag}(B\\#H)=B\otimes 1_{H}$. Define a map $\psi$ from ${\rm diag}(B\\#H)$ to $B$ by sending $x\otimes 1_{H}$ to $x$ for any $x\in B$. It is easy to check that $\psi$ is a graded braided Hopf algebra isomorphism in ${}^{H}_{H}{\mathcal{Y}D}$. (ii) If $A$ is a pointed Hopf algebra of Nichols type, then ${\rm diag}(A)={\mathcal{B}}(V)$ is a Nichols algebra in ${}^{kG}_{kG}{\mathcal{Y}D}$ and the coradical of $A$ is group algebra $kG$. Therefore $A\cong{\mathcal{B}}(V)\\#kG$ as graded Hopf algebras. By [AS98, Lemma 2.5], $A_{(0)}=kG$ and $A_{(1)}=V.$ Conversely, clearly, ${\rm diag}({\mathcal{B}}(V)\\#kG)=$ $({\mathcal{B}}(V)\\#kG)^{{\rm co}kG}=$ ${\mathcal{B}}(V)\\#1\cong$ ${\mathcal{B}}(V)$ by (i) and the coradical of ${\mathcal{B}}(V)\\#kG$ is $({\mathcal{B}}(V)\\#kG)_{(0)}=kG$. $\Box$ ###### Lemma 2.2. If $H=kG$ is a group algebra and $M$ is an $H$-Hopf bimodule, then pointed Hopf algebra $H[M]$ of type one is a Hopf algebra of Nichols type. In particular, one-type-co-path Hopf algebra $kG[kQ_{1}^{c},G,r,\overrightarrow{\rho},u]$ is a pointed Hopf algebra of Nichols type and ${\rm diag}(kG[kQ_{1}^{c},G,r,\overrightarrow{\rho},u])$ $={\mathcal{B}}(kQ_{1}^{1},{\rm ad}(G,r,\overrightarrow{\rho},u))$. Proof. By Proposition 1.3 there exists an ${\rm RSR}(G,r\overrightarrow{\rho},u)$ such that $M\cong(kQ_{1}^{c},G,r\overrightarrow{\rho},u)$ as $kG$-Hopf bimodules. Thus $kG[M]\cong kG[kQ_{1}^{c}]$ as graded Hopf algebras by [ZZC, Lemma 1.6]. Therefore, it is enough to show that $kG[kQ_{1}^{c}]$ is of Nichols type. Let $A:=kG[kQ_{1}^{c}]$ and $R:={\rm diag}(kG[kQ_{1}^{c}])$. Obviously, $R_{(0)}=k.$ Now we show $R_{(1)}=kQ_{1}^{1}$. Obviously, $kQ_{1}^{1}\subseteq R_{(1)}$. Let $\alpha=\sum_{p=1}^{n}k_{p}b^{(p)}\in R_{(1)}$, where $b^{(p)}$ is an arrow from $x^{(p)}$ to $y^{(p)}$ with $0\not=k_{p}\in k$, and $b^{(1)},b^{(2)},\cdots,b^{(n)}$ are different each other. Therefore $\sum_{p=1}^{n}k_{p}b^{(p)}\otimes 1=\sum_{p=1}^{n}k_{p}b^{(p)}\otimes x^{p}$, which implies $x^{(p)}=1$ for $1\leq p\leq n.$ Thus $\alpha\in kQ_{1}^{1}.$ We next show that $R$ is generated by $R_{(1)}$ as algebras. Let $\mu$ denote the multiplication and let $B$ denote the algebra generated by $kQ_{1}^{1}$ in $kG[kQ_{1}^{c}]$. Obviously, $B\subseteq R.$ It follows from argument in [ZZC, Subsection 3.1] that $\alpha^{+}:=\mu(id\otimes\iota_{0})$ is an algebraic isomorphism from $R\\#kG$ to $kG[kQ_{1}^{c}]$. For any $x,y\in G$ and any arrow $a_{y,x}$ from $x$ to $y$, see $\displaystyle a_{y,x}$ $\displaystyle=$ $\displaystyle x\cdot a_{x^{-1}y,1}=\mu(x\otimes a_{x^{-1}y,1})=\mu(\alpha^{+}(1\\#x)\otimes\alpha^{+}(a_{x^{-1}y,1}\\#1))$ $\displaystyle=$ $\displaystyle\alpha^{+}(\mu((1\\#x)\otimes(a_{x^{-1}y,1}\\#1)))=\alpha^{+}(x\rhd a_{x^{-1}y,1}\\#x)\in\alpha^{+}(B\\#kG).$ Therefore $\alpha^{+}(B\\#kG)=\alpha^{+}(R\\#kG)$ and so $B=R.$ It is enough to show $P(R)=kQ_{1}^{1}$, where $P(R)$ denotes the set of all primitive elements in $R$. For any $a\in Q_{1}^{1}$ with $\delta^{-}(a)=y$ and $\delta^{+}(a)=1$, see $\Delta_{R}(a)=$ $(\omega\otimes id)\Delta_{A}(a)=$ $1\otimes a+a\otimes 1$ (see [ZZC, Section 3] or [Ra]), i.e. $kQ_{1}^{1}\subseteq P(R)$, where $\omega=\mu_{A}(id\otimes\iota_{0}\pi_{0}S)\Delta_{A}$. Conversely, we shall show $P(R)\subseteq kQ_{1}^{1}$ by the following two steps. Obviously, $kG\cap P(R)=0$ and $P(R)$ is a graded subspace of $R$. (i) Assume that $\alpha=a_{x_{n}x_{n-1}}a_{x_{n-1}x_{n-2}}\cdots a_{x_{1}x_{0}}$ is a path from vertex $x_{0}$, via arrows $a_{x_{1}x_{0}}$ $,\cdots,$ $a_{x_{n-1}x_{n-2}}$, $a_{x_{n}x_{n-1}}$, to vertex $x_{n}$. Then $\omega(\alpha)=\alpha\cdot x_{0}^{-1}$. (ii) Let $v=\sum_{p=1}^{m}k_{p}\alpha_{p}\in P(R)$, where $\alpha_{p}=b_{x_{n}x_{n-1}}^{(p)}b_{x_{n-1}x_{n-2}}^{(p)}\cdots b_{x_{1}x_{0}}^{(p)}$ is a path with $n>1$, $k_{p}\in k$ for $p=1,2,\cdots,m$, and $b_{x_{j}x_{j-1}}^{(p)}$ is an arrow from vertex $x_{j-1}$ to vertex $x_{j}$ for $j=1,2,\cdots,n$. We shall show that $k_{p}=0$ for $p=1,2,\cdots,m.$ Indeed, $\displaystyle\Delta_{R}(v)$ $\displaystyle=$ $\displaystyle\sum_{p=1}^{m}\sum_{j=0}^{n}k_{p}(b_{x_{n}x_{n-1}}^{(p)}b_{x_{n-1}x_{n-}}^{(p)}\cdots b_{x_{j+1}x_{j}}^{(p)})\cdot(x^{(p)}_{j})^{-1}\otimes b_{x_{j}x_{j-1}}^{(p)}b_{x_{j-1}x_{j-2}}^{(p)}\cdots b_{x_{1}x_{0}}^{(p)}$ $\displaystyle=$ $\displaystyle\sum_{p=1}^{m}k_{p}(\alpha_{p}\otimes 1+1\otimes\alpha_{p}).$ This implies $\displaystyle\sum_{p=1}^{m}k_{p}(b_{x_{n}x_{n-1}}^{(p)}b_{x_{n-1}x_{n-2}}^{(p)}\cdots b_{x_{j+1}x_{j}}^{(p)})\cdot(x^{(p)}_{j})^{-1}\otimes b_{x_{j}x_{j-1}}^{(p)}b_{x_{j-1}x_{j-2}}^{(p)}\cdots b_{x_{1}x_{0}}^{(p)}$ $\displaystyle=$ $\displaystyle 0,$ (2.1) for $j=1,2,\cdots,n-1$, because of their length. For any $j$ with $1\leq j\leq n-1$, assume that $\\{p\mid b_{x_{j}x_{j-1}}^{(p)}b_{x_{j-1}x_{j-2}}^{(p)}\cdots b_{x_{1}x_{0}}^{(p)}$ $=b_{x_{j}x_{j-1}}^{(1)}b_{x_{j-1}x_{j-2}}^{(1)}\cdots b_{x_{1}x_{0}}^{(1)}\\}$ $=\\{1,2,\cdots,m_{1}\\}$ without loss generality. Therefore, by (2.1), $\sum_{p=1}^{m_{1}}k_{p}(b_{x_{n}x_{n-1}}^{(p)}b_{x_{n-1}x_{n-2}}^{(p)}\cdots b_{x_{j+1}x_{j}}^{(p)})\cdot(x^{(p)}_{j})^{-1}\otimes b_{x_{j}x_{j-1}}^{(p)}b_{x_{j-1}x_{j-2}}^{(p)}\cdots b_{x_{1}x_{0}}^{(p)}=0,$ since $b_{x_{j}x_{j-1}}^{(q)}b_{x_{j-1}x_{j-2}}^{(q)}\cdots b_{x_{1}x_{0}}^{(q)}\not=b_{x_{j}x_{j-1}}^{(1)}b_{x_{j-1}x_{j-2}}^{(1)}\cdots b_{x_{1}x_{0}}^{(1)}$ for $q=m_{1}+1,\cdots,m.$ This implies $\sum_{p=1}^{m_{1}}k_{p}b_{x_{n}x_{n-1}}^{(p)}b_{x_{n-1}x_{n-2}}^{(p)}\cdots b_{x_{j+1}x_{j}}^{(p)}=0$ and $k_{p}=0$ for $p=1,2,\cdots,m_{1}$. Similarly, we can show $k_{p}=0$ for $p=m_{1}+1,\cdots,m.$ $\Box$ By [BD, Theorem 4.3.2], the category ${}^{H}_{H}{\mathcal{Y}D}$ of Yetter- Drinfeld modules is equivalent to the category ${}^{H}_{H}\\!{\mathcal{M}}^{H}_{H}$ of $H$-Hopf bimodule, where $H$ is a Hopf algebra with bijective antipode. Let $T$ and $U$ be the two corresponding functors. For any $N\in^{H}_{H}\\!{\mathcal{Y}D}$, according to [BD, Prop.4.2.1], $T(N):=N\rtimes H=N\otimes H$ as vector spaces, and the actions and coactions are given as follows: the left (co)actions are diagonal and right (co)actions are induced by $H$. Explicitly, $g\cdot(x\otimes h):=g\cdot x\otimes gh$, $(x\otimes h)\cdot g=x\otimes hg$, $\delta^{-}_{N\rtimes H}(x\otimes h):=\sum_{x}x_{(-1)}h\otimes x_{(0)}\otimes h;$ $\delta^{+}_{N\rtimes H}(x\otimes h):=x\otimes h\otimes h$, where $\delta_{N}^{-}(x)=\sum_{x}x_{(-1)}\otimes x_{(0)}$, $x\in N,h,g\in H.$ For any $M\in^{H}_{H}\\!{\mathcal{M}}^{H}_{H}$, according to [BD, Equ. (7) and (21)], $U(M)$ is the coinvariant of $M$ as a vector space, i.e., $U(M):=M^{{\rm co}H}:=\\{x\in M\mid\delta_{N}^{+}(x)=x\otimes 1\\}$. The left action is left adjoint action and the left coaction is the restricted coaction of the original coaction of $M$. That is, $\alpha_{U(M)}^{-}(h\otimes x)=h\triangleright_{\rm ad}x:=\alpha_{M}^{+}(\alpha_{M}^{-}(h\otimes x)\otimes h^{-1})=(h\cdot x)\cdot h^{-1}$ and $\delta_{U(M)}^{-}(x)=\delta_{M}^{-}(x)$ for any $h\in H,x\in U(M).$ In fact, $TU(M)=U(M)\rtimes H$ and $UT(N)=N\otimes 1_{H}$. Let $\lambda_{N}$ be a map from $N\otimes 1_{H}$ to $N$ by sending $x\otimes 1_{H}$ to $x$ for any $x\in N$, and let $\nu_{M}$ be a map from $U(M)\rtimes H$ to $M$ by sending $x\otimes h$ to $\alpha_{M}^{+}(x\otimes h)=x\cdot h$ for any $x\in U(M)$ and $h\in H$. $\lambda$ and $\nu$ are the natural isomorphisms from functor $UT$ to $id$ and from functor $TU$ to $id$, respectively. Note that the inverse of $\nu_{M}$ is $(\alpha^{+}_{M}\otimes id)(id\otimes S\otimes id)(\delta^{+}_{M}\otimes id)\delta_{M}^{+}.$ Remark: We have $U(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)=(kQ_{1}^{1},{\rm ad}(G,r,\overrightarrow{\rho},u))$ by the proof of Lemma 2.2. ###### Lemma 2.3. Assume that $\phi$ is a Hopf algebra isomorphism from $H$ to $H^{\prime}$. Let $N\in{}^{H^{\prime}}_{H^{\prime}}{\mathcal{Y}D}$ and $M\in{}^{H^{\prime}}_{H^{\prime}}\\!{\mathcal{M}}^{H^{\prime}}_{H^{\prime}}$. Then $T({}{{}^{\phi}}^{-1}_{\phi}\\!N)\cong{}{{}^{\phi}}^{-1}_{\phi}T(N){{}^{\phi}}^{-1}_{\phi}\hbox{\ \ in \ }{}^{H}_{H}\\!{\mathcal{M}}^{H}_{H}\hbox{ \ \ and \ \ }U({}{{}^{\phi}}^{-1}_{\phi}M{{}^{\phi}}^{-1}_{\phi})\cong{{}^{\phi}}^{-1}_{\phi}U(M)\hbox{\ \ in }\ {}^{H}_{H}{\mathcal{Y}D}.$ Proof. The first isomorphism is given by sending $x\otimes h$ to $x\otimes\phi(h)$ for any $x\in N,h\in H$; the second one is identity. $\Box$ ###### Proposition 2.4. (i) If $N$ is a Yetter-Drinfeld $kG$-module, then there exists an ${\rm RSR}(G,r,$ $\overrightarrow{\rho},u)$ such that $N\cong(kQ_{1}^{1},ad(G,r,\overrightarrow{\rho},u))$ as Yetter-Drinfeld $kG$-modules. (ii) If ${\mathcal{B}}(N)$ is a Nichols algebra in ${}^{kG}_{kG}{\mathcal{Y}D}$, then there exists an ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ such that ${\mathcal{B}}(N)\cong{\mathcal{B}}(kQ_{1}^{1},ad(G,r,\overrightarrow{\rho},u))$ as graded braided Hopf algebras in ${}^{kG}_{kG}{\mathcal{Y}D}$. Proof. (i) Since $T(N)$ is a $kG$-Hopf bimodule, it follows from Proposition 1.3 that there exists an ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ such that $T(N)\cong(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)$ as $kG$-Hopf bimodules. Thus, $N\cong UT(N)$ $\cong U(kQ_{1}^{c},G,r,\overrightarrow{\rho},u)=(kQ_{1}^{1},ad(G,r,\overrightarrow{\rho},u))$ as Yetter-Drinfeld $kG$-modules by [BD, Equ. (7) and (21)]. (ii) It follows from (i) and [AS02, Cor.2.3 ]. $\Box$ ###### Lemma 2.5. Assume that $\phi$ is a Hopf algebra isomorphism from $H$ to $H^{\prime}$. Let $R$ and $R^{\prime}$ be graded braided Hopf algebras in ${}^{H}_{H}{\mathcal{Y}D}$ and ${}^{H^{\prime}}_{H^{\prime}}{\mathcal{Y}D}$ with $R_{(0)}=k1_{R}$ and $R_{(0)}^{\prime}=k1_{R^{\prime}}$, respectively. If $R$ and ${}^{\phi^{-1}}_{\phi}R^{\prime}$ are isomorphic as graded braided Hopf algebras in ${}^{H}_{H}{\mathcal{Y}D}$, then biproducts $R\\#H\cong R^{\prime}\\#H^{\prime}$ as graded Hopf algebras. Proof. Let $\psi$ be a graded braided Hopf algebra isomorphism from $R$ to ${}^{\phi^{-1}}_{\phi}R^{\prime}$ in ${}^{H}_{H}{\mathcal{Y}D}$. Define a map $\nu$ from $R\\#H$ to $R^{\prime}\\#H^{\prime}$ by sending $x\otimes h$ to $\psi(x)\otimes\phi(h)$ for any $x\in R,h\in H$. It is easy to check that $\nu$ is an isomorphism of graded Hopf algebras. $\Box$ ###### Theorem 3. If $A$ is a pointed Hopf algebra of Nichols type with coradical $kG$, a group algebra, then there exists a unique ${\rm RSR}(G,r,\overrightarrow{\rho},u)$, up to isomorphism, such that $A\cong kG[kQ_{1}^{c},G,r,\overrightarrow{\rho},u]$ as graded Hopf algebras. Proof. By Lemma 2.1, $A\cong{\mathcal{B}}(V)\\#kG$ as graded Hopf algebras. By Proposition 2.4 (ii), there exists a ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ such that ${\mathcal{B}}(V)\cong{\mathcal{B}}(kQ_{1}^{1},{\rm ad}(G,r,\overrightarrow{\rho},u))$ as graded braided Hopf algebras in ${}^{kG}_{kG}\\!{\mathcal{Y}D}$. Thus $\displaystyle kG[kQ_{1}^{c},G,r,\overrightarrow{\rho},u]$ $\displaystyle\cong$ $\displaystyle{\mathcal{B}}(kQ_{1}^{1},{\rm ad}(G,r,\overrightarrow{\rho},u))\\#kG\ \ (\hbox{by Lemma \ref{2.2}})$ $\displaystyle\cong$ $\displaystyle{\mathcal{B}}(V)\\#kG\ \ (\hbox{by Lemma \ref{2.5}})$ $\displaystyle\cong$ $\displaystyle A.$ The uniqueness follows from [ZZC, Theorem 3]. $\Box$ Considering Theorem 3 and Lemma 2.2, we have that $A$ is a pointed Hopf algebra of Nichols type if and only if $A$ is a pointed Hopf algebra of type one. ## 3 Classification of Nichols Algebras ###### Theorem 4. Let $(G,r,\overrightarrow{\rho},u)$ and $(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$ be two RSR’s. Then the following statements are equivalent: (i) ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$. (ii) There exists a Hopf algebra isomorphism $\phi:kG\rightarrow kG^{\prime}$ such that $(kQ_{1}^{1},ad(G,r,\overrightarrow{\rho},u))$ $\cong\ $ ${}_{\phi}^{\phi^{-1}}(kQ_{1}^{\prime}{}^{1},ad(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime}))$ as Yetter-Drinfeld $kG$-modules. (iii) There is a Hopf algebra isomorphism $\phi:kG\rightarrow kG^{\prime}$ such that ${\mathcal{B}}(kQ_{1}^{1},{\rm ad}(G,r,\overrightarrow{\rho},u))$ $\cong\ $ ${}_{\phi}^{\phi^{-1}}{\mathcal{B}}(kQ_{1}^{\prime}{}^{1},{\rm ad}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime}))$ as graded braided Hopf algebra in ${}^{kG}_{kG}{\mathcal{Y}D}$. Proof. ${\rm(i)}\Rightarrow{\rm(ii)}$. See ${}_{\phi}^{\phi^{-1}}(kQ_{1}^{\prime}{}^{1},ad(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime}))$ $\displaystyle=$ ${}_{\phi}^{\phi^{-1}}U(kQ_{1}^{\prime}{}^{c},G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})\ \ \ (\hbox{ by the remark before Lemma \ref{2.3}})$ $\displaystyle\cong$ $\displaystyle U(_{\phi}^{\phi^{-1}}(kQ_{1}^{\prime}{}^{c},G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})_{\phi}^{\phi^{-1}})\ \ \ (\hbox{ by Lemma \ref{2.3}})$ $\displaystyle\cong$ $\displaystyle U((kQ_{1}^{c},G,r,\overrightarrow{\rho},u))\ \ \ (\hbox{ by Theorem \ref{1}})$ $\displaystyle=$ $\displaystyle(kQ_{1}^{1},ad(G,r,\overrightarrow{\rho},u)).$ ${\rm(ii)}\Rightarrow{\rm(iii)}$. By [ZZC, Lemma 2.7], ${\mathcal{B}}(kQ_{1}^{1},{\rm ad}(G,r,\overrightarrow{\rho},u))$ $\cong$ ${\mathcal{B}}(^{\phi^{-1}}_{\phi}({kQ_{1}^{\prime}}^{1},{\rm ad}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})))$ $\cong$ $\ {}^{\phi^{-1}}_{\phi}\\!{\mathcal{B}}({kQ_{1}^{\prime}}^{1},{\rm ad}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime}))$ as graded braided Hopf algebras in ${}^{kG}_{kG}{\mathcal{Y}D}$. ${\rm(iii)}\Rightarrow{\rm(i)}$. ${\mathcal{B}}(kQ_{1}^{1})\\#kG\cong{\mathcal{B}}(kQ_{1}^{\prime}{}^{1})\\#kG^{\prime}$ as graded Hopf algebras by Lemma 2.5. Thus $kG[kQ_{1}^{c}]\cong{\mathcal{B}}(kQ_{1}^{1})\\#kG\cong{\mathcal{B}}(kQ_{1}^{\prime}{}^{1})\\#kG^{\prime}\cong kG^{\prime}[kQ^{\prime}_{1}{}^{c}]$ as graded Hopf algebras by Lemma 2.2. Now (i) follows from Theorem 2. $\Box$ Up to now we have classified all Nichols algebras by means of RSR’s. In other words, ramification systems with irreducible representations uniquely determine their corresponding Nichols algebras up to pull-push graded braided Hopf algebra isomorphisms. ## 4 The classification of RSR over symmetric groups Let ${\rm ad}_{h}^{-}$ and ${\rm ad}_{h}^{+}$ denote the left and right adjoint actions, respectively. That is, ${\rm ad}_{h}^{-}(x):=hxh^{-1}$ for any $x\in G.$ Let $\gamma_{C}=\mid\\!Z_{s}\\!\mid$ when $s\in C\in{\mathcal{K}}(G)$. Aut$G$ and Inn$G$ denote the automorphism group and inner automorphism group of $G.$ ###### Definition 4.1. Let ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$ be two ${\rm RSR}$’s. If there exists a bijective map $\phi_{C}$ from $I_{C}(r,u)$ to $I_{C}(r^{\prime},u^{\prime})$ such that $\rho^{(i)}_{C}\cong\rho^{\prime}{}^{(\phi_{C}(i))}_{C}$ for any $i\in I_{C}(r,u)$ and $C\in{\mathcal{K}}_{r}(G)$ with $G=G^{\prime}$, $r=r^{\prime}$ and $u=u^{\prime}$, then ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$ are said to be of the same type. Furthermore, if let $\widehat{Z_{u(C)}}=\\{\xi_{u(C)}^{(i)}\mid i=1,2,\cdots,\gamma_{C}\\}$ and $n_{C}^{(i)}:=$ $\mid\\{j\mid\rho_{C}^{(j)}\cong\xi_{u(C)}^{(i)}\\}\\!\mid$ for any $C\in\mathcal{K}_{r}(G)$ and $1\leq i\leq\gamma_{u(C)}\ $, then $\\{(n_{C}^{(1)},n_{C}^{(2)},\cdots,n_{C}^{(\gamma_{C})})\\}_{C\in\mathcal{K}_{r}(G)}$ is called the type of ${\rm RSR}(G,r,\overrightarrow{\rho},$ $u)$. ###### Lemma 4.2. If $\mathrm{Aut}G=\mathrm{Inn}G$, for example, $G=\mathbb{S}_{n}$ with $n\not=6$, then ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u)$ are isomorphic if and only if they have the same type. Proof. By [Zh82, Theorem 1.12.7] or [ZWW, Proposition 1.1 (ii)], $\mathrm{Aut}G=\mathrm{Inn}G$ when $G=\mathbb{S}_{n}$ with $n\not=6$. Let ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$ denote ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u)$ with $G=G^{\prime}$, $r=r^{\prime}$ and $u=u^{\prime}$ for convenience. If ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u)$ have the same type, then there exists a bijective map $\phi_{C}$ from $I_{C}(r,u)$ to $I_{\phi(C)}(r^{\prime},u^{\prime})$ such that $\rho^{\prime}{}^{(\phi_{C}(i))}_{C}\cong\rho^{(i)}_{C}$ for any $i\in I_{C}(r,u)$, $C\in{\mathcal{K}}_{r}(G)$. Therefore they are isomorphic. Conversely, if ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u)$ are isomorphic, then there exist a $\phi\in{\rm Aut}(G)$, $h_{C}\in G$ and a bijective map $\phi_{C}:I_{C}(r,u)\rightarrow I_{\phi(C)}(r^{\prime},u^{\prime})$ such that $u^{\prime}(\phi(C))=\phi{\rm ad}^{+}_{h_{C}}(u(C))$ and $\rho^{\prime}{}_{\phi(C)}^{(\phi_{C}(i))}\phi{\rm ad}^{+}_{h_{C}}\cong\rho_{C}^{(i)}$ for any $C\in{\mathcal{K}}_{r}(G)$ and $i\in I_{C}(r,u)$. Since $\phi{\rm ad}_{h_{C}}^{+}\in{\rm Aut}G$ $={\rm Inn}G$, there exists a $g_{C}\in G$ such that $\phi{\rm ad}_{h_{C}}^{+}={\rm ad}_{g_{C}}^{+}$. Therefore $u(C)={\rm ad}^{+}_{g_{C}}(u(C))$ and $\rho^{\prime}{}_{C}^{(\phi_{C}(i))}{\rm ad}^{+}_{g_{C}}\cong\rho_{C}^{(i)}$. That is, $g_{C}\in Z_{u(C)}$ and $\chi^{\prime}{}_{C}^{(\phi_{C}(i))}{\rm ad}^{+}_{g_{C}}(h)$ $=\chi^{\prime}{}_{C}^{(\phi_{C}(i))}(g_{C}^{-1}hg_{C})$ $=\chi^{\prime}{}_{C}^{(\phi_{C}(i))}(h)=\chi_{C}^{(i)}(h)$ for any $h\in Z_{u(C)}$, where $\chi^{\prime}{}_{C}^{(\phi_{C}(i))}$ and $\chi_{C}^{(i)}$ denote the characters of $\rho^{\prime}{}_{C}^{(\phi_{C}(i))}$ and $\rho_{C}^{(i)}$, respectively. Consequently, $\chi^{\prime}{}_{C}^{(\phi_{C}(i))}=\chi_{C}^{(i)}$ and $\rho^{\prime}{}_{C}^{(\phi_{C}(i))}\cong\rho_{C}^{(i)}$. This implies that ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ and ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u)$ have the same type. $\Box$ For a given ramification $r$ of $G$, let $\Omega(G,r)$ be the set of all RSR’s of G with the ramification $r$, namely, $\Omega(G,r):=\\{(G,r,\overrightarrow{\rho},u)\mid(G,r,\overrightarrow{\rho},u)\mbox{\ is\ an }{\rm RSR}\\}$. Let ${\mathcal{N}}(G,r)$ be the number of isomorphic classes in $\Omega(G,r)$. ###### Theorem 5. Given a group $G$ and a ramification $r$ of $G$. Assume $\mathrm{Aut}G=\mathrm{Inn}G$, for example, $G=\mathbb{S}_{n}$ with $n\not=6$. Let $u_{0}$ be a fixed map from ${\mathcal{K}}(G)\rightarrow G$ with $u_{0}(C)\in C$ for any $C\in{\mathcal{K}}(G)$. Let $\bar{\Omega}(G,r,u_{0})$ denote the set consisting of all elements with distinct type in $\\{(G,r,\overrightarrow{\rho},u_{0})\mid(G,r,\overrightarrow{\rho},u_{0})$ is an RSR $\\}$. Then $\bar{\Omega}(G,r,u_{0})$ becomes the representative system of isomorphic classes in $\Omega(G,r).$ Proof. For any ${\rm RSR}(G,r,\overrightarrow{\rho},u)$, there exists ${\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u_{0})$ such that ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong{\rm RSR}(G,r,\overrightarrow{\rho^{\prime}},u_{0})$ by Proposition 1.5. Using Lemma 4.2, we complete the proof. $\Box$ ###### Corollary 4.3. $\displaystyle{\mathcal{N}}(G,r)$ $\displaystyle=$ $\displaystyle\prod\limits_{C\in\mathcal{K}_{r}(G)}\tau_{C},$ where $\tau_{C}=$ the number of elements in set $\\{(n_{1},n_{2},\cdots,n_{\gamma_{C}})\in{\mathbb{N}}^{\gamma_{C}}\mid n_{1}{\rm deg}\xi_{1}+n_{2}{\rm deg}\xi_{2}+\cdots+n_{\gamma_{C}}{\rm deg}\xi_{\gamma_{C}}=r_{C}\\}$ for any $C\in{\mathcal{K}}_{r}(G)$. Remark: For a given finite Hopf quiver $(Q,G,r)$ over symmetric group $G={\mathbb{S}}_{n}$ with $n\not=6$, every path algebra $T_{(kG)^{*}}(kQ_{1}^{a})$ over quiver $(Q,G,r)$ exactly admits ${\mathcal{N}}(G,r)$ non-isomorphic graded Hopf algebra structures; every path coalgebra $T_{kG}^{c}(kQ_{1}^{c})$ over quiver $(Q,G,r)$ exactly admits ${\mathcal{N}}(G,r)$ non-isomorphic graded Hopf algebra structures. ## 5 Appendix We now consider the dual case of Theorem 4. If $Q$ is finite, then $(kQ_{1}^{a},G,r,\overrightarrow{\rho},u)$ is a $(kG)^{*}$-Hopf bimodule with comodule operations $\delta^{-}$ and $\delta^{+}$. Define a new left $(kG)^{*}$-coaction on $kQ_{1}^{a}$ given by $\delta^{-}_{\rm coad}(x):=\sum_{x}x_{(-1)}S(x_{(0)(1)})\otimes x_{(0)(0)},\ \ \ \hbox{ for any }x\in kQ_{1}^{a},$ i.e. adjoint coaction. With this left $(kG)^{*}$-coaction and the original left (arrow) $(kG)^{*}$-action $\alpha^{-}$, $kQ_{1}^{a}$ is a Yetter-Drinfeld $(kG)^{*}$-module. Let $kQ^{1a}_{1}$ denote the subspace spanned by $Q_{1}^{1}$ in $kQ_{1}^{a}$. It is clear that $\xi_{kQ_{1}^{c}}(kQ_{1}^{1})=(kQ_{1}^{1a})^{*}$, where $\xi_{kQ_{1}^{c}}$ was defined in [ZZC, Lemma 1.7]. Thus $kQ_{1}^{1a}$ is a Yetter-Drinfeld $(kG)^{*}$-submodule of $kQ_{1}^{a}$, denoted by $(kQ_{1}^{1a},$ ${\rm coad}(G,r,\overrightarrow{\rho},u))$, which is isomorphic to the dual of $(kQ_{1}^{1},ad(G,r,$ $\overrightarrow{\rho},u))$ as Yetter-Drinfeld $(kG)^{*}$-modules. Therefore we have the dual case of Theorem 4. ###### Proposition 5.1. Let $(G,r,\overrightarrow{\rho},u)$ and $(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$ be two RSR’s. Then the following statements are equivalent: (i) ${\rm RSR}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${\rm RSR}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime})$. (ii) There exists a Hopf algebra isomorphism $\phi:(kG)^{*}\rightarrow(kG^{\prime})^{*}$ such that $(kQ_{1}^{1a},$ ${\rm coad}(G,r,\overrightarrow{\rho},u))\cong\ _{\phi}^{\phi^{-1}}(kQ_{1}^{\prime}{}^{1a},{\rm coad}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime}))$ as Yetter-Drinfeld $(kG)^{*}$-modules. (iii) There is a Hopf algebra isomorphism $\phi:(kG)^{*}\rightarrow(kG^{\prime})^{*}$ such that ${\mathcal{B}}(kQ_{1}^{1a},$ ${\rm coad}(G,r,\overrightarrow{\rho},u)$ $\cong$ ${}_{\phi}^{\phi^{-1}}{\mathcal{B}}(kQ_{1}^{\prime}{}^{1a},{\rm coad}(G^{\prime},r^{\prime},\overrightarrow{\rho^{\prime}},u^{\prime}))$ as graded braided Hopf algebras in ${}^{(kG)^{*}}_{(kG)^{*}}{\mathcal{Y}D}$. Proof. Obviously, (ii) and Theorem 4 (ii) are equivalent; (iii) and Theorem 4 (iii) are equivalent. $\Box$ Acknowledgement: We would like to thank Prof. N. Andruskiewitsch for his help. The first author and the third author were financially supported by the Australian Research Council, and the second author was supported by NSF of China (10771183) and Sino-German project (GZ310). S.C.Z thanks the Department of Mathematics, University of Queensland for hospitality. ## References * [AS98] N. Andruskiewitsch and H. J. Schneider, Lifting of quantum linear spaces and pointed Hopf algebras of order $p^{3}$, J. Alg. 209 (1998), 645–691. * [AS02] N. Andruskiewitsch and H. J. Schneider, Pointed Hopf algebras, new directions in Hopf algebras, edited by S. Montgomery and H.J. Schneider, Cambradge University Press, 2002. * [AS00] N. Andruskiewitsch and H. J. Schneider, Finite quantum groups and Cartan matrices, Adv. Math. 154 (2000), 1–45. * [AS05] N. Andruskiewitsch and H. J. Schneider, On the classification of finite-dimensional pointed Hopf algebras, preprint math.QA/0502157. * [AZ07] N. Andruskiewitsch and Shouchuan Zhang, On pointed Hopf algebras associated to some conjugacy classes in $S_{n}$, Proc. Amer. Math. Soc. 135 (2007), 2723-2731. * [ARS] M. Auslander, I. Reiten and S. O. Smal$\phi,$ Representation theory of Artin algebras, Cambridge University Press, Cambridge, 1995\. * [BD] Y. Bespalov and B. Drabant, Hopf (Bi-)Modules and Crossed Modules in Braided Monoidal Categories, J. Pure and Appl. Alg. 123 (1998), 105–129. * [CR02] C. Cibils and M. Rosso, Hopf quivers, J. Alg. 254 (2002), 241-251. * [CR97] C. Cibils and M. Rosso, Algebres des chemins quantiques, Adv. Math. 125 (1997), 171–199. * [DPR] R. Dijkgraaf, V. Pasquier and P. Roche, Quasi Hopf algebras, group cohomology and orbifold models, Nuclear Phys. B Proc. Suppl. 18B (1991), pp. 60–72. * [ENO] P. Etingof, D. Nikshych and V. Ostrik, On fusion categories, Ann. Math. 162, 581–642 (2005). * [EO] P. Etingof and V. Ostrik, Finite tensor categories, Mosc. Math. J. 4 (2004), no. 3, 627–654, 782–783. Also see math.QA/0301027. * [Ga] Matthias R. Gaberdiel, An algebraic approach to logarithmic conformal field theory, Int.J.Mod.Phys. A18 (2003) 4593-4638. Also see hep-th/0111260. * [H1] I. Heckenberger, The Weyl groupoid of a Nichols algebra of diagonal type, Inventiones Math. 164, 175–188 (2006). Also see math.QA/0411477. * [H2] I. Heckenberger, Classification of arithmetic root systems, preprint math.QA/0605795. * [N] W. Nichols, Bialgebras of type one, Comm. Alg. 6 (1978), 1521–1552. * [OZ] F. Van Oystaeyen and P. Zhang, Quiver Hopf algebras, J. Alg. 280 (2004), 577–589. * [Ra] D. E. Radford, The structure of Hopf algebras with a projection, J. Alg. 92 (1985), 322–347. * [Ri] C. M. Ringel, Tame algebras and integral quadratic forms, Springer-Verlag, Berlin, 1984. * [W] S. L. Woronowicz, Differential calculus on compactmatrix pseudogroups (quantum groups), Comm. Math. Phys. 122 (1989), 125-170. * [ZZWC] Shouchuan Zhang, Y-Z Zhang, Peng Wang, Jing Cheng, On Pointed Hopf Algebras with Weyl Groups of Exceptional Type, Preprint arXiv:0804.2602. * [ZZC] Shouchuan Zhang, Y-Z Zhang and H. X. Chen, Classification of PM Quiver Hopf Algebras, J. Alg. and Its Appl. 6 (2007)(6), 919-950. Also see in math. QA/0410150. * [ZWW] Shouchuan Zhang, Min Wu and Hengtai Wang, Classification of Ramification Systems for Symmetric Groups, Acta Math. Sinica, 51 (2008) 2, 253–264. Also in math.QA/0612508. * [ZZ] Shouchuan Zhang and Y-Z Zhang, Structures and Representations of Generalized Path Algebras, Algebras and Representation Theory, 10(2) (2007), 117-134. * [Zh82] Yuan-Da Zhang, The Structure of Finite Groups (I), Science Press, Beijing, 1982.

arxiv-papers

2024-09-04T02:48:53.995142

0802.3524

# Gravitational Radiation from Primordial Helical MHD Turbulence Tina Kahniashvili CCPP, New York University, 4 Washington Plaza, New York, NY 10003, USA Department of Physics, Kansas State University, 116 Cardwell Hall, Manhattan, KS 66506, USA Department of Physics, Laurentian University, Ramsey Lake Road, Sudbury, ON P3E 2C6, Canada National Abastumani Astrophysical Observatory, 2A Kazbegi Ave, Tbilisi, GE-0160, Georgia Grigol Gogoberidze Centre for Plasma Astrophysics, K.U. Leuven, Celestijnenlaan 200B, 3001 Leuven, Belgium Department of Physics, Kansas State University, 116 Cardwell Hall, Manhattan, KS 66506, USA National Abastumani Astrophysical Observatory, 2A Kazbegi Ave, Tbilisi, GE-0160, Georgia Bharat Ratra Department of Physics, Kansas State University, 116 Cardwell Hall, Manhattan, KS 66506, USA ( KSUPT-08/1) ###### Abstract We consider gravitational waves (GWs) generated by primordial inverse-cascade helical magneto-hydrodynamical (MHD) turbulence produced by bubble collisions at the electroweak phase transitions (EWPT). Compared to the unmagnetized EWPT case, the spectrum of MHD-turbulence-generated GWs peaks at lower frequency with larger amplitude and can be detected by the proposed Laser Interferometer Space Antenna (LISA). ###### pacs: 98.70.Vc, 98.80.-k When detected, primordial cosmological GWs will provide a very valuable probe of the very early Universe sources . Various mechanisms that generate such GWs have been discussed: quantum fluctuations inflation ; bubble wall motion and collisions during phase transitions bubble ; kos1 ; cosmic strings strings ; cosmological magnetic fields magnet ; cdk04 ; cd06 ; and plasma turbulence kmk02 ; dolgov ; kgr05 ; gkk07 . From the direct detection point of view, GWs generated during the EWPT are promising since their peak frequency lies in or near the LISA lisa frequency band m00 , however, to produce a detectable signal the EWPT must be strong enough detection ; gs06 ; grojean . Currently discussed EWPT models do not predict an observable GW signal from bubble collisions grojean , nor for GWs produced by unmagnetized turbulence gkk07 . Here we study the generation of GWs during a first-order EWPT assuming that bubble collisions produce helical MHD turbulence.111Kinetic or magnetic helicity generation at the EWPT is studied in Refs. helicity . Previously we studied generation of GWs by direct-cascade turbulence and found that, due to parity violation in the early Universe, the induced GWs are circularly polarized kgr05 . Polarized GWs are present in other models gw-pol , and the polarization of the GW background is in principle observable, either directly seto or through the CMB cdk04 ; a . In the case of unmagnetized hydrodynamical turbulence the peak frequency of the GW power spectrum is determined by the inverse turn-over time of the largest eddy and the energy- scale when the GW is generated. Recently discussed modifications of the standard EWPT model place the transition at a higher energy-scale gs06 . As a result, the GW power spectrum peak frequency is shifted to higher frequency which, since the GW spectrum is sharply peaked, reduces the possibility of detection by LISA. On the other hand, in the case of MHD turbulence the presence of an energy inverse-cascade leads to an increase in the effective size of the largest eddy (now associated with an helical magnetic field), and can result in the GW power spectrum peaking in the LISA band, with amplitude large enough to be detected by LISA. We adapt the technique developed in Ref. gkk07 to study this case here. We model MHD turbulence and obtain the GW spectrum by using an analogy with the theory of sound wave production by hydrodynamical turbulence L52 ; P52 ; G ; my75 . Since the turbulent fluctuations are stochastic, so are the generated GWs. The GW energy density is m00 $\displaystyle\rho_{\rm GW}({\bf x})$ $\displaystyle=$ $\displaystyle\frac{1}{32\pi G}\langle\partial_{t}h_{ij}({\mathbf{x}},t)\partial_{t}h_{ij}({\mathbf{x}},t)\rangle$ $\displaystyle=$ $\displaystyle\frac{G}{2\pi}\int{\rm d}^{3}{\bf x}^{\prime}{\rm d}^{3}{\bf x}^{\prime\prime}\frac{\langle\partial_{t}S_{ij}({\mathbf{x}}^{\prime},t^{\prime})\partial_{t}S_{ij}({\mathbf{x}}^{\prime\prime},t^{\prime\prime})\rangle}{|{\bf x}-{\bf x}^{\prime}||{\bf x}-{\bf x}^{\prime\prime}|}.$ Here the times $t^{\prime(\prime\prime)}=t-|{\bf x}-{\bf x}^{\prime(\prime\prime)}|$, $i$ and $j$ are spatial indices (repeated indices are summed), the source $S_{ij}({\mathbf{x}},t)=T_{ij}({\mathbf{x}},t)-\delta_{ij}T^{k}_{k}({\mathbf{x}},t)/3$ is the traceless part of the stress-energy tensor $T_{ij}$, $G$ is the gravitational constant, and we use natural units with $\hbar=1=c$. We assume that the turbulence exists for a time short enough to neglect the cosmological expansion during GW production. We consider metric perturbations in the far- field limit (i.e. for $x\gg d$, where $d$ is a characteristic length-scale of the source region), where GWs are the only metric perturbations W , and replace $|{\bf x}-{\bf x}^{\prime}|$ by $|\bf x|$ in Eq. (Gravitational Radiation from Primordial Helical MHD Turbulence). If the turbulence is stationary then the GW spectral energy density $I({\bf x},\omega)$ ($\rho_{\rm GW}({\bf x})=\int{\rm d}\omega I({\bf x},\omega)$ where $\omega$ is the angular frequency) is gkk07 $I({\bf x},\omega)=\frac{4\pi^{2}\omega^{2}G{\rm w}^{2}}{|{\bf x}|^{2}}\int{\rm d}^{3}{\bf x}^{\prime}H_{ijij}\left({\bf x}^{\prime},\frac{{\bf x}}{|{\bf x}|}\omega,\omega\right).$ (2) Here $H_{ijij}({\bf x}^{\prime},{\bf k},\omega)$ (where ${\bf k}$ is a proper wavevector) is the (double traced) four-dimensional Fourier transform of the two-point time-delayed fourth-order correlation tensor, $\langle S_{ij}({\bf x}^{\prime},t)S_{lm}({\bf x}^{\prime\prime},t+\tau)\rangle/{\rm w}^{2}$, with respect to ${\bf x}^{\prime\prime}-{\bf x}^{\prime}$ and $\tau$, where ${\rm w}=\rho+p$ is the enthalpy density and $p$ and $\rho$ the pressure and energy density of the plasma. We assume that primordial MHD turbulence is generated at time $t_{\star}$ at proper length-scale $l_{0}=2\pi/k_{0}$ with characteristic velocity perturbation $v_{0}$.222We assume that the usual and magnetic Reynolds numbers are much greater than unity on scales $\sim l_{0}$, otherwise there is no turbulence. Throughout this paper the symbol $\sim$ represents equality to the accuracy of a dimensionless multiplicative constant of order unity. The dynamics of MHD turbulence is dominated by Alfvén waves for which the magnetic and kinetic energy densities are in approximate equipartition B03 . In this case $v_{0}\sim b_{0}$, where $b_{0}=B_{0}/\sqrt{4\pi{\rm w}}$ is the characteristic magnetic field perturbation expressed in velocity units. While MHD turbulence is isotropic on large scales, it is locally anisotropic on small scales s83 , resulting in small-scale anisotropy in the generated GW background. However, GWs are generated mainly by the largest eddies gkk07 so we adopt an isotropic turbulence model, and thus the magnetic field two-point correlation function is $\langle b_{i}^{\star}({\bf k},t)b_{j}({\bf k^{\prime}},t+\tau)\rangle=F_{ij}^{M}\\!({\bf k},t)f(\eta(k),\tau)\delta({\bf k}-{\bf k^{\prime}})$, with my75 $F_{ij}^{M}\\!({\bf k},\tau)=P_{ij}({\bf k})\frac{E^{M}\\!(k,t)}{4\pi k^{2}}+i\varepsilon_{ijl}{k_{l}}\frac{H^{M}\\!(k,t)}{8\pi k^{2}}.$ (3) Here $P_{ij}({\bf k})=\delta_{ij}-{k_{i}k_{j}}/{k^{2}}$, and $E^{M}\\!(k,t)$ and $H^{M}\\!(k,t)$ are the magnetic field energy and helicity densities. The Schawarz inequality implies $|H^{M}\\!(k,t)|\leq 2E^{M}\\!(k,t)/k$ B03 . For the total magnetic energy ${\mathcal{E}}_{M}(t)=\int E^{M}\\!(k,t)dk$ and helicity ${\mathcal{H}}_{M}(t)=\int H^{M}\\!(k,t)dk$ we get ${\mathcal{H}}_{M}(t)\leq 2\xi_{M}(t){\mathcal{E}}_{M}(t)$, where $\xi_{M}(t)\equiv\int E^{M}(k,t)k^{-1}dk/{\mathcal{E}}_{M}(t)$ is the magnetic-eddy correlation length. $\eta(k)$ is an autocorrelation function that determines the characteristic function $f(\eta(k),\tau)$ that describes the temporal decorrelation of turbulent fluctuations. In the following we use $f(\eta(k),\tau)=\exp\left(-{\pi}\eta^{2}(k)\tau^{2}/4\right)$ K64 . After generation primordial turbulence freely decays. We adopt the decaying MHD turbulence model of Refs. BM99 ; CHB05 . For non-zero initial magnetic helicity turbulence decay is a two stage process. First decay stage dynamics is governed by a direct cascade of energy density lasting for a time $\tau_{s0}=s_{0}\tau_{0}$, several times ($s_{0}\sim 3-5$) longer then the characteristic largest-eddy turn-over time $\tau_{0}=l_{0}/v_{0}=2\pi/k_{0}v_{0}$. During the first stage energy density flows from large to small scales and finally dissipates on scales $\sim l_{d}=2\pi/k_{d}$ ($k_{d}\gg k_{0}$) where one of the Reynolds numbers becomes of order unity. Due to the selective decay effect B03 magnetic helicity is nearly conserved during this stage CHB05 . To compute the GWs generated by decaying MHD turbulence, we assume that decaying turbulence lasting for time $\tau_{s0}$ is equivalent to stationary turbulence lasting for time $\tau_{s0}/2$. This can be justified using the Proudman P52 ; my75 argument for (unmagnetized) hydrodynamical turbulence. Consequently, when computing the emitted GWs we ignore the time dependence of $E^{M}(k,t)$ and $H^{M}(k,t)$. We also assume small initial magnetic helicity, $\alpha_{\star}\equiv{\mathcal{H}}_{M}(t_{\star})/[2\xi_{M}(t_{\star}){\mathcal{E}}_{M}(t_{\star})]\ll 1$. For $E^{M}(k,t)$ and $\eta(k)$ we use the Kolmogorov model, $E^{M}\\!(k,t)\sim\varepsilon^{2/3}k^{-5/3},~{}~{}~{}\eta(k)=\varepsilon^{1/3}k^{2/3}/\sqrt{2\pi},$ (4) for $k_{0}<k<k_{d}$. Here $\varepsilon\sim k_{0}v_{0}^{3}$ is the energy dissipation rate per unit enthalpy. At the end of the first stage turbulence relaxes to a maximally helical state, $\alpha_{s0}\sim 1$ CHB05 ; jedamzik . Accounting for conservation of magnetic helicity, the characteristic velocity and magnetic field perturbations at this stage are $v_{1}\sim\alpha_{\star}^{1/2}v_{0}$ and $b_{1}\sim\alpha_{\star}^{1/2}b_{0}$. Second stage dynamics is governed by a magnetic helicity inverse cascade. If both Reynolds numbers are large at the end of the first stage, magnetic helicity is conserved during the second stage. The magnetic eddy correlation length evolves as $\xi_{M}(t)\sim l_{0}\sqrt{1+t/\tau_{1}}$ BM99 ; CHB05 where $\tau_{1}\sim l_{0}/v_{1}=\tau_{0}/\sqrt{\alpha_{\star}}$ is the characteristic energy containing eddy turn-over time at the beginning of the second stage. The magnetic ${\mathcal{E}}_{M}(t)$ and kinetic ${\mathcal{E}}_{K}(t)$ energy densities evolve as BM99 ; CHB05 $\displaystyle{\mathcal{E}}_{M}(t)$ $\displaystyle\propto$ $\displaystyle(1+t/\tau_{1})^{-1/2}{\rm w}b_{1}^{2},$ $\displaystyle{\mathcal{E}}_{K}(t)$ $\displaystyle\propto$ $\displaystyle(1+t/\tau_{1})^{-1}{\rm w}v_{1}^{2}.$ (5) These imply that the characteristic turn-over ($\tau_{\rm to}$) and cascade ($\tau_{\rm cas}$) timescales evolve as $\displaystyle\tau_{\rm to}\sim\tau_{\rm cas}\sim\tau_{1}(1+t/\tau_{1}).$ (6) To compute the GWs emitted during the second stage we use the stationary turbulence model that has the same GW output. Introducing the characteristic wavenumber $k_{\xi}(t)=2\pi/\xi_{M}(t)$ and using Eqs. (5) we find ${\mathcal{E}}_{M}\sim{\rm w}v_{1}^{2}k_{\xi}(t)/k_{0}$ and ${\mathcal{E}}_{K}\sim{\rm w}v_{1}^{2}[k_{\xi}(t)/k_{0}]^{2}$ since $b_{1}\sim v_{1}$. The time when turbulence is present on scale $\xi_{M}(t)$ is determined by Eq. (6) which can be rewritten as $\tau_{\rm cas}\sim\tau_{1}[k_{0}/k_{\xi}(t)]^{2}$. So instead of considering decaying turbulence, we consider stationary turbulence with a scale-dependent duration time (time during which the magnetic energy is present at the scale), $\tau_{s1}\sim\tau_{1}[k_{0}/k]^{2}$ (for $k=k_{\xi}$ this coincides with $\tau_{\rm cas}$). The expression for ${\mathcal{E}}_{M}$ yields the time-independent $E^{M}\\!(k,t)=C_{1}v_{1}^{2}/k_{0}=kH^{M}\\!(k,t)/2,~{}~{}~{}k_{S}<k<k_{0}.$ (7) Here $C_{1}$ is a constant of order unity, $k_{S}$ is the smallest wavenumber where the inverse cascade stops, and the second equation follows from saturating the causality condition. For the second stage autocorrelation function, which is inversely proportional to the turn-over time (6), we assume $\eta(k)=(k/k_{0})^{2}/\sqrt{2\pi}\tau_{1}$. At the largest scales there is no efficient dissipation mechanism, so the inverse cascade will be stopped at scale $l_{S}(t)=2\pi/k_{S}$ where either the cascade timescale $\tau_{\rm cas}$ reaches the expansion timescale $H^{-1}_{\star}=H^{-1}(t_{\star})$, or when the characteristic length scale $\xi_{M}(t)\sim l_{S}$ reaches the Hubble radius. These conditions are $\alpha_{\star}^{-1/2}l_{S}^{2}/v_{0}l_{0}\leq H^{-1}_{\star}$ or $l_{S}\leq H^{-1}_{\star}$ (the cascade time is scale dependent and maximal at $k=k_{S}$). Defining $\gamma=l_{0}/H^{-1}_{\star}$ ( $\gamma\leq 1$), it is easy to see that the first condition is fulfilled first and consequently $k_{0}/k_{S}\leq(v_{0}/\gamma)^{1/2}\alpha_{\star}^{1/4}$. To have an inverse cascade requires $k_{0}/k_{S}\geq 1$, leading to a constraint on initial helicity, $\gamma\leq M\alpha_{\star}^{1/2}$ (where $M=v_{0}$ is the turbulence Mach number). The magnetic field perturbation stress-energy tensor is $T_{ij}^{M}({\mathbf{x}},t)={\rm w}b_{i}({\mathbf{x}},t)b_{j}({\mathbf{x}},t)$. For the first decay stage we compute for this magnetic part and then double the result to account for approximate magnetic and kinetic energy equipartition for Alfvén waves. During the second stage, according to Eqs. (5), kinetic energy can be neglected compared to magnetic energy. To compute $H_{ijij}({\bf k},\omega)$ we assume Millionshchikov quasi-normality my75 and use the convolution theorem (for details see Sec. III of Ref. gkk07 ). Using the ($k\rightarrow 0$) aero- acoustic approximation, which is accurate for low Mach number ($M\leq 1$, (and slightly overestimates GWs amplitude for the Mach number approaching unity ($M\rightarrow 1$) gkk07 ), we find $\displaystyle\\!\\!\\!\\!\\!\\!\\!H_{ijij}({\bf k},\omega)\approx H_{ijij}(0,\omega)=~{}~{}~{}~{}~{}~{}~{}~{}~{}$ (8) $\displaystyle\\!\\!\\!\\!\\!\\!\\!\\!\frac{7C_{1}^{2}M^{3}\alpha_{\star}^{3/2}\\!}{6\pi^{3/2}k_{0}}\\!\\!\int_{k_{S}}^{k_{0}}\\!\\!\frac{{\rm d}k}{k^{4}}\exp\\!\left(-\frac{\omega^{2}k_{0}^{2}}{\alpha_{\star}M^{2}k^{4}}\right)\\!{\rm erfc}\\!\left(-\frac{\omega k_{0}}{\sqrt{\alpha_{\star}}Mk^{2}}\right).$ The integral is dominated by the contribution of large scale ($k\simeq k_{S}$) perturbations and is maximal at $\omega_{\rm max}^{(II)}\sim\alpha_{\star}^{1/2}Mk_{S}^{2}/k_{0}=2\pi H_{\star}$. For the first-stage direct-cascade turbulence the peak frequency is $\omega_{\rm max}^{(I)}\sim k_{0}M$ gkk07 . To determine the peak frequency at the current epoch we need to account for the cosmological expansion which decreases the GW amplitude and frequency by the factor $a_{\star}/a_{0}$, where $a_{\star}$ and $a_{0}$ are the values of the cosmological scale factor at the GW generation and current epochs The total GW energy spectrum at a given space-time event is obtained by integrating over all source regions with a light-like separation from that event, and includes contributions from GW generated during the first and second stages. For the first stage (with duration time $\tau^{(I)}_{T}=s_{0}\tau_{0}$) $\rho_{GW}^{(I)}(\omega)$ is given by Eqs. (21) and (A3) of Ref. gkk07 . For the second stage contribution we must account for the scale dependence of the cascade time. The total GW fractional energy density parameter at the moment of emission $\Omega_{{\rm GW},\star}$ is $105~{}H_{\star}^{4}\omega^{3}\sum_{m}\tau_{T}^{(m)}H_{ijij}^{(m)}(0,\omega_{\star})/H_{0}^{2}$ gkk07 . Here the index $m$ runs over $I$ and $II$ for the first and second decay stages, $H_{0}$ $\omega_{\star}$ is an angular frequency at the moment of emission. The current GW amplitude is related to the current fractional energy density parameter through $h_{C}(f)=1.26\times 10^{-18}\left({1\,{\rm Hz}}/{f}\right)\left[h_{0}^{2}\Omega_{GW}(f)\right]^{1/2}$ (where $h_{0}$ is the current Hubble parameter $H_{0}$ in units of 100 ${\rm km}\,{\rm sec}^{-1}{\rm Mpc}^{-1}$) m00 , and $\displaystyle h_{C}(f)$ $\displaystyle\simeq$ $\displaystyle 2\times 10^{-14}\left(\frac{100\,{\rm GeV}}{T_{*}}\right)\left(\frac{100}{g_{*}}\right)^{1/3}$ (9) $\displaystyle\times\sum_{m}\left[\tau_{T}^{(m)}\omega_{\star}H_{\star}^{4}H_{ijij}^{(m)}(0,\omega_{\star})\right]^{1/2}.~{}~{}~{}~{}~{}~{}~{}~{}~{}~{}$ Here the linear frequency $f=(a_{\star}/a_{0})f_{\star}$ with $f_{\star}=\omega_{\star}/2\pi$, $T_{\star}$ and $g_{\star}$ are the temperature and effective number of relativistic (all fields) degrees of freedom at scale factor $a_{\star}$. Figure 1: The spectrum of gravitational radiation from MHD turbulence for $g_{\ast}=100,~{}T_{\ast}=100~{}{\rm GeV},~{}\gamma=0.01$, and $M=1/\sqrt{3}$, for four different initial magnetic helicity values, $\alpha_{\star}=0$ (solid line), $\alpha_{\star}=0.02$ (dashed line), $\alpha_{\star}=0.05$ (dash-dotted line), and $\alpha_{\star}=0.1$ (dotted line). The bold line is the LISA design sensitivity curve. Figure 1 shows $h_{C}(f)$ for a few initial magnetic helicity values. GWs emitted during direct-cascade unmagnetized turbulence peak at current $f_{\rm max}^{(I)}\simeq M\nu_{\star}$ gkk07 . We find that the MHD-inverse-cascade generated GW (current epoch) peak frequency is determined by cosmology parameters, $f_{\rm max}^{(II)}=H_{\star}a_{0}/a_{\star}=1.6\times 10^{-5}\,{\rm Hz}\,\left({g_{*}}/{100}\right)^{1/6}\left({T_{*}}/{100\,{\rm GeV}}\right)$ and is independent of turbulence parameters. On the other hand, $f_{\rm max}^{(II)}=\gamma f_{\rm max}^{(I)}/M$ is shifted to lower frequency compared to the unmagnetized case. From Eq. (9), the amplitude of MHD- turbulence-generated GWs at the peak is a factor $\sim\alpha_{\star}^{9/8}\gamma^{-3/4}M^{3/4}$ larger than that in the unmagnetized case. When modeling turbulence we used the Biskamp and Muller model, BM99 ; CHB05 . If we adopt the helical MHD turbulence model of Banerjee and Jedamzik jedamzik (also see Refs. campanelli ) the GW peak frequency remains the same while the amplitude of the signal doubles. Figure 1 shows that even for small values of magnetic helicity the main contribution to the GW energy density is from the second, inverse-cascade stage. The GWs will be strongly polarized since magnetic helicity is maximal at the end of the first stage kgr05 . LISA should be able to detect such GW polarization seto . Unlike the unmagnetized case due to the second (inverse- cascade) stage contribution the GW amplitude is large enough at $10^{-4}$ Hz to be detectable by LISA. If the EWPT occurs at higher energies ($T_{\star}>100$ Gev) the peak is shifted to higher frequency, closer to LISA sensitivity peak, which leads to a stronger signal. Our formalism is applicable for GW production at an earlier QCD phase transition, assuming the presence of colored magnetic fields QCD , or for any other phase transitions bubble ; the peak frequency will be shifted according to the changes in $T_{\star}$ and $g_{\star}$. The GW signal estimated here exceeds that from bubble collisions kos1 ; detection ; gs06 or from hydrodynamical, unmagnetized turbulence kmk02 ; dolgov ; gkk07 . Of course, this strong signal assumes initial non-zero (although small) magnetic helicity, so detection of polarized GWs by LISA will indicate parity violation during the EWPT as proposed in Refs. helicity . We greatly appreciate useful comments from A. Gruzinov and the referees. We acknowledge helpful discussions with R. Durrer, G. Gabadadze, and A. Kosowsky. T. K. acknowledges the hospitality of the Abdus Salam International Center for Theoretical Physics. G. G. and T. K. acknowledge partial support from INTAS 061000017-9258 and Georgian NSF ST06/4-096 grants. B. R. acknowledges US DOE grant DE-FG03-99EP4103. ## References * (1) S. Chongchitnan and G. Efstathiou, Phys. Rev. D 73, 083511 (2006); C. J. Hogan, AIP Conf. Proc. 873, 30 (2006); A. Buonanno, arXiv:0709.4682 [qr-qc]. * (2) L. P. Grishchuk, Sov. Phys. JETP 40, 409 (1975) [Zh. Eksp. Teor. Fiz. 67, 825 (1974)]; A. Starobinsky, Sov. Phys. JETP Lett. 30, 682 (1979) [Pisma Zh. Eksp. Teor. Fiz. 30, 719 (1979)]; V. A. Rubakov, M.V. Sazhin, and A. V. Veryaskin, Phys. Lett. B 115, 189 (1982); B. Allen, Phys. Rev. D 37, 2078 (1988); B. Ratra, Phys. Rev. D 45 1913 (1992); M. Giovannini, Phys. Rev. D 60, 123511 (1999). * (3) E. Witten, Phys. Rev. D 30, 272 (1984); C. J. Hogan, Mon. Not. R. Astron. Soc. 218, 629 (1986); M. S. Turner and F. Wilczek, Phys. Rev. Lett. 65, 3080 (1990); A. Kosowsky, M. S. Turner, and R. Watkins, Phys. Rev. D 45, 4514 (1992). * (4) M. Kamionkowski, A. Kosowsky, and M. S. Turner, Phys. Rev. D 49, 2837 (1994); R. Apreda, et al., Class. Quant. Grav. 18, L155 (2001); J. F. Dufaux, et al., Phys. Rev. D 76, 123517 (2007); S. J. Huber and T. Konstandin, arXiv:0709.2091 [hep-ph]; C. Caprini, R. Durrer, and G. Servant, arXiv:0711.2593 [astro-ph]; R. Easther, J. T. Giblin, and E. A. Lim, arXiv:0712.2991 [astro-ph]; J. Garcia-Bellido, D. G. Figueroa and A. Sastre, Phys. Rev. D 77, 043517 (2008); A. Megevand, arXiv:0804.0391 [astro-ph]. . * (5) T. Vachaspati and A. Vilenkin, Phys. Rev. D 31, 3052 (1985); M. P. Infante and N. Sánchez, Phys. Rev. D 61, 083515 (2000). * (6) D. V. Deriagin, et al., Mon. Not. R. Astron. Soc. 229, 357 (1987); R. Durrer, P. Ferreira, and T. Kahniashvili, Phys. Rev. D 61, 043001 (2000); A. Lewis, Phys. Rev. D. 70, 043011 (2004). * (7) C. Caprini, R. Durrer, and T. Kahniashvili, Phys. Rev. D 69, 063006 (2004). * (8) C. Caprini and R. Durrer, Phys. Rev. D 74, 063521 (2006). * (9) A. Kosowsky, A. Mack, and T. Kahniashvili, Phys. Rev. D 66, 024030 (2002). * (10) A. D. Dolgov, D. Grasso, and A. Nicolis, Phys. Rev. D 66, 103505 (2002). * (11) T. Kahniashvili, G. Gogoberidze, and B. Ratra, Phys. Rev. Lett. 95, 151301 (2005). * (12) G. Gogoberidze, T. Kahniashvili, and A. Kosowsky, Phys. Rev. D 76, 083002 (2007). * (13) http://lisa.nasa.gov/ * (14) M. Maggiore, Phys. Rept. 331, 28 (2000). * (15) R. Apreda, et al., Nucl. Phys. B 631, 342 (2002); A. Nicolis, Class. Quantum Grav. 21, L27 (2004). * (16) C. Grojean and G. Servant, Phys. Rev. D 75 043507 (2007). * (17) C. Delaunay, C. Grojean, and J. D. Wells, arXiv:0711.2511 [hep-ph]. * (18) J. Cornwall, Phys. Rev. D 56, 6146 (1997); M. Giovannini and M. E. Shaposhnikov, Phys. Rev. D 57, 2186 (1998); G. B. Field and S. M. Carroll, Phys. Rev. D 62, 103008 (2000); T. Vachaspati, Phys. Rev. Lett. 87, 251302 (2001); G. Sigl, Phys. Rev. D 66, 123002 (2002); K. Subramanian and A. Brandenburg, Phys. Rev. Lett. 93, 205001 (2004); L. Campanelli and M. Giannotti, Phys. Rev. D 72, 123001 (2005); V. B. Semikoz and D. D. Sokoloff, Astron. Astrophys. 413, L53 (2005); A. Diaz-Gil, et al., arXiv:0712.4263 [hep-ph]; L. Campanelli, arXiv:0805.0575 [astro-ph]. * (19) A. Lue, L. M. Wang, and M. Kamionkowski, Phys. Rev. Lett. 83, 1506 (1999); D. Lyth, C. Quimbay, and Y. Rodriguez, JHEP 0503, 016 (2005); M. Satoh, S. Kanno, and J. Soda, Phys. Rev. D 77, 023526 (2008). * (20) N. Seto, Phys. Rev. Lett. 97, 151101 (2006). * (21) S. H. S. Alexander, M. E. Peskin, and M. M. Sheikh-Jabbari, Phys. Rev. Lett. 96, 081301 (2006); S. Saito, K. Ichiki, and A. Taruya, J. Cosmol. Astropart. Phys. 0709, 002 (2007). * (22) A. S. Monin and A. M. Yaglom, Statistical Fluid Mechanics (MIT Press, Cambridge, MA, 1975). * (23) M. E. Goldstein, Aeroacoustics (McGraw-Hill, New York, 1976). * (24) M. J. Lighthill, Proc. R. Soc. London Ser. A 211, 564 (1952), 222, 1 (1954). * (25) I. Proudman, Proc. R. Soc. London Ser. A 214, 119 (1952). * (26) S. Weinberg, Gravitation and Cosmology (Wiley & Sons, New York, 1972). * (27) D. Biskamp, Magnetohydrodynamic Turbulence (Cambridge University, Cambridge, 2003); M. K. Verma, Phys. Rept. 401, 229 (2004). * (28) J. V. Shebalin, W. H. Matthaeus, and D. Montgomery, J. Plasma Phys. 29, 525 (1983); P. Goldreich and S. Sridhar, Astrophys. J. 438, 763 (1995). * (29) R. H. Kraichnan, Phys. Fluids 7, 1163 (1964). * (30) D. Biskamp and W.-C. Mueller, Phys. Rev. Lett. 83, 2195 (1999), Phys. Plasma 7, 4889 (2000). * (31) M. Christensson, M. Hindmarsh, and A. Brandenburg, Phys. Rev. E 64, 056405 (2001), Astron. Nachrichten 326, 393 (2005). * (32) R. Banerjee and K. Jedamzik, Phys. Rev. D 70, 123003 (2004). * (33) D. T. Son, Phys. Rev. D 59, 063008 (1999); L. Campanelli, Phys. Rev. Lett. 98, 251302 (2007). * (34) M. H. P. van Putten, Phys. Rev. D 50, 6640 (1994).

arxiv-papers

2024-09-04T02:48:54.003310

0802.3653

# A Note on Walking Versus Waiting Anthony B. Morton (February 2008) To what extent is a traveller (called Justin, say) better off to wait for a bus rather than just start walking—particularly when the bus headway is of a similar order of magnitude to the walking time and Justin does not know the precise arrival time of the bus? The recent analysis by Chen et al [1] goes some way toward answering this question; however, there are additional valuable insights that can be gained from their approach. In essence this is a decision-theoretic problem, where the aim is to compare various strategies for getting to one’s destination. Following Chen et al [1], it is supposed that the comparison hinges solely on the expectation of overall travel time, and there is no additional penalty attached to walking. We may list some of the available strategies as follows: * • Strategy A: Wait indefinitely until a bus arrives. * • Strategy B: Begin walking immediately and do not wait for a bus. * • Strategy C: Wait for a predetermined time interval $T_{W}$, and if no bus has arrived in this time then walk. (Naturally, strategies A and B can themselves be thought of as the extreme limiting cases of strategy C as $T_{W}\rightarrow\infty$ and $T_{W}\rightarrow 0$ respectively. However, we distinguish them for the sake of discussion.) Following the nomenclature of [1], let $d$ denote the journey distance, $v_{w}$ Justin’s average walking speed and $v_{b}$ the average speed of a bus. The overall travel time is a random variable and will be denoted $T_{t}$. As indicated in [1], if the decision is purely between strategies A and B, the problem has an easy, one might say trivial, solution: if the expected arrival time for the bus is $T$ minutes in the future, Justin should wait if $T\leq d/v_{w}-d/v_{b}$ and walk otherwise. The more interesting questions are (a) could there exist values $T_{W}$ for which strategy C is superior to either A or B; and (b) does the presence of intermediate stops affect the conclusion? The purpose of this note is to show, by direct analysis for (almost) arbitrary arrival probabilities, that the answer to (a) is indeed no in most practical cases, but that the answer to (b) is not quite as clear-cut as the authors in [1] argue. ## The Case of No Intermediate Stops The following expression is provided in [1] for Justin’s expected travel time under strategy C, where $p(t)$ is the density function for the probability a bus arrives $t$ minutes from now: $E[T_{t}]=\int_{0}^{T_{W}}\left(\frac{d}{v_{b}}+\tau\right)p(\tau)\,\mathrm{d}\tau+\left(1-\int_{0}^{T_{W}}p(\tau)\,\mathrm{d}\tau\right)\left(\frac{d}{v_{w}}+T_{W}\right).$ (1) In [1], this expression is used to determine the waiting time $T_{W}$ such that $E[T_{t}]$ is equal to the time $d/v_{w}$ required under strategy B. However, it is also of interest to optimise (1) directly. The derivative of (1) with respect to $T_{W}$ is $\frac{\partial}{\partial T_{W}}E[T_{t}]=\left(1-\int_{0}^{T_{W}}p(\tau)\,\mathrm{d}\tau\right)-\left(\frac{d}{v_{w}}-\frac{d}{v_{b}}\right)p(T_{W}).$ (2) The first term in parentheses is the complementary c.d.f. of the arrival time for the bus; that is, the probability that the bus has _not_ arrived by time $T_{W}$. Denote this function by $R(t)$: $R(t)=1-\int_{0}^{t}p(\tau)\,\mathrm{d}\tau.$ (3) Also for brevity, define $T_{\delta}$ as the difference between the travel time on foot and the travel time by bus: $T_{\delta}=\frac{d}{v_{w}}-\frac{d}{v_{b}}.$ (4) Then the derivative (2) may be expressed as $\frac{\partial}{\partial T_{W}}E[T_{t}]=R(T_{W})-T_{\delta}\,p(T_{W}).$ (5) Zeros of (5) occur at points where $\frac{1}{T_{\delta}}=\frac{p(T_{W})}{R(T_{W})}=\lambda(T_{W}).$ (6) In reliability theory where one is concerned with failure-time distributions, the function $\lambda(t)=p(t)/R(t)$ is known as the _hazard rate_ : the density for the probability of failure given a component has not failed up to time $t$ [2]. But $\lambda(t)$ also has a natural interpretation for arrival times, where it may be termed the _appearance rate_ for the distribution $p(t)$. By definition $\lambda(t)$ is nonnegative, but depending on the underlying distribution it may behave in a variety of ways: it may increase, decrease, attain local maxima or minima, or be constant. More precisely, for any $T\in\mathbb{R}_{+}\cup\\{\infty\\}$ any nonnegative real measurable function on $(0,T)$ can be the appearance rate for a possible arrival time distribution $p(t)$; if $T$ is finite then take $p(t)=R(t)=0$ for $t\geq T$. An increasing $\lambda(t)$ is characteristic of many waiting scenarios; it represents the case where the conditional probability increases with time spent waiting. By contrast, a constant $\lambda(t)$ characterises the exponential arrival time distribution, where time spent waiting does not affect the conditional likelihood of a bus arriving. Equation (6) indicates that $E[T_{t}]$ as a function of $T_{W}$ is stationary at any point where the appearance rate $\lambda(T_{W})$ equals the reciprocal of $T_{\delta}$. To determine whether this is a minimum or a maximum, one calculates the second derivative (assuming for simplicity that $p(t)$ is dfferentiable) $\frac{\partial^{2}}{\partial T_{W}^{2}}E[T_{t}]=-p(T_{W})-T_{\delta}\,p^{\prime}(T_{W})$ (7) and inspects the sign of this quantity. Since $p(t)$ and $T_{\delta}$ are both positive (and $p(T_{W})$ is nonzero by (6)) this is equivalent to the sign of $-\left(\frac{1}{T_{\delta}}+\frac{p^{\prime}(T_{W})}{p(T_{W})}\right)=-\left(\frac{p(T_{W})}{R(T_{W})}+\frac{p^{\prime}(T_{W})}{p(T_{W})}\right)=-\frac{\lambda^{\prime}(T_{W})}{\lambda(T_{W})}=-T_{\delta}\,\lambda^{\prime}(T_{W}).$ (8) Accordingly, any stationary point of $E[T_{t}]$ is a minimum when it coincides with a falling appearance rate, and a maximum when it coincides with a rising appearance rate. It is thus established that, when faced with a typical probability distribution for bus arrivals having a rising appearance rate, the optimal strategy reduces to a choice between strategies A and B, and hence to a straightforward comparision of expected arrival time with $T_{\delta}$ as indicated above. For example, if it is known that buses run punctually every $T$ minutes but one does not know the actual arrival times, the uncertainty in arrival time can be modelled as a uniform probability density $p(t)$ between $0$ and $T$, with mean $T/2$. The corresponding appearance rate is $\lambda(t)=\frac{1}{T-t},\qquad t\in(0,T)$ (9) an increasing function of $t$. One may then distinguish three cases: 1. 1. If $T<T_{\delta}$, the expected travel time $E[T_{t}]$ has no stationary points, and is in fact a decreasing function of wait time. In this case strategy A is optimal: one should wait for the next bus. 2. 2. If $T_{\delta}<T<2T_{\delta}$, the expected travel time has one stationary point at $T_{W}^{*}=T-T_{\delta}$, which is a maximum. The expected travel time increases with waiting time up to $T_{W}^{*}$ and then decreases. But since the mean bus arrival time is $T/2<T_{\delta}$, the waiting strategy A is still preferable to an initial decision to walk. 3. 3. If $T>2T_{\delta}$, then expected travel time again reaches a maximum at $T_{W}^{*}=T-T_{\delta}$. But now, it is better to follow strategy B and not wait for a bus at all. Perhaps the most interesting situation here is the marginal case where $T=2T_{\delta}$: the interval between buses is twice the difference between walking time and bus travel time. In this case it is equally good (or bad) to wait as to walk, but having decided to wait, one should not give up and walk, as this will result in a worse outcome (averaged over all eventualities). Not all practically realisable probability distributions have rising appearance rates for all $t$. Consider, for example, the situation where Justin has arrived at the bus stop one minute after the (known) scheduled arrival time for the bus, but knows that buses sometimes run up to five minutes late. Under reasonable assumptions this leads to a falling $\lambda(t)$ for the first four minutes, and if $p(0)$ is sufficiently large and the service relatively infrequent, this will result in an optimal time $T_{W}^{*}$ for which Justin should remain at the bus stop and then start walking. However, if the next bus can be counted on to arrive in a time comparable to $T_{\delta}$, it will still be advantageous to keep waiting for the next bus, and the lazy mathematician still wins. One may also contemplate the special case where $\lambda(t)\equiv\lambda$ is constant, and bus arrivals are a Poisson process. The mean time between arrivals is $T=1/\lambda$; if this is greater than $T_{\delta}$ then it is better to walk, and if less then it is better to wait, as is intuitively clear. More interesting is the case where $T$ happens to just equal $T_{\delta}$: then, the derivative (5) vanishes and the expected travel time is equal to the walking time $d/v_{w}$, _regardless_ of the time spent waiting. What this means is, no matter how long Justin decides in advance to wait, over a large number of journeys he can expect to spend no more or less time than if he had decided to walk at the outset. Nonetheless, Poisson arrivals are generally an unrealistic assumption except in some cases of very frequent service, where it is generally better to catch the bus in any event. ## Intermediate Stops In [1] it is argued that the conclusion is unchanged when Justin has the additional option of walking to an intermediate stop and catching the bus there. If, for example, one re-evaluates the decision between strategies A and B at a distance $d_{1}$ from the start of the journey, then the walking time is $(d-d_{1})/v_{w}$ and the bus travel time is $(d-d_{1})/v_{b}$. However, the expected waiting time for the bus (assuming it does not pass by _en route_) is now reduced by $T_{1}=d_{1}/v_{w}-d_{1}/v_{b}$, the time that passes while walking less the additional time taken by the bus to reach the next stop. The terms involving $d_{1}$ cancel, and one is left with a decision identical to the original one. Under these circumstances, one would “rather save energy” and act as though the intermediate stop did not exist, since the outcome is the same. This conclusion is also intuitively evident. The authors argue that the same reasoning applies to walk-and-wait strategies analogous to strategy C, at least when the wait time $T_{W}$ at the next stop is chosen to make the expected travel time equal to the walking time. However, there appears to be a circularity involved here—if the expected travel time is fixed as $d/v_{w}$ _a priori_ , then naturally it will be observed to be the same whether or not Justin walks to another stop. Again, some more insight into the problem is gained by seeking stationary points of $E[T_{t}]$ with respect to the free variables. Suppose strategy C is modified so that one walks a distance $d_{1}$ to another stop, then waits for a maximum time $T_{W}$. One may now distinguish two cases, depending on whether or not a bus passes by _en route_. Denote this event by $M$ (‘miss the bus’) and its non-occurrence by $\bar{M}$; the relevant probabilities are $\mathrm{Pr}\left(M\right)=\int_{0}^{T_{1}}p(\tau)\,\mathrm{d}\tau,\qquad\mathrm{Pr}\left(\bar{M}\right)=1-\mathrm{Pr}\left(M\right)=R(T_{1}),\qquad T_{1}=\frac{d_{1}}{v_{w}}-\frac{d_{1}}{v_{b}}.$ (10) If no bus passes by _en route_ , then the expected travel time is similar to (1), but with a time offset: $\displaystyle E[T_{t}|\bar{M}]=\frac{d_{1}}{v_{w}}$ $\displaystyle+\int_{0}^{T_{W}}\left(\frac{d-d_{1}}{v_{b}}+\tau\right)\frac{p(\tau+T_{1})}{\mathrm{Pr}\left(\bar{M}\right)}\,\mathrm{d}\tau$ $\displaystyle+\left(1-\int_{0}^{T_{W}}\frac{p(\tau+T_{1})}{\mathrm{Pr}\left(\bar{M}\right)}\,\mathrm{d}\tau\right)\left(\frac{d-d_{1}}{v_{w}}+T_{W}\right).$ (11) Even if the bus does pass by, it is possible that Justin may catch it anyway, if he is vigilant and there is a stop close enough, or if the bus is delayed, or if he takes a short cut. Let the probability of catching a bus if one turns up on the way be $P_{C}$. On the other hand, if Justin misses the bus it may be assumed there is no advantage in waiting for the next one (else he would have preferred strategy A at the outset). Thus the expected travel time conditional on $M$ is $E[T_{t}|M]=P_{C}\int_{0}^{T_{1}}\left(\tau+\frac{d}{v_{b}}\right)\frac{p(\tau)}{\mathrm{Pr}\left(M\right)}\,\mathrm{d}\tau+(1-P_{C})\frac{d}{v_{w}}.$ (12) (Note that within the integral, $\tau$ is the time at which the caught bus reaches the _starting_ point, rather than Justin’s current position, since $p(\tau)$ is the p.d.f. of bus arrivals at a _fixed_ point on the route. This is a subtle point and easily overlooked. In terms of $\tau$, Justin’s overall travel time if he catches the bus is $\tau$ plus the time taken by the bus to cover the entire journey, regardless of where Justin is when he catches it.) The overall expected travel time is $E[T_{t}]=\mathrm{Pr}\left(M\right)E[T_{t}|M]+\mathrm{Pr}\left(\bar{M}\right)E[T_{t}|\bar{M}],$ (13) where $E[T_{t}|M]$ and $E[T_{t}|\bar{M}]$ are given by (12) and (11) respectively. Again, one can differentiate to find optimal values of wait time $T_{W}$ at the more distant stop. (This is made easier by noting that $E[T_{t}|M]$ does not vary with $T_{W}$.) One finds that $\displaystyle\frac{\partial}{\partial T_{W}}E[T_{t}]$ $\displaystyle=R(T_{W}+T_{1})-T_{\delta 1}p(T_{W}+T_{1})$ (14) $\displaystyle\mbox{and}\qquad\frac{\partial^{2}}{\partial T_{W}^{2}}E[T_{t}]$ $\displaystyle=-p(T_{W}+T_{1})-T_{\delta 1}p^{\prime}(T_{W}+T_{1}),$ (15) where $T_{\delta 1}=(d-d_{1})\left(\frac{1}{v_{w}}-\frac{1}{v_{b}}\right)=T_{\delta}-T_{1}$ (16) is the difference between travel time on foot and by bus starting from the more distant stop. These are exactly equivalent to (5) and (7), with $T_{W}+T_{1}$ in place of $T_{W}$ and $T_{\delta 1}$ in place of $T_{\delta}$. Accordingly, the analysis of the previous section is essentially unchanged, as is the conclusion: that in most waiting scenarios the optimal value of $T_{W}$ is either zero or arbitrarily large. But now there is an additional variable in the problem, because Justin will generally have a number of stops to choose from between the starting point and the destination. For the sake of simplicity, suppose the stop spacing is small enough that $d_{1}$ can be regarded as a continuous variable. Then we can try and optimise with respect to $d_{1}$, obtaining after a little work $\frac{\partial}{\partial d_{1}}E[T_{t}]=q^{2}(d-d_{1})\left[(1-P_{C})p(T_{1})-p(T_{1}+T_{W})\right]$ (17) where $q=1/v_{w}-1/v_{b}$ is a constant. The above discussion suggests directing attention to the cases $T_{W}=0$ and $T_{W}\rightarrow\infty$. If $T_{W}=0$, then (17) vanishes unless $P_{C}>0$; that is, when there is a nonzero chance of catching a bus while walking to the next stop. Assuming then that $P_{C}>0$ and $T_{W}=0$, (17) is found to be zero when $p(T_{1})=0$ and negative when $p(T_{1})>0$. (Recall that $T_{1}$ is proportional to $d_{1}$.) The result is intuitively evident: as long as there is a chance of catching a bus on the way (however small), walking a distance $d_{1}$ to an intermediate stop always reduces the expected travel time $E[T_{t}]$ relative to strategy B, where one does not attempt to catch a bus. The actual reduction can be quantified by integrating (17) with respect to $d_{1}$. Due to the factor $(d-d_{1})$ in (17), the marginal reduction in travel time is greatest at the start of the journey, all other things being equal. Now consider the case $T_{W}\rightarrow\infty$. One may presume that there is a limit to the time one must wait for a bus, so that $p(T_{1}+T_{W})$ becomes zero for $T_{W}$ sufficiently large. It is also fair to assume that $P_{C}<1$. Then it follows that for $T_{W}$ sufficiently large, (17) is zero when $p(T_{1})=0$, but is _positive_ when $p(T_{1})>0$. It follows that walking to an intermediate stop only to wait indefinitely at that stop yields a worse expected outcome than waiting at the starting point, consistent with common sense. Again, (17) can be used to judge how much one is worse off due to the chance of missing the bus on the way. The more interesting case to consider, of course, is a direct comparison between strategy A (waiting indefinitely at the starting point) and a strategy of walking but trying to catch a bus on the way. From above, the best possible strategy of the latter variety is the case $T_{W}=0$ and $d_{1}=d$: a little manipulation of (13) yields the result $E_{*}[T_{t}]=\frac{d}{v_{w}}-P_{C}\int_{0}^{T_{\delta}}(T_{\delta}-\tau)p(\tau)\,\mathrm{d}\tau.$ (18) Under strategy A, of course, the expected travel time is just the bus travel time, plus the expected arrival time for the bus: $E_{A}[T_{t}]=\frac{d}{v_{b}}+\int_{0}^{\infty}\tau p(\tau)\,\mathrm{d}\tau.$ (19) The difference in expected travel time is then $E_{A}[T_{t}]-E_{*}[T_{t}]=P_{C}T_{\delta}\left(1-R(T_{\delta})\right)+(1-P_{C})\int_{0}^{T_{\delta}}\tau p(\tau)\,\mathrm{d}\tau+\int_{T_{\delta}}^{\infty}\tau p(\tau)\,\mathrm{d}\tau-T_{\delta}.$ (20) If this quantity is positive, it represents the expected time saving if one walks instead of waiting. (It will be seen that all terms are in fact nonnegative apart from the last one, $T_{\delta}$; which, however, is relatively large.) For example, let $p(t)$ be the uniform distribution based on a bus headway $T$. The mean arrival time is $T/2$ and hence $E_{A}[T_{t}]=d/v_{b}+T/2$. If $T>T_{\delta}$, then $E_{*}[T_{t}]=\frac{d}{v_{w}}-P_{C}\frac{T_{\delta}^{2}}{2T}$ (21) and this improves on $E_{A}[T_{t}]$ if $P_{C}>2\frac{T}{T_{\delta}}-\left(\frac{T}{T_{\delta}}\right)^{2}.$ (22) As $T/T_{\delta}$ increases from 1 to 2, the minimum probability $P_{C}$ decreases from 1 to zero; recall that when $T/T_{\delta}>2$ strategy A is worse than strategy B, which is equivalent to the above strategy with $P_{C}=0$. If $T<T_{\delta}$, then $E_{*}[T_{t}]=\frac{d}{v_{w}}-P_{C}\left(T_{\delta}-\frac{T}{2}\right)$ (23) and the condition for this to improve on $E_{A}[T_{t}]$ reduces to $P_{C}>1$, a contradiction. Accordingly, strategy A is still the best choice for uniform arrival probabilities with headway $T<T_{\delta}$. ## Conclusion Closer analysis of the model and results in [1] for a traveller at a bus stop confirms that the lazy mathematician does indeed win in many cases. The exact criteria for when it is better to walk can be formulated in terms of an appearance rate function, and if one disregards the existence of intermediate stops, an arrival probability with a rising appearance rate is sufficient to make waiting the best strategy. It has also been shown, however, that allowing for the possibility of catching the bus at an intermediate stop can restore the walking strategy to optimality. In the end, the decision whether to walk or wait is not always clear-cut, and relies on one’s expectations of being able to catch a bus on the run, as well as of a bus turning up in the first place. In the particular case where one knows the (uniform) headway but not the exact timetable, it has been clearly established that one should wait if the headway is less than the walking time (less bus travel time), and should walk if the headway is more than twice this much. This leaves a substantial window where it may be better to wait or to walk, depending on one’s confidence in being able to catch up to a passing bus. ## References * [1] J.G. Chen, S.D. Kominers, and R.W. Sinnott. Walk versus wait: The lazy mathematician wins. arXiv.org Mathematics, January 2008. http://arxiv.org/abs/0801.0297. * [2] R. Billinton and R.N. Allan. Reliability Evaluation of Engineering Systems. Plenum, second edition, 1992.

arxiv-papers

2024-09-04T02:48:54.009054

0802.3850

# Spin-polarized collision of deuterium and tritium: Relativistic Kinematics Thomas B. Bahder William C. McCorkle Charles M. Bowden Research Facility Weapons Sciences Directorate, Army Aviation and Missile Research, Development, and Engineering Center, Redstone Arsenal, Alabama 35898-5000 William V. Dent, Jr Dent International Research, Inc. 3000 Turnberry Drive Hampton Cove, AL 35763 thomas.bahder@us.army.mil ###### Abstract We investigate the relativistic kinematics of the spin-polarized collision of deuterium incident on tritium, producing 4He and a neutron. Within the context of special relativity, we apply the conservation of four momentum and the conservation of intrinsic spin, which leads to a system of ten equations. We impose initial conditions such that the deuterium is moving along the x-axis, the tritium is stationary at the origin of coordinates, and the classical spin vector of the deuterium (spin magnitude = 1) is along the $+$z-axis, while the classical spin vector for tritium (spin magnitude = 1/2) is along the $-$z-axis. We expand the ten conservation equations to second order in velocities and we solve them for the velocity components of the neutron, its unit-spin-orientation vector, and the velocity components of the 4He nucleus, as a function of the incident deuterium energy. We find that this analytic solution agrees closely with the numerical solution of the ten (unexpanded) equations. For a given energy of deuterium, we find that there are two solutions, each solution having a unique velocity for the emitted neutron and helium nucleus. The two solutions are related to each other by reflection in the plane perpendicular to the deuterium spin and containing the initial deuterium velocity vector. ## I Introduction Sources of neutrons are currently used to excite nuclei in an unknown sample of material in order to determine its elemental composition. The technique is called neutron activation analysis, where neutrons are incident on target nuclei and a unique spectrum of gamma rays are emitted for each element that is present in the sample Alfassi1990 . In this technique, measurements are made of emitted gamma rays that are either emitted almost instantaneously (prompt gamma-ray), or, that are delayed. The prompt gamma-rays that are emitted almost instantaneously come from a compound nucleus that is formed when the neutron is captured by the target nucleus in the sample. The technique is called prompt gamma-ray neutron activation analysis (PGNAA). When delayed gamma rays are measured, they are emitted by the decay of a radioactive intermediate state formed by the neutron irradiation. In this case, the technique is called delayed gamma-ray neutron activation analysis (DGNAA). In both cases, analysis of the composite emitted gamma ray spectrum often allows a precise determination of the elemental content of the sample Alfassi1990 . A critical component in this type of detection is the source of neutrons. Ideally, we would want a narrow beam of neutrons that can be directed at the sample so that the damaging effects of neutron radiation are limited to a small solid angle. A reaction that is commonly used for neutron production is the collision of deuterium with tritium Ohlsen1967 , $d+t\rightarrow n+{}^{4}{He}$ (1) However, with unpolarized deuterium and tritium this reaction generally produces an isotropic distribution of emitted neutrons, as opposed to the narrow beam mentioned above. If the deuterium and tritium nuclei were spin polarized before collision, then the symmetry of the collision, $\overrightarrow{T}(\overrightarrow{d},\overrightarrow{n})\,{}^{4}He$, would be reduced, and we would expect that there may be preferential directions for the emitted neutrons. Quantum mechanical calculations of this collision process are quite complicated Ohlsen1967 ; Kulsrud1982 . However, it is possible to use a considerably simpler approach to analyze some aspects of the kinematics of the spin-polarized deuterium and spin-polarized tritium collision. We use conservation of four-vector momentum and conservation of intrinsic spin tensor, within the context of special relativity, to investigate the kinematics of this collision. This reaction in Eq.(1) has the advantage that it has a large cross section, which peaks at 5 b Ohlsen1967 ; Bonar1968 for a deuterium energy of 107 keV. It is believed that at low-energy the reaction, T(d,n)${}^{4}He$, proceeds via a short-lived intermediate resonant state of ${}^{5}He$ with a spin angular momentum $J=3/2$, and that this state quickly decays into a neutron and a ${}^{4}He$ nucleus. At low energies, the reaction proceeds via zero orbital angular momentum, $l=0$, so that the total spin $J=3/2$ must arise from intrinsic spin of the reactants Kulsrud1982 . Therefore, at low energies, the deuterium nuclear spin $s_{d}=1$, and tritium nuclear spin, $s_{t}=1/2$, must be aligned to produce a total angular momentum $J=3/2$ in the initial state. The products of the reaction are a neutron, with nuclear spin $s_{n}=1/2$, and ${}^{4}He$ which has nuclear spin $s_{h}=0$. The total intrinsic spin of the product state (helium and neutron) is $s=1/2$, so the rest of the angular momentum, the quantity $J=3/2-1/2=1$, must be carried off as orbital angular momentum of the product state. Alternatively, there is also some small amplitude for the reaction to proceed, with zero orbital angular momentum, when the deuterium and tritium spins are anti-aligned. In this case, there is zero orbital angular momentum carried off by the products. We consider this latter case in the work below. The formalism we use was initially applied by J. L. Synge to discuss elastic collisions (without transmutation of mass) of particles having intrinsic spin Synge1965 . Here, we use Synge’s formalism to analyze the fundamentally inelastic collision, $\overrightarrow{T}(\overrightarrow{d},\overrightarrow{n})\,{}^{4}He$, when the reactants have a definite state of spin, in the sense of special relativity, where spin is represented as a four-tensor. The treatment we give below assumes that the deuterium has a zero impact parameter on the tritium, and therefore this limits our analysis to the case where the spin of deuterium and tritium are anti-aligned, i.e., the spins are pointing in opposite directions. Unfortunately, the condition of zero impact parameter is not easily relaxed because of complications in special relativity associated with the speed of propagation of interactions and the requirement of relativistic invariance Synge1965 . As described above, at low energies, the zero impact parameter (zero orbital angular momentum in initial state) is probably the most important in the collision process since the reaction $\overrightarrow{T}(\overrightarrow{d},\overrightarrow{n})\,{}^{4}He$ will only occur when the deuterium and tritium are in close spatial proximity, due to the short range nature of the nuclear forces. Therefore, our goal here is to analyze the dependence of the velocity components (angular dependence) of the emitted neutron and helium nuclei, on the initial conditions, which are specified by initial velocity and initial spin orientation for deuterium and tritium. We apply the conservation of four momentum and conservation of intrinsic spin four-tensor. ## II Conservation of four momentum The total four momentum before the collision is equal to the total four momentum after the collision $p_{k}^{(d)}+p_{k}^{(t)}=p_{k}^{(n)}+p_{k}^{(h)}$ (2) where the superscripts $d$, $t$, $n$, and $h$ label the four momentum vectors for deuterium, tritium, neutron, and helium, respectively. The subscript $k=1,2,3,4$ labels the space-time components of the four-vector. We follow the convention used by Synge and take the fourth component to be imaginary, thereby omitting the need to explicitly introduce a metric tensor for space- time. So the momentum four-vector has the form $p_{k}=(p_{\alpha},p_{4})=m(\frac{\gamma}{c}\,u_{\alpha},i\,\gamma)$ (3) where $\gamma=(1-(u_{\alpha}\,u_{\alpha})/c^{2})^{-1/2}$. Greek indices take values $\alpha=1,2,3$, and Latin indices always take values $k=1,2,3,4$. We use the convention that repeated indices are summed over their respective ranges. Here, $u_{\alpha}=d\,x_{\alpha}/d\,t$ are the ordinary three velocity components. Equation (2) is a conservation law that is valid in any inertial frame of reference. For our purposes, we assume that the laboratory frame is an inertial frame. The conservation of four momentum leads to four equations. ## III Angular Momentum Tensor In four dimensional space-time, the total angular momentum is an antisymmetric second rank tensor, $H_{mn}=H^{orb}_{mn}+H^{spin}_{mn}$, with six independent components, for $m,n=1,2,3,4$, where $H^{orb}_{mn}$ is the orbital angular momentum and $H^{spin}_{mn}$ is the spin angular momentum Synge1965 . The spin is represented by a four tensor because in four dimensions (space-time) there is no four-vector (of angular momentum) that is dual to the tensor $H_{mn}$. As discussed above, we assume that in the laboratory frame of reference the impact parameter for deuterium on tritium is zero, so there is zero orbital angular momentum in the collision, $H^{orb}_{mn}=0$. In what follows, we drop the superscript and we use $H_{mn}$ to denote the intrinsic spin four-tensor, $H^{spin}_{mn}$. For a particle that is not experiencing any forces, the intrinsic spin four- tensor is conserved along the particle’s world line Synge1965 . When a collision occurs at a space-time point, the sum of the intrinsic spin four- tensors before the collision must be equal to the sum of these tensors after the collision, since the particles do not experience forces before or after the collision. We assume that the range of forces between particles is negligible, so that we take the collision to occur at a point. The intrinsic spin four-tensor can be expressed in terms of the particle’s four velocity, $\lambda_{n}=(\frac{\gamma}{c}\,u_{\alpha},i\,\gamma)$ (4) and a unit spin four-vector, $s_{n}$, according to Synge1965 $H_{kl}=i\,\Omega\,\epsilon_{klmn}\,\lambda_{m}\,s_{n}$ (5) where $\Omega$ is the magnitude of the intrinsic spin, the four-velocity satisfies $\lambda_{n}\,\lambda_{n}=-1$ and the unit spin four-vector satisfies, $s_{n}s_{n}=1.$ (6) The Levi-Civita symbol, $\epsilon_{klmn}$, is antisymmetric with respect to interchange of adjacent indices, it satisfies $\epsilon_{1234}=+1$ and has value -1 for odd permutations of its argument, so that $H_{kl}=i\,\Omega\,\,\left(\begin{array}[]{cccc}0&(s_{4}\lambda_{3}-s_{3}\lambda_{4})&(s_{2}\lambda_{4}-s_{4}\lambda_{2})&(s_{3}\lambda_{2}-s_{2}\lambda_{3})\\\ (s_{3}\lambda_{4}-s_{4}\lambda_{3})&0&(s_{4}\lambda_{1}-s_{1}\lambda_{4})&(s_{1}\lambda_{3}-s_{3}\lambda_{1})\\\ (s_{4}\lambda_{2}-s_{2}\lambda_{4})&(s_{1}\lambda_{4}-s_{4}\lambda_{1})&0&(s_{2}\lambda_{1}-s_{1}\lambda_{2})\\\ (s_{2}\lambda_{3}-s_{3}\lambda_{2})&(s_{3}\lambda_{1}-s_{1}\lambda_{3})&(s_{1}\lambda_{2}-s_{2}\lambda_{1})&0\end{array}\right)$ (7) The intrinsic spin four-tensor has the following frame-independent invariant, $\frac{1}{2}H_{kl}\,H_{kl}=\Omega^{2}.$ (8) Furthermore, the intrinsic spin four-tensor is orthogonal to the four velocity $H_{mn}\,\lambda_{n}=0$ (9) by virtue of its construction in Eq. (5). The four components of the unit spin four-vector, $s_{n}$, are related to the four-velocity of the particle, $\lambda_{n}$, by the orthogonality condition Synge1965 $\lambda_{n}\,s_{n}=0$ (10) Equations (9) and (10) state that the physical spin vector lies in the three- dimensional hypersurface that is the three-dimensional physical space. Equations (5 ) through (10) are tensor equations and hence they are valid in any frame of reference. The fourth component of the unit spin four-vector, $s_{4}$ can be expressed in terms of the other three components by use of the orthogonality condition in Eq. (10), $s_{4}=\frac{i}{c}(s_{1}\,u_{1}+s_{2}\,u_{2}+s_{3}\,u_{3})$ (11) Using Eq. (11) to eliminate $s_{4}$ from Eq. (6), we have a relation between the three components of spin, $s_{\alpha}$, and the three velocity, $u_{\beta}$, which must be satisfied $s_{\alpha}\,s_{\alpha}=1+\frac{1}{c^{2}}(s_{\beta}\,u_{\beta})^{2}$ (12) If we define three-component quantities by ${\bf s}=(s_{1},s_{2},s_{3})$ and ${\bf u}=(u_{1},u_{2},u_{3})$, then we can write suggestively that the quantity ${\bf s}$ is normalized according to ${\bf s}^{2}=1+\frac{1}{c^{2}}({\bf s}\cdot{\bf u})^{2}$ (13) Equation (12) or (13) shows that the three spatial components ${\bf s}$ of the unit spin four-vector, $s_{n}$, satisfy a normalization relation that depends on the particle’s three velocity components, ${\bf u}$. When the particle is at rest, ${\bf u}=0$, the three components satisfy ${\bf s}\cdot{\bf s}=1$, as expected. In the low velociy limit, we can interpret the three component quantity, ${\bf S}=\Omega{\bf s}$, as the classical spin vector, where $\Omega$ is the magnitude of the particle’s spin. Of course, the transformation properties of the three components, ${\bf S}$, depend on velocity and hence do not transform as a true three-vector. ## IV Intrinsic Spin Conservation We consider the collision, $\overrightarrow{T}(\overrightarrow{d},\overrightarrow{n})\,{}^{4}He$ where the initial conditions are such that both the deuterium and tritium spins are anti-aligned. The magnitude of total spin for deuterium is assumed to be $\Omega_{d}=1$ and the magnitude of spin for the tritium nucleus is taken to be $\Omega_{t}=1/2$. The products of this inelastic collision are a ${}^{4}He$ nucleus, whose total spin is $\Omega_{h}=0$, and a neutron, whose total spin is $\Omega_{n}=1/2$. The conservation of intrinsic spin before and after the collision is given by $H^{d}_{mn}+H^{t}_{mn}=H^{h}_{mn}+H^{n}_{mn}$ (14) where $H^{h}_{mn}=0$ because the spin of He is zero. Each of the intrinsic spin four-tensors in Eq. (14) satisfies Eq. (9) by virtue of their construction according to Eq. (5). Each spin tensor $H_{mn}$ is a $4\times 4$ antisymmetric matrix. Therefore, Eqs. (14) constitute six independent equations that relate the spin orientations to the velocity components, before and after the collision. Equations (14) are tensor equations that are valid in any inertial frame of reference. ## V Initial Conditions and Solutions We have a total of ten conservation equations, where four of them are the conservation of four momentum given by Eq. (2) and six of them are the conservation of intrinsic spin given by Eq. (14). In solving the kinematic problem of the deuterium-tritium collision, the magnitude of spin for each particle is known, as described above. The orientation of the spins and velocities of deuterium and tritium are initial conditions that can be selected arbitrarily. We solve for the velocity and spin orientation components of the neutron, and the velocity components of the 4He, in the laboratory frame of reference. For initial conditions, we take the tritium to be stationary at the origin of coordinates with velocity $u_{\alpha}^{t}=(0,0,0)$, and we take the deuterium to be moving along the $+x$-axis, with velocity components specified by $u_{\alpha}^{d}=(u,0,0)$, where $u>0$. We take the deuterium spin direction to be along the $+z$-axis, specified by the spin orientation three-vector as $s_{\alpha}^{d}=(0,0,1)$ and we take the tritium spin orientation three-vector to be $s_{\alpha}^{t}=(0,0,-1)$. The deuterium is moving such that ${\bf s}\cdot{\bf u}=0$, so ${\bf s}^{2}=1$, and the initial condition $s^{d}_{\alpha}=(0,0,1)$ satisfies Eq.(13). (As described above, the magnitude of the spin for every particle is fixed.) We solve for the following quantities: velocity and spin orientation components of the neutron, $u^{n}_{\alpha}=(u^{n}_{x},u^{n}_{y},u^{n}_{z})$ and $s^{n}_{\alpha}=(s^{n}_{x},s^{n}_{y},s^{n}_{z})$, respectively, and velocity of 4He, $u^{h}_{\alpha}=(u^{h}_{x},u^{h}_{y},u^{h}_{z})$. Note that we use subscripts $x,y,z$ and 1,2,3 interchangeably to denote components. Figure 1: The angle of neutron as a function of deuterium kinetic energy in the laboratory frame of reference for solution 1. Figure 2: The angle of helium as a function of deuterium kinetic energy in the laboratory frame of reference for solution 1. Figure 3: The angle of neutron as a function of deuterium kinetic energy in the laboratory frame of reference for solution 2. Figure 4: The angle of helium as a function of deuterium kinetic energy in the laboratory frame of reference for solution 2. We have a total of ten nonlinear equations, given by Eq. (2) and (14). There are a total of nine unknowns: three velocity components for the neutron, $u^{n}_{\alpha}$, three velocity components for the 4He, and three spin- orientation vector components for the neutron spin, $s^{n}_{\alpha}$. A key point is that, before solving the ten Eqs. (2) and (14), we must eliminate from Eqs. (14) the fourth component of the spin-orientation vector for each particle using Eq. (11). In this way, only the three spatial components, $s^{n}_{\alpha}$, of the spin-orientation four-vector $s^{n}_{m}$, appear in the intrinsic spin tensors in Eqs. (14). In the solutions of our ten equations, the spin-orientation vector for the neutron, ${\bf s}^{(n)}=(s^{n}_{x},s^{n}_{x},s^{n}_{x})$, and the neutron velocity, $u^{n}_{\alpha}$, must still satisfy Eq. (12) or (13). We have checked that our solutions satisfy Eq. (12). Having eliminated the components, $s^{d}_{4}$, $s^{t}_{4}$, $s^{n}_{4}$, and $s^{h}_{4}$, the ten Eqs. (2) and (14) are solved by expanding them in a Taylor series to second order in the velocity components, $u^{n}_{\alpha}$, $u^{h}_{\alpha}$ and $u$, which we assume are small quantities compared to the speed of light, $c$. At least second order in the velocities must be kept to allow for the transmutation of mass, i.e., inelastic collision process. We find that there are two solutions to these ten equations, and each solution has a unique velocity for the neutron and for the 4He nucleus. The two solutions, which we call solution 1 and solution 2, are related by reflection in the $x-y$ plane, which is perpendicular to the deuterium spin, $s_{\alpha}^{d}=(0,0,1)$, and contains the deuterium velocity vector, $u^{d}=(u,0,0)$. Note that the signs of the velocity components for the two solutions are as follows: $\displaystyle{\rm solution\,\,1:}\,\,\,\,$ $\displaystyle u^{n}_{x}>0,\,\,u^{n}_{y}=0,\,\,u^{n}_{z}>0;\,\,\,\,u^{h}_{x}<0,\,\,u^{h}_{y}=0,\,\,u^{h}_{z}<0$ $\displaystyle{\rm solution\,\,2:}\,\,\,\,$ $\displaystyle u^{n}_{x}>0,\,\,u^{n}_{y}=0,\,\,u^{n}_{z}<0;\,\,\,\,u^{h}_{x}<0,\,\,u^{h}_{y}=0,\,\,u^{h}_{z}>0$ Furthermore, we solve numerically the ten nonlinear (unexpanded) equations, given by Eq. (2) and (14), and compare these numerical solutions to the analytic solutions. The two solutions agree very well. The solutions show that the emitted neutron and helium have velocity vectors that lie in the x-z plane. Figures 1 and 2 show the angle of the emitted neutron and helium, respectively, as a function of incident deuterium kinetic energy, for solution 1. Figures 3 and 4 show that same quantities for solution 2, which is related to solution 1 by reflection, as described above. The numerical data used in our calculations is given in Table I. At the maximum cross section, where the deuteron kinetic energy is 107 keV, the angle of the neutron is $\theta_{n}=\arctan(u_{3}/u_{1})=82.85$ degrees and the angle of the helium is $\theta_{h}=\arctan(u_{3}/u_{1})=89.88$ degrees. ## VI Discussion Within the context of special relativity, we have applied the conservation of four momentum and the conservation of intrinsic spin to a collision of deuterium with a stationary tritium atom, in the laboratory frame of reference. We have assumed that the deuterium and tritium nuclei have spins that are anti-aligned and have zero impact parameter, corresponding to zero orbital angular momentum in the initial state. We have verified, as intuitively expected, that there are unique (classical) solutions for neutron (and helium) velocities, i.e., there are unique directions for the emitted neutron and helium. Specifically, for a given energy of incident deuteron, there are two solutions that are related by reflection symmetry, see the above discussion. We have obtained the angular dependence of the emitted neutron and helium as a function of the incident deuterium kinetic energy. Our results may be compared with a quantum calculation Kulsrud1982 given by Kulsrud et al. Table 1: Numerical data taken from Refs.atomicMasses and conversions . deuterium mass $m_{d}$ 2.01410177785 u tritium mass $m_{t}$ 3.01604927767 u neutron mass $m_{n}$ 1.00866491574 u 4He mass $m_{h}$ 4.00260325415 u speed of light $c$ 299792458 m/s MeV per u – 931.49402823303 MeV/u From our classical relativistic analysis, it seems clear that we could in principle relax the constraint of zero impact parameter, so that we could impose some amount of orbital angular momentum for the deuterium and helium. Within a classical context (as opposed to a quantum calculation) the amount of orbital angular momentum imposed in the initial conditions is arbitrary and is a continuous quantity (not a discrete quantity, as in quantum calculations)related to the impact parameter. As described earlier, assuming non-zero angular momentum in the initial state would allow us to treat the case where the deuterium and tritium spins are aligned, which is presumably the more important case. As described earlier, in such as case, we will have orbital angular momentum appear in the products of the collision (neutron and helium). The continuum of possible angular momentum in the initial state would then lead to a continuum of angular momentum in the possible final (product) states. So the neutrons would not have a unqiue direction, but instead, they would have a distribution of possible velocities. The assumed initial probability distributions of impact parameters would then lead to a probability distributions for neutron velocities and angles. As described earlier, this is believed to be the more important case. We leave this for possible future work. ## VII acknowledgments This work was sponsored in part by ILIR at the AMRDEC. ## References * (1) See for example, Z.B. Alfassi, Activation Analysis, Volumes I and II. CRC Press, Boca Raton, Florida, USA, (1990). See also the web sites, http://archaeometry.missouri.edu/naa_overview.htmlandhttp://www.reak.bme.hu/nti/Education/Wigner_Course/WignerManuals/Budapest/NEUTRON_ACTIVATION_ANALYSIS.htm and http://web.missouri.edu/~umcreactorweb/pages/ac_elemlist.shtml#anchor5. * (2) G. G. Ohlsen, Phys. Rev. 164 1268 (1967). * (3) D. C. Bonar, C. W. drake, R. D. Headrick, V. W. Hughes, Phys. Rev. 174, 1200 (1968). * (4) R. M. Kulsrud, H. P. Furth, and M. Goldhaber, Phys. Rev. Lett. 49, 1248 (1982). * (5) J. L. Synge, Relativity: The Special Theory, Second Edition, North-Holland (1965). * (6) A. H. Wapstra, G. Audi, and C. Thibault. Nuclear Physics A729, 129 (2003), see http://www.nndc.bnl.gov/nsr/nsrlink.jsp?2003WA32,B and also G. Audi, A.H. Wapstra, and C. Thibault. Nuclear Physics A729, 337 (2003), see http://www.nndc.bnl.gov/nsr/nsrlink.jsp?2003AU03,B. Also, see the Los Alamos web page http://www.nndc.bnl.gov/masses/mass.mas03. * (7) See for conversion factors, http://physics.nist.gov/cuu/Document/factors_2006.pdf

arxiv-papers

2024-09-04T02:48:54.015418

0802.3903

# Constraining Galaxy Evolution With Bulge-Disk-Bar Decomposition Tim Weinzirl1, Shardha Jogee1, Fabio D. Barazza2 ###### Abstract Structural decomposition of galaxies into bulge, disk, and bar components is important to address a number of scientific problems. Measuring bulge, disk, and bar structural parameters will set constraints on the violent and secular processes of galaxy assembly and recurrent bar formation and dissolution models. It can also help to quantify the fraction and properties of bulgeless galaxies (those systems having no bulge or only a relatively insignificant disky-pseudobulges), which defy galaxy formation paradigms requiring almost every disk galaxy to have a classical bulge at its core. We demonstrate a proof of concept and show early results of our ongoing three- component bulge-disk-bar decomposition of NIR images for a sample of three complementary samples spanning different epochs and different environments (field and cluster). In contrast to most early studies, which only attempt two-component bulge-disk decomposition, we fit three components using GALFIT: a bulge, a disk, and a bar. We show that it is important to include the bar component, as this can significantly lower the bulge-to-total luminosity ratio (_B/T_), in many cases by a factor of two or more, thus effectively changing the Hubble type of a galaxy from early to late. 11footnotetext: The University of Texas, Department of Astronomy, 1 University Station, C1400, Austin, Texas 78712-025922footnotetext: Laboratoire d’Astrophysique, École Polytechnique Fédérale de Lausanne (EPFL), Observatoire, 1290 Sauverny, Switzerland ## 1\. Introduction The formation of galaxies is a classic problem in astrophysics. Contemporary galaxy formation models combine the well-established Lambda-Cold Dark Matter (LCDM) cosmology, which describes behavior of dark matter on very large scales, with baryonic physics to model galaxy formation. In the early Universe, pockets of dark matter decoupled from the Hubble flow, collapsed into virialized halos, and then clustered hierarchically into larger structures. Meanwhile, gas aggregated in the interiors of the halos to form rotating disks, which are the building blocks of galaxies (Navarro & Steinmetz, 2002; Cole et al. 2000). Such disks were destroyed during mergers of their parent halos, leaving behind classical de Vaucouleurs bulges. Spiral disk galaxies formed subsequently as gaseous disks accreted around spheroids (Burkert & Naab, 2004). Troubling inconsistencies exist between real galaxies and LCDM models of galaxy formation. One issue is the angular momentum problem; simulated galaxy disks have smaller scalelengths and, therefore, less specific angular momentum than their counterparts in nature (D’Onghia & Burkert, 2006). A second problem is the severe under prediction in the fraction of galaxies with low bulge-to- total mass ratio ($B/T<$0.2) and of so-called bulgeless galaxies, which lack a classical bulge. Simulated spiral galaxies feature prominent classical bulges in their cores. Such predictions are in glaring contradiction with emerging observations that suggest 15-20% of disk galaxies out to z$\sim$0.3 are bulgeless (Kautsch et al. 2006; Barazza et al. 2007) There are many unanswered questions about the assembly of bulges, the distribution of $B/T$, and the properties of so-called bulgeless galaxies with low $B/T$. How do properties, such as disk scalelengths, mass, kinematics, colors, and star formation histories vary across galaxies of different $B/T$, ranging from bulge-dominated systems to quasi-bulgeless systems? Are quasi- bulgeless systems confined to low mass systems with high specific star formation rates, while classical bulges populate high mass systems? How do the fraction, mass function, and structural properties of galaxies with different $B/T$ vary across environments with different large-scale cosmological overdensities? If environment plays a central part in suppressing bulge formation, then differences would be expected in the properties of bulgeless galaxies in different environments, such as field versus dense galaxy clusters. How does the frequency and properties of galaxies with low $B/T$ as a function of redshift over $z=0.2-0.8$ compare to the recently reported merger history of galaxies over this epoch (Jogee et al 2007)? Answering these questions will help us to understand the reasons behind the apparent failure of LCDM galaxy formation models, and shed light on how galaxies assemble. Progress is possible by observationally constraining properties of enigmatic bulgeless galaxies. A powerful technique for measuring the structural properties (e.g. scalelengths, Sérsic indexes, _B/T_) of galaxies is the decomposition of the 2D light distribution into separate structural components with GALFIT (Peng et al. 2002). Most earlier work has only performed 2D bulge- disk decomposition, but because late-type spirals have been shown to have higher optical bar fractions than early-type galaxies (Barazza et al. 2007), it is important to include the bar when analyzing disk-dominated systems. Bars can contain a significant fraction of light, so failure to account for bars could lead to inflated _B/T_ (Laurikainen et al. 2006). ## 2\. Methodology and Samples We perform three-component decomposition of the 2D galaxy light distribution, while taking the PSF into account, with GALFIT. Since GALFIT utilizes a non- linear least squares algorithm, initial guess parameters are required for each component GALFIT attempts to fit. While reasonable initial guesses can be generated by inspection in many cases with common tools (e.g. IRAF), this is time-consuming and inefficient for large samples. In practice, we break three- component decomposition into three separate invocations of GALFIT. We first perform one and two-component fits to constrain the bulge and disk parameters. The single-component fit models the entire galaxy with only a Sérsic bulge component. In addition to constraining the bulge structural parameters, the total luminosity of the object is also determined. A two-component fit, consisting of a Sérsic bulge and exponential disk, is then made based on the output of the previous fit. If GALFIT is allowed to do an unconstrained two-component bulge-disk fit in a strongly barred galaxy, it will often try to fit the bar by artificially stretching the disk along the bar PA. In order to get physically meaningful two-component fits, we therefore constrain the fit by fixing the position angle and axis ratio (b/a) of the outer disk to values pre-determined by fitting an ellipse to the outermost disk isophote. Finally, a three-component bulge-bar-disk fit is performed, using the two- component fits as initial guesses for the bulge parameters, and fixing the disk b/a and PA as before. Bars are modeled with elongated, low-index Sérsic components using initial guesses for the size and position angle estimated from the images. All objects are subjected to the three-component fits, regardless of whether they appear by eye to possess a bar. If there is independent evidence for an AGN or nuclear cluster, a point source is fitted as a fourth component. In order to decide which of the two or three-component fit is better, a number of criteria are used. 1) If the one or two-component residuals show a bar signature that is removed in the three-component residual, then the three component fit is favored; 2) Structural parameters (scalelength, Sérsic index, b/a) of the bar fit must well behaved; 3) Visual evidence of a strong bar in the input images favors the three-component fit. Weak bars are often not visually prominent, but for such bars, the changes in the disk or bulge parameters, between the two and three component fits, are small; (4) In addition, we test the robustness of the three-component solution by varying the initial guesses to check that the same solution is converged upon. In order to address the questions outlined in §1, we are applying the three- component decomposition to three complementary samples, which span different epochs and different environments (field and cluster): (1) a $z\sim$ 0 sample of $\sim$ 200 galaxies with Hubble types S0 to Sm drawn from the OSU Bright Spiral Galaxy Survey (OSUBSG) (Eskridge et al. 2002) and UKIDSS (McLure et al. 2006); (2) a sample of galaxies in the dense environment of the Coma cluster from our ACS Treasury survey (Carter et al 2007); and a sample of early disk galaxies out $z\sim$ 2 with deep NICMOS imaging (180 orbits). ## 3\. Preliminary Findings For the two-component fits, we have performed consistency checks by testing our decomposition on samples of galaxies with published results from the Millennium Galaxy Catalog (Driver et al. 2007) and the New York University Value-Added Catalog (Blanton et al. 2005) for the Sloan Digital Sky Survey. For three-component fits, we have performed similar tests on a few galaxies drawn from small samples with published three-component bulge-bar-disk fits, drawn from Laurikainen et al. (2006) and Reese et al. (2007). An example of our method is presented in Figure 1, which illustrates the complete three-step decomposition for NGC 4643. We now summarize our preliminary findings: 1. 1. Luminosity is conserved between the two and three-component fits. 2. 2. Modeling the bar in the three-component fits forces a reshuffling of luminosity. Generally, the bulge declines in luminosity, whereas light can be taken from, or added back, to the disk. The reshuffling of light occurs because the two-component model adjusts the bulge and disk accordingly to compensate for the bar, which can include artificially elongating and brightening the bulge. Accounting for the bar returns the bulge and disk parameters closer to their true values. 3. 3. Inclusion of the bar can reduce bulge fractional luminosity _B/T_ by a factor of two or more. Larger changes in bulge luminosities (a factor of 10 or more) occur in cases where a prominent bar influences the two-component fit to very much overstate the bulge luminosity. The bulge-disk fits in such extreme cases underscore the importance of including the bar in 2D luminosity decomposition. 4. 4. The scalelength of the disk is generally unchanged by including the bar. However in a few cases, the two-component disk structure can be erroneous, as in the case of NGC 4643, shown in Figure 1. We have provided a proof of concept of our ongoing three-component bulge-disk- bar decomposition with GALFIT. We are optimistic about our on-going work, which will be described in Weinzirl et al. 2008 (in prep). Figure 1.: Shown is the complete three-step decomposition for NGC 4643. From top to bottom, the rows show the fits from the one, two, and three-component decompositions. The residuals for the one and two-component fit show a distinct bar signature. In Step 2, the fitted disk has an unphysically large scalelength ($335\hbox{${}^{\prime\prime}$}$) that does not match the galaxy. Due to its resulting low surface brightness, the fittted disk is hard to see, and ellipses are drawn to show its PA and b/a. In Step 3, the addition of the bar component restores the disk scalength to a reasonable value. The fit parameters are presented in Table 1. Table 1.: Fit parameters for NGC 4643 Fit | | $r_{e}$ or h (′′) | n | b/a | Position Angle | Fractional light ---|---|---|---|---|---|--- Step 1 | Sersic | 27.90 | 4.44 | 0.80 | -51.08 | 100% Step 2 | Bulge | 23.86 | 4.16 | 0.80 | -51.08 | 34.6% | Disk | 335.88 | 1.0 | 0.84 | 66.94 | 65.4% Step 3 | Bulge | 5.43 | 2.53 | 0.90 | 60.52 | 25.0 % | Disk | 48.22 | 1.0 | 0.84 | 66.94 | 54.1 % | Bar | 21.30 | 0.62 | 0.37 | -45.84 | 20.9 % ### Acknowledgments. TW and SJ acknowledge support from NSF grad AST-0607748, LTSA grant NAG5-13063, and HST-GO-10861 from STScI, which is operated by AURA, Inc., for NASA, under NAS5-26555. ## References * (1) Barazza F. D., Jogee, S., Marinova, I. 2007, ApJ, accepted, arXiv:0710.4674 * (2) Blanton, M. R., et al. 2005, AJ, 129, 2562 * (3) Burkert, A., & Naab, T. 2004, Coevolution of Black Holes and Galaxies, 422 * (4) Carter, D. et al. 2007, ApJ, submitted * (5) Cole, S. M., Baugh, C., Frenk, C., Lacey, C., Benson, A. 2000, MNRAS, 319, 168 * (6) D’Onghia, E. & Burkert, A. 2006, MNRAS, 372, 1525 * (7) Driver, S. P., Liske, J., Graham, A. W., 2007, ApJL, accepted (astro-ph/0701728) * (8) Eskridge, P. B., et al. 2002, ApJS, 143, 73 * (9) Jogee et al. 2007, Proceedings of ”Formation and Evolution of Galaxy Disks”, held in Rome, 1-5 October 2007 (arXiv:0802.3901) * (10) Kautsch, S. J., Grebel, E. K., Barazza, F. D., & Gallagher, J. S., III 2006, A&A, 445, 765 * (11) Laurikainen, E., Salo, H., Buta, R., Knapen, J., Speltincx, T., & Block, D. 2006, AJ, 132, 2634 * (12) McLure, R. J., et al. 2006, MNRAS, 372, 357 * (13) Navarro, J.F., Steinmetz, M. 2000, ApJ, 538, 477 * (14) Peng C. Y., Ho L. C., Impey C. D., Rix H.-W., 2002, AJ, 124, 266 * (15) Reese, A. S., Williams, T. B., Sellwood, J. A., Barnes, E. I., & Powell, B. A. 2007, AJ, 133, 2846

arxiv-papers

2024-09-04T02:48:54.020169

0802.4246

# Physical realization of coupled Hilbert-space mirrors for quantum-state engineering E. S. Kyoseva Department of Physics, Sofia University, James Bourchier 5 blvd., 1164 Sofia, Bulgaria N. V. Vitanov Department of Physics, Sofia University, James Bourchier 5 blvd., 1164 Sofia, Bulgaria Institute of Solid State Physics, Bulgarian Academy of Sciences, Tsarigradsko chaussée 72, 1784 Sofia, Bulgaria B. W. Shore Fachbereich Physik, Technische Universität Kaiserslautern, 67653 Kaiserslautern, Germany Permanent address: 618 Escondido Circle, Livermore, CA 94550, USA ###### Abstract Manipulation of superpositions of discrete quantum states has a mathematical counterpart in the motion of a unit-length statevector in an $N$-dimensional Hilbert space. Any such statevector motion can be regarded as a succession of two-dimensional rotations. But the desired statevector change can also be treated as a succession of reflections, the generalization of Householder transformations. In multidimensional Hilbert space such reflection sequences offer more efficient procedures for statevector manipulation than do sequences of rotations. We here show how such reflections can be designed for a system with two degenerate levels – a generalization of the traditional two-state atom – that allows the construction of propagators for angular momentum states. We use the Morris-Shore transformation to express the propagator in terms of Morris-Shore basis states and Cayley-Klein parameters, which allows us to connect properties of laser pulses to Hilbert-space motion. Under suitable conditions on the couplings and the common detuning, the propagators within each set of degenerate states represent products of generalized Householder reflections, with orthogonal vectors. We propose physical realizations of this novel geometrical object with resonant, near-resonant and far-off-resonant laser pulses. We give several examples of implementations in real atoms or molecules. Householder reflection, quantum state engineering, coherent excitation, degenerate levels ###### pacs: 03.67.Lx; 32.80.Bx; 33.80.Be ## I Introduction Manipulation of discrete quantum states has long held interest, most recently for application to quantum information processing QI . In the simplest realizations one deals with a nondegenerate two-state system (a “two-level atom”AE ; Shore ) and through pulsed resonant coherent excitation produces a specified superposition of the two states, starting from a single state. Following the availability of laser light sources, much attention centered on producing complete excitation, i.e. complete population transfer from the ground state to an excited state Shore . More recently interest has shifted to the production of specific superpositions of the two states, and to producing transitions between superposition states. In the context of quantum information a two-state system serves as a qubit; more general tasks involve multiple quantum states (a qunit), possibly driven by pulsed nonresonant light. The goal of quantum-state manipulation is to realign the statevector $\Psi(t)$ from some given Hilbert-space direction $\Psi(t_{i})$ at an initial time $t_{i}$ to some prescribed state $\Psi(t_{f})$ at a final time $t_{f}$, each of these statevectors being defined by the set of complex-valued probability amplitudes $C_{n}(t)$ associated with a set of quantum states $|\psi_{n}\rangle$. Expressed in matrix form, the goal is to obtain a propagator matrix $\mathsf{U}(t_{f},t_{i})$ that transforms the vector $\mathbf{C}(t)$ of components $C_{n}(t)$, $\mathbf{C}(t_{f})=\mathsf{U}(t_{f},t_{i})\mathbf{C}(t_{i}).$ (1) As noted in the following section, there exists an extensive literature describing analytic solutions to the two-state system for a variety of pulsed- excitation scenarios. There exist also numerous analytic solutions to multistate systems. Here we consider, as a particular generalization of the two-state system, a two-level system involving two sets of degenerate sublevels: a less energetic set of $N$ states $|\psi_{n}\rangle\ (n=1,2,\ldots,N)$ with common energy $E_{N}$ (the _lower_ set), and a more energetic set of $M$ states $|\varphi_{m}\rangle\ (m=1,2,\ldots,M)$ that share energy $E_{M}$ (the _upper_ set). Thus we consider a Hilbert space of dimension $N+M$. For definiteness we here assume that $N\geqq M$. We assume that interactions produced by laser fields can induce direct transitions between lower sublevels and upper sublevels, but not directly within either manifold of states (i.e. we allow electric-dipole transitions only). Because the statevector has unit length at all times, any such change can be regarded as a rotation, and any allowable motion in several dimensions can be decomposed into a succession of two-dimensional rotations, analogous to the three Euler angles that define an arbitrary rotation in Euclidean space. However, length-preserving changes in multi-dimensional space can also be produced by a succession of reflections. When expressed in matrix form these are generalizations of the Householder reflections used in matrix calculus Householder . As has been shown previously, such unitary operations can be implemented very efficiently using coherent excitation techniques Kyoseva ; Ivanov ; as subsequently noted Ivanov2 , the use of quantum Householder reflections (QHR) permits efficient quantum-state engineering of transitions between arbitrary superpositions. The QHR implementation proposed earlier Kyoseva ; Ivanov requires a particular multistate linkage pattern, in which a set of $N$ low-lying degenerate states all link, via radiative interaction, with a _single_ upper state – a generalization of the tripod linkage termed an _N-pod_. Such a linkage pattern occurs with a lower level having angular momentum $J=1$ (three sublevels) excited to an upper level having $J=0$ (a single sublevel) but the pattern is difficult to realize for more than 3 lower states. It is therefore desirable to extend the QHR technique to more general linkage patterns. The present manuscript describes a procedure that allows implementation of the QHR when there are multiple states in the upper set. Such an extension makes possible the application of QHR to arbitrary angular momentum states, as occur with free atoms and molecules. As we show, the propagator for such situations is not a single QHR, but a product of QHRs, moreover with orthogonal vectors. The expression for the resulting propagator has a clear geometric interpretation as the effect of a succession of reflections, i.e. _coupled mirrors_. The key to this extension of the QHR is a transformation of the underlying basis states, the so-called Morris-Shore (MS) transformation MS ; Vitanov MW ; other MS . This replaces the original system, with its multiple linkages between states, by a set of independent nondegenerate two-state systems, thereby allowing us to utilize the considerable literature of analytic solutions to two-state systems and produce analytic solutions for degenerate multistate systems Kyoseva . In this paper we present the solution of the degenerate two-level problem in a simple closed form involving sums of projectors of MS states. These expressions are useful for deriving analytical solutions, generalizations of known two-state solutions, for systems having degenerate levels. For certain conditions, wherein the transition probabilities between states from different sets vanish, the propagator for states of the lower set is given by a product of QHRs with orthogonal vectors, each vector being a bright state; a similar property applies to the upper set. We discuss properties of this novel geometric object, the coupled mirrors. To illustrate the procedure we develop a useful explicit analytic formalism for two upper states and present some examples. ## II The degenerate two-level system ### II.1 Two non-degenerate states We consider the controlled alteration of a multistate system, expressed as a redirection of the statevector, induced by a set of laser pulses. The simplest example of coherent excitation occurs when there are just two nondegenerate states, indexed 1 and 2. The time-dependent Schrödinger equation prescribes the changes in Hilbert space as (here and henceforth we set $\hbar=1$, thereby making no distinction between energy and frequency units) $i\frac{d}{dt}\mathbf{C}(t)=\mathsf{H}(t)\mathbf{C}(t),$ (2) where the column vector $\mathbf{C}(t)$ has the complex-valued probability amplitudes $C_{n}(t)$ as elements, $\mathbf{C}(t)=[C_{1}(t),C_{2}(t)]^{T}$. Commonly one neglects in the Hamiltonian $\mathsf{H}(t)$ elements that vary rapidly compared with the characteristic response times of the system, the so- called rotating-wave approximation (RWA) Shore . Then the Hamiltonian matrix, with suitable choice of energy zero-point, reads $\mathsf{H}(t)=\left[\begin{array}[]{cc}0&\frac{{}_{1}}{{}^{2}}\Omega(t)\\\ \frac{{}_{1}}{{}^{2}}\Omega(t)^{\ast}&\Delta(t)\end{array}\right].$ (3) Here the slowly-varying function $\Omega(t)$, the Rabi frequency, quantifies the coupling between the two states. For a laser-driven electric-dipole transition in an atom or a molecule, $\Omega(t)$ is proportional to the transition dipole moment and the electric-field envelope. The detuning $\Delta(t)$ measures the frequency offset of the carrier laser frequency $\omega$ from the Bohr transition frequency $\omega_{0}$, $\Delta=\omega_{0}-\omega$. A time dependence in the detuning can be introduced by both $\omega$ (due to chirping femto ) and $\omega_{0}$ (due to Stark and Zeeman shifts). When the transition is exactly resonant the associated detuning vanishes, $\Delta=0$. Then the response depends only upon the initial conditions and upon the time-integrated Rabi frequency (the temporal pulse area), $A(t)=\int_{0}^{t}dt^{\prime}\,|\Omega(t^{\prime})|.$ (4) At all times the probability amplitudes are expressible as trigonometric functions of $A(t)$; when $A(t)$ is some even-integer multiple of $\pi$ (e.g., a $2\pi$__ pulse), the probabilities repeat their initial values. Thus the response, for resonant excitation, does not depend on any details of the pulse shape. Of particular use are the $\pi$ pulses, which produce complete population inversion between the two states, and half-$\pi$ pulses that create an equal coherent superposition of the two states. Resonant pulses having precise temporal area have had wide application, most notably in nuclear magnetic resonance NMR and coherent atomic excitation Shore . They are now a common tool in quantum information processing QI . A variety of nonresonant ($\Delta\neq 0$) pulses also lead to exact analytic expressions for the probability amplitudes $C_{1}(t)$ and $C_{2}(t)$. Amongst the soluble two-state models are the models of Rabi Rabi , Landau-Zener LZ , Rosen-Zener RZ , Allen-Eberly AE ; Hioe , Bambini-Berman BB , Demkov-Kunike DK , Demkov Demkov , Nikitin Nikitin , and Carroll-Hioe CH . Methods for approximate solutions are also available, such as the perturbation theory, the adiabatic approximation STIRAP , the Magnus approximation Mag54 , and the Dykhne-Davis-Pechukas approximation DDP . The latter, in particular, is a very useful tool for deriving very accurate approximations in various cases of interest, e.g., transform-limited Vasilev-Gaussian and chirped Gaussian pulses Vasilev-Gaussian-chirped . The two-state dynamics acquires interesting new features when the pulsed field possesses some symmetries symmetry , or when it is a sequence of identical pulses (pulse train) train . The two-state atom provides a very basic model of coherent excitation, widely used because of the relative simplicity with which it can be treated mathematically and fabricated experimentally. However, with this mastery has come interest in more general systems, involving more than two discrete quantum states. It is with a class of these that the present paper deals. Specifically, we describe a straightforward technique that allows efficient redirection of the statevector for such a system by means of laser pulses, and the use of the many exact and approximate solutions available for two-state systems. ### II.2 Multiple states as two degenerate levels An important extension is the linear chain of $N$ states, each state being linked only to its nearest neighbor. The simplest of this is the three-state chain, involving two independent pulses, the mathematics of which has been extensively reported three state , as have lengthier chains chains . Here we consider an extension in which each of the states of the two-state system is replaced by a degenerate set of states that together form an energy _level_. Such situations occur commonly when one deals with angular-momentum eigenstates, as happens with the electronic structure of atoms in free space. For a given angular momentum $J$ there are $2J+1$ magnetic sublevels that, in the absence of an external electric or magnetic field, all have the same energy. We shall consider the possibility that there be no nonzero elements of the Hamiltonian matrix linking any states of the same energy; nonzero couplings occur only between states from the lower set and the states in the upper set. We make no restriction on the number of states that connect with any single state. This generalizes the usual situation of angular momentum states excited by electric-dipole radiation, when any state can link to no more than three other states (the selection rules $\Delta m=-1,0,+1$). Figure 1 illustrates the linkage pattern we consider. Figure 1: (Color online) Schematic linkage pattern for multistate system consisting of two coupled sets of degenerate levels. We shall assume that we can control the magnitude and the phase of each interaction (e.g. each electric field envelope), and that _every Rabi frequency has the same time dependence_ , which we write as $\Omega_{mn}(t)=2V_{mn}f(t),$ (5) where $V_{mn}$ is a complex-valued constant and $f(t)$ a real-valued pulse- shaped function of time bounded by unity. We assume that the duration of the pulse is shorter than any decoherence time, so that the dynamics is governed by the Schrödinger equation. Such is the situation for excitations of free atoms and molecules by picosecond or femtosecond pulses femto . We shall also assume that each transition has exactly the same detuning $\Delta$. In the simplest situations, those associated with angular-momentum states, three distinct fields can be distinguished by their polarization, while sharing a common carrier frequency. By utilizing three independent polarization directions of a laser pulse we can ensure three linkages with common time dependence. More general linkage patterns, still within the model described here, are also possible other MS . ## III Exact analytical solution ### III.1 The RWA Hamiltonian In the RWA the Schrödinger equation provides a prescription for the time dependence of the probability amplitudes $C_{n}(t)$. Written in matrix form it is $i\frac{d}{dt}\mathbf{C}(t)=\mathsf{H}(t)\mathbf{C}(t),$ (6) where the column vector $\mathbf{C}(t)$ has $C_{n}(t)$ as elements. We shall arrange these elements with those of the lower set first, followed by the upper set. Then the RWA Hamiltonian has the block structure $\mathsf{H}(t)=\begin{bmatrix}\mathsf{O}&\mathsf{V}f(t)\\\ \mathsf{V}^{\dagger}f(t)&\mathsf{D}(t)\end{bmatrix}.$ (7) Here $\mathsf{O}$ denotes the $N$-dimensional square null matrix; the zeros signify that the lower states do not interact with each other and that they all have the same energy, which we take as the zero-point of our energy scale. The matrix $\mathsf{V}$ is an $(N\times M)$-dimensional matrix whose elements $V_{nm}$ are the magnitudes of the couplings between the lower and the upper states; $\mathsf{V}^{{\dagger}}$ is its hermitian conjugate. Lastly, $\mathsf{D}(t)$ is an $M$-dimensional square diagonal matrix whose elements are all equal to the shared detuning $\Delta(t)$, $\mathsf{D}(t)=\Delta(t)\mathsf{I},$ (8) where $\mathsf{I}$ is the unit matrix. The diagonal nature of $\mathsf{D}$ indicates the absence of interaction of the upper states amongst themselves. We allow the detuning $\Delta(t)$ to vary with time, bearing in mind use of known analytic solutions to the two-state model with frequency-swept detuning. It proves useful to write the matrix of interactions $\mathsf{V}$ as a row vector of $N$-dimensional column vectors $\left|V_{n}\right\rangle=\left[V_{1n},V_{2n},\ldots,V_{Nn}\right]^{T}$, $\mathsf{V}=\begin{bmatrix}V_{11}&V_{12}&\cdots&V_{1M}\\\ V_{21}&V_{22}&\cdots&V_{2M}\\\ \cdots&\cdots&\ddots&\vdots\\\ V_{N1}&V_{N2}&\cdots&V_{NM}\end{bmatrix}=[\left|V_{1}\right\rangle,\left|V_{2}\right\rangle,\ldots,\left|V_{M}\right\rangle].$ (9) ### III.2 The Morris-Shore transformation A significant property of the linkage pattern shown in Fig. 1 is that a transformation of Hilbert-space coordinates – the Morris-Shore (MS) transformation – reduces the dynamics to that of a set of independent two- state systems, together with decoupled states MS . When $N>M$, as is the case with Fig. 1, the $N_{0}=N-M$ additional states are a part of the lower-level manifold. They have no connection with excited states, and hence they cannot produce excitation followed by fluorescence; they are termed _dark states_ dark state , by contrast to the states that can produce excitation (and thence fluorescence), the _bright states_. The number of bright states is the lesser of $N$ and $M$, in this case $M$. Figure 2 shows the new linkage pattern in the MS basis. Figure 2: (Color online) Linkages of Fig. 1 transformed by the Morris-Shore transformation into a set of $M$ independent nondegenerate two-state systems, with couplings $\lambda_{n}$, and a set of $N_{0}=N-M$ decoupled (dark) states. Because, by assumption, all elements of the interaction matrix share a common time dependence, the required MS transformation is achieved by a constant unitary transformation $\mathsf{S}=\left[\begin{array}[]{cc}\mathsf{A}&\mathsf{O}\\\ \mathsf{O}&\mathsf{B}\end{array}\right].$ (10) Here $\mathsf{A}$ is a unitary $N$-dimensional square matrix ($\mathsf{AA}^{\dagger}=\mathsf{A}^{\dagger}\mathsf{A}=\mathsf{I}$) which transforms the lower set of states, and $\mathsf{B}$ is a unitary $M$-dimensional square matrix ($\mathsf{B}\mathsf{B}^{\dagger}=\mathsf{B}^{\dagger}\mathsf{B}=\mathsf{I}$) which transforms the upper set. This transformation casts the dynamics into the MS basis, with new MS bright lower states $\\{|\alpha_{m}\rangle\\}_{m=1}^{M}$, dark lower states $\\{|\gamma_{k}\rangle\\}_{k=1}^{N_{0}}$, and upper states $\\{|\beta_{m}\rangle\\}_{m=1}^{M}$. The transformed MS Hamiltonian has the form $\widetilde{\mathsf{H}}(t)=\mathsf{S}\mathsf{H}(t)\mathsf{S}^{\dagger}=\left[\begin{array}[]{cc}\mathsf{O}&\widetilde{\mathsf{V}}f(t)\\\ \widetilde{\mathsf{V}}^{\dagger}f(t)&\mathsf{D}(t)\end{array}\right],$ (11) where $\widetilde{\mathsf{V}}=\mathsf{A}\mathsf{V}\mathsf{B}^{\dagger}.$ (12) Because the matrices $\widetilde{\mathsf{V}}\widetilde{\mathsf{V}}^{\dagger}=\mathsf{A}\mathsf{V}\mathsf{V}^{\dagger}\mathsf{A}^{\dagger}$ and $\widetilde{\mathsf{V}}^{\dagger}\widetilde{\mathsf{V}}=\mathsf{B}\mathsf{V}^{\dagger}\mathsf{V}\mathsf{B}^{\dagger}$ must be diagonal (possibly after removing null rows and rearanging the basis states), the matrices $\mathsf{A}$ and $\mathsf{B}$ are defined by the conditions that they diagonalize $\mathsf{V}\mathsf{V}^{\dagger}$ and $\mathsf{V}^{\dagger}\mathsf{V}$, respectively. $\mathsf{V}^{\dagger}\mathsf{V}$ has $M$ generally nonzero eigenvalues $\lambda_{n}^{2}$ ($n=1,2,\ldots,M$). The $N$-dimensional matrix $\mathsf{V}\mathsf{V}^{\dagger}$ has the same $M$ eigenvalues as $\mathsf{V}^{\dagger}\mathsf{V}$ and additional $N_{0}=N-M$ zero eigenvalues. From the vector form (9) of $\mathsf{V}$ we obtain $\displaystyle\mathsf{VV}^{\dagger}$ $\displaystyle=$ $\displaystyle\sum_{n=1}^{M}\left|V_{n}\right\rangle\left\langle V_{n}\right|,$ (13a) $\displaystyle\mathsf{V}^{\dagger}\mathsf{V}$ $\displaystyle=$ $\displaystyle\left[\begin{array}[]{cccc}\left\langle V_{1}|V_{1}\right\rangle&\left\langle V_{1}|V_{2}\right\rangle&\cdots&\left\langle V_{1}|V_{M}\right\rangle\\\ \left\langle V_{2}|V_{1}\right\rangle&\left\langle V_{2}|V_{2}\right\rangle&\cdots&\left\langle V_{2}|V_{M}\right\rangle\\\ \vdots&\vdots&\ddots&\vdots\\\ \left\langle V_{M}|V_{1}\right\rangle&\left\langle V_{M}|V_{2}\right\rangle&\cdots&\left\langle V_{M}|V_{M}\right\rangle\end{array}\right].$ (13f) $\mathsf{V}^{\dagger}\mathsf{V}$ is the Gram matrix Gantmacher for the set of vectors $|V_{n}\rangle$, ($n=1,2,\ldots,M$). Thus if all these vectors are linearly independent then $\det\mathsf{V}^{\dagger}\mathsf{V}\neq 0$ and all eigenvalues of $\mathsf{V}^{\dagger}\mathsf{V}$ are nonzero Gantmacher ; however, this assumption of independence is unnecessary. The MS Hamiltonian (11) has the explicit form $\widetilde{\mathsf{H}}(t)=\left[\begin{array}[]{c|c}\mathsf{O}&\mathsf{O}\\\ \hline\cr\mathsf{O}&\begin{array}[]{cccccccc}0&0&\cdots&0&\lambda_{1}f(t)&0&\cdots&0\\\ 0&0&\cdots&0&0&\lambda_{2}f(t)&\cdots&0\\\ \vdots&\vdots&\ddots&\vdots&\vdots&\vdots&\ddots&\vdots\\\ 0&0&\cdots&0&0&0&\cdots&\lambda_{M}f(t)\\\ \lambda_{1}f(t)&0&\cdots&0&\Delta&0&\cdots&0\\\ 0&\lambda_{2}f(t)&\cdots&0&0&\Delta&\cdots&0\\\ \vdots&\vdots&\ddots&\vdots&\vdots&\vdots&\ddots&\vdots\\\ 0&0&\cdots&\lambda_{M}f(t)&0&0&\cdots&\Delta\end{array}\end{array}\right].$ (14) The structure of $\widetilde{\mathsf{H}}(t)$ shows that in the MS basis the dynamics decomposes into sets of $N_{0}$ decoupled single states and $M$ independent two-state systems $|\alpha_{n}\rangle\leftrightarrow|\beta_{n}\rangle$ ($n=1,2,\ldots,M$), each composed of a lower state $|\alpha_{n}\rangle$ and an upper state $|\beta_{n}\rangle$, and driven by the two-state RWA Hamiltonians $\widetilde{\mathsf{h}}_{n}=\left[\begin{array}[]{cc}0&\lambda_{n}f(t)\\\ \lambda_{n}f(t)&\Delta\end{array}\right],\quad(n=1,2,\ldots,M).$ (15) Each of these two-state Hamiltonians has the same detuning $\Delta$, but they differ in the couplings $\lambda_{n}$. Each of the new lower states $|\alpha_{n}\rangle$ is an eigenstate of $\mathsf{V}\mathsf{V}^{\dagger}$ that corresponds to a specific eigenvalue $\lambda_{n}^{2}$ of $\mathsf{V}\mathsf{V}^{\dagger}$. Similarly the new upper state $|\beta_{n}\rangle$ is the eigenstate of $\mathsf{V}^{\dagger}\mathsf{V}$, corresponding to the same eigenvalue $\lambda_{n}^{2}$. The square root of this common eigenvalue, $\lambda_{n}$, represents the coupling (half the Rabi frequency) in the respective independent MS two-state system $|\alpha_{n}\rangle\leftrightarrow|\beta_{n}\rangle$. The $N_{0}$ zero eigenvalues of $\mathsf{V}\mathsf{V}^{\dagger}$ correspond to decoupled (dark) states in the lower set (we assume thoughout that $N\geqq M$; therefore, dark states, if any, are in the lower set). The dark states are decoupled from the dynamical evolution because they are driven by one-dimensional null Hamiltonians. ### III.3 The propagator in the original basis The eigenvectors of $\mathsf{V}\mathsf{V}^{\dagger}$ and $\mathsf{V}^{\dagger}\mathsf{V}$ form the transformation matrices $\mathsf{A}$ and $\mathsf{B}$, $\mathsf{A}=\left[\begin{array}[]{c}\langle\gamma_{1}|\\\ \vdots\\\ \langle\gamma_{N_{0}}|\\\ \langle\alpha_{1}|\\\ \vdots\\\ \langle\alpha_{M}|\end{array}\right],\quad\mathsf{B}=\left[\begin{array}[]{c}\langle\beta_{1}|\\\ \vdots\\\ \langle\beta_{M}|\end{array}\right].$ (16) They obey the completeness relations $\displaystyle\sum_{n=1}^{M}|\alpha_{n}\rangle\langle\alpha_{n}|+\sum_{k=1}^{N_{0}}|\gamma_{k}\rangle\langle\gamma_{k}|=\mathsf{I},$ (17a) $\displaystyle\sum_{n=1}^{M}|\beta_{n}\rangle\langle\beta_{n}|=\mathsf{I}.$ (17b) The propagator $\widetilde{\mathsf{u}}_{n}(t_{f},t_{i})$ for the independent MS two-state system $|\alpha_{n}\rangle\leftrightarrow|\beta_{n}\rangle$ ($n=1,2,\ldots,M$) is unitary and therefore can be parameterized in terms of the complex-valued Cayley-Klein parameters $a_{n}$ and $b_{n}$, $\widetilde{\mathsf{u}}_{n}(t_{f},t_{i})=\begin{bmatrix}a_{n}&b_{n}\\\ -b_{n}^{\ast}e^{-i\delta}&a_{n}^{\ast}e^{-i\delta}\end{bmatrix},$ (18) where $\left|a_{n}\right|^{2}+\left|b_{n}\right|^{2}=1$ and $\delta=\int_{t_{i}}^{t_{f}}\Delta(t^{\prime})dt^{\prime}$. In what follows we express the desired control of statevector motion in terms of constraints on these two-state Cayley-Klein parameters. The unimportant phase factor $e^{-i\delta}$ originates from the chosen representation of the Hamiltonian (7), which facilitates the application of the MS transformation. In the interaction representation, where the diagonal elements are nullified and the detunings appear in phase factors multiplying the couplings, the factor $e^{-i\delta}$ disappears. (Note: This phase factor has been omitted in Eqs. (17), (18) and (20) of Ref. Kyoseva , where it should have been associated with $a^{\ast}$ and $b^{\ast}$; however, it did not affect any result there.) By taking into account the MS propagators (18) for the two-state MS systems, the ordering of the states (16), and the MS Hamiltonian (14), we write the propagator of the full system in the MS basis as $\widetilde{\mathsf{U}}=\left[\begin{array}[]{c|c}\mathsf{I}&\mathsf{O}\\\ \hline\cr\mathsf{O}&\begin{array}[]{cccccccc}a_{1}&0&\cdots&0&b_{1}&0&\cdots&0\\\ 0&a_{2}&\cdots&0&0&b_{2}&\cdots&0\\\ \vdots&\vdots&\ddots&\vdots&\vdots&\vdots&\ddots&\vdots\\\ 0&0&\cdots&a_{M}&0&0&\cdots&b_{M}\\\ -b_{1}^{\ast}e^{-i\delta}&0&\cdots&0&a_{1}^{\ast}e^{-i\delta}&0&\cdots&0\\\ 0&-b_{2}^{\ast}e^{-i\delta}&\cdots&0&0&a_{2}^{\ast}e^{-i\delta}&\cdots&0\\\ \vdots&\vdots&\ddots&\vdots&\vdots&\vdots&\ddots&\vdots\\\ 0&0&\cdots&-b_{M}^{\ast}e^{-i\delta}&0&0&\cdots&a_{M}^{\ast}e^{-i\delta}\end{array}\end{array}\right].$ (19) It is straightforward to show that the propagator in the original basis $\mathsf{U}=\mathsf{S}^{\dagger}\widetilde{\mathsf{U}}\mathsf{S}$ reads $\mathsf{U}=\left[\begin{array}[]{cc}\mathsf{U}_{N}&\mathsf{U}_{NM}\\\ \mathsf{U}_{MN}&\mathsf{U}_{M}\end{array}\right],$ (20a) where $\displaystyle\mathsf{U}_{N}$ $\displaystyle=$ $\displaystyle\sum_{n=1}^{M}a_{n}|\alpha_{n}\rangle\langle\alpha_{n}|+\sum_{k=1}^{N_{0}}|\gamma_{k}\rangle\langle\gamma_{k}|,$ (20b) $\displaystyle\mathsf{U}_{NM}$ $\displaystyle=$ $\displaystyle\sum_{n=1}^{M}b_{n}|\alpha_{n}\rangle\langle\beta_{n}|,$ (20c) $\displaystyle\mathsf{U}_{MN}$ $\displaystyle=$ $\displaystyle-e^{-i\delta}\sum_{n=1}^{M}b_{n}^{\ast}|\beta_{n}\rangle\langle\alpha_{n}|,$ (20d) $\displaystyle\mathsf{U}_{M}$ $\displaystyle=$ $\displaystyle e^{-i\delta}\sum_{n=1}^{M}a_{n}^{\ast}|\beta_{n}\rangle\langle\beta_{n}|.$ (20e) The propagator $\mathsf{U}_{N}$ connects states within the lower set, $\mathsf{U}_{M}$ connects states within the upper set, and $\mathsf{U}_{NM}$ and $\mathsf{U}_{MN}$ mix states from the lower and upper sets. By using the completeness relations (17a) and (17b) we find $\displaystyle\mathsf{U}_{N}$ $\displaystyle=$ $\displaystyle\mathsf{I}+\sum_{n=1}^{M}(a_{n}-1)|\alpha_{n}\rangle\langle\alpha_{n}|,$ (21a) $\displaystyle\mathsf{U}_{M}$ $\displaystyle=$ $\displaystyle e^{-i\delta}\left[\mathsf{I}+\sum_{n=1}^{M}(a_{n}^{\ast}-1)|\beta_{n}\rangle\langle\beta_{n}|\right].$ (21b) Hence the propagator $\mathsf{U}_{N}$ does not depend on the decoupled states $|\gamma_{k}\rangle$ $(k=1,2,\ldots,N_{0})$. This has to be expected because, owing to their degeneracy, the choice of the decoupled states is not unique: any superposition of them is also a zero-eigenvalue eigenstate of $\mathsf{V}\mathsf{V}^{\dagger}$. Because the dynamics in the original basis must not depend on such a leeway of choice, the full propagator $\mathsf{U}$ must not depend on the decoupled states at all, and this is indeed the case. ### III.4 Analytical solutions for degenerate levels Equation (20a) expresses the dynamics of the degenerate two-level system in terms of the dynamics of the $M$ independent non-degenerate two-state systems, each with the same detuning $\Delta(t)$ and pulse shape $f(t)$ of the couplings but with different coupling strengths $\lambda_{n}$. Therefore, Eq. (20a) allows one to generalize any analytical non-degenerate two-state solution to a pair of degenerate levels. Recently, such a generalization of the Landau-Zener model to two degenerate levels has been presented Vasilev . This generalization displays several interesting properties, for instance, not all transition probabilities between degenerate states are defined for infinite time duration. Other analytical solutions involving two degenerate levels have been derived for five chainwise-coupled states in $M$ or $W$ linkage configurations Vitanov MW . We present below another interesting aspect of the solution (20): its geometrical nature. ## IV Quantum-state reflections ### IV.1 Coupled reflections Of particular significance is the special case when the Cayley-Klein parameters $b_{n}$ are all equal to zero, $b_{n}=0\quad(n=1,2,\ldots,M);$ (22) then all transition probabilities in the MS basis, as well as these in the original basis from the lower set to the upper set, vanish, i.e. $\mathsf{U}_{NM}=\mathsf{U}_{MN}=0$. Equation (22) implies that $\left|a_{n}\right|=1$, or $a_{n}=e^{i\phi_{n}}\quad(n=1,2,\ldots,M),$ (23) for all MS two-state systems, where $\phi_{n}$ are arbitrary phases. After substituting Eqs. (22) and (23) in Eqs. (20), the propagator in the original basis reads $\displaystyle\mathsf{U}$ $\displaystyle=$ $\displaystyle\left[\begin{array}[]{cc}\mathsf{U}_{N}&\mathsf{O}\\\ \mathsf{O}&\mathsf{U}_{M}\end{array}\right],$ (24c) $\displaystyle\mathsf{U}_{N}$ $\displaystyle=$ $\displaystyle\mathsf{I}+\sum_{n=1}^{M}(e^{i\phi_{n}}-1)|\alpha_{n}\rangle\langle\alpha_{n}|,$ (24d) $\displaystyle\mathsf{U}_{M}$ $\displaystyle=$ $\displaystyle e^{-i\delta}\left[\mathsf{I}+\sum_{n=1}^{M}(e^{-i\phi_{n}}-1)|\beta_{n}\rangle\langle\beta_{n}|\right],$ (24e) where the propagator $\mathsf{U}_{N}$ operates in the lower set of states and $\mathsf{U}_{M}$ acts in the upper set. Taking into account that the coupled lower states $|\alpha_{n}\rangle\ (n=1,2,\ldots,M)$ are orthonormal basis vectors, i.e. $\langle\alpha_{n}|\alpha_{k}\rangle=\delta_{nk}$, as are the upper MS states $|\beta_{n}\rangle$ $(n=1,2,\ldots,M)$, we rewrite $\mathsf{U}_{N}$ and $\mathsf{U}_{M}$ as the products $\displaystyle\mathsf{U}_{N}$ $\displaystyle=$ $\displaystyle\prod_{n=1}^{M}\mathsf{M}(\alpha_{n},\phi_{n}),$ (25a) $\displaystyle\mathsf{U}_{M}$ $\displaystyle=$ $\displaystyle e^{-i\delta}\prod_{n=1}^{M}\mathsf{M}(\beta_{n},-\phi_{n}),$ (25b) where $\mathsf{M}(\nu;\phi)=\mathsf{I}+(e^{i\phi}-1)|\nu\rangle\langle\nu|.$ (26) These individual matrices are, in fact, generalizations of the matrices used to produce Householder reflection in matrix computations Householder . We refer to them as _generalized quantum Householder reflection_ (QHR) operators Kyoseva ; Ivanov ; Ivanov2 . The generalized QHR operator (26) is unitary, $\mathsf{M}(\nu;\phi)^{-1}=\mathsf{M}(\nu;\phi)^{\dagger}=\mathsf{M}(\nu;-\phi)$, and its determinant has unit magnitude, $\det\mathsf{M}(\nu;\phi)=e^{i\phi}$. For $\phi=0$ the QHR operator reduces to the identity, $\mathsf{M}(\nu;0)=\mathsf{I}$, while for $\phi=\pi$, the QHR operator reduces to the standard reflection, $\mathsf{M}(\nu;\pi)=\mathsf{I}-2|\nu\rangle\langle\nu|.$ (27) The orthogonality of the QHR vectors $|\alpha_{n}\rangle$ in Eq. (25a) automatically ensures the commutation of the QHRs, $\left[\mathsf{M}(\alpha_{n},\phi_{n}),\mathsf{M}(\alpha_{m},\phi_{m})\right]=0.$ (28) Therefore their ordering in the product in Eq. (25a) is unimportant. The same argument applies to Eq. (25b). The importance of QHR derives from the fact that any $N$-dimensional unitary matrix can be decomposed into a set of at most $N$ generalized QHRs Ivanov . The resulting ease with which pulse sequences can be designed to realize the QHR therefore enables one to synthesize any desired unitary transformation of a qunit state, for example a quantum Fourier transform Ivanov or transition between any two pure or mixed qunit states Ivanov2 . ### IV.2 Special case: orthogonal interaction vectors In the special case when the interaction vectors $|V_{n}\rangle$ are _orthogonal_ , $\langle V_{m}|V_{n}\rangle=|V_{n}|^{2}\delta_{mn}$, the MS eigenvalues and the MS states simplify greatly. Then the matrix $\mathsf{V}^{\dagger}\mathsf{V}$ of Eq. (13f) becomes diagonal, and hence its eigenvalues are $\lambda_{n}^{2}=\left|V_{n}\right|^{2}$ ($n=1,2,\ldots,M$). Moreover, the eigenstates of $\mathsf{V}^{\dagger}\mathsf{V}$ – the MS states in the upper set – coincide with the original states, $|\beta_{n}\rangle\equiv|\varphi_{n}\rangle,\quad(n=1,2,\ldots,M).$ (29) The coupled MS states in the lower set – the eigenstates of $\mathsf{VV}^{\dagger}$ – are readily found from Eq. (13a), $|\alpha_{n}\rangle=\frac{1}{\left|V_{n}\right|}\left|V_{n}\right\rangle\equiv|\hat{V}_{n}\rangle.$ (30) The propagator $U_{N}$ in the lower set is a product of QHRs, with the normalized interaction vectors $|\hat{V}_{n}\rangle$ serving as QHR vectors, whereas the propagator $\mathsf{U}_{M}$ in the upper set is a phase gate, $\displaystyle\mathsf{U}_{N}$ $\displaystyle=$ $\displaystyle\prod_{n=1}^{M}\mathsf{M}(\hat{V}_{n},\phi_{n}),$ (31a) $\displaystyle\mathsf{U}_{M}$ $\displaystyle=$ $\displaystyle e^{-i\delta}\sum_{n=1}^{M}e^{-i\phi_{n}}|\varphi_{n}\rangle\langle\varphi_{n}|.$ (31b) The advantage of having orthogonal interaction vectors $|V_{n}\rangle$ is that they serve as QHR vectors. The implication is that in order to construct a pre-selected coupled-QHR transformation (31a), the required couplings are directly obtained from Eq. (30). Otherwise, in the general case of non- orthogonal interaction vectors, a set of pre-selected QHR vectors $|\alpha_{n}\rangle$ ($n=1,2,\ldots,M$), defined as the eigenvectors of $\mathsf{VV}^{\dagger}$, demand numerical derivation of the required couplings from Eq. (13a). ### IV.3 Realizations #### IV.3.1 Off-resonant hyperbolic-secant pulses The condition (23) can be realized with the Rosen-Zener model RZ ; Kyoseva , which assumes constant detuning and hyperbolic-secant time dependence $f(t)$ for the couplings, with pulse duration $T$, $\displaystyle f(t)$ $\displaystyle=$ $\displaystyle\mbox{sech}(t/T),$ (32a) $\displaystyle\Delta(t)$ $\displaystyle=$ $\displaystyle\text{const}.$ (32b) The independent MS two-state systems share the same detuning $\Delta$ and pulse shape $f(t)$, but have different MS couplings $\lambda_{n}$. For the Rosen-Zener model the Cayley-Klein parameters $a_{n}$ ($n=1,2,...,M$) read RZ ; Kyoseva $a_{n}=\frac{\Gamma^{2}\left(\frac{1}{2}+\frac{1}{2}i\Delta T\right)}{\Gamma\left(\frac{1}{2}+\lambda_{n}T+\frac{1}{2}i\Delta T\right)\Gamma\left(\frac{1}{2}-\lambda_{n}T+\frac{1}{2}i\Delta T\right)},$ (33) where $\Gamma(z)$ is Euler’s $\Gamma$-function. Using the reflection formula $\Gamma(1/2+z)\Gamma(1/2-z)=\pi/\cos\pi z$, we find $\left|a_{n}\right|^{2}=1-\frac{\sin^{2}(\pi\lambda_{n}T)}{\cosh^{2}\left(\frac{1}{2}\pi\Delta T\right)}.$ (34) It follows that the condition $\left|a_{n}\right|=1$ is satisfied when $\lambda_{n}T=l$ $(l=0,1,2,\ldots)$. The phase $\phi_{n}$ of $a_{n}=e^{i\phi_{n}}$ depends on the detuning $\Delta$, but not on the corresponding coupling $\lambda_{n}$, and for an arbitrary integer $l$ we find Kyoseva $\phi_{n}=2\arg\prod_{k=0}^{l-1}\left[\Delta T+i\left(2k+1\right)\right].$ (35) Hence the QHR phase $\phi_{n}$ can be produced by an appropriate choice of the detuning $\Delta$. This result shows that even though the couplings for the MS two-state systems are not the same, the phases $\phi_{n}$ of the Cayley-Klein parameters $a_{n}$ coincide, $\phi_{n}\equiv\phi$. This feature is unique for the sech pulse. For other non-resonant pulses, e.g. Gaussian Vasilev-Gaussian , the phase $\phi_{n}$ would depend also on the coupling and therefore will be generally different for each MS two-state system. #### IV.3.2 Resonant pulses For exact resonance ($\Delta=0$), the Cayley-Klein parameter $a_{n}$ reads $a_{n}=\cos(A_{n}/2)$, where the pulse area is $A_{n}=2\lambda_{n}\int_{-\infty}^{\infty}f(t)dt$. When the pulse area is $A_{n}=2(2l+1)\pi$ ($l=0,1,2,\ldots$), the phase $\phi_{n}$ is equal to $\pi$; hence we obtain a physical realization for the standard QHR (27). This result is not resticted to the sech pulse (32a) but it is valid for any pulse shape with such an area Kyoseva ; Ivanov . When the pulse areas are multiples of $4\pi$, the phases $\phi_{n}$ vanish, $a_{n}=1$, and the corresponding QHRs reduce to the identity. Resonant pulses therefore do not produce variable QHR phases $\phi_{n}$, which can be used as free parameters. #### IV.3.3 Far-off-resonant pulses Far-off-resonant pulses provide the opportunity for easy adjustment of the phase $\phi_{n}$, albeit only approximately. Then the condition (23) is fulfilled automatically because of the smallness of the transition probabilities in each MS system. Specifically, if the common detuning $\Delta$ exceeds sufficiently much the largest MS coupling $\lambda_{n}$, then all transition probabilities will be negligibly small, $\left|b_{n}\right|\ll 1$. By adiabatic elimination of each upper MS state one finds $\phi_{n}\approx\frac{\lambda_{n}^{2}}{\Delta}\int_{-\infty}^{\infty}f^{2}(t)dt.$ (36) For a sech pulse the integral is equal to $2T$, and for a Gaussian to $\sqrt{\pi/2}$. Because each MS coupling $\lambda_{n}$ is generally different, each phase $\phi_{n}$ will also be different. Any desired phase $\phi_{n}$, or a set of such phases, can easily be produced by choosing the original couplings $V_{mn}$, and hence the MS couplings $\lambda_{n}$, appropriately. ## V Two degenerate upper states Above, we described the dynamics of the degenerate two-level quantum system in the general case when the lower and upper levels had arbitrary degeneracies, $N$ and $M$ respectively. In this section, we will illustrate these results with a specific example: when the upper set consists of just two degenerate states, i.e. $M=2$. This case is insteresting because of the possible implementations in different real physical systems, several examples of which will be presented below. Moreover, this special case allows for an elegant analytical treatment. ### V.1 General case We retain the notation for the lower states $|\psi_{n}\rangle$ $(n=1,2,...,N)$, and we denote the two upper states $|\varphi^{\prime}\rangle$ and $|\varphi^{\prime\prime}\rangle$. The interaction matrix (9) reads $\mathsf{V}=\left[\begin{array}[]{cc}V_{1}^{\prime}&V_{1}^{\prime\prime}\\\ V_{2}^{\prime}&V_{2}^{\prime\prime}\\\ \vdots&\vdots\\\ V_{N}^{\prime}&V_{N}^{\prime\prime}\end{array}\right]\equiv\left[\left|V^{\prime}\right\rangle,\left|V^{\prime\prime}\right\rangle\right],$ (37) where $\left|V^{\prime}\right\rangle$ and $\left|V^{\prime\prime}\right\rangle$ are $N$-dimensional interaction vectors comprising the couplings between the lower states and the corresponding upper state. The product $\mathsf{V}^{{\dagger}}\mathsf{V}$ reads $\mathsf{V}^{\dagger}\mathsf{V}=\left[\begin{array}[]{cc}\left|V^{\prime}\right|^{2}&\left\langle V^{\prime}|V^{\prime\prime}\right\rangle\\\ \left\langle V^{\prime\prime}|V^{\prime}\right\rangle&\left|V^{\prime\prime}\right|^{2}\end{array}\right].$ (38) With the introduction of parameters $\theta$ and $\sigma$ through the definitions $\displaystyle\frac{2|\langle V^{\prime}|V^{\prime\prime}\rangle|}{|V^{\prime\prime}|^{2}-|V^{\prime}|^{2}}$ $\displaystyle=$ $\displaystyle\tan 2\theta\quad(0<\theta<\pi/2),$ (39a) $\displaystyle\arg\langle V^{\prime}|V^{\prime\prime}\rangle$ $\displaystyle=$ $\displaystyle\sigma,$ (39b) we write the eigenvalues $\lambda_{m}$ and the associated eigenvectors $|\beta_{m}\rangle$ ($m=1,2$) within the upper set as $\displaystyle\lambda_{1,2}^{2}$ $\displaystyle=$ $\displaystyle\frac{|V^{\prime}|^{2}+|V^{\prime\prime}|^{2}}{2}\pm\frac{|V^{\prime}|^{2}-|V^{\prime\prime}|^{2}}{2\cos 2\theta},$ (40a) $\displaystyle|\beta_{1}\rangle$ $\displaystyle=$ $\displaystyle\left[\begin{array}[]{c}\cos\theta\\\ -e^{-i\sigma}\sin\theta\end{array}\right],\qquad|\beta_{2}\rangle=\left[\begin{array}[]{c}e^{i\sigma}\sin\theta\\\ \cos\theta\end{array}\right].$ (40f) The next step is to find the MS states $|\alpha_{1}\rangle$ and $|\alpha_{2}\rangle$ within the lower set of states. They are the eigenstates of the $N$-dimensional matrix, $\mathsf{V}\mathsf{V}^{{\dagger}}=|V^{\prime}\rangle\langle V^{\prime}|+|V^{\prime\prime}\rangle\langle V^{\prime\prime}|,$ (41) which correspond to the (nonzero) eigenvalues (40a). We construct them as superpositions of the interaction vectors $|V^{\prime}\rangle$ and $|V^{\prime\prime}\rangle$, and find after simple algebra $\displaystyle|\alpha_{1}\rangle$ $\displaystyle=$ $\displaystyle\frac{1}{\lambda_{1}}\left(\cos\theta\left|V^{\prime}\right\rangle-e^{-i\sigma}\sin\theta|V^{\prime\prime}\rangle\right),$ (42a) $\displaystyle|\alpha_{2}\rangle$ $\displaystyle=$ $\displaystyle\frac{1}{\lambda_{2}}\left(e^{i\sigma}\sin\theta|V^{\prime}\rangle+\cos\theta|V^{\prime\prime}\rangle\right).$ (42b) ### V.2 The propagators After we have found the explicit form of the MS states (40f) and (42), we obtain the exact form of the propagators $\mathsf{U}_{N}$ and $\mathsf{U}_{M}$, $\displaystyle\mathsf{U}_{N}$ $\displaystyle=$ $\displaystyle\mathsf{M}(\alpha_{1},\phi_{1})\mathsf{M}(\alpha_{2},\phi_{2})$ (43a) $\displaystyle=$ $\displaystyle\mathsf{I}+(e^{i\phi_{1}}-1)|\alpha_{1}\rangle\langle\alpha_{1}|+(e^{i\phi_{2}}-1)|\alpha_{2}\rangle\langle\alpha_{2}|,$ (43b) $\displaystyle\mathsf{U}_{M}$ $\displaystyle=$ $\displaystyle e^{-i\delta}\left[e^{-i\phi_{1}}|\beta_{1}\rangle\langle\beta_{1}|+e^{-i\phi_{2}}|\beta_{2}\rangle\langle\beta_{2}|\right],$ (43c) with $\phi_{1}$ and $\phi_{2}$ being the phases of the Cayley-Klein parameters for the MS two-state propagators (18). It is easy to verify that the bright states (42a) and (42b) are eigenstates of the propagator $\mathsf{U}_{N}$ with eigenvalues $e^{i\phi_{1}}$ and $e^{i\phi_{2}}$, respectively. Physically this means that if the qunit starts in one of these states it will end up in this same state, acquiring only a phase factor. This occurs because of the conditions (22) and the independence of the different MS two-state systems. If the QHR phases are equal, $\phi_{1}=\phi_{2}$, then any superposition of $|\alpha_{1}\rangle$ and $|\alpha_{2}\rangle$ is also an eigenvector of $\mathsf{U}_{N}$. The other eigenstates of the propagator are all degenerate, with unit eigenvalue, and they are orthogonal to $|\alpha_{1}\rangle$ and $|\alpha_{2}\rangle$. For a qutrit ($N=3$) there is only one such eigenvector (up to an unimportant global phase factor) and it is proportional to $|\alpha_{1}\rangle\times|\alpha_{2}\rangle$. For higher-dimensional qunits, any vector in a hyperplane orthogonal to $|\alpha_{1}\rangle$ and $|\alpha_{2}\rangle$ is an eigenvector of $\mathsf{U}_{N}$. In the special case when the vectors $\left|V^{\prime}\right\rangle$ and $\left|V^{\prime\prime}\right\rangle$ are orthogonal, $\left\langle V^{\prime}|V^{\prime\prime}\right\rangle=0$, the expressions simplify. Then $\theta=0$ and $\sigma=0$, and hence $\displaystyle\lambda_{1}$ $\displaystyle=$ $\displaystyle|V^{\prime}|,\quad\lambda_{2}=|V^{\prime\prime}|,$ (44a) $\displaystyle|\alpha_{1}\rangle$ $\displaystyle=$ $\displaystyle|\hat{V}^{\prime}\rangle,\quad|\alpha_{2}\rangle=|\hat{V}^{\prime\prime}\rangle,$ (44b) $\displaystyle|\beta_{1}\rangle$ $\displaystyle=$ $\displaystyle|\varphi_{1}\rangle,\quad|\beta_{2}\rangle=|\varphi_{2}\rangle.$ (44c) Then the propagator $\mathsf{U}_{N}$ in the lower set is a product of QHRs, in which the interaction vectors $|V^{\prime}\rangle$ and $|V^{\prime\prime}\rangle$ serve as QHR vectors, while the propagator $\mathsf{U}_{M}$ in the upper set is a phase gate, $\displaystyle\mathsf{U}_{N}$ $\displaystyle=$ $\displaystyle\mathsf{M}(\hat{V}^{\prime},\phi_{1})\mathsf{M}(\hat{V}^{\prime\prime},\phi_{2}),$ (45a) $\displaystyle\mathsf{U}_{M}$ $\displaystyle=$ $\displaystyle e^{-i\delta}\begin{bmatrix}e^{-i\phi_{1}}&0\\\ 0&e^{-i\phi_{2}}\end{bmatrix}.$ (45b) ### V.3 Examples Following are examples of linkage patterns amongst angular-momentum states, which allow application of the QHR theory. #### V.3.1 Two levels, $J=3/2\leftrightarrow J=1/2$ Figure 3: (Color online) Linkage pattern for the four degenerate magnetic sublevels of $J=3/2$, the lower set, shown coupled by arbitrary polarization of electric-dipole radiation to the two sublevels of $J=1/2$, the upper set. The states are labeled by their magnetic quantum number. Figure 3 shows linkage patterns possible with arbitrary polarization between the four magnetic sublevels of $J=3/2$, the lower set, and the two of $J=1/2$, the upper set. The interaction matrix has the elements (with the Clebsch- Gordan coefficients included), $\mathsf{V}=\frac{1}{\sqrt{6}}\begin{bmatrix}\sqrt{3}V_{+}&0\\\ -\sqrt{2}V_{0}&V_{+}\\\ V_{-}&-\sqrt{2}V_{0}\\\ 0&\sqrt{3}V_{-}\end{bmatrix},$ (46) where the subscripts $+$, $-$ and 0 refer to right circular ($\sigma^{+}$), left circular ($\sigma^{-}$) and linear ($\pi$) polarizations. Figure 4: (Color online) Time evolution of the numerically calculated populations of the magnetic sublevels of a $J=3/2$ level coherently coupled to a $J=1/2$ level by three polarized ($\sigma^{+}$, $\sigma^{-}$ and $\pi$) pulsed laser fields, with sech shape $f(t)=\text{sech}(t/T)$, $V_{-}=V_{0}=V_{+}=8.5T^{-1}$ and detuning $\Delta=80T^{-1}$. Then $\theta=\pi/4$, $\sigma=\pi$, $\lambda_{1}T=106.3$, $\lambda_{2}T=38.2$, $\phi_{1}=2.65772$, $\phi_{2}=0.954776$. The arrows on the right indicate the values derived by the QHR theory. Figure 4 shows an example of time evolution of the populations of the magnetic sublevels in the $J=3/2$ level, starting with all population in state $|\psi_{-3/2}\rangle$. The conditions $a_{n}=e^{i\phi_{n}}\ (n=1,2)$ are realized approximately, by using large detuning from the upper $J=1/2$ level. In the end of the interaction, the numerically calculated populations are seen to approach the values predicted by the QHR theory (the arrows on the right). #### V.3.2 Two levels, $J=2\leftrightarrow J=1$ The linkage pattern for electric-dipole couplings between sublevels of $J=2$, the lower set, and $J=1$, the upper set will, for polarization expressed as a combination of left- and right-circular polarization, appear as two uncoupled systems: three states form a $\Lambda$-linkage, while five form an M-linkage, as shown in Fig. 5. The interaction matrix is $\mathsf{V}=\frac{1}{\sqrt{10}}\begin{bmatrix}\sqrt{6}V_{+}&0\\\ V_{-}&V_{+}\\\ 0&\sqrt{6}V_{-}\end{bmatrix}.$ (47) Then $\tan 2\theta=|V_{+}V_{-}|/[5(|V_{-}|^{2}-|V_{+}|^{2})]$ and $\sigma=\arg V_{+}-\arg V_{-}$. Figure 5: (Color online) Linkage pattern for five states in an $M$ configuration. #### V.3.3 Three levels, $J=0\leftrightarrow J=1\leftrightarrow J=0$ The MS transformation can be applied not only to a pair of degenerate levels but also to a ladder of degenerate levels, as long as there is only a single detuning. Figure 6 illustrates an example in an angular momentum basis between the magnetic sublevels of three levels with angular momenta $J=0,1,0$. The interaction matrix has the form $\mathsf{V}=\begin{bmatrix}V_{+}^{\prime}&V_{+}^{\prime\prime}\\\ V_{0}^{\prime}&V_{0}^{\prime\prime}\\\ V_{-}^{\prime}&V_{-}^{\prime\prime}.\end{bmatrix},$ (48) and the formalism of this section applies. In this case all fields can be changed independently. Figure 6: (Color online) Linkage pattern for three-level ladder involving a degenerate middle level. The two ends of the chain, magnetic sublevels with $J=0$ have the same detuning from the three intermediate sublevels of $J=1$. The various linkages are invoked by adjusting the direction of the polarization with respect to the quantization axis. The sublevels with $J=1$ form the lower set, while those with $J=0$ form the upper set. ## VI Discussion and conclusions We have here extended the earlier work on QHR in Hilbert space to allow more general linkage patterns between the quantum states, with particular attention to degenerate sublevels that occur with angular momentum states. The extension relies on the use of the Morris-Shore transformation to reduce the original multi-linkage Hamiltonian to a set of independent two-state systems, thereby allowing the utilization of various known two-state analytic solutions. Such solutions, when expressed in terms of Cayley-Klein parameters, readily lead to conditions upon the pulse areas and the time-varying detunings of the excitation pulses. We have found that the propagator within the lower (or upper) set of degenerate states represents coupled quantum-state mirrors. The realization of these coupled QHRs within the lower (or upper) set of states requires certain conditions on the interaction parameters, specifically that all transition probabiities between the lower and upper sets must vanish. We have proposed three physical realizations of this condition: with resonant, near-resonant (hyperbolic-secant), and far-off-resonant pulses. Resonant and near-resonant pulses provide exactly zero transition probabilities, but offer less flexibility in the QHR phases $\phi_{n}$; moreover they require a carefully chosen pulse area in each MS two-state system, which leads to a number of conditions on the interaction parameters. Far-off-resonant pulses fulfill the zero-probability conditions only approximately, but offer much more flexibility. Then the only restriction is for sufficiently large detuning, without specific constraints on pulse areas, because the zero-probability conditions are fulfilled simultaneously in all MS two-state systems. For angular-momentum states, there are six independent interaction parameters: three polarization amplitudes, two relative phases between different polarizations, and the common detuning. Therefore the constructed QHR has six free parameters (with the far-off-resonance realization). For a more general linkage, the number of independent parameters can be much larger. In a subsequent paper multiQHR , we shall describe the mathematical aspects of this largely unknown operator of coupled reflections, which however, as we have shown here, arises naturally in quantum systems. In particular, we shall show how one can factorize, and therefore synthesize, an arbitrary U($N$) propagator by such objects. The procedure is more efficient than a set of rotations that would produce the same transformation. We conclude by pointing out that the confinement of the statevector evolution to the lower set of states, and the availability of simple and efficient tools for its engineering, such as coupled QHRs, can be an essential ingredient for decoherence-free quantum computing decoherence . ###### Acknowledgements. Quantum information processing, and the needed manipulation of statevectors in Hilbert space, has long interested Sir Peter Knight, in whose honor the present issue of Journal of Modern Optics has been assembled. We are pleased to offer the present article in his honor. This work has been supported by the EU ToK project CAMEL (Grant No. MTKD-CT-2004-014427), the EU RTN project EMALI (Grant No. MRTN-CT-2006-035369), and the Alexander von Humboldt Foundation. ## References * (1) M.A. Nielsen and I.L. Chuang, _Quantum Computation and Quantum Information_ (Cambridge University Press, Cambridge, 2000); D. Bouwmeester, A. Ekert and A. Zeilinger, _The Physics of Quantum Information: Quantum Cryptography, Quantum Teleportation, Quantum Computation_ (Springer-Verlag, Berlin, 2000); S. Bose, P.L. Knight, M. Murao, M.B. Plenio and V. Vedral, Phil. Trans. Roy. Soc. Lond. A 356 1823 (1998). * (2) L.Allen and J.H. Eberly, _Optical Resonance and Two-Level Atoms_ (Dover, New York, 1987). * (3) B.W. Shore, _The Theory of Coherent Atomic Excitation_ (Wiley, New York, 1990). * (4) A.S. Householder, J. ACM 5 339 (1958); J.H. Wilkinson, Comput. J. 3 23 (1960); J.H. Wilkinson, Numer. Math. 4 354 (1962); J.M. Ortega, Numer. Math. 5 211 (1963); D.J. Mueller, Numer. Math. 8 72 (1966). * (5) E.S. Kyoseva and N.V. Vitanov, Phys. Rev. A 73 023420 (2006). * (6) P.A. Ivanov, E.S. Kyoseva and N.V. Vitanov, Phys. Rev. A 74 022323 (2006). * (7) P.A. Ivanov, B.T. Torosov and N.V. Vitanov, Phys. Rev. A 75 012323 (2007). * (8) J.R. Morris and B.W. Shore, Phys. Rev. A 27 906 (1983). * (9) N.V. Vitanov, J. Phys. B 33 2333 (2000); N.V. Vitanov, Z. Kis and B.W. Shore, Phys. Rev. A 68 063414 (2003). * (10) Z. Kis, A. Karpati, B.W. Shore and N.V. Vitanov, Phys. Rev. A 70 053405 (2004); Z. Kis, N.V. Vitanov, A. Karpati, C. Barthel and K. Bergmann, Phys. Rev. A 72 033403 (2005); A.A. Rangelov, N.V. Vitanov and B.W. Shore, Phys. Rev. A 74 053402 (2006). * (11) J.C. Diels and W. Rudolph, _Ultrashort Laser Pulse Phenomena: Fundamentals, Techniques, and Applications on a Femtosecond Time Scale_ (San Diego: Academic, 1996); M. Wollenhaupt, V. Engel and T. Baumert, Annu. Rev. Phys. Chem. 56 25(2005); T. Brixner, T. Pfeifer, G. Gerber, M. Wollenhaupt and T. Baumert, in _Femtosecond Laser Spectroscopy_ , edited by Hannaford P (New York: Springer, 2005), Chap. 9. * (12) C.P. Slichter, _Principles of Magnetic Resonance_ (Berlin: Springer, 1990). * (13) I.I. Rabi, Phys. Rev. 51 652 (1937). * (14) L.D. Landau, Physik Z. Sowjetunion 2 46 (1932); C. Zener, Proc. R. Soc. Lond. Ser. A 137 696 (1932). * (15) N. Rosen and C. Zener, Phys. Rev. 40 502 (1932). * (16) F.T. Hioe, Phys. Rev. A 30 2100 (1984). * (17) A. Bambini and P.R. Berman, Phys. Rev. A 23 2496 (1981). * (18) Yu.N. Demkov and M. Kunike, Vestn. Leningr. Univ. Fiz. Khim. 16 39 (1969); see also F.T. Hioe and C.E. Carroll, Phys. Rev. A. 32 1541 (1985); J. Zakrzewski, Phys. Rev. A 32 3748 (1985). * (19) Yu.N. Demkov, Sov.Phys.-JETP 18 138 (1964); N.V. Vitanov, J. Phys. B 26 L53, erratum ibid. 26 2085 (1993). * (20) E.E. Nikitin, Opt. Spectrosc. 13 431 (1962); E.E. Nikitin, Discuss. Faraday Soc. 33 14 (1962); E.E. Nikitin, Adv. Quantum Chem. 5 135 (1970); N.V. Vitanov, J. Phys. B 27 1791 (1994). * (21) C.E. Carroll and F.T. Hioe, J. Phys. A: Math. Gen. 19 3579 (1986). * (22) N.V. Vitanov, T. Halfmann, B.W. Shore and K. Bergmann, Ann. Rev. Phys. Chem. 52 763 (2001); N.V. Vitanov, M. Fleischhauer, B.W. Shore and K. Bergmann, Adv. At. Mol. Opt. Phys. 46 55 (2001). * (23) W. Magnus W., Comm. Pure Appl. Math. 7 649 (1954). * (24) A.M. Dykhne, Sov. Phys. JETP 11 411 (1960); J.P. Davis and P. Pechukas, J. Chem. Phys. 64 3129 (1976). * (25) G.S. Vasilev and N.V. Vitanov, Phys. Rev. A 70 053407 (2004). * (26) G.S. Vasilev and N.V. Vitanov, J. Chem. Phys. 123 174106 (2005). * (27) N.V. Vitanov and P.L. Knight, Opt. Commun. 121 31 (1995); N.V. Vitanov and P.L. Knight, J. Phys. B 28 1905 (1995). * (28) N.V. Vitanov and P.L. Knight, Phys. Rev. A 52 2245 (1995). * (29) J.R. Ackerhalt, J.H. Eberly and B.W. Shore, Phys. Rev. A 19 248 (1979); P.M. Radmore and P.L. Knight, J. Phys. B 15 561 (1982); C.E. Carroll and F.T. Hioe, Phys. Rev. A 36 724 (1987); C.E. Carroll and F.T. Hioe,J. Math. Phys. 29 487 (1988). * (30) J.H. Eberly, B.W. Shore, Z. Bialynicka-Birula and I. Bialynicki-Birula, Phys. Rev. A 16 2038 (1977); Z. Bialynicka-Birula, I. Bialynicki-Birula, J.H. Eberly and B.W. Shore, Phys. Rev. A 16 2048 (1977); B.W. Shore and J.H. Eberly, Opt. Commun. 24 83 (1978); R.J. Cook and B.W. Shore, Phys. Rev. A 20 539 (1979). * (31) E. Arimondo, Progress in Optics 35 259 (1996). * (32) F.R. Gantmacher, _Matrix Theory_ (Springer, Berlin, 1986). * (33) G.S. Vasilev, S.S. Ivanov and N.V. Vitanov, Phys. Rev. A 75 013417 (2007). * (34) P.A. Ivanov, to be published. * (35) A. Beige, D. Braun, B. Tregenna and P.L. Knight, Phys. Rev. Lett. 85 1762 (2000).

arxiv-papers

2024-09-04T02:48:54.035437